#1113 (moving pistons) and #1115 (break particles, which gave particle_sprite
a position for its biome tint) each passed alone; together main no longer
compiled. A moving block now takes its tint at its own block position, as
vanilla's moving-block renderer reads the level's tint there.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Verify and CI parity pass (the earlier online_mode failure was a race and passed on rerun). Moving blocks are drawn and solid; entity pushing is in FEATURES.md LATER 4h.
🤖 Agent: merged. Verify passes. Proof: the B5 integration of these eight heads (parity --changed 102/102, lan_e2e 46–47/48 with only order-dependent flakes), and the conflict resolution in tools/lan_e2e.py checked def by def against each branch.
🤖 Agent: merged. Verify passes; local parity --changed origin/main passes (101 harnesses; #1159's one online_mode diff reran clean and passes on main). Screenshots and tick-by-tick results in the body.
🤖 Agent: merged. Verify passes; local parity --changed origin/main passes (101 harnesses; #1159's one online_mode diff reran clean and passes on main). Screenshots and tick-by-tick results in the body.
A new Horizon Extras option, Lighting > Dynamic lights (`dynamicLights`,
off by default): a torch, lantern or other light-emitting item held by a
player or mob, dropped as an item, or any entity on fire lights the blocks,
entities and hand around it, as LambDynamicLights and OptiFine do.
Client-only and at render time: the light engine, chunk meshes and saves
never see it. Each frame the game picks the 16 sources nearest the camera
(dynamic_light::select) and passes them in Globals; terrain.wgsl and
entity.wgsl raise each vertex's block light to `level - distance` when that
is brighter. Off, the count in globals.time.z is 0, nothing is gathered and
the shaders' loop runs zero times, so no pipeline is rebuilt either way.
An item glows like its block's default state (torch 14, lantern and
glowstone 15, lit redstone torch 7); a lava bucket 15; burning entities 15.
Fixes#814
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The guest came home into its Nether portal's frame, with an obsidian
ceiling three blocks up, and the host onto the End portal it had set at
spawn; both stayed there for the rest of the run. Under that ceiling a
double jump rises about 1.2 blocks at most, and whether the flight check
saw it depended on timing: `flight` failed in full --server-thread runs
(portals runs only there) and passed alone. The check now clears the
frame and the End portal in the overworld, and puts the guest back where
it started, also when it fails.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On single player's server thread the broken log's drop and its pickup
both come from the server, a few ticks after `break` answers: the
pickup could land before the scenario looked for the drop, which it
then never found (StopIteration). Step 3 now waits up to 20 ticks for
the logs in the inventory or the drop to walk to.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Step 2 clones a stack in the creative tab and closes the screen. The
clone lived only on the client: ItemPickerMenu's carried stack is the
inventoryMenu's, and AbstractContainerMenu.removed clears the carried
stack only for a ServerPlayer, so vanilla's client still holds it when
the survival inventory opens. With a stack in hand the middle button's
release sends CLONE (AbstractContainerScreen.mouseReleased); the server
finds the carried stack it was told about wrong and sends the empty one
(broadcastChanges). On single player's server thread that answer takes
a round trip, as in vanilla, so step 3 now waits up to 20 ticks for it.
The in-frame world puts the clone away when the menu closes.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
From b9f3821f (fix/server-thread-default): the host window draws what
it was sent on the server thread, and agent_control_e2e waits two ticks
where the server answers a click.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On a client of a server (single player's server thread, a LAN guest) the
placed block comes with the server's answer: `place` waits up to a second
for it before reporting no change. agent_control_e2e tries the next free
spot when the tree's trunk above the broken log hides a spot's faces.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit 9821bb7ef5)
- The command harness (HORIZON_COMMAND_SCRIPT) drives the world itself,
without the event loop, so it does not start the server thread.
- `status` lists the minecarts a client of a server was sent.
- save_roundtrip: single player's server thread finishes the chunks its
player tracks, as vanilla's integrated server does, including some that
vanilla's server (view distance 4) left before their terrain or light.
Those had no blocks, light or heightmaps to compare yet.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit 2818c89f24)
Single player's world now runs on the integrated server thread unless
asked otherwise, as vanilla's does, so a world tick no longer holds up
a frame. `--no-server-thread` or `HORIZON_SERVER_THREAD=0` keeps the
world in the frame as an escape hatch; `--server-thread` stays.
lan_e2e.py sets the host's mode explicitly either way.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit 4ff23b686d)
🤖 Agent: merged. Verify passes; CI parity's only failure is the known dedicated_server pearl flake. Measured panning against OpenAL Soft's stereo output in the body.
🤖 Agent: merged. Verify passes. CI parity's only failure is server_player_targets, a server harness this client rendering change can't touch; it has flaked before (its log said '109 lines identical' on #1118). Screenshots in the body.
🤖 Agent: merged. Verify and CI parity pass. Mod browser part 1/3: Modrinth source and checksum verification. The CurseForge source stays inactive until a key is configured.
A player who joined already in creative could not fly: the remote
client kept only instabuild and the flying speed of
ClientboundPlayerAbilitiesPacket, and the motion step clears flying for
a player that may not fly. The packet also comes before the client has
a player (it is made at the first teleport). Single player creative on
its server thread could not fly at all; a LAN guest only after a game
mode change.
- The remote client keeps the last abilities and applies flying, may
fly, invulnerable and the flying speed to the player, also when the
player is made (ClientPacketListener.handlePlayerAbilities).
- When its flight turns on or off (double jump, spectator, landing) it
sends ServerboundPlayerAbilitiesPacket (LocalPlayer.onUpdateAbilities).
- The hub keeps it as the guest's flying when the guest may fly
(handlePlayerAbilities), so the save has it, and a game mode change
sets it as GameType.updatePlayerAbilities does.
- lan_e2e `flight`: the guest, and on the server thread the host's
window, fly in creative, stay up, and stop in survival.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla plays a block use's sound with `level.playSound(player, ...)`: the
player who clicked hears it from their own client's use, and the server
sends it to everyone else. Horizon's guest waited for the server's sound,
one round trip after its predicted door had already opened.
The guest now plays the sounds of the use it predicts (doors, trapdoors,
fence gates, levers, buttons), and the server sends the sounds of such a
use to the other players only. The control socket's `status` counts the
positional sounds each game played (`heard`).
lan_e2e `predict`: with the host frozen, the guest already hears its door;
once the server answers, the guest and the host have each heard it once.
Fixes#869
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A client of a server sent the bow's use but never started using the item
itself, so letting go of the button sent no RELEASE_USE_ITEM: the server
kept the guest drawing and never shot. BowItem.use runs on the client too
(startUsingItem, with an arrow or in creative), and Minecraft.handleKeybinds
releases the item in use once the button is up.
The guest now draws on its own side as well (held until the button lets go
or another slot is picked), so the server shoots when it lets go.
lan_e2e: `arrows` passes both ways now (the guest's arrow hurts the host);
the `pvp` and `arrows` checks allow the victim natural regeneration while
pvp is off (it healed 19.3 -> 20 and failed the exact check).
Fixes#484
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla's MultiPlayerGameMode.useItemOn runs the use on the client's own
level inside startPrediction, records each block it changes under the
packet's sequence in BlockStatePredictionHandler, and keeps them until
ClientboundBlockChangedAckPacket settles them (a use the server refused
goes back). A Horizon guest only sent the use: its door opened one round
trip later.
The guest already ran the use on a dry copy of its level to know whether
the click ends there; the blocks that use changes are now kept on its
level as predictions under the use's sequence, as the break prediction
does, when they are all hand-operated blocks: a wooden or copper door,
trapdoor or fence gate, a lever or a button (an iron door has no hand use
and changes nothing). The sound still comes from the server.
lan_e2e `predict`: with the host process frozen, the oak door the guest
clicks is already open on the guest's side (both halves), stays open on
both sides once the server answers, and an iron door stays shut.
Fixes#869
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A remote player's projectiles never hit other players: a player's shots
only looked for mobs and the host, a guest's flew as if the host had shot
it (so it could not hit the host until it left the host's box), and a hit
on a guest from another guest ran as the shooter, who has no hub to reach
the victim. No bow PvP on a server or between LAN players.
A player's arrows, snowballs and eggs now meet every player, a guest's
flying from that guest (Projectile.getOwner). A hit on another player runs
as the victim's hurt, refused when the pvp game rule is off or the teams
are allies without friendly fire (ServerPlayer.canHarmPlayer), which turns
the arrow back as a raised shield does (AbstractArrow.onHitEntity).
Pearls, potions and wind charges fly past players as before.
lan_e2e `arrows`: the guest shoots the host and the host the guest with a
bow, both hurt; with pvp off, no harm.
Part of #484
Co-Authored-By: Claude Opus 5.5 <[email protected]>
LivingEntity.die credits the kill to lastHurtByPlayer: the killed
statistic, the mob_kills custom statistic, the kill score criteria and the
player_killed_entity trigger belong to that player. Horizon kept only
whether a player had hurt the mob, and the death ran as the host, so a LAN
or server guest's kills counted in the host's statistics and advancements.
The mobs a player hurt now remember which player (the acting guest's
connection, or the host), and the killer's side of the death runs as that
guest.
lan_e2e `kills`: a pig the guest kills shows in the guest's
minecraft.killed:minecraft.pig score, not the host's.
Part of #484
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The hub's players are server players too, and in vanilla the armour they
wear counts against every hit (LivingEntity.collectEquipmentChanges puts
each piece's attribute modifiers on the player, getDamageAfterArmorAbsorb
reads them). Horizon only put them on the host: a guest who put armour on
in its inventory menu (the hub moves it), joined wearing it, respawned with
it kept or came through a portal (a new body each time) was hurt as if it
wore nothing. Iron armour, 15 points, should take a 10-damage blow down to
6; a guest took the full 10.
Each guest now keeps its own list of applied equipment modifiers, swapped
in with the rest of its state for its turn, and its armour and main hand
are put on its attributes whenever they differ from what they were put on
for (a new body starts with none).
lan_e2e `worn`: the guest shift-clicks iron armour on in its inventory and
takes 10 mob-attack damage: 6 with it, 10 without (failed before: 10 and 10).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit e0549f88ce)
ServerExplosion.hurtEntities takes every entity in range, players
included, but only the host player was hurt and pushed: LAN guests stood
in a blast untouched. The player's part is now its own function, run for
the host and then for each guest in the level, swapped in as the player.
It also follows Player.hurtServer more closely for everyone: the damage
goes through a raised shield (an explosion is blocked like any attack
from where the blast is, the shield takes wear), the push is reduced by
explosion knockback resistance, and a spectator or a creative player
flying is not pushed.
tools/lan_e2e.py gains a `blast` check: primed TNT five blocks from a
survival guest hurts it (20 -> 4.6) and pushes it away; facing it behind
a raised shield, the guest takes nothing. Before this change the guest
took no damage and no push.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit 0782a598fe)
PiglinAi.throwItemsTowardPlayer aims at NEAREST_VISIBLE_PLAYER: the
nearest living, non-spectator player within 16 blocks. Only the host was
considered, so a LAN guest who bartered saw the loot fly to the host.
lan_e2e's new barter check: the guest, in gold boots, drops a gold ingot
to a piglin in a glass corridor with the host on the far side; the
piglin stays calm and the barter reaches the guest. It fails on the old
build (the loot lands on the host's side).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit 896f12f732)
Player.attack on a #minecraft:redirectable_projectile (a ghast's fireball,
a wind charge) deflects it with ProjectileDeflection.AIM_DEFLECT: its
movement becomes the attacker's unit view vector and the attacker its
owner, with the no-damage sound. The crosshair finds these projectiles a
block round their box (AbstractHurtingProjectile.getPickRadius), a remote
client sends the attack for them, and the host turns one a guest strikes
along that guest's view. A fireball a player turned back kills the ghast
it hits (Ghast.isReflectedFireball: no fire immunity, 1000 damage).
lan_e2e.py --check deflect: the host strikes a caged ghast's fireball and
the ghast dies; before, the fireball hit the host.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit 6ba02f5205)
SmallFireball.onHitBlock places fire with Level.setBlockAndUpdate and
BaseFireBlock.getState (soul fire on soul soil and soul sand), and a mob's
fireball only when mobGriefing is on. Horizon wrote a plain fire state
straight into the chunk: no block updates and no scheduled fire tick, so
the fire never aged and never went out, even on stone or nether bricks
where vanilla's goes out within about half a minute (FireBlock.tick). Every
blaze fireball that missed left a fire that stayed for good: fortress
floors round a spawner filled with fire that kept everyone near it burning.
The fire is now the level's fire state for the spot, set with updates
(the fire's tick is scheduled on place), and only with mobGriefing when a
mob shot it.
lan_e2e `fireball`: a dispenser shoots a fire charge onto stone; both
players see the fire and it goes out by itself (failed before: still burning
after 1200 ticks; now out after about 480).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit 5b7235c24b)
🤖 Agent: merged. Verify passes. B5's local integration of all its PRs on main: parity --changed 102/102, plus lan_e2e (flakes listed in FEATURES.md LATER 4l).
🤖 Agent: merged. Verify passes. B5's local integration of all its PRs on main: parity --changed 102/102, plus lan_e2e (flakes listed in FEATURES.md LATER 4l).
🤖 Agent: merged. Verify passes. B5's local integration of all its PRs on main: parity --changed 102/102, plus lan_e2e (flakes listed in FEATURES.md LATER 4l).
🤖 Agent: merged. Verify passes. B5's local integration of all its PRs on main: parity --changed 102/102, plus lan_e2e (flakes listed in FEATURES.md LATER 4l).
🤖 Agent: merged. Verify passes. B5's local integration of all its PRs on main: parity --changed 102/102, plus lan_e2e (flakes listed in FEATURES.md LATER 4l).
🤖 Agent: merged. Verify passes. B5's local integration of all its PRs on main: parity --changed 102/102, plus lan_e2e (flakes listed in FEATURES.md LATER 4l).
🤖 Agent: merged. Verify passes. B5's local integration of all its PRs on main: parity --changed 102/102, plus lan_e2e (flakes listed in FEATURES.md LATER 4l).
🤖 Agent: merged. Verify passes. B5's local integration of all its PRs on main: parity --changed 102/102, plus lan_e2e (flakes listed in FEATURES.md LATER 4l).
The Shader Pack setting cycled through what shaderpacks/ held, so with the
folder empty (or the packs in the launcher's version folder, as brabus_ had
them) it silently stayed Off. The row now opens a Shader Packs page: a
sidebar section in the Horizon menus, Video Settings > Shader Packs... in
the Classic ones. It shows the game home's shaderpacks/ path and an Open
Folder button (made if missing), says when there is no pack, and lists one
switch per pack. A pack that fails to load says why on its row. The list is
read again every second while the page shows.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
GL fills the channels a fragment shader leaves out; wgpu refuses a pipeline
whose output has fewer channels than its target. Steadfast's final pass
writes `out vec3` and failed to build in the game.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Preprocessor (GLSL and shaders.properties), as JCPP does:
- a #define replaces an earlier macro (Bloop, Steadfast: DefineRedefined);
- in #if, a float counts as its integer part (Steadfast's
shaders.properties float token), and a macro always defined empty counts
as 0 (Iris's MC_GL_VENDOR_OTHER, Bloop's TAA_USE_YCOCG);
- an #if may run to the end of the input (Bloop's taa.fsh);
- #error "..." lexes in skipped blocks (Photon);
- shaders.properties keeps its empty lines and hides backslashes from the
preprocessor (IRIS_PASSTHROUGHBACKSLASH), so a continued value no longer
swallows the next directive or line (Photon, Bliss, Steadfast's custom
uniforms; Complementary's isCold uniform is now complete).
Legaliser: % through overloads (naga won't take ivec2 % int: Noble),
sqrt of a negative constant left to the GPU (Bloop's NaN), constants naga
can't fold (?:, mix: Bliss) become globals set in main, and 16/8-bit types
(u16vec2 & co., Noble on NVIDIA) become their 32-bit ones.
horizon-client: real_packs_load runs the game's own load over the packs in
HORIZON_TEST_SHADERPACKS (the compatibility table's source).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A Mods page between Versions and Modpacks: resource packs, shader packs and
modpacks as tabs, a search field, Show everything, and Modrinth / CurseForge
once CurseForge has a key. Results are index rows with summary, author,
downloads, source, licence and how well each runs (with the reason), Page and
Install (Import for a modpack, which then shows on the Modpacks page).
Installed packs list their version, source, file and the checksum they were
checked against, with Update and Remove. Searches, installs and update checks
run on threads; stale search answers are dropped.
Fixes#577
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Verify passes. B5's local integration of all its PRs on main: parity --changed 102/102 (the explosion-dependent 22 rerun after rebasing onto #1101), and lan_e2e passes except checks that fail only after earlier checks and pass alone (in FEATURES.md LATER 4l).
🤖 Agent: merged. Verify passes. B5's local integration of all its PRs on main: parity --changed 102/102 (the explosion-dependent 22 rerun after rebasing onto #1101), and lan_e2e passes except checks that fail only after earlier checks and pass alone (in FEATURES.md LATER 4l).
browser::installed installs a pack and its required dependencies into
resourcepacks/ or shaderpacks/, every file checked before any download
(a dependency that is a Java mod stops it) and checksum-verified on the
way in; mod-browser.json records source, version, file and checksum.
Update swaps the file, remove deletes only what the browser installed (a
player's own file of the same name is never overwritten). A modpack is
downloaded, checked, and imported through the Modpacks page's pipeline.
The CLI gains `mods list|search|install|update|remove`.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The launcher's core gains `browser`: search Modrinth for what Horizon loads
(resource packs, Iris/OptiFine shader packs, modpacks; never Java mod jars),
say how well each result runs, pick the newest version that runs, and
download a file only from Modrinth's or CurseForge's file servers, with a
plain file name, kept only when its size and SHA-512 (SHA-1 on CurseForge)
match. CurseForge is a pluggable source that stays off until an API key
exists (HORIZON_CURSEFORGE_API_KEY at run time or build time; the launcher
workflow passes the repository secret). docs/mod-browser.md.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Three differences from vanilla's Axolotl kept the brain from matching:
- The constructor's setPathfindingMalus(WATER, 0) was missing, so every
swim target in water failed DefaultRandomPos's malus check and idle
axolotls never picked a walk target.
- Axolotl.getWalkTargetValue returns 0 everywhere; Horizon used Animal's
light-based value, which picked the brightest candidate instead of the
first.
- Axolotl.playAttackSound (entity.axolotl.attack at pitch 1) never played.
The mob AI harness gets #64's two scenarios, in a pool sunk into the ground
so no water flows out of it: hunting a cod, and playing dead after repeated
hits. Both now match vanilla on every tick (393/393 mob-ticks, was 26).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Verify passes. CI parity's failures are only the known flakes: world_spawn (vanilla gives several spawn heights for one seed across runs) and dedicated_server (play-pearl). This client rendering change can't move either. B2's integration passed every rendering harness, and before/after screenshots are in the body.
🤖 Agent: merged. Verify passes. CI parity's failures are only the known flakes: world_spawn (vanilla gives several spawn heights for one seed across runs) and dedicated_server (play-pearl). This client rendering change can't move either. B2's integration passed every rendering harness, and before/after screenshots are in the body.
On a server or a world open to LAN, a player saw only their own crack.
The server sent each new stage to the others
(ClientboundBlockDestruction), but a client of a server ignored the
packet, and the host of a LAN world neither drew its guests' cracks nor
sent its own to them.
Vanilla keeps the cracks per breaker (ClientLevel.destroyBlockProgress:
a progress outside 0..9 removes it, 400 ticks without news drops it) and
draws, within 32 blocks, the furthest one on each block
(LevelExtractor.extractBlockDestroyAnimation). Horizon now does the
same: a client of a server takes the packet, a LAN host records its
guests' stages and sends its own (each new stage, then -1 when it
stops), and every crack is drawn with its own stage. A guest's crack
also goes when the block it digs is gone before its STOP_DESTROY_BLOCK
(its own tick broke it), as ServerPlayerGameMode.tick clears it on air:
the others kept its last stage for 400 ticks. Each crack vertex carries
its stage sprite's atlas origin in its light bytes (the crumbling pass
is unlit); the shader keeps the shared sprite size.
lan_e2e `cracks`: each side watches the block the other mines and sees
its crack grow; no crack is left once the blocks break. The screenshots
from both sides show the cracks.
Fixes#1096
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Verify passes. Local integration of #1121–#1125 on main 2d586fac: parity --changed, all 101 affected harnesses pass. Screenshots in the body.
🤖 Agent: merged. Verify passes. Local integration of #1121–#1125 on main 2d586fac: parity --changed, all 101 affected harnesses pass. Screenshots in the body.
🤖 Agent: merged. Verify passes. Local integration of #1121–#1125 on main 2d586fac: parity --changed, all 101 affected harnesses pass. Screenshots in the body.
🤖 Agent: merged. Verify passes. Local integration of #1121–#1125 on main 2d586fac: parity --changed, all 101 affected harnesses pass. Screenshots in the body.
🤖 Agent: merged. Verify passes. Local integration of #1121–#1125 on main 2d586fac: parity --changed, all 101 affected harnesses pass. Screenshots in the body.
PotentSulfurBlock.animateTick sends two sulfur_bubbles up through a water
source above a wet potent sulfur block, and one time in ten plays
block.potent_sulfur.noxious_gas. SulfurBubbleParticle rises at -0.04
gravity with 0.85 friction, wiggles sideways, grows from 0.02 to 0.15 and
pops when it leaves a water source, stops rising or travels 3 blocks. Its
provider passes yAux as the Z speed, which is kept.
The particle providers harness gets water worlds for sulfur_bubbles,
bubble and bubble_column_up.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Verify and CI parity pass, and the independent review is OK: 12 update tests; v0.2.0.1 ranks below every 0.2 tag; a 0.2 install is never offered a dated tag.
From 0.2 the game is tagged vX.Y.Z and the launcher [email protected] (docs/versioning.md). The site
took only vYYYY.MM.DD as a game release and launcher-v as a launcher, so it would keep offering
launcher 0.16 and showing the last dated patch notes.
src/versions.js holds the tag test and order (every 0.2 tag above every tag from before, semver
within, a pre-release before its release), shared by build.mjs and main.js; the download button
picks the newest launcher by tag, the notes the newest stable game releases, and the launcher
lookup falls back to launcher@ tags before launcher-v. Nothing shows "0.2" until a 0.2 release
exists. `npm test` runs its tests (node --test).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Ludovic decided (2026-10-21) that the next release is 0.2 for both the game and the launcher, with one
clear scheme instead of vYYYY.MM.DD(.N) and launcher-vX.Y.Z. This page is the design the other 0.2
pull requests implement:
- tags vX.Y.Z and [email protected] (launcher-v0.2.0 is taken), on a line in .github/release-line.txt;
betas stay pre-releases under their final tag; semver pre-release tags reserved for previews;
- one order for every reader: each 0.2 tag above every tag from before;
- how release.yml, launcher.yml and promote.yml map onto it (HORIZON_VERSION, HORIZON_LAUNCHER_TAG);
- what launchers and servers from before 0.2 do with it, and the bridges that move them
(launcher-v0.17.0, the last dated game release made latest before v0.2.0);
- the orchestrator's steps for the day Ludovic says go, freeze off included.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Verify and CI parity pass, and an independent review found no blocking issue. The release, launcher and promote workflows stay disabled with RELEASES_PAUSED=1. Without .github/release-line.txt the train decides exactly as before (checked against every existing tag).
🤖 Agent: merged. Verify and CI parity pass, and an independent review found no blocking issue. The release, launcher and promote workflows stay disabled with RELEASES_PAUSED=1. Without .github/release-line.txt the train decides exactly as before (checked against every existing tag).
🤖 Agent: merged. Verify passes. CI parity's only failure is world_spawn seed -2080727421395798566, where vanilla itself gives several heights across runs (also on main; see #1112). Screenshots in the body.
🤖 Agent: merged. Thank you @feeldev12! Verify passes. A 2-line fix to benchmark tooling: the world folder is created before the marker is written, so slow devices no longer race the async world open.
An ignored test reads vanilla's advancements and item tags from the
prepared resources (HORIZON_RESOURCES) and reports what an inventory
change and a recipe unlock test now. On the Mac mini (release):
inventory_changed has 1914 criteria, the index is built once in 0.75 ms
with the tree (off the tick), and a nine-stack inventory tests 67 of them
(found in 16 us); recipe_unlocked has 1739, an iron pickaxe tests 1.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ServerWaypointManager.removePlayer disconnects every waypoint the player
was receiving: a player who goes to another level gets an untrack for each
player it tracked in the one it left. Horizon's per-level managers tick
with the players in their own level only, so the untrack had no connection
to go to and the player kept the old level's markers.
Each level's manager now also gets the connections of the players in the
other levels, for these untracks.
The player-targets and dimension-players harnesses compare the waypoints
as each step leaves them (the last packet for each waypoint): arriving in a
level, vanilla sends the same waypoint up to three times within the step,
and the packet by packet order is test_locator_bar_parity's. Both pass:
123 and 40 lines identical.
Part of #843
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Two main-thread costs from the hub tick (found on the companion branch with
HORIZON_SPIKES, commit 060c795e; its companion and metering parts stay there):
- Every inventory change tested all ~1,900 inventory_changed criteria, and
every recipe unlocked (an iron ingot unlocks dozens) all ~1,300
recipe_unlocked ones: 1.5-5 ms on such a tick. The tree now indexes
inventory_changed criteria by the items that can pass them (tags resolved;
the few without an item list are always tried) and recipe_unlocked
criteria by recipe, so a change tests the 20-40 that could match. Same
answers: a criterion with an item-limited predicate can only pass with one
of those items held.
- Each log line was an unbuffered stderr write plus a latest.log write under
a mutex on the calling thread; a full pipe or a slow disk held the tick.
The logger now formats the line and keeps it for crash reports on the
calling thread and hands the I/O to a "Log writer" thread; flush() waits
for it (crash report, exit).
(cherry picked from commit 060c795e, companion parts left out)
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla keeps a ServerWaypointManager per level: a player in another level
is neither a transmitter nor a receiver there. Horizon passed every player
to one manager whatever their dimension, so a player who went to the
Nether stayed tracked at their Nether coordinates by the players left in
the Overworld (`track Bob block 0 100 0`), and nothing was re-tracked when
they came back.
Each level now has its own manager, ticked with the players in that level:
a player changing level leaves the old one (untracked for everyone there,
as ServerLevel.removePlayerImmediately) and joins the new one
(addDuringTeleport). The player-targets and dimension-players harnesses
now compare the waypoints too, instead of leaving them out.
Fixes#843
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A guest's body is rebuilt when it respawns, and its hurt and attack logs
went with the old one: the control socket's hurtlog lost the hurts that
killed it whenever the death and the respawn fell between two reads.
The logs now pass to the new body, and a dead guest's are read too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit df9fa5f018)
Every hurt a player takes (PlayerMotion::hurt_blocking, which all player
damage goes through) is kept: the damage type, the amount before blocking
and armour, what a raised blocking item stopped, the health and absorption
lost, the armour and armour toughness it met, the fire ticks, the health
left and, for a mob's blow, the knockback it gave. Every melee blow on a
mob is kept too: the damage after the attack strength
(getAttackStrengthScale), the strength, a critical hit, the mob, whether it
landed and killed. The last 64 of each are kept.
`hurtlog` on the control socket hands them out, per player (the host and
each LAN guest), and clears them: a harness can read what a fight did hit
by hit and compare it with vanilla's numbers.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit fdfa952379)
OminousItemSpawner.addAdditionalSaveData keeps its item and
spawn_item_after_ticks; Horizon saved neither, so an ominous trial
spawner's key or potion floating mid-wave vanished when the world closed
or its chunk unloaded. They are now written with the chunk's entities
(UUID and unmodelled fields kept), read back as live spawners whose tick
count starts again from 0 (Entity.tickCount isn't saved), and leave the
level with their chunk.
Fixes#462
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A sleeping guest stood on top of its bed for everyone else: the server
sent its pose but never its bed (LivingEntity.SLEEPING_POS goes only to
the sleeper itself), Horizon's clients had no sleeping pose for other
players, and the host's own models of its guests ignored the bed. A
player paired with the sleeper later (the host walking up from afar)
saw the same.
The tracked state of a sleeping player (the host or a guest) now holds
its bed, so pairing and updates send it, and its removal sends the empty
position back; the clients and the host's guest models lie the player in
the bed as LivingEntityRenderer.setupRotations does for Pose.SLEEPING
(the villagers' sleeping pose, by the bed's facing).
lan_e2e `sleep`: the host draws the sleeping guest lying, the guest the
sleeping host, and neither once they are up.
Fixes#1045
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Like the dragon, a spike's crystal gets its id from the entity counter as
EndCrystal's constructor does, instead of only once a client tracks it: a
player the level serves without a client can then aim at it. The fight's
control status lists each crystal's id and position.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit e32e5031c1)
The dragon read the window's own player for everything: network players
(LAN guests, players on a server) were never targeted, strafed, charged,
knocked back, bitten, breathed on or hit by its fireballs.
- It sees the level's players (the window's and every guest in the level)
through `TargetingConditions.forCombat` as each phase asks
(`Level.getNearestPlayer`): alive, neither creative nor spectator, none on
Peaceful. The holding pattern strafes the enemy nearest the egg, the
landing approach comes in opposite it, the perched dragon scans within 20
blocks (10 up or down) and charges within 150.
- The strafe keeps its own target (`attackTarget`) and goes back to the
holding pattern when it leaves the level.
- A destroyed crystal sends a holding dragon after the player who destroyed
it, else the nearest player (`EnderDragon.onCrystalDestroyed`), when it can
attack them.
- Wing knockback and the head and neck bites, the breath and fireball clouds
and its fireballs act on every player alike; what hits a guest runs as
that guest, its knockback sent to its client.
- The dragon takes its id and its parts' when it is created, so a guest can
attack them in a world without network players too.
A unit test covers the targeting rules (spectators and creative players
passed over, ranges, Peaceful, the nearest to the egg).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit 2eef4e4bf5)
Primed TNT (and falling blocks, projectiles) spawned from /summon get
their entity id only when first tracked, so the summon found no new id
and answered "Unable to summon entity" though the entity was there and
went off. The summon now also counts those lists, and names the type:
"Summoned new Primed TNT", as vanilla.
Test: a TNT summon in the command execution corpus (world group:
125/126 match vanilla; the one left, /waypoint list, was there before).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit dd733684bc)
The world spawn was only filled in when the world was first saved, from
the player's spawn point at that moment: a bed slept in before then became
the world spawn, and anything asking for the world spawn earlier (a
guest's respawn) fell back on the host's spawn point instead.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit 03e621ed3a)
SetPotionFunction was skipped, so the potions a loot table gives came
out with no contents: a piglin's barter never held fire resistance (the
drinkable and splash ones, 16 of 469), only uncraftable potions. The
merchant and /loot code had it; the shared loot evaluator now does too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit 1b2767122d)
- A perched dragon (scanning, attacking, flaming, hovering) shrugs arrows
off as `AbstractDragonSittingPhase.onHurt` does: no damage, and the arrow
is set alight for a second and turns back instead of being spent. Arrows
that hit a part or a crystal deal the arrow's damage with its Power bonus,
and turn back when the hit is refused (a dying dragon, an invulnerable
crystal).
- The damage cooldown is `LivingEntity.hurtServer`'s: within it a stronger
hit lands its excess over the last one instead of being dropped; the hurt
sound plays only for a new hit.
- Explosions hit as `ServerExplosion.hurtEntities` does: the dragon itself
(its body) and then each part, each by its distance and its exposure
(`getSeenPercent`, so cover protects it), through the cooldown; and every
end crystal in reach blows up, not only the first.
`Dragon::hurt` and `Dragon::explosion_hits` are pure, with unit tests for
the perched arrows, the cooldown's excess, the parts' quarter plus one,
the take-off after a quarter of its health, and exposure.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit c730ed9ed8)
On a dedicated server (and single player's server thread) the End is ticked
with no local player, and `tick_fight` returned before spawning the dragon:
players in the End found crystals but never a dragon.
- The fight finds its players as `EnderDragonFight.updatePlayers` does, every
20 ticks: every player alive within 192 blocks of (0, 128, 0), the host and
the network players in the level alike. Scanning the state, the respawn
sequence, `createNewDragon` and the crystal count run while it has one;
the dragon and crystals tick whenever the End does.
- The dragon is tracked to network players (`minecraft:ender_dragon`, its
health and `DATA_PHASE` as vanilla pairs them, no attribute instances) and
reserves its eight part ids after its own, as `EnderDragon`'s constructor
does, so clients can attack the parts; a network player's attack on a part
or an end crystal is aimed at it.
tools/test_server_dragon_parity.py sends a lone vanilla-codec client to the
End of vanilla's dedicated server and of Horizon's: the fight starts and the
dragon is tracked with the same spawn data, synched data entries and
attributes. VanillaProtocolClient's entity mode answers a scene's teleport
with the acceptance alone, as the 26.3 client does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit 618db927d9)
release.yml builds every game release with HORIZON_VERSION, the tag without its v
(docs/versioning.md), as launcher.yml already does for the launcher: no version-bump commits.
- horizon_core::release::{RELEASE, VERSION}: the built-in version, or Cargo.toml's for a build from
source. (horizon_core::version stays Minecraft's, generated by gen_version_consts.py.)
- horizon-server and horizon-prepare --version, the server's /api/v1/status, horizon-ctl's MCP
serverInfo and the "Horizon/<version>" user agents use it; they all said 0.1.0.
- The client's crash reports and telemetry say v0.2.3 from the built-in version wherever the game
is installed; releases from before 0.2 (no version built in) still read the launcher's folder name.
- The launcher's benchmark telemetry used Cargo.toml's version instead of the release's.
- release.yml checks that horizon-prepare and horizon-server --version say the release's version.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Verify passes. Local integration of #1109–#1112 on main ae21bd54: parity --changed 142/142. After the rebase onto #1094: 101/101, with screenshots in the body.
🤖 Agent: merged. Verify passes. Local integration of #1109–#1112 on main ae21bd54: parity --changed 142/142. CI parity's world_spawn diff (seed -2080727421395798566) also happens on main: vanilla itself gives y64/65/66/72 across runs. The other failure is the known dedicated_server pearl flake.
🤖 Agent: merged. Verify passes. Local integration of #1109–#1112 on main ae21bd54: parity --changed 142/142, with before/after item counts in the body.
The station screens' hand-written quick move put a payment stack's first item
into the beacon, fell back to the other inventory half when an enchanting,
grindstone, stonecutter, cartography or anvil rule's own slots were full, and
sent every input back hotbar-last-first. Follow each menu's quickMoveStack
(EnchantmentMenu, BeaconMenu, GrindstoneMenu, StonecutterMenu,
CartographyTableMenu, ItemCombinerMenu with SmithingMenu's
canMoveIntoInputSlots), repeated as doClick's QUICK_MOVE does.
A local brewing stand's clicks now run on the verified menu engine, as a
server's already did, so its shift-clicks, drags (quick craft) and double
clicks are BrewingStandMenu's, which test_menu_click_parity covers.
Part of #275
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The menu style is one process-wide flag, and options tests switch it to lay
out a page in Horizon's or Classic's style. Running in parallel, one test's
switch moved another's widgets (classic_graphics_presets_set_the_distances
failed 4 of 200 runs of the options tests). Under cfg(test) the flag is a
thread local, Classic first; the game keeps the shared atomic.
Fixes#802
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The title box is an unbordered, centred EditBox (black, no shadow) in
vanilla. Horizon kept Ctrl/Cmd+A's selection but never drew it, and centred
the title together with its cursor. Draw it with ui::append_field (now with
a text shadow flag): the value centred alone, the cursor after it, and the
inverted highlight. horizon-ctl gets the super (Cmd) and F6 key names, and
A/C/V/X with the edit modifier held act on the open book.
Part of #275
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Classic Extras laid every setting out as a full-width button. Pair each
section's buttons as vanilla's OptionsList.addSmall does (the same layout as
Video Settings), and shorten "Lower sensitivity while zoomed" to "Slower
turning" so it fits a half-width button.
Horizon's Extras page shows a switch's own settings (Zoom key, Minimap size,
FPS corner...) only while that switch is on, so it stays one row per extra
as the list grows. Classic keeps every button, as vanilla screens do.
Fixes#812Fixes#815
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A world's session.lock is an flock held by the open file. Closing the file
releases it only once every copy of the descriptor is gone, and a process
spawned on another thread holds a copy until it execs. Under a parallel
cargo test (crash reports spawn uname/sw_vers), deleting a world right after
dropping its lock could then fail with "already open". Unlock explicitly
in Drop, as vanilla releases the FileLock before closing the channel.
Fixes#749
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A centred mono source played at full gain on each channel; OpenAL Soft
(vanilla's LWJGL build, HRTF off) renders stereo pairwise: 0.596 per side
ahead, 0.404 behind, 0.5 above or below. Explosions were 4.5 dB louder than
vanilla. Pan with the camera's forward and up vectors, play a mono output
unpanned, and release the limiter over 0.2 s like OpenAL Soft's.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
docs/versioning.md: with .github/release-line.txt (MAJOR.MINOR) the train tags game releases
vMAJOR.MINOR.PATCH and launcher releases [email protected], the patch after the newest
release on the line; the line's first release goes out at once (still not while RELEASES_PAUSED is
set). Without the file nothing changes: vYYYY.MM.DD[.N] and launcher-vX.Y.Z. The file is not added
here; the switch to 0.2 is a separate one-line pull request for the day Ludovic says go.
- tag_key: every 0.2 tag ranks above every dated or launcher-v tag; semver order within, a
pre-release before its release. release_channels.py uses the same key and tag test, so promotion
and "latest only moves forward" work across the switch.
- A line's first release says so in its notes; 0.2.0's explains the numbering and how a server
that says it is up to date takes it (horizon-server update --version v0.2.0).
- launcher.yml strips either prefix for HORIZON_LAUNCHER_VERSION and passes HORIZON_LAUNCHER_TAG,
which the launcher uses to know its own line.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
From 0.2 the game is tagged vX.Y.Z and the launcher [email protected] (docs/versioning.md). The launcher
compared every tag number by number, so it would never offer game v0.2.0 over v2026.10.19.15 nor
update itself to 0.2.0 from 0.16.0 (launcher-v0.2.0 is also taken by an old release).
- release_order(): tags from before 0.2 (vYYYY.MM.DD[.N], launcher-vX.Y.Z, bare versions) compare as
before; every 0.2 tag follows them, in semver's order (a pre-release before its release).
- own_tag(): the tag this launcher was released as. launcher.yml will pass HORIZON_LAUNCHER_TAG; a
release with only HORIZON_LAUNCHER_VERSION was a launcher-v one; a build from source is on 0.2.
So the last launcher-v release (the bridge old launchers can see) updates itself to launcher@.
- The matching-refs fallback looks for launcher@ tags first, then launcher-v.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
From 0.2 Horizon's game releases are tagged vMAJOR.MINOR.PATCH (docs/versioning.md). newer() compared
tags number by number, so v0.2.0 looked older than v2026.10.19.15 and an installed server would never
offer or install it. Every semver tag now outranks every date tag; within semver, patch and minor
compare as numbers and a pre-release (v0.3.0-rc.1) comes before its release. Pruning old releases
sorts with the same order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Verify passes. Local integration of #1101–#1107 on main caeffffb: parity --changed 302/304, and the two failures (block_particles, entity_model) fail identically on plain main. CI parity's secure_chat failure on #1101 logged 'transcripts identical' (flaky).
🤖 Agent: merged. Verify passes. Local integration of #1101–#1107 on main caeffffb: parity --changed 302/304, and the two failures (block_particles, entity_model) fail identically on plain main. CI parity's secure_chat failure on #1101 logged 'transcripts identical' (flaky).
`open` replies once the world starts opening on a worker, so the save
folder may not exist yet when the benchmark writes horizon-benchmark.txt
into it. On a slower device (an Android phone) the write failed with
"No such file or directory" and the run ended before the first scene.
With the camera above the build height (or below the bottom), vanilla's
SectionOcclusionGraph.initializeQueueForFullUpdate seeds the search with
the top (bottom) layer of sections around the camera instead of the
camera's own section. Horizon started from the camera's section and could
only step one section past the world's edge, so from y 352 up (two
sections above the top) it reached nothing and every chunk vanished.
Fixes#1037
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Verify passes; CI parity passes except dedicated_server, which fails the same way on main (play-pearl teleported 0 vs 1, the known flake in FEATURES.md LATER).
🤖 Agent: merged. Verify passes; CI parity passes except dedicated_server, which fails the same way on main (play-pearl teleported 0 vs 1, the known flake in FEATURES.md LATER).
🤖 Agent: merged. Verify passes; CI parity passes except dedicated_server, which fails the same way on main (play-pearl teleported 0 vs 1, the known flake in FEATURES.md LATER).
Moving blocks are minecraft:moving_piston blocks with a block entity.
The client drew nothing for them and gave them no collision, so pistons
on a clock flickered and players walked through them (#935).
The world now keeps a copy of the level's moving-piston block entities.
SceneCollisions::moving_piston_shape returns their shape
(PistonMovingBlockEntity.getCollisionShape, already in piston_shape),
and game/moving_pistons.rs ports PistonHeadRenderer: the moved block at
its interpolated offset, the moving head's short state, and a retracting
piston's head and extended base.
Entity pushing (moveCollidedEntities) is not ported yet.
Fixes#935
Co-Authored-By: Claude Opus 5.5 <[email protected]>
BoatRenderer draws its water_patch part with RenderTypes.waterMask (depth
only) unless the boat is under water, so translucent water behind it fails
the depth test inside the hull. Horizon loaded the part but never drew it:
the water surface showed through the boat. The patch now draws with a
depth-only Water mask pipeline after the entities and before translucent
terrain (rafts have none, as RaftRenderer).
Turning in a boat stepped the camera at the tick rate: positionRider adds
the boat's deltaRotation to the passenger's yRot after its yRotO was kept,
and LocalPlayer.getViewYRot interpolates a passenger's yaw between them.
The camera now spreads the turn over the tick's frames.
Fixes#1016
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The camera's biome alone gave the sky, fog and water fog colours, so they
jumped at every biome boundary (most visibly the Nether fog). Vanilla's
EnvironmentAttributeProbe weights the noise biomes of a 6x6x6 quart
neighbourhood around the camera with GaussianSampler's kernel and lerps
the spatially interpolated attributes between them
(SpatialAttributeInterpolator, ARGB.srgbLerp); unloaded chunks count as
plains.
Fixes#797
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A thrown splash or lingering potion or experience bottle was a stand-in
terrain particle spawned at its tick position: darkened by the terrain
particle colour, without the potion's colour, two overlapping copies and
no interpolation between ticks. Draw it as vanilla's ThrownItemRenderer
does: the item model (GROUND, with the potion tint) facing the camera,
interpolated, lit at its light probe, hidden for its first two ticks near
the camera (ThrowableProjectile.shouldRenderAtSqrDistance). Witch potions
draw as the splash potion of their effect instead of a bare bottle.
Fixes#818
Co-Authored-By: Claude Opus 5.5 <[email protected]>
EndCrystal.tick advances the crystal's time wherever it is; only the fire
at its block needs a dragon fight. Horizon advanced it only during the End
fight, so a crystal placed elsewhere kept a fixed time and jumped back
every tick (it flickered instead of turning).
Fixes#1051
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Breaking, mining, landing and other block particles took the block's up
face and a fixed grass or foliage colour, so grass blocks crumbled as
tinted grass tops (moss-like). Vanilla's TerrainParticle uses the model's
particle material (first variant / first multipart selector) and tint
source 0's colorAsTerrainParticle: the biome-blended in-world colour,
none for grass blocks (BlockTintSources.grassBlock) or blocks without a
source. The in-wall screen overlay and block markers use the particle
material too.
Fixes#1034
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The glide flag is the server's: LivingEntity.updateFallFlying clears it when
canGlide fails (the player is on the ground), from the position the client
sent, and the client gets the change on its next tick. So in vanilla a glide
that lands goes on for one tick on the ground, with travelFallFlying's drag
instead of ground friction, and a jump on that tick (with its sprint boost)
keeps the speed: the elytra hop. Horizon ended the glide on the landing tick,
so each hop lost about half of the speed to ground friction.
Fixes#1056
Co-Authored-By: Claude Opus 5.5 <[email protected]>
BeehiveDecorator stores 2 or 3 bees in the nest's block entity. World
generation kept them, but trees placed in the live level (a sapling grown next
to flowers, which picks the *_bees_005 trees, and /place feature) dropped them:
LevelFeatureWorld had no store_beehive_occupants, so the nest had no block
entity at all. Breaking it released nothing and bees could not enter it.
LevelAccess gets load_block_entity, the game loads the occupants into the
nest's block entity, and the occupant NBT is shared with world generation.
Fixes#1097
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Player.attack adds causeExtraKnockback after the hurt: the weapon's knockback
enchantment (halved) plus 0.5 for a sprinting full-strength hit, pushed along
the attacker's yaw (sin, -cos) rather than away from it, which lets a player
steer a hit target by turning while attacking (knockback displacement).
Horizon had only the enchantment part, on mobs; sprinting hits added nothing
and players (LAN PvP) got no extra knockback at all. Now:
- player hits on mobs and players: one extra knockback of that strength with
LivingEntity.knockback (knockback resistance, lift only from the ground),
the attack.knockback sound, and the attacker's sprint stops;
- the knockback a player takes (mob and PvP hits, dealDefaultKnockback with
0.4) reads its knockback resistance (netherite) and only lifts from the
ground: an airborne player kept being thrown up by every hit.
Fixes#504
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The player's ATTACK_DAMAGE and ATTACK_SPEED modifiers change in
LivingEntity.detectEquipmentUpdates, once a tick, while Player.attack reads
the weapon's enchantments and smash bonus from the stack in hand at the
attack. So swapping hotbar slots just before an attack hits with the old
item's damage and charge and the new item's enchantments (a Wind Burst or
Breach mace after another weapon): attribute swapping. Horizon read both from
the item in hand at the attack.
Also Player.tick's reset: switching to a different item now resets the attack
strength the damage uses (only the indicator's ticker reset before), for the
host and for LAN guests.
Fixes#505
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Two Blast Protection pieces now cancel explosion knockback (#1049):
- ServerExplosion.hurtEntities scales knockback by EXPLOSION_KNOCKBACK_RESISTANCE
for living entities. The player's knockback ignored it, and mobs used
KNOCKBACK_RESISTANCE (iron golems were never pushed; vanilla pushes them,
every type's explosion resistance base is 0).
- EnchantmentAttributeEffect.idForSlot: the modifier id ends with the slot,
so the same enchantment on two armor pieces adds up. The id was
<id>.<level>, so the second piece replaced the first.
- hitPlayers skips creative players who are flying: no knockback for them.
Fixes#1049
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Minecraft.handleKeybinds only calls startUseItem while the player is not
using an item: use clicks are consumed and held use doesn't repeat. Horizon
repeated held use every 4 ticks regardless, so with a shield raised in the off
hand, aiming at a block placed the main hand's block (or spawn egg, boat...).
Fixes#1050
Co-Authored-By: Claude Opus 5.5 <[email protected]>
LivingEntity.updateFallFlying starts with Entity.checkFallDistanceAccumulation:
while gliding and falling slower than 0.5 blocks a tick, the fall distance is
capped at 1. Horizon kept accumulating it for the whole glide, so landing after
gliding down (or after climbing and gliding back) dealt the full drop's damage.
Fixes#1046
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Verify and CI parity pass on the rebased head (b591c9dc), with before/after frame times in the body (rain 561 → 759 FPS, snow 663 → 981 on the Mac mini).
Low leaves can hide every face of the trunk's bottom log from where the
walk stops ("no face of block ... is visible from here"). The scenario
now breaks inside the tree loop and moves on to the next tree, as it
already did for one it cannot reach.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- survival and ui asserted that the Create New World click itself
answered `playing`. Since 7f4193f8 the in-frame world is prepared off
the UI thread behind the title screen, so the click answers `title`
and both scenarios failed in the frame (main's default) every time.
They now wait for `loaded`, as the other scenarios do.
- The survival spawn is random within the spawn radius, on the first
full top face below the MOTION_BLOCKING height
(PlayerSpawnFinder.getLevelRespawnPos), so a tree's leaves too: from
there no tree was reachable within goto's three-block drops. The
scenario now drops the player to the ground under the leaves first.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla's spawn chunks load by ticket level, so the level loading screen's
chunk map fills in Chebyshev rings. While the loading screen is up,
Horizon now picks chunks ring by ring (nearest first inside a ring)
instead of by Euclidean distance, which grew a disc.
Fixes#770
Co-Authored-By: Claude Opus 5.5 <[email protected]>
check_xp makes the host a spectator, out of the orbs' way, and left it
one for the rest of the run. A spectator sinks (Player.travel keeps 0.6
of its vertical speed each tick, about 0.2 blocks from standing), so the
next check that made it survival again, swings, landed it inside the
floor with the guest teleported beside it: the guest's feet were in the
floor's stone, which hid the dirt it tills from every angle ("no face of
block ... is visible from here"). It failed only after the earlier
checks, never alone.
check_xp now ends in survival where the host stood, as check_trials
already does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ItemEntity.tick merges a dropped stack with mergeable items whose boxes
meet its own grown by half a block sideways, every 2 ticks while it moves
between blocks and every 40 otherwise (mergeWithNeighbours, tryToMerge,
merge). Horizon never merged, so fresh chunks, where water and falling
blocks break leaf litter, moss and plants, filled up with single items.
FallingBlockEntity.tick also places (or breaks) the block within the
entity's own tick. Horizon moved every falling block first and placed them
after, so a block falling right behind another landed in the same cell and
broke into an item: stacked sand and gravel in fresh chunks dropped.
Fixes#996
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The oracle prints every ProtoChunk postprocessing position and chunk_dump
prints Horizon's, so a chunk that schedules different fluid or shape
updates after generation (the blocks that later tick and drop items in
fresh chunks, #996) fails the harness.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Verify passes. The change runs only on the server thread (opt-in today), and the PR body's proof shows lan_e2e --server-thread going from 35/39 to 38/39. #1072's switch needs it.
🤖 Agent: merged after integration C: the desktop ran 322/325 harnesses on main + #1088–#1093, and its 3 failures were the known dedicated_server flake plus the two display harnesses; world_reopen and save_roundtrip pass on the Mac mini (main + the six branches).
🤖 Agent: merged after integration C: the desktop ran 322/325 harnesses on main + #1088–#1093, and its 3 failures were the known dedicated_server flake plus the two display harnesses; world_reopen and save_roundtrip pass on the Mac mini (main + the six branches).
🤖 Agent: merged after integration C: the desktop ran 322/325 harnesses on main + #1088–#1093, and its 3 failures were the known dedicated_server flake plus the two display harnesses; world_reopen and save_roundtrip pass on the Mac mini (main + the six branches).
🤖 Agent: merged after integration C: the desktop ran 322/325 harnesses on main + #1088–#1093, and its 3 failures were the known dedicated_server flake plus the two display harnesses; world_reopen and save_roundtrip pass on the Mac mini (main + the six branches).
🤖 Agent: merged after integration C: the desktop ran 322/325 harnesses on main + #1088–#1093, and its 3 failures were the known dedicated_server flake plus the two display harnesses; world_reopen and save_roundtrip pass on the Mac mini (main + the six branches).
🤖 Agent: merged after integration C: the desktop ran 322/325 harnesses on main + #1088–#1093, and its 3 failures were the known dedicated_server flake plus the two display harnesses; world_reopen and save_roundtrip pass on the Mac mini (main + the six branches).
SectionOcclusionGraph.runUpdates, both rules Horizon left out:
- a section leads on through whatever any face it was entered through
connects to (Node.sourceDirections), not only the first one;
- beyond 3 sections from the camera, a step is taken only if the line
from the section's corner towards the camera, in 28-block steps while
farther than 60 blocks, stays in sections the search has reached.
The search takes the camera position for the ray test. Exact equivalent
of vanilla's, faster: one ray per axis per node (both steps along an
axis share the corner; while the camera stays in the searched-from
section no point falls in a neighbour the loop adds), an answer per
section corner during a full search (neighbours share corners; reached
sections are only added, so a line that stayed in them still does, and
one that failed is checked again from the sample it failed at, stepping
over the earlier ones with the same additions), and an integer floor
instead of libm's on x86-64. Tests compare every cell with a literal
port of runUpdates (map of nodes, FIFO, JOML arithmetic), and a resumed
line with one checked from its corner.
Fixes#782
Co-Authored-By: Claude Opus 5.5 <[email protected]>
WeatherEffectRenderer reads level.getHeight(MOTION_BLOCKING, x, z) for
every column within the weather radius. Horizon scanned each column block
by block from 10 blocks above the radius down to the ground, with a
string search per block, every frame: about half a millisecond of CPU per
frame in rain on a Ryzen, more on laptops (#899). Client chunks now keep
a MOTION_BLOCKING map next to their WORLD_SURFACE one, built on the first
read and dropped by writes that can move it, from the facts world
generation already uses (blocks_motion_in_heightmap or a fluid).
This is also vanilla's rule where the scan differed: slabs, stairs and
other motion-blocking blocks without a full collision box now stop rain,
and a roof more than 20 blocks above the camera does too.
Tiling a column at each texture repeat no longer allocates two vectors
per column.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
owner_console_command ran the window's commands as the world's owner with
the hub taken out, so a selector saw only the owner (`team join … @a` left
the guests out, and friendly fire still hurt) and `defaultgamemode` never
reached the hub (guests joined in the world's old mode). It now puts the
hub back while the command runs, as run_guest_commands does for a guest.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Title backdrop built on a worker (fixes#896). Integration proof (main + #1079#1081#1082#1083#1085#1086#1087), 2026-10-21: desktop parity 103/105; the 2 others (world_reopen, save_roundtrip) need a display and pass on the Mac mini. Verify passed.
🤖 Agent: merged. Creative pick block in inventories (fixes#1048, Visivel). Integration proof (main + #1079#1081#1082#1083#1085#1086#1087), 2026-10-21: desktop parity 103/105; the 2 others (world_reopen, save_roundtrip) need a display and pass on the Mac mini. Verify passed.
🤖 Agent: merged. Firework tooltips (fixes#1047, Visivel). Integration proof (main + #1079#1081#1082#1083#1085#1086#1087), 2026-10-21: desktop parity 103/105; the 2 others (world_reopen, save_roundtrip) need a display and pass on the Mac mini. Verify passed.
🤖 Agent: merged. Cobwebs slow falling blocks (fixes#933). Integration proof (main + #1079#1081#1082#1083#1085#1086#1087), 2026-10-21: desktop parity 103/105; the 2 others (world_reopen, save_roundtrip) need a display and pass on the Mac mini. Verify passed.
🤖 Agent: merged. Stars built once (fixes#1054, Visivel and rdbtCVS). Integration proof (main + #1079#1081#1082#1083#1085#1086#1087), 2026-10-21: desktop parity 103/105; the 2 others (world_reopen, save_roundtrip) need a display and pass on the Mac mini. Verify passed.
🤖 Agent: merged. Blocks can't be placed into entities (fixes#1052, Visivel). Integration proof (main + #1079#1081#1082#1083#1085#1086#1087), 2026-10-21: desktop parity 103/105; the 2 others (world_reopen, save_roundtrip) need a display and pass on the Mac mini. Verify passed.
🤖 Agent: merged. Shears carve pumpkins (fixes#1067, reported by Tydragon). Integration proof (main + #1079#1081#1082#1083#1085#1086#1087), 2026-10-21: desktop parity 103/105; the 2 others (world_reopen, save_roundtrip) need a display and pass on the Mac mini. Verify passed.
- sprites: vanilla models name the missing texture on purpose (block/air.json's
particle is minecraft:missingno), so using it is not worth a warning (#853).
- audio: when the resources hold no sounds.json, the game plays nothing on a
working output and the log said only which device opened. It now says the
sounds were not copied and how to fix it (#1035, #1076).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A player installed Java 25 but JAVA_HOME still named an older JDK, so
Prepare stopped with "Java 25 is needed", and the launcher showed that
error at the foot of the page, out of sight until the window was enlarged.
- horizon-prepare java::find: after JAVA_HOME, the PATH and macOS's
java_home, look in the folders JDK installers use (Program Files vendors
on Windows, /usr/lib/jvm on Linux, JavaVirtualMachines on macOS, ~/.jdks,
SDKMAN), JDKs named for the pinned major first.
- launcher: the preparation's progress and error sit right under the
buttons instead of below the rows.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The client JAR ships an empty font/include/unifont.json; the real provider
list and font/unifont.zip, unifont_jp.zip and unifont_pua.zip (1.9 MB) come
only from the asset index, which preparation never copied. Every character
the bitmap fonts lack showed as a missing-glyph box (a server's "▸").
Full preparation now imports them after extracting the JAR, like the title
panorama, and `horizon-prepare fonts` repairs prepared resources.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Mojang republished the 26.3 version JSON again (sha1 61fff131), with a new
asset index 34 (0c8f43b2) that changes only 142 translation files. A launcher
keeps one indexes/34.json, so every install made since then, official
launcher or Prism, failed the "pinned asset index" check: no sounds and no
title panorama.
- research-lock.json: accept the new version JSON, and list the asset
indexes of every accepted version JSON (accepted_asset_index_sha1).
- horizon-prepare: an index the lock does not know is still used, with a
line in the log, since every object copied from it is checked against its
own SHA-1; the sounds now require minecraft/sounds.json in the index.
- tools/lab.py: the same for the dev tools.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Measured on an Intel HD 620 (#897): rain and snow cost about 3 ms of GPU
per frame there, most of it from the weather radius; FXAA about 0.8 ms;
cave culling saves about 0.6 ms; mipmaps cost nothing measurable and
clouds about 0.1 ms. A new Cost::Saves tag ("Saves time", in the success
colour) marks a setting that makes frames faster when on.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The integrated and dedicated server loops paced every tick at the tick rate,
so /tick sprint ran at 20 tps and never reported, and the integrated server
recorded no tick times for /tick query. They now share
MinecraftServer.runServer's pacing: no wait while sprinting, the sprint's
report sent to the operators as the server's, and tick times recorded.
The hub broadcasts ClientboundTickingStepPacket when a step starts or stops
(ServerTickRateManager.updateStepTicks) and sends a joining player the
steps left, not 0.
The camera, hand, view bobbing and the player's own model use
Camera.getCameraEntityPartialTicks (getGameTimeDeltaPartialTick(true)), so
they stay smooth while the level is frozen instead of moving at 20 Hz.
Fixes#1038
Co-Authored-By: Claude Opus 5.5 <[email protected]>
start_backdrop built VanillaTerrain on the event loop, so the title stopped
drawing for the whole build: 2.1 s cold and 0.86 s warm on an M4 Pro (fast
profile), far longer on slow CPUs. The worker now builds it and finds the
spot; the title draws its panorama until the generator is installed.
Fixes#896
Co-Authored-By: Claude Opus 5.5 <[email protected]>
BlockItem.canPlace now checks every entity whose blocksBuilding is set: mobs,
armor stands (not markers), boats, minecarts, end crystals, primed TNT and
falling blocks, not only players.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Linux dedicated servers lower their world-generation and packet-writer threads behind the server thread; nothing changes elsewhere. verify passed; parity: every harness passes except dedicated_server, which is flaky on main with the same failure modes (7-run A/B against main in the last comment).
The middle button (keyPickItem) over an inventory or the creative screen no
longer clicks like the left button: with infinite materials it clones the
hovered stack at full size (CLONE) or, holding a stack, drags full stacks into
slots (quick craft type 2); otherwise it does nothing. Agents can middle-click
widgets (click ID middle), and creative slots are ui widgets now.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Rockets list their flight duration and each run of equal explosions (shape,
colours, fade colours, trail, twinkle); firework stars list their explosion.
Both honour TooltipDisplay. New harness: tools/test_firework_tooltip_parity.py.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A falling block's move now runs vanilla's checkInsideBlocks along its path: a
cobweb it passes or lands in makes it stuck (0.25, 0.05, 0.25), which scales
its next move and zeroes its velocity, and resets its fall distance.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Right-clicking a pumpkin with shears now carves it like vanilla: a carved
pumpkin facing the clicked side (the side facing the player on the top and
bottom), four pumpkin seeds from carve/pumpkin popped out of that side, the
carve sound, one point of wear, the shear game event and the used
statistic. A pumpkin carved on a golem base builds the golem, as
CarvedPumpkinBlock.onPlace does; the dispenser shares that placement.
use_item_on parity: pumpkin cases on any face with the player turned any
way (the oracle and fixture take an optional yaw and face).
Fixes#1067
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Verify passed. Integration proof (main + #1041#1043#1021#1020#1032#659#1029#1066), 2026-10-21: 97 of 100 harnesses passed on the desktop. world_reopen and save_roundtrip need a display there; they pass on the Mac mini with #1041/#1043. dedicated_server is a pre-existing flake: it fails on plain main too, and is pass/fail in 2 runs on the integration.
🤖 Agent: merged. Verify passed. Integration proof (main + #1041#1043#1021#1020#1032#659#1029#1066), 2026-10-21: 97 of 100 harnesses passed on the desktop. world_reopen and save_roundtrip need a display there; they pass on the Mac mini with #1041/#1043. dedicated_server is a pre-existing flake: it fails on plain main too, and is pass/fail in 2 runs on the integration.
🤖 Agent: merged. Verify passed. Integration proof (main + #1041#1043#1021#1020#1032#659#1029#1066), 2026-10-21: 97 of 100 harnesses passed on the desktop. world_reopen and save_roundtrip need a display there; they pass on the Mac mini with #1041/#1043. dedicated_server is a pre-existing flake: it fails on plain main too, and is pass/fail in 2 runs on the integration.
🤖 Agent: merged, thanks @rdbtCVS! World saves are spread over frames and keep the session lock until every write is durable (with my small clippy fix). Integration proof (main + #1041#1043#1021#1020#1032#659#1029#1066), 2026-10-21: 97 of 100 harnesses passed on the desktop. world_reopen and save_roundtrip need a display there; they pass on the Mac mini with #1041/#1043. dedicated_server is a pre-existing flake: it fails on plain main too, and is pass/fail in 2 runs on the integration.
🤖 Agent: merged, thanks @rdbtCVS! Local worlds are prepared on a worker; the loading screen keeps drawing. Integration proof (main + #1041#1043#1021#1020#1032#659#1029#1066), 2026-10-21: 97 of 100 harnesses passed on the desktop. world_reopen and save_roundtrip need a display there; they pass on the Mac mini with #1041/#1043. dedicated_server is a pre-existing flake: it fails on plain main too, and is pass/fail in 2 runs on the integration.
🤖 Agent: merged, thanks @ItsPotatoTime! The action bar, title, subtitle and boss bar names now resolve legacy § codes, like the sidebar since #1003 and like vanilla's StringDecomposer.
🤖 Agent: merged. Every merge condition holds: the latest stable game release (v2026.10.19.15) and launcher release (launcher-v0.16.0) verify against the release key, legacy_releases.py --check passes, and launcher and horizon-server built from this branch install signed, pinned-legacy and beta releases and refuse an edited SHA256SUMS, an altered archive and a missing signature. verify passed; parity proof in the last comment.
Lithium, FerriteCore, ImmediatelyFast, Gnetum, ScalableLux, Structure
Layout Optimizer, AsyncParticles and Ixeris: per technique, the vanilla
cost it targets, whether Horizon already avoids it, and whether it is
worth measuring. From public descriptions only; no mod code read.
Fixes#991
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged inert. Without the MACOS_SIGN_IDENTITY variable (not set: Ludovic's Developer ID and notary setup don't exist yet), every new signing/notarizing step is skipped and the release, launcher, sign and promote pipelines are unchanged; verify and parity passed on this head, a launcher dry run passed with signing skipped, and today's ad hoc launcher bundle passes the new self-update signature check. Setting it up is in FEATURES.md LATER (4b).
Servers send § codes inside a component's text. Vanilla's font applies them to every literal, but the action bar, title, subtitle and boss bar names drew the component raw, so the codes showed as glyphs. Resolve them before measuring and drawing, as the sidebar already does.
Co-Authored-By: cairn-code (DeepSeek V4.1 Flash) <[email protected]>
🤖 Agent: merged, thanks @ItsPotatoTime! Prefers an 8-bit non-sRGB surface (Bgra8Unorm/Rgba8Unorm) instead of RADV's fp16 ExtendedSrgb one, so colours are no longer washed out on Wayland. Mac and Windows already got Bgra8Unorm. Verify passed; parity doesn't cover surface formats.
SkyRenderer.buildStars builds the 1500 star quads once into a static
buffer; Horizon regenerated them on the CPU and uploaded about 200 KB of
vertices every frame while stars were out, which made nights cost more
than days (#1054). The stars now live in a static vertex buffer built
when the renderer starts, drawn after the sun and moon with one small
uniform: the view projection times the sky's rotation, and the star
brightness.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Rgba16Float is the first non-sRGB format RADV advertises on Wayland, and wgpu gives it the linear ExtendedSrgb color space. The renderer writes gamma-encoded values, so the compositor read them as linear light and washed every colour out.
Prefer Bgra8Unorm or Rgba8Unorm and keep the old choice as the fallback. A screenshot on a surface that is not 32-bit now reports an error instead of copying with an 8-bit row stride, which aborted wgpu validation.
Co-Authored-By: cairn-code (DeepSeek V4.1 Flash) <[email protected]>
🤖 Agent: merged. Docs only (README and website/docs): the Servers page is in launcher 0.15+, and releases go from beta to stable. Verify passed; parity doesn't cover docs. This also redeploys playhorizon.dev, which still showed the 10-19 14:29 build.
horizon_core::VANILLA_VERSION is a re-export, which left the datafix harness's
PARITY_RUST_SCOPE open: its six shards reran on any Rust change.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
lab.VERSION, lab.GAME (.local/minecraft/<version>) and lab.RESOURCES
(.local/horizon-client/<version>) come from the lock, and lab.version_info()
reads the pinned client jar's version.json. Every harness, generator and
tool that named 26.3 (resource folders, jar and source paths, result
records, the vanilla client's --version and asset index, the bare login
client's protocol, the benchmark's options.txt data version, the packet id
table's header) uses them, so a candidate worktree pinned to another version
runs them unchanged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
tools/gen_version_consts.py writes what the pinned client jar's version.json
states (name, id, protocol, data version, pack formats, Java, stable, build
time) into horizon_core::version, and every Rust site reads it from there:
PROTOCOL_VERSION, GAME_VERSION, DATA_VERSION, the save's and the datafixer's
version name, /version, the crash report, the legacy ping, the dedicated
server's start line, the title screens, the default resource folder, the
resource pack format, and the launcher's and horizon-ctl's fallbacks. CI
checks the file against the pinned jar.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- crates/horizon-release-sign/legacy-releases.txt pins the SHA-256 of every archive of the 131 releases published
before signing (GitHub's upload digests, written by tools/ci/legacy_releases.py). The launcher and
horizon-server update install those releases only with exactly those files, so rolling back to one works and
nobody can swap its files later; every other release needs its signed SHA256SUMS.
- the_launcher_ships_with_a_release_key and horizon_server_ships_with_a_release_key fail while
release-keys.txt lists no key, so this can't ship refusing everything.
- Test tags moved to v2099.* so they never collide with a pinned release.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The launcher (game installs and its own updates) and `horizon-server update` fetch a release's
SHA256SUMS and SHA256SUMS.sig, check the signature against the keys built in from
crates/horizon-release-sign/release-keys.txt and the archive against the signed SHA-256, and
refuse anything unsigned, signed for another tag or by another key, or that doesn't match, with a
message saying why. Merge only once release-keys.txt lists the key and a signed stable release
exists: until then every download is refused.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
With the repository variable MACOS_SIGN_IDENTITY set, the launcher's Horizon.app and .dmg and the
game's three programs are signed with the hardened runtime and a timestamp, notarized and (app,
dmg) stapled, from a keychain of the Mac mini's own that keychain.sh unlocks with its password from
1Password. Unset, everything is signed ad hoc as before. The launcher checks an updated bundle's
signature before and after swapping it in, and puts the old one back when it doesn't hold.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged on verify. Test-only (a Python test fixture and a temp-folder name inside a #[test] fn), so no harness can see it; it turns main's verify green again after #235 and removes a horizon-server test flake.
a_stopped_server_saves_and_keeps_comments (settings.rs) and
writes_properties_and_syncs_back (config.rs) both used
horizon-server-config-<pid>, so when they ran at the same time one
removed or rewrote the other's horizon.toml and
writes_properties_and_syncs_back failed now and then. The settings test
now has its own folder.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
#235 made Game::capture read the surface size, warp the pointer to the
window's centre and set the HUD cursor, so the fixture's copy of the method
no longer compiled and verify failed on main. The fixture now stubs
winit's PhysicalPosition/PhysicalSize, Window::inner_size and
set_cursor_position, HudBatch and Game's renderer and cursor, and checks
the recentring: one warp on grab, two on release, the HUD cursor at the
centre, and nothing when releasing an already released pointer.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged. Recentres the cursor like vanilla's MouseHandler.grabMouse/releaseMouse. CI verify and parity passed; Ludovic asked on 10-21 to clear the PR queue autonomously. If the cursor misbehaves on a Wayland compositor, file it and we'll follow up.
🤖 Agent: merged. Reviewed the storage host: keys are one [a-z0-9_.-] segment, ids are [a-z0-9_], and the quota, key cap and per-call write budget all hold. Thanks @Yvonne-Aizawa!
🤖 Agent: merged. pr_ready now checks the touched crates and their dependents instead of the whole workspace; CI keeps its workspace-wide checks. The earlier verify failure was an unrelated supervisor test. Thanks @rdbtCVS!
🤖 Agent: merged after verify and CI parity passed (modpacks-11: Chipped's 11 vines carried as vines; the 11 blockstate errors are gone from the game log).
- Reject a second running mod with an id already in use: two mods
sharing an id would also share capability storage's data_dir,
letting one read, overwrite or delete the other's state.
- Cap storage to 64 keys per mod, not just 4 MiB: zero- or near-zero-
byte values could otherwise flood the directory with files the
byte quota never sees.
- Add a 1 MiB per-call write budget, separate from the 4 MiB on-disk
quota: a loop replacing one key stays under the quota on every
check (the total on disk never grows) while still doing unbounded
synchronous disk I/O per call, which fuel never meters.
Co-Authored-By: Claude Sonnet 5 <[email protected]>
A mod with capability storage gets get(key)/set(key, value)/delete(key),
backed by real files beside horizon-mods/ (.horizon-mod-data/<id>/, 4 MiB
quota per mod). Keys are a single path segment so they can't leave the
mod's own directory. Unlike in-memory mod state, this survives a
cargo-horizon-mod dev reload, since it's keyed by the mod's path and id
rather than its Store.
🤖 Agent: merged after verify passed and a local check on main + #988: chunk_generation, command_exec and mob_ai (the three harnesses that failed in CI on its old head, under load) all pass; modpacks-10's integration ran 301/301. Formations' structures generate from the noise column.
WaterlogWhenReplacingWaterProcessor (16 lists: the bottom pillars of its
villages and outposts) waterlogs a block that can be where the world holds
water, makes it dry where it does not, and gives a block placed in water a
block tick. Carry used to drop it; it is now the game's
horizon:waterlog_in_world_water, in the chunk being generated as the mod's.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify passed, security-1's dry runs on the branch, and an independent review whose findings were all fixed. Every release goes out as a beta and is promoted to stable after 6 hours if its smoke test passed and no release-blocker issue is open; launcher-v0.14.0 (stable by default) is already out.
🤖 Agent: merged after verify passed, security-1's dry runs on the branch, and an independent review whose findings were all fixed. CODEOWNERS for the release pipeline and update code. The protection itself (tools/ci/protect_pipeline.sh --apply) waits for the switch to the GitHub App.
🤖 Agent: merged after verify passed, security-1's dry runs on the branch, and an independent review whose findings were all fixed. docs/security.md: the security model in plain words, with what is live and what still waits.
🤖 Agent: merged after verify passed, security-1's dry runs on the branch, and an independent review whose findings were all fixed. Release builds on the Mac mini start from a clean checkout.
🤖 Agent: merged after verify passed (docs and website only; the site build fails on broken links, anchors or images). The README is a shorter front page that links to the new docs.
🤖 Agent: merged after verify passed (docs and website only; the site build fails on broken links, anchors or images). The website gets a Docs section with 11 pages; /server redirects to /docs/hosting.
One-paragraph pitch, download and install, what works today with links to the
site's documentation, hosting and modpacks in two lines each, then build from
source, contributing and the project status table (moved lower, unchanged in
form so the site and tests/test_parity.py still read it). The stale release
notes section and launcher version are gone: the release notes are linked.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
/docs has a left navigation and eleven pages written for players first, then
hosters, then contributors: what Horizon is, install, settings, multiplayer,
logs and bug reports, hosting (the launcher's Servers page and the command
line), modpacks, mods and data packs, performance, contributing and the FAQ.
/server redirects to /docs/hosting. The build checks every link between the
pages and every image.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify passed (74 + 9 + 1 launcher tests; launcher-only). The Servers page: create, start/stop, players, settings, backups, updates, run at login, and Friends over the internet with Tailscale (device sharing, an optional rule limiting shared friends to the game port, never Funnel). Ludovic's real-tailnet check is described in the PR.
🤖 Agent: merged after verify passed (launcher-only; no parity harness depends on it). The launcher's server backend and `horizon-launcher servers …` CLI.
🤖 Agent: merged after verify and CI parity passed (40 server tests; the whole create → start → join → whitelist → backup → restore → update → stop flow ran on the Mac mini and the ThinkPad). Adds `horizon-server config`, `backups --json`, `update --check --json`; fixes whitelisting on offline servers and updates as root in containers.
repurposed_structures:structure_void_processor returns null for a
template's structure_void blocks, which is what vanilla's block_ignore
does with blocks: [structure_void]. Carry writes it that way, so the
82 lists that use it (its mineshafts, ocean village and temples, the
jungle pyramid's hidden room, swamp village) no longer place structure
voids over the terrain.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
CloseOffAirSourcesProcessor walls off the air and the other liquids beside
a template's fluids, so its ocean mineshafts keep their water in. The game
runs it as horizon:close_off_air_sources (the same weighted blocks and the
mod's random, neighbours written in their sections as the mod does), and
carry maps the mod's 4 uses to it. The weighted block list is shared with
the fluid fix.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A mod's file for a vanilla, convention or other mod's tag adds its values in
load order; one with "replace": true now drops vanilla's entries and the
earlier mods' first, as a higher data pack does, and the log says which mod
replaced what (kept with the cached root, so a reused root warns too).
Carry keeps such tags instead of setting them aside: Towns and Towers'
seven minecraft:has_structure biome tags load. Mods still own no other
minecraft content.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
26.3's EnterBlockTrigger reads a block list from 'blocks' and skips
1.21.1's 'block', so a carried criterion matched any block the player
entered: Regions Unexplored's Downer, From the Tree Tops and Light as a
Frog were granted on the first dive into the sea. Carry renames the
field (20 advancements in Regions Unexplored, with its boat recipes'
in-water criteria).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Without the oracle, a block's template is guessed from its blockstate's
property names, and a vine's (east, north, south, up, west) fit
cobblestone_wall. The game then refused Chipped's 11 vine blockstates:
a wall's sides are none, low or tall, so 'true' is an unknown multipart
property value. block_template looks at the sides' values and makes those
vines; Luminax's 32 walls stay walls.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify passed and a local check on main + #983 + #984 (clippy clean; chunk_generation and locator_bar, the two harnesses that failed in CI on the old head, both pass; modpacks-10's integration ran 301/301). Structures built under the sea are flooded and waterlogged; Better Ocean Monuments' processors carried.
🤖 Agent: merged after verify passed and a local check on main + #983 + #984 (clippy clean; chunk_generation and locator_bar, the two harnesses that failed in CI on the old head, both pass; modpacks-10's integration ran 301/301). Structures built under the sea are flooded and waterlogged; Better Ocean Monuments' processors carried.
The relative-dimension check teleported the player from the Nether's
80 100 -40 to the overworld's 644 100 -316, fresh ocean chunks at seed 1.
On some runs vanilla broke some of their kelp and seagrass while they
loaded (its chunk loading is asynchronous), which rolled those blocks'
loot tables, so /random reset * later counted 12 or 13 sequences
against Horizon's 11. 1 230 0 in the Nether lands at 12 230 4, terrain
loaded since the start: the scaling is still checked, and 10 runs in a
row pass (5 of the 16 runs before failed this way).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- A table of what is live, waiting to merge, or waiting for Ludovic, so the page doesn't read as if it all were.
- Forks: their checks need a maintainer's approval, and run-parity runs them on the Mac mini; the Mac mini
section no longer says forks don't run there, and says what a clean release checkout does and doesn't fix.
- Blockers hold every beta while open; releases from before signing stay installable with pinned files.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
What an attacker could do before (instructions hidden in reports, a skipped check that counted as
passed, releases straight to every player, downloads trusting GitHub alone, pipeline files changed
like any other), what stops it now (untrusted text, verify always running, beta then stable,
signed updates, code owners, the agents' own GitHub App), what still needs Ludovic, and what it
doesn't solve. Written for the website's Contributing page; CONTRIBUTING.md links it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Its work folder is shared with pull requests' parity jobs; with clean: false, whatever they left (target/, an
untracked .cargo/config.toml, a build script's output) could end up in a release. release.yml and launcher.yml
now always check out clean; target/ is rebuilt from scratch, and vanilla-data.sh relinks .local/ from its
preparations outside the checkout.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- backend.rs downloads, installs and self-updates; links.json says where from; launcher/Cargo.toml and
crates/horizon-server/Cargo.toml pick the crates that check downloads.
- protect_pipeline.sh: ludovic111 is the only owner and can't approve his own pull requests, so he is the one
bypass_pull_request_allowances user. enforce_admins stays on, so verify is still required for everyone.
--apply reads the protection back and fails if GitHub didn't keep it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
.github/, tools/ci/, the release tools, the signing crate and the code that checks updates are
Ludovic's to review. tools/ci/protect_pipeline.sh prints (and with --apply sets) main's protection
with code owners' reviews required and no other approvals, keeping the required verify check;
it is for the orchestrator to apply once agents work as the GitHub App.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify passed. The secret was set first (RELEASE_SIGNING_KEY in the release environment), so the next release is signed with this key.
- A release-blocker issue's text is never read: its author can edit it after a maintainer labels it, so an
open blocker holds every beta, and one closed as completed still holds the betas published before it closed.
- Even --force never promotes a beta that isn't newer than the stable release: horizon-server update and
install.sh download GitHub's latest.
- When the oldest beta newer than the stable release has waited 24 hours, promote.yml opens a "Promotion to
stable has stalled" issue with the last reason; the next promotion closes it.
- promote.yml: persist-credentials: false; issues: write for the stall issue.
- Website: the newest 100 releases, plus GitHub's latest (the stable one) when a day of betas pushes it out.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Game releases go out as betas (GitHub pre-releases, never "latest") with the Mac mini's smoke
report attached. promote.yml (every 30 minutes, or by hand from a phone) runs
tools/release_channels.py, which makes the newest beta stable and latest once it has been out 6
hours, passed its smoke test and no release-blocker issue covers it, with notes rewritten to cover
everything since the last stable release. release_train.py counts merges from the newest release of
either channel, and a beta's notes say it is one.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The private key was generated by tools/ci/new_release_key.sh straight into
the release environment's RELEASE_SIGNING_KEY secret (never on disk or
printed); from now on every release's SHA256SUMS is signed with it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify passed, with security-1's dry runs on the branch and an independent review (no regression: parity still runs on every PR from this repo; releases go out unsigned with a warning until the signing key is set).
🤖 Agent: merged after verify passed, with security-1's dry runs on the branch and an independent review (no regression: parity still runs on every PR from this repo; releases go out unsigned with a warning until the signing key is set).
🤖 Agent: merged after verify passed, with security-1's dry runs on the branch and an independent review (no regression: parity still runs on every PR from this repo; releases go out unsigned with a warning until the signing key is set).
As root in a user namespace (an unprivileged LXC or rootless container), GNU tar
keeps the archive's owners by default and fails on IDs it can't map, so the update
stopped at "tar couldn't unpack". Found driving the launcher's Update now from one.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify and CI parity passed, plus modpacks-10's local integration (parity 301/301, structure-start oracle 150/150 and 134/134). Towns and Towers' namespaces carried in its manifest.
🤖 Agent: merged after verify and CI parity passed, plus modpacks-10's local integration (all nine branches on main: parity 301/301, structure-start oracle 150/150 and 134/134). Content mods may own namespaces besides their id; two mods may not share one.
🤖 Agent: merged after verify and CI parity passed, plus modpacks-10's local integration (all nine branches on main: parity 301/301, structure-start oracle 150/150 and 134/134). Mod loot tables from every namespace, so modded structure chests are filled.
The render thread summarized every observed chunk (256 columns, seven
light reads per block) and gathered each mesh job's inputs (neighbour
edges, visited columns, cloned spans) itself, within a budget of about
1 ms and 2 ms a frame but at least one chunk or tile: a large tile or a
chunk on a slow machine still cost tens of milliseconds in one frame.
Both now run on the far-terrain worker. Lod::observe hands the chunk
over (the latest per column, queued once), and the worker returns its
summary with its reductions (Done::Observed): an unchanged one is
dropped, a new one is kept and applied as a reduction was. A mesh job
takes a snapshot of the active tiles, their samples and the visited
summaries (Arc maps, cheap to clone) and gathers every tile's inputs
on the worker, as it did before inputs moved to the render thread.
At most OBSERVE_AHEAD resident chunks wait at once, since an edit
copies a chunk the worker still holds.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
1.21.1 read a {min, max} without a type as a uniform number provider; 26.3
dispatches on type alone. The game read such rolls and counts as 0, so
chests of 304 carried ATM10 tables (When Dungeons Arise 140, Repurposed
Structures 50, Structory and its towers 78, ...) came out empty or short.
Carry now types them in pools (rolls, bonus_rolls) and in the functions
that take one, nested bounds too; ranges such as limit_count's and item
predicates' counts stay ranges.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify passed and CI parity passed on the stack head (#995; local 98/98). perf-frame-9's terrain LOD follow-ups: band and black cliffs fixed, far quads −36 %, per-frame budgets, LOD cap on non-Apple integrated GPUs. LODs off matches main (screenshots, FPS).
🤖 Agent: merged after verify passed and CI parity passed on the stack head (#995; local 98/98). perf-frame-9's terrain LOD follow-ups: band and black cliffs fixed, far quads −36 %, per-frame budgets, LOD cap on non-Apple integrated GPUs. LODs off matches main (screenshots, FPS).
ring writes Ed25519 keys as PKCS#8 v2 (with the public key), which Ubuntu's OpenSSL 3.0 refuses
("unsupported"); verify caught it on ubuntu-latest. keygen now writes v1, the 32-byte seed alone,
checked to give the same public key as ring's document.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify passed and CI parity passed on the stack head (#995; local 98/98). perf-frame-9's terrain LOD follow-ups: band and black cliffs fixed, far quads −36 %, per-frame budgets, LOD cap on non-Apple integrated GPUs. LODs off matches main (screenshots, FPS).
🤖 Agent: merged after verify passed and CI parity passed on the stack head (#995; local 98/98). perf-frame-9's terrain LOD follow-ups: band and black cliffs fixed, far quads −36 %, per-frame budgets, LOD cap on non-Apple integrated GPUs. LODs off matches main (screenshots, FPS).
The game resolves names as vanilla does, Mojang first, so on an offline server a name someone owns got that
account's UUID and the player, who joins under the offline UUID, was refused. horizon-server now writes the
offline entry and has the game reload the whitelist.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Any repository code running earlier in the job that holds the key could reach it (a build patched
through Cargo.toml, or a step writing $GITHUB_ENV or $GITHUB_PATH for the next one). The `sign`
job checks out only tools/ci/sign_release.sh, which signs with the runner's OpenSSL; the publish
job, without the key, checks the result with horizon-release-sign as launchers will.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Game releases now go out as betas and become stable 6 hours later (promote.yml). The launcher
installs new versions from the stable channel unless "Beta builds" is on in Settings; turning it
on checks again at once. The Versions page, `versions` and the MCP server show each build's
channel. The website's patch notes list stable releases only.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
crates/horizon-release-sign signs a release's checksums with an Ed25519 key that exists only as a
secret of the `release` environment, and checks them as launchers will, against the public keys
built in from release-keys.txt. Dry runs sign with a throwaway key. Until release-keys.txt lists a
key, releases keep going out with SHA256SUMS alone and a warning; once it does, a missing or
unknown key fails the release before anything is published. tools/ci/new_release_key.sh pipes a
new key straight into the secret and prints only the public key. ring and base64 are already in
the lockfile.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Labels started `checks` runs whose verify job was skipped, and GitHub counts a skipped required
check as passed: auto-merge merged #823 and #837 before the real verify finished. Labels no longer
start `checks` (the fork-only run-parity label has its own workflow), and verify has no condition.
The parity job moves into a reusable workflow both call.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify passed and local parity (perf-server-12: `tools/parity.py --changed origin/main` 98/98 with the three branches together, sculk / natural_spawn / spawn_rules harnesses pass; before/after numbers in the PR). CI parity was still queued on a busy runner.
🤖 Agent: merged after verify passed and local parity (perf-server-12: `tools/parity.py --changed origin/main` 98/98 with the three branches together, sculk / natural_spawn / spawn_rules harnesses pass; before/after numbers in the PR). CI parity was still queued on a busy runner.
horizon_launcher::servers keeps the list of servers and drives horizon-server
for them: create (init, then config set), start, stop, backups, settings, the
service, and the running supervisor's control API for the live status, the log
and console commands. The programs come from an installed, prepared version,
laid out as install.sh does (server-runtime/releases, current, game-files), so
horizon-server update and its rollback work on them unchanged; the game files
are a copy of the player's own prepared files.
Problems come back in plain words with the fix to offer (a busy port with a
free one, missing game files with Prepare, the EULA).
horizon-launcher servers list|create|set-up|run|forget does the same from a
terminal.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
`config [--json]` prints the settings; `config set KEY=VALUE...` writes them through
toml_edit, checked as server.toml is checked (nothing is written if one is wrong). A
running server takes the difficulty, the default game mode and the whitelist at once
through its console, and the backup settings because the supervisor reads them each
time; the rest waits for the next start, and the command says which.
Also `backups --json` and `update --check --json`, for the launcher's Servers page.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify passed; cloud_render, environment_render, weather_render and 923 client tests pass, with before/after screenshots of Complementary Unbound. Java shader packs get real eye brightness, custom uniforms, Iris's sun path and tilt, and clouds=off/sun/moon flags. Part of #944 (reported by brabus_); shadows and gbuffers still to come.
🤖 Agent: merged after the verify rerun passed (the first run's failure was a flake; horizon-server tests pass 39/39 locally) and CI parity passed. Old releases are pruned automatically after an update.
🤖 Agent: merged after verify and CI parity passed (modpacks-10, reviewed with local harness proof in the agent's report). Carries Formations' structures.
🤖 Agent: merged after verify and CI parity passed (modpacks-10, reviewed with local harness proof in the agent's report). Carries the mods' pillar and fluid-fix processors in ported packs.
🤖 Agent: merged after verify and CI parity passed (modpacks-10, reviewed with local harness proof in the agent's report). Structure pillars and fluid fixes placed outside the template, like the mods do.
🤖 Agent: merged after verify passed and the HUD harness with new §-coded cases (34/220 on main → all pass). The sidebar applies § formatting codes like vanilla. Fixes#937.
🤖 Agent: merged after verify passed and local proof (menu_click fails on main 7978/8000, passes with the fix; place_recipe passes). The player's armour equip sound and event follow every vanilla path. Fixes#986 (reported by famescare).
Complementary Unbound drew clouds and light shafts inside terrain and a
second sun because the full-screen passes got wrong inputs:
- eyeBrightness was a constant (0, 240), so packs saw open sky
underground. It is now the block and sky light of the camera entity's
eye block (CommonUniforms.getEyeBrightness), and eyeBrightnessSmooth
follows it with the pack's eyeBrightnessHalflife (SmoothedFloat).
- Custom uniforms (uniform.* / variable.* in shaders.properties) were all
0: Complementary's eyeBrightnessM, isEyeInCave, starter, framemod8 and
the rest. horizon-shaderpack's uniforms.rs parses and evaluates them
each frame like Iris (stareval priorities, IrisFunctions, SmoothFloat).
- sunPosition, moonPosition, shadowLightPosition, sunAngle and
shadowAngle now follow CelestialUniforms: the sky's sun and moon angles
turned by Y -90, the pack's sunPathRotation about Z, then the angle. The
old sun vector pointed north-south instead of east-west.
- While a Java pack runs, sunPathRotation tilts Horizon's sun, moon and
stars as Iris tilts vanilla's, and clouds = off | fast | fancy, sun and
moon = false apply. Complementary and Sildur's set clouds = off; the
vanilla clouds Horizon drew anyway wrote depth the packs' light shafts
and fog took for terrain.
Fixes#944
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged, thanks Online5277! Ludovic decided: Classic menus by default for fresh profiles (saved choices and HORIZON_MENU_STYLE still win). We merged current main in; fmt, clippy -D warnings and 1,022 client tests pass, and verify passes. Fixes#548.
VanillaPlaceRecipeOracle prints VanillaMenuOracle.state, so the equips
field the menu_click harness compares broke its lines. Only the
menu_click oracle's own lines carry it now.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
LivingEntity.onEquipItem now runs for the player wherever vanilla calls
it: the inventory menu's ArmorSlot and off-hand slot (setByPlayer, in the
menu engine, so clicks, shift-clicks, number keys, drags and LAN guests'
clicks), right-click equipping (Equippable.swapWithEquipmentSlot), the
creative screen (every changed slot, as handleSetCreativeModeSlot sets it),
the swap-hands key, dispensers (now also a shield into the off hand) and
/item and /loot on armour slots. It plays Equippable.equipSound (the
generic armour sound by default) on PLAYERS when the item lands in its own
slot, and emits equip/unequip game events for armour; nothing for
spectators, on a client, or when the item is unchanged.
The menu click harness's oracle now records vanilla's onEquipItem calls.
Fixes#986
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla's Font draws every literal through StringDecomposer.iterateFormatted:
a § code styles what follows and is not drawn, a lone trailing § is
dropped. The sidebar drew owner names, team affixes, scores and the title
raw, so servers' sidebars showed § and the code letters. They now go
through Text::resolve_legacy, as the tab list already does.
The HUD harness's sidebar cases gain legacy-coded owners, affixes, titles
and fixed score formats.
Fixes#937
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify passed. Remote players' projectiles record their owner: ender pearls teleport and hurt the thrower, mobs target the shooter; docs/server.md row updated; part of #484.
🤖 Agent: merged after verify passed. A remote player's item used on a block is used once (a snowball thrown at the ground flew 2× main hand / 3× off hand); part of #484.
🤖 Agent: merged after verify passed. Chat commands from a dead player run, as in vanilla's handleChatCommand (fixes#973); dedicated_server transcript 604/604 lines identical.
A projectile a guest shot was recorded as the local player's (Who::Player), which on a
server means the host, or nobody on a dedicated server. So a guest's ender pearl
teleported the host on a LAN world and did nothing on a dedicated server, what it hit
answered to the host, and clients were told the projectile had no owner.
Projectiles now remember the guest who shot them (Projectile.guest, set when the
guest's turn ends). Clients get the guest as the owner (Projectile.getAddEntityPacket),
and the hit runs as that guest: their pearl teleports and hurts them, and the mob hit
answers to them. ThrownEnderpearl's endermite also needs monsters to spawn, and comes
where the player was.
The dedicated-server harness checks that the snowballs' owner is the player and throws
an ender pearl at the player's feet: vanilla teleports it and takes 5 health, main did
neither.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A vanilla client sends a right click as one packet per hand and part
(Minecraft.startUseItem): UseItemOn for the block's part and ItemStack.useOn, then
UseItem for Item.use when that passed, the main hand first. The server ran the whole
click, both hands, for UseItemOn, then the item again for UseItem: a snowball thrown
at the ground flew twice, three times from the off hand.
Each packet now does its own part with its own hand
(ServerGamePacketListenerImpl.handleUseItemOn / handleUseItem): UseItemOn stops at
useOn (no throwing, drinking, eating, raising, equipping, buckets or an ender eye off a
frame), and neither touches entities (Interact does). Horizon's own client, which
counted on the server trying the off hand, now sends the off hand's packets when the
main hand's use passes, as vanilla's does. UseItem's rotation is the player's now
(absSnapRotationTo) without sending the client a teleport back.
The dedicated-server harness right-clicks the ground with a snowball in the main hand,
then in the off hand, and counts the snowballs against vanilla's (2 and 3 on main,
1 and 1 now).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Player attacks between the host and guests, or two guests, only checked the pvp game
rule. ServerPlayer.canHarmPlayer also asks the victim's team (Player.canHarmPlayer):
an ally on a team with friendlyFire false takes no damage, and the attacker hears the
no-damage sound.
lan_e2e's pvp check puts both players on such a team and checks that neither hurts
the other.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A dead guest's packets were all dropped but the respawn request, so a command typed on
the death screen never ran. ServerGamePacketListenerImpl.handleChatCommand and
handleSignedChatCommand don't look at the player's health: the commands now run.
The dedicated-server harness sends `/trigger nope` from the death screen and compares
the reply with vanilla's before respawning.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
An install kept every release it ever unpacked in releases/. Once the game is up on the
new release (or a stopped server is switched), the update now deletes the old ones,
keeping the release in use, the one `previous` names (rollback's target), the one the
running horizon-server came from, and the `updates.keep` newest of the others (1 by
default, a new server.toml setting). A release that can't be deleted is reported and
left for the next update.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify passed. If the game crashes natively after its Vulkan adapter line and before its first frame, the launcher switches the Graphics API setting from Auto to DX12 and tells the player; it doesn't relaunch by itself.
🤖 Agent: merged after verify passed. DirectX 12 for Windows: Auto tries Vulkan first and falls back to DX12 when Vulkan has no adapter or its device request fails; Video Settings > Graphics API (Windows only) and HORIZON_GRAPHICS_API override it. We merged main in (kept #771's LOD row and lodDistance next to graphicsApi); clippy -D warnings and 1,021 client tests pass, including the WGSL→HLSL translation of every shader. Not yet run on real Windows hardware.
- Integrated GPUs other than Apple's draw far terrain to 128 chunks at
most, whatever LOD Distance says; its description says so, and the
status reports the distance in use (`lod.effective_distance`).
- On machines with four threads or fewer, far terrain starts no detail and
one sample job per 500 ms while the near terrain loads, so the near
chunks and the frame get the cores first.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- Upload far meshes within about 2 ms a frame (at least one); the rest
wait, with their removals and the cutout, and apply in order.
- Gather mesh inputs across frames within about 2 ms, from a snapshot of
the active tiles' samples, instead of hundreds at once.
- Select tiles only when the plan, samples, covered columns or dirty tiles
changed; add the covered stand-ins only after a replan.
- Find the tiles touching an invalidated chunk per level, not by scanning
every active tile; share tile samples with mesh jobs (Arc) instead of
cloning them.
- Recompute the full-detail columns only when their completeness, the
center or the radius changed.
- Observe chunks for the far summaries within about 1 ms a frame (at least
one), unloaded ones first, with each block state's kind read once instead
of matching block names per block.
LODs off is unchanged: none of this runs without far terrain.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Far tiles mesh every gap in their columns: caves and aquifers in visited and
refined chunks, and the gaps kelp and seagrass leave in water. Most are closed
on every side, so their floors, ceilings and walls cost vertices and never
show.
Before meshing, a tile fills each gap below a column's top that no
neighbouring column's open air looks into (every neighbour's top is at or
above the gap's top) with the material below it, or with the fluid above it
(kelp in water). Cave mouths stay open; unknown neighbours and cells meeting
full-detail or finer terrain count as open.
Seed 1234, RD12, LOD 64, overview: 4.93 M to 3.17 M far-terrain quads.
Follow-up to Online5277's terrain LODs (#771).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Full-detail sections leaving the render distance stopped being drawn at once,
while the distant tiles still left their columns out until the next mesh job
finished: for a few seconds after moving (longer on slow machines) a hole
opened there, lined with the tiles' seal walls. Those took the light recorded
for buried blocks, none, so the hole showed as a dark band.
Lod::take_cutout hands the renderer the columns the finished meshes leave
out. Their resident sections stay drawn, even past the render distance, and
sections unloaded while still left out are released only once the tiles
cover them. With LODs off the cutout is empty and nothing changes.
Side faces above the top of the column they face open onto air. Reduced
columns and the seals expose sides that were buried, so those parts take at
least the sky light over that top (black cliffs in far tiles). Faces into a
cave keep their recorded light.
Follow-up to Online5277's terrain LODs (#771).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify passed and CI parity passed 322/323; the one failure, dedicated_server (two extra block updates at respawn, a server-only path this GPU change can't reach), passed 2/2 locally on main + this PR and 2/2 on main. Vendors wgpu-hal 30.0.1 with the upstream fix for gfx-rs/wgpu#9845; should fix#746 on HD 520/530, to be confirmed by the reporters on the release.
🤖 Agent: merged after verify passed (benchmark data and docs only). Four-way equal-work benchmark: Horizon uses 3.8–4.8× less RAM and starts in ~8 s; Paper's average tick is shorter at every count, which is now perf work.
Its formations:simple becomes the game's horizon:formations (its pool and
placement), and its single pool element vanilla's with its ground_level
under horizon:ground_level. Its pieced structure (one) stays out; its
processors are still dropped.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The structure context gets ChunkGenerator.getBaseColumn's blocks
(StartContext::noise_column: air, water, lava, solid or other, from the
noise's minimum Y, air past its ends), which Formations' placements test.
horizon:formations is Formations' SimpleStructure: one element of its pool
at a random spot of the chunk and rotation, at the height its placement
finds under the element's corners and centre (surface, ceiling, on water
or lava, in caves on the floor, under the ceiling, buried, on lava), as
the mod draws and checks them. Single pool elements take a ground level
(`horizon:ground_level`, 1 by default as vanilla's), which moves the piece
and its terrain adaptation as Formations' does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify passed. The site's nav links Host a server; README gains a hosting paragraph; tools/server.py and docs/server-quickstart.txt retired.
🤖 Agent: merged after verify and CI parity passed. horizon-server update (SHA-256 check, world backup, switch, health wait, automatic rollback) and rollback; 6-hourly update notice in log, status and panel. Proven end to end on the Mac mini and the ThinkPad (healthy update, crash rollback, timeout rollback, manual rollback, real GitHub download); 37 unit tests.
Better Ocean Monuments' ten processors (no fields; their behaviour is in
its Java) become vanilla's rule, the game's horizon:pick (one new
nextFloat per chance), horizon:flood_air, horizon:waterlog,
horizon:keep_world_block and horizon:yung_pillar; Better Witch Huts' log
and fence legs become horizon:yung_pillar. Its monuments are no longer
dry.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Repurposed Structures' and Moog's pillar_processor, Repurposed's
bottom_pillar_processor and close_off_fluid_sources_processor, YUNG's
pillar_processor (Better Fortresses, Better Jungle Temples) and Better
Fortresses' stair_pillar_processor become the game's horizon:pillar,
horizon:bottom_pillar, horizon:close_off_fluid_sources,
horizon:yung_pillar and horizon:stair_pillar. A pillar's own processor
list is written in, with the Java processors in it dropped.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Towns and Towers ships its 60 structures under kaisyn and towns_and_towers,
not its mod id t_and_t, so carry left them outside. A namespace is now the
mod's when its own files (its id's, its tag additions, the namespaces
already claimed) name ids in it and no mod of the pack has it as its id.
Compat files for other mods and pack mods' configs stay outside; of ATM10's
42 jars only T&T claims any. A replacing tag of its own is an addition.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Adds the game's side of Better Ocean Monuments' Java processors:
- horizon:flood_air: the template's air below sea level becomes water
(its AirProcessor; the monuments' templates are mostly air, so they
stood dry);
- horizon:waterlog: below sea level, blocks that can be waterlogged are,
and water and lava on the edge of the chunk being generated are marked
for postprocessing (its WaterlogProcessor);
- horizon:keep_world_block: a placeholder takes the world's block (its
StructureVoidProcessor);
- horizon:pick gains `redraw` (a new nextFloat for each chance, as its
random processors draw) and `keep_properties` (Block.withPropertiesOf,
for its copper stairs and slabs);
- horizon:yung_pillar gains `keep_properties`, for Better Witch Huts'
fence legs.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A content mod's manifest may list namespaces its files use besides its id
(Towns and Towers ships kaisyn and towns_and_towers under mod id t_and_t).
They are the mod's as its id is: files load, loaders read them. minecraft,
c and horizon cannot be listed. Two mods that are on owning one namespace
is an error naming both; the game keeps the first and leaves the other's
content out.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The loot loader read data/minecraft/ only, so every chest in a carried
mod's structure that names its mod's loot table opened empty. It now reads
every namespace (a vanilla root has only minecraft). Integers written as
floats (rolls 3.0, as Towns and Towers writes them) read as vanilla's int
codecs read them. 534 of the 675 non-block tables in ATM10's carried mods
now fill a chest; vanilla's chest fills are byte-identical.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Adds the game's side of the mods' Java processors that write blocks around
a structure while its template is processed:
- horizon:pillar and horizon:bottom_pillar: Repurposed Structures' and
Moog's PillarProcessor and BottomPillarProcessor, a column from a trigger
block (or the bottom layer) down to the ground, with their nested lists
and Moog's height randomizer;
- horizon:close_off_fluid_sources: their fluid fix, which walls off source
blocks next to the structure's open blocks;
- horizon:yung_pillar and horizon:stair_pillar: YUNG's PillarProcessor
(Better Fortresses, Better Jungle Temples) and Better Fortresses' stair
supports.
They write through PlacementContext::set_chunk_block, the
ProtoChunk.setBlockState the mods call (heightmaps kept), and only inside
the chunk being generated, as the mods check WorldGenRegion's centre.
Processor lists nested in pillars are capped at 16 levels.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify and CI parity passed, and a combined check with rdbt's #796 on current main (fmt, clippy, 359 tests, 14 world-gen harnesses). YUNG/Moog terrain adaptation around structures (Better Fortresses netherrack inside bridges 39–41 % → 2.5–7 %), its carry, and YUNG modifiers.
🤖 Agent: merged after verify and CI parity passed, and a combined check with rdbt's #796 on current main (fmt, clippy, 359 tests, 14 world-gen harnesses). YUNG/Moog terrain adaptation around structures (Better Fortresses netherrack inside bridges 39–41 % → 2.5–7 %), its carry, and YUNG modifiers.
🤖 Agent: merged after verify and CI parity passed, and a combined check with rdbt's #796 on current main (fmt, clippy, 359 tests, 14 world-gen harnesses). YUNG/Moog terrain adaptation around structures (Better Fortresses netherrack inside bridges 39–41 % → 2.5–7 %), its carry, and YUNG modifiers.
🤖 Agent: merged, thanks rdbt! Bulk density arithmetic, the shared Beardifier kernel and Perlin column reuse: 12.685 → 10.069 ms/chunk Overworld terrain (−20.6 %, one worker). We merged current main in (density/tests.rs: kept both test sets) and checked it together with #955/#956/#959: fmt, clippy -D warnings, 359 core/world/port-pack tests, and 14 world-gen harnesses (density, noise, structure_density, server_world_gen, world_spawn, decorated_region, …) all pass.
🤖 Agent: merged after verify passed. The renderer logs each start-up step once per device (driver, requested features and limits, surface, pipelines, first acquire/submit/present), so a native crash's last log line names the step.
🤖 Agent: merged, thanks Online5277! Terrain LOD lands off by default. With LODs off it matches main (0 differing pixels in 6 RD16 views, same FPS/GPU ms on the HD 620, same time to full render distance); CI verify and parity pass. We merged main in three times and adapted it to #905/#939 and the packed texture coordinates. The follow-ups from the review (edge band, integrated-GPU cost, lighter far meshes) we're taking on ourselves.
A native crash in a graphics driver leaves no crash report, only the game's
last lines. When the game ends with no report after its Vulkan adapter line and
before its first frame (Windows only), set graphicsApi:dx12 in its graphics
settings (the game home's and the run's own, only where it was auto) and tell
the player to press Play again, instead of the crash notice.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The Windows client builds wgpu's dx12 backend. With graphicsApi:auto (the
default), the renderer opens Vulkan and falls back to DirectX 12 when Vulkan
has no adapter (no Vulkan driver) or its device request fails; vulkan and dx12
pick one (horizon-graphics.txt, Video Settings > Graphics API on Windows, or
HORIZON_GRAPHICS_API; WGPU_BACKEND still overrides). Other systems keep their
one backend.
DirectX 12 keeps wgpu's indirect-draw validation: it is what passes each
draw's first instance to shaders there. Occlusion culling stays Vulkan-only and
flat draws Metal-only, as before; multi-draw and the GPU timer use DirectX 12's
own. gpu-allocator now resolves to the windows crate wgpu-hal's dx12 code
uses (0.62): with cpal's 0.54 it did not compile. A test translates every WGSL
shader to HLSL as the dx12 backend does for FXC.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A native crash in the graphics driver leaves no panic or crash report: the
last log line is all a player can send. After the adapter line, log the
driver, the features and limits the device is requested with, the surface
configuration, the pipelines (with multi-draw, occlusion culling and the GPU
timer on or off) and the first frame's acquire, submit and present, once per
device.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
wgpu 30 creates a Vulkan descriptor pool with no pool sizes while it creates
the device. Vulkan allows that only since 1.3.215; Intel's Windows driver for
6th-10th gen integrated GPUs (igvk64.dll, HD 520/530, UHD 620/630) reads
past the empty array and crashes with 0xc0000005 inside request_device, right
after the game logs its renderer (gfx-rs/wgpu#9845). Apply the upstream fix
(gfx-rs/wgpu#10124: one sampler entry when the pool would be empty), which no
wgpu release has yet, through [patch.crates-io].
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The crates.io sources of wgpu-hal 30.0.1, minus its examples, lock file and
VCS metadata, so the next commit can apply one upstream fix that has no
release yet. Excluded from the workspace: it keeps its own lints.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify passed and local proof (sculk harness). Sculk drops every unloaded chunk's blocks in one pass: worst tick 583–609 → 38–44 ms with 40 bots.
🤖 Agent: merged after verify passed and local proof (heightmap, chunk_stage, lan, …). Chunk-packet heightmaps are built on the warm-up threads; bytes identical; lan_chunks −11–13 %.
🤖 Agent: merged after verify passed and local proof on the integration commit (8 harnesses, lan_e2e both modes). The server thread runs ahead of the workers on macOS and Windows, like vanilla's Thread.setPriority(8): 40 bots tick p99 101.5/77.7 → 32.8/32.7 ms, ticks over 50 ms 166/146 → 1/2.
🤖 Agent: merged after verify passed. The host clears the hazards box, so a dead guest no longer leaves it behind for pvp/hurt (8/10 → 0/3 failures under --server-thread).
🤖 Agent: merged, thanks Online5277! Server chat now goes through the shared chat component, so the wheel and Page Up/Down scroll it. We merged main into the branch (main had moved the atlas to self.gui.atlas); clippy -D warnings, the 6 chat tests and the LAN chat check (both modes) pass, and verify passes. Fixes#810.
🤖 Agent: merged, thanks LookImEcho-Rick! It was still marked draft, but its description was complete and our checks were done, so we marked it ready. /execute in <dimension> now moves the player across dimensions like vanilla (fixes#592). With main merged in and the #947 fix (#962), command_exec, command_parse, server_dimension_players and dedicated_server pass locally, and verify passes. If you meant to add more, a follow-up PR is welcome.
🤖 Agent: merged after verify passed and parity --changed 311/311 locally. Lithostitched apply_random, condition and set_block in the game: CTOV farms grow carrots, potatoes and beetroot, paths and mossy cobblestone appear.
🤖 Agent: merged after verify passed and parity --changed 311/311 locally on the integration commit. YUNG modifiers, piece_in_range and dead-end pools (game side).
🤖 Agent: merged after verify and CI parity passed. The carry reads data files with comments as Minecraft does; six Better Fortresses pools come back (265 → 3,765 pieces over seed 1234's Nether centre).
🤖 Agent: merged after verify passed and local proof on one integration commit of modpacks-9's PRs: parity --changed 311/311, structure-start oracle 3,153/3,153. CI parity's only failure was secure_chat, which this change doesn't touch. Stronghold rings 3/128 off → 0/128 on seeds 1234, 2, 3, 42, -7.
🤖 Agent: merged after verify and CI parity passed (command_exec 764/764 three times locally, 20 affected harnesses). /summon no longer spawns other pending entities and takes their id; the harness settle waits for entities like vanilla's. Fixes#947.
tools/server.py and docs/server-quickstart.txt are gone (horizon-server replaces them);
build_release.py builds only the client starter, and the docs point to docs/hosting.md.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The void can kill the guest before the hazards check tidies its stone box,
and the LAN server drops a dead guest's packets, so the box stayed: the
guest suffocated in it during pvp, and the hurt check's pig with it. The
host runs the tidying fills instead.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
MinecraftServer.spin raises its server thread (Thread.setPriority(8)) when
there are more than four processors. The new horizon-threads crate makes the
same request for the dedicated server's thread and the integrated server's:
user-interactive QoS on macOS, above-normal priority on Windows, nothing on
Linux (an unprivileged thread can't raise itself).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
`horizon-server run --dir srv` (what init prints) started the game in srv/ with
--control-socket srv/.horizon/engine.json, which the game resolved inside srv/,
so it crashed at every start. spawn now passes absolute paths.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A Vega 11 on Windows (Vulkan) panicked at launch in bytemuck's cast_slice
with TargetAlignmentGreaterAndInputNotAligned, starting with the release
that shipped occlusion culling. The section-test results and the GPU
timer's timestamps were cast in place from the mapped range, which isn't
guaranteed to be 4- or 8-byte aligned. Copy the words out instead.
Fixes#953
Co-Authored-By: Claude Opus 5.5 <[email protected]>
apply_random becomes horizon:apply_random: its weighted lists, or its set
of lists at weight 1, with the inline lists carried like any other (their
Java processors dropped and counted). condition becomes horizon:condition
when the game runs every condition in it (true, all_of, any_of, not,
random_chance, matching_blocks with plain property values), with its then
and else lists carried. set_block with a simple or weighted state provider
and no block entity change becomes horizon:set_block. Anything else stays
Java and is dropped as before.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
horizon:apply_random runs one processor list drawn by weight
(WeightedList.getRandom); horizon:condition runs `then` or `else` by a
test of the block (true, all_of, any_of, not, random_chance and
matching_blocks on the placed or the existing block). Both take their
random from Lithostitched's RandomSettings: seeded with the world seed
plus the name's hash, forked positionally at the block, the piece or the
structure's pivot. process_block now gets the piece's position for it.
horizon:set_block is Lithostitched's set_block with a simple or weighted
state provider: the block becomes the state (drawn at the chosen position
from the unsalted seed), keeping the properties it shares with the old one
unless preserve_state is off.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged, thanks rdbt! Terrain sampler scratch buffers are reused between chunks; byte-identical output across the 22 world-gen parity harnesses on the combined commit.
🤖 Agent: merged, thanks rdbt! Spectator scroll-wheel flight speed now follows vanilla (6,791 oracle cases); verified on the combined commit with local_input and the client tests. Fixes#945.
lithostitched:reference runs the processors of other lists in turn.
Carry now writes their processors in its place, in order and references
in them too, from the mod's own processor lists. A reference to a tag or
to a list the mod lacks stays, and is dropped and counted as before.
ChoiceTheorem's Overhauled Village builds most of its lists this way.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Lod::update sorted every unsampled plan node each frame to take at most
four (none while every sample slot is busy), and woke the workers every
frame. It now skips the selection while the slots are full, keeps only
the best 2 * (SAMPLE_JOBS + pending) entries before sorting (a key is
listed at most twice, so the same tiles are taken in the same order),
takes the best fallback with min(), and wakes the workers only when a job
was queued. A finished chunk's detail now goes straight to the tile at
each level 0-4 holding it instead of scanning every ready tile (the others
never overlapped it).
Mac mini, RD12, LOD 256 while far terrain generates, `sample` of the main
thread over 30 s, alternating: Lod::update 10.4 % and 12.3 % of the
samples before, 3.4 % and 3.5 % after; update_world 3497 and 3373
samples before, 668 and 864 after.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
YUNG's API and MoogsStructureLib add their own kernel to the Beardifier:
a size and falloff of its own, carving or burying above and below a piece's
floor, around rigid pieces and junctions. Better Strongholds buries its
stronghold, Better Fortresses carves the netherrack over its bridges and
Moog's mega fortress carves a band around its corridors. A structure or a
pool element holding `horizon:enhanced_terrain_adaptation` now gets that
kernel (horizon_core's TerrainKernel), summed after vanilla's density as
the libraries' mixins do; YUNG's also grows the start's box by its radius.
Vanilla data never has the key, so vanilla worlds are unchanged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Better Nether Fortresses relies on three parts of YUNG's API the game did
not have:
- piece_in_range: another piece (of some templates, or any) inside the
piece's box grown sideways, below and above.
- Modifiers, applied once the layout is done, in piece order and with the
assembler's random: a condition, then actions on the piece itself.
transform swaps its template for one picked by setFeatureSeed(x, y, x)
and moves it by an offset turned with the piece; delay_generation moves
it to the end of the piece list.
- Dead-end pools: a piece that got no child though it has other jigsaws
is taken out (its parent's junction, its room and its count given back)
and a piece of its dead-end pool is chosen for the same parent jigsaw.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Minecraft reads data JSON with Gson's lenient reader, which skips `//`,
`#` and `/* */` comments. Six of YUNG's template pools have them (Better
Fortresses' bridges, halls and hall ends; Better Jungle Temples' floors),
so carry refused them as unreadable and the structures lost those pieces:
seed 1234's seven Better Fortresses had 265 pieces in all, now 3,765, and
five Better Jungle Temples 170, now 196. A file that parses once its
comments are gone is carried as plain JSON.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ChunkGeneratorStructureState's ring searches call findBiomeHorizontal,
which samples its 57x57 quarts in order through one resolver; vanilla's
climate index starts each lookup from the previous result and keeps it on
a tie. Horizon built a fresh resolver for every quart, so ties went the
other way: three of seed 1234's 128 strongholds were up to 6 chunks off
(the first one 96 blocks from vanilla's). StartContext::noise_biomes
samples a batch through one resolver and the search hands it the square.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
YUNG's single elements keep their dead-end pool and their modifiers under
horizon:yungsapi, which the game's assembler now reads: the modifier's
condition, and its transform and delay_generation actions on the piece
itself. piece_in_range carries as is. A modifier aimed at another target
than self, or another action, is counted lost.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
YUNG's API's `enhanced_terrain_adaptation` (its presets spelled out, NONE
as size 0) and MoogsStructureLib's (with its band) were counted lost. Carry
now writes them as `horizon:enhanced_terrain_adaptation` on the structure
and on the pool elements that have their own: YUNG's with every piece's
junctions and the start's box grown by the radius, Moog's with only the
terrain-matching pieces' junctions, as each library does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged after verify and CI parity passed. tick_blocks finds the motion-blocking height from the top of the terrain: 1.98 → 1.36 ms per tick with 10 bots on the Mac mini (−31%).
🤖 Agent: merged after verify and CI parity passed. Non-living entities (TNT, falling blocks, boats, frames…) now tick only in entity-ticking chunks, as in vanilla.
🤖 Agent: merged after verify and CI parity passed; 301/301 harnesses locally on the combination with #951/#952. Dedicated servers now spawn mobs around every player (0 → ~70 per bot, 139–144 for two, matching vanilla's caps).
Coming back to a level, the entities saved with its chunks wait in
World::loaded_entities until the next tick spawns them, as vanilla's
PersistentEntitySectionManager.processPendingLoads does. /summon pushed its
compound onto that queue and drained all of it, so the first waiting
entity took the id counted as the summon's root: after a Nether round trip
`/summon minecraft:armor_stand ...` reported "Summoned new Pig". Summon now
spawns only its own stack and leaves the queue to the tick.
The command harness's settle now waits for the entities of the chunks
around the player as well, like the oracle's settle
(areEntitiesActuallyLoadedAndTicking), and a `portal` op sends the player
through the nether portal it stands in (Entity.handlePortal:
NetherPortalBlock.getPortalDestination, teleportToPortalDestination). The
corpus's new `levels` group makes the round trip and checks the Overworld's
pig and a summon's feedback on the way back.
Fixes#947
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The soft 4 ms yield per frame (also with Unlimited budgets) made a saved
RD32 world reach full render distance about 1 s later on the Mac mini
(streaming time after launch, 4 alternating pairs: main 1.9 s mean, yield
2.8 s; t100 from a first batch of 3 pairs: 6.1 s vs 7.1 s) without better
loading frame times (max 133-210 ms vs 31-150 ms). Without a budget every
finished mesh uploads at once again, as on main. The budgeted path keeps
this branch's fixes: only loaded chunks' sections wait, and sizes count the
24-byte terrain vertex.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: thanks rdbt! Verify passes. A per-worker memo of the carver biome lookups in front of the shared mutex: the resolved biome per origin is a pure function of seed, dimension and position, so results can't change, and the test checks eviction leaves the terrain identical. Paired benchmarks: +3.3–3.9 % terrain throughput at six workers (16 pairs, two sites), up to +5.9 % at size 32, no single-worker regression.
getHeightmapPos(MOTION_BLOCKING) walked every column from y 319 down,
one block lookup per y. Each precipitation roll (3 in 48 per ticking
chunk per tick) and each isRainingAt paid for it: with the chunks of 10
players ticking it was most of tick_blocks. Only air lies above the
WORLD_SURFACE heightmap the chunk already keeps, and air neither blocks
motion nor holds a fluid, so the scan starts there. LevelAccess gains
world_surface (the build limit by default, so other levels scan as before).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ServerLevel.tick ticks an entity only where the chunk tickets make it
entity ticking (inEntityTickingRange, and EntityTickList holds only the
entities of entity-ticking chunks). Mobs, items and minecarts already
checked it (#931); these six entity kinds ticked wherever they were
loaded, so TNT went off and sand fell past the simulation distance.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Natural spawning returned without a local player, so a dedicated or
--server-thread server spawned no mobs and a LAN host spawned them only
around itself. ServerChunkCache.tickChunks now takes its spawning chunks
from every player in the level (ChunkMap.collectSpawningChunks), the
global caps from the natural spawn chunk tracker's size, and the local
caps from each player (LocalMobCapCalculator). Inhabited time and
thunder follow the same chunks (tickSpawningChunk), the wandering trader
picks a random living player (getRandomPlayer), and mobs despawn by
their distance to the nearest non-spectator of any kind.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
main's #880 added Source::level and Source::elsewhere for server players
teleported into another level; this branch adds Source::dimension and
moves the local player. Kept dimension (it covers what elsewhere was read
for) and level; /tp tries main's teleport_across for a server's players
first, then this branch's local-player level switch.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: reviewed; verify passes. Ports removeWhenFarAway = false for Villager, WanderingTrader and AbstractGolem (iron/snow/copper golems, shulkers). Seed 1234 plains village: main 0 villagers, this PR 14, vanilla 26.3 server 14. 95/96 affected harnesses pass; world_spawn timed out on a busy machine, then 59/60 seeds alone (the one off is the known chunk-order flake #308). Fixes#917.
🤖 Agent: reviewed; verify passes. Re-checked all four perf-server-10 PRs merged on current main (after #887#931#912#916): fmt/clippy clean, horizon-client tests 893 passed, lan, dedicated_server, server_world_gen, chunk_stage, world_reopen, server_entities, server_player_targets, server_dimension_players pass, lan_e2e growth/plates/eating/hurt/chat/light pass in both modes (eating with --server-thread failed once in the long sequence, then passed twice alone). Numbers in the PR.
🤖 Agent: reviewed; verify passes. Re-checked all four perf-server-10 PRs merged on current main (after #887#931#912#916): fmt/clippy clean, horizon-client tests 893 passed, lan, dedicated_server, server_world_gen, chunk_stage, world_reopen, server_entities, server_player_targets, server_dimension_players pass, lan_e2e growth/plates/eating/hurt/chat/light pass in both modes (eating with --server-thread failed once in the long sequence, then passed twice alone). Numbers in the PR.
🤖 Agent: reviewed; verify passes. Re-checked all four perf-server-10 PRs merged on current main (after #887#931#912#916): fmt/clippy clean, horizon-client tests 893 passed, lan, dedicated_server, server_world_gen, chunk_stage, world_reopen, server_entities, server_player_targets, server_dimension_players pass, lan_e2e growth/plates/eating/hurt/chat/light pass in both modes (eating with --server-thread failed once in the long sequence, then passed twice alone). Numbers in the PR.
🤖 Agent: reviewed; verify passes. Re-checked all four perf-server-10 PRs merged on current main (after #887#931#912#916): fmt/clippy clean, horizon-client tests 893 passed, lan, dedicated_server, server_world_gen, chunk_stage, world_reopen, server_entities, server_player_targets, server_dimension_players pass, lan_e2e growth/plates/eating/hurt/chat/light pass in both modes (eating with --server-thread failed once in the long sequence, then passed twice alone). Numbers in the PR.
Sections arrive back to front (visible is sorted by distance, then key), so
the merged list only needs sorting when a distant tile adds translucent
draws, and ties now break the same way as main's reversed visible order:
with LODs off the draw order and cost are exactly main's.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: reviewed; verify passes. Main merged into the stack (processors.rs tests kept from both sides; port-pack tests pass on each layer, clippy clean). Parity: all 269 game-side harnesses on modpacks-8's integration commit, structure-start oracle 35/35.
🤖 Agent: reviewed; verify passes. Main merged into the stack (processors.rs tests kept from both sides; port-pack tests pass on each layer, clippy clean). Parity: all 269 game-side harnesses on modpacks-8's integration commit, structure-start oracle 35/35.
🤖 Agent: reviewed; verify passes. Main merged into the stack (processors.rs tests kept from both sides; port-pack tests pass on each layer, clippy clean). Parity: all 269 game-side harnesses on modpacks-8's integration commit, structure-start oracle 35/35.
🤖 Agent: reviewed; verify passes. Main merged into the stack (processors.rs tests kept from both sides; port-pack tests pass on each layer, clippy clean). Parity: all 269 game-side harnesses on modpacks-8's integration commit, structure-start oracle 35/35.
processors.rs tests: keep both #926's trial spawner and vault tests and
#920's Repurposed rule-test test.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Keeps #905's single command buffer and #939's occlusion culling: distant
tiles draw in phase 1, so phase 2 stays exactly the deferred sections the
culling test reads back, and the flat path's origins include the tiles in
main's relocated upload block.
Adapts to main's APIs: block sprites and models now live in RenderAssets
(WorkerContext::render), BlockFacts behind Resources::facts(), and
terrain vertices hold the UV as 16-bit fixed point (#df58a474). Merged
distant faces repeat their sprite once per block, which the [0, 2) section
UVs cannot hold, so tile vertices now store UVs from the face's corner in
1/32-block steps (TerrainVertex::tiled, faces split at 1024 blocks) and the
instance's origin.w carries the UV step: 1/32768 for sections (decodes
exactly as before), 1/32 for tiles.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: reviewed; verify passes. HD 620 (Vulkan/Mesa) culling off → on: RD16 +22–42 % FPS (peak unchanged, culling paused), RD32 +14–109 %; 1 % lows and CPU ms equal or better except RD32 peak's 1 % low (38.8 → 37.4). Screenshots: static views within main-vs-main noise, 21 frames after 30–40° camera jumps identical (no popping). parity section_mesh, translucency, light, hud locally; CI parity covers the rest. On only on Vulkan devices with SHADER_EARLY_DEPTH_TEST, INDIRECT_FIRST_INSTANCE and fragment-writable storage; Metal draws as before. Not yet checked on NVIDIA/AMD or Windows Vulkan: HORIZON_GPU_CULL=0 turns it off.
Sections that terrain in front hides entirely are drawn with an instance
count of 0. Phase 1 draws the sections the last test read back saw; in the
same pass every other section's box (and an eighth of phase 1's) is drawn
against phase 1's depth, front faces only, with a fragment shader forced
past an early depth test that marks the section seen. Phase 2, in a second
pass, draws what the test saw. The test is exact per pixel and needs no depth
read or Hi-Z, and every solid draw stays one entry of one list, so the CPU
records as many draws as before (Mesa unrolls a multi-draw per entry on
Intel Gen9). Culling pauses while the last result hid under a tenth of the
solid indices, with one frame in 64 testing everything.
HD 620 at RD16: +23-42% FPS in five of perf-frame-6's six views, the peak
view unchanged; RD32: +14-109%. Pixels unchanged, also on camera jumps.
Vulkan only: naga's MSL has no early fragment tests. HORIZON_GPU_CULL=0
turns it off, `all` culls every frame; `section-cull on|off` in a game.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: reviewed; verify passes. Release dry run on this branch (run 39167763008): linux, macos, windows builds and publish all succeed; horizon-server-linux-x86_64.tar.gz (28 MB) built and import-checked (no libasound/libudev). Agent's proof: 880/879 client tests in both builds, dedicated_server/server_entities/lan on the server build, install → init → start → bot join on the ThinkPad without the libraries.
🤖 Agent: reviewed; verify passes. Equal-work server benchmark: fixed exploring speed, 300 s warm-up, chunk-work comparison per run. Horizon 1.8/4.7/11.7/23.8 ms per tick at 1/10/25/50 bots vs vanilla 11.3/34.3/50.5/stopped by its watchdog; single runs on the ThinkPad, caveats in the README.
🤖 Agent: reviewed; verify and parity pass. tools/vanilla_client.py runs on macOS (vsync off so a hidden window doesn't block the join); the owner's vanilla 26.3 client joined a Horizon server and stayed connected 4 min.
Explore bots paced at 0.8 blocks a second and a 300-second warm-up, so both servers serve the same
paths; the work table shows it. Horizon ran as horizon-engine (the server build, from its archive).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: thanks rdbt! Verify passes. Checked locally merged with current main: fmt and clippy clean, cargo test -p horizon-client (885 passed), 14 harnesses including your two new oracles (player_sync, interaction_input, local_input, sprint_input, item_use, item_use_start, item_use_lifecycle, use_item_on, teleport_consume, lan, dedicated_server, server_player_combat, server_player_targets, server_entities), and lan_e2e pvp, hurt, eat, eating, clicks, chat, respawn in both modes: all pass. Fixes#903.
Village villagers are placed from the jigsaw templates without
PersistenceRequired (the village's iron golem has it), so Mob.checkDespawn
discarded every villager of a village generated more than 128 blocks from
the player, and the rest after a while beyond 32 blocks. Vanilla's
Villager, WanderingTrader and AbstractGolem (iron, snow and copper golems,
shulkers) override removeWhenFarAway to false.
Fixes#917.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: reviewed; verify passes. parity --changed 95/95 before the rebase; the 9 block-tick harnesses (dedicated_server byte-identical, 591 lines) and lan_e2e plates/growth/time in both modes after it; growth fails on main. Fixes#890: blocks tick from vanilla's ticket levels around every player. MSPT with 10 bots rises from ~1.6 to ~14 ms because blocks now tick at all on a server (vanilla: 40).
ChoiceTheorem's Overhauled Village swaps blocks in its templates through
Lithostitched's block_swap processor (137 lists): a map of block to
block, the new one taking Block.withPropertiesOf the old state, and the
block left alone when the new one is not registered. Carry dropped it.
- Game: horizon:block_swap, the same map; the new block's default state
with each property of the old one it has too (by name, when the value
is one of its own).
- Carry: lithostitched:block_swap becomes it.
ATM10: processors carried 136 -> 273 of 1,427 (dropped 1,291 -> 1,154).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Better Dungeons' small dungeon templates hold several chests; its Java
chest processor keeps the first, then each other one on nextFloat <= 0.2
until two have stayed, and turns the rest into cave air. Carry dropped it,
so every chest stayed.
- Game: horizon:count_cap. Of `blocks`, the first `min` a placement meets
stay, then each stays on settings.getRandom(pos).nextFloat() <= chance
until `max` have; the others become `replacement`. Counted per
placement (PlaceSettings, reset in process_infos), like the mod's
DungeonContext, set up per StructureTemplate.placeInWorld.
- Carry: betterdungeons:small_dungeon_chest_processor becomes it with the
mod's config defaults (1, 2, 0.2; chest and trapped chest).
Seed 1234, five small dungeons decorated: 12 chests -> 6.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Better Desert Temples builds its temples with placeholder blocks (orange,
red and white glass, blue and yellow concrete, red wool, polished diorite,
blank pots) that its Java processors turn into suspicious sand, sand and
sandstone, upside-down stairs, cobwebs, terracotta, TNT and pots with
sherds. Carry dropped them, so the placeholders stayed in the world.
- Game: a horizon:pick processor. A block of `block` takes the first
pick whose `below` is past one settings.getRandom(pos).nextFloat(),
else `default`; an outcome is a state or one of several by nextInt
(Util.getRandom), a new block entity or the block's own, and NBT lists
of picked strings (one float each).
- Carry: the fixed swaps become vanilla's rule; the random ones
horizon:pick with the mod's bounds (BlockStateRandomizer's sums as
floats). The processors that read or change the world around (pillars,
Diorite's waterlogging) stay dropped.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Without a local player (dedicated server, --server-thread) no block ticked:
tick_blocks returned early, so crops, fluids, redstone, pistons and pressure
plates stood still, and a LAN host ticked only around itself.
ticket_levels ports DistanceManager's levels: PLAYER_LOADING at 31 within the
view distance and PLAYER_SIMULATION at 31 - simulation distance for every
player in the level (the local player, LAN guests, a server's players), and
FORCED at 31. Random ticks run in simulation entity-ticking chunks that are
ticking chunks, scheduled ticks, block events and block entities in block
ticking ones, entities in entity ticking ones; the sculk blocks follow the
same rule. Saved ticks unpack when a chunk starts ticking
(startTickingChunk), not when it loads.
The hub keeps the tick-rate state current while nobody is in the game, so a
player joining a frozen server is told once (updateJoiningPlayer).
dedicated_server parity froze nothing while the client was sent its chunks:
saved fluid ticks there (a lava fall that never settles) ran before or after
the chunk was sent depending on each server's loading speed, and it failed on
main too. It now freezes the game until the client has its chunks.
lan_e2e gets a growth check: water flows and wheat grows next to a guest
1000 blocks from the host.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Moog's trial spawner and vault randomizing processors keep the block and
change its block entity. The templates were saved with state that breaks
them in a new world: a vault's server_data holds
state_updating_resumes_at far in the future (it never opens) and a
spawner's spawn_data an empty entity (it spawns nothing of its own).
- Game: a horizon:edit block entity modifier takes `remove`'s keys out,
then merges `data` in. append_static now merges as CompoundTag.merge
does (compounds in both are merged), not a shallow replace.
- Carry: each processor becomes vanilla's rule processor, one rule per
block state (12 trial spawner, 32 vault), with horizon:edit: the
spawner's normal and ominous configs by id, the vault's loot table and
key (the ominous ones on ominous vaults). trial_spawner/ files get their
mobs' 1.21.1 ArmorItems/HandItems as 26.3's equipment.
ATM10: 6 more processors carried (125 of 1,427).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The tool picks this machine's natives (macOS arm64 or Linux), adds -XstartOnFirstThread on macOS
and turns vsync off there: with it on, an unseen window (a locked session) blocks in
SDL_GL_SwapWindow before the client joins. docs/server.md records the join.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Repurposed Structures' jungle fortress and End stronghold still write
vertical_distance_from_start_piece, but no codec of the mod (or of Moog's
Structure Lib) reads it any more: its own structures ignore it. Carry
drops it without counting it in rules_lost, which now lists only rules
the mods do run.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
betterdungeons:small_nether_dungeon is YUNG's jigsaw structure (the same
fields and YungJigsawManager assembly) behind the mod's
smallNetherDungeons config switch, which ships on. Carry maps it as it
maps yungsapi:yung_jigsaw: minecraft:jigsaw with horizon:yungsapi.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Repurposed Structures' 10 overworld mineshafts (birch, desert, icy,
jungle, ocean...) are its generic jigsaw with one change in
MineshaftStructure.postLayoutAdjustments: the layout is not centred on
the chunk; its top goes 15 blocks under the lowest ocean floor at the
chunk's middle and 16 blocks to each side (taking one off after each
side, as the mod does), never up, within y_allowance.
- Game: "layout": "mineshaft" in horizon:piece_limited runs that step.
- Carry: repurposed_structures:generic_mineshaft becomes minecraft:jigsaw
with the piece-limited layout and that key, as its generic jigsaw does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A merged or closed pull request's queued runs held the self-hosted runners
until someone cancelled them by hand. pull_request_target lets the token
cancel runs of fork pull requests too; the job checks out nothing. It
matches the branch and the head commit, so a fork's branch named like
another doesn't cancel that one's runs.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Two things kept Repurposed Structures' mineshaft processor lists from
loading in 26.3, and with them the mod's 5 Nether mineshaft pools:
- Its three rule tests written in Java. Carry renames them to the game's
horizon:matter_phase, horizon:piece_origin_axis_aligned_linear_pos and
horizon:y_value (213 in 75 lists), counted in the report's
rule_tests_carried.
- 1.21.1's minecraft:chain, which 26.3 calls minecraft:iron_chain (its
datafixers rename it in worlds). Carry renames it in recipes, loot
tables and worldgen files: 112 files in 5 of ATM10's mods.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Repurposed Structures' mineshafts and Nether monument test blocks with
three rule tests of their own. Vanilla, and the game, refuse a processor
list naming a type they do not know, and with it every pool naming the
list: the 5 Nether mineshaft start pools in a ported pack never load.
The game now runs them under its own types, with the mod's fields:
- horizon:matter_phase (MatterPhaseRuleTest): solid (not air, no fluid,
canOcclude), liquid, air, and air or liquid with rails and chains,
optionally inverted. Structure processors now bind the block
properties to read canOcclude.
- horizon:piece_origin_axis_aligned_linear_pos: vanilla's axis-aligned
chance over the block's distance from its template's origin.
- horizon:y_value: the block's world height in a range.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Conflicts: the hub's player state keeps both main's attributes (#857) and
#887's living_flags; remote.rs and lan.rs debug JSON keep swinging and
using; the guests' model tick keeps both the swing update and the item
in use; lan_e2e.py checks unioned. remote_player_eating reads main's
Option<Consumable> (#799): any consumable with particles, its own use
time and should_emit, as the local player does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: reviewed; verify passes. Mac mini RD32 standing, 4 alternating pairs: 116.1 → 163.3 FPS, CPU 7.09 → 4.98 ms/frame. Screenshots of 6 views plus HUD and inventory: no pixel differs beyond main-vs-main noise. parity --changed 95/95 locally. No early return after the copies encoder is taken, so staged section copies can't be dropped.
horizon-client's cpal output and its controllers (gilrs, the DualSense's HID output) move behind the
audio and controllers features, on by default, so the game is unchanged. Built with
--no-default-features it links neither libasound.so.2 nor libudev.so.1. release.yml packages that
build as horizon-engine in horizon-server-linux-x86_64.tar.gz (xtask package-server, checked by
check_imports.py --server); install.sh installs it on Linux (and --archive installs a local one);
horizon-server already prefers horizon-engine. The Dockerfile uses it too, and CI lints it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: reviewed; verify passes. Parity: 298/298 harnesses on the integration commit 234bd101 (main + #907/#908/#910); vanilla structure-start oracle 35/35 with all five modpacks-7 changes. Merged in the required order #907 → #908 → #909 → #910 → #911.
🤖 Agent: reviewed; verify passes. Parity: 298/298 harnesses on the integration commit 234bd101 (main + #907/#908/#910); vanilla structure-start oracle 35/35 with all five modpacks-7 changes. Merged in the required order #907 → #908 → #909 → #910 → #911.
🤖 Agent: reviewed; verify passes. Parity: 298/298 harnesses on the integration commit 234bd101 (main + #907/#908/#910); vanilla structure-start oracle 35/35 with all five modpacks-7 changes. Merged in the required order #907 → #908 → #909 → #910 → #911.
🤖 Agent: reviewed; verify passes. Parity: 298/298 harnesses on the integration commit 234bd101 (main + #907/#908/#910); vanilla structure-start oracle 35/35 with all five modpacks-7 changes. Merged in the required order #907 → #908 → #909 → #910 → #911.
🤖 Agent: reviewed; verify passes. Parity: 298/298 harnesses on the integration commit 234bd101 (main + #907/#908/#910); vanilla structure-start oracle 35/35 with all five modpacks-7 changes. Merged in the required order #907 → #908 → #909 → #910 → #911.
🤖 Agent: reviewed; verify passes (client --help exits 0 without loading the game, so the start check only catches a program the system can't load). Linux installs now name the missing ALSA/udev libraries; services run through the installer's current link.
A dedicated server printed refused logins on its console as vanilla
does ("[Server thread/INFO]: Disconnecting <name> ..."), but wrote them
to logs/latest.log under the connection thread's own name
("[horizon-lan-conn/INFO]"): the file logger used the OS thread's name,
not the one vanilla logs the line from. The engine's lines now go to the
file under the same thread as on the console.
The console's times were UTC while logs/latest.log and vanilla's
console (Log4j's %d{HH:mm:ss}) use local time, so the two disagreed by
the time zone, as did the supervisor's own lines in
logs/supervisor.log and the web panel's console. Both now print local
time. Backup names keep their UTC time, as documented.
Found by the end-to-end install check on the Mac mini and the ThinkPad.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The first published benchmark for the working version (docs/server.md
section 10, item 4). It ran the v2026.10.18.15 release, as install.sh
installs it, against vanilla 26.3 with 1, 10, 25 and 50 mixed bots. The
servers ran on the ThinkPad (quiet, under nice 19) and the bots on the
Mac mini over the tailnet. Vanilla's default game rules applied on both;
Horizon's server runs the locator bar now.
Both servers kept 20 TPS throughout. Horizon used 3.7 to 5.2 times less
memory. Its average tick was shorter with 1, 10 and 25 bots and 18 %
longer with 50, while it generated about twice the new terrain a second
(the exploring bots fly as fast as each server sends chunks). Its P99
was lower at every count.
benchmarks/server-mixed-2026-10-18/ holds the README, the report and the
machine's load; docs/server.md links it from sections 9 and 10.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On the server thread, lan_e2e --check light failed its border step (#891):
the guest's border chunk still had block light 0 next to glowstone placed
in the chunk past it. The server was right; the check was not.
same_light() waits until the host's light equals the guest's, but on the
server thread the host's game is a client too, and both are dark until
the light update arrives. They agreed at once, same_light returned, and
the next line read the guest's light before the update came.
The check now waits up to 100 ticks for the guest's border light, then
compares the host's with it.
With temporary logging on the server thread: the glowstone lands in a
finished chunk that isn't lit (the ring past what the guest tracks), so
the edit queues a full light job for its neighbours. The border chunk's
job then relights sections 15 to 17, and Hub::broadcast_light_changes
sends them to both clients as a tracked border chunk, as
ChunkMap.isChunkOnTrackedBorder does. Nothing in the engine changes.
Fixes#891
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: Merged. Bit-exact against the per-corner oracle (unit test over a dense grid, 7 seeds, wrapping extremes), CI parity passed, and the surface pass is about 10% faster in chunkgen_bench. Thanks rdbt!
On a Linux without ALSA's library (most minimal server images), the
game exits at once with the loader's "error while loading shared
libraries: libasound.so.2". The supervisor took that for a crash and
restarted it five times over two and a half minutes before giving up
with "see logs/latest.log", which holds nothing about it.
Now:
- `run` starts the game with --help first (it loads no game files) and,
if the system can't load it, stops at once with the loader's line and
the packages to install on Debian/Ubuntu, Fedora, Arch and NixOS;
- `init` prints the same as a warning after "Game:";
- install.sh lists the libraries `ldd` can't find, with the same advice;
- docs/server.md says so in the install section.
Also, a service installed without the installer's launcher on the PATH
ran `<dir>/releases/<tag>/horizon-server`: Linux reports the running
program with links resolved, so the unit pinned that release and an
update never reached it. The fallback now goes through `<dir>/current`
while it points at that release.
Found by the end-to-end install check on the ThinkPad (NixOS, no
nix-ld libraries) with v2026.10.18.15.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: Merged. Thanks rdbt for the consolidation and for merging main yourself! On de6f15e6: fmt, workspace clippy, horizon-client tests and the client build pass, and 19/19 covering harnesses pass (options, world_reopen, lan, dedicated_server, block_removal, section_mesh, resource_pack, menu_click, screen_layout, the pack, world-options, chat and container screens, and the consumable ones). We'll handle the two follow-ups noted in the comment above (the 120 s fatal timeout, and the backdrop terrain built on the event loop).
init and run looked for asset-index.json to recognise a prepared folder,
but horizon-prepare writes that file only with the optional sounds
(`horizon-prepare sounds`). A server prepared the usual way, without
sounds, was never found: init said "no prepared game files" and listed
the folder that held them. They are now recognised by
horizon-resources.json, the manifest every preparation writes and the
one the game itself reads.
Found by installing v2026.10.18.15 with install.sh on the Mac mini and
preparing from the release folder.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: Merged. Matches 26.3's KeyboardHandler.keyPress: the F3 toggle isn't conditional on a world or the absence of a screen, and the title counts as an open screen so it gets no gameplay keys. Fixes#811. Thanks rdbt!
wgpu-core ends its Metal command buffer at every pass: each render or
compute pass costs two command buffers (the pass and its barriers), each
run of copies between passes one more, and Metal creates and commits every
one on its own. The frame made about 18: the section copies in their own
encoder, the queue's pending writes for the HUD's fire animation, copies
before and after the flatten pass, and the screen effects' and HUD's
vertex uploads between later passes. At render distance 32 that cost 2-5 ms
of submit per frame, mostly waiting on the command queue's lock.
The frame now records after the section copies in their encoder, stages
the fire animation and the HUD's glyph patches through the frame's belt
instead of Queue::write_texture, and records every upload (screen
effects, HUD, flatten inputs, indirect draws) before the first pass.
The commands and their order within each pass are unchanged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A `fly` benchmark replied with frame times only. With single player on its
server thread, the world's ticks no longer show in the frames, so a flight
could not tell how long the server took.
The summary now has tick_ms and tick_ms_max: the world's smoothed tick time
(the F3 server line) averaged and at its highest over the flight. On the
server thread it is the server's.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Block triggers counted only the host's own player and the minecarts, so a
guest stepping on a plate did nothing. The guests in the level, not
spectators, now trigger them through Entity.checkInsideBlocks and are
counted by the plates' entity queries. On the server thread no block
ticks run at all yet (#890).
lan_e2e.py plates: a stone plate under the guest, on both sides.
Part of #848
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: Merged. Fixes#861: a ctl tp/fly move is sent even while the destination chunk isn't loaded, the single-player owner is exempt from the move check as in vanilla, and wait-ready waits for the level around the player. lan, dedicated_server, server_entities and server_dimension_players pass.
🤖 Agent: Merged. CI parity and verify passed, and on current main with this branch, horizon-client clippy is clean and all its tests pass (873 lib + integration). Consumables now finish as in vanilla 26.3.
With the world on its server thread, a horizon-ctl `tp` or `fly` moved the
window's player to chunks the server had not sent. The client skips its
tick while its own chunk isn't loaded, so it never sent the move, and the
server kept streaming the old area: "loading terrain..." forever.
- The client sends a move made where nothing is loaded (by horizon-ctl;
a player who walks there has already sent it), so the server streams
the chunks there.
- The integrated server doesn't check its owner's moves, as vanilla's
`shouldCheckPlayerMovement` (`isSingleplayerOwner`); a long `tp` was
"moved too quickly".
- `wait-ready` on a server also waits for the level around the player
(`PlayerLoaded`); the status shows it as `server.loaded`.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
BrewingStandBlock opens BrewingStandMenu for a guest: the host clicks it on
the menu engine over the stand's slots (one bottle per slot, ingredient and
fuel slots, quick moves by item) and sends its four data slots
(BrewingStandBlockEntity.dataAccess). The guest opens minecraft:brewing_stand
menus, maps their slots, predicts clicks on Kind::Brewing and shows the
times. Reported by Tydragon00.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Neither side sent LivingEntity.DATA_LIVING_ENTITY_FLAGS for players, so
no one saw another player eat, draw a bow or raise a shield. The host now
sends the item in use and its hand for itself and every guest (always:
the trackers send only the entries that changed, and stopping is a change
back to 0), and draws its guests' bodies using their items. A guest's
client starts and stops the other players' item use from the flags and
drops their eating crumbs at their mouth (Consumable.onConsumeTick); the
eating sounds already come from the server.
lan_e2e.py eating: the host and the guest each see the other eat bread,
and stop.
Part of #803
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The host ignored ServerboundPunchPacket, so a guest mining or hitting
swung only on its own screen; only the first swing a guest's use ever
started went out, the host's own swings never did, the host drew no swing
on guests' bodies and a guest's client ignored
ClientboundSwingAnimationPacket.
The host now swings the guest's hand on a punch (handlePunch), sends each
tick's guest swings to the players around and the host's own to the
guests, and swings and ticks the guests' bodies it draws. A use the server
ran that swung goes back to the guest too: Horizon's client does not run
uses itself, so it never predicted that swing (the hoe). The guest's
client swings itself or the other entity on the packet.
lan_e2e.py swings: mining swings seen from each side, and the guest's own
hoe swing.
Part of #848
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The host told guests of a hurt mob with EntityEvent 2, which 26.3's
clients read as a kinetic hit, not a hurt. It now sends
ClientboundDamageEventPacket (ServerLevel.broadcastDamageEvent) with the
damage type and the attacker, and the guest's client restarts the hurt
flash and jolts the walk animation (LivingEntity.handleDamageEvent).
lan_e2e.py hurt: a pig the guest hits and one the host hits flash on
both sides.
Part of #848
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The guest's client closed every menu it had no screen for, so hoppers
(MenuType.HOPPER) and dispensers and droppers (GENERIC_3x3) did nothing.
They open as their 5 and 9 slot screens now, drawn by the menu's type
and under the server's title.
lan_e2e.py machines: each opens with its slots and keeps a stack the guest
shift-clicks in.
Part of #848
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The host sent the overworld clock's rate even while /time pause held it
(ServerClockInstance.packNetworkState sends 0 when paused), so a guest's
day kept running. A guest also moved its clock by the rounded rate each
client tick and never caught up when the server's game-time sync moved
game_time on, so a guest whose game lagged fell behind for good.
ClientClockManager.tick advances the clock by the game time passed, with
its partial tick; the guest now does the same.
lan_e2e.py time: the paused clock and a guest stalled for 4 seconds.
Part of #848
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A client joined to a server (and the window on the server thread) keeps the
server's chat lines in its own list. The event log only counted the local
chat's lines, so `wait-for chat` never saw a line from the server.
The remote client counts the lines it receives. The event log reads that
count and those lines while connected.
lan_e2e gets a `chat` check: a line the host sends reaches the guest's
event log, and on the server thread the host window's too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
They are Repurposed's generic jigsaw with a start check of their own:
carried as minecraft:jigsaw with the piece-limited layout and the
check named in horizon:start_checks.repurposed.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Repurposed Structures' city_nether, shipwreck_nether and stronghold_end
types are its generic jigsaw with one more start check each. The
`repurposed` key of horizon:start_checks names the check:
- city_nether: no block from sea level + 20 to sea level + 45 at the
corners of the eight chunks around;
- shipwreck_nether: no solid, filled block in a 33x30x33 grid above the start;
- stronghold_end: land at least 45 above the bottom at 70 and 35 blocks
in each direction and at the start.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Carry used to take every processor written in Java out of its list (#822),
so spawners kept what the template has: an empty spawner in Moog's
templates, a placeholder pig in Better Dungeons'. The new processors module
rewrites the ones with a data form before that:
- Moog's Structure Lib's and Repurposed Structures' spawner randomizing
processors become horizon:spawner_mobs, with Repurposed's rs_spawners list
written in (weight-0 entries left out, as its loader does);
- YUNG's Better Dungeons and Better Desert Temples spawner processors (a
spawner or a placeholder block becomes a spawner of a fixed mob) become
vanilla's rule processor with an append_static block entity, the entity in
26.3's shape (equipment, drop_chances);
- Moog's vanilla loot swap is a no-op while its "replace vanilla" preset is
off, as they ship, and is taken out.
Each is counted in carry-report.json's processors_carried. The game's own
horizon: types stay in the lists.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Moog's Structure Lib and Repurposed Structures register a spawner
randomizing processor in Java: each spawner of the template gets one mob of
a weighted list, picked with settings.getRandom(pos).nextFloat(), and the
spawner NBT they write (Delay ... SpawnRange as shorts, the mob as the only
potential and the next spawn, optional custom spawn rules). With nothing to
pick the spawner becomes the replacement block.
horizon:spawner_mobs does the same from data; horizon-port-pack writes it in
place of theirs. Without it carry drops the processors, and Moog's templates
keep their spawners empty.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A rule processor's position predicate and block entity modifier types are
identifiers: vanilla reads "always_true" as minecraft:always_true. Horizon
matched the full string, so a list writing the short form failed, and with it
every pool naming the list. ChoiceTheorem's Overhauled Village writes
"always_true" in four village pools (plains, plains fortified, halloween), whose
houses then never generated.
The two decoders move out of Rule's decode into PosRule::decode and
Modifier::decode, which normalize the type first.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
With single player on its server thread, the window is a client of the
server. The agent's `command` went out as the owner's chat and returned no
output, so harnesses that read a command's feedback (lan_e2e trials) failed.
Control `give` and `time` changed only the window's copy of the world.
The server now runs them as single player's console does without a server
thread (`as_console`): as the world's owner, in the owner's level. The
feedback goes to the owner's chat and comes back to the agent. `give` runs
`give @s`, and `time` sets the server's clock too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: Merged after #855 and the real verify. block_removal 4000/4000 including 319 mining cases; dispenser, crafter, composter, mace and use_item_on pass.
🤖 Agent: Merged after the real verify passed. Prism's metadata renames native jars (same sha1), so prepare now also looks under Prism's name, still checked against Mojang's sha1, with nothing downloaded. Launcher and horizon-prepare tests pass.
With HORIZON_SPIKES=MS, a dedicated or integrated server tick longer than
MS ms is logged with the server thread's CPU time over the tick and the
tick's spans (tick_mobs, streaming, ...). A tick with much more wall time
than CPU time waited for a lock or for a core.
The spans recorded on a server thread were never taken before, so they
grew for as long as the server ran with HORIZON_SPIKES set.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Every spawn attempt (`NaturalSpawner.getRandomPosWithin`) asks for the
WORLD_SURFACE height of a column. Horizon scanned the column down for it
each time, a chunk-map lookup and a block-name compare per block. Vanilla
reads the chunk's heightmap.
Chunks now keep WORLD_SURFACE: built on the first read, and dropped by a
`set` that can move a column's surface (`Heightmap.update` leaves writes
two or more below it alone). Natural spawning, the spawn rules' bat check
and beacons read it.
The unused `heights` field (never written after `Chunk::empty`) is gone.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ServerLevel.removePlayerImmediately and addRespawnedPlayer: when a player
respawns, every other player tracking it gets RemoveEntities and the next
tick pairs it again as a new entity. The hub only cleared the respawned
player's own tracker, so the others kept the dead body, lying down, wherever
the player went (the host window too, on the server thread).
The in-frame LAN host's death now reaches its guests (LivingEntity.die's
EntityEvent DEATH), and its respawn goes through the same path.
lan_e2e gets a `respawn` check: the host dies and respawns (and on the
server thread the guest too), and the other side's client draws it dead,
then alive. `status` lists the other players a client draws dead.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
test_block_removal_parity gains a "mine" removal: VanillaDispenserOracle
calls the real ServerPlayerGameMode.destroyBlock for a survival player holding
a pickaxe, axe, shovel, sword, shears, a mace or nothing, and Horizon's
dispense fixture calls break_block, the path every player's mining takes (the
single player game, the server thread and LAN guests). It compares the drops,
the spilled contents, experience, sounds, events, statistics and the held
stack. On main it shows every mined container losing its contents (#844, fixed
by #855) and three gaps of Horizon's own, fixed here:
- ItemStack.mineBlock: an item with a tool component counts as used, and wears
its damage_per_block (2 for swords, the mace and the trident), also the mace
and trident, whose tool component has no rules. Shears wear on every block
but fire (ShearsItem.mineBlock).
- DecoratedPotBlock.playerWillDestroy: a pot mined with an item in
breaks_decorated_pots, unless an enchantment prevents it, cracks first and
shatters.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The generator's structure-start cache was one Mutex<HashMap>. Each
generated chunk reads the 289 start chunks around it for its references,
and the server thread reads the starts of the chunks its mobs and spawn
attempts are in (`structures_reaching`). While workers generated, the
server thread spent most of its start lookups waiting for that lock.
The cache is now 64 RwLock shards keyed by a hash of the chunk: a lookup
read-locks one shard. Vanilla keeps the starts on each chunk
(`StructureManager.getStartForStructure` -> `chunk.getStartForStructure`),
so its lookups share no lock either. The first starts built for a chunk
are still the ones everyone keeps.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: Merged after the real verify passed. Waypoints::tick −33% (166 → 111 µs median), with byte-identical packets (530,834, same hash). locator_bar, lan, dedicated_server and server_dimension_players pass.
🤖 Agent: Merged. Entities::tick −37% (112–116 → 71–75 µs median per tick at 40 bots + 570 entities), with the same 14,159,154 packets before and after. 9 covering harnesses pass (server_entities, lan, dedicated_server, server_dimension_players, locator_bar and the entity ones); main has no lan/ changes since.
🤖 Agent: Merged after the real verify passed. Combined check on this branch: fmt, clippy, workspace tests, and the worldgen, structure_density, chunk_generation and world_spawn harnesses pass. modpacks-6's own integration passed 298/298 harnesses with the start oracle at 35/35.
🤖 Agent: Merged. Mesh scheduling while flying at RD32: 0.62/0.72 → 0.36/0.22 ms per frame; at rest with ~12,000 waiting sections: 0.30–0.91 → 0.06–0.22 ms. Same checks as #865.
🤖 Agent: Merged. Terrain vertices go 28→24 B: vertex memory −14%, Metal GPU 0.62→0.54 ms. HD 620 FPS is unchanged (vertex size isn't its bottleneck). Screenshots: the only differing pixels are on water, at main-vs-main noise level, with identical drawn sections. 95/95 covering harnesses on an integration with #866 and #867.
A moving player updates its connection with every other player, three
times a tick (updateWaypoint, then updatePlayer at the position change and
at ServerChunkCache.move), so with 40 players there are about 4800
updates a tick. Each looked the connection up twice and both players up
four to six times. Now an update finds the connection's place once and
resolves the two players once, then checks, updates or remakes it there.
The same packets go out, in the same order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Entity tracking still asks every client about every entity each tick
(vanilla re-checks each pair whenever the player sends a move, so for
moving players it does the same), but the check is cheaper:
- each entity's position, id, range and chunk are read once a tick into
a compact list, not from its EntityState once per client;
- each client looks up each of the tick's chunks once in the chunks it
holds, not once per entity in it;
- an entity with no changes and no client starting to track it skips the
loop over the clients.
The decisions and the packets are the same, in the same order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A value back at its default is left out of an entity's state, so the
tracker never told clients that a player stopped crouching or a mob stopped
burning: Entity's shared flags and pose now go back to 0. On a client of a
server (the host window on the server thread), another player's crouch flag
now reaches their model and name tag.
From #731's server-thread list; lan_e2e's walls check passed in-frame and
failed with --server-thread.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: Merged after the real verify passed. Output is byte-identical (palettes / region files), and 17 covering harnesses passed with both PRs on main 256f7c20 (since then main has only #751, a settings UI change).
🤖 Agent: Merged after the real verify passed. Output is byte-identical (palettes / region files), and 17 covering harnesses passed with both PRs on main 256f7c20 (since then main has only #751, a settings UI change).
Prism's metadata moves a native jar's classifier into its artifact name
(com.mojang:jtracy:1.14.38:natives-windows becomes
com.mojang:jtracy-natives-windows:1.14.38) and Prism stores libraries at their
name's path, so every native jar of a Prism 26.3 install is in
com/mojang/jtracy-natives-windows/1.14.38/jtracy-natives-windows-1.14.38.jar,
not at the version JSON's path. The bytes and sha1 are Mojang's. Preparing
from a linked Prism folder stopped on the first one (jtracy on Windows).
horizon-prepare now looks there when the version JSON's path is missing; the
copy is still checked against the version JSON's sha1. The missing-file hints
no longer tell players of other launchers to install the official one.
Reported by famescare on Discord.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ClientboundUpdateAttributesPacket, as vanilla's ServerEntity sends it: every
client-syncable instance when an entity is paired, the changed ones with the
dirty entity data, and a player's dirty set to the player itself.
- Mobs and armor stands report the instances vanilla would have created
(movement speed, finalizeSpawn's bases and modifiers, the baby modifier,
effects', equipment's and /attribute's), players their LiveAttributes.
- Guests get the per-tick movement speed instance and the creative reach
modifiers (ServerPlayer.updatePlayerAttributes).
- cube::set_size sets the attribute bases as AbstractCubeMob.setSize does,
so a summoned slime has its size's health and speed.
- server_entities compares attributes again, the player's own too, with a
slime and a scaled pig in its scene.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Reads every chunk of a saved dimension, then times SaveDir's save_later and
flush against the same encoding and compression done on one thread, and
checks the written chunks byte for byte.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Sections waiting to be remeshed were one hash set that the render thread
walked section by section every frame, though every check (column loaded,
in view, its 3x3 lit, its region) holds for a whole column. At RD 32 about
12,000 sections wait at the edge of the lit area at any time. They are now
kept per column (a bit per section index), so each frame looks at each
column once and meshes all of its ready sections together.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Relighting block edits and relit chunks collected every changed cell,
sorted and de-duplicated them, then marked the sections around each cell
for remeshing. Vanilla's light engine reports changed sections to
LevelRenderer instead. LightChanges now keeps, per changed section, which
of 27 border classes (low face, inside, high face per axis) its changed
cells fall in; that gives exactly the same dirty sections and light log
without keeping or sorting the cells.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
TerrainVertex goes from 28 to 24 bytes: the UV becomes two u16 in 32768
steps per unit, which hold model UVs (multiples of 1/16 and their halves)
exactly and decode exactly in terrain.wgsl. The position stays a float:
fixed point moves the edges of offset plants and rotated elements by a
fraction of a pixel.
Section vertex memory drops 14% (RD 16: 380 -> 325 MB, arena 503 -> 369 MB);
pixels are unchanged on Metal and Vulkan outside animated water.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: Merged. CI parity passed 94 of 95. The one miss, tooltip_display, passes locally on main with this branch, and so does options. Thanks LookImEcho-Rick!
The writer thread encoded (NBT) and zlib-compressed every chunk of a batch
one after another before writing them. A batch of at least 8 chunks is now
shared out to scoped helper threads (at most half the hardware threads),
which take the next chunk as they finish one. The results are put back in
submission order and written exactly as before, so the region files are
byte-identical: each chunk's bytes depend only on the chunk, and sector
allocation follows the same write order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
palette_ids (behind PalettedContainer::pack, every saved and sent section,
and from_values, every generated one) scanned the palette for each of the
4096 values while it had up to 16 entries, then hashed every value with
SipHash. A run of one value is now matched once, and past 16 entries a
thread-local table indexed by value (block states and biomes are below
65536) replaces the hash map; only the palette's entries are set, and they
are cleared before returning. The output is unchanged: palette in
first-occurrence order and the same ids.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: Merged. tick_lan is 40–55% cheaper and MSPT about 20% lower (40 bots, ~600 entities). The agent ran 95/95 harnesses (--changed) with #842 on main 4b43242a. The orchestrator then ran clippy and the lan tests on current main with both: clean.
🤖 Agent: Merged. light_chunk is 28% faster, light parity shows 0 differing values against vanilla 26.3, and CI parity passed on a merge with main after #824's light changes. Thanks rdbt!
🤖 Agent: Merged. Reviewed against vanilla: landing comes from Entity.move's vertical collision, the landing position is blockPosition(), velocity is zeroed per axis, and placement goes through Level.isUnobstructed with the collision shape. Locally on this head, fmt and clippy are clean, and contextual_collision, level_place, placement, powder_snow_collision and world_movement all PASS. Thanks LookImEcho-Rick!
Their jigsaw structures, the Nether variants too, get a
horizon:piece_limited key for the game's piece-limited assembler:
y_allowance, burying_type, cannot_spawn_in_liquid, the pools whose
fallbacks ignore the bounds, the Nether variants' land search and ledge,
and each structure's piece counts from the mod's rs_pieces_spawn_counts
or msl_pieces_spawn_counts files. use_bounding_box_hack becomes vanilla's
use_expansion_hack. Lithostitched's jigsaw stays vanilla's.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
YUNG's enhanced random spread becomes vanilla's random spread with its
enhanced exclusion zone as horizon:exclusion_zone; the desert and jungle
temples' placements add horizon:avoid_biomes (no ocean or river within 48
blocks of the chunk's middle at y 120); Better Strongholds' adds
horizon:ring_sections. A YUNG's structure the game can start gets
horizon:disables with the vanilla type its mod's mixin turns off; Better
Mineshafts, a Java structure type, leaves vanilla's mineshafts in place.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A jigsaw structure with "horizon:piece_limited" is laid out as
PieceLimitedJigsawManager does (the same code in both mods): a fresh
random seeded by the chunk, the start piece drawn uniformly and moved to
the surface at its center, a volume 120 blocks down and 180 up, children
drawn by weight from the pools' entries in order, pieces capped per
template and some required (retried up to 40 times), pools whose
fallbacks ignore the bounds, y_allowance clipping. After the layout, the
structure's postLayoutAdjustments: the pieces centred on the chunk corner,
then buried (lowest corner, average land, lowest side) or, for the Nether
variants, moved onto the highest or lowest land or a fixed height.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: Merged. CI parity passed 88 of 89 harnesses. The one miss, dedicated_server (a known intermittent failure), passes locally on this head. Thanks rdbt!
EntitySelector.findPlayers and EntityArgument's player targets on a
hosted world or dedicated server: every guest other than the one in place
is Target::Guest, found by name, UUID or selector, and what a command does
to them (give, gamemode, tp, kill, effect give/clear, data) runs with
their state swapped in, their own messages going to their chat. The
console, operators and command blocks can all target them; execute
as/at reads their position, rotation and eye height; a teleport into
another level (execute in, the console's Overworld, a destination in
another level) moves them there.
as_guest splits into enter_guest/leave_guest so that guest turns nest,
and a guest's commands run with the hub back in self.lan (queued per
packet), so they reach the other players (and /kick works for ops).
Item and advancement names and player names in feedback take vanilla's
component shape; /kill on a guest sends vanilla's death message.
Harness: tools/test_server_player_targets_parity.py.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ServerWaypointManager's per-tick work for every moving player copied each
connection out of its row (its icon's style string too), scanned the row
again to put it back, copied the lists of players and waypoints it walked,
and looked players up through SipHash. Connections are now updated where
they are (the same row order and the same packets), the lists are walked
by index and the maps use FxHash.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Every guest had its own copy of what it was last told about every entity,
so each tick compared each entity's data and built and encoded its update
packets once per guest. A level now keeps one ServerEntity state per
entity (ChunkMap.entityMap), whose changes are worked out once a tick,
encoded once and sent to every guest tracking it; a guest's Tracker is the
set of entities it is paired with. The update cadence is the entity's own
tickCount, as vanilla's, and late clients are paired with the data the
others hold. The per-tick copy of every entity state, the id lists and the
packet lists are gone too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Three keys a carried mod's data can hold, which vanilla data never has:
- "horizon:ring_sections" on a random spread (BetterStrongholdsPlacement):
starts only in the odd rings around the origin, with Java's int math.
- "horizon:avoid_biomes" (BetterDesertTemplePlacement and
BetterJungleTemplePlacement): no start when one of the biomes is found
around the chunk's middle block, as findBiomeHorizontal samples it.
- "horizon:disables" on a structure: every structure of the listed types
fails to start, as YUNG's DisableVanilla*Mixin makes
ChunkGenerator.tryGenerateStructure return false; in a set of several
structures the next one is tried, as in vanilla.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged: decoration_order::vanilla_order on flat arrays (RD32 8.14→1.06 ms per call; ticket order unchanged, checked against the old version). Combined on current main with #834/#835: clippy, client + world tests PASS.
🤖 Agent: merged: section uploads through one reused staging buffer and one encoder, as vanilla's SectionRenderDispatcher (RD32 exploring: FPS 43.5→56.0, CPU 23.7→18.6 ms/frame, p99 122→106 ms; sections drawn identical in 12 views). Combined on current main with #834/#835: clippy, client + world tests PASS.
Rust's bool parser takes only true/false, so the fixture failed on Python's
True/False; Java's Boolean.parseBoolean ignores case.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A jigsaw structure with "horizon:yungsapi" (min_y/max_y) assembles the way
YUNG's API's JigsawManager does: the start piece by priority then weight
from the shuffled raw entries, children breadth-first, pool or fallback per
depth, and per-element rules from the element's "horizon:yungsapi": a name
capped by max_count, a depth range, a priority, ignore_bounds and a
condition (true, false, not, all_of, any_of, depth, altitude,
random_chance, rotation, piece_in_horizontal_direction).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla reads structure templates with NbtAccounter.unlimitedHeap()
(TemplateSource.readStructure). Horizon used the default 1,000,000-node
cap, so a template past it loaded as empty: YUNG's Better Ocean
Monuments' start piece (115x64x126 blocks) never started. Limits::templates
keeps the 256 MiB byte cap and drops the node cap, for worldgen and /place.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged: LAN guests and the server-thread game keep each chunk's light and get vanilla light updates (travel CPU −17%, light_chunk −35%; lan_e2e light PASS with and without --server-thread).
🤖 Agent: merged: brewing stands tick in the level's single ticker list, in vanilla's order. perf-ticks-7's four PRs together on main d85a3603: clippy, 861 client + world tests, 19 harnesses (light, lan, dedicated_server, save_roundtrip, world_reopen, disk, datafix ×6, world, server_world_gen, block_entity ×4, chunk_stage) PASS.
🤖 Agent: merged: chunks saved in place in their region file, as vanilla (save into a full region 80.3→7.1 ms). perf-ticks-7's four PRs together on main d85a3603: clippy, 861 client + world tests, 19 harnesses (light, lan, dedicated_server, save_roundtrip, world_reopen, disk, datafix ×6, world, server_world_gen, block_entity ×4, chunk_stage) PASS.
Queue::write_buffer creates a staging buffer for every call: on Metal a
kernel call to make the buffer, then page faults while the mesh is copied
into fresh memory. While exploring at render distance 32 the render thread
made two such writes per uploaded section, plus one for its flat-draw ids.
Section meshes and flat ids now go through their own StagingBelt (4 MiB
mapped buffers, reused once the GPU has read them) into one encoder that
the frame's submission runs ahead of its own commands. A page of the
section arena that grows records its copy into that encoder too, so copies
recorded before the grow reach the grown buffer. A frame that is not drawn
still submits the copies, and more than 64 MiB of them are submitted early.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A remote client (a LAN guest, or the game on its server thread) threw away
the light in each LevelChunkWithLight and relit the column and its
neighbours itself. It now applies the sky and block layers as
ClientPacketListener.applyLightData does, and relights only block changes,
as vanilla's client does.
The hub now sends ClientboundLightUpdatePacket for the light sections that
changed in a column to the players for whom it is on the tracked border
(ChunkHolder.broadcastChanges), so light from blocks a guest doesn't track
reaches it.
horizon-ctl gains 'light X1 Y1 Z1 X2 Y2 Z2', and lan_e2e a 'light' check
comparing the guest's light with the host's.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Lithostitched's add_structure_set_entries and add_template_pool_elements
modifiers become horizon:add_to files under worldgen/<kind>/horizon_added/,
upgraded as the mod's own pools are. Their load conditions (lithostitched's
predicate, NeoForge's, else Fabric's) are checked against the mods ported
along; other modifier types and tag targets are counted, not carried.
ChoiceTheorem's Overhauled Village adds its villages and outposts to
vanilla's sets in Java code; carry writes them with its default config
(small, medium and large villages weighted 10, 4 and 1, outposts 1).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A content mod's structure set or template pool file holding
"horizon:add_to" appends its entries to the sets or pools it names
(vanilla's, another mod's) instead of being one of its own, as
lithostitched's add_structure_set_entries and add_template_pool_elements
modifiers do on NeoForge and Fabric. The content root merges them after
every mod's own files, in mod and file order; a name nobody defines is
skipped. Vanilla data never has the key.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged: worldgen block states rewritten to 26.3's format (1,005 files in 17 mods). Combined check of main + modpacks-5's six: clippy, tests, worldgen structure_density block_properties registry_fluid chunk_stage PASS.
🤖 Agent: merged secure chat as vanilla 26.3's server (chat sessions, signed chat and commands, enforce-secure-profile; secure_chat harness identical to vanilla offline and online; 7 server harnesses pass).
🤖 Agent: merged: the integrated server follows the window's render distance past 32 (fixes#804), and the lan_e2e hazards check now finds the guest by name (the server-thread hunger gap was a test bug).
IntegratedServer.tickServer: the server takes the window's render and
simulation distances when they change ("Changing view distance to N, from
M"), tells every player (SetChunkCacheRadius, SetSimulationDistance), and
each player's chunks follow, up to what they asked for. The integrated
server no longer clamps the view distance to 32, and a remote client sends
its render distance again when it changes (Options.broadcastOptions),
past 32 too. Fixes#804.
lan_e2e: a distance check (--server-thread), and the guest is found by
name in lan status, where the server thread lists the owner too: the
hazards check read the owner's food (the "no hunger from moving" gap).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
tick_block_entity_anims rebuilt a map of every drawn block entity, sorted
it and swept the animation states against it every tick. The renderer now
keeps a revision that changes when its block entities do, and the sorted
ticker list is rebuilt only then. The animation states use FxHashMap.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Chat session updates with their profile key checked against the services'
keys, signed chat and the message chain, last-seen acknowledgements, the
signature cache, signed say/me/msg, enforce-secure-profile, and vanilla's
log lines. Offline servers and players without a chat session chat
unsigned, as on vanilla. tools/test_secure_chat_parity.py compares both
servers with test keys and a mock services server.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Repurposed Structures' and Moog's jigsaws check biomes and terrain height in
a radius and refuse to start in liquid; their spreads keep a minimum
distance from the world origin and away from several other sets. Carry maps
them to vanilla's jigsaw and random spread and writes those rules under
horizon: keys; the game now runs them (start_checks.rs, the spread's
min distance and exclusion sets). Vanilla data has none of the keys.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The game, like vanilla, refuses a processor list holding a type it does not
know, and with it every pool naming the list, so those structures did not
generate at all. Carry now takes the library types out of processor lists
(files and inline ones in pool elements) and counts them by type in the
carry report: the structures generate, undecorated. On ATM10's carried
mods: 1,427 processors of 146 types; pool decode failures 31 -> 0.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Repurposed Structures' and Moog's jigsaws and spreads add rules on top of
vanilla's; mapped to vanilla they were lost, and Moog's spreads structures
1.65 times as far apart as its files say, so MVS alone started 9x as often
as on NeoForge. Carry now writes the start rules under
horizon:start_checks, the minimum distance and the super exclusion zone's
sets (from the mod's structure set tags) under horizon: placement keys,
and scales Moog's spacing. Layout rules (piece clipping, burying) are still
lost and counted in the carry report's rules_lost.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
1.21.1 writes a block or fluid state in worldgen JSON as {"Name", "Properties"};
26.3's codec reads {"id", "properties"} and refuses the old shape, so every
processor list, feature or pool holding one failed to decode. Carry now
rewrites them. On ATM10's carried mods: 1,005 files in 17 mods.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The game and the assets step each parsed facts.json into a JSON tree, then
copied each state into an NBT tree, before building the registry. With a
pack's 214,000 states each step peaked at 816 MB for a 139 MB registry.
The registry now deserializes the rows it needs straight from the file
(each row's state and its air and fluid columns, and the defaults):
registry peak 816 -> 153 MB, assets peak 816 -> 235 MB, RSS after the
assets step 1,344 -> 442 MB. Vanilla: 138 -> 27 MB.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Saving a chunk read every chunk of its region file and wrote them all to a
new file. The writer now writes only the saved chunk, into the first free
run of sectors (RegionBitmap), points the header at it and then frees its
old sectors (RegionFile.write / clear), with one flush per region and batch.
Readers and the writer share one open file per region. A chunk queued again
before the writer reached it is written once, by its newest batch
(IOWorker.pendingWrites).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Brewing stands were the last server ticker with a pass of its own over the
whole block entity map every tick. They now tick in their place in
Level.blockEntityTickers, as BrewingStandBlock.getTicker makes them: in
list order, only in block-ticking range and inside the world border, with
their bottle states written as each one ticks.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged: creaking heart scan only in sections whose palette holds a heart (240→3 ms per 15 s). Combined check of main + perf-ticks-6's six: clippy, tests, 8 harnesses PASS.
🤖 Agent: merged: vanilla FULL ticket square per server player (RD32 finished chunks 8809→4761, spawn CPU 153→98 s). Combined check of main + perf-ticks-6's six: clippy, tests, 8 harnesses PASS.
vanilla_order replays vanilla's ticket scheduler over every chunk in view
each time the player crosses into another chunk, on the render thread. At
render distance 32 that is about 200,000 level updates in SipHash maps:
8 ms per call. Over dense grids covering the square it can touch, the
same replay takes 1 ms. A test checks the order against the hash map
version for several centres and radii.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A content mod's data/<ns>/worldgen/ files were copied into the content root,
where nothing reads them: the worldgen loader reads bundles. The build now
adds a mod's structures, structure sets, template pools, processor lists and
configured and placed features to their bundles, its data/<ns>/structure/
templates to worldgen/structure/<ns>/, and its biome tags to the biome tag
bundle. The loader refuses a biome tag naming what it lacks (another mod's
biomes, a loader's convention tags), so those entries are dropped, as they are
from block and item tags. Biomes, noise and density functions stay out: they
would change vanilla's terrain.
Structure and placement types Horizon does not have stay unsupported, as
before; carry-over maps the common mod ones next.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
26.3's copper and iron vein rules both bind minecraft:overworld/ore_vein/richness
eagerly, so the material stage sampled the same root over the same chunk volume
twice. Bindings of one compiled root now share its buffer; sampling is
deterministic, so the values are identical.
The two vein density reads per solid block went through
DensityVolume::index_of_block, which divides and takes the remainder by the
step on each axis. The stage's volumes are unit-step, so index them directly
(falling back for strided volumes).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A remote client's chat now sends ServerboundCommandSuggestionPacket for an
argument node whose provider is minecraft:ask_server (bossbar ids, data
storage, datapacks, team names, functions...), as ClientSuggestionProvider.customSuggestion
does. The suggestions are done when ClientboundCommandSuggestionsPacket
answers that id; its completions join the local ones (Suggestions.merge).
Replies to older requests are dropped.
The remote tree takes each node's provider from the packet only: a local
provider the server didn't send stood in for the server's before.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A dedicated server draws nothing, like the integrated server, yet its
streaming still meshed every section it lit (Streaming::mesh followed only
options.integrated). It now skips meshing too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
TitleCommand sends the targeted player ClientboundSetTitleText,
SetSubtitleText, SetActionBarText, SetTitlesAnimation and ClearTitles. A
guest's /title (the window's own player on the server thread too) only
changed the server's HUD; it is now sent to the guest. The remote client
showed titles and action bars as a bare line and ignored subtitles, times
and clearing; it now feeds the HUD's titles as ClientPacketListener does
(the action bar of a SystemChat overlay too) and ticks them (Gui.tick).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The hub held SetHealth back once a guest died, so the guest's health
stayed at its last positive value under the death screen. ServerPlayer
sends it every tick it changes, dead or not, and only when the health or
the food level changes or the saturation reaches or leaves zero: the hub
sent it on every change of saturation too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
DistanceManager.PlayerTicketTracker gives every chunk within the server's
view distance (Chebyshev) of a player a PLAYER_LOADING ticket at level 31,
so FULL (33) reaches two chunks past it and FEATURES and TERRAIN one and
two more. A server player's streaming area was a circle of radius
(view distance + 2) * 3 / 2 + 1, finished one further: 2.1x the chunks at
view distance 12. It is now that square, and lan status reports the
server world's chunk counts.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged vanilla-order block entity ticking (new harness: 21,590/21,590 ticks match vanilla; block-entity share of the tick 19.5%→9.6% with 50 bots; 90/90 harnesses locally).
Follow-up to #773 (consume effects on food):
- ApplyStatusEffectsConsumeEffect always draws its float and skips the
effects when the draw is >= probability (#773 skipped the draw at 1.0
and applied on <=).
- on_consume_effects run for every consumable, not only food: milk can be
drunk (clear_all_effects, then the bucket), honey removes poison and
leaves its glass bottle, stews their bowl (UseRemainder).
- The consumable's own timing, sound, animation and particles
(consume_seconds, sound, has_consume_particles) through horizon-world's
Consumable.emit, in vanilla's draw order; FoodProperties.onConsume's
sound and burp; the used statistic; the eat or drink game event after
the effects.
- teleport_randomly (chorus fruit) through horizon-world's verified
TeleportConsume port, with LivingEntity.randomTeleport for the player
(border clamp, entities_can_teleport_to ground, collision and liquid
checks), the teleport game event and sound, level event 2017's
directional particles and entity event 46's portal particles.
- tools/agent_control_e2e.py: a consume scenario (golden apple at full
hunger, milk, honey, chorus fruit).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
CommandSourceStack.broadcastToAdmins beyond a guest on a dedicated server:
- The dedicated server's console tells every operator what its commands
did, as `[Server: ...]`, when `broadcast-console-to-ops` is on
(DedicatedServer.shouldInformAdmins). This covers the game's commands
and the dedicated-only ones vanilla broadcasts (op, deop, ban, pardon,
kick, whitelist, save-all, save-on/off, stop).
- In a world open to LAN, the host, an operator when commands are
allowed, hears of its guests' commands, and op guests hear of the
host's.
- A player's name in the line is their display name, with the `/tell`
suggestion, the entity hover and the insertion (Player.getDisplayName).
New harness test_server_ops_feedback_parity: two vanilla-codec
operators' system chat while the console and one of them run commands,
deop and op, toggle send_command_feedback and save, with
broadcast-console-to-ops on and off. LocatorBarClient can record system
chat (`chat on`). lan_e2e's feedback check also asserts the host is told.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On a dedicated server, `execute in <dimension> run tp @s ...` from a player
moved them inside the level they were in. It now does what vanilla's
TeleportCommand does: ServerPlayer.teleport into the other level, with the
Respawn packet keeping all data and no portal sound, then the feedback. The
players there are tracked to them and they are removed from the ones they
left.
tools/test_server_dimension_players_parity.py joins two players to a vanilla
and a Horizon server and compares, step by step, the players each is sent and
loses and the dimensions it respawns into, while they go to the Nether and
the End by command and through an End portal. LocatorBarClient gains
`players on` and `still NAME` (the position reminder a still client sends).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
`generate-structures=false` (server.properties, stored by #733) and the
world screen's Generate Structures were written to the world's settings
but the generator ignored them. VanillaTerrain now builds its generator
from the world's `generate_structures`
(OverworldGenerator::for_world): without structures there are no starts,
terrain adaptation or structure pieces, as in
ChunkStatusTasks.generateStructureStarts, and `locate structure` and eyes
of ender find nothing (ChunkGenerator.findNearestMapStructure).
The server world gen harness's world without structures now also
compares `locate structure` for a village and a mineshaft and the saved
chunks' structure starts and references.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A Horizon client of a server places what PR #766 sends: primed TNT with
its fuse, falling blocks with their block state and start position,
arrows (critical, in the ground), snowballs, eggs, pearls, fireballs,
wither skulls, wind charges and llama spit, tridents, firework rockets
and end crystals with their beam and base. They interpolate like mobs,
turn with the move packets, and are cleared with the connection, a
respawn or a dimension change.
`status` lists them (`projectiles`), and lan_e2e has a `projectiles`
check: the guest sees the host's primed TNT, end crystal and landed
arrow where the host has them, and a falling block lands on both sides.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
`/summon minecraft:tnt` and primed TNT in a saved chunk were kept as an
unsimulated entity. They now become primed TNT with the fuse (80 when
absent) and motion they were saved with (PrimedTnt.readAdditionalSaveData).
The entity harness runs its scene for a fixed 200 ticks (`tick step`),
so what still changes when the client stops recording (a fuse) ends the
same on both servers. Its scene has two primed TNT (one falling), a
floating falling block, an arrow that lands and an end crystal. Unit
tests cover arrows' position and rotation, fireballs' bundled power and
precise position syncs.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
These entities had no id in Horizon, so a server could not send them to
its clients. Each now has one, given from the entity counter before it is
first tracked, and goes to guests as vanilla's ServerEntity sends it:
- primed TNT with its fuse (sent each tick it changes) and falling blocks
with their block state and start position; both sync their exact
position while they touch a block (requiresPrecisePosition);
- arrows and tridents send their position and rotation every update
(AbstractArrow), with their flags, pierce level, in-ground state,
loyalty and foil; their owner is the spawn data;
- fireballs, wither skulls and wind charges bundle their velocity with
ClientboundProjectilePowerPacket (new in horizon-protocol);
- snowballs, eggs, pearls, llama spit, shulker bullets and splash potions;
- firework rockets with their item, the player they boost and whether
they were shot at an angle; end crystals with their beam and base.
The entity harness's scene gains primed TNT, a falling block, an arrow, a
snowball and a trident that land, a fireball, a firework and a crystal.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ClientPacketListener.handleAddEntity for the objects a server tracks: a
Horizon client of a server (a LAN host, a dedicated server, single
player's server thread) now places the boats, minecarts, armor stands,
item frames and paintings it is sent, where the server says.
- Spawn: carts by kind, boats and rafts by wood (with a chest), armor
stands, frames and paintings at their block with the direction from the
spawn data.
- Entity data: frame items and rotation, painting variants (a registry
holder), armor stand flags and health, boat paddles, vehicle hurt.
- Equipment: what an armor stand wears and holds.
- Moves interpolate like mobs'; rotation and ground state follow.
- They are cleared with the connection, a respawn or a dimension change.
`status` lists them (`objects`), and lan_e2e has an `objects` check: the
host summons them, the guest sees each as the host has it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A server sent its clients only mobs and dropped items. Boats, every kind of
minecart, armor stands, item frames, paintings and the entities Horizon keeps
without simulating are now tracked as ServerEntity does: the AddEntity data
(hanging entities at their block, with their direction), the synched data that
differs from the defaults, equipment, passengers (ClientboundSetPassengersPacket
at pairing and on change; a passenger syncs only its rotation and resyncs its
position when it dismounts), and the type's update interval and range.
Also vanilla-exact on the way, as the new harness found:
- mobs pair with their health (always non-default) and Mob flags (no AI, left-handed);
- /summon without NBT summons any type from an empty compound;
- armor stands move before gravity (LivingEntity.travelInAir), so a stand at rest
keeps vanilla's -0.0784 vertical motion.
tools/test_server_entities_parity.py compares, from a vanilla-codec client, what
vanilla's dedicated server and Horizon's track for a scene of 20 entities.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
scan_for_hearts read all 4,096 blocks of every non-air section of each new
chunk on the server thread to find creaking hearts. A section whose palette
holds no heart state (PalettedContainer.maybeHas) cannot have one, so it is
skipped. Same block entities, same order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged the six LAN guest fixes for #696 (#722–#727), combined on main and checked locally (clippy, tests, parity lan/dedicated_server/mob_ai, lan_e2e clicks/sleep/push/eat/xp).
These entities were drawn every frame wherever they were in the loaded
world: every armor stand, item frame and painting of every loaded village
or base went through its model, item and per-vertex occlusion test even
far behind the camera. EntityRenderer.shouldRender now decides, as for
mobs and minecarts:
- Entity.shouldRenderAtSqrDistance from the current position: 64 times the
bounding box's mean edge times the entity distance option (boats about
71 blocks, primed TNT 63, paintings by size); item frames 1024 blocks;
armor stands 256 times the box's mean edge whatever the option (4 for a
marker's empty box); tridents ten times the default (AbstractArrow);
firework rockets 64 blocks.
- the interpolated culling box grown by 0.5 in the view; tridents grow it
by 1.5 more (ThrownTridentRenderer), armor stands with a dragon head by
0.5 (LivingEntityRenderer), item frames holding a map use the block-wide
box.
The fishing bobber is unchanged: vanilla never frustum-culls it
(FishingHookRenderer.affectedByCulling) and Horizon only draws the
player's own, always within its 64 blocks.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The section occlusion search (SectionOcclusionGraph) ran on the render
thread whenever a section was meshed or the camera entered another
section: at render distance 32 that is a breadth-first walk over about
124,000 cells, and while exploring it ran nearly every frame.
As in vanilla 26.3, the full search now runs off the render thread. A
worker keeps its own copy of every section's face connectivity, fed the
changes with each job, and searches over dense arrays around the camera.
Frames use the last finished graph and never wait for the next. The search
starts at the top of the frame, as soon as the camera is known, so it
usually lands before the frame builds its visible list. A section meshed in
between propagates into the graph the frame has (vanilla's
runPartialUpdate), so a wall that opens shows what is behind it at once; a
section that closes goes at the next full search. A graph that never
reached the camera's section (after a teleport, say) is not used: until the
next one lands every section in the frustum is drawn.
The search itself is unchanged: the same box, steps and first-entry rule,
checked cell for cell against the previous code in a test.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
light_chunk's sky and block spreads read LightTables::cells for every
neighbour, and the fill and both spreads dominated lighting. Pack each
cell into two bytes (occlusion type and dampening) with a per-type
blocking mask, step neighbours by linear offset with an axis bounds
check, and precompute the per-column present table.
1156-column probe: 1.81 -> 1.31 ms/col (~28%). Vanilla light parity is
unchanged (1,179,648 sky/block values, 0 differences).
entity_motion_events looked up each moving mob in a SipHash map every
tick. The keys are mob ids the server gives out, so the fast hasher is
safe there.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Their particles, ambient sounds and the turning display mob are for a
window. A dedicated or integrated server has none, yet made a display
mob for every spawner it ticked.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The list kept its slots for the whole tick; sweeping it every tick cost
two hash lookups per ticker. Now it is swept only when a slot was freed
or a block entity was taken out and not brought back.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
BlockSection::get and set check the section index and the state id once,
then read and write through new PackedStorage::read/write, which skip the
storage-level bounds and value checks and the Result that get/set keep for
external callers. A section index is below 4,096 and a palette id always
fits the width bits_for_palette chose; both methods debug_assert that
contract. Used for the hot read, the hot write and the growth re-encode.
Benchmark (ignored tests, best of alternating-order rounds, fast profile):
BlockSection::set 13.15 -> 9.12 ns/write (1.44x), get 2.33 -> 1.45 ns/read
(1.61x). See docs/performance/paletted-container-access-results.md.
World::block_entities keeps vanilla's Level.blockEntityTickers beside the
map: a slot when a block entity whose state has a ticker is added, kept
when it is replaced, freed when it is removed, unloaded or its state loses
the ticker; a loaded chunk's in its block entity map's order. The tick
visits that list once, in block-ticking chunks within the world border,
instead of scanning the whole map once per kind.
New harness block_entity_tickers: real LevelChunks of a server level
against BlockEntities.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged by Ludovic's AI agent. CI parity passed (harness paths stay synchronous). Smoke test of the new path on current main (macOS, --offscreen, normal launch): resources loaded on the worker in 9.3 s, renderer swapped, title screen up and running for 60 s, no panic. Thanks @rdbtCVS!
The NBT tables stream, the fluid tables reuse the game's facts and the content
root is kept between starts (#753, #754, #756). Horizon with ATM10's 40 data-tier
mods against the same mods on NeoForge 21.1.259. Worldgen carry-over and what it
does not cover yet (#763, #764). Chipped's harness row with its self-drop tables
(#762). The overlay cache line said it was cached; it is now.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
BlockSection::set resolved the old cell through get and then probed the
section palette three times per non-trivial write: keying the old state by
id, testing the incoming state and finding it. The slot at the written index
already is the old state's id, so read it straight from storage and resolve
only the incoming state, with a single map probe. Palette reuse, free slots
and growth are unchanged, so palette, counts, free list and NBT are identical.
PalettedContainer::unpack and unpack_u16 matched the palette variant inside
the per-cell loop; dispatch Single/List/Global once before it.
Measured against copies of the pre-optimization code (ignored tests):
BlockSection::set 12.73 -> 11.37 ns/write (1.12x), unpack list 1.956 ->
1.860 us/call (1.05x), global 2.255 -> 2.172 (1.04x). See
docs/performance/paletted-container-access-results.md.
set_matches_baseline_under_random_edits drives both implementations over the
same 50k-edit streams for 1/2/16/17/24/64 distinct states, with periodic
whole-section collapses, and asserts equal results, bits, palette, counts,
free list and NBT.
Structure mods build on libraries whose structure, placement and pool element
types are vanilla's with extra rules: Lithostitched's, Moog's Structure Lib's
and Repurposed Structures' jigsaw structures, their advanced random spreads,
versioned and ocean-bottom single pool elements. Carry rewrites them as
vanilla's types; the extra rules (no start in liquid, a minimum distance from
the origin) are lost, so a structure may start where the mod would not. Types
with no vanilla counterpart (YUNG's, Formations', the nether variants) are left
as they are and counted in the carry report.
The structures' biomes name the loader's convention tags (#c:is_plains). fetch
now also gets the pack's NeoForge jar from NeoForge's Maven repository, checked
against its SHA-1, and carry copies the convention biome tags each mod's
structures use, with those they include, into the mod.
ATM10's 42 carried mods: 1,061 files rewritten, 103 convention biome tags.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Chipped registers its blocks' loot in code: its jar ships 7 loot tables for
about 7,000 blocks, so carried over they dropped nothing, while in the NeoForge
oracle every one drops itself. When a declared block has no loot table in the
jar and the oracle shows vanilla's dropSelf (one of itself with every tool,
nothing only where the state needs a tool it did not get), carry writes that
table in 26.3's shape. When only the states with one property value drop it
and the rest drop nothing (a door's lower half), the entry gets vanilla's
match_block condition on that value, as createDoorTable writes. Chipped:
6,843 tables added.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The nether-portal section precheck from #661 answered "does this section's
palette hold a portal?" per section, per tick. Keep the answer per resident
chunk in `portal_sections`, keyed by a weak handle: an edit or a chunk
replacement rebuilds it, and it is dropped when a chunk unloads and cleared
when the block facts change, so a portal block always forces the sweep. The
entity sweep now finds portal blocks without the per-tick map.
This is a scan microbenchmark, not a whole-game claim: steady improves 2.4x
to 17x over the fixtures and the run parity-checks the cached and per-tick
answers. Rescanning every chunk every tick (cold, reload, edit_all) is slower
because a miss reads all 24 sections of a chunk; see
docs/portal-scan-benchmark.md.
Halves the CPU of deflating a chunk column and makes small packets about
20 % cheaper. The compressed bytes change; what they inflate to doesn't.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
getFiddledDistance restarts a six-step LinearCongruentialGenerator mix for
each of the eight quart corners. Its first two steps depend only on the X
and Y corner offsets, so quart_at_block now mixes the two X states and the
four XY states once, then runs the remaining steps across the eight corners.
next(m, salt + offset) == next(m, salt) + offset, so a coordinate folds into
its step with one addition.
Bit-identical to BiomeManager.getBiome and getFiddledDistance; a reference
oracle test pins the equivalence over a dense grid and wrapping extremes.
🤖 Agent: merged by Ludovic's AI agent: perf-server-5. Before/after in the PR body; agent parity 87/87 stacked; orchestrator combined check with the perf-ticks-5 stack on current main: clippy clean, client+protocol tests pass (one flaky world-delete test, #749), dedicated_server/lan/chunk_stage/block_entity_storage PASS.
🤖 Agent: merged by Ludovic's AI agent: perf-server-5. Before/after in the PR body; agent parity 87/87 stacked; orchestrator combined check with the perf-ticks-5 stack on current main: clippy clean, client+protocol tests pass (one flaky world-delete test, #749), dedicated_server/lan/chunk_stage/block_entity_storage PASS.
resource_root built the content root again on every start. With ATM10's 40
carried data-tier mods that is about 25 s on a quiet machine (50 s on a busy
one) and the biggest memory peak of the start, above 3 GB of footprint. The
root now records a stamp of what it was built from (this build of the game,
the base's manifest and rewritten tables, every mod file by size and time) and
later starts with the same stamp use it as it is.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
FluidResources decoded gameplay-facts.json a second time for three per-state
columns the game's BlockFacts already has. With ATM10's carried mods that file
is 211 MB: the second decode took a few seconds and peaked about 500 MB above
what the game keeps. MotionResources::load_with_facts takes the game's facts;
load still decodes them itself for tests and tools.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged by Ludovic's AI agent: verify passed, parity proven (CI or perf-frame-2's local run), before/after numbers and screenshots in the PR body.
🤖 Agent: merged by Ludovic's AI agent: verify passed, parity proven (CI or perf-frame-2's local run), before/after numbers and screenshots in the PR body.
🤖 Agent: merged by Ludovic's AI agent: verify passed, parity proven (CI or perf-frame-2's local run), before/after numbers and screenshots in the PR body.
The title's GUI atlas and its panorama both come from the resource root and
are only needed to draw the title and menus, but they lived apart: the atlas
on Game and the panorama inside PackResources. Gather them into UiResources so
Stage 2 can load the title bundle before the world bundle. No behaviour change.
Part of #712 (stage 2a).
A Horizon game with content mods builds its content root and block behaviour
snapshot on its first start, in its game folder, and reuses them afterwards.
perf_bench starts every game from an empty folder, so it only measured first
starts. --warm starts the game once to the title screen first, unmeasured.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged by Ludovic's AI agent once CI parity passed; screenshots identical to main (perf-frame-2), CPU frame time about 40% lower on a chest field.
occlusion.nbt, cached-collision.nbt, cached-support.nbt, block-properties.nbt and
movement-properties.nbt each carry the identity of every block state, and every
loader parsed them into an NBT tree first: 240 MB per table for ATM10's 153,511
states, about 1.4 GB allocated and freed while the game starts.
nbt::read_named_streaming hands a named list to a visitor element by element.
registry::BlockTableFile reads a table that way, checking each state against the
registry as it comes, and loaders build from a checked registry::BlockTable.
from_nbt stays for trees built in memory (tests, fixtures).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Each player's writer thread compressed every packet it sent, so a chunk
column sent to fifty players was deflated fifty times; compression was
most of the writer threads' time and about half of the server's CPU.
Bodies of 8 KiB or more (chunk columns, mostly) now go through a shared
cache of compressed frame contents keyed by their length, two
independently keyed 64-bit hashes and the threshold, holding the newest
32 MiB. The bytes on the wire are the same: frame::encode is split into
frame::contents and frame::prefixed, and ConnWriter::send_contents
sends a frame from its contents.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The synchronous startup loaded every resource before the first window
existed, so the game hung on a blank screen for seconds. An ordinary
launch now enters the event loop at once, opens the window, and draws a
loading screen with the current stage and a progress bar while a
resource-loader thread runs the existing load sequence. When it finishes,
audio is created on the render thread and the real renderer, game and
title screen replace the loading screen on the same window.
Harness, dedicated and scripted runs (--frames, --world-frames,
--benchmark, --control, --dedicated, HORIZON_MENU_STYLE=classic and the
parity fixtures) keep the fully-synchronous path unchanged, so their
output stays byte-identical. title_shown still fires on the real title,
not on the loading screen.
🤖 Agent: merged by Ludovic's AI agent once verify passed (the earlier failure was a flaky client test #721 doesn't touch; main passes it). Launcher tests run locally on main + #719 + #721: all pass. Thanks @rdbtCVS!
🤖 Agent: merged by Ludovic's AI agent. CI doesn't build the launcher on PRs, so launcher tests were run locally on main + #719 + #721 (cargo test --locked, macOS): all pass. Thanks @rdbtCVS!
Six scenes from one fixed-seed world (seed 8675309): a plains village among
savanna hills, sunset over it and the river, moonrise over its lit square, a
crimson forest with a zombified piglin, tropical fish over a warm-ocean coral
reef, and survival at dusk with a zombie and the full HUD. Plus Horizon's own
title, Singleplayer, Video options and pause screens.
Captured with horizon-ctl screenshot from the offscreen client (fullscreen
2560x1286 on the Mac mini, default graphics settings, GUI scale Auto), scenes
cropped to 16:9 and menus to 16:10 (left-aligned) and scaled to the old sizes
(1600x900, 1280x800). The scene sources are JPEGs in docs/media again, the
names scripts/assets.mjs already reads; the README's WebP copies are the
rebuilt website media. Alt texts and LEGAL.md follow the new pictures.
The launcher screenshots are unchanged: the Mac mini's session is locked, so
no window capture works.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged by Ludovic's AI agent. CI parity passed on the head; combined with current main (and #680) locally: workspace clippy clean, horizon-client 826 + horizon-world tests pass, mob_ai/save_roundtrip/spawner PASS. Thanks @Online5277!
A Prism or MultiMC instance keeps no jar of its own: the client sits in the
launcher root's shared libraries, two or three folders above `instances/`,
and the game is downloaded only once the instance is launched. Choosing
`%APPDATA%\PrismLauncher\instances` (or an instance's `minecraft`/`.minecraft`
folder) as the Minecraft folder failed Prepare when the jar was not there yet,
and the error only listed every launcher without saying what to do.
resolve now also looks in the launcher root above an `instances` folder, so a
folder deeper than the four walked above the chosen one still finds the
libraries. When nothing is found and the chosen folder is inside a launcher
Horizon knows, the error names it and says to launch the instance once so it
downloads the game.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
🤖 Agent: merged by Ludovic's AI agent after local checks on a6592eb1: workspace clippy clean; horizon-client/world/protocol tests pass; lan, chunk_stage, dedicated_server, block_entity_storage, chunk_generation PASS.
Settings gains a one-line Game arguments field, kept as `game_args` in
settings.json and appended to the client's command line after the
launcher's own, so a player can pass things like `--server-thread` or
`--render-distance 40`. Quotes group a word with spaces in it.
`horizon-launcher game-args [<line>]` shows or sets the line,
`play --arg <ARG>` passes one for a single run, and the MCP launch_client
tool takes an args array. The same splitter feeds all three.
Closes#716.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
tick_hoppers read the block under every plain container block entity
each tick (chests, barrels, shulker boxes, dispensers, droppers) to find
the hoppers. A hopper's block entity is the only five-slot one
(BlockEntity::for_block, and BlockEntity::id already names it the
hopper), so only those are looked up now; a position still needs a
hopper block to tick.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On a LAN world or a server, guests never got experience: orbs followed and
were taken only by the host's own player (none on a dedicated server), and
guests weren't sent orbs at all.
- Each orb follows the nearest living, non-spectator player within 8
blocks, the host or a guest (ExperienceOrb.followNearbyPlayer).
- A guest takes an orb as Player.aiStep does: one of those in its box grown
by 1, 0.5, 1, at random, at most every 2 ticks (takeXpDelay), repairing
with Mending first; the rest goes to the guest's experience (sent with
SetExperience as before), the level-up sound every fifth level.
TakeItemEntity goes to the players around, the pickup sound plays at the
orb.
- Orbs have entity ids and are tracked to guests (AddEntity, DATA_VALUE).
The client shows a server's orbs, moves them with the server's packets and
flies them to their collector.
tools/lan_e2e.py xp: three 7-point orbs near the guest: it sees three and
ends at level 2 (21 points); the host gets none.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A mob's push of a guest went to the guest as SetEntityMotion carrying the
server copy's velocity, which nothing damps: every tick of contact added
to it and replaced what the guest's client was doing. Vanilla pushes
players on their own client only (EntitySelector.pushableBy): the server
no longer sends those pushes, and a server client now pushes its player
away from every living entity it touches (Entity.push, team rules, no
spectators, bats or passengers of the same vehicle).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Minecarts were drawn every frame wherever they were: at render distance 32
every chest minecart of every loaded mineshaft went through the rail
lookups, display block and model, and the per-vertex terrain occlusion
test. Vanilla draws a cart only within 64 times its bounding box's mean
edge (about 57 blocks at entity distance 100%), measured from its current
position, and only when its interpolated box, grown by 0.5 and stretched
up by the display block's offset, is in the view
(EntityRenderer.shouldRender, AbstractMinecartRenderer
.getBoundingBoxForCulling). The cart the player rides always draws.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On a LAN world or a server, other entities' heads flickered between two
directions. The client wrote ClientboundRotateHeadPacket's head yaw into
the entity's yRot and the move packets' yRot into the body yaw, never moved
the head's previous yaw and pitch on, and never ran the body rotation, so
each frame interpolated the head from stale values.
As vanilla: move packets interpolate the position, yRot and xRot over three
ticks (LinearInterpolationHandler), RotateHead the head yaw over three
(lerpHeadTo), each tick keeps the previous head rotation for the frames, and
the body follows (a mob's BodyRotationControl, a player's walk direction
within 50 degrees of the look). AddEntity sets yRot and xRot and puts the
body and head at the packet's head yaw (recreateFromPacket).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On a LAN world or a server, a guest's clicks that went through the menu
engine (placing a carried stack, quick craft drags, double clicks, number
keys, Q) ran only on the guest's own copy and were never sent: the item
showed in the chest or furnace, the server still had it in the inventory,
and the next menu brought it back.
menu_ops now sends every click of the server's open menu as vanilla's
MultiPlayerGameMode.handleContainerInput does: the engine predicts it on
the client's copy, then a ContainerClick goes out with the slots it changed
and the carried stack after it (a drag is its three kinds of clicks). The
plain inventory and crafting table are predicted too, the Shift double click
sends each QUICK_MOVE, and a prediction no longer drops, crafts or makes maps
on the client (the server does).
tools/lan_e2e.py clicks: real mouse presses put a stack into a chest and a
furnace, drag planks over the crafting grid and craft; the server has each.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
At the end of each tick the game thread handed every player's held
packets to that player's writer thread, waking each one: a system call
per player, and on a busy machine often the rest of the core's time
slice to the woken writer. Now the batches go to one post thread in a
single send, and it wakes the writers. A packet sent to a player while
its batch is still with the post thread goes the same way, so each
player's packets keep their order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ChunkEncoder::body_in writes the LevelChunkWithLight body that
Clientbound::encode would give, into one buffer sized up front: the
cached section bytes and light layers are copied once, uniform light
layers are filled in place, and the heightmaps are shared rather than
cloned. Before, each chunk sent built a packet (growing its buffer,
cloning every 2048-byte layer), then encode copied all of it again.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A server client sent UseItem and never started the use itself, so food
showed no eating animation or crumbs, and holding right click sent a new
UseItem every 4 ticks. Like MultiPlayerGameMode.useItem, the client now
starts eating (Consumable.startConsuming), counts the use down for the
animation and its particles, and starts no other use until it ends; the
server still eats the item. Looking at another player no longer stops
the item from being used (Player.interactOn passes).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The bed's sleeping state, timer and respawn point were the host's fields,
so a guest clicking a bed put the host to sleep and moved the host's
spawn. as_guest now swaps them with the guest's own; the guest gets its
sleeping pose and bed (entity data 14), wakes on STOP_SLEEPING with the
WAKE_UP animation, and the night skips only once enough players sleep
(players_sleeping_percentage). Guests respawn at their own bed, else the
world spawn. Server clients lie in the bed from the synced position.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
wgpu replays an indirect multi-draw on Metal as one draw call per entry, so
every frame encoded a draw per visible face range and translucent section
(9,300 at render distance 32). Where multi-draw is emulated (Metal) and a
frame has at least 4,096 section draws, a compute pass now copies the
frame's section draws into one index list, in draw order, and each run of
sections in the same arena page is one draw call: 9,309 draw calls become
about 750 at RD32. Smaller frames keep the per-section draws, where the
list's GPU cost (about 0.4 ms on an M4 Pro at RD16) outweighs the CPU saved.
The picture is unchanged by construction: the same triangles from the same
vertices in the same order through the same shader code. Only where a
vertex finds its section's origin changes, from the instance attribute to
a table: a section's vertices fill whole 256-vertex blocks of its page, each
block names the section, and the section's camera-relative origin is
computed on the CPU exactly as before.
Vulkan keeps the native multi-draw. HORIZON_FLAT_DRAWS=1 draws every frame
flat (any backend that supports it), 0 never; the control command
`section-draws flat|indirect` switches in a running game for A/B checks.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Apple GPUs time only pass boundaries (no TIMESTAMP_QUERY_INSIDE_ENCODERS),
so the GPU timer was off on every Mac and benchmarks reported no GPU time.
Where only pass boundaries can be timed, the frame is now bracketed by two
empty compute passes that write the timestamps.
HORIZON_SPIKES also names the renderer's parts (occlusion, visibility,
encode, submit), and the status reply counts the frame's section draws and
the draw calls that submit them.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
TickCtx gained `collision` here; main's two newer initializers
(lan_actions respawn, server_travel's server_ctx) now pass it too, as
every other one does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
DedicatedServerProperties.WorldDimensionData: a world that doesn't exist
yet gets level-type's preset (flat or normal; an unknown name warns and
falls back to minecraft:normal), a flat world's generator-settings
(FlatLevelGeneratorSettings: layers, biome, lakes, features; a layer that
doesn't parse is logged and left out, and settings without layers use the
preset's classic ones) and generate-structures, written to
world_gen_settings.dat before the world opens, with vanilla's log lines.
Horizon ignored all three and always made a normal world.
New harness server_world_gen compares world_gen_settings.dat and the lines
logged with vanilla's for five server.properties.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On a server, single player's server thread included, the inventory
had no recipe book: the window ignored the recipe book packets, and
the server ignored a guest's ServerboundPlaceRecipePacket.
- horizon-protocol decodes the recipe book's packets and the recipe
displays they carry: RecipeBookAdd / Remove / Settings,
PlaceGhostRecipe and the serverbound PlaceRecipe. Every 26.3
`recipe_display` and `slot_display` type is decoded (shaped,
shapeless, furnace, stonecutter and smithing displays; empty,
any_fuel, with_any_potion, only_with_component, item, item_stack,
tag, dyed, smithing_trim, with_remainder and composite slots).
- The remote client keeps the book the server sends (ClientRecipeBook:
added, highlighted, removed, replaced), shows the unlock toast,
applies the open and filter settings, and counts the menu the
server last sent for the craftable filter. Clicking a recipe sends
PlaceRecipe; a ghost recipe from the server fades in.
- The server handles a guest's PlaceRecipe in the inventory grid and
the crafting table (ServerPlaceRecipe on the menu engine, for known
recipes only) and answers with the slots, or a ghost recipe when the
items are missing.
- The screen state reports the recipe book's buttons.
lan_e2e.py gains a `recipes` check: the guest gets oak logs, opens the
book, clicks the planks recipe, and a log lands in the grid.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
/give, /clear, /gamemode, /experience, /advancement, /playsound and the
whisper-to-yourself lines printed the host's placeholder name for the
acting player. They now use the acting player's name (a guest's own while
their command runs), as target_name and the scoreboard do.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
While a guest's command or action runs, the score holder, team, display
name and death and kill criteria of the acting player use the guest's own
name (ScoreHolder.getScoreboardName), as on vanilla's server.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
CommandSourceStack.broadcastToAdmins: on a dedicated server, a guest's
command success (sendSuccess with broadcast) is logged on the console as
"[Name: message]" when log_admin_commands is on, and sent gray and italic
to every other operator when send_command_feedback is on. The queries and
lists vanilla sends with broadcast false (commands/admin.rs) stay with the
player who asked. Lines are queued while the guest acts and sent once the
command returns.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ditto without --norsrc --noextattr --noqtn stores an AppleDouble ._ entry
beside every file that has extended attributes, so the macOS release zip
held hundreds of them (horizon-macos-arm64/._horizon-ctl, tools/._*.py).
Package it as packaging/macos/package.sh already does, and fail the job if
any ._ or __MACOSX entry is left.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The window dropped ClientboundCommandsPacket ("argument nodes are not
decoded"), so on a server, single player's server thread included,
the chat was a bare input line without completion, colouring or usage.
- horizon-protocol decodes argument nodes: the 26.3
`command_argument_type` ids and every parser's properties
(ArgumentTypeInfo.deserializeFromNetwork), plus the custom
suggestion provider.
- command::network turns the tree into a Dispatcher. A node at a path
the client's vanilla tree also has takes that node's argument type
and suggestion provider; any other node takes the type its parser
and properties describe.
- On a server the chat is ChatScreen over that tree, so Tab completes,
arguments are coloured, usage and errors show, and Enter sends the
command to the server. Player-name suggestions come from the
server's player list. The old bare input line is gone.
- The screen state reports the chat input (`chat_input`).
lan_e2e.py gains a `commands` check: the guest completes `/gamem` to
`/gamemode` and `/gamemode cr` to `/gamemode creative` with Tab, and
the command runs.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On a server (the integrated server thread and the dedicated server)
the host has no player of its own, so effects that only touched the
host's player reached nobody; on a LAN world they reached only the
host. Beacons, the warden's and shriekers' darkness, the elder
guardian's mining fatigue, splash and lingering potions' splash and
the raid victory's Hero of the Village now run for each player in
the level, guests through `as_guest`.
The elder guardian's curse sends the guest the GUARDIAN_ELDER_EFFECT
game event, which the remote client now shows (the apparition and
the curse sound).
lan_e2e.py gains an `effects` check: a speed beacon next to both
players must give both of them speed.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
With an explicit Present Mode (Mailbox, Immediate), the VSync option did
nothing: the setting replaced vanilla's choice even with VSync on, so a
window kept running uncapped. Vanilla's VSync means Fifo, in a window and in
fullscreen alike. Now VSync on always presents with Fifo, and the Present
Mode setting chooses only while VSync is off (Auto keeps vanilla's choice:
Immediate, else Mailbox, else Fifo). The choice is one tested function,
window::chosen_present_mode.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The LAN chunk encoder's section cache (#623) and dimension range (#668)
are kept; a cached section's wire bytes use the registry's palette width,
set once per process before any encoder exists, so the cache stays keyed
on the section alone. The sculk scan's palette check takes the u32 state.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
LAUNCHER_PATHS and launcher.yml's push paths now list horizon-bench,
horizon-friends, horizon-mods and horizon-port-pack too, the launcher's
path packages per cargo metadata. A change there (like the horizon-mods
dependency that left launcher/Cargo.lock stale) then builds and releases
the launcher instead of going unnoticed.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
horizon-port-pack now depends on horizon-mods (1f1d5a5a), which the
launcher's own lock file didn't have, so every --locked build of the launcher
failed.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A server (the dedicated server and single player's server thread) now
holds a level per dimension, like vanilla's `MinecraftServer.levels`. Each
server tick runs every level with players in it, one at a time, by swapping
its fields into place (game/levels.rs), so the world tick code runs
unchanged. Before this, a portal did nothing while two or more players were
online.
- Levels: the parked levels keep the world, streaming, save, block ticks,
entities, block entity state, raids, sculk and creaking scans. The
Overworld stays loaded; another level is saved and closed after 300 ticks
with nobody in it. Saving saves every level.
- Hub: `tick` is split into `poll` (the network, once per tick),
`tick_level` (the packets, chunks, block changes, containers and entities
of the players in the level in place) and `end_tick` (latency). Each guest
has a dimension; time, weather, sounds and events reach only the players
in their level, and the clock and weather last sent are kept per level.
- Travel: every player has their own portal state; a portal moves only that
player into the destination's level (`ServerPlayer.teleport` packets as
before). Respawning brings a player back to the Overworld's level.
- Player data saves and loads each player's dimension, and a returning
player is placed in their own level.
- tools/lan_e2e.py --server-thread --check portals: the guest goes to the
Nether while the host stays, then the host goes to the End while the guest
is in the Nether, and both come back.
The in-frame game keeps its one level that follows its player.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On Windows the launcher links another launcher's client jar with a hard link
(a copy across volumes) and its libraries and assets with a junction, since
symlinks need developer mode. canonicalize() of a hard link is its own path,
so the sources tests, which compared canonical paths, failed on Windows and
blocked every launcher release since launcher-v0.4.0. They now check that the
linked jar is the launcher's file (a byte written through one shows in the
other), which holds for a symlink and a hard link alike.
The modpack profile test compares paths as paths rather than strings ending
in "a/carry".
mklink /J's paths are always quoted: Rust only quotes arguments with spaces,
and cmd reads an unquoted & or ^ in a user folder as its own syntax.
CI runs every launcher test target, not just the library's.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The first criterion trigger of a world's first tick loaded the
advancement tree: about 1,400 JSON files read and parsed on the game
thread, about 60 ms of a 125 ms first tick. open_world now starts that
read on a thread, and the first advancement_tree() takes its result,
the same tree as before.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A guest's action runs on the host's game with the guest's state swapped in
(`as_guest`). The swap restored the open screen but not what opening one
does to the window: `capture(false)` released the host's pointer, so the
cursor showed and the camera stopped whenever a guest opened a chest, a
crafting table or any other menu. `keys.clear()` dropped the keys the host
was holding, and a sign, book, command block or mount menu the guest opened
was drawn on the host's screen.
- `capture` leaves the window alone while a guest acts.
- `as_guest` keeps the host's held keys, sign editor, book, command block
screen and mount menu, as it already kept the open screen.
- `state` reports `pointer_grabbed` (mouse movement turns the camera).
- tools/lan_e2e.py: `chest` and `crafting` check that the host keeps its
screen and its mouse look; a new `sign` check places a sign as the guest.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The hub tests every player's sent-chunk set for every entity and every
column of the view each tick, and each player's tracked-entity map for
every entity: tens of thousands of SipHash lookups a tick with 50
players. Both now use the Fx hash. Their iteration order was random
before and carries no meaning.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Every chunk packet scanned all 4,096 blocks of each section holding a
block entity block to list them, though the section's wire bytes were
already cached. The cache now keeps those blocks (index and state, in
scan order) beside the bytes, made in the same pass that encodes them.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
encode_in asks whether each of the nine surrounding columns' 24 sections
holds anything, to decide which light layers exist: 216 cache lookups,
each locking the cache and SipHashing a section's address. The column's
sections now share one lock, and the cache uses the Fx hash.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The hub looked up an entity's tracking range (a binary search over type
names) for every player and every entity each tick: about 87,000 string
comparisons a tick with 50 players. The ranges are now worked out once
per entity per tick; the visibility test is the same.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The generation scan walked every area's square of columns, so with 50
players standing near each other it tested about 100,000 columns a tick
(each against every player's range) to find the few hundred around them.
The squares are now merged into one run per row first. The candidates
are the same: they were sorted and deduplicated before.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
tick_hoppers read the block under every block entity in the loaded world
each tick to find the hoppers. A hopper's block entity is always a plain
container (BlockEntity::Chest), so only those need the lookup.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
PlayerAdvancements.flushDirty only works when it is the first packet or
progress changed. Horizon walked the whole advancement tree for every
network player every tick to find nothing new: a fifth of the server
tick with 50 players.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Interpolated animated sprites (`"interpolate": true`) mix two frames into
every mip level with `f32::round`. Baseline x86-64 has no rounding
instruction (it needs SSE4.1), so each byte was a call into libm's roundf:
7.8 to 8.5% of the render thread's samples on a Kaby Lake laptop.
Round with a truncation and an exact fraction test instead. The result
is bit-identical to `f32::round` for every value the blend can produce,
checked over all frame byte pairs at 144 blend times.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
PlayerList.tick broadcasts UPDATE_LATENCY once ++sendAllPlayerInfoIn > 600;
the hub sent it every 100 ticks. The extra updates landed in the
dedicated_server harness's respawn window and failed it at random.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Online mode checks accounts since v2026.10.17.07; until #589, players added
before they join get offline UUIDs, with a workaround. Portals follow a
lone player since v2026.10.17.09. Copied game files need v2026.10.17.07 or
newer. Section 10 links the 2026-10-17 benchmark and says plainly that 25
moving players are too many for now. Links to other docs are absolute, so
they work both on GitHub and on the site.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Port CachedUserNameToIdResolver to the dedicated server, in place of the
in-memory name cache: usercache.json in the universe folder is loaded at
startup (bad entries skipped, a bad date warned about), players are added
when they join and kept for a month, and the file keeps the 1000 most
recently used, newest first, as Gson writes it.
The console's op, deop, ban, pardon and whitelist find players by name
through it, in any case. A name it doesn't hold is asked of the profile
service (profiles::find_by_name) on offline servers too, as vanilla does;
one the service doesn't know is the offline player of the lower-case name
on an offline server, and "That player does not exist" on an online one.
tools/test_user_cache_parity.py runs vanilla's server and Horizon's with a
mock profile service and compares the replies, the names looked up and the
files; the dedicated-server harness now compares usercache.json too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
With oracle/facts.json, a mod the oracle did not load (a library it
needs is missing) lost every block: none of its ids were in the facts.
It now keeps the guesses from its blockstates, as without an oracle.
ATM10: Regions Unexplored (TerraBlender is CurseForge-only) gets its
805 blocks back.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
horizon-bots and tools/server_bench.py at 1, 10, 25, 50 and 100 bots,
walk and explore, Horizon (main at 123987c4) and vanilla 26.3 taking turns.
The Mac mini was overloaded by other work (load 55-141), so only the 1-
and 10-bot rows are a fair comparison; the README says so and gives the
load of every run. It also records where Horizon's server thread goes at
25 walking bots: 96 % in ChunkEncoder::encode_in.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
At render distance 32 the section arena holds about 1.5 GB of vertices
(2.1 GB of GPU memory in all while flying). Vulkan drivers fail an
allocation once video memory is full instead of paging like OpenGL
drivers, and wgpu's default error handler turns that into a panic: a
4 GB laptop GPU that also drives the desktop crashed a few seconds into
an RD32 world.
Arena pages are now created inside an out-of-memory error scope. A page
that doesn't fit in video memory is created mappable, which the
allocator places in host-visible system memory, so the terrain still
draws (more slowly) as OpenGL drivers do. A page that can't grow keeps
its contents and a smaller new page is tried; with no memory left at
all, the section is skipped rather than the game ending.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla skips a resource with an invalid path, a data file that does not
parse, one at a time; Horizon's content loader refuses the whole mod
set. carry now keeps them in outside/ with the reason (report
`unreadable`): upper-case paths, JSON that does not parse, shared
tags with "replace": true. A byte-order mark is dropped.
ATM10's 42 carried mods: 19 such files in 7 mods (Mekanism Tools'
ry_UA lang, CTOV's empty tag, Towns and Towers' replacing tags, broken
template pools, recipes and a lang file), and the 42 load as carried.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
lan/chunks.rs: keep the per-section encode cache and the cached
heightmaps, applied to the level's section range from #668. Heightmaps
use client_heightmaps_in with the level's min_y and height, and
set_dimension drops cached heightmaps. The light layers go over the
level's sections and drop sky light where the level has none.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- perf_bench.py samples the physical footprint on macOS as well as
RSS. RSS counts the pages the allocator freed and marked reusable,
so the six mods' "+1.1 GB" is about 0.5 GB of footprint.
- examples/resource_memory.rs: live heap, peak, RSS and footprint after
each resource loader, for a prepared or a content-mod root.
- BlockRegistry indexes ids by a hash of the state, checked against
the states, instead of keeping every state twice: -22 MB live for
vanilla, -124 MB for ATM10's 42 mods (193,762 states).
- docs/modpacks.md: the per-loader breakdown and the next cut (the
per-state tables parsed whole at startup).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Port 26.3's ServerWaypointManager for the players of a served world: each
player who receives waypoints gets ClientboundTrackedWaypointPacket for
every other player who transmits one, as a block (in view), a chunk (one
chunk out) or an azimuth (really far), with the update thresholds,
breaking rules and team colours vanilla uses. Crouching, invisibility and
a carved pumpkin or mob head hide a player; spectators see everyone and
are seen by no one; the locator_bar game rule turns it all off and on.
tools/test_locator_bar_parity.py runs vanilla's server and Horizon's with
two vanilla-codec players and compares every waypoint packet each receives.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
AvatarRenderer.getArmPose gives THROW_TRIDENT to the hand using a
trident: HumanoidModel lifts that arm over the head (xRot * 0.5 - pi)
and leaves the other arm's pose alone (affectsOffhandPose). Horizon
used the plain item pose, so the held trident, drawn with its throwing
model, pointed down through the body. An aggressive drowned raises its
trident the same way (DrownedRenderer), and zombie arms leave that arm
alone (AnimationUtils.animateZombieArms).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
FloatGoal's constructor lets the navigation float, so Mob.jumpInLiquid
adds 0.04F a tick; Horizon set that flag only for a few species, and
every other FloatGoal user (pigs, cows, wolves, cats, horses, witches,
illagers...) got the non-floater 0.3 and bounced out of the water.
Register it from the goal (and from the constructors that set it:
allay, armadillo, camel, goat, parrot, wither), and add 0.04 as the
float vanilla adds.
Mobs that drop back into water now splash (Entity.doWaterSplashEffect
from updateFluidInteraction, skipped on the first tick), with their
swim splash sound. The new mob_ai deep_water scenarios also caught:
horses register addBehaviourGoals last (none for the skeleton horse, no
mount panic for the zombie horse), RandomStandGoal waits the horse
ambient interval (400 for every AbstractHorse), and the cat's sit-on-
block goal draws its delay only for tame cats once the last ran out.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
With the wide state ids, ATM10's 42 data-tier mods (193,762 states)
reach the world once these loaders stop refusing them:
- Block sprites: eight texture arrays of 2048 layers instead of four
(9,790 block textures for the 42).
- Per-state tables read as NBT: the table limits cover 2^20 states, and
the collision table in level data uses them.
- Worldgen block tags: 2^20 blocks instead of 8,192 (mineable/pickaxe
holds every stone block of a pack).
- As vanilla's loaders do, a blockstate definition that does not bind,
a missing or broken block model of a mod, and a recipe that does not
parse are logged and skipped instead of dropping every mod (or
stopping the game).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Chunk sections keep block states as u32 and size the global palette
to the registry: 16 bits for vanilla's 35,723 states, wider when
content mods take the registry past 65,536. The content-mod limit
goes from 2^16 to 2^20 states.
The LAN host writes chunks at the registry's width, so with vanilla
(or mods within 16 bits) the wire is unchanged. Remote chunks are read
at the server's width, 16 from a vanilla server, and widened to ours.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
#656 handles ClientboundChangeDifficultyPacket and refreshes the World
Options screen too; this branch's own arm became unreachable.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
What works on main today (install, game files, first run, server.toml, the commands, the web panel,
backups, services, Docker built by hand) and, plainly, what doesn't yet: offline mode only, no
published image, no secure chat or locator bar, the open PRs that are coming. The site renders the
guide into /server at build time, so it has one source. The CSP hashes now cover every page's inline
script, so the privacy page's theme script is allowed too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The player's model never read the swimming pose: sprint-swimming and
crawling drew it upright and walking. Add LivingEntity.swimAmount
(eased 0.09 a tick toward isVisuallySwimming), AvatarRenderer's swim
rotation (tipped by the pitch in water, -90 out of it, laid along the
swimming box) and HumanoidModel.setupSwimAnimation (the 26-unit arm
stroke, head down, flutter kick). The avatar rotations now follow
LivingEntityRenderer.setupRotations' death and upside-down turns, as in
vanilla.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
level/data.rs, level/mod.rs: load, load_with_behaviour_json and
load_with_gameplay all go through one cached() helper (try the snapshot,
build and store on a miss), so #607's client path keeps this cache.
Resolution from contrib-review-1's validated integration, without #601.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Walk and explore passes at 1 to 100 bots on the ThinkPad (bots on the Mac
mini), vanilla with the locator bar off, plus an explore pass at a rate both
servers keep up with. The README has the method, the tables, the profile of
what remains at 25 walking bots and the earlier runs; docs/server.md section
9 and the Scale row summarise it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
encode_in built the light layers twice (encode filled them for every
section, then encode_in replaced them with the ones that exist around the
column). On the ThinkPad at 25 walking bots light_data was 8 % of the server
thread. The light is now built once, and a section's packed DataLayers stay in
the section cache while the section is unchanged (relighting moves it to a new
address, as an edit does). Uniform light is filled directly, not kept.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The server holds one level at a time, as the in-frame game does. While one
player is online (single player on the server thread, or a world with one
player in it), a Nether or End portal swaps the server's level the
in-frame way (switch_level, split out of enter_dimension) and the player
is moved as ServerPlayer.teleport moves it into another level: a Respawn
packet keeping all data, the difficulty, permission level, teleport,
abilities, level info, inventory, effects and the portal travel sound,
then a teleport to the exit once the destination has loaded (the Nether
portal found or built, the End platform, the Overworld's spawn column).
Respawning away from the Overworld takes the level back there first.
What the hub sends follows the level:
- Chunks carry the level's sections only (y 0 to 255 in the Nether and the
End), without sky light where the dimension has none, and heightmaps
from the level's floor. Until now every level went out with the
Overworld's 24 sections, which a client in the Nether or the End reads
64 blocks off.
- The dimension type follows the level the hub names; the default spawn
position is the Overworld's.
- A level switch keeps the block change log (an open hub kept none after
a portal) and the sculk watch list.
Worlds whose player is in the Nether or the End now open on the server
thread too.
The network client maps the End to the End (it drew it as the Overworld)
and shows the death screen on its own PlayerCombatKill, as
ClientPacketListener.handlePlayerCombatKill does, whatever health came
last.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ClientboundUpdateAdvancementsPacket and ClientboundAwardStatsPacket are
new in the protocol crate, and the window applies them as vanilla's client
does:
- ClientAdvancements.update: a reset replaces the tree, added advancements
join it with their place on the tab (DisplayInfo x, y), removed ones go,
each changed one takes the progress sent, and one that is done in a
non-reset update with show set gets its toast. The advancements screen
lays a server's tree out where the server placed it and shows everything
it sent.
- handleAwardStats: the counters the server sends, by registry name.
Opening the statistics screen asks for them (StatsScreen.init's
REQUEST_STATS).
Joining and leaving a server start from empty advancements and statistics.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A Modpacks page between Versions and Patch notes (docs/modpacks.md, the
working version's launcher step):
- Import a pack file (.mrpack, CurseForge zip) or an instance folder (Prism,
MultiMC, CurseForge): horizon-port-pack's import, inventory (Modrinth by
file hash) and report into ports/<slug>/ in the game's home.
- The coverage report: ported, partial, not yet and not needed, with every
mod's class, tier, licence and reason, filtered by status.
- Port: the portable mods' jars (content, data tier) the player's own way
(their folder, the pack's links, Modrinth's CDN; what only CurseForge has
needs its key and is said so), carried over for the selected version's
prepared tables; the pipeline's log shows while it runs.
- Play: the game as the pack's own profile, ports/<slug>/home/ (its saves,
options.txt copied from the player's the first time) with the carried-over
mods as dev mods. The session flow is shared with Play (start_client).
- The same from the command line: horizon-launcher modpack list | import |
port | play | remove.
Nothing is uploaded or shared; jars and ports stay in the game's home.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The launcher's "Import modpack" needs the pipeline as a library and a way to
get the jars a pack names without redistributing anything.
- The steps move into the library (pipeline.rs) with a line callback, so the
command line and the launcher run the same code; main.rs is a thin CLI.
- fetch: the jars of the portable mods (content, data tier) into
<work>/jars/, the player's own downloads: copied from a folder they have
(by SHA-1, or a CurseForge app instance's minecraftinstance.json), else
from the pack's own link (.mrpack hosts only) or the same file on
Modrinth's public CDN found by SHA-1, kept only when the hash matches.
What only CurseForge has is listed as needing a CurseForge API key.
<work>/fetch.json records each file.
- port: every step after the import for the portable mods (fetch, the
inventory again with the jars read, carry, report).
- A CurseForge zip whose overrides hold Crash Assistant's modlist.json (All
the Mods 10 does) takes its jar names, mod ids and SHA-1s instead of bare
project ids, so Modrinth answers for most files.
- inventory goes on offline when Modrinth cannot be reached.
All the Mods 10 8.3's CurseForge zip: 496 files named; of the 54 data-tier
content mods 42 download from Modrinth's CDN (84 MB) and 12 need the key or
the player's instance; 42 carried over (partial: the harness is a dev step).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Single player on its server thread, LAN guests and dedicated server players
are network players, and the server gave them none of these:
- PlayerAdvancements: each player's progress loads from and saves to
players/advancements/<uuid>.json (an integrated server's owner keeps
single player's file). The game's triggers, awards and the advancement
commands work on the player the game acts for (as_guest swaps the
progress in), every tick runs the tick, location and inventory triggers,
and flushDirty sends ClientboundUpdateAdvancementsPacket with what became
visible or hidden and the progress that changed. A player who has been
here gets their saved advancements in the first packet; a new one gets
the same first tick as before. Announcements go to everyone
(PlayerList.broadcastSystemMessage).
- ServerStatsCounter: per-tick and movement statistics are counted for each
player, saved to players/stats/<uuid>.json, follow the scoreboard
criteria under the player's name, and go out in
ClientboundAwardStatsPacket when the client asks (REQUEST_STATS), all of
them the first time.
- ServerRecipeBook: the player file's recipeBook is kept, sent on joining
(settings and every known recipe), and recipes unlocked later go out in
ClientboundRecipeBookAddPacket.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Main's startup now falls back to vanilla resources when packs fail; the
behaviours snapshot folder is set just before.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
With single player on its server thread, the window is the server's client
and dropped several packets the in-frame game never needed:
- ClientboundBlockEntityDataPacket and the block entities of chunk packets:
the block entity of that type loads the tag (signs, spawners), as
handleBlockEntityData and LevelChunk.replaceWithPacketData do.
- ClientboundChangeDifficultyPacket: the level data's difficulty and lock.
- ClientboundSetDefaultSpawnPositionPacket: the respawn data spawn
compasses point at.
- ClientboundSetChunkCacheCenterPacket: ClientChunkCache's view centre;
columns outside the storage (the server's radius + 3) are ignored or
dropped, as ClientChunkCache.Storage does.
- ClientboundAnimatePacket: wake up, and the critical and magic critical
hit emitters.
- ClientboundTickingStatePacket / ClientboundTickingStepPacket (new in the
protocol crate): the level's TickRateManager. A frozen level holds its
time and non-player entities, and the server's rate only ever slows the
client timer down (Minecraft.getTickTargetMillis).
- ClientboundServerDataPacket: the server list entry keeps a valid icon
(ServerList.saveSingleServer); single player has no entry.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Path searches (vanilla's PathFinder with WalkNodeEvaluator) were 14 % of
the world tick with 80 villagers. Two costs vanilla doesn't have:
- the node index and the per-query path type and collision caches used the
standard SipHash; vanilla keys them by packed position in fastutil maps.
They use the Fx hasher now (nothing iterates them, so no order changes);
- a position's static path type (its block, the one below and up to 24
neighbours, `getPathTypeStatic`) was worked out again for every mob box
that covered it. Vanilla caches the block part level-wide
(`PathTypeCache`); the finder now remembers each position's type for the
query, during which the world doesn't change.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
When chunks arrive, the sculk index looks through them for sensors,
shriekers and catalysts. It unpacked and tested all 4,096 blocks of every
section that wasn't uniform: 10 to 18 ms in a tick after a world loaded,
more over a larger render distance (#489).
A section whose palette holds no sculk state has no sculk block, so it is
passed over without reading its blocks; the others are scanned as before,
and the same blocks are found.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Every tick, each item and mob that can use a portal sweeps the blocks its
box passed through since the last tick, looking for a nether portal block
(`Entity.checkInsideBlocks` reaching `NetherPortalBlock.entityInside`).
Each sweep allocated a visited set and read every block of the path.
Every block a sweep visits lies in the box covering the entity's box at
both ends. When no section that box touches has a nether portal state in
its palette, no visited block is a portal: the sweep is skipped. Sections
are checked once per tick, and a box over more than 64 sections (an
entity teleported far) is swept as before.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Block collision scans (vanilla's BlockCollisions) are most of every mob's
move, and moves were the largest share of the world tick with 600 mobs.
Per cell and per hit, the scan did work vanilla's JIT makes cheap:
- the cursor divided its index by the box width and height for every cell;
it now counts x, then y, then z, as Cursor3D does (same cells, same order,
same budget);
- the unit cube test cloned the shared unit cube to compare identities
(`VoxelShape::is_block` compares with it in place), and built a
normalised box to test it against the entity box (`intersects_corners`);
- every non-empty block looked its state up in the contextual rule map
(scaffolding, liquids, powder snow); a flag in the scan table now says
which states have one;
- a moved shape decomposed its occupancy into boxes again; the moved
shape's boxes are its own boxes moved, the same values.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A loaded world had none of its structure starts in the generator's cache.
The first time natural spawning (or anything else) asked which structures
hold a position in a chunk referencing a village, the game thread built the
village start: 60 to 120 ms in one tick, once right after the world opened
and again a few ticks later for the next start (#489).
Vanilla reads the starts with the chunk. The streaming worker that loads a
saved chunk now builds the starts it references into the same cache, so the
game thread finds them there. The starts are the same: the same function
builds them from the same seed and chunk, only earlier and on a worker.
The cache now keeps the first starts built for a chunk when two threads
build them at once, so every user shares one start and its pieces'
placement state, as vanilla's single StructureStart does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
lan_actions.rs: clear the old body's potions (main), then respawn with
the tick context (this branch). hub.rs: keep main's effect and chunk
tracking reset before this branch's vanilla respawn sequence.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Horizon has no locator bar yet, so equal-work runs turn vanilla's off
(--gamerule locator_bar=false): its waypoint updates cost every move
a pass over every other player.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The bots acknowledge chunk batches with ChunkBatchSizeCalculator's
desired chunks per tick, from how long each batch takes to arrive,
instead of a fixed 25. A bot moves only while it holds the chunk it
stands in, as vanilla's client stops ticking its player in a chunk it
hasn't received, so no bot outruns a server's chunk sending. Each bot
keeps the chunks it holds, so the report separates chunks that entered
its view from chunks sent again while it still held them, and counts
the moves the server refused and the ticks it waited for its chunk.
server_bench.py runs one pass per behaviour (walk, explore), spawns
every bot on the world spawn itself (respawn_radius 0) so both servers
see the same paths, can run the bots on another machine (--bots-via,
--bots-bin, --connect) and the servers under nice, and reports new,
dropped and re-sent chunks per second, pull-backs and the server's
"moved too quickly" lines.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
PlayerChunkSender: one unacknowledged batch until the client's first
acknowledgement and ten after it, batches capped at the client's
desired chunks per tick (not twice that), and the quota reset to one
chunk whenever every batch is acknowledged.
ServerGamePacketListenerImpl.handleMovePlayer: a move further from
where the tick began (firstGood) than 100 blocks squared per move
packet (300 while gliding; the count falls back to 1 past 5 packets)
logs "<name> moved too quickly!" and pulls the player back, as the
player_movement_check and elytra_movement_check game rules allow.
NaN coordinates or non-finite angles and two position moves in one
client tick disconnect with invalid_player_movement, positions are
clamped to the world's limits, an unanswered teleport is sent again
after 20 ticks, and accepting a teleport runs the move it carries.
The move is then replayed against the blocks: a survival or adventure
player who ends more than 0.25 blocks short of the target (horizontally;
vanilla drops the vertical part) logs "<name> moved wrongly!" and is
sent back unless they were already stuck, and any move into a block
the old box did not touch (isEntityCollidingWithAnythingNew) is sent
back silently. Spectators and sleepers skip both. Angles wrap as
Mth.wrapDegrees does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Every PR of the startup batch adds a different tools/profile_client_startup.py,
so any two of them conflict. This one becomes tools/profile_shared_json.py; its
guide and tools/profile_gameplay_decode.py follow.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Every PR of the startup batch adds a different tools/profile_client_startup.py,
so any two of them conflict. This one measures only Behaviours::load, so it
becomes tools/profile_behaviour_cache.py, and --cache-dir passes the snapshot
folder to the example so warm loads can be measured.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Loading the eight carried-over mods in the game (not only the probe) showed
two limits real packs reach. Block textures past 2048 layers are another PR;
this one handles block states and measures a pack.
- Chunk sections store a block state in 16 bits. Vanilla 26.3 has 35,723
states; Macaw's four mods add 36,154, the eight 41,389. Past 65,536 ids
wrapped onto other blocks (a trapdoor became a hanging sign).
mod_content::build now refuses a set of mods past MAX_STATES and names the
states per mod; the game then starts without content mods, as for any
refused content.
- tools/mod_port_harness.py probes each mod on its own against vanilla, as a
player adding it would load it.
- tools/perf_bench.py --game neoforge: Minecraft 1.21.1 on NeoForge 21.1.259
with --mod-jars and --deps, driven by a dev-only NeoForge port of the bench
mod (mods/horizon/src/neoforge-bench); --port-work loads a work folder's
carried-over mods into Horizon. tools/neoforge_oracle.py installs the
client once under .local/neoforge-oracle (the installer, then the vanilla
libraries it inherits, sha1-checked) and builds the driver. A NeoForge
start that crashes before the driver loads (FML reading fml.toml while
creating it) is retried once.
Measured (docs/modpacks.md): six ported mods in Horizon against the same
mods on NeoForge, Mac mini M4 Pro, 2 games per row on a loaded machine. The
world is ready 3.1x sooner at 8 and 16 chunks, the flights run 1.7x and
3.6x the frames, and Horizon's RSS is not lower (+1.1 GB with the mods).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Every packet sent to a guest woke its writer thread, and at 25 walking bots
those wake-ups (mpsc notify, semaphore_signal) were about a quarter of the
server thread's tick. While tick_lan runs, packets for each guest are held
and handed over in one Out::Packets at the end; a send outside a tick, or a
disconnect, first hands over what is held, so the bytes and their order are
unchanged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
options_ui.rs: #585 put Mods... in the telemetry slot; keep it and put
Horizon Extras... in the credits slot as this branch intended.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Right-clicking a furnace as a guest did nothing: the host only handed
chests, barrels, hoppers, dispensers and shulker boxes over. It now opens
the furnace's own menu (minecraft:furnace, blast_furnace or smoker),
runs the guest's clicks through the host's AbstractFurnaceMenu (fuel slot,
take-only result, shift-clicks by recipe and fuel, a new input restarts
cooking) and keeps the client's progress arrows in step with
ContainerSetData. The dedicated server harness now opens a furnace and
smelts an ore in it on both servers.
Fixes#540
Co-Authored-By: Claude Opus 5.5 <[email protected]>
component_nbt wrapped the numbers of a mixed list under the empty key, and
the NBT writer then wrapped those wrappers again, as ListTag.wrapIfNeeded
does. A vanilla client reading /give's reply ("Gave 1 [Raw Iron] to ...")
unwrapped once, found a {"": 1} argument and dropped the connection. The
list is now left to the writer.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
world_spawn now takes about 1480 s on a loaded Mac mini runner, just
under the old 1500 s limit, so PRs failed on a timeout rather than a
mismatch (#567, #514).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The resource folder is shared: worktrees and parity harnesses link one copy,
and the launcher copies a prepared folder to each new compatible version. A
snapshot there would be read by builds that resolve the tables differently
while the VERSION constant stays the same. Keep it in the game's own
.horizon-cache (set by the client; tools, tests and servers don't cache), add
the running executable to its key, and skip hashing the resources when there
is no snapshot. Postcard without default features drops heapless and the
unmaintained atomic-polyfill from the lockfile.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On a server the F key swapped the hands only on the client, and the
inventory menu's off-hand slot (InventoryMenu slot 45) was missing from the
client's slot table, so clicks on it went nowhere and the server's updates
of it were dropped (#543).
- F sends ServerboundPlayerActionPacket SWAP_ITEM_WITH_OFFHAND and predicts
nothing, as vanilla's Minecraft.handleKeybinds does (not for spectators).
- Inventory menu slot 45 is the off hand; SetPlayerInventory slot 40 too.
- The server swaps only for non-spectators and stops the item in use
(ServerGamePacketListenerImpl.handlePlayerAction).
Fixes#543
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A guest's STOP_DESTROY_BLOCK cleared the digging before the host's own
count was done, so a block the client broke stayed on the server and the
player fell through or froze on it. The host now follows
ServerPlayerGameMode: it breaks the block when its progress since START is
at least 0.7 and otherwise finishes the dig alone (the delayed destroy).
Block acks go out after the tick's block updates, and the client keeps
predicted blocks until the server acknowledges them
(BlockStatePredictionHandler), so a refused break comes back instead of
being lost.
Fixes#500
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A vanilla 26.3 client that died on a Horizon server stayed on its loading
screen after pressing Respawn (#566): ClientPacketListener.handleRespawn
starts a LevelLoadTracker that waits for the level info's "level chunks
load start" game event and the chunk under the player, and Horizon's
respawn sent neither the level info nor the chunk cache centre.
The hub's respawn now follows PlayerList.respawn: the entities paired with
the dead player are removed (one packet each), its chunks are dropped
without ForgetLevelChunk (as PlayerChunkSender does for a dead player) and
sent again, the abilities, then Respawn (with the last death location),
the teleport to the spawn facing its direction, the default spawn,
difficulty, experience, post effects, the level info, the permission
level, the chunk cache centre and the inventory. A dying guest also gets
the death animation event and no health update while dead, as vanilla's
ServerPlayer.die does.
Guests' commands also ran at the host's permission level: an operator on
a dedicated server could not run operator commands (and LAN guests ran
the host's level). They now run at the guest's own level, as the command
tree they are sent says.
The dedicated server harness covers it: after trading, the vanilla-codec
client is made an operator by the console, runs /kill, respawns, and the
packets from the death screen to the respawned inventory are compared
with vanilla's, then the level must load where it respawned. The client
also stops sending a movement packet beside each teleport acceptance
during the respawn: the 26.3 client sends the acceptance alone, and
vanilla's server kicks a second position in the same tick.
Fixes#566
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Chunk packets took nearly all of the server's tick with players moving
about: every send re-encoded the column for that player (each section
unpacked and re-paletted with linear searches, light layers packed a
nibble at a time, heightmaps primed again) and unpacked all 9 x 24
sections around it to find which light layers exist. At 50 walking
bots that was ~85 % of the tick, 280-470 ms ticks against vanilla's
33 ms on the benchmark machine.
The encoder now keeps what a resident section gives the wire (its
bytes, whether it holds block entity blocks, whether it is all air) by
the section's address. Sections are immutable behind Arc and edits go
through Arc::make_mut, so an edited section moves to a new address,
and the cache's Weak keeps an old address from being reused: the same
address is the same unchanged section. Heightmaps are kept per column
the same way. Block entity tags are still read for every packet. The
bytes sent don't change.
Also: Paletted::from_values uses an index table instead of linear
palette searches (the same palette order and bits), the all-air check
reads the palette first and stops at the first block, uniform sections
skip unpacking, and light layers pack two blocks per byte in one pass.
On a busy Mac mini with 50 walking bots (`tick query`): about 400 ms
ticks with the faster palette and air checks alone, 14 ms with the
caches.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
With single player on its server thread, the window is the server's
client, and some of what it answered was its own empty side:
- `horizon-ctl save` and `save-all` returned ok without saving: the
window's save skips worlds it is connected to. They now save on the
server thread (`Integrated::call`).
- `horizon-ctl chat-log` read the window's local chat, which never
showed what the server said (command feedback, joins). It reads the
connection's chat, as the agent's chat query already did.
- The game's own messages (Open to LAN, screenshots...) went to that
local chat, which is not drawn while connected: they now join the
connection's chat.
- `state` names the world (the server thread's `LevelName`).
- The client follows its permission level (EntityEvent 24..28,
LocalPlayer.setPermissionLevel), which the commands the chat offers
and checks depend on.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The game reads the data generator's registries.json (and commands.json)
next to the resources (#573). Versions whose resources the launcher copied
from another installed version, or that were prepared before the preparer
put the reports there, have them only in the scratch output of the version
that ran the data generator, if anywhere. At start, on `versions`/`update`
and before starting the game or a server, the launcher now copies them in
from the version's own scratch output or any installed version pinning the
same server jar, so existing installs can join servers again without
preparing again.
Complements #576 (the preparer and the client side of #573).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Each dimension change builds the destination's generator again, and with
it a new structure state, so the 128 stronghold ring searches ran again on
every return to the Overworld (about 1.5 s on two workers) and held up the
first new chunks. The ring positions depend only on the shared generator
data, the seed and the dimension's generator definition, so the generator
data now keeps them by those (the last eight), and later generators for
the same world and dimension share them.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Single player on its server thread could not be opened to the LAN: the
window is a client of that server, and Open to LAN, /publish and
`horizon-ctl lan` tried to start a second hub on the window's side.
The server thread now publishes itself, as IntegratedServer.publishServer
does:
- The window asks its server with `Integrated::call` (executeBlocking):
a task run between two ticks, while paused too, with its answer.
- Publishing binds the port at once (a busy port fails right away), then
the hub starts listening and announcing the world (LanServerPinger)
with the MOTD "<owner> - <world>". Until then only the owner may log
in; afterwards anyone may, and status pings are answered.
- Unpublishing stops the listener and sends every player but the owner
away (multiplayer.disconnect.server_shutdown).
- /publish and /unpublish run on the server for the player who types
them, and the integrated server's command tree (with publish, without
the dedicated-only commands) is what its players get.
- The owner's level follows Allow Commands; other players get game
master commands only when guests may use commands and the world allows
them (IntegratedServer.getCustomPermissionLevel). Every player's
command runs at their own level.
- The world options screen reads and applies its settings on the
server: difficulty, default and personal game mode, Allow Commands and
the LAN settings; the window shows the outcome.
- A published world does not pause (Minecraft.pause needs
!isPublished), and F3+Escape follows the same rule.
- Memory connections are never compressed, whatever the threshold
(Connection.isMemoryConnection), so network players get the usual 256.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
An online-mode server checks players by their account's UUID, so op,
whitelist, ban and pardon now resolve a name as vanilla's
CachedUserNameToIdResolver does: a player on the server, then one seen
since the start (in memory; vanilla also keeps usercache.json), then
the services' profiles/getByName endpoint through the discovery
document. A name with no account is "That player does not exist", with
vanilla's "Couldn't find profile with name" warning. Offline servers
keep the offline profiles.
horizon-server does the same while the server is stopped (op,
whitelist add, and the players init writes), and removes entries by
the name already on the list. `init --offline` (or HORIZON_ONLINE=false)
sets online-mode off; init's note now warns about offline servers only.
The online-mode harness adds a sixth login: `whitelist add` for the
player who joined earlier needs no lookup and lets them in, and an
unknown name is looked up and refused, on vanilla's server and on
Horizon's (43 checks).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A dedicated server reads registries.json and commands.json (the data
generator's reports) next to the prepared resources, and otherwise only
finds them in the .local/datagen folder of a development checkout. A
prepared folder copied or mounted on its own, as the Docker image
expects, stopped the server at start with "registries.json not found
next to the resources". The prepare step now copies both reports into
the folder (again whenever the server jar changes).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
online-mode=true now authenticates players as vanilla's
ServerLoginPacketListenerImpl does. The server makes a 1024-bit RSA key
pair at startup ("Generating keypair"), sends it with a random challenge,
checks the challenge and decrypts the shared secret from the client's key
packet, encrypts the connection with AES/CFB8, and asks the session
service's verify endpoint (hasJoined, through the services discovery
document) whether the player joined with the server's hash, from the
"User Authenticator #N" thread. A verified profile keeps its UUID and
properties, so skins show in the player list; an unverified one is
refused with "Failed to verify username!", and an unreachable service
with authservers_down. The offline-mode warnings are only printed in
offline mode.
Refused logins are logged as vanilla logs them: "Disconnecting
<name> (/<address>): <reason>" and "<name> (/<address>) lost
connection: <reason>", with log-ips respected.
horizon-protocol gains RSA key generation (Miller-Rabin primes,
e = 65537), the X.509 encoding of the public key and PKCS#1 v1.5
decryption. test_online_mode_parity.py runs its logins against Horizon's
server too, plus a verified player the whitelist refuses, and compares
the two servers' log lines about the player.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Every block sprite is a layer of one texture array, and GPUs give an array
at most 2048 layers (Metal, most Vulkan drivers). Vanilla takes 1250; eight
real NeoForge mods carried over (docs/modpacks.md) take 2183, and the game
stopped at start with "GPU supports 2048 texture layers; 2200 required".
- Sprites: a still texture with the same pixels and mip settings as one
already loaded reuses its layer (mods copy vanilla's textures under their
own names): 17 layers fewer for those mods, none for vanilla.
- The renderer splits the layers over up to four arrays of 2048 (8192
sprites). terrain.wgsl picks the array per fragment with derivatives taken
in uniform control flow; a pipeline constant keeps vanilla, which needs one
array, on the same single textureSample as before.
- HORIZON_SPRITE_SPLIT (dev only) forces a smaller split: vanilla drawn from
three arrays of 512 renders the same blocks as from one.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Resolve game.rs and titles.rs against the resources split (#537): layer
names and item names now come from self.pack, the atlas from self.atlas.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
#528 was tested before #537 moved the asset registry from Resources to
the reloadable pack, so main stopped compiling after both merged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla draws a section when its column is within the render distance of
the camera's column with a column of slack on each axis
(SectionOcclusionGraph.isInViewDistance → ChunkTrackingView): 533
columns at render distance 12. Horizon meshed a circle of radius 12 (441
columns) and drew sections whose centre lay within 16 x distance + 24
blocks, so the edge of the world, mostly in fog, missed columns vanilla
shows, and fell short of vanilla's corners.
`streaming::in_view_distance` is vanilla's test. Streaming meshes the
columns it accepts around the player (and forced ones), and the
renderer's column walk draws the ones it accepts around the camera.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A player who joins over the network, which is now single player's own
player on its server thread, had no status effects: what they drank, ate
or were given went to the host, and their client never heard of any.
- Each guest has their own effects. `as_guest` trades them, and what
their body was given for them (attribute modifiers, armor protection,
enchantment modifiers), with the host's while the guest acts.
- They tick on the server with the guest's body (LivingEntity.baseTick's
tickEffects), spectators included.
- The client is told as vanilla's ServerPlayer does: every effect when
placed (PlayerList.sendActivePlayerEffects), then added (blended),
updated, refreshed every 600 ticks left, and removed. An effect that
only counted down is not sent again.
- The client keeps its own effects from those packets
(LivingEntity.forceAddEffect), counts them down (tickClient), and its
predicted movement, HUD icons and night vision follow them. A respawn
starts without effects on both sides.
- Guests save and load them in their player file.
Saving also stops writing effects that have run out: the loaded player
compound kept the old `active_effects`, and they came back on reopen.
Potions in flight now tick on a server without a local player too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Resolve minecarts.rs against the resources split (#537): the block model
catalog now lives in self.pack. Fill MobData's new villager_hats and pois
fields in the sulfur cube render test.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On its server thread a single-player world is joined like a server, so
the Horizon pause menu named it "Multiplayer" with a "Disconnect" button,
and F3+Escape opened the full menu as on a server. The server thread now
hands the window the world's name with its connection, the pause menu
shows that name and "Save and quit", and `has_singleplayer_server`
(Minecraft.hasSingleplayerServer) covers both ways of running a world for
the pause checks.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The integrated server draws nothing, but it built the window's full HUD
atlas (GUI, item, block, entity and particle textures plus the run-time
slots). It now builds an atlas of the font sheets only, reads the glyph
widths from it and drops the pixels: text still measures the same (books
split their pages with it), without holding the textures.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
`report <work>` writes coverage.json and coverage.md: per mod ported,
partial, not yet or not needed with the reason, and the totals by class,
tier and licence family. It reads the inventory, each mod's carry report
and its harness result: ported is a green harness with nothing left out
(blocks with a class of their own are counted and named); partial lists
the failing checks, the blocks left out and what the templates miss; not
yet names the API tier or the dependency it waits for; not needed covers
libraries, performance mods and pack tooling. Unclassified mods are not
yet, never not needed.
`import` also reads Crash Assistant's modlist.json, the list ATM10
publishes in its repository (mod ids, names, versions, Modrinth hashes,
the loader entry), and the inventory uses its mod ids and names when
there is no jar.
ATM10 (496 files, the eight mods we downloaded carried and checked):
ported 1, partial 7, not yet 367, not needed 121.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The harness the modpack design calls for, for data-tier mods:
- tools/neoforge_oracle.py installs a NeoForge 21.1.259 server (ATM10's)
under .local/neoforge-oracle, builds the exporter in
mods/horizon/src/neoforge-oracle (our code, never shipped), runs it
headless with the user's jars and writes every block (class chain,
states, per-state hardness, light, tool need, map colour, sounds,
outline and collision; drops for hand, diamond tools, shears and silk
touch; placement; BlockSetType and WoodType), item, recipe and tag of
the mods' namespaces, plus the vanilla blocks for templates. About 30 s.
- examples/mod_port_probe.rs builds the content root from the carried
mods and dumps what Horizon's loaders make of the same things.
- tools/mod_port_harness.py compares them per mod and writes
<work>/mods/<id>/harness/result.json. 1.21.1 sound events map to 26.3's
through vanilla blocks both have, and vanilla items newer than 1.21.1
are not counted against a recipe.
On ATM10's own jars (MIT): Macaw's Trapdoors passes everything; Doors,
Fences, Paths, Glassential, Factory Blocks and Cloud Glass pass every
drop, recipe, tag and placement and fail only on what API 0.2 cannot say
(shapes and sound events of their own, blocks of classes no vanilla block
shares). Jars, facts and results stay in the work folder.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Carrying over seven real ATM10 mods (Cloud Glass, Factory Blocks,
Glassential, Macaw's Doors, Fences, Paths and Trapdoors) showed what the
guesses missed:
- Loot tables: 1.21.1's `conditions` lists, `functions` and
`condition`/`function` kinds are now 26.3's single `condition`,
`modifier` and `type`; silk-touch and shears tool matches become
`minecraft:tool/can_silk_touch` and `can_shear`. Before, every drop
condition was ignored (glass dropped itself by hand, upper door halves
dropped doors, double slabs dropped one).
- With the oracle's facts (<work>/oracle/facts.json, from
tools/neoforge_oracle.py in the next PR), every registered block gets
the vanilla template of the same vanilla class and states that matches
its shapes, sounds, BlockSetType and WoodType best, and the mod's own
hardness, blast resistance, light, tool need, friction, map colour and
sounds. Items get stack size and rarity. The carry report lists what a
template does not reproduce and why a block was left out.
- Convention tags a recipe uses (c:dyes/red) are written from the
oracle's tags, vanilla and own members only; the jar's own c tags stay
inside the mod.
- Without the oracle, blockstates that omit properties their models
ignore (powered, waterlogged) still find their template, and names pick
between fences and panes, doors and trapdoors.
- `version = "${file.jarVersion}"` is read from MANIFEST.MF.
- `--resources <prepared>` limits templates to blocks Horizon has.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Loading seven real NeoForge mods carried over by horizon-port-pack (about
1,000 blocks) broke the content root in four places:
- The per-state NBT tables (cached collision, occlusion, movement, worldgen
block properties and support) passed the NBT reader's 1M-node cap once
mod states were appended. They are trusted prepared tables with a row
per state, so they now read and write with Limits::tables() (16M nodes,
256 MiB) everywhere they are loaded.
- The overlay linked the base root by the path it was given, so a relative
base gave dangling links. The base is made absolute first.
- Carried-over tags name blocks the port left out and other mods' items.
Vanilla fails such a tag as a whole; entries naming ids nobody registers
(or tags nobody defines) are now dropped, so the rest loads.
- Mods use the convention tags (c:dyes/red, c:glass_blocks/colorless).
Content mods may add to data/c/tags/ as to vanilla's, and with content
mods loaded the tag loaders read the c namespace and the mods' own.
Stonecutter and smithing recipes in mod namespaces load too (ids
<ns>:<name>), as crafting and cooking ones already did. With no content
mod nothing changes: the loaders read the minecraft namespace only.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The per-frame section walk iterated a SipHash map of every resident
section, then looked each visible one up twice more by key. Resident
sections now live in a dense list grouped by chunk column: one distance
and frustum test rejects a whole column (exact: a section's box lies in
its column's box and the plane test rounds monotonically), the rest are
reached by slot, and the visible list sorts on integer keys with ties
broken by section key, so the order no longer depends on hash order.
FaceGroups looks its axes up in a table instead of searching for them.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
What the dedicated server does today (measured against vanilla 26.3), the horizon-server supervisor and
its packaging, server.toml, the CLI and TUI, the web panel, backups and updates, the control API and MCP,
the bot benchmark plan against vanilla and Paper, the working version for 2026-10-23, and the PRs that
build it (#551 to #556).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
horizon-bots is a load client on horizon-protocol for any 26.3 server: offline login, configuration,
keep-alives, pings, teleports, chunk-batch acknowledgements, and movement at 20 Hz. The bots play in
creative and fly above their spawn point, so they need no world of their own and no server refuses their
moves; idle, walk, explore and mixed behaviours are deterministic from a seed.
tools/server_bench.py runs the same bots on Horizon's server (horizon-server run) and on the pinned vanilla
jar, each fresh from the same server.properties, and measures MSPT from each server's own /tick query,
RSS and CPU of the process tree, the chunks the bots receive, and the start time.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
One page embedded in the binary (crates/horizon-server/panel: HTML, CSS and JS, no framework or CDN): status,
live TPS, MSPT, RAM and CPU graphs over the last 10 minutes, players with op and kick, the console, and
backups with restore, plus restart and stop.
It listens on [panel] bind, 127.0.0.1:8130 by default; a tailnet address reaches it from the owner's
devices, and any other address is refused unless allow_public is set. `horizon-server panel` prints a
one-time login link (10 minutes) that becomes an HttpOnly, SameSite=Strict session cookie. The Host header
must name the panel, every POST needs X-Horizon: 1, failed logins are rate-limited, and the page is served
with a strict Content-Security-Policy. The control API gains GET /api/v1/metrics and, with the token
only, POST /api/v1/panel/login.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The window now joins its server thread through an in-memory pipe, as
vanilla's integrated server player does (LocalChannel), rather than a TCP
socket on 127.0.0.1. The server opens no port until it is opened to the LAN.
- horizon-protocol: Connection runs over TCP or one end of a Pipe (a
blocking byte pipe with read timeouts and shutdown; dropping an end
closes it).
- Lan: NetConfig.listen false starts no listener; connect_local serves the
other end of a pipe on its own thread, through the same handshake, login
and configuration as a socket.
- Session::connect_local joins through a pipe.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
With --server-thread or HORIZON_SERVER_THREAD=1, opening a single-player
world starts a game of its own on a "Server thread" over the shared
resources, as vanilla's IntegratedServer does, and the window joins it as a
player over a local connection. The world tick then never runs inside a
frame.
- The server listens on 127.0.0.1 only, lets in the window's profile alone,
without compression, and makes it an operator when the world allows
commands.
- The window's player is the world's own player: it loads from and saves to
players/data/<singleplayer_uuid>.dat, so the world opens the same way in
either mode.
- A pause screen pauses the server (IntegratedServer.tickServer), and
leaving the world or quitting waits for the server to save.
- The integrated server does not mesh sections it never draws.
- Worlds whose player is outside the Overworld open the usual way for now.
- The bench summary reports the 0.1% low.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
run() loads the shared resources and the pack, then hands them to a
constructor that builds a game with no world open. The same constructor will
build an integrated server's game on its own thread. The struct literal moves
unchanged; Options derives Clone so a server can start from the client's.
No behaviour change.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Resources now holds only what is loaded once and never changes (blocks,
items, recipes, biomes, mobs, collision, motion) and is Send + Sync behind
an Arc, so an integrated server thread can share it with the client.
What a resource pack reload replaces moves to PackResources (assets,
meshing context, item renderer, names, panorama, clouds, forcefield), built
by one loader used at startup and on reload. The client-only mutable state
(the HUD atlas with its font cache and skin slots, fog state, screen effects)
becomes the game's own fields.
No behaviour change.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
wgpu validates every indirect draw with a compute pass and a per-draw
batcher before the render pass, for untrusted WebGPU content. The section
draws are the only indirect draws and are in bounds by construction; the
validation cost about 0.1 ms of the render thread per frame at render
distance 32 on Metal. Debug builds keep it, and
WGPU_VALIDATION_INDIRECT_CALL=1 turns it back on.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
AcquirePoi (job sites, beds, meeting points) and the bed searches read
every block of the sections in a 48-block sphere. Like vanilla's
PoiManager, each section now keeps the list of its POI blocks, rebuilt
from its blocks after an edit (a weak reference detects edited and
replaced sections). The searches visit the same positions in the same
order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- packaging/server/install.sh (Linux, macOS) and install.ps1 (Windows) install the latest release (or
--version) from GitHub, check the SHA-256 GitHub publishes for it, and keep releases side by side:
<dir>/current points at the one in use, so an update or a rollback is one link, and the game files
prepared from the owner's Minecraft stay in <dir>/game-files across updates.
- packaging/server/Dockerfile builds the server and the engine; the prepared game files are mounted
read-only, and the first start writes server.toml from HORIZON_EULA, HORIZON_WORLD, HORIZON_SEED,
HORIZON_PORT, HORIZON_WHITELIST, HORIZON_PLAYERS and HORIZON_OP (`init` reads them too).
- `horizon-server service install|uninstall|print` writes and loads a systemd unit (user, or --system),
a launchd agent, or a Windows task at logon. It runs the installer's launcher, so updates apply on
the next restart.
- Releases build horizon-server and ship it beside horizon-ctl and horizon-prepare; PLAY.txt says how
to run a server.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
`backup [NAME]` zips the world and the settings into backups/; on a running server it turns saving
off, flushes, zips and turns saving back on, so the game stays up. `restore BACKUP` puts a backup's
world in place and keeps the current one in backups/replaced-<time>/; a running server restarts with
it. `[backups] every = "6h"`, `keep = 12` in server.toml back up on a schedule and prune the
automatic backups only. The control API gains GET/POST /api/v1/backups and POST /api/v1/restore.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The supervisor serves a small HTTP API on 127.0.0.1 (a token in .horizon/api.json, mode 0600, and a
Host check): status, logs, console commands, stop and restart. `start` runs the server in the
background and returns once the game is up; `restart` and `cmd` wait for it too. A sampler reads the
game's tick rate, tick times, players and chunks every second over the existing control endpoint
(`--control-socket`), and RAM and CPU from the system.
The dedicated loop now records its tick times in the game clock, so `/tick query` reports real
numbers on a dedicated server, as it does in single player. The changes to game/dedicated.rs are
additive: the log capture for command replies, the tick counter and the control requests
(game/dedicated/status.rs).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The first part of hosting a Horizon server with one small program
(docs/server.md). horizon-server is a new 1.3 MB binary with no window,
GPU, audio or controller code; it runs the game headless
(horizon-client --dedicated) as its engine.
- horizon-server-files: server.properties, eula.txt and the user lists,
moved out of horizon-client unchanged so both programs share vanilla's
exact reading and writing; horizon_client::server re-exports it.
- server.toml: typed settings for what owners change, [properties] for
any other vanilla key, vanilla 26.3's defaults (a test checks them
against ServerProperties). It is written into server.properties before
every start; keys the game changes there itself (/whitelist on) are
copied back afterwards, keeping the owner's comments (toml_edit).
`import` turns a vanilla server.properties into server.toml.
- init: the EULA (only a yes writes eula=true), world, seed, port (checked
free), whitelist and first operator; every question has a flag, so it
also runs unattended.
- run: the console on stdin, a clean `stop` on Ctrl-C, SIGTERM or SIGHUP
(the game runs in its own process group), and a restart when the game
crashes: 5 s, doubling to 5 min, giving up after 5 crashes in 10 min.
One supervisor per folder (a file lock).
- op, deop and whitelist for a stopped server, with vanilla's messages.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The threaded test added seed + 1 for seed = i64::MAX, which overflows in debug
builds (CI's verify).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
carry <work> [<mod id>...] turns each content mod with a jar into a folder
content mod (API 0.2) in <work>/mods/<id>/carry/, for local use:
- assets/<id>/, data/<id>/ and data/minecraft/tags/ are copied; other
namespaces and vanilla data go to outside/ for review; no class file is
copied.
- Vanilla-type recipes get 26.3 ingredients ("id", "#tag", lists);
custom types and NeoForge conditions are kept and listed.
- Item models get the items/ client definitions 1.21.4 introduced.
- Each block is declared from a vanilla template guessed from its
blockstate properties; unknown property sets are listed. Items
without a block are declared from minecraft:stick.
The test builds a small NeoForge jar, carries it over, and reads the
result with horizon-mods' own Content reader.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A CLI for the first steps of docs/modpacks.md's pipeline:
- import: a Modrinth .mrpack, a CurseForge zip or a Prism / MultiMC
instance (or a folder of jars) to pack.json: Minecraft and loader
versions, each file with its hash and sides, and the overrides.
- inventory: Modrinth metadata by SHA-1 (cached in modrinth.json, so
reruns work offline), and the jars when the user's instance has them
(neoforge.mods.toml / mods.toml / fabric.mod.json and a file census,
without decompiling). Each mod gets a class and API tier from the
reviewed ATM10 table, then rules on categories, sides and the census,
and a dependency layer. Writes inventory.json and inventory.csv.
Only metadata is downloaded; jars are never fetched.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The game builds the content root from the enabled content mods at startup.
Loaders that only read the minecraft namespace now also read the content
mods' namespaces (from the root's marker file): block loot tables, recipes,
blockstates, language files, and the HUD atlas's item sprites. /give and the
command registries know the new items and blocks. A block without a
blockstate of its own takes its template's, so every state has a model.
Adds examples/ruby-content, a folder content mod: a ruby block and a ruby
item with recipes, loot, tags, models, textures and names.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
horizon_client::mod_content builds the resource root for content mods as
an overlay over the base root (vanilla, or the resource-pack overlay):
- the base is linked;
- the block and item tables are rewritten, each mod block getting its
template's rows appended. That covers 13 JSON tables (facts, gameplay,
behaviours, models, render, landing, items, command registries, mob
data, worldgen flags, structure, survival) and 6 NBT ones (occlusion,
collision, support, movement, worldgen block properties). Names,
translation keys and in-table state ids move over, and the declared
overrides apply;
- each mod's own assets/ and data/ are copied in;
- tag additions are merged into the tag files and both block-tag bundles.
Vanilla state ids never move: mod states come after them, mods and blocks
in id order. Rewritten JSON keeps its key order. With no content mod the
root is the base itself. The game doesn't call this yet.
Tested over the real prepared resources: every loader reads the overlay
(Assets, BlockFacts, Items, collision, visibility, light, motion, worldgen
GeneratorData). The mod blocks carry their templates' facts with the
overrides applied, and vanilla's rows and ids are unchanged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A .hmod may carry assets/ and data/ beside mod.wasm, or instead of it. The
manifest needs api = "0.2" and the new registries capability ("add blocks
and items to your worlds"). Content mods have no WIT interface: code still
links API 0.1.
horizon_mods::content::Content reads a mod's content and checks it:
- plain resource paths only;
- 65,536 files, 16 MiB each, 512 MiB in all;
- files in the mod's own namespace, plus additions to vanilla tags;
- blocks and items declared in data/<id>/horizon/{block,item}/*.json, each
copying a vanilla template with typed overrides (hardness, resistance,
light, tool, friction, map colour, sounds; stack size, rarity), unknown
keys refused.
Mods::load refuses an enabled mod whose content fails these checks, with
the reason on its row. A content-only mod is on with no engine and no hooks.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Loading a world read every file name under a folder pack's data/ to find
functions, function tags, predicates and item modifiers. The built-in pack
holds about 9,900 other data files, so this walk took 190-340 ms on the
render thread at every world load. A folder pack now walks
data/<namespace>/<kind>/ for those four kinds only, in the same order as
before; zip packs keep filtering their index.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
docs/modpacks.md: how any pack (ATM10 down to small Fabric packs) gets to
Horizon: import (.mrpack, CurseForge zip or instance, Prism, packwiz),
inventory and classification, dependency order, mechanical carry-over of
data and assets, one agent per mod for code with a per-mod harness against
a dev-only NeoForge/Fabric oracle, and a coverage report. Also the mod API
gaps against 0.1, the registries 0.2 design (template blocks through the
overlay root), licence handling, how "10x" is measured against NeoForge,
and the scope for 2026-10-23.
docs/modpacks/atm10-mods.csv: ATM10's 495 jars from the pack's own
crash_assistant modlist (2026-10-15), with Modrinth metadata looked up by
hash: class, API tier, licence, sides. Metadata only.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Fills an area with N workers, optionally flies the centre along +x, and
reports wall time, chunks generated, sections meshed, how often a column
near the player is still unlit, and what Streaming::update costs the
calling thread. Allocations can be counted with
HORIZON_BENCH_COUNT_ALLOCATIONS=1 at build time.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The macOS release build has failed since 05:54 UTC on 2026-10-16: on the
self-hosted Mac mini runner the game drew its 1500 frames in about 2.5 s,
before the world had started generating (chunks 0, player null), so no
release has been published since v2026.10.16.03.
--world-frames N exits after N frames drawn with the player in the world.
It is otherwise a frame-limited run (classic menus, --timeout-seconds), and
--frames keeps its meaning for the harnesses.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The 128 stronghold ring positions each need a biome search over a 57x57
quart area. The first worker to ask computed all of them inside a OnceLock
while every other worker blocked on it: about 2 s on an M4 Pro with 12 of
13 workers idle, after every world load and every return to the Overworld.
Vanilla's generateRingPositions runs each search as its own task on the
background executor. Now the ring chunks and search randoms are planned
serially (cheap), and every thread that needs the rings takes searches
until none are left, then waits for the rest. Positions are unchanged:
each search only uses its own forked random.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The mesher groups each section's solid quads by the axis side they face
(Sodium's ModelQuadFacing) and records each group's nearest plane. When
the eye is behind every plane of a group, back-face culling would discard
all of it after vertex shading, so the renderer leaves the group out of
the section's draws. Seen from the village path, 29-36% fewer solid quads
reach the GPU at render distances 8-32; the picture is unchanged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
BlockCollisionScan asked every cursor cell for its collision shape,
moved it into place and intersected it with Shapes.create(box), even
for air. Mob movement and every other Entity.collide spent most of
their time there. Now:
- CollisionBlock carries empty_shape, set by the registry for states
with no contextual rule and an empty cached shape (air, plants): the
scan skips them without asking. Providers that build CollisionBlock
themselves (the query-order fixture) leave it false, so their query
logs are unchanged.
- A shape whose bounds miss the box is skipped before moving it:
joinIsNotEmpty(AND) of the two is false anyway.
- Shapes.create(box) is built the first time a shape is not the unit
cube (finite bounds always make one), not for every scan.
- SceneCollisions hands the scan the resident chunk itself, so a cell
costs one section read instead of a chunk-map lookup.
Results are unchanged: running both paths side by side over 900,000
mob moves in a 600-mob village gave bit-identical movement.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
PoiTypes.forState ran per palette entry of every section a villager
searched (and per block of the sections that hold one), with a string
match on the block name each time and a per-search hash map in front.
At world load, when every villager looks for a job site, bed and
meeting point at once, that was 16 % of the tick. The POI type of
every state is now worked out once, the first time a villager looks,
and read from a table after that.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The probe builds MobWorld itself; it now keeps the same SeenGrid over
its snapshot for the mob ticks, so the mob_ai harness checks pushes
through the grid against vanilla.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Entity.checkSupportingBlock runs findSupportingBlock after every move of
a mob on the ground. Horizon looked up and built the collision shape of
every cell of the cursor (27 to 48 per call, twice when it falls back
to where the mob came from). Like BlockCollisions' cursor, a block whose
shape stays inside its own cell (neither hasLargeCollisionShape nor a
moving piston) can only meet the area when its cell does, so only those
cells, and large shapes such as fences, are shaped now. The pick is a
total order, so visiting column by column finds the same block.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
LivingEntity.pushEntities compared each mob's box with every other mob
of the tick's snapshot: quadratic, about 1 ms per tick with 600 mobs.
The snapshot is now also kept in a grid of 4-block columns by each box's
minimum corner, updated as each mob finishes its tick, and the push
reads only the columns its box can reach. Candidates come back in
snapshot order, so the pushes, their order and their sums are the same:
over 800,000 pushes in a 600-mob village both ways gave the same
targets in the same order. A query wider than the grid (or not finite)
falls back to the full scan.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Drop this branch's own staging buffer (gfx/uploads.rs); the sprite uploads
go through the StagingBelt from #501. What stays is the animation mip-chain
cache holding every vanilla frame, with tests.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
Renderer::render's buffer writes (globals, section origins and draws,
entity, cloud, celestial, border, debug-line and screen-effect vertices)
and the HUD's vertices now go through the frame's staging belt instead of
one Queue::write_buffer staging buffer each. The lightmap is uploaded only
when its pixels change instead of every frame.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Every frame that crossed a tick re-uploaded the animated block sprites, and
that took about 1.8 ms on an M4 Pro (more than the world tick itself):
- The mip-chain cache held 8 MiB, which fits 256 of the 594 vanilla
animation frames (64-pixel layers, chains stored with spare capacity), so
the other 338 frames had their resize and mip chain rebuilt on the render
thread whenever their animation reached them. The budget is now 32 MiB and
chains are allocated at their exact size: all 594 frames fit in 13 MB.
- Each changed sprite mip level went through its own Queue::write_texture,
which allocates and frees a GPU staging buffer per call (a few hundred a
tick). They are now packed into one reused buffer, written once, and
copied with one submission per tick.
The texels are unchanged: a GPU test compares the batched copies with
write_texture over several flushes, and the existing cache tests compare
cached and generated chains.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Queue::write_texture creates a new staging buffer for every call, a kernel
call each on Metal. Animated sprites write every changed frame's mip
levels once per tick, so tick frames spent about 1.3 ms in buffer creation.
The copies now go through one wgpu StagingBelt recorded into the frame's
encoder; a GPU readback test checks they write the same texels as
write_texture on every mip level.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Nobody could reproduce it, and no issue tracks it, yet every release since
2026-10-14 repeated it. Players noticed (rdbt on Discord).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The rules by category under a search box, a switch or number box each,
with vanilla's tooltip, number validation (Brigadier readInt and the
rule's range, red text, Done held) and ConfirmScreen before discarding.
Done sets the changed rules and tells the player, as handleSetGameRule
does. The button follows the commands the player has, as updateButton.
The control socket's `type` now reaches World Options' boxes.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The Lock button toggles the wanted lock; Apply with a new lock asks first
(PopupScreen "Lock World Difficulty"), Yes locks the level's difficulty
(difficulty_settings.locked, now read and written), No returns to the
options. Restrictions opens RestrictionsScreen: the chat restriction the
Chat option makes, with its way to Chat Settings, and the four chat
permissions. A locked or hardcore level keeps both the difficulty and the
lock inactive, and LAN guests are told about the lock.
Part of #464
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ItemContainerContents.addToTooltip: an item with a container (a shulker box,
or a chest Ctrl + Pick Block copied) lists its first five stacks as
"count x name" and how many more there are. Spawner.appendHoverText: a
spawner or trial spawner item names the entity its block_entity_data spawns,
else how to set one. ItemStack.OP_NBT_WARNING: for a player with game master
permission, an item carrying data of an op-only block entity type (command
blocks, lecterns, signs, spawners) warns that it might run commands.
The item argument now refuses block_entity_data without an id or with an
unknown block entity type, with vanilla's TypedEntityData messages.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ServerGamePacketListenerImpl.addBlockDataToItem: with Ctrl held in creative,
Pick Block gives the block's item the block entity's data. What each type
turns into item components (removeComponentsFromTag and
collectImplicitComponents: container, custom_name, lock, container_loot,
sign text, waxed, bees, banner patterns, pot decorations, skull profile)
leaves its NBT; the rest becomes minecraft:block_entity_data, tagged with
the type's id. Placing the item does the reverse
(BlockItem.updateCustomBlockEntityTag, op-only types only for a game
master, then applyComponentsFromItemStack), for every block entity type,
which also fills placed chests, barrels, hoppers and the like from their
container component, as vanilla does, where only shulker boxes were.
The client sends includeData on Ctrl in multiplayer, a picked banner keeps
its patterns without Ctrl (AbstractBannerBlock.getCloneItemStack), a picked
beehive carries its honey level, a placed command block with data keeps its
own auto and output tracking, and a placed waxed sign opens no editor.
tools/test_pick_block_data_parity.py compares both directions with vanilla's
own block entities: 148/148 cases.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The control socket's key command went straight to key_pressed, so Escape
paused the game instead of closing the chat a server connection opens
(remote_chat_key handles the real keyboard).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ServerGamePacketListenerImpl.handleUseItem takes the client's rotation and
uses the held item, nothing else. The host ran the whole right click, with
the block the guest had looked at before: a guest that turned from an
ender chest to a double chest could get the ender chest opened again.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
PlayerDetector reads level.getPlayers(): the guests count for a trial
spawner's detection and a vault's connected players, and a key a guest
inserts rewards the guest's UUID.
Part of #462
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Feedback about the player (target_name) names the guest a command runs
for. A client reading a server's chat now unwraps NbtOps' {"": value}
wrapper of any bare value, not only strings, so translation arguments that
are numbers show ("Replaced 1 slot(s)", the coordinates). horizon-ctl
messages reads a server's chat too.
Part of #435
Co-Authored-By: Claude Opus 5.5 <[email protected]>
SubmitNodeCollection.submitNameTag: a name that is not discrete is drawn
in full over the world and, faded on its background, again with
RenderPipelines.TEXT_SEE_THROUGH (no depth test) after the rest of the
world; a discrete one (a crouching player) only once, faded, on its
background. The background is ARGB.color(0.25, black): alpha 0x3F.
Part of #435
Co-Authored-By: Claude Opus 5.5 <[email protected]>
With "View what would be sent" open in a short window, "Send nothing" and
"Save choices" sat below the overlay's scroll range. The card now fits the
window (max height 100%): its body scrolls, and the answers row stays pinned
at its foot.
Fixes#460
Co-Authored-By: Claude Opus 5.5 <[email protected]>
`wait` ended with release_agent_keys like every other task, so
`key w down` + `wait 40` + `wait 20` walked for the first wait only.
A wait holds nothing: it leaves keys and buttons alone. They stay held
until `key ... up`, `release` or `stop`. Documented in horizon-ctl's
key/wait help (and so the MCP tools) and docs/agent-control.md.
Fixes#469
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
GameRenderer.render3dHud draws DebugCrosshairRenderer after the hand and
the screen effects, with the hudFov projection, no bob and FogMode.NONE.
Horizon drew it among the world's on-top gizmos, so the axes bobbed with
View Bobbing.
Fixes#475
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Player.attack on a player: the host hits the guest on its crosshair, a guest
hits the host or another guest. The hit is worked out as the attacker
(Game::melee_damage, shared with attacking mobs) and lands on the victim as a
player_attack (blocking, armour, knockback to the guest's client), when the
pvp game rule allows it (ServerPlayer.canHarmPlayer).
Part of #435
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The hub's container menus now say where their slots live (lan::menu::Slots):
one block entity, both halves of a double chest (CompoundContainer, the right
half first) or the guest's own ender inventory (PlayerEnderChestContainer,
saved as EnderItems). The host's ender inventory stays the host's while a
guest acts. A Horizon client shows a server's chest-like menu with the rows
and title the server sent, and keeps its slots off a block that holds no such
container.
Part of #435
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The host now runs the rest of the server's body tick for each guest:
LivingEntity.baseTick's drowning and lava, fire blocks, suffocation
(isInWall) and the void (checkBelowWorld, 4 a tick, creative too), and per
move handleMovePlayer's jump (ServerPlayer.jumpFromGround's exhaustion) and
checkMovementStatistics (sprinting, swimming). The first move of a tick now
starts from the ground state before it, so take-offs are seen. Hurt sounds
are heard at the guest; lan status lists each guest's health and food.
Part of #435
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Since #372 a guest's commands need the world's Allow Commands as well as
the LAN's; the test's survival world had none, so every guest command (tp,
gamemode, setblock) was refused and the checks built on them failed.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Each mob sees the nearest player (host or guest), as Level.getNearestPlayer
gives it, and keeps a guest it targets or that hurt it in the last 100 ticks.
Its hits, pushes, fire and launches on that player run with the guest swapped
in; mobs' projectiles hit guests too, and knockback reaches the guest's client
(SetEntityMotion on its own entity).
Part of #435
Co-Authored-By: Claude Opus 5.5 <[email protected]>
vanilla_world_check.py (used by the light harness) defaulted to port 25671, so two parity runs on one machine failed with 'FAILED TO BIND TO PORT'. It now picks a free port unless --port is given. The save round trip, datafix and upgrade_world tools derived their port from the pid or a random number; they ask the OS for a free one like test_dedicated_server_parity does.
Fixes#461
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
A creative player with game master commands (Player.canUseGameMasterBlocks)
places command, chain and repeating command blocks and opens
CommandBlockEditScreen on them: the command, its previous output with the
O/X tracking toggle, Impulse/Chain/Repeat, Conditional and Needs
Redstone/Always Active, applied as handleSetCommandBlock does.
The block entity (CommandBlockEntity with BaseCommandBlock) saves and
loads its vanilla fields. CommandBlock.tick runs impulse blocks on a
rising signal, repeating blocks every tick while powered or always
active, and the chain in front (executeChain), honouring conditional
blocks, command_blocks_work and max_command_sequence_length. Commands
run through the command system as the block (position, facing, game
master level); their feedback becomes the block's last output and the
[@: ...] broadcast under command_block_output. Comparators read the
success count.
Fixes#444
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A mod's commands ran but the chat box parsed against vanilla's tree only,
so `/coords` showed red as unknown and never completed. The game now keeps
the chat's tree as vanilla's plus a literal per running mod's command (every
permission level, optional greedy `args`), rebuilt when mods start, reload
or stop.
Fixes#456
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla's AbstractSliderButton keeps its own position while held and
only snaps to the option's stepped value in resetValue (on release, or
600 ms after the last change for sliders that apply late).
Fixes#385
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A player seen with Bad Omen or Trial Omen turns a trial spawner ominous
(tryDetectPlayers' ominous branch): Bad Omen becomes Trial Omen, level
events 3020, the spawner's mobs vanish (3012) and the ominous config
starts over. While active, an ominous spawner drops an ominous item
spawner over a player or one of its mobs every 160 ticks
(spawnOminousOminousItemSpawner, getDispensingItems seeded per 30x20x30
cell); it floats, grows and turns for 60 to 120 ticks, then the item
falls or is fired down (OminousItemSpawner).
Zombies, piglins and zoglins read IsBaby from their tag, so the baby
zombie trial spawners spawn babies, and zombies save it.
The trial spawner harness covers the ominous branch, Trial Omen, the
discarded mobs and the item spawners' items, positions and delays.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Generated chunks now keep the block entities Horizon does not model
(decorated pots, brushable blocks) as placed, like a loaded chunk's, so a
trial chamber pot keeps its loot table and sherds. The loot table is
rolled into the pot's item the first time it is read
(RandomizableContainer.unpackLootTable: the seed, else the table's random
sequence, else the level random), so breaking the pot drops it. Pots
render their sherds from the saved block entity, and DecoratedPotBlock's
useItemOn / useWithoutItem insert items (wobble, sounds, dust plume,
game event) or wobble and refuse.
Part of #223.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
LevelEventHandler 3011-3021 (spawn, detect, eject, ominous and item spawner
particles and sounds, vault activation and deactivation), the trial
spawner's client tick (state particles, the spinning display mob, the
ambient sound) and the vault's client tick (idle smoke and flames, the
connection particles towards the player, the ambient sound).
Part of #223.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vaults tick in the game now, and an idle one pushes
state_updating_resumes_at to gameTime + 20 on every update, so Horizon,
which saves later than vanilla, stores a later value. Stage 3 already
treats it as volatile.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vaults are block entities now (loaded from saves and structure templates,
so trial chambers keep their ominous configs) and tick every game tick:
their state changes, sounds, level events and ejected rewards reach the
world, keys go in through VaultBlock.useItemOn before the item's own use,
and VaultRenderer draws the display item spinning inside the cage.
Part of #223.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
vault::Vault is VaultBlockEntity with its config, server and shared data:
load and save, Server.tick (display item cycling, VaultState.tickAndGetNext,
ejection, transitions with their sounds and level events) and
VaultBlock.useItemOn / tryInsertKey (keys, rewarded players, fail sounds,
the reward roll with its random sequence).
tools/test_vault_parity.py runs 3,000 cases against vanilla through a new
mode=vault of the natural spawn oracle and fixture.
Part of #223.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
EditWorldScreen's SystemToast goes through vanilla's ToastManager: it
slides in from the right over 600 ms (squared), shows for 5 s times the
notification display time, slides out, and plays ui.toast.in / out. The
in-game toasts' slide moves to a shared toasts module for it.
Fixes#384
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The world list's search, Edit World's name and Create New World's name and
seed kept only their String and appended to it. They now edit through an
EditBox (ui::Caret keeps the focused field's cursor, selection and scroll
between frames): EditBox.keyPressed's moves and deletions (Ctrl by word,
Shift to select), Ctrl+A/C/V/X on the selection, onClick placing the
cursor under the pointer (Shift extends, a double click within 250 ms
selects the word) and onDrag selecting.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
LivingEntity.drop swings the main hand for every drop, so throwing a
slot's item with Q over the inventory, a container or a creative tab
plays the swing (the inventory's player model shows it). Horizon only
swung for the drop key in game; Player.drop's port now swings itself.
Fixes#442
Co-Authored-By: Claude Opus 5.5 <[email protected]>
LevelLoadingScreen while the player waits for its chunks: the title's
panorama, blur and menu background when a world opens (Reason.OTHER),
the nether portal sprite after a nether portal, Loading terrain... above
the 17x17 chunk status map (ChunkLoadStatusView, radius 8, COLORS) and
the 200x2 progress bar easing towards LevelLoadProgressTracker's value.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
KeyboardHandler.keyPress sends a key repeat (InputConstants.REPEAT) to
Screen.keyPressed and KeyMapping.click like a press, so holding key.drop
keeps throwing the held item, the hovered slot's item in a menu, and
the hovered item of a creative tab. Horizon dropped every repeat; the
drop key's now reach key_pressed. horizon-ctl `key KEY repeat` sends
one auto-repeat, for scripted checks.
Fixes#443
Co-Authored-By: Claude Opus 5.5 <[email protected]>
WorldOptionsScreen in a HeaderAndFooterLayout: the title in the 33 pixel
header, the content in a ScrollableLayout over the in-world list
background between the separators, the footer buttons at h - 27.
Bold underlined category headings (FocusableTextWidget), the difficulty
button 130 wide beside its lock, Edit Game Rules... and Restrictions...,
the port EditBox always shown with its hint and error colour, and the
buttons' active states from updateButton and the update*Button methods.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
pre_present_notify ties winit's RedrawRequested to the compositor's
frame callbacks on Wayland, one per refresh, so a window drew at the
monitor's rate whatever the present mode. Only Fifo (VSync) asks for
that pacing now; Mailbox and Immediate draw as fast as the frame limit
allows, like vanilla's GLFW window.
Fixes#441
Co-Authored-By: Claude Opus 5.5 <[email protected]>
CenteredIcon draws its 15x15 sprite at x + 20/2 - 15/2 = x + 3 (integer
division), y + 3, with the widget's alpha only: Horizon drew it at +2
and tinted it grey.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
WorldListEntry.extractContent dims a hovered row's icon (0xA0909090) and
draws world_list/join over it, join_highlighted under the pointer;
mouseClicked joins the world when the click lands on the icon.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Allow Commands is the integrated server's setting; a dedicated server's
guests run what their permission level allows (dedicated_server parity).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
When no folder has the pinned version and the selected Microsoft account
owns Minecraft: Java Edition (the mcstore entitlement, asked again with the
saved token), Prepare downloads it the way other launchers do: the version
JSON from piston-meta by its pinned sha1, the client jar, this system's
libraries (the JSON's os.name rules) and the asset index with its objects
(the sounds), each checked against its sha1, written whole, 8 at a time,
into the data folder's minecraft/ in the official layout. The next
preparations find it there first. An asset that fails is a line in the log;
anything else is an error.
Not signed in, or signed in without the game, the not-found error says so as
its last line. Ownership stays required: there is no other path to the
download, and nothing of Mojang's is part of Horizon.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Brigadier's getCompletionSuggestions lists every child whatever its
requirement (command_parse compares it at permission 4, 2 and 0); only the
tree the server sends the client is pruned (Commands.sendCommands). The
player's chat now asks client_suggestions, which skips what the player
may not use; guests' ask_server requests keep brigadier's.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The official launcher crashes on some Linux systems (rdbt's Arch), and
Horizon only read its layout. Preparing now finds the pinned version in the
chosen folder (or a folder above it, so an instance folder works), then in
every launcher's default folder on Linux (Flatpak included), macOS and
Windows:
- the official launcher: versions/<v>/<v>.jar
- Prism Launcher, MultiMC: the jar their meta/net.minecraft/<v>.json names
(a Maven name in libraries/), libraries/com/mojang/minecraft/<v>/...
- the Modrinth App: meta/versions/<v>/<v>.jar, meta/libraries
- ATLauncher, GDLauncher (data/, Next's datastore/):
libraries/net/minecraft/client/<v>/client-<v>.jar
The client jar is checked against the lock's sha1. horizon-prepare reads the
official layout and pins the version JSON's own sha1, so files found
elsewhere are linked into that layout in the data folder
(linked-minecraft/), with the pinned version JSON from piston-meta (by its
sha1, checked) when the launcher keeps only its own format or a rewrite.
Nothing else is copied or downloaded. Every preparer, old Python ones
included, reads the linked folder as an official one.
When nothing is found, the error lists the chosen folder and every installed
launcher's, with what each lacks.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Prepare used to stop with "Minecraft 26.3 is not in <dir>. Play 26.3 once from
the official launcher, or choose another folder." followed by an unrelated log
line ("Checking Java and your Minecraft install"). It now lists every folder it
looked in with what each one lacks (no such folder, no versions/26.3/26.3.json,
no 26.3.jar) and the two ways out, one per line. A missing client jar is caught
here too, instead of later inside horizon-prepare.
Only a preparer process's "Exited with ..." gets the last log line appended;
the launcher's own errors are full sentences.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A folder holding shaders/, or a .zip, in shaderpacks/ now loads from the
Shader Packs screen: every planned begin, prepare, deferred, composite
and final pass is translated and drawn each frame over Horizon's world,
with Iris's formats, clears and buffer flips.
The targets keep OpenGL's layout (the translator flips gl_Position and
remaps its depth), the world enters as OpenGL depth, and the pack gets
OpenGL matrices. Textures share four samplers by kind (Metal allows 16
per stage), and isnan/isinf test bits (Metal's fast math folds x != x).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
GitHub's releases page sorts a day's tags as text, so v2026.10.15.9 showed
above v2026.10.15.18 (reported by rdbt on Discord). Tags now count .01, .02...
Every reader (release_train.tag_key, the launcher's compare_versions) compares
the parts as integers, so older unpadded tags still order correctly.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Delete opens JoinMultiplayerScreen's ConfirmScreen (title, wrapped
warning, Delete and Cancel; Escape answers no) and only removes the
server on yes. Direct Connection puts its label at y 100 and the box at
116 like DirectJoinServerScreen, and Add/Edit Server its boxes at 66 and
106 like ManageServerScreen, labels one pixel in.
Fixes#380, #381
Co-Authored-By: Claude Opus 5.5 <[email protected]>
OnlineServerEntry draws the server's 64x64 favicon (from servers.dat, then
from each ping, saved back like updateServerList), the description as a
styled component with legacy codes, formatPlayerCount's dark grey with
grey numbers, and on hover the join and move arrows over the icon. The
icon's right half joins, its left quarters move the server up or down.
Fixes#365, #378, #370
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The begin, prepare, deferred, composite and final passes of a program
folder: what each writes (the later of DRAWBUFFERS and RENDERTARGETS),
which side of every ping-ponged colour target it reads (Iris's
BufferFlipper, with flip.* overrides), mipmapped targets, scale.*, and
the targets' formats, clearing and sizes from the consts of every pass
in Iris's order. The preprocessor now keeps the directive comments of
active code, as Iris's does, so a DRAWBUFFERS in an #if'd out branch
doesn't count.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Java .properties parsing, the two reads Iris makes (the raw file for
screens, sliders and profiles; the preprocessed one, with enabled options
and the environment's macros defined, for everything else), option
screens, profiles and the current one, program.*.enabled through a port
of Iris's BooleanParser, image formats for shadowcomp's writeonly
images, and blend overrides.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The legaliser of docs/shader-packs.md step 1, on a syntax tree (glsl-lang)
instead of the spike's tokens: attribute/varying to located in/out with
matrix and array varyings split, loose uniforms in one std140 block that
extends across a pass, combined samplers split into texture + sampler
bindings (through function parameters, respecting locals that hide them),
compatibility built-ins and Iris's core-profile names declared, legacy
texture and shadow calls, hoisted global initialisers, sized unsized
arrays, isnan/isinf/modf helpers, and dead code after a jump dropped (the
spike's 'naga constructor bug' in BSL was code after discard). naga then
parses, validates and writes WGSL. The programs of one pass share a Link
so locations, bindings and the uniform layout agree.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A new crate for the OptiFine/Iris translation (docs/shader-packs.md, order
step 1, first half): open a pack folder or zip (shaders/ found as Iris finds
it), list its programs per dimension folder, build the #include graph,
discover options per include component (OptionAnnotatedSource rules: boolean
defines referenced by #ifdef, value defines and configurable consts with a
[...] list, conflicting defaults dropped), write the player's values into
the lines, and expand Iris's standard macros with pp-rs, keeping line
numbers so errors name file:line.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The default debug profile enables 3d_crosshair while the overlay is open:
GameRenderer draws DebugCrosshairRenderer's axes one block ahead of the
camera (0.01 x GUI scale long; X red, Y green, Z light blue, 2 px over a
4 px black line) and Hud.extractCrosshair skips the 2D crosshair and its
attack indicator. Hidden with reduced debug info, in third person and
with F1.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The overlay is drawn after screens and toasts, so it covers menus, and on
the title and options screens too (Gui.extractRenderState); F1 hides it
only in game. It lays out at the debug GUI scale (debugGuiScale, default
half the largest automatic scale) and shows the default profile's entries
sorted by id, split into vanilla's two columns: fps, version, position,
section position, performance impactors, system specs (CPU, display,
window, GPU, backend) and the integrated server's tick time.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
GraphicsPreset.apply also sets render distance (8/16/32), simulation
distance (6/12/12), chunk update priority, smooth lighting and menu blur,
and its cloud range (32/64/128) and Fabulous entity distance (1.25)
differ from what Horizon set. Dragging the classic preset slider only
moved the label: the preset now applies whenever its value changes
(Options.applyGraphicsPreset), in both menu styles.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The single-player owner was always at permission level 4. Like
IntegratedServer.getProfilePermissions, the owner now gets OWNER (4)
only when the world allows commands, else ALL (0): /gamemode, /give and
friends answer "Unknown or incomplete command" and are not suggested
(the client's tree only holds usable nodes, Commands.sendCommands).
allowCommands is read from level.dat, set by World Options, and worlds
made without the Create New World screen take its default (Creative only).
The server console, the command harness and horizon-ctl `command` keep
the owner's level.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Save and Quit to Title kept boats, minecarts, armor stands, item frames,
paintings and other non-mob entities in memory. Opening the world again
loaded them from disk a second time and the next save wrote both copies,
so each reopen doubled them until the game crawled. Closing and opening a
world now let go of every entity of the level (MinecraftServer.stopServer
unloads them with the level). tools/test_world_reopen_parity.py reopens a
world three times and checks its minecarts stay the same.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The host drew each guest with the host's own armour (player_armor_layers),
and a guest ignored ClientboundSetEquipmentPacket, so it drew the host and
other guests bare-handed and without armour. Player models now carry what
they wear in extra.head..feet: the host fills it from the guest's armour,
the guest from SetEquipment (main and off hand too, for mobs as well), and
both draw HumanoidArmorLayer from it. The host also draws what a guest holds.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The player model (third person, the inventory's preview) was only ticked by
the single-player tick; on a server it was never created.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Two Horizon games under horizon-ctl, each with its own scratch home: the
host opens a survival world to LAN with commands, the guest joins by
address, and each check (tab, self, armour, nametags, chest, crafting,
damage) is reported on its own with screenshots of both sides.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A guest could not craft: the host's inventory menu for guests had no
crafting slots (0-4 were always empty), and a crafting table the guest
used switched the host's own grid to 3x3 and opened nothing for the guest.
The guest's inventory menu (window 0) and a crafting table window
(minecraft:crafting) now run the host's vanilla-checked menu engine
(horizon_client::menu, Kind::Inventory / Kind::Crafting) over the guest's
inventory, kept between clicks so drags work, with the usual click
corrections. The 2x2 grid and its result are real slots of the guest's
inventory menu, so the slot sync carries them. Closing a menu puts its grid
back into the inventory (dropped if full); a guest leaving keeps it too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
No player had a name tag: only mobs with a custom name drew one. The other
players now get theirs on both sides (the guests on the host, the players a
server sends on a client), shown as LivingEntityRenderer.shouldShowName
allows: no custom name or crosshair needed, a crouching player's only within
32 blocks, team visibility and colours by player name, the below-name score.
The host also joined its own LAN world as "Player": the guests' player
list, name tags and chat, the MOTD and the offline UUID now use the host's
profile name, as IntegratedServer does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Nothing hurt a guest: the host only set a guest's position from its move
packets, so a fall did nothing, and their body never ticked on the server
(no damage clock, no regeneration, no food tick).
Each tick the host now runs a guest's server-side body like ServerPlayer:
the damage clock and Entity.baseTick's fluid phase, then
ServerPlayer.doCheckFallDamage for every move packet accepted that tick
(the supporting block, the fall distance and the landing, through the same
landing code as the host's own falls, so water resets a fall and lava
halves it), then peaceful regeneration and the food tick.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A chest a guest opened opened on the host only, and never closed: the
host ran the guest's open as its own player's (one opener, a local block
event), nothing was sent to the guests, and the guest's close was not
counted. Chests now count their openers like ContainerOpenersCounter (the
host and each guest; a guest's close, a menu it replaces and its leaving
take it off) and the block event goes to the guests too (shulker boxes as
well). A guest's client handles ClientboundBlockEventPacket and ticks its
block entities (ClientLevel.tickBlockEntities, without the server copy's
bell sound and gateway ticks), so lids move on its screen.
The control socket's block reply gives a chest's or shulker box's lid
openness (lid, 0..1).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
In singleplayer the middle button did nothing outside creative, and in
creative it overwrote the selected hotbar slot. It now runs the server's
handlePickItemFromBlock (tryPickItem): a stack of the block's item
already in the hotbar is selected; one in the main inventory is swapped
into getSuitableHotbarSlot (the first empty hotbar slot from the selected
one, else the first unenchanted); only creative adds the item when none
is there, moving what the slot held to a free slot. Servers already got
the packet and do this themselves.
Fixes#367.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
GuiGraphicsExtractor.textHighlight paints an inverting white fill, then
adds pure blue (GUI_TEXT_HIGHLIGHT, additive). Horizon only inverted, so
selections lacked the blue: black text on a light box, not blue text.
The anvil's name box skips the inversion (setInvertHighlightedTextColor
false), as in AnvilScreen.
The creative search and anvil name box now draw like EditBox: the cursor
blinks from when the field took the focus (300 ms on, 300 ms off), turns
into a bar once the field is full, and Select All highlights the value.
Sign lines draw their Select All highlight too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The Extras page's Inventory section gains "Sort buttons" (sort, off) and
"Sort key" (sortKey, unbound). A small button right of a chest, barrel,
ender chest or shulker box title sorts the container; one right of the
"Inventory" line sorts the main inventory (the hotbar stays). The key
sorts the part under the cursor.
inv_sort::plan merges partial stacks, then orders the stacks by creative
tab order (the search tab's), then name, full stacks first, using only
left PICKUP clicks with an empty hand: the game sends them as ordinary
clicks, so vanilla servers accept them. A unit test replays the plans
through the menu engine on 40 random chests.
The Extras page binds any extra key the way it bound the zoom key
(Item::key_name, Extras::set_key).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The Extras page gains an Inventory section with "Mouse Tweaks controls"
(mouseTweaks, off) and "Scroll wheel moves items" (mouseTweaksWheel). In
the player inventory and chest-like containers, locally and on servers:
- right-drag holding a stack puts one item in each slot entered, at once;
- left-drag holding a stack gathers the slots of the same item into the
hand, while vanilla's drag covers one slot (a drag over empty slots
still splits as vanilla does); with Shift they are quick-moved;
- Shift + left-drag with an empty hand quick-moves every slot entered;
- the wheel over a slot moves one item to the other side (down) or
brings one back (up).
Every action is ordinary slot clicks: the menu engine's, or ContainerClick
packets for any slot (remote_slot_click, factored out of remote_click).
With the extra off each hook returns at once.
horizon-ctl gains drag-at and scroll-at, which send the window events of
a mouse drag and of wheel steps, to test these and vanilla's drags.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
CreativeModeInventoryScreen.slotClicked moves the search box cursor to
the end and sets the highlight to 0 on any click of a display slot, so
the next typed character replaces the whole query. Horizon kept the
query and appended to it.
horizon-ctl's type now edits the creative search as typing does (the
selection and the 50 character limit apply) instead of appending.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
While a held stack is dragged over slots, each dragged slot shows the
stack it would get over a white highlight, its count yellow when the slot
caps it, and the cursor shows what stays in hand (extractSlot,
recalculateQuickCraftRemaining, extractCarriedItem); a drag over a single
slot hides that slot's item. Slots join the drag by
shouldAddSlotToQuickCraft (same item or empty, placeable, canDragTo, and
not more slots than carried items for a right drag), so the preview counts
the slots the menu will fill.
The menu engine computes both (Menu::quick_craft_accepts and
quick_craft_preview) for the inventory, crafting table, chests, hopper,
dispenser, shulker box, furnaces, crafter, loom, trading and creative
screens; the slot and carried-stack drawing of those screens goes through
draw_slot_item and draw_carried.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
AbstractContainerScreen.mouseReleased: a Shift double click QUICK_MOVEs
every slot of the clicked slot's container that holds the item the last
Shift click moved (lastQuickMoved), when canTakeItemForPickAll allows the
clicked slot. Menu::quick_move_all runs it on the menu engine;
test_menu_click_parity generates it as a QUICK_MOVE_ALL op that the oracle
runs with the real Slot.mayPickup, Slot.container and
AbstractContainerMenu.canItemQuickReplace.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The item tabs handled clicks on press with their own rules: Shift sent the
item to the inventory, the hotbar row swapped with the carried stack, a
carried stack clicked on the tab was always deleted, and there were no
drags. They now go through the same mouseClicked/mouseReleased as the other
container screens, over ItemPickerMenu:
- display slots: slotClicked's CONTAINER branch (creative::click_display):
an empty hand takes the shown stack (a full one with Shift), the same item
grows by one, fills with Shift or shrinks with a right click, another item
is deleted or loses one; number keys fill a hotbar slot, Q drops copies;
- the hotbar row: the menu engine's new ItemPicker kind, whose
quickMoveStack deletes the stack (Shift-click on the hotbar from an item
tab), and whose display slots take no drags or PICKUP_ALL.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Tests pin player 1 (0x04), the default blue (0, 0, 128) and the instant
bit (0x20) to hid-playstation.c and SDL; the docs say what Horizon resets
on quit (more than SDL or the kernel do) and that the mute LED is never
touched.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- The pad's audio device is also found by Windows' "Wireless Controller"
endpoint name (the four channels rule out a DualShock 4); a name test.
- docs/controllers.md: the haptics need reports without the rumble flags
(SDL, nondebug/dualsense, dse-haptics), which Horizon now always sends while
they are open; Windows needs the Quadraphonic speaker setup; on Linux ALSA
can't open the card while PipeWire holds it; Bluetooth haptics (report
0x32, SAxense) are documented and not done.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- A loaded crossbow and the boosts (rocket, pig, strider) get a weapon
catch that breaks in the zone where the trigger threshold presses, so the
click is the shot (the boost was a feedback wall that never gave way).
- Draws start at zone 2 and stop at 6 of 8 (held for as long as you like).
- A shield in the use hand weighs (feedback 2, 3); disabled by an axe, the
trigger locks (feedback 1, 8), as Deathloop does for a jammed gun.
- Mining buzzes slower on harder blocks (40/30/20/10 Hz).
- A 0 Hz vibration frees the trigger (Nielk1's factory); tests pin the
bytes to DSX++'s and dualsensectl's examples.
- docs/controllers.md: the table with the effect of each action and why.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- Every report carries the motors; still motors send no rumble flags, as SDL
stops rumble (a report without them ends the emulation and re-arms the
audio haptics), so a trigger change no longer cuts a vibration.
- Legacy vibration at half strength (SDL); the Edge always takes V2; an
unreadable firmware version falls back to legacy.
- The first report after opening clears the motor/trigger attenuation
(flags 1 bit 6, byte 36; dualsensectl, pydualsense).
- The LED reset goes in a report of its own before any colour
(hid-playstation's dualsense_reset_leds); over Bluetooth the colour waits
for the connection animation (SDL's sensor clock check).
- Bluetooth output only where gilrs keeps reading the pad (Linux), else with
HORIZON_DUALSENSE_BT=1: any output switches the pad to its full report.
- USB is preferred when the pad is on both.
- Tests pin the reset report and its CRC (zlib), the flags and the clock offset.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The lightbar shows the player's health (green to amber to red) or the
dimension, at half brightness, chosen in the controller options (Lightbar:
Health / Dimension / Off; health by default). The player LEDs show player 1.
When Horizon lets go of the pad, the lights it changed go back to a host's
defaults.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Each rumble event plays a short shape instead of one flat block: a hit
starts at full strength and dies away, an explosion rolls on, a totem
swells, a broken block is a tick. They drive the motors' emulation in the
HID report.
On USB the pad is also a four-channel audio device whose channels 3 and 4
drive its actuators: with HORIZON_DUALSENSE_HAPTICS=1 the shapes play there
as waveforms (80 Hz and 200 Hz tones for the heavy and light parts),
channels 1 and 2 silent, until that path is confirmed on a real pad.
docs/controllers.md says what each OS and transport allows.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Each trigger feels the action bound to it: resistance growing while a bow
draws, a trident charges or a crossbow loads, a gun trigger's click on a
loaded crossbow, a wall where pulling boosts (a rocket while gliding, the
stick of a ridden pig or strider), and a light buzz while mining, scaled by
the block's hardness. Free in menus, paused, unfocused, on another pad,
with controllers off, and on quit. On by default, "Adaptive Triggers" in
the controller options turns them off.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- First run (and when CONSENT_VERSION changes): a question over the launcher
with four switches, all off, what each sends in plain words, what is never
sent, "View what would be sent" (the exact JSON), and "Send nothing" /
"Save choices".
- Settings → Data collection: the same switches any time, the install ID and
what waits, "View what would be sent" and "Delete my data" (server deletion,
outbox cleared, new install ID).
- The outbox is sent at start and on close (at most 3 s), only when there is a
server (none yet) and only what the player allows.
- CLI `horizon-launcher data status|allow|deny|show|delete|send|bug|benchmark`
and MCP tools data_collection, set_data_collection, data_preview,
delete_my_data, report_bug. Bug reports and benchmark uploads are printed for
the player to confirm; logs are scrubbed and lose chat lines.
- horizon-telemetry joins the launcher's paths (launcher.yml, release_train,
pr_ready).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The horizon-dualsense crate builds the DualSense's output report (USB 0x02,
Bluetooth 0x31 with its CRC-32 seeded with 0xA2) and writes it with hidapi
from a thread. The game sends its rumble pulses through it when the active
pad is a DualSense, so vibration now works on macOS too, and falls back to
gilrs force feedback when the pad can't be opened. Quitting frees the
triggers and stops the motors.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
SpriteContents.AnimationState.drawToAtlas mixes an animation with
"interpolate": true towards its next frame every tick
(animate_sprite_interpolate.fsh: mix(current, next, subFrame / time),
per mip level, into the 8-bit atlas). Horizon stepped the frames: the
sprite upload now mixes the two frames' mip chains the same way and
re-uploads every tick while the mix changes.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The design sections now describe the implementation: ids for rows,
switches and segments, the redraw wait, wait-for --since and event KIND,
the events log (subscribe comes next), the launcher proxying through
horizon-ctl mcp. A before/after table and the PRs of each step replace
the plan.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The workstation screens (anvil, grindstone, stonecutter, smithing,
enchanting, cartography, loom, beacon, villager trading, brewing stand),
the horse/llama/camel inventory and the creative item grid read Shift from
the sneak mapping (`keys`). Behind a screen a Shift press never reaches the
mapping (KeyboardHandler.keyPress: only releases do) and Right Shift isn't
mapped at all, so shift-click there worked only when Shift was already held
before the screen opened. They now read `has_shift_down` like
AbstractContainerScreen (`InputWithModifiers.hasShiftDown`), and the recipe
book reads Ctrl the same way (`Minecraft.hasControlDown`).
`status` reports `physical_keys` so scripts can check what the OS delivered.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A Shader Pack row cycles Off and every folder in shaderpacks/; the game
loads the pick on the next frame. Not part of the presets, off by default.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Converting a proto chunk for saving (VanillaTerrain::saved_proto copies it
block by block into a column, then encodes it) ran on the main thread: in
Streaming::save during the autosave, and in Streaming::update for every far
proto unloaded while exploring. The protos are now checked out (like a
decoration job does) and saved by a worker, ahead of every other job, then
handed back. The explicit save and quitting still save them in place, after
waiting for the saves in flight; dropping the streamer waits for them too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A light job for a chunk that was already lit (a neighbour loaded, or a block
changed in an unlit neighbour) dirtied every section of the 3x3 columns
around it: 216 remeshes, whether or not any light changed. It now writes only
the sections whose light changed and dirties the sections around the changed
positions, as the incremental relight already does. A chunk lit for the
first time still dirties its 3x3 neighbourhood.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
GameLevel::heightmap found getHeight by testing every block from the top
of the world down (with a block tag lookup each), for every spawnable chunk
every tick: a quarter of the main thread in a village. Sections holding
only air are now passed over whole; air counts for no heightmap, so the
height is the same.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
With shaderPack set in horizon-graphics.txt, the world renders into
world_post's target with its depth kept, each composite stage of the
pack (gfx/pack_stages.rs) draws a full-screen pass reading the scene
colour and depth, the previous stage and the Frame uniform (camera,
matrices, time, sun and moon, sky and fog colours and distances,
weather, camera state, dimension), and the last output goes to the
screen through the usual copy/FXAA, under the hand, screen effects and
HUD. The game loads the pack from shaderpacks/, checks its files twice
a second and on F3 + T, and keeps the last good pack on errors, which
go to chat. Without a pack nothing changes: no target, no extra pass,
depth discarded as before.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
shader_pack.rs reads shaderpacks/<pack>/pack.toml, the player's option
values (shaderpacks/<pack>.txt) and each composite stage's WGSL, puts
the api 1 prelude (bindings, the Frame uniform, depth helpers) before
it and has naga parse and validate it, with errors pointing into the
pack's own file. Options are WGSL override constants. A file stamp
tells when the pack was edited, for hot reload. Nothing draws with it
yet. examples/shaderpacks/horizon-sample is the MIT sample pack.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
wait-for screen NAME|widget ID|loaded|chat TEXT|event KIND replies once
the game gets there (--timeout, default 30 s, ends in an error of kind
timeout; --since SEQ counts chat and events from an earlier point), so
scripts and agents stop polling. events [--since SEQ] lists what
happened, oldest first, numbered: screens, chat lines, deaths, respawns,
items gained or lost, dimensions, worlds opened and closed. The state is
compared once per event loop turn and only under --control, so watching
never changes the game. respawn now says when the death screen's button
is not active yet instead of clicking it to no effect.
horizon-ctl and its MCP gain both; tools/agent_control_e2e.py gains an
events scenario.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
options-get [KEY...] gives every options.txt option by its vanilla key
with its value and what it accepts (type, range, step or choices; enums
by name, key bindings as key.keyboard.w). options-set KEY VALUE sets one,
applies it at once and saves options.txt as the options screen does
(language and resource packs reload too); a value the option refuses is
an error naming what it accepts, and the option keeps its value. After a
set, ui reads the redrawn screen.
horizon-ctl and its MCP gain both; tools/agent_control_e2e.py gains an
options scenario.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Without an options.txt the game ran unlimited (260), while an options.txt
without maxFps, like vanilla, gives 120. With VSync off (or rendering
offscreen) that kept a core busy drawing frames nobody sees.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
An agent now needs one MCP server. horizon-launcher mcp runs the selected
version's horizon-ctl mcp as a child and offers its tools as game_<name>
(all but launch, which launch_client covers), passing calls and image
results through. launch_client starts the game with a control socket in
the launcher's data folder, takes world, creative and offscreen, and
returns once the socket answers. A game tool called while no game listens
says to call launch_client.
tools/launcher_mcp_e2e.py checks it end to end against this checkout's
builds in a scratch launcher home.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The Discord server banner (page + PNG), the server icon render and Kira's avatar (SVG master +
render) now live in the private kira repo (ludovic111/kira#1). xtask icons no longer renders them;
the brand README points there, and CONTRIBUTING states what belongs in this repository.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
`ui` lists every widget the last frame drew (title menus, pause, options
in both menu styles, inventories and menus) with an id, its kind, label,
bounds and value; `click ID` clicks one through the same window events
as the mouse, `set ID VALUE` moves a slider along its track. Ids come
from labels (FOV: 70 is fov), world/server/language rows carry their
names, Horizon switches, sliders and segments their row's caption, and
inventory slots are slot-ID with their stacks. After a click, the next
widget command waits for a redrawn frame.
horizon-ctl and its MCP gain ui, click and set, and launch --offscreen.
tools/agent_control_e2e.py gains a ui scenario and runs offscreen on
macOS; both scenarios now handle Singleplayer opening Create New World
when there are no worlds.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
docs/mods.md: the sandboxed WebAssembly runtime, capabilities, hooks, budgets,
authoring, versions, multiplayer and distribution, with measured hook costs.
docs/research/java-mod-compat.md: why Fabric/NeoForge jars cannot run, data
import, which mods to port natively first, a porting guide outline.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Manifest and .hmod reading, capability checks against the component's imports,
fuel and memory limits per call, a trapping mod stopped alone, hot reload.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
horizon-mod-sdk wraps wit-bindgen: a Mod trait with default hooks and export_mod!.
coords-hud shows coordinates, facing and the time of day, with /where and /coords.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
tools/port_report.py turns a diff into work items: rank 1 blocks the switch (facts,
packets, DataFixers, data component types, report formats, tables to regenerate),
rank 2 and 3 one item per Horizon file that ports changed vanilla code (with or
without a harness), rank 4 new content, rank 5 summaries. version_port.py report
writes report.md and report.json; port prepares, diffs and reports in one go.
The class map now resolves names of both versions. The doc records the 26.2 -> 26.3
replay against Horizon's own code.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
tools/class_index.py hashes every method's bytecode with constant-pool references
resolved and branch targets as instruction indices, lambdas folded into their
declaring method; class diffs list moved classes. tools/vanilla_map.py maps vanilla
classes to the Rust files whose comments name them and the harnesses whose oracles
or probes reach them. version_port.py prepare indexes, diff reports both.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
detect reads Mojang's manifest; prepare downloads and verifies a version into the
version store (outside the repository), runs its data generator and reads the
jar's own version facts; diff compares facts, packets, registries, blocks, item
components, commands, data packs, assets and lang.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
What the launcher, the game socket, horizon-ctl and the two MCP servers
can do today, and one model for what comes next: shared names and JSON
shapes, a widget tree with actions by id, options get/set, waits, an
event stream, the launcher's MCP proxying the game, errors with a kind,
and the safety rules.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Trial spawner block entities are kept (generated chunks and saves), tick on
the server with the level random and random sequences, spawn persistent
mobs, eject their rewards as items and set their block state, with their
level events and shutter sounds.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
mobs::trial_spawner: TrialSpawner and its state data, the registry
configs (data/*/trial_spawner), the six states, player detection with the
visual line of sight, mob spawns and reward ejection, read and saved as
vanilla does (a field that fails to decode takes its default alone).
tools/test_trial_spawner_parity.py ticks 2,500 random trial spawners in
VanillaNaturalSpawnOracle and natural_spawn_fixture and compares states,
spawns, events and ejected items line by line.
Also: copper grates have no visual shape; a slime's or magma cube's
obstruction box scales with the Size its spawn data sets; LootData::roll_split
for an empty-context getRandomItems.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Scraping and waxing off copper already swing the arm, show the scrape and
wax-off particles and play the axe sound on main (#59, #104). Two gaps were
left in BlockTransformer.transformBlock:
- playerHasBlockingItemUseIntent: with a shield in the off hand, right-clicking
with an axe raises the shield instead of scraping, unless sneaking.
- The block_change game event, which sculk sensors pick up.
Fixes#222
Co-Authored-By: Claude Opus 5.5 <[email protected]>
FirstPersonHandsAndItemsRenderer draws both hands: the off-hand item sits
on the other arm with its own equip height and swing, uses its own pose
when it is the hand in use (eating, blocking, drawing a bow), and
evaluateWhichHandsToRender hides a hand while a bow or crossbow is drawn
or a charged crossbow is held in the main hand. Only the main hand's empty
arm shows, as in vanilla.
Part of #276
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Minecraft.startUseItem tries each InteractionHand in turn: useItemOn, then
useItem, stopping at the first that succeeds or fails. Item uses now act on
the hand that started them (Game::hand), so blocks, food, potions, buckets,
spawn eggs and the rest work from the off hand, the swing and the
re-equip animation follow that hand, and the empty-hand block use
(useWithoutItem) stays main-hand only. The respawn anchor passes on the
main hand when the off hand holds glowstone, as vanilla does, replacing the
special case for it.
The use_item_on harness now puts items in the off hand too.
Part of #276
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Player.getProjectile: what the weapon supports in the off hand, then in
the main hand (getHeldProjectile; crossbows also take firework rockets
there), else the first arrows in the inventory. Crossbows only looked
for arrows in the inventory, so a rocket in the off hand was never
loaded. A charged rocket is shot at an angle from 0.15 under the eyes
(CrossbowItem.createProjectile), every shot of a load holding one at
power 1.6 (getShootingPower), for three durability (getDurabilityUse).
Part of #276
Co-Authored-By: Claude Opus 5.5 <[email protected]>
AbstractArrow.onHitEntity: an entity that takes no damage from the arrow
(a raised shield, invulnerability) turns it back with
ProjectileDeflection.REVERSE at power 0.2: a tenth of the speed, a half
turn give or take ten degrees from the arrow's own random; one left
almost still drops its pickup and is discarded. Arrows used to vanish.
The arrow tick now stops at the entity it hits: the game hurts the
entity, then arrows::finish_entity_hit spends the arrow (with its hit
sound, from the arrow's random) or turns it back, and ends the tick
(drag, gravity, fluids). The damage is taken from the speed at the hit,
the crit bonus from the arrow's random, the entity search follows
ProjectileUtil.getManyEntityHitResult (the box first, then
computeMargin's margin, which needs tickCount), Punch pushes only what
was hurt, and a refused burning arrow leaves the player's fire as it
was. Mobs' refusals keep the old ending until their hurt results follow
hurtServer.
The arrow harness gains an entity in the way that takes or refuses the
damage: squarely, grazing, slow enough to drop.
Part of #277
Co-Authored-By: Claude Opus 5.5 <[email protected]>
LivingEntity.hurtServer's playHurtSound goes through Player.playSound,
which leaves the player itself out; the player hears its hurt sound from
the client's handleDamageEvent, and a blocked hit sends BlocksAttacks'
block sound (onBlocked) instead of the damage event. Queue the local hurt
sound on the damage broadcast rather than on the server's hurt sound.
Part of #277
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- The fishing rod switches to its cast model while its hook is out
(FishingRodCast), in first person, third person and the hotbar.
- FishingHook.catchingFish's particles: splashes where fish tease the bait,
the fishing-particle trail and bubbles of a fish swimming up, and the
bubble and fishing burst when it bites.
- FishingHookRenderer: the line is 16 black `lines` segments of the window's
line width (2.5 px at 1080p) from a quarter block above the hook to the
rod hand (on the near plane in first person, beside the body in third),
and the bobber is centred on the hook.
Fixes#265
Co-Authored-By: Claude Opus 5.5 <[email protected]>
MinecartSoundInstance follows every minecart in the level (entity.minecart.riding,
louder with speed, silent when the tick rate is frozen), and getting in one starts
RidingMinecartSoundInstance's two inside sounds (normal and underwater). Hitting a
minecart or a boat plays Player.attack's strong/weak (and knockback) sounds.
The sound_engine harness now replays both instances against the vanilla client.
Fixes#227
Co-Authored-By: Claude Opus 5.5 <[email protected]>
DebugScreenOverlay.toggleLightmapTexture and showLightmapTexture: the 16x16
lightmap at 64x64 on a black border, 2 pixels from the corner, sky light up
and block light right, as vanilla's flipped blit. It shares the bottom of the
screen with the F3+2 and F3+3 charts, so each switch turns the others off.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
CreateWorldScreen as in 26.3: a MenuTabBar (tab sprites, header
separators, underline) over the transparent tab header, each tab's
GridLayout aligned at (0.5, 1/6) between it and a 33-pixel footer with
Create New World and Cancel.
Game: the world name box with its save-folder tooltip, Game Mode
(Survival, Hardcore, Creative), Difficulty and Allow Commands, locked
in Hardcore. World: World Type (Default, Superflat), Customize, the seed
box with its hint, and the Generate Structures / Bonus Chest switches.
More: Game Rules, Experiments and Data Packs. Tooltips follow vanilla's.
The choices reach the new world: hardcore, difficulty and allowCommands
in level.dat, and world_gen_settings with the flat preset, structures and
bonus chest written before streaming starts. Re-Create reads them back.
Fixes#281
Co-Authored-By: Claude Opus 5.5 <[email protected]>
DebugScreenOverlay's FpsDebugChart, TpsDebugChart, PingDebugChart and
BandwidthDebugChart with vanilla's scales, colours and places: 240 samples
(LocalSampleLogger), F3+2 and F3+3 toggles that show the overlay and exclude
each other, the overlay's "Debug charts" lines, and the play-state ping
request and pong response packets the ping chart uses on servers.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
SelectWorldScreen's list now sits at vanilla's 49..h-60 with the list
background, header and footer separators, white selection outline and
the widget/scroller sprites at the row's right + 8. The scroller follows
the pointer while held (AbstractScrollArea.mouseDragged), the wheel moves
half a row, and the pack screen's lists can be dragged the same way.
Part of #282
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Changes to horizon-core or horizon-world make ~280-300 harnesses rerun, which takes
60-90 min on the Mac mini runners; at 45 min those runs were cancelled and proved nothing
(#269, #280, #292 today).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
JoinMultiplayerScreen's layout: the title centred in the 33-pixel header, the
ServerSelectionList between the separators (list background, 305-wide rows,
white selection outline, scissored rows, the scroller sprites), the footer
buttons centred in 60 pixels. The list always ends with the LAN header,
"Scanning for games on your local network" over LoadingDotsText's dots,
then the LAN games. Saved servers show two MOTD lines, the player count and
the ping, pinging or unreachable icon; Edit and Delete need a saved server.
The list scrolls with the wheel and a draggable scrollbar (ui::Scroll, after
AbstractScrollArea).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The autosave queued its chunks without waiting (#284), then saved the
entities with the waiting call. The region writer is one FIFO thread,
so the main thread still waited for every chunk queued before the
entities: a 1.9 s frame at the 5-minute autosave on a world with ~900
columns. Like PersistentEntitySectionManager.autoSave (saveAll's flush
only on save-all and quit), the autosave now queues the entities too.
Part of #224
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Screen.extractBlurredBackground: every classic screen reached from the title
(world selection and creation, multiplayer, options) blurs the panorama with
vanilla's minecraft:blur post chain, three horizontal and vertical box blur
passes of menuBackgroundBlurriness pixels, then draws menu_background over it
instead of a flat dark fill. The title screen itself stays sharp.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
FireworkEntityRenderer submits the rocket's item in the GROUND context,
turned to the camera (and over on its side when shot at an angle), at the
packed light of its eye. Horizon drew a flat 0.5-block sprite at full
brightness, so rockets glowed at night and had no item shading.
Fixes#279
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Options.cutoutLeaves, ambientOcclusion and biomeBlendRadius call
LevelExtractor.allChanged when they change. Horizon only handed the new
value to the mesher, so after turning See-Through Leaves off the loaded
sections kept their see-through leaves, pale against the sky, beside the
opaque ones of sections rebuilt later.
Fixes#273
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla 26.3's Projection passes its planes reversed and every pipeline
tests GREATER_THAN_OR_EQUAL. Horizon kept forward depth with a 0.05 near
plane, so far geometry lost precision and fire_side's quad, 0.01 pixel in
front of the block it burns on, z-fought with it.
Fixes#272
Co-Authored-By: Claude Opus 5.5 <[email protected]>
BeehiveBlockEntity (crate::beehive): up to three bees stored as entity
tags, out when their time is up (not at night, in the rain or through a
blocked front unless in an emergency), honey for nectar, the flower they
knew; vanilla's bees / flower_pos save format. Bees find a hive with room
(BeeLocateHiveGoal), fly home and go in (BeeEnterHiveGoal through
MobEvent::EnterHive). In the game hives tick, breaking one lets the bees
out angry unless smoked, fire empties it, dispensers release them.
mob_ai: bee_hive_night, _rain, _morning and _honey match vanilla.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
arrows::tick never set the projectile's previous position (Entity.setOldPosAndRot
at the start of every tick); Projectile::tick does, but arrows don't go
through it. The renderer lerped from the spawn point to the current
position every frame, so a flying or stuck arrow jumped along its whole
path. The arrow harness now prints the old position too.
Fixes#278
Co-Authored-By: Claude Opus 5.5 <[email protected]>
RecipeToast keeps the station and result of each display unlocked while it
is up, draws the station at 0.6 scale in the plaque's corner over the result,
takes turns through them over the display time and stays up five more seconds
after each unlock. Toasts slide in and out on ToastInstance's squared curve,
honour notificationDisplayTime and play ui.toast.in / ui.toast.out.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The oracle started a scenario once hasLightWork() had been false for six
ticks, but light updates reach the light thread through the chunk task
dispatcher, where hasLightWork() can't see them. A drowned's first
sun-burn roll (Mob.isSunBurnTick draws only when the brightness at its
eyes is above 0.5) then read the dirt's light under a freshly filled
pool or the water's, depending on thread timing: one random draw more or
less, and drowned_pool / drowned_pool_swimmer failed at random.
Queue ThreadedLevelLightEngine.waitForPendingTasks on every forced chunk
after building the scene and tick only once they have all completed.
Fixes#290
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A port of vanilla 26.3 Bee: FlyingMoveControl and flying navigation,
hover wandering, flower search and pollination (nectar), growing crops,
tempting and breeding, persistent anger with the hurt-by alert, the sting
(poison, stinger lost, later death), drowning, staying home at night and
in rain, roll, textures and NBT. Hives as homes are not simulated yet.
FlyingMoveControl clamps the pitch like Entity.setXRot, so a rotlerp
that wraps to 359 gives 90 as in vanilla.
Mob AI parity: 9 bee scenarios, 3609/3609 ticks; full harness 120/120.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Bats fell into the generic fallback AI and walked like land animals; their
animation states were never started. Port vanilla Bat: hanging from a solid
block until a player comes within 4 blocks or the block goes, the take-off
level event, customServerAiStep's flight towards random nearby air, Bat.tick
(held to the ceiling while resting, vertical speed scaled by 0.6 in flight),
setupAnimationStates (flying and resting keyframes), waking when hurt,
BatFlags in the save, no fall damage, movement events without sounds, the
lower voice pitch, and no omnidirectional air drag (vanilla gives that only
to bees, parrots and sulfur cubes).
Mob AI parity harness: six bat scenarios, 6/6 match (1906/1906 mob-ticks).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Striders fell into the generic fallback AI, whose FloatGoal jumped them
up and down in lava. Port vanilla Strider: its goals (no FloatGoal,
StriderGoToLavaGoal, RandomStrollGoal every 60 ticks, LookAtPlayerGoal for
striders), floatStrider, standing on lava sources through the living
collision context and the half-block liquid shape, travelling in air where
lava holds it up, StriderPathNavigation's lava destinations, the walk target
value, going cold (suffocating speed modifier, cold texture) away from warm
blocks, the happy/retreat sounds and a lava-proof checkFallDamage.
Mobs now keep Entity.mainSupportingBlockPos from their last move, as
getBlockStateOnLegacy reads it.
Mob AI parity harness: five strider scenarios, 5/5 match (1105/1105
mob-ticks; 0/5 and 5/1105 before).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Mob.readAdditionalSaveData assigns persistenceRequired from the tag,
overriding the constructor's setPersistenceRequired (copper golems,
elder guardians). Horizon ORed it in, so a copper golem vanilla saved
with 0 came back with 1. The monument's elder guardian, which Horizon
builds from NBT where vanilla calls the constructor, now carries the
flag in its spawn NBT.
Fixes#289
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Generation asked for columns within the view distance + 3 or two of a
finished chunk, but a column that came back finished (loaded from the
save) was kept only within + 2 or one of a finished chunk. After a
teleport, saved columns in between were loaded, dropped, asked for and
loaded again on every frame, forever: 8 of them at render distance 16
held 8 of the 39 in-flight generation slots and kept the disk and a
worker busy. The requested, kept and unloaded ranges now live in Ranges
and nest; requested_columns_stay_loaded checks it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On macOS the Horizon client renders offscreen (--offscreen) with audio off,
from the logged-in session; Linux keeps vscreen and the null ALSA device.
The vanilla side was already a headless dedicated server. --bin-dir is
resolved, since the client runs from the run folder. parity.py no longer
skips save_roundtrip off Linux; it stays local-only (no rendering over SSH).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A new chunk relights its eight neighbours, and every finished light job
dirties all sections of the 3x3 columns around it. Meshing waited only for
the column itself to be lit, so sections near the loading frontier were
meshed with half-settled light and again after each neighbour's relight:
about 2.25 meshes per section while a world loads. Wait until no column of the
3x3 is being lit or queued for light. The final meshes are the same; they are
built once.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Same rule as perf_bench: with status.streaming, the last sections are drawn
once no stage has held work for half a second, so the remeshing that
follows a load (fluids settling, random ticks) isn't counted as loading.
Without it, meshing still has to go quiet.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Builds that report status.streaming are loaded once chunk counts are
stable and no stage (generation, decoration, light, meshing) has held work
for half a second. Meshing going quiet also waited through the remeshing
that follows a load (fluids settling, random ticks: 1500 falling to 60
meshes/s for ~150 s at RD16), which is simulation, not loading. Older
builds keep the meshing-rate rule.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Sections wait for their neighbours' light before they are meshed, so the
chunk counts settled while edge sections were still undrawn.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Each game runs --repeats times (3) per render distance and writes one v1
report per distance (benchmarks/perf/<label>-rd<N>.json): figures as means
with standard deviation and 95% CI, per-flight frame entries with the fixed
histogram, the hardware without the host name. frames() and summary() match
crates/horizon-bench value for value (tested on the same inputs).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Each session rewrites benchmarks/launcher-<time>-session.json in the data
folder (report format v1): start to first frame, release list fetch,
install/update and prepare durations, Play to the game's first frame (the
game now logs its first frame), RSS, UI frame times, binary size.
horizon-launcher benchmark starts the launcher N times (cold, then warm),
switches pages, measures idle CPU and memory and the release list fetch,
and writes one report with means and 95% CIs. Nothing is uploaded.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
One versioned JSON report for every benchmark Horizon writes (the game's
--benchmark, the launcher's measurements, tools/perf_bench.py): the local log
a player can attach and, with consent, what the telemetry upload sends.
Summaries over repeats (mean, stdev, 95% CI) and fixed-edge frame-time
histograms, never raw frame streams; no host, user or path in uploads.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Horizon saved a chunk only once a player changed it and never saved proto
chunks, so leaving an area dropped decorated chunks while the final chunks
next to them stayed loaded (or were regenerated later). Coming back, those
chunks were generated again and their decoration was skipped beside a final
neighbour: whole chunks of a bastion lost their pieces, other chunks their
features. Like vanilla, generated chunks are now saved, proto chunks that
decoration wrote to are saved with their status (terrain or features) and
restored onto the chunk generated again, columns that finish after the player
left are saved, and saving waits for the decorations in flight.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The mesh loop dropped every dirty section of a loaded column outside the
render distance. Columns load up to three chunks past it, and a column
loaded from a save with its light runs no light job, so nothing asked for
its sections again: once the player came closer it stayed invisible until
a block in it changed. Those sections now wait and are put back when the
centre, radius or forced chunks change.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
`status` gains `streaming` (queued jobs, chunks generating, protos waiting for
decoration, decorating, finishing, lighting, meshing) and `dirty_light`, so a
slow load shows which stage it waits on.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla trims each line and drops only the hidden splash (hash 125780783);
a blank line stays a (blank) splash.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
PalettedContainer::pack hashed all 4096 values with SipHash to build the
palette. It now shares from_values' scan, which only switches to a map
past 16 distinct values. Same palette order, same output.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Every mesh job cloned the whole FaceVisibility (every state, every face
VoxelShape and the occlusion cache) and dropped it afterwards: about half
of the workers' busy time while a world loads. Each worker now keeps its
copy, like vanilla's per-thread occlusion cache, and clones a new one only
when the mesh tables change.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
tools/perf_bench.py runs Horizon, the vanilla 26.3 client and vanilla with
Sodium, Lithium, FerriteCore and ImmediatelyFast through the same steps:
fresh world (seed 1234, creative) per render distance, start to title, start
to in-world, then per scene teleport, chunk load, settle and a fixed 192-block
flight. Every frame time is kept and one function computes average FPS, 1% and
0.1% lows, p50/p95/p99/p99.9 and a frame-time histogram for every game; RSS is
sampled from outside the process for every game. One options.txt (vanilla's
keys, Fancy preset values with a Custom preset so the render distance holds)
is written for all of them.
Vanilla is driven by a dev-only Fabric mod (mods/horizon/src/bench) compiled
against the user's own client jar: it creates the world with
WorldOpenFlows.createFreshLevel, teleports with /tp and places the player on
the path every frame as horizon-ctl fly does. Nothing of the game is
redistributed and the performance mods are fetched from Modrinth with the
hashes pinned in packs/mods.lock.json.
Horizon side: `fly ... PITCH raw` also replies with every frame time, and
--offscreen drives frames from the event loop, so a window nobody sees (a
locked macOS session) still renders.
docs/benchmark.md describes what is measured, how fairness is kept, the
caveats and how to reproduce it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Sharing no longer needs a relay. The panel shows the full invite code in
groups (monospace, wrapped) with Copy and Stop sharing, then who can join:
friends anywhere when the router opened the port or this computer has a
public address; otherwise what to do (UPnP, the port to forward and the
LAN address, a relay, a VPN; CGNAT called out) with a link to the
troubleshooting docs. A relay set in horizon.txt (friendsRelay) or
HORIZON_RELAY_URL adds its short code. horizon.txt keeps other lines when
the menu style is saved.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A read timeout restarts with every byte, so a peer trickling one byte at
a time could hold the host's 8 handshake slots (or relay threads) for
hours. The host's openings and the relay's first line now end after 10 s
in all; the relay closes connections past RELAY_MAX_CONNECTIONS (4096).
The docs say what happens to a router mapping when the game crashes.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The game joins `friend:CODE` with either code: a full invite is dialled
directly, then through the relay it names; a short code goes through the
relay the player set. Join a friend gains a Relay (optional) field, saved
to horizon.txt (friendsRelay) and used for hosting too; the code field
shows the end of long codes and says what is wrong with one (typo, other
version, short code without a relay). The launcher validates both kinds.
tools/test_friends.py now runs (a) the code alone with no relay anywhere
and (b) a relay run locally with the short code; routers and STUN stay
out of tests.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Hub::admit read the entity counter before incrementing it, while the game
increments first (the current value is the last id in use). The first
guest after Open to LAN got the host's own entity id, so its client took
the host's AddEntity for itself and never showed the host.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The Horizon pause menu gains "Invite friends" beside Open to LAN. It opens
the world to LAN if it isn't (Open to LAN's defaults) and shares that port
behind an invite code through the relay; the menu shows the code with Copy
and Stop sharing, who is here and why the code changed. Closing the LAN
server ends the code; stopping sharing closes LAN again if sharing opened
it. `horizon-ctl friends host|stop|status` drives it in harnesses.
Without HORIZON_RELAY_URL (the default until a relay is deployed) the button
is disabled and says the feature isn't available yet.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The game joins `friend:CODE` addresses: the network thread resolves the code
through the relay, agrees on the key, settles on a direct or relayed stream
and connects the client to the tunnel's loopback port; everything after is
an ordinary join. Errors (unknown or ended code, wrong secret, host not
answering, version mismatch, rate limits) end on the Disconnected screen in
plain words. Shown and logged addresses hide the code's secret half.
Horizon's multiplayer page gains "Join a friend" (a code field); the classic
screens keep vanilla's buttons. The launcher's Play page gains a "Join a
friend" row (paste works) that starts the game straight into the join;
`horizon-launcher play --join CODE` does the same.
tools/test_friends.py runs a relay, a host Horizon sharing a world and a
guest Horizon, over plain LAN (--baseline), the direct path and the forced
relay path: chat, movement and blocks both ways, then the ended invite.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Ludovic can't pay for a hosted relay, so nothing depends on one now. The
host's game finds how it can be reached on its own: its interface
addresses (LAN, VPNs such as Tailscale or ZeroTier, IPv6), a port opened
on the router for an hour at a time (PCP, then NAT-PMP, then UPnP IGD;
renewed while sharing, deleted on stop) and its public address from STUN
(Cloudflare's and Google's public servers by default, configurable). The
invite code carries those endpoints and an 80-bit secret in base32 groups
with a checksum; guests dial every endpoint at once and run the same
SPAKE2 + ChaCha20-Poly1305 handshake, no rendezvous involved. When nothing
can reach the host it says why and what to do (UPnP, the port to forward,
a relay, a VPN; CGNAT and double NAT detected).
The relay stays as something anyone can run (binary, Dockerfile, systemd
unit; Railway config removed): with one configured the host also gets a
short code, and the full code names the relay so guests fall back to it.
No relay by default.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A dual-stack listener reports them as ::ffff:a.b.c.d, which the other end
can't dial from an IPv6-only socket.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
horizon-friends: codes (Crockford base32, nameplate + secret), the relay's
JSON-line protocol, SPAKE2 then ChaCha20-Poly1305 frames keyed off the
code's secret, direct connections (candidates, TCP simultaneous open from
a reuse-port socket) with the relay as fallback, and the host and guest
ends that carry the game's TCP stream between loopback ports.
horizon-relay: the service (codes, rendezvous, byte relay, rate limits,
bandwidth caps, idle timeouts, counts only), its Dockerfile and Railway
config. Nothing is deployed and nothing in the game uses it yet.
Design: docs/play-with-friends.md.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
/privacy says what the launcher may send when a player allows it (performance
summaries, crash and bug reports, benchmark uploads), why, on what basis,
where it is kept and for how long, and how to delete it. The footer links it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
crates/horizon-telemetry-server, its own workspace like the launcher's (the
game's workspace never builds a web server):
- POST /v1/batch (zstd, decoded and checked by horizon-telemetry), DELETE
/v1/installs/<id>, GET /v1/aggregate (cached 10 min, CORS for the website),
GET /healthz.
- SQLite: machine descriptions stored once, one row of columns per session
with 80-byte summary blobs, zstd bodies for crash/bug/benchmark, daily totals
rebuilt when a day changes and kept after raw rows expire (180 days). About
190 bytes per session on disk.
- Per-address token bucket (addresses in memory only), per-install daily cap,
groups under 5 install-days hidden from the totals.
- Tests: refusals, rate limit, daily cap, deletion (rows and totals), retention,
storage size, and end to end with the launcher's uploader.
- Dockerfile and a path-filtered CI workflow.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The game never sends anything itself; it keeps items in the outbox for the
launcher, and only for kinds the player allows:
- Crash reports: crash::deliver keeps a scrubbed copy (error, stack, thread,
Horizon version, graphics adapter and render settings; never the log, world,
server or resource packs; the world's name and folder and the server address
are scrubbed wherever they appear). Without consent crash-reports/ stays
local, as before.
- Performance: a Session records frame times while playing and the integrated
server's tick times into fixed-size Timings, with the time to the title
screen and to the world; on leaving a world or quitting, a session of a
minute or more becomes a summary (under 1 KiB) when the player allows it.
- The GPU, its driver and backend come from the wgpu adapter.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Builds on Cosmic's #59 (player fluid travel, swimming, air supply) and
changes only what differs from vanilla 26.3 or what it lacks:
- BubbleColumnBlock.entityInside for the player (isPrecise, onAbove /
onInsideBubbleColumn, not while flying) with its splash/bubble particles.
- LocalPlayer.goDownInWater before LivingEntity.aiStep's velocity cleanup.
- The aiStep jump phase as vanilla: nothing while flying, the fluid jump
threshold from the eye height (also in travelInLava).
- Player.travel: swim steering samples y + 1.0 - 0.1, and LivingEntity.travel
checks fluids before gliding; water and lava no longer end a glide.
- LocalPlayer.isUnderWater is wasUnderwater; sprint conditions get shallow
water, flight and gliding; no flight toggle while swimming; the base tick
runs before local crouching; the desired pose puts swimming before gliding.
- decreaseAirSupply creates the oxygen_bonus instance (getAttribute).
- Swim sounds and muffled steps in water; entity event 67 drown bubbles.
- Scene motion harness: swim_corpus (player travel in water and lava, bubble
columns, air supply) and sprint-swim cases for --sprint-input.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The real-pack benchmark on the production resource_root path stays as the
reproducible measurement. On the Mac mini it saves 1.18 s (Faithful 32x) and
5.29 s (FPBR) per startup with the pack enabled.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A Horizon graphics setting (Performance, off by default, batteryMode in
horizon-graphics.txt): at most 60 frames per second in a world, 30 in menus
without a world or behind other windows, and half the automatic worker
threads. It only changes when frames are drawn; ticks and what frames show
are untouched.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
horizon-telemetry, shared by the launcher, the game and the ingest server:
- consent.json in the game folder: four switches, all off until the player
answers, a consent version that re-asks when what is collected changes, the
answer's time, and a random install ID made at the first yes.
- Payloads: perf summaries (percentiles + 14 frame-time buckets from a
fixed-size Timings recorder), crash and bug reports, and benchmark reports
carried as docs/benchmark-report.md defines them; size caps per kind.
- A scrubber for user folders, user names, e-mail and IP addresses, UUIDs and
caller-known private terms.
- The outbox: only allowed kinds get in, daily caps, 200 items / 2 MiB.
- zstd batches with bounded decoding; the uploader re-reads the consent and
drops withdrawn items before connecting, backs off and caps batches a day.
- tests/consent_gating.rs: a local mock server sees zero connections with the
consent missing, outdated or withdrawn.
No server is set (DEFAULT_ENDPOINT is None): nothing is sent yet.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
pr_ready is the one self-check to run before asking for review: from the
files a branch changes it runs rustfmt, clippy with warnings denied, the
tests of the touched crates (mapped through cargo metadata), the launcher's
own workspace when it is affected, the tokens.css check, the Cargo.lock
rule, the Python tests and the public-tree check, and the vanilla harnesses
the change can affect, picked by tools/parity.py exactly as CI's parity job
picks them. It runs locally or on a fleet machine through ~/fleet/rbuild;
on the Mac mini the harnesses use the parity runner's own workspace and
scripts. It prints one Markdown block for the pull request body and exits
non-zero on any failure.
tools/parity.py gains --changed-files for copies without Git history, and
tools/ci/vanilla-data.sh accepts the preparation key from outside.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Releasing the pointer only ungrabbed it, so the cursor showed up wherever
the system had it: on Wayland, where the locked pointer last stopped (often
where the previous screen was closed), elsewhere wherever the hidden pointer
had drifted. Opening the inventory, a chest, the pause menu, chat, a sign or
a book could put it anywhere.
Game::capture now follows MouseHandler.grabMouse/releaseMouse: both move the
pointer to the window centre, and releasing also sets Horizon's own cursor
(what every screen hit-tests and draws hover with) to that point. On Wayland
the position can only be set as the locked pointer's hint, applied by the
compositor as the lock goes, so it is set before ungrabbing; X11, Windows
and macOS warp, so it is set again once the pointer is free. Releasing an
already released pointer does nothing, so screen-to-screen changes keep the
cursor where it is, as in vanilla, and nothing warps while the window is
unfocused (GLFW's glfwSetCursorPos does the same), so losing focus never
pulls the pointer back into the game.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The Horizon-style Singleplayer page had no way to delete a save. A Delete
(danger) button now shows beside Play when a world is selected and opens a
confirmation page in the design language: the world's name and last-played
date in a notice led by the danger dot, "It will be lost forever. This
can't be undone.", then Delete world / Cancel. After deleting, the list
refreshes and the selection moves to the world that took the deleted row
(or the new last one). A failure shows a Horizon error page whose Back
returns to the question.
Both menu styles share world_management::delete, now a port of
LevelStorageAccess.deleteLevel: it takes the world's session lock first, so
a world open in this or another instance is refused, deletes everything but
the lock (retrying five times, 500 ms apart), then releases and removes the
lock and the folder. Only a single plain folder name directly in saves is
accepted, and links inside the world are removed, never followed.
Leaving a world now flushes the region writer and releases the session
lock (vanilla closes the storage access), so the world just played can be
deleted from the title.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla 26.3 has a MapItem per explorer map (buried_treasure_map,
abandoned_camp_map, woodland_mansion_map...) and treats any stack with a
map_id as a map. Its ExplorationMapFunction finds the nearest structure of
the tag, makes an unlimited tracking map centred on it, draws the biome
preview (MapItem.renderBiomePreviewMap) and adds the target decoration
(MAP_DECORATIONS, drawn by tickCarriedBy). Horizon only recognised
filled_map, turned trade maps into filled_map, cleared every decoration but
the player's each tick, and chest loot only flagged the map as found: the
treasure maps of shipwrecks and underwater ruins and the maps of 26.3's
abandoned camps had no map id, no target and no picture.
- map_id: any stack with the component.
- tick_maps: the item's map_decorations (red X, mansion, camp...) are drawn
as addDecoration places them (off the map only when unlimited).
- render_biome_preview: vanilla's water outline picture from the noise
biomes at sea level (new generator query noise_biomes, fuzzed through
BiomeManager's zoom) and #water_on_map_outlines.
- Chest loot (rolled with the chunk here) keeps the function's search in
minecraft:custom_data; opening the chest runs it from the chest (vanilla
runs it when the chest is unpacked on first open), else reaching the
inventory runs it from the player. Nothing found drops the item, as the
tables' filtered map_id check does.
- Trades keep their map item.
Fixes#147
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Releases now ship as often as changes land, so nothing they leave behind may
grow without bound. Workflow uploads keep 3 days (nightly parity logs 7); the
releases carry their own assets.
tools/ci/runner-gc.sh, run every 6 hours by launchd on the Mac mini from a copy
of main, skips a runner in a job, removes a checkout's cargo build folders past
12 GB and harness work folders under .local/ untouched for 3 days, and keeps the
2 newest shared preparations while no runner works. The runners build through
the machine's sccache, so what it removes comes back from the cache.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Ludovic wants releases as often as changes land. A kind now releases when
player-facing pull requests merged since its last release and the newest of
them merged 15 minutes ago (a burst of merges becomes one release), or the
oldest an hour ago when merges never pause; release-now still ships at once.
There is no minimum time between releases. The scheduled checks run every 10
minutes (game at :02, launcher at :07) so a settled release goes out without
another merge.
Release titles keep proper nouns (Horizon's mark, Microsoft, LAN, ...) when
they lowercase the clauses after the first.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The classic title screen's banner now reads "RUST EDITION". It is rebuilt
at load time from the player's own textures/gui/title/edition.png:
"EDITION" is kept column for column, the "T" is lifted out of it, and R, U
and S are Horizon's own glyphs (rectangles in title_edition.rs) drawn the
way the texture's letters are built, in shades sampled from it. "RUST"
spans "JAVA"'s columns, so the banner stays centred. Only vanilla 26.3's
exact texture (pixel fingerprint) is rebuilt: a resource pack's banner is
left as drawn. No Mojang pixels are added to the repository.
The logo and banner are now blitted as LogoRenderer does: the logo's top
44 of 64 rows and the banner's top 14 of 16 rows at y = 30 + 44 - 7
(the banner was 20 px too low and squeezed 16 rows into 14).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The icon files kept their names while the server lets browsers cache /favicon*
for a week, so visitors still saw the old mark. The pages now link each icon
with its content hash (?v=…), and the build writes /favicon.ico (the 32-pixel
PNG in an ICO container) for browsers that ask for it on their own.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla's ClientPacketListener.handleTakeItemEntity adds an
ItemPickupParticle for every take: the taken entity's render state
(extracted at partial tick 1) moves to the collector's x, (y + eyeY) / 2, z
along ((life + partial) / 3)^2 over three ticks, the target lerped between
ticks (ItemPickupParticleGroup). Horizon removed taken entities at once.
game/pickups.rs keeps those particles (joining the group at the end of the
particle tick, as ParticleEngine.tick does) and the item, orb and arrow
renderers draw them frozen at the moving position. Takes by the player
(items, orbs, arrows), by mobs (allays, villagers, piglins, sulfur cubes,
raid banners), by LAN guests on the host, and ClientboundTakeItemEntity on
a LAN guest or server all add one.
Fixes#142
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla plays the hurt sound on the client from broadcastDamageEvent
(LivingEntity.handleDamageEvent -> Player.getHurtSound, the damage type's
DamageEffects sound, pitch (r - r) * 0.2 + 1). Horizon only played
entity.player.hurt by hand for falls and mob hits, so sweet berry bushes
(poking -> hurt_sweet_berry_bush), fire and lava (burning -> hurt_on_fire),
drowning, freezing, cactus and the rest were silent.
The combat port's broadcast_damage now carries its damage source
(HealthEvent::DamageBroadcast, same trace line as before) and the game plays
hurt_sound(effects) for it; the two hand-written plays are gone.
Fixes#144
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Every offline Horizon player joined as Player, the LAN host's own offline
profile, so the host refused every guest with "You logged in from another
location". Vanilla's client defaults --username to Player + millis % 1000
and the launcher passes the account's name. Horizon now generates
Player + five digits once and keeps it in horizon-player.txt next to
options.txt (same name, same offline UUID, same inventory on a friend's
world). --name, HORIZON_PLAYER_NAME and a signed-in account still win.
tools/test_lan_lifecycle.py: two default guests join one host together,
keep their names across rejoins, and a guest named Player is refused.
Fixes#195
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Every page and the in-world screens veil the backdrop at 0.90; the home page at 0.80, with its
secondary text in ink_soft (Ui::muted). Muted text over the backdrop is BACKDROP_MUTED. The home
page's faint wordmark is now whole: as wide as the content column, above the footnote, instead
of cut off at the bottom right.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
HORIZON_MSA_MICROSOFT_ENDPOINT (debug builds only) points the device-code and token requests at
a local mock, so the restyled code panel can be seen and screenshotted without Microsoft.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
macOS: packaging/macos/package.sh (run by launcher.yml) builds Horizon.app (Info.plist, bundle
id dev.horizon.launcher, macOS 11+, the .icns), signs it ad hoc, and makes the two release
assets: horizon-launcher-macos-arm64.dmg, the installer (the app and an Applications link over
a background, laid out by dmgbuild without a GUI), and the .zip the launcher updates from. A
launcher inside the app replaces the whole bundle: copied beside it with ditto, swapped in by
renames, restarted; a bare binary, as older launchers are, still finds its binary in the zip.
A macOS test zips, unpacks, swaps and re-verifies a signed bundle. Signing for real is a
MACOS_SIGN_IDENTITY away (packaging/macos/README.md, with what Gatekeeper asks meanwhile).
Linux: release builds keep horizon-launcher.desktop (with the binary's current path),
horizon.desktop (hidden, for the game window's icon) and the hicolor icons under
$XDG_DATA_HOME, without root; `horizon-launcher desktop-entry --remove` takes them out and
stops it, `desktop-entry` puts them back.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The README banner is redrawn in the new design (docs/media/banner.html), and the launcher and
menu screenshots (README and website) are the restyled ones. The website's launcher section
gains a Launcher tab: its screenshots no longer show the old background, whose licence was
never confirmed.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The mark is an H whose crossbar is the horizon: a violet line edge to edge, the stems above it
and their reflection below at a third, on whole units of a 16-unit grid so it stays sharp at
16 px. docs/media/brand holds the SVG masters (square icon, macOS tile, avatars, the mark for
light and dark backgrounds), the two alternatives that were shown with it, the GitHub social
preview and the Discord banner, and a README of what goes where and how to upload what only
Ludovic's accounts can.
`cargo run -p xtask -- icons` renders the masters (rsvg-convert) and packs, in plain Rust:
packaging/icons/horizon.ico and horizon.icns, Linux's hicolor icons 16–512 and the scalable
SVG, the game's run-time window icon, the avatars and the website's favicons.
packaging/linux has .desktop entries for the launcher and the game.
On Windows, build.rs embeds the .ico as icon resource 1 in horizon-client (and its tools) and
horizon-launcher (winresource, a Windows-only build dependency); gpui loads that resource as
the launcher's window icon. The game window sets the icon itself (X11, Windows) and names
itself `horizon` on Wayland so the desktop entry matches.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The page now reads the design tokens generated from crates/horizon-theme (src/tokens.css, put
before style.css by build.mjs), and follows the system's light or dark appearance, with a
toggle that remembers the choice. The full-screen sunset hero is replaced by type: a headline,
one paragraph, the download, and an index list of what the page holds. Every section is a mono
label, a display heading and its content as numbered rows with one-pixel rules: the facts, the
fourteen subsystems, the launcher's features, the community links. Rows fade a soft plate in
under the pointer; content arrives with a short fade, rise and unblur. The footer ends on the
giant faint wordmark. The 404 page and the social preview (og.html) follow.
Instrument Serif, the hero images, the parallax, the statement fill and the magnetic buttons
are gone; the icons come from Horizon's new mark (xtask icons). Touch targets are 44 px high.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Only Horizon's own style changes (what HORIZON_MENU_STYLE=horizon and "Menu style: Horizon"
show); vanilla's classic screens and the HUD are untouched.
The shared components in ui_theme follow the launcher: pill buttons, the sidebar as text items
with a soft sliding pill under the mark and a name instead of a tinted panel, the segmented
control's ink pill, a 36x20 switch, a thin ink slider, fields without borders, display-size page
titles, mono capitals for status lines and section labels. Settings rows, worlds and servers are
numbered index lists (01, 02…) with one-pixel rules; list rows fade a soft plate in under the
pointer and keep it under the selection. Plates are translucent ink so they read the same over
any backdrop. The veil over the live backdrop is flat (no gradients), lighter on the home page,
which ends on the faint wordmark. The opening's wordmark is "horizon" in Geist.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A top bar (the mark, Horizon, the pages as text) replaces the sidebar; a mono footer carries
the launcher's version, its update and the community links. Every page is a display title, a
short muted line, then an index list: Play's resources and folders, the versions with their
names large and their size and state in mono, the patch notes, the settings. Buttons are pills,
the main action is the one violet thing on the screen, selection is an ink pill, the console,
notices and the crash report sit on soft plates without borders. The Play page ends with the
faint wordmark, half set below the window's edge.
The sunset photograph and its veils are gone (and the `image` dependency with them): the
window is the page colour, light or dark with the system. The opening draws the wordmark
letter by letter, then the launcher rises into place. HORIZON_LAUNCHER_APPEARANCE=light|dark
overrides the system's appearance, for screenshots and tests.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The veil goes to 0.90 (0.80 on the title's home page, which keeps to ink and ink_soft), and
muted text over the backdrop is lifted to #9fa3ae; a test checks every pair over a backdrop of
luminance 0.77 (the snowiest title panorama measured reaches 0.49), bare and on a row's plate.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
docs/design-system.md describes the new design language (after mkdir.directory: type as the
interface, numbered index lists with hairline rules, one violet accent, space instead of
borders, quiet motion, a faint wordmark) and its tokens, components, surfaces and mark.
crates/horizon-theme now holds the whole system: the two palettes (cool neutrals, light and
dark, each text colour checked against WCAG AA on every background by a test), the type scale
with web sizes, spacing, radii, rules, control heights and motion with CSS easing curves.
Instrument Serif is gone; display type is Geist at medium weight.
`cargo run -p xtask -- tokens` writes website/src/tokens.css from the crate, and a test fails
when the file is out of date. The launcher and the game's menus read the new names through
small adapters; the restyles follow in their own changes.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A saved spawner's block entity id is minecraft:mob_spawner, not its block's
name, so decode_block_entity found no block entity for it and dropped it:
every spawner loaded from a save (a vanilla world's dungeons and mineshafts,
or Horizon's own after a reload) lost its spawn data, and the server sent
its chunk with no update tag. The id now maps to the spawner block.
The chunk packet lists block entities in LevelChunk.blockEntities' order,
which Horizon reproduces from the position hashes; positions sharing a
bucket keep their insertion order, which for a loaded chunk is the save's
block_entities list (SerializableChunkData -> pendingBlockEntities). That
list order is now kept with the chunk and used as the insertion order.
dedicated_server: 542 transcript lines identical (was 10 differing).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Horizon averaged each mip level's alpha plainly, so a cutout texture's
transparent edges grew past the alpha test with every level: the cactus
side's transparent border columns opened see-through slits along its
corners from about 40 blocks, more than vanilla shows. The sprite array's
mips now come from a port of MipmapGenerator.generateMipLevels: the
texture's .mcmeta mipmap_strategy (auto: cutout when a pixel is fully
transparent) and alpha_cutoff_bias, solidify or the dark fill on the base,
meanLinear or darkenedAlphaBlend, and scaleAlphaToCoverage keeping every
cutout level's alpha-test coverage at the base level's.
tools/test_mipmap_parity.py runs vanilla's MipmapGenerator and the port on
every static 16x16 block texture and a sample of item textures.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
EnderDragonFight's boss event sets createWorldFog, and while such a bar
shows AtmosphericFogEnvironment.setupFog pulls the fog in (start at most
10, end at most 96, sky and cloud fog ending there): Horizon never did, so
the End stayed clear during the fight. FogInputs gains boss_world_fog,
set while the dragon's bar shows.
FogRenderer's void darkness starts voidDarknessOnsetRange (32) above
level.getMinY(), which is 0 in the Nether and the End; Horizon used the
Overworld's -64 everywhere, so the End's fog never darkened below y 32.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The crystals were a hand-built mesh without normals (so unshaded), without
the bedrock base and without beams. They now draw the baked
minecraft:end_crystal#main layer under EndCrystalRenderer's transform
(scale 2, down 0.5, not flipped) with EndCrystalModel.setupAnim's poses
(end_crystal::poses: the base shown per showsBottom, the glass raised by
half the bob, outer glass, inner glass and cube turned with rotateBy), the
entity light at their block, and the beam to their target
(submitCrystalBeams from the target's centre up to the crystal) while the
spikes are being summoned back. Crystals placed outside the End draw too.
The entity model harness no longer skips end_crystal: the oracle runs
EndCrystalModel.setupAnim on EndCrystalRenderState and the fixture poses
the layer with end_crystal::poses.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Hud.extractCrosshair draws every crosshair sprite with
RenderPipelines.CROSSHAIR, whose BlendFunction.INVERT (ONE_MINUS_DST_COLOR,
ONE_MINUS_SRC_COLOR, ONE, ZERO) inverts the colour under it; Horizon drew it
alpha-blended white. GuiDraw sprites now carry their pipeline, the HUD batch
records the vertex ranges that invert and the HUD renderer draws them with a
second pipeline.
The attack indicator (crosshair or hotbar, per the Attack Indicator option)
now shows Player.getAttackStrengthScale(0) from the attack strength ticker
instead of a constant full strength, and the full indicator shows at full
strength on a living entity under the crosshair with a weapon slower than
MIN_CROSSHAIR_ATTACK_SPEED.
The HUD harness records every sprite's pipeline on both sides and gives the
crosshair a stand-in crosshairPickEntity (alive or not) and attack delay.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The host sends Creeper.DATA_SWELL_DIR and DATA_IS_IGNITED with the
creeper's entity data, and a guest runs the client half of Creeper.tick
on them, so the swell and white flash CreeperRenderer draws now show on
the guest too. Creeper.readAdditionalSaveData's ignited lights the fuse
(summon creeper ~ ~ ~ {ignited:1b}).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Players could never pick up arrows: arrows had no pickup rule and no
item, nothing checked a player touching them, and arrows from saves were
loaded into a separate list that never ticked.
Arrows now carry AbstractArrow's pickup rule and pickupItemStack. The
player's bow and crossbow give them the drawn ammo (ProjectileWeaponItem
.useAmmo: creative, Infinity and multishot's extra arrows draw intangible
ammo, so those arrows are CREATIVE_ONLY), setOwner(player) turns
DISALLOWED into ALLOWED, dispensers make them ALLOWED, mobs' stay
DISALLOWED. Each tick the player touches the arrows in its box grown by
(1, 0.5, 1) (Player.aiStep): a stuck arrow that has stopped shaking goes
into the inventory (ALLOWED), or just goes for a creative player
(CREATIVE_ONLY, or ALLOWED with a full inventory), with the item pickup
sound at the arrow.
Saved arrows load as live arrows (life, shake, inGround, inBlockState,
pickup, item, fire) and live arrows are saved, so stuck arrows survive a
reload and despawn on time; the legacy never-ticked arrow list is gone,
and /kill, /tp and /data reach the live arrows.
The arrow harness adds who shot the arrow, intangible ammo and
dispensers' ALLOWED, then touches in flight, while shaking and when stuck
by survival and creative players with empty, partly and fully filled
inventories, against the real setOwner and playerTouch.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Arrows never caught fire: nothing ignited them (lava, fire blocks, a
Flame bow's projectile_spawned ignite), so they never set mobs, TNT,
campfires or candles alight. Flame only worked for mobs' bows, as a
5-second ignite on hit.
Arrows now keep Entity.remainingFireTicks: the blocks along their path go
through InsideBlockEffectApplier.StepBasedCollector (fire and soul fire
fireIgnite to 8 s, lava LAVA_IGNITE to 15 s, water EXTINGUISH with the
extinguish sound), the player's bow and mobs' bonus ignite them for 100 s
at spawn, and the fire burns down in flight but not while stuck. A
burning arrow sets its target alight for 5 s, primes TNT, lights
campfires, candles and candle cakes (onProjectileHit, mob_griefing for
mobs' arrows), and blows up a TNT minecart; arrows now hit minecarts like
other entities (VehicleEntity damage). Burning arrows show
FlameFeatureRenderer's fire quads.
The arrow harness adds lava pools, lava/water walls, fire and soul fire,
Flame arrows, and burning/non-burning hits on TNT, campfires, candles and
candle cakes against the real blocks' onProjectileHit.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Arrows flew through water with a point check for the 0.6 drag and nothing
else: no bubbles, no splash, no currents, no bubble columns, and the
stuck arrow logic (nudging back by the normalized motion, no falling
out, a 1200-tick age) was not AbstractArrow's.
Port AbstractArrow.tick for arrows (arrows.rs, state in
Projectile::arrow): the in-block check, the water inertia from the last
base tick's wasTouchingWater with four bubbles behind it, crit trails,
lerped rotation, the COLLIDER clip with entity hits first, the effects
of the blocks along the path (bubble columns push, water extinguishes),
air drag and gravity, then the base tick's updateFluidInteraction
(currents while not stuck, splash sound/particles), sticking 0.05 back
along each moving axis with the arrow's own random for the hit sound,
falling out when the block goes, and despawning 1200 ticks after
sticking.
New harness tools/test_arrow_parity.py drives real vanilla Arrow entities
(VanillaArrowOracle) and Horizon's arrows::tick (examples/arrow_probe.rs)
over the resident scene.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The player's movement treated every fluid as empty: is_in_water was
always false, LivingEntity.checkFallDamage's updateFluidInteraction was
a no-op and Entity.move's fall-reset ray ignored water. A player landing
in water (or a waterlogged block, a bubble column, kelp) still took the
whole fall.
Bind vanilla's Entity.updateFluidInteraction to the player over the
existing horizon-world kernels (EntityFluidInteraction heights, eyes and
currents, entity_fluid_tick, water_splash, fluid_flow): run it in the
base tick and from checkFallDamage, read wasTouchingWater for
isInWater, query the water's (first-query cached) shape in the
FALLDAMAGE_RESETTING ray, halve the fall in lava after the first tick,
and play the splash sound with bubble and splash particles.
The scene motion harness gains water/lava fixtures (fluid_tick plus
scripted falls through the real Entity.move) against the real vanilla
updateFluidInteraction with the #water/#lava tags bound.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Classifies the online-mode login outcomes from #161: invalid session,
banned from playing online, multiplayer disabled, no account signed in
and the server failing to verify the username are never retried;
unreachable session servers (the client's or the server's) are. A retry
of an online-mode server joins with the account held in memory for the
whole run.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On the Disconnected screen after a lost multiplayer connection, a
countdown "Reconnecting in N s" with a Cancel button, then the same
address again. Attempts back off (5, 10, 20, 40, 60 s) and stop after
five; the count resets only after a minute connected. Only failed
connections and the server's shutdown, restart, time-out or full
reasons are retried, never kicks, bans, whitelists, version refusals or
duplicate logins: the network thread now says whether the server gave
the reason (login refusals included).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
nightly.yml runs every harness from scratch on main at 03:00 Paris time on the Mac mini's runner, after
preparing the resources again if the preparer changed. tools/ci/parity_issues.py compares the results with
the previous audit: a harness that passed and now fails gets one issue labelled parity-regression (an open
one is commented on), and an issue closes when its harness passes again. The bugs session picks them up.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
launcher.yml releases the launcher by itself: after a push to main that changes launcher/,
crates/horizon-prepare or crates/horizon-theme, hourly, and by hand, with the game's rules
(tools/release_train.py decide --kind launcher: player-facing pull requests and a 3-hour cooldown,
release-now, skip-release, RELEASES_PAUSED). launcher-vX.Y.Z is the next patch, or the next minor after a
pull request labelled feature. Linux on ubuntu-22.04, Windows on GitHub, macOS on the Mac mini's runner;
published with --latest=false.
No version-bump commits: VERSION reads HORIZON_LAUNCHER_VERSION with option_env!, falling back to
Cargo.toml's, and --version, the User-Agent and the self-update comparison use it; each build checks
--version says the release's version.
With several game releases a day, the newest launcher release soon falls out of the newest 100 (launcher)
or 30 (website) releases: both now look the newest launcher-v tags up one by one when the list has none.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla's resource packs refuse symbolic links inside a pack: with linked sounds the vanilla sound oracles found
no sound files while Rust did (sound_event, sound_engine on #133's parity run).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Checked on the Mac mini: 1500 offscreen frames in 12 s, 81 chunks generated, 4153 sections meshed, a player.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The first audit on the Mac mini failed 7 audio harnesses (no sounds in the prepared resources: the Mac mini has
no official launcher for horizon-prepare sounds to copy from) and save_roundtrip (it needs vscreen). The
sounds depend only on the Minecraft version: kept once in shared-local/sounds/<version>/ and linked into each
preparation.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A second runner (ludovics-mac-mini-2, same labels) shares the first one's vanilla data through a link, so
vanilla-data.sh prepares under a lock; a 30-minute audit no longer holds a release's macOS build back.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
release.yml now releases the game by itself: after every green checks run of a push to main, every hour
(a release the cooldown held goes out without another merge) and by hand. tools/release_train.py decides
from the pull requests merged since the last published game release: a release when one changed what
players run and the last release is 3 hours old, at once for a pull request labelled release-now;
skip-release pull requests don't count. RELEASES_PAUSED=1 stops it. Tags vYYYY.MM.DD, then .1, .2 the
same UTC day.
Linux (ubuntu-22.04) and Windows build on GitHub with Cargo caches, macOS on the Mac mini's runner
(MAC_RUNNER=macos-latest sends it to GitHub). Each builds horizon-client, horizon-ctl and horizon-prepare
(all three named: --bin selects only the named binaries), checks all three exist, checks their libraries,
starts each one, packages with xtask and archives; the Mac also runs the game offscreen for 300 frames on
a new world with resources prepared from its own Minecraft. The notes: New (label feature), Fixed (the
rest), Known issues (docs/known-issues.md). A hand-started run off main, or with dry_run, builds
everything and writes the notes but publishes nothing.
docs/release-flow.md describes the flow, the labels, pausing and rolling back; the launcher gets a test
that a release turned back into a draft is never offered.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The first parity run kept only .local/horizon-client per preparer version; the exporters' outputs beside it
stayed in the checkout, and the dedicated server, which finds registries.json at
<resources>/../../datagen, didn't find it next to the prepared folder (world_spawn failed).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
verify keeps its dependencies and workspace build in a cache saved by main's runs (Swatinem/rust-cache).
The new parity job runs on the Mac mini's self-hosted runner (label horizon-game), which holds the
maintainer's own Minecraft files: tools/ci/vanilla-data.sh links .local/ to the runner's copy and prepares
the resources with the pull request's own horizon-prepare, once per preparer version; then
tools/parity.py --changed <base> runs only the harnesses whose traced dependencies changed since the merge
base and that have no pass for their current inputs. A fork's pull request runs it only when a maintainer
adds run-parity, once per label, on the commit it had then.
parity.py: --changed REF, and linux_only in tools/parity_hosts.json for harnesses that differ on other
platforms for a reason that is not a Horizon bug (skipped there with the reason, unless named).
docs/release-flow.md starts with the pull request checks and merging.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The player's movement treated every fluid as empty: is_in_water was
always false, LivingEntity.checkFallDamage's updateFluidInteraction was
a no-op and Entity.move's fall-reset ray ignored water. A player landing
in water (or a waterlogged block, a bubble column, kelp) still took the
whole fall.
Bind vanilla's Entity.updateFluidInteraction to the player over the
existing horizon-world kernels (EntityFluidInteraction heights, eyes and
currents, entity_fluid_tick, water_splash, fluid_flow): run it in the
base tick and from checkFallDamage, read wasTouchingWater for
isInWater, query the water's (first-query cached) shape in the
FALLDAMAGE_RESETTING ray, halve the fall in lava after the first tick,
and play the splash sound with bubble and splash particles.
The scene motion harness gains water/lava fixtures (fluid_tick plus
scripted falls through the real Entity.move) against the real vanilla
updateFluidInteraction with the #water/#lava tags bound.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A mace hit after falling more than 1.5 blocks is now a smash attack
(MaceItem): the fall-based damage bonus plus Density, mace_smash damage,
the stopped fall with its damage waived, the ground/air smash sounds, the
dust pillars (level event 2013) and the knockback of the creatures around
the target, then the fall reset and landing particles. Breach thins the
target's armor (CombatRules.getDamageAfterAbsorb), Wind Burst bursts at
the player's feet (ExplodeEffect, trigger), and weapons wear by their
weapon component and award the used statistic.
Vanilla details matched on the way: the hurt knockback is 0.4F and reads
the live knockback-resistance attribute, explosion exposure uses the
exact COLLIDER clip, a burst posts the explode game event and ends the
impulse grace unless it is a wind charge's, visible armor stands feel
bursts that affect block-like entities, attack sounds play at the player.
tools/test_mace_parity.py compares the hit with VanillaDispenserOracle's
real MaceItem / hurtServer / post-attack effects: 3000/3000.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A right click on a bed in the Nether or the End tried to sleep in it (and
set the respawn point) as in the Overworld.
AbstractBedBlock.useWithoutItem reads the dimension's bed rule, which in
the Nether and the End is destroy_on_use: BedBlock.destroyOnUse removes the
head, then the foot behind it if it is still the bed, and explodes at the
head (power 5, fire, BLOCK interaction, the bad_respawn_point damage source:
"Intentional Game Design"); StrawBedBlock.destroyBed plays
block.straw_bed.break_leave and removes it.
tools/test_use_item_on_parity.py clicks beds and straw beds (head or foot,
whole or missing their other half, among blocks) in the Nether and the End,
through vanilla's real explosion; the oracle gains the End.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A right click with glowstone on a respawn anchor did nothing and a charged
anchor never became the respawn point: the block reported the use and the
game dropped it.
RespawnAnchorBlock is now played out (game/respawn_anchor.rs): useItemOn
charges with glowstone (the block change, block.respawn_anchor.charge, the
used item and the charge_respawn_anchor criterion), passing to the off
hand's glowstone; useWithoutItem sets the respawn point in the Nether
(block.minecraft.set_spawn once, respawn_anchor.set_spawn) and elsewhere
explodes: the anchor goes, then a power 5 fiery BLOCK explosion with the
bad_respawn_point damage source whose centre resists like water when water
would flow into it. Dying with an anchor as the respawn point travels to
its dimension and respawns at findStandUpPosition (DismountHelper's safe and
unsafe passes, facing the anchor with calculateLookAtYaw), using a charge
and playing respawn_anchor.deplete; a missing or empty anchor falls back to
the world spawn with block.minecraft.spawn.not_valid. The point is saved in
the player's respawn compound with its dimension; a bed or /spawnpoint
replaces it. Dying in the Nether without one now respawns in the Overworld.
Explosions gained what these need: a damage type, a block resistance
override, BaseFireBlock.getState for their fire, the explosion radius in the
block loot context (survives_explosion and explosion_decay drawing from the
table's random sequence) and TntBlock.wasExploded without a game event.
tools/test_use_item_on_parity.py now charges, uses and blows up anchors in
both dimensions through vanilla's real ServerExplosion on the stand-in
level, and respawns at anchors among blocks through
ServerPlayer.findRespawnAndUseSpawnBlock.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Device-code OAuth, Xbox Live, XSTS, Minecraft login, ownership and profile;
secrets in the OS keychain (oo7 on Linux, keyring on macOS/Windows), accounts.json
without tokens; Settings account rows, 'Playing as' on the Play page; CLI
'account' and MCP account tools; the game gets a 0600 session file it deletes.
Client ID is configuration (HORIZON_MSA_CLIENT_ID), empty = not configured.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Players' skins on online servers come from https://textures.minecraft.net;
the old plain-HTTP fetch left everyone with their default skin.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Read the signed-in account from a session file (--session-file, deleted
on read), answer ClientboundHelloPacket like vanilla: shared secret, server
hash, session join through the services discovery document (authlib's
error mapping and messages), RSA-wrapped key packet, AES/CFB8 both ways.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A corner map of the loaded chunks in vanilla's map colours with the map
item's height and water shading: square or round, three sizes, four
zooms, north up or turning with the player, with the player arrow. In
the Nether, underground and under a roof it maps the floor at the
player's height and cuts walls. Each loaded chunk keeps a 16x16 tile,
filled nearest first within 0.3 ms a frame and rescanned after block
changes; the image is recomposed only when the view or a tile changed
and uploaded into its own atlas slot. Nothing runs while it is off.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The zoom key went to the zoom on every screen, so with zoom on, Z could
not be typed in screens with text fields that share the game's key
handler. Now it zooms only while playing with no sign or book open, lets
go when a screen opens, and always passes the key on, like a vanilla key
mapping. A vanilla mapping bound to the same key still acts; the Extras
page marks the zoom key "Also bound" and names it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Agents press keys and buttons through the control socket, which skipped
the keyboard and mouse hooks; under control the window may also lack
focus without pausing, so the zoom no longer lets go then.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
An Extras section in Horizon's Options holds optional additions saved in
horizon-extras.txt beside options.txt. Zoom divides the field-of-view
target vanilla's Camera.tickFov eases toward (the spyglass path), with a
wheel-adjustable level and lower mouse sensitivity; scoping keeps the
spyglass view. The FPS counter (with minimum and 1% low) and the keystroke
overlay (WASD, space, mouse buttons with clicks per second) draw in the
HUD layer, hidden with F1. docs/extras.md describes all the extras.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The health baseline (HealthWatch) is forgotten when there is no player,
on the title screen, and whenever a new player entity takes over (spawn
and respawn, joining a server, End and Nether dimension changes), so
arriving with less health than the last player had isn't damage.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Short, light pulses through gilrs force feedback, on by default
(Vibration in horizon-controller.txt): damage taken (stronger with the
damage, from any source, here or on a server), a block broken, an
explosion nearby (fading over four times its radius), a totem of undying.
Only the pad that is playing vibrates; macOS has no force feedback in
gilrs and stays still. pad status reports the last vibration asked for.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Spawners were never ticked: Horizon kept their block entity as loaded NBT,
and a freshly generated dungeon or mineshaft spawner lost even that, so no
mob ever came out, nothing turned inside and a spawn egg used on one spawned
a mob next to it instead of setting the spawner's.
The spawner is now a modelled block entity (mobs::spawner): BaseSpawner's
load and save with SpawnData's codec (normalised ids, custom spawn rules,
weighted potentials, the next data shared with the one potential when the
potentials are missing), serverTick (the player range, spawner_blocks_work,
the delay, getOrCreateNextSpawnData, Pos, the spawn box against blocks and
solid shulkers, custom light rules and peaceful, SpawnPlacements, the nearby
cap, the yaw, checkSpawnRules and checkSpawnObstruction, finalizeSpawn for
an id alone, the 2004 burst, entity_place and the spawn poof) and delay with
its block event, in the level random's order. The client copy keeps its own
delay for the spin and puffs smoke and flames; SpawnerRenderer draws the
next mob small and turning. SpawnEggItem.useOn on a spawner sets its mob
(setEntityId), tells the player when spawner_blocks_work is off and fails in
peaceful for monsters. Breaking one pops 15 + nextInt(15) + nextInt(15)
experience whatever the tool. LAN guests get the update tag, the block event
and the entity event.
tools/test_spawner_parity.py ticks vanilla's SpawnerBlockEntity on the
natural spawn oracle's level against mobs::spawner on the same worlds (6,000
cases); tools/test_use_item_on_parity.py now covers spawn eggs on spawners.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Since #52 the shared vertex carried an RGBA8_SNORM face normal for
entity.wgsl's directional light, so every section vertex grew from 32 to
36 bytes although vs_terrain never reads it. Vanilla's terrain uses
DefaultVertexFormat.BLOCK (position, colour, UV0, UV2: 28 bytes): the face
shade is baked into the colour by ModelBlockRenderer and terrain.vsh only
multiplies it by the lightmap.
Sections now mesh into TerrainVertex: position, UV, the sprite layer as a
u16 (bit 15 marks opaque leaves), the two light bytes as a u16 and the
RGBA8 colour, 28 bytes. Every value the shader reads is unchanged; entity
vertices keep their normal.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The status reply's renderer block gains the section vertex arena's buffer
bytes, the bytes resident sections use in it, and the backend allocator's
allocated/reserved device memory (Vulkan), to measure terrain memory.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
options.txt, horizon.txt, horizon-graphics.txt and the rest showed up as
untracked after running the game from the checkout.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Ui::interact records every enabled control (sidebar items, switches,
sliders, segments) as a D-pad focus target, apart from the labelled
button list horizon-ctl clicks by name.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The left stick moves a cursor drawn over the screen, the D-pad steps
between slots and buttons (focus_step), and A / X / the right stick send
the window the pointer events a mouse would, so every screen keeps its
own mouse behaviour (drags, sliders, scrollbars). Y quick-moves in
inventories, LB/RB switch creative tabs, B and Start go back. Focus lands
on the first button, or the held hotbar slot, when a menu opens while the
pad is in use; moving the real mouse hands the pointer back.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
With --bin flags, cargo builds only the named binaries of every selected
package, so -p horizon-prepare built nothing and the import checks failed on
target/release/horizon-prepare (every run since #92).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla samples keys once per tick, so a stick click shorter than 50 ms
(sprint) could come and go between two ticks and be lost: its release
waits for the next tick.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
`cargo run -p xtask -- package BINARY OUTDIR` does what package.sh did:
OUTDIR wiped and recreated, the `git archive HEAD tools
mods/horizon/src/test/java crates/horizon-prepare research-lock.json LICENSE`
tree extracted (the tar read in Rust, executables kept executable), the client,
horizon-ctl and horizon-prepare copied (.exe beside an .exe, the same error when
one is missing) and PLAY.txt. release.yml calls it; package.sh is gone.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla 26.3 has no gamepad input, so this is Horizon's own, on Bedrock's
default layout (docs/controllers.md). gilrs reads the pads; the pure part
(bindings, radial deadzones, the look curve, button edges, the
horizon-controller.txt settings file) is in gamepad.rs with unit tests.
Buttons act as the vanilla key mapping they're bound to, through the same
path a rebound keyboard key takes (the keyboard's post-rebinding handling
moves into Game::mapped_key). The left stick is an analog move vector,
the right stick turns the view per frame, LB/RB step the hotbar, B and
Start leave screens and A respawns. horizon-ctl pad injects a virtual pad.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
tools/fetch_sounds.py (and the lab.py functions it used) becomes
`horizon-prepare sounds` (horizon_prepare::sounds): the pinned asset index,
sounds.json and every .ogg from the player's own launcher install, each SHA-1
checked, into .local/horizon-client/<version>. The launcher runs it as part of
Prepare (in-process, the version's own binary when its sources differ and it
knows the subcommand, in-process for older versions), and a launcher install
without the assets only logs that the game will be silent. Player docs no
longer mention Python for sound.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The hurt red and the white row's alphas come from overlay.rs, which
reproduces OverlayTexture's texels (checked against values printed by
the 26.3 JVM) and is prepended to entity.wgsl; a test parses and
validates the composed shader with naga.
Swelling creepers flash white as CreeperRenderer.getWhiteOverlayProgress
says (the white row's column, red winning while hurt; layers keep the
red only), and CreeperRenderer.scale's wobble uses the unclamped
swelling as vanilla does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla 26.3 has no gamepad code; Horizon's own design follows Bedrock's
default controller layout. Mapping, menus, glyphs, options file, rumble,
Steam Deck and the PR plan.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The player's portal timer reset to zero on any tick outside a portal and
came back from a trip with a 300-tick cooldown that only ran down outside
portals; the overworld exit search covered 32 blocks of loaded chunks;
nothing but the player could use a portal.
The player now follows Entity.setAsInsidePortal / handlePortal with
PortalProcessor (decay by 4, the cooldown of 10 that a portal keeps
resetting, players_nether_portal_default/creative_delay), and
NetherPortalBlock.getPortalDestination through the new
horizon_world::level::portal_travel: PortalForcer.findClosestPortalPosition
(16 / 128 blocks, loaded and saved chunks), createPortal,
BlockUtil.getLargestRectangleAround, PortalShape.getRelativePosition and
findCollisionFreePosition, with the turned yaw and motion and
block.portal.travel. LocalPlayer.handlePortalTransitionEffect closes
screens and plays block.portal.trigger. Items and mobs go through after one
tick with a 300-tick cooldown and wait as saved entities until their
dimension loads around the exit.
tools/test_nether_portal_parity.py compares the destination with vanilla's
getPortalDestination over two headless levels; the scene motion harness
gains portal fixtures for the player's processor and cooldown.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The entity model parity harness now draws each posed model on both sides
under LivingEntityRenderer's root transform (body yaw, scale(-1, -1, 1),
the renderer's own scale) and compares every vertex's normal as the
signed bytes written to the vertex buffer: vanilla's ModelPart.render /
Cube.compile / PoseStack.Pose.transformNormal into a capturing
VertexConsumer, packed with BufferBuilder's own normalIntValue, against
Horizon's mobs::render::draw RGBA8_SNORM normals. A part-scale override
covers negative and non-uniform part scales (trusted sign flips and
untrusted 1/scale normalisation). A one-step difference is accepted only
where vanilla's float sits at a truncation step; a posed position
checksum per drawn part is compared too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
entity.fsh mixes the OverlayTexture texel into the textured, cardinal-lit
colour and only then multiplies by the lightmap. Horizon applied the lightmap
first and mixed a 0.3 red into the vertex colour, so the red was multiplied
by the texture and never darkened: hurt mobs glowed red at night and in caves.
The vertex stage now passes the lightmap colour and the overlay texel
(0xB2FF0000 hurt red, white alpha 63 for OverlayTexture.u(1.0)) to the
fragment stage, which mixes the overlay, then the lightmap, then glint and fog.
Layers vanilla submits with OverlayTexture.NO_OVERLAY no longer take the hurt
tint: armour (EquipmentLayerRenderer, glint stays), held items
(ItemInHandLayer) and the enderman's block (CarriedBlockLayer). The TNT
minecart flash uses the white overlay on the cart's light
(TntMinecartRenderer.submitWhiteSolidBlock) instead of plain full-bright white.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Horizon searched its own spiral of 48-block steps from 0 0 for a dry,
non-leaf top block, so seed 1891623539983467248 spawned at -48 -48 where
vanilla puts the world spawn at 144 71 144 (the player then lands within
the respawn radius, 152 145 in the report).
Vanilla (MinecraftServer.setInitialSpawn): NoiseBasedChunkGenerator.
getOrigin runs NoiseSpawnFinder over the noise settings' spawn_target
points (SpawnTargetPoint.sampleFitness, a 2048/512 then 512/32 radial
search biased towards 0 0); an 11 by 11 chunk spiral around that chunk
takes the first column PlayerSpawnFinder.getSpawnPosInChunk accepts
(getLevelRespawnPos: MOTION_BLOCKING, WORLD_SURFACE and OCEAN_FLOOR
heights, no liquid, a full-topped collision shape); without one, the
origin chunk's centre at the spawn height. The first player is then
placed by PlayerSpawnFinder.findSpawn: the respawn-radius square in a
random coprime order, a free player box with liquids colliding, else the
spawn with fixupSpawnHeight; adventure mode uses the spawn as is.
horizon-world gains worldgen::spawn_finder and OverworldGenerator::origin;
the client's world_spawn module and game/initial_spawn.rs follow
setInitialSpawn and PlayerSpawnFinder, for single player and the
dedicated server alike. Flat worlds use FlatLevelSource's origin and
spawn height.
tools/test_world_spawn_parity.py compares, for the report's seed, the
chunk harness's seeds and 50 random ones, the origins
(VanillaWorldSpawnOracle against the spawn_origin example) and the
level.dat spawn of worlds created by the vanilla dedicated server and by
Horizon's. Vanilla is not deterministic on chunk corners where trees of
neighbouring chunks overlap (decoration order follows its worker
threads), so a mismatching seed reruns vanilla up to three times.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A right click with food that is also a block item tried to eat it before
placing it: in creative always, in survival whenever the player was
hungry. BlockItem.useOn places first and only eats a consumable that could
not be placed (counting the use, as ItemStack.useOn does). Picking a sweet
berry bush or cave vines dropped nothing: the harvest loot
(Block.dropFromBlockInteractLootTable) is now rolled and popped. Bone meal
on a bush that can still grow grows it (SweetBerryBushBlock.useItemOn
passes it on) instead of picking it, and counts the use with its
item_interact_finish vibration. Every block placement played its sound
twice; BlockItem.place's level event is now the only one.
tools/test_use_item_on_parity.py drives vanilla's real
ServerPlayerGameMode.useItemOn (and Minecraft.startUseItem's useItem
fallback) against use_item_as_player in a headless game.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
mouse-move DX DY (MCP mouse_move) adds relative motion to MouseHandler's
accumulated movement, so the next frame turns the player with
turnPlayer's sensitivity curve ((s*0.6F+0.2F)^3*8, no 8 while scoping),
the inverted axes and the cinematic camera, then Entity.turn. The motion
counts as a grabbed pointer in an active window until the real pointer
is grabbed or released again.
turn and look-at (and look-at-block/look-at-entity) take --over SECONDS:
the turn becomes that motion fed over the time with ease-in-out (the
shorter way round for look-at), replying once the last of it has been
turned. Without --over they still jump. turn is now an MCP tool. stop
cancels a turn in progress.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The debug crash from #90 printed a stand-in report; it now ends the game
with a DebugCrash error that main reports as "Manually triggered debug
crash" with the stack where it was raised. The game keeps the error that
ends it as an error, not a string, so main can tell. Leading std backtrace
frames are dropped (macOS keeps them).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Entity.applyEffectsFromBlocks now runs WebBlock.entityInside and
SweetBerryBushBlock.entityInside for the player through
Player.makeStuckInBlock (no slowdown while flying, impulse context reset),
with Weaving's lighter cobweb hold and the sweet_berry_bush damage of a
grown bush while the known movement is at least 0.003 on X or Z. The scene
motion harness gains cobweb and berry-bush fixtures run through vanilla's
real applyEffectsFromBlocks.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The game writes logs/latest.log in its home like vanilla's log4j config
(client-1.21.2.xml): [HH:mm:ss] [thread/LEVEL]: message, rolled at start and
at midnight to logs/<date>-<n>.log.gz, seven a day. Messages go through the
log facade and still print their old text on stderr.
A main-thread panic or a fatal error writes
crash-reports/crash-<date>_<time>-client.txt (error, stack, Horizon and
Minecraft versions, OS, CPU, memory, GPU adapter, render settings, resource
packs, world, last 50 log lines). Development builds crash on purpose with
HORIZON_DEBUG_CRASH.
The launcher shows "Horizon crashed" when the game it started ends badly,
with Open report, Copy report, Open logs folder and Report on Discord;
Settings gets a Logs row. horizon-launcher logs and horizon-ctl logs
(--lines N, --crash) print the log's end or the newest report, and both MCP
servers have a logs tool.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Fullscreen: applying settings (any option button, joining a world,
travelling, reloading packs) compared the Fullscreen option with what the
window reported and forced the window back to the option. A window made
fullscreen by the compositor (a Hyprland keybind or window rule) left
fullscreen on the next button click. The window now follows blaze3d
Window's fullscreenRequested/fullscreen pair: it is only asked to change
when the option changes (updateFullscreenIfChanged), a change the window
system makes is adopted by the option (updateFullscreenState ->
Minecraft.fullscreenStateChanged), and the window is created fullscreen
when options.txt says so.
Clipboard: Ctrl+A/C/V/X (Cmd on macOS) in the chat, signs, book and quill
(page and title), anvil, creative and recipe-book searches, the language
and pack searches, the LAN port, world name and seed, and the server name
and address boxes, through the arboard clipboard (Wayland data-control,
falling back to X11). Shortcuts follow InputWithModifiers.isCopy/isPaste/
isCut/isSelectAll on the modifier state and the layout's letter; pasted
text goes through each field's own vanilla rules: StringUtil.filterText
and setMaxLength for EditBox, line breaks kept and the 1024-character and
14-line limits for MultiLineEditBox, formatting stripped and whole-or-
nothing width checks for TextFieldHelper on signs.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The new particle types now come from the events that spawn them in
vanilla.
- Level events 3003 (wax on), 3004 (wax off), 3005 (scrape):
ParticleUtils.spawnParticlesOnBlockFaces, three to five particles on
each face, from the client level's random. They are sent by honeycomb
on a block (HoneycombItem.useOn), axe scraping and wax removal
(BlockTransformer's TransformParticle.send), dispensed honeycomb, and
copper golems being waxed or scraped.
- Level event 3002: a struck lightning rod sends sparks along its axis
(ParticleUtils.spawnParticlesAlongAxis); without an axis they cover
the block's faces.
- Eating: LivingEntity.spawnItemParticles, five crumbs every
four ticks and sixteen when consumed (Consumable.onConsume), placed in
front of the eyes and thrown along the look, as item particles in full
colour. They were greyed terrain crumbs above the eyes. They now
follow Consumable.shouldEmitParticlesAndSounds with the remaining ticks
before the countdown, as LivingEntity does: six times over 32 ticks,
not seven, each one a tick earlier.
- Snowballs (Snowball.handleEntityEvent(3)): eight item crumbs at rest.
Eggs (ThrownEgg): eight scattered by up to 0.04 a tick.
- Slime, magma cube and sulfur cube landings (AbstractCubeMob.tick):
size × 16 item_slime, flame or sulfur_cube_goo particles around the
body, from draws the mob's random also makes on the server.
New harness tools/test_emitter_particles_parity.py: the real
ParticleUtils emitters and LivingEntity.spawnItemParticles.
Part of #63
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On Linux, once the launcher's file has been replaced under the running
process (an update from the command line or another window), current_exe()
is "<path> (deleted)" and self_replace failed with "No such file or
directory". The new binary now goes to the path the launcher started from
(copy beside it, rename over it; Windows still uses self_replace), and an
update the file already has is not downloaded again. The message reads
"Updated to X. Restart the launcher to use it."; Restart execs the new
binary with the same arguments (spawn and exit on Windows) and says so if
it cannot start.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The launcher prepares a version in-process when it ships the preparer sources
the launcher was built with, with the version's own horizon-prepare when they
differ, and with the Python script only for versions made before it. Release
folders ship horizon-prepare and its sources; resources are reused between
versions with the same preparer. The game's messages, PLAY.txt, the docs and
release builds name horizon-prepare, and tools/prepare_native_client.py is a
wrapper of it for the development tools.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A port of tools/prepare_native_client.py and the helpers it imported from
lab.py and the test_*_parity.py modules: the player's pinned game files are
imported and checked against research-lock.json, the client's assets and data
copied, the vanilla exporters compiled and run with the JDK (JAVA_HOME, the
PATH, or java_home on macOS, at the pinned java_major), and the resources and
their manifest written byte for byte as the script wrote them, with its
sha256-keyed skips. tools/compare_prepare.py checks that against the Python
preparer.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Horizon's engine knew 91 of vanilla's 128 particle types. These come from
their real providers, constructor to tick:
- sweep_attack (AttackSweepParticle)
- glow, wax_on, wax_off, electric_spark, scrape (GlowParticle and its
providers)
- squid_ink, glow_squid_ink (SquidInkParticle: sinks through air)
- item, item_slime, item_snowball, item_cobweb, sulfur_cube_goo
(BreakingItemParticle: a quarter of the item's ground-model sprite in
full colour, unlike terrain crumbs)
- flash (FireworkParticles.OverlayParticle, its sine pulse and fading
alpha), which firework explosions now show as vanilla's Starter does
- firefly (FireflyParticle: wanders, fades in and out, vanishes inside
blocks)
- trial_spawner_detection(_ominous) (TrialSpawnerDetectionParticle)
- trial_omen, raid_omen (SpellParticle)
- ominous_spawning (FlyStraightTowardsParticle, colour lerped like
ARGB.srgbLerp)
- spit (ExplodeParticle falling at 0.5), fishing (WakeParticle),
dust_plume (DustPlumeParticle)
- red/orange/yellow_poplar_leaves (FallingLeavesParticle.PoplarProvider)
- pause_mob_growth, reset_mob_growth (SimpleVerticalParticle)
- block_marker (BlockMarker)
ParticleType.getOverrideLimiter now lists every type vanilla registers
with it (gust, sonic_boom, sculk charges, campfire smoke, vault_connection
and the new ones were missing), so they show beyond 32 blocks.
The /particle commands take the new types; item particles resolve the
stack's ground model.
The particle_providers harness now builds all 120 types Horizon implements
through the real providers (1428 cases).
Part of #63
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Villagers showed one angry, happy or heart particle where vanilla's
AbstractVillager.addParticlesAroundSelf shows five, spread over the body
with Gaussian motion. Taming and love bursts were spread about the middle
of the body rather than over it, and drew from the world random.
particles::around_self places a burst as the client handlers do: per
particle three Gaussian speeds, then getRandomX(1), getRandomY() + y,
getRandomZ(1). Burst holds the counts, speeds and heights vanilla uses,
and entity_event_burst maps entity events to them. The client side uses
the particle engine's random, so no server draw moves.
- villager hurt by a player, failed breeding (13), breeding hearts and
the child's (12), trades, job sites and beds (14): five particles about
the villager, not one 1.8 above its feet.
- wolf, cat, parrot, nautilus and horse taming (7/6): seven hearts or
smoke about the body.
- allay duplication (18): three hearts about the allay.
- llama growth from food: one particle at rest about the body.
- animals in love: the heart moves at the Gaussian speeds already drawn,
exactly where vanilla puts it; forced growth's particle sits where it
was placed (sendParticles with no spread) instead of being jittered.
New harness tools/test_entity_particles_parity.py: the real client
handleEntityEvent of Villager, Animal, TamableAnimal, AbstractHorse,
Ocelot, Dolphin and Allay against entity_event_burst and around_self.
Part of #63
Co-Authored-By: Claude Opus 5.5 <[email protected]>
F3 only toggled the debug overlay. 26.3 makes the debug keys rebindable
mappings (key.debug.*): the modifier (F3) held with another key runs
handleDebugKeys, released alone toggles the overlay unless it was used as
a modifier; F3+Escape pauses without the menu; F3+C held ten seconds
crashes the game.
- debug_keys: KeyboardHandler.keyPress's state (usedDebugKeyAsModifier,
the modifier's down state, debugCrashKeyTime) and tick, with the actions
in handleDebugKeys' order; unit tests replay key sequences.
- gizmos: 26.3's gizmo lines (colour, width in pixels, near-plane clip as
GizmoFeatureRenderer.buildLines), ChunkBorderRenderer's geometry and
EntityHitboxDebugRenderer's boxes, eye lines and view arrows; a coloured
line pipeline draws them, depth tested or always on top.
- game/debug_keys: the actions. F3+A rebuilds every section, B hitboxes,
C copies the location (/execute in ... run tp @s, %.2f half up), D
clears chat, G chunk borders, H advanced tooltips (options.txt), I copies
/setblock or /summon with data (Shift: client side), N toggles spectator
and the previous mode with /gamemode's feedback, P pause on lost focus,
T reloads resource packs, V dumps the version, X improved transparency;
F4 without permission says debug.gamemodes.error. Messages carry the bold
[Debug]: prefix in yellow (red for the crash countdown).
- PauseScreen(false): only "Game Paused" at the top; the
pauseOnLostFocus option now decides whether losing focus pauses.
- The clipboard through arboard; `horizon-ctl status` reports the debug
state and the text last copied.
Fixes#73
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Every vanilla Particle owns a RandomSource.create() of its own; providers
draw from ParticleEngine.random, a few from ClientLevel.random, and spells
from SpellParticle.RANDOM. Horizon drew everything from one generator, so
draws landed on the wrong stream. Particles now carry their own random,
and provider, level and spell draws use theirs, as in vanilla.
Fixed against the real providers:
- gust / small_gust: SingleQuadParticle(x, y, z) constructor, so the
12..15 tick lifetime comes out of the right draws.
- explosion_emitter, gust emitters: NoRenderParticle draws no quad size.
- effect, instant_effect, entity_effect, witch, infested: x/z motion from
SpellParticle.RANDOM; the witch's brightness from the provider.
- landing_honey, landing_obsidian_tear, falling_nectar, spore_blossom_air,
crimson_spore, warped_spore, white_ash, sculk_charge(_pop), dolphin:
provider draws (lifetimes, motion, alpha) from the provider random, after
or before the sprite as the provider orders them; crimson spore motion
scaled by the float constants; composter lifetime from the level random.
- enchanted_hit: MagicProvider's 0.3/0.8 tint after the constructor tick;
damage_indicator lives 20 ticks (DamageIndicatorProvider).
- note: colour from Mth.sin's table.
- shriek: one random sprite, not animated.
- trail: Particle(x, y, z, xa, ya, za) constructor, ARGB.scaleRGB on the
8-bit channels, Mth.lerp towards the target including the last tick.
- firework sparks: full bright (SimpleAnimatedParticle.getLightCoords).
New parity harness tools/test_particle_providers_parity.py: the real
ParticleResources providers build every type Horizon implements in a
stand-in ClientLevel (air, optionally a stone floor); the oracle recovers
each particle's clock-seeded random and the probe replays it; state,
requested particles and sounds are compared tick by tick.
Part of #63
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Powder snow was inert: the player's collision context never carried its
fall distance or boots, block-inside effects never ran for the player,
frozen ticks stayed 0 and the empty bucket only took fluids.
Player (PlayerMotion, LivingEntity.aiStep order):
- PowderSnowBlock.getCollisionShape gets the player's fall distance and
leather boots: it sinks through, walks on top in leather boots (sinks
when crouching), lands on the 0.9-high shape after a fall of more
than 2.5 blocks.
- Entity.applyEffectsFromBlocks over this tick's movement path (the
ported block sweep, StepBasedCollector steps): entityInside makes it
stuck (0.9, 1.5, 0.9) while its feet are in powder snow and sends the
server's snowflake from the level random; FREEZE sets isInPowderSnow
and adds a frozen tick per step; EXTINGUISH destroys the snow if the
player burns and puts the fire out. ServerPlayer.onInsideBlock fires
enter_block for every block entered (which also unlocks the boat
recipes in water, and the end gateway advancement).
- baseTick's wasInPowderSnow lets leather boots climb out (living_air).
- Server freezing: thawing 2 a tick, the powder_snow frost modifier
(-0.05 x percent frozen, not over air), freeze damage every 40 ticks
when fully frozen; leather armor (#freeze_immune_wearables) and
spectators don't freeze.
- HUD: the powder snow outline overlay and frozen hearts.
Items and blocks:
- BucketItem.use on powder snow: pickupBlock (2001), used statistic,
fill sound, FLUID_PICKUP, createFilledResult, filled_bucket trigger.
- SolidBucketItem: the empty_powder_snow place sound and an empty bucket
back in survival.
- Falling blocks land on powder snow, or end up inside it and break
after a fall of more than 2.5 blocks.
test_native_scene_motion gains powder snow fixtures ("snow_tick": the
oracle runs vanilla's real applyEffectsFromBlocks and freezing) and
prints frozen ticks, the snow flags and the stuck multiplier.
Fixes#62
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Taming any wolf also granted The Whole Pack and A Complete Catalogue: entity
predicates treated minecraft:components as always true, so every
tame_animal criterion of those advancements (one per wolf or cat variant)
matched the first animal tamed. EntityPredicate tests them with
DataComponentExactPredicate against what the entity reports through
Entity.get (Wolf.get(WOLF_VARIANT) and so on): mobs now report their variant
component (wolf, cat, frog, pig, cow, chicken, zombie nautilus) and the
predicate compares it, an absent component never matching.
Checked against the real EntityPredicate of every criterion of the variant
advancements on real entities of every variant by
tools/test_entity_components_parity.py.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Choosing packs in the Resource Packs screen only reloaded the GUI atlas
(and only when it kept its size), item models and languages; block
textures and models changed after a restart, and sounds, clouds and the
world border never did. Block textures were also squeezed to 16x16, and
the pack overlay took every file from the top pack, so a pack's own
en_us.json, sounds.json, font or atlas definition replaced vanilla's
instead of adding to it (all other text, or every vanilla sound, gone).
- Minecraft.reloadResourcePacks: rebuild block sprites and models, mesh
tables, the shared atlas, item models, sounds, languages, GUI
metadata, clouds.png and forcefield.png from the new overlay; swap the
mesher's context (workers read it per job), mesh every loaded section
again (LevelRenderer.allChanged) and build a renderer with the new
textures. Nothing switches if loading fails.
- The block sprite array takes the largest texture's resolution (power
of two, at most 256 MiB), so 32x/64x packs keep their detail;
animations upload at that size.
- Files vanilla reads as resource stacks are merged in the overlay:
ClientLanguage (keys override), SoundManager (sounds add up unless an
event replaces; a sounds.json Gson rejects is skipped whole),
FontManager (top pack's providers first), SpriteSourceList (sources
bottom up). Languages are read from every namespace, not only
minecraft.
- Missing or broken block textures use vanilla's missing texture; packs
that still break startup fall back to the default resources
(Minecraft.rollbackResourcePacks).
- Open Pack Folder opens the folder on macOS and Windows too.
tools/test_resource_pack_parity.py checks pack discovery, overlays,
filters, which pack provides a file and the merged languages, sounds,
fonts and atlases against vanilla's FolderRepositorySource,
MultiPackResourceManager and reload listeners.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Quick move, the recipe book's place-all, its search box and the hold-Shift
tooltips read self.keys, which holds remapped and toggled keys: with toggle
sneak, a Left Shift press behind a screen is dropped by the sneak mapping and
sync_toggle then removes ShiftLeft. Vanilla reads the event's own modifiers
(InputWithModifiers.hasShiftDown) or Screen.hasShiftDown: the physical keys.
Track them in physical_keys and read them through has_shift_down.
Clicking outside the window did nothing. AbstractContainerScreen.mouseClicked
sends slot -999 when hasClickedOutside (AbstractRecipeBookScreen's rule when
the screen has a book): a held stack starts a quick craft drag whose release
sends PICKUP -999, which AbstractContainerMenu.doClick turns into a drop of
the stack (left) or one item (right), thrown from the hand with a swing. The
mouse handling is now horizon_client::container_screen, checked against the
real screens by tools/test_container_screen_parity.py; servers get the same
ContainerClick.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Vanilla keeps xBob and yBob on the LocalPlayer: applyInput eases them
halfway to the player's rotation every tick, and
FirstPersonHandsAndItemsRenderer turns both hands by
(viewXRot - xBob) * 0.1 degrees about X, then (viewYRot - yBob) * 0.1 about
Y, before drawing either (FirstPersonHandsAndItems.extractRenderState
interpolates the bobs). Horizon drew the hands fixed to the view.
- first_person::HandSway: the bobs, applyInput's tick and the rotation.
- The game ticks it with the hands and resets it where vanilla makes a new
LocalPlayer (joining, respawning, changing dimension, leaving the End);
hand_geometry turns the arm, held items and maps by it.
The first-person harness's steps can turn the player: the oracle sets the
LocalPlayer's rotation and calls the real applyInput, and both sides print
xBob, yBob, xBobO and yBobO every tick (turning scenarios plus random turns
in the random cases).
Fixes#66
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Placing fire with a fire charge played item.firecharge.use at the flint and
steel's pitch (nextFloat * 0.4 + 0.8, one draw). FireChargeItem.playSound
uses (nextFloat - nextFloat) * 0.2 + 1.0, two draws from the level random.
The pitch is now horizon_client::items::fire_starter_pitch, checked against
the real FireChargeItem.playSound by tools/test_fire_starter_sound_parity.py.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Game rules lived twice: Level (loaded when the world opened, written to
data/minecraft/game_rules.dat by Level::write) and the rules the game and
/gamerule use. Every save wrote the file first with the rules loaded at
open, then again with the current ones; a failed second write (on Windows a
file just renamed into place can be held briefly by the indexer or an
antivirus) was only logged, leaving the old rules on disk, so the world
opened with them again.
- sync_level copies the current rules into Level, and Level::write is the
only writer of game_rules.dat (written as vanilla writes saved data:
`_new`, then renamed over the old file).
- The world is saved when the application ends without a window close
(ApplicationHandler::exiting): on macOS, Quit (Cmd+Q) ends the
application through applicationWillTerminate, while vanilla's GLFW window
turns it into a window close and Minecraft.stop saves the world. Window
close, `horizon-ctl quit` and exiting share save_for_exit, which saves
once.
Tests: a unit test writes changed rules through Level::write and reads them
back (GameRules::load and Level::read); the save round-trip harness has
Horizon's player change a rule with /gamerule before saving and requires the
new value in the world vanilla then loads.
Fixes#74
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Animal.mobInteract plays playEatingSound(), which is empty in Animal and
overridden only by armadillos, cats, foxes, frogs, goats, pigs, sniffers and
sulfur cubes (plus nautili). Horizon fell back to entity.generic.eat for every
other animal (cows, sheep, chickens, rabbits, turtles, bees...), with a level
random pitch instead of each class's volume and pitch.
- mobs::ai::eating_sound: the overrides with their exact volume and pitch
(makeSound with the voice pitch, playSound(1, 1), or Level.playSound with
Mth.randomBetween(0.8, 1.2)); silent otherwise.
- mobs::ai::feed_animal: Animal.mobInteract's food part, shared by the game
and the mob AI probe; Strider.mobInteract adds entity.strider.eat.
- AgeableMob.ageUp(seconds, true): forcedAge and the forced-age particle
timer, whose positions draw from the mob's random on the server.
- Tame wolves and cats healed with food: TamableAnimal.feed (wolves silent).
- Parrots: Parrot.isFood is false; wild ones take parrot food with
entity.parrot.eat and a one-in-ten taming chance.
- Baby pigs and chickens use their sound variant's baby_sounds.
The mob AI harness gains a `feed TICK MOB ITEM` event (the oracle calls the
real Animal.mobInteract) and feeding scenarios for adult and baby cows, pigs,
sheep and chickens.
Fixes#68
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Right-clicking a composter with wheat (or any compostable item) did
nothing: ComposterBlock.useItemOn wasn't implemented, only emptying a
full composter. Hoppers and droppers didn't see the composter as a
container either, and a hopper under one sucked items through it.
- useItemOn: below level 7 a compostable item adds a layer with the
chance of its compostable component's provider (evaluated in the
block-interaction loot context, drawing from the level random), then
level event 1500, the used statistic and one item consumed (none in
creative); level 7 still consumes the action; a full composter falls
through to useWithoutItem and empties.
- Level event 1500 (handleFill): composter.fill(_success) and ten
composter particles above the compost.
- getContainer: the input container takes one compostable item from
above (hoppers, droppers, crafters), the output container gives a
hopper below the bone meal of a full composter, emptying it twice as
vanilla's setChanged calls do (and losing it when the hopper is full);
levels 7 and other faces refuse.
tools/test_composter_parity.py runs right-clicks, hoppers above and
below, and droppers into composters at every level through
VanillaDispenserOracle (new "use" and "hopper" actions) and the game's
dispense fixture.
Fixes#70
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The cloud mesh radius came from the render distance, the far plane was
capped at max(4 x render distance, 1024) and the cloud fog end was a
constant 2048, so lowering the render distance cut the clouds short.
Vanilla ties all three to options.cloudRange: CloudRenderer.prepare
builds ceil(range * 16 / 12) cells, Camera.depthFar is
max(4 x render distance, range * 16) (also the hand projection, the
frustum and the world border), and AtmosphericFogEnvironment ends
cloud fog at min(range * 16, visual/cloud_fog_end_distance).
The cloud render harness now checks the radius for Cloud Distance
2..128 against the shipped mesh builder, and the environment harness
covers visual/cloud_fog_end_distance.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A click in the furnace menu reset the cooking timer whatever it did.
AbstractFurnaceBlockEntity.setItem restarts cooking only when the input
slot gets something other than the same item with the same components,
and menu clicks that take part of the input or merge into it never call
it. Furnace::input_clicked applies that after each click, with the new
input's total cook time.
test_menu_click_parity gains furnace_be, blast_furnace_be and smoker_be:
the server's furnace menus over real block entities, comparing the
cooking data slots after every click.
Fixes#69
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Horizon stretched a quarter of rain.png over each whole column, up to
20 blocks tall, so streaks were about five times too long. Vanilla
advances v by a quarter per block and lets the sampler repeat, so the
texture repeats every 4 blocks. Port createRainColumnInstance,
createSnowColumnInstance and prepareInstances exactly (seeded column
speed, game-time scroll, snow drift and brightening, camera-block size
table, square radius from the Weather Radius option, alpha falloff,
precipitation at the camera's y), and split each quad at the texture
seams so the atlas sprite repeats.
tools/test_weather_render_parity.py compares the columns and quads
with the shipped renderer.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Bind resident environment kernels and real fire clocks, fix sprite uploads and flame material, restore swim speed, throttle item lava audio, and emit fire extinguish events. Reset world-entry progress state without deleting saved data. Exact axolotl replay parity remains deferred to issue #64.
The README keeps the parity table, and its known differences move to a
Known issues section. The website drops the What's next section and its
nav links, and no longer fetches or requires a Roadmap section. The
launcher drops its What's next page, `next` subcommand and `roadmap`
MCP tool.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
Lists the GitHub contributors of ludovic111/horizon (bots left out, the
owner first) with avatar, login, commit count and a link to the profile.
Built in by build.mjs like the releases (the build still succeeds when
GitHub does not answer, and the block stays hidden until the page loads
it) and refreshed by the page from the API. Adds a nav and menu entry,
lets avatars.githubusercontent.com through img-src, and updates the
README, LEGAL.md and the privacy note.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
A chunk in the decoration range whose diagonal neighbour falls outside the
generation range never gets its 3x3 protos, so it must not hold back the
higher-ranked chunks within two of it. Render distances 2, 5 and 12 load and
keep loading while flying without stalling.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A new harness decorates a 13x13-chunk area around each site, on both sides,
in the order vanilla's chunk scheduler uses around a player standing there
(chunk_dump HORIZON_LIST_ORDER prints decoration_order::vanilla_order), and
compares every block state, quart biome, block entity and scheduled tick:
the two reported spots on seed 1891623539983467248 (taiga trees at
204 -239; a stronghold with sculk and a geode at 4870 401), eleven tree-heavy
biomes and the nearest mineshaft, village, ancient city and trial chambers of
two seeds. 21 sites, 3,549 chunks, 0 differences.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
StructurePiece.placeBlock schedules a tick for the fluid of every block it
sets (the lava of a stronghold's portal room, a fortress's lava wells, water
in a monument or a mineshaft); in a WorldGenRegion it lands in the proto
chunk and runs once the chunk is live. The port treated it as a no-op, so
those fluids sat still until a neighbour changed.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Biome decoration is order dependent: a chunk's features write into its
neighbours, so a tree, ore blob or sculk patch placed first changes where the
next chunk's features fit and how many random draws they take. Horizon's
streamer decorated whichever chunk had its 3x3 protos ready, nearest first but
spread over non-overlapping regions for parallelism; around a teleported
player that swapped 45% of the overlapping chunk pairs against vanilla (whose
own runs swap 6%), and grew visibly different forests (seed
1891623539983467248 at 204 108 -239), ore veins, sculk and geodes.
worldgen::decoration_order replays vanilla's chunk scheduler for a standing
player: loading tickets ring by ring in fastutil hash order, four in flight,
FULL tasks decorating their 3x3 squares x-outer by level, INITIALIZE_LIGHT
tasks themselves, a ticket released once its 5x5 square is full. The streamer
ranks chunks with it whenever the player's chunk or the render distance
changes and decorates a chunk only when no lower-ranked chunk within two of it
is still waiting, so chunks further apart still decorate in parallel. Around
that spot the decorated blocks that differ from a vanilla server fall from
about 9000 to about 1000 (two vanilla runs differ by about 200).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
VanillaTerrain::finish dropped the proto chunk's block and fluid ticks; vanilla's
LevelChunk takes them over from the ProtoChunk (unpackBlockTicks), so the leaf
re-measurements and the fluid ticks of lakes and springs run once the chunk is
live. They now travel in the chunk's ChunkExtras, like a saved chunk's ticks,
and register with the level's scheduler on arrival; block ticks are translated
from worldgen's default state ids to the game's block indices.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Both dumps list each proto chunk's scheduled block and fluid ticks in
scheduling order (`t block|fluid NAME X Y Z PRIORITY`) and the harness
compares them: they run once the chunk is live, and the leaf ticks of
updateShapeAtEdge were missing without any block differing at FEATURES.
HORIZON_ORDER names a file of `x z` lines both sides decorate in, instead of
row-major: decoration is order dependent (a chunk's features write into its
neighbours), and vanilla's chunk system decorates around a player in its own
order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
StructureTemplate.updateShapeAtEdge (every tree, structure pieces) and
Block.updateFromNeighbourShapes run updateShape on a WorldGenRegion, where
LeavesBlock.updateShape schedules a one-tick block tick to re-measure its
distance and waterlogged blocks schedule fluid ticks: vanilla keeps them in
ProtoChunkTicks and runs them once the chunk is live. Horizon ran updateShape
against a throwaway tick scheduler and lost them, so leaves where trees
overlap kept the distance their own tree gave them (one more than vanilla's
for thousands of leaves in a taiga), and could decay where vanilla's don't.
level::Ticks gains a capture mode; the decoration world schedules what a
shape update captured, in order, before the update's own setBlock.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Conflicts with #51: entity.wgsl keeps the crumbling sprite UVs and
samples with the per-face colour; gfx.rs keeps crumbling_sprite and
outline_params and puts the Nether flag in time.y; game.rs keeps #51's
breaking and cloud geometry. #51's new vertices (breaking decals,
cloud faces, the batching test) get the new zero normal, so they stay
unshaded.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
test_dedicated_server_parity summons a librarian beside the spawn and the
vanilla protocol client right-clicks it: the OpenScreen packet (title as
Entity.getDisplayName, UUIDs left out) and the MerchantOffers packet must
match vanilla's server. Villagers keep Xp, LastRestock and RestocksToday
before their offers are made (vanilla's null offers).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The host hands a trading screen a guest's interaction opens to that guest:
OpenScreen (minecraft:merchant), MerchantOffers when the offers change,
the select trade packet, container clicks run on the merchant menu engine,
the TRADE reputation event and prices for the guest's UUID, the payment back
on close. As a client, Horizon shows a server's trading window from its
MerchantOffers and sends select trade and clicks. Stored enchantments, dyed
colours and map ids now travel in item stacks.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The trading screen runs MerchantMenu on the menu engine with the villager's
merchant::Session: level-ups add offers while it is open, prices follow the
TRADE reputation event a tick later, orbs drop at the villager with its own
random, the yes/no voice and regeneration as vanilla; special prices stay
after closing. Offers use the world's RandomSequences (now saved as
random_sequences.dat), the villager's position, its namespaced type (desert,
snow... trades) and exploration maps found through /locate.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Resolve the gfx.rs conflict with #42 and keep the shared Globals uniform
consistent: Rust, terrain.wgsl and entity.wgsl now all end with
time, crumbling_sprite, outline_params. The automatic merge left
terrain.wgsl reading outline_params at crumbling_sprite's offset. A test
compares the three declarations.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
VanillaTradingOracle runs villagers, wandering traders and MerchantMenu on a
real flat server; trading_fixture runs horizon_client::merchant and a new
merchant kind of the menu engine (MerchantContainer, MerchantResultSlot,
tryMoveItems, quick move) over a merchant::Session. Generation now covers
random dyes, random potions, stew effects, exploration maps (a MapMaker hook),
sum/binomial providers and the trade's max uses / xp / discount providers;
getRecipeFor honours the selection hint as vanilla, Offers leave defaults out.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Choosing Classic in Options > Interface left no way back. The classic
Video Settings page, which already starts with Horizon's graphics row,
gets a Horizon-only "Menu Style: Classic" button below it: it switches to
Horizon's menus, saves horizon.txt and opens Options > Interface. The
classic main Options screen keeps vanilla's exact layout.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The client's MotionWorld.apply_movement_emission pushed a footstep event
on every Entity.move, and the game played the block's step sound for
each. Gravity moves a grounded player every tick, so standing on grass
played block.grass.step 20 times a second, overlapping into an endless
rustle (19 instances alive at once). The audio stream had been muted by a
saved WirePlumber setting until v2026.10.13.1, which hid it.
Route the player through horizon_world::movement_emission (vanilla's
applyMovementEmissionAndPlaySound: moveDist, nextStep, the supporting
block checks) with Player.getMovementEmission (nothing while flying or
sneaking on the ground). The status sound census now lists the events
holding channels, which is how the loop was found.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The creative-tabs.json export failed to build the functional blocks tab (no
data-pack registries: banner patterns), and its fallback read decompiled
sources from a path that no longer exists, so the tab was empty and anvils,
crafting and smithing tables were missing from the tabs and the search. The
search also matched only an approximate display name.
VanillaCreativeTabsExporter now builds every tab with vanilla's own
CreativeModeTabs.tryRebuildTabContents over the vanilla data pack's
registries and writes creative-inventory.json: stacks with components, and
the search tab's creative tooltip lines and tags. The client searches them
as SessionSearchTrees does (FullTextSearchTree, IdSearchTree after '#').
Changing tab empties the search box, as in selectTab.
The scrollbar only jumped to the pressed point; holding it now scrolls as
the mouse moves, anywhere, until release (mouseClicked/mouseDragged), and
the wheel and the scroller follow ItemPickerMenu.
tools/test_creative_parity.py compares every tab and 6,793 searches with
the vanilla client's own search trees. horizon-ctl `type` also fills the
creative search box.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Hardware LineList forced one-pixel strokes, while the guessed 0.002 endpoint growth did not separate the edges from block surfaces. Render screen-space strips with vanilla framebuffer width scaling and perspective/view depth layering, leaving the shape edges unchanged. Apply normal outline fog and add three regression tests.
Offline gate passed on this main-based branch. Actual vanilla GLSL vs Horizon WGSL GPU comparison covers 3,072 vertices; maximum projected difference is 0.0036621 pixels. No code from PR #42 is included.
README gets an Agent control section (out of the roadmap: it's done), the
native client guide describes every command group, the blocking semantics,
exit codes, launch discovery, MCP and the two transports, and PLAY.txt tells
release players about horizon-ctl.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
release.yml builds horizon-ctl beside the client (checked for system-only
imports like it), packaging/package.sh copies it from beside the client binary
and refuses to package without it (the macOS folders made by hand go through
the same script), and package_native_client.py adds it to private archives.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On a virtual screen: create a world through the title menus, walk to a tree,
break a log, craft planks in the inventory screen, place one and read it
back, once with horizon-ctl commands and once over horizon-ctl mcp.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
One command table drives horizon-ctl help, COMMAND --help (with examples)
and the MCP tools. Replies stay one JSON line; errors are error objects with
exit status 1 (game), 2 (usage) or 3 (unreachable). script runs a file or
stdin with shell-like quoting. launch finds the game beside horizon-ctl or
the newest launcher install (--version, --game), with its prepared resources
and the game home, and --wait waits for the world. horizon-ctl mcp serves
MCP over stdio in the launcher's style, one tool per command, with the
screenshot returned as an image. Existing commands and replies are unchanged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
game/agent.rs adds commands an agent can play with: state, looking-at,
entities, block, blocks, find-block, inventory, screen and messages read the
game as JSON; move, jump, goto (A* over standable blocks), look-at, attack,
break, place, interact, use-item, select, drop, open-inventory, click-button,
click-at, command, respawn, pause, wait and stop act like a player. Actions
that take time run as a task advanced every frame and reply when done, with
what happened (break reports drops and pickups, place the block placed).
Buttons are clicked by label: the classic and Horizon button drawers record
what they draw each frame (widgets.rs), and a click is a synthesized pointer
move and press, so it behaves exactly like the mouse on every screen.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Windows has no Unix domain sockets, so the control endpoint there is a random
127.0.0.1 port described by a JSON file (port and a 128-bit token, readable by
the user only); a client's first line is auth TOKEN. Linux and macOS keep the
Unix socket, and any endpoint path ending in .json picks the TCP transport, so
it is tested on Linux too. control::connect and Stream serve both.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Launcher 0.1.3 auto-installed v2026.10.13.1 unprepared; after updating to 0.1.4 nothing installs
it again, so install's reuse never ran for it. backend::reuse_prepared_everywhere() applies the
same reuse (same compatibility checks, copied then renamed) to every installed version that is
not prepared, oldest first. It runs when the window starts (on a thread; the page refreshes when
it is done), in the CLI's `versions` and `update`, and in the MCP list_versions and check_updates
tools. The partial folder is now named per copy, and a version prepared meanwhile is left as it is,
since the window's start and an install can copy at the same time.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
After installing a version (CLI install and update, the window's automatic install, MCP),
when it is not prepared and another installed version with the same Minecraft version is,
the launcher copies that version's prepared resources. Players who auto-updated to
v2026.10.13.1 no longer prepare again, and can play it without their Minecraft folder.
The prepared output depends on the Horizon version as well as Minecraft's: the manifest
records the hashes of the vanilla exporter sources it was made with, and
prepare_native_client.py writes every file and the manifest's schema. So resources are reused
only when both versions have the same pinned jars in research-lock.json, the same
prepare_native_client.py and the same exporter sources the manifest lists. They are copied
(prepare rewrites files in place, so hard links would tie the versions together), into a
partial folder renamed when complete. install and the MCP tool say when it is prepared already.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- A banner in the website's design (Instrument Serif wordmark over a dimmed
sunset, purple accent), rendered from docs/media/banner.html with Playwright.
- Screenshots of Horizon's own menus and graphics settings, the in-game scenes
as WebP (from the website's media), and the launcher 0.1.3 Play and Patch
notes pages; the JPEGs the README no longer uses are removed.
- A "What changed" section matching the v2026.10.13.1 release notes.
- Download only through the launcher (playhorizon.dev or launcher-v releases).
- Roadmap: performance optimisation, mod support and agent control.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The crack overlay and particle emitters assumed a unit cube. Use selected baked model quads and projected decal UVs for cracks, outline bounds for mining crumbs, and per-box density for the destruction burst. Keep translucent overlays after translucent terrain.
Distinguish model offsets from outline offsets and preserve double-precision shape coordinates. Add regression tests and actual-client differential oracles: 2,740 model cases and 2,740 particle cases, each passing three runs after the upstream merge. Particle enqueue positions match exactly; model max error is 1.40395e-7. Offline gate, camera/first-person/cloud comparisons and Vulkan smoke checks pass.
Shared CI runners hit GitHub's anonymous rate limit in the smoke test (launcher 0.1.2's
macOS build failed twice on it). Players never set the variable, so nothing changes for them.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The launcher's palette, violet accent, Instrument Serif and Geist, without its sidebar
and settings rows: a full-viewport hero over Horizon's own sunset (slow drift,
parallax, grain), then sections for what it is (a statement that lights up as it
scrolls), a screenshot mosaic, the parity count and a grid of the 14 subsystems, the
launcher beside Horizon's own menus, the latest patch notes, the roadmap, community
and the footer. Reveals, staggered letters and magnetic buttons stop under
prefers-reduced-motion; a translucent top bar and a full-screen phone menu replace
the sidebar.
playhorizon.dev: canonical, og:url, absolute og/twitter images, sitemap.xml and its
robots.txt line, and a 301 from *.up.railway.app (Railway's healthcheck host is still
served).
The full-page screenshots are no longer committed (about 17 MB): `npm run shots`
writes them to the gitignored shots/. The unused blurred background is gone; LEGAL.md
lists the hero and the settings screenshot.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
links.json gains a website key; the launcher shows a quiet Website link under
Join Discord and lists it in the links CLI command and MCP tool.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
lab.prepare imported every client library in the version JSON, ignoring
its OS rules, so a Linux install failed on the macOS-only
java-objc-bridge the official launcher never downloads there. The rule
evaluation moves from runtime.py into lab.py and one selector,
client_library_artifacts, now serves preparation, the runtime and the
model baking harness: OS rules as the launcher applies them, and the
natives for this architecture (classifier suffixes, plain natives being
the x64 build only when an ARM variant is listed). Required libraries
are still checksummed and a missing one still errors.
Mojang republished 26.3.json twice (702fe591, then 11f1aefe, the current
one) with only assetIndex changed: non-English translations, same jars
and libraries. The lock pins the current file and accepts both earlier
ones; preparation takes the launcher's copy so it matches the asset
index installed with it, and an unknown republish errors clearly.
Fixes#41
Co-Authored-By: Claude Opus 5.5 <[email protected]>
website/ holds Horizon's site: the horizon-theme tokens and bundled fonts, the
launcher's sidebar, rows, buttons and segmented control, over Horizon's own frozen
landscape. The parity table, harness count and roadmap are generated from the README
at build time; releases, patch notes, the roadmap and links.json are refreshed in the
browser from GitHub, with the build's copy as fallback. The download button picks the
newest launcher-v* asset for the visitor's OS.
Legal: the Minecraft disclaimer in the hero, footer and 404; no Minecraft logo or
wordmark; no game files; the launcher's background (licence unconfirmed) and the
screenshots showing it are left out; LEGAL.md lists every asset and its licence. No
cookies, analytics or third-party assets.
Railway: root directory website, Dockerfile (Node build, Caddy on $PORT with a strict
CSP). Checked with Playwright at 360-1920 px and 1250x700 (docs/widths/); Lighthouse
96/100/96/92 on mobile, 100/100/96/92 on desktop.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Villager POI search: main's per-search state classification replaces this
branch's state table (same sections skipped, same positions in order).
status keeps main's mob and block-entity counts and this branch's pending
uploads.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- links.json on main gives the Discord and sponsor links (built-in fallback offline,
empty hides the button); 'links' command and MCP tool.
- Install new versions automatically is on unless explicitly turned off (new settings
key, so 0.1.1's written default does not count as a choice).
- Settings: Horizon / Classic menu style, the horizon.txt menuStyle line the game reads;
'menu-style' command and MCP tool.
- launcher workflow: install latest into a temp home, check imports (Windows, Linux client)
and that the client starts, on all three platforms; also runs on launcher-* branches.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
The Horizon options shell's Video section now holds the graphics settings:
a Simple / Advanced switch (remembered in horizon-graphics.txt), then the
level's rows, each with its frame-time cost and one-line description
(ui_theme::described_row). Horizon settings draw as switches, sliders or
value buttons like vanilla options. The Classic style keeps the Video
Settings buttons to the same screens. Main's menu blur and the world's
resolution-scale pass are separate (post.wgsl, world_post.wgsl).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
At render distance 32 the single section arena grew past the 4 GiB buffer
limit and the renderer panicked. Sections now live in pages of at most
256 MiB (or the device's buffer limit); draws keep their order and bind each
page once per run of consecutive draws in it, so translucent sections stay
back to front. docs/graphics.md describes the settings, the benchmark and
what changed; graphics_bench.py --compare prints a before/after table and
keeps each game's log.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
AcquirePoi scanned every block within 48 of the villager (about 460,000),
naming each with string matches, every 20 to 40 ticks for each villager missing
a home, job site or meeting point. Villages gain such villagers over time
(babies, more villagers than beds), so the game thread's work grew with play
time: in a village, 18% of samples at the start and 30-42% eight minutes on,
frame rate dropping and staying low.
Block states are classified once per search, and 16^3 sections whose palette
holds no POI block are skipped whole (PoiManager only visits sections with
records). The same positions are visited in the same order, so the result,
ties in the distance sort included, is unchanged. Same session after: 18%
throughout, frames at the vsync cap.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
With a resolution scale other than 100% or FXAA on, the world renders into
its own target at the scaled size and one full-screen pass copies it
(bilinear) or anti-aliases it (FXAA 3.11) onto the screen; the hand, screen
effects and HUD stay at full resolution. Off, the world renders straight to
the screen as before. Vanilla's Mipmap Levels option now limits the mip
levels block textures sample (the default 4 is the full chain, as before).
Both are in the Advanced screen.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Video Settings opens two Horizon screens (graphics_ui): Simple has the
Performance / Balanced / Quality / Ultra presets with render distance, FOV,
max framerate, VSync and GUI scale; Advanced groups every knob under
Rendering, Quality, Display, Performance and Overlay. Each row shows a
one-line description and its frame-time cost, and applies live. Layout lives
in graphics_ui, every colour, size and widget call in graphics_style, so a
menu theme can restyle them in one place.
Horizon-only knobs (horizon-graphics.txt, never options.txt): preset, cave
culling, chunk updates and upload budget per frame (nearest sections first),
sky, fog, weather, animated textures, present mode, frame pacing, worker
threads and the performance overlay (FPS, frame/CPU/GPU times and a
frame-time graph). Defaults keep vanilla's rendering; vanilla options keep
their options.txt keys.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- Unloaded chunks and entity chunks are encoded and written by a background
region writer, in order; loads see queued records first, and synchronous
saves wait for the queue. Deleting entity chunks a region doesn't hold no
longer rewrites the region file.
- Mobs render within Entity.shouldRenderAtSqrDistance (64 x bounding-box size
x entity distance) and inside the view frustum (EntityRenderer.shouldRender);
block entities outside the frustum are skipped, beams and structure blocks
excepted.
- Minecarts and dropped items tick only in entity-ticking chunks, like mobs.
- Villager POI searches skip sections whose palette has no POI block.
- Streaming bookkeeping uses the Fx hash and checks a column's light
readiness once per frame instead of once per dirty section.
- HORIZON_SPIKES=MS prints the timed spans of slow frames; frame summaries
give the phase split of the slowest 1%.
RD 32 village flight on the 4080: 81 -> 230 fps; 1% low 5.8 -> 18.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A column that finished loading or generating after the player moved out of
range was dropped, but its block entities and saved entities were added to the
world first. The entities spawned in a column that never became resident, so
nothing saved or removed them: travelling, mobs piled up (141 to 5,028 in an
8-minute flight), and when the column loaded later its entities spawned again,
so villages near the view edge doubled their villagers. Such a column is now
dropped with everything in it. Mesh results for a column unloaded meanwhile are
discarded too, rather than uploaded with nothing left to remove them.
Decoded static sounds were kept for the whole session: every sound heard, up to
1.4 GiB of samples, and a village plays hundreds. They now stay within 64 MiB,
least recently used out, and are decoded on a thread of their own instead of the
game loop (vanilla's buffers load asynchronously too). Stopping a source that
is still waiting for its sound stops it, as alSourceStop does, so a late decode
cannot start it.
status reports mobs, block entities, resident GPU sections and the sound
engine's tables, for soak tests.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The --sprint-input scene drives the real LocalPlayer.aiStep sprint and
jump path from keyboard state. Add sprint-jump runs: sprint held or
started by double-tapping forward, jump tapped, held through landings or
pressed first, at five headings, with too little food to sprint, under a
two-block ceiling and up onto a slab. All 293,175 states match.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- Terrain sections live in one vertex arena (first-fit free list, grown on the
GPU) and draw with one multi_draw_indexed_indirect per pass instead of a
buffer bind and draw per section; per-frame scratch vectors are reused.
- The SectionOcclusionGraph search runs over dense arrays around the camera
instead of hash sets, with per-column visibility.
- Mobs outside the entity-ticking chunks (Chebyshev distance of the
simulation distance, DistanceManager.inEntityTickingRange) only check
despawning, as in ServerLevel.tick; before, every loaded mob ticked.
- World chunks use the Fx hash; sky darkening is sampled once per time and
weather instead of once per mob; creaking-heart scans compare state ids.
- Horizon-only graphics settings (horizon-graphics.txt): frame pacing without
spinning, present mode, animated textures, worker threads.
RD 32 village flight on the 4080: 3.1 -> 81 fps.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The local player's clicks now go through a port of Minecraft.startAttack,
continueAttack and handleKeybinds' attack part with MultiPlayerGameMode's
block breaking (first_person.rs): attacking the air swings (and sets the
miss time), creative breaking swings every tick through its destroy delay,
survival mining keeps swinging after a block breaks, swords cannot break
blocks in creative, and attacking an entity in creative sets the delay.
The swing is LivingEntity.SwingState (restart rules, durations under haste
and mining fatigue, forward wrap on restart); the equip height follows
FirstPersonHandsAndItems with the item swap ticker, and placing or
starting to use an item drops the hand (itemUsed). Mining fatigue scales
by Math.pow(0.3, amplifier + 1) as in 26.3.
test_first_person_parity drives the real startAttack, continueAttack,
startUseItem, MultiPlayerGameMode breaking, SwingState, the tickers and
FirstPersonHandsAndItems over 4,192 scripted click sequences.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
GameRenderer.render3dHud sets up the hand's projection at Camera's hudFov
(70, changed only by death and fluids), never the FOV option or the
sprint, speed and bow modifier. The hand pass now has its own globals
with that projection (a port of JOML setPerspective, as Projection does).
The camera harness compares the world and hand projection matrices at FOV
options 30, 70, 110 and others with vanilla's.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
horizon-ctl gains 'fly' (a constant-speed camera path that replies with the
frame summary); frame summaries add CPU time per frame (waits for the GPU,
display and limiter excluded), per-phase averages and GPU time from timestamp
queries. --offscreen renders into a texture on the discrete GPU so the 4080
can be measured under the headless compositor. Baselines: lavapipe on vscreen
and the RTX 4080 offscreen, seed 1234.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The Windows launcher was built without crt-static and imported VCRUNTIME140.dll,
which only the Visual C++ redistributable installs: players without it got a
missing-DLL dialog before anything started. The game itself links only system
DLLs (vulkan-1.dll is loaded at run time).
- launcher.yml links the C runtime on Windows.
- tools/check_imports.py reads PE import and delay-import tables (and ELF
loader, needed libraries and run paths); release.yml and launcher.yml fail
when a Windows binary imports a non-system DLL.
- release.yml builds Linux on Ubuntu 22.04: the NixOS-made Linux release asked
for a /nix/store loader, so it ran nowhere but NixOS.
- No adapter: the game says the Vulkan driver is missing and what to install,
and on Windows keeps the console open until Enter.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
PipeWire remembers mute per application name. A test run's stream, muted once
in the mixer, left "PipeWire ALSA [horizon-client]" muted for every later game,
the release played from the launcher included: no sound at all, with the device,
the stream and the mixer all working.
- The launcher starts the game as "Horizon" (media.role Game), its own entry.
- The game opens no output on a vscreen, where agents run it (HORIZON_AUDIO=1
forces it, for a null sink), so nobody needs to mute it again.
- Any device sample format is accepted (the mixer converts from f32): a plain
ALSA default reporting i16 used to fail.
- Other output devices are tried when the default does not open, and the log
names the output opened or every reason none did.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-10-13 15:02:15 +02:00
1436 changed files with 338623 additions and 16029 deletions
<img src="docs/media/banner.jpg" alt="Horizon: Minecraft Java Edition 26.3, rewritten from scratch in Rust" width="100%">
<img src="docs/media/banner.webp" alt="Horizon: Minecraft: Java Edition 26.3, rewritten in Rust and checked against vanilla by 281 differential harnesses" width="100%">
A Rust rewrite of Minecraft: Java Edition 26.3 built for one-to-one vanilla parity, checked against the game itself with differential tests.
Every subsystem matches vanilla in all 281 parity harnesses, on Linux and macOS: 100% of what the harnesses compare (see [Project status](#project-status)). The native client is playable on Linux, macOS (Apple Silicon) and Windows.
Horizon is Minecraft: Java Edition 26.3 rewritten in Rust: a native game that reads the same worlds, follows the same rules and generates the same terrain as the Java game, and is checked against it by 281 differential harnesses. You play it through a small launcher, on the files from your own copy of Minecraft. It is free, open source and unofficial.
> Not an official Minecraft product. Not approved by or associated with Mojang or Microsoft. Horizon contains no Minecraft code, assets or data: you need your own legally obtained copy of Minecraft: Java Edition 26.3.
## Download
**[Download the Horizon launcher](https://github.com/ludovic111/horizon/releases?q=launcher-v&expanded=true)** for Linux x86-64, macOS Apple Silicon or Windows x64: take the newest `launcher-v*` release (currently [launcher-v0.1.1](https://github.com/ludovic111/horizon/releases/tag/launcher-v0.1.1)); the launcher then updates itself. The launcher is the way to get Horizon: it installs and updates the game, prepares its resources from your own copy of Minecraft, and starts the client or a dedicated server.
Horizon is installed through **the Horizon launcher**. Get it from [playhorizon.dev](https://playhorizon.dev) (the Download button picks your system) or from [the launcher releases](https://github.com/ludovic111/horizon/releases?q=launcher-v&expanded=true), for Linux x86-64, macOS on Apple Silicon or Windows x64. The launcher installs and updates the game, prepares its files from your own copy of Minecraft, and updates itself.
**You need** Minecraft: Java Edition installed with the official launcher, with 26.3 played once; Python 3.10 or newer and Java 25 (a JDK, e.g. Eclipse Temurin 25) for the one-time resource preparation; and a GPU with Vulkan on Linux and Windows, or any Apple Silicon Mac.
### The launcher
**You need** Minecraft: Java Edition 26.3, played once (the official launcher, Prism, MultiMC and other launchers all work); a Java 25 JDK, for the one-time preparation; and a GPU with Vulkan on Linux and Windows, or any Apple Silicon Mac. Step by step, with macOS's first-open prompt and the first launch: [Install and first launch](https://playhorizon.dev/docs/install).
<table>
<tr>
<td width="50%"><img src="launcher/docs/play.png" alt="The launcher's Play page"></td>
<td width="50%"><img src="docs/media/launcher-play.webp" alt="The launcher's Play page: a version ready to play, its summary, Client and Server, the version picker and Play"></td>
It lists Horizon's releases and installs the build for your platform, shows each release's patch notes and the roadmap, updates Horizon and itself, and doubles as a command-line tool and an MCP server. See [the launcher's README](launcher/README.md).
## What works today
## Screenshots
The [documentation](https://playhorizon.dev/docs/) covers each of these.
Captured in the Rust client itself, with vanilla assets and no mods or shaders.
- **The game:** singleplayer with vanilla's world generation, mobs, villagers, raids, redstone, every command, the HUD and the recipe book, on saves in vanilla's format. Worlds from 1.8 to 26.2 are upgraded as vanilla does. [Status table](#project-status).
- **Horizon's own extras:** new menus (vanilla's are one setting away), graphics presets, distant terrain up to 1024 chunks, controller support, optional zoom, minimap and more, all off by default where they would change the vanilla look. See [Settings](https://playhorizon.dev/docs/settings).
- **Resource packs, shader packs and data packs:** resource packs work, Java shader packs run their full-screen effects (experimental), and data packs load functions, predicates and item modifiers.
- **Playing with others:** invite friends from the pause menu with a code, Open to LAN, and joining offline-mode servers. Horizon cannot sign in to a Microsoft account yet, so it cannot join servers that check accounts. See [Multiplayer](https://playhorizon.dev/docs/multiplayer).
- **Mods:** Horizon has its own mod system (Rust compiled to WebAssembly). Java mod jars do not run. See [Mods and data packs](https://playhorizon.dev/docs/mods).
Screenshots, captured in the Rust client itself with vanilla assets and no mods or shaders:
<table>
<tr>
<td colspan="2"><img src="docs/media/sunset.jpg" alt="Sunset over a savanna river valley and a plains village"></td>
<td colspan="2"><img src="docs/media/sunset.webp" alt="Sunset over a savanna river valley and a plains village"></td>
</tr>
<tr>
<td width="50%"><img src="docs/media/village.jpg" alt="A plains village with farms, an iron golem and animals"></td>
<td width="50%"><img src="docs/media/survival.jpg" alt="Survival at dusk: a zombie attacks near the village"></td>
<td width="50%"><img src="docs/media/menu-title.webp" alt="Horizon's title screen over a veiled snowy landscape"></td>
<td width="50%"><img src="docs/media/settings.webp" alt="The Video settings in their Simple view: a preset, render distance, field of view, framerate, VSync and GUI scale, each with what it costs"></td>
</tr>
<tr>
<td width="50%"><img src="docs/media/nether.jpg" alt="A crimson forest in the Nether with a zombified piglin"></td>
<td width="50%"><img src="docs/media/night.jpg" alt="Moonrise over a torch-lit village and river"></td>
</tr>
<tr>
<td width="50%"><img src="docs/media/underwater.jpg" alt="Salmon swimming in a river"></td>
<td width="50%" valign="middle">
Vanilla world generation ported to Rust (terrain, aquifers, carvers, features, villages and the Nether), every block model with smooth lighting and biome tints, the vanilla sky and fog, mobs, the HUD, chat and advancements: everything in these frames is drawn by Horizon.
</td>
<td width="50%"><img src="docs/media/village.webp" alt="A plains village among savanna hills, with farms, villagers and an iron golem"></td>
<td width="50%"><img src="docs/media/nether.webp" alt="A crimson forest in the Nether with a zombified piglin"></td>
</tr>
</table>
## What it does
## Host a server
Horizon reimplements the Minecraft Java client and its game logic in Rust, one subsystem at a time, and compares each piece with the vanilla game. Each `tools/test_*_parity.py` harness drives the same inputs through Horizon and through the vanilla client and server bytecode from your own copy of the game, then checks that the outputs are identical. It is for players and developers interested in game reimplementation, and for anyone who wants to see how far a behaviour-exact native rewrite can go.
The launcher's **Servers** page creates, starts, backs up and updates a Horizon server with buttons, and invites friends over Tailscale. On a VPS, `horizon-server` installs with one command and runs without Java. Guide: [Host a server](https://playhorizon.dev/docs/hosting) ([reference](docs/hosting.md)).
The [native client](docs/native-client.md) opens on a vanilla-style title screen with world creation and selection, and generates worlds with the Rust port of the vanilla generators: noise terrain, aquifers, surface rules, carvers and placed features, all checked against vanilla. It streams and lights chunks on worker threads and renders every vanilla block model with smooth lighting, biome tints, the vanilla sky, sun, moon, stars and clouds. Gameplay includes:
## Modpacks
- validated player physics, health and hunger;
- mining with vanilla tool rules and loot tables;
- placing, crafting, chests and furnaces;
- dropped items, particles and sounds;
- passive and hostile mobs with combat and vanilla AI, including guardians, shulkers, illagers, ravagers, the wither, the ender dragon fight, the breeze, the creaking, the sniffer, the happy ghast, llamas and the copper golem;
- villagers that take jobs, trade, restock, meet at the bell and sleep in beds, on a port of vanilla's brain;
- raids and pillager patrols;
- dispensers, crafters, the loom and the recipe book;
- every command, scoreboards and teams, the player list, subtitles and the HUD overlays;
- boats, minecarts, item frames, paintings and armor stands;
- shields, maps, books, fireworks, beacons, advancements and statistics;
- Anvil saves, joining vanilla servers, and opening a world to LAN.
The launcher's **Modpacks** page imports a Modrinth or CurseForge pack, shows which of its mods Horizon can carry over, and plays the ones it can as their own profile. Today that means content mods made of data and assets; mods with Java code are not ported yet. See [Modpacks](https://playhorizon.dev/docs/modpacks).
## Agent control
AI agents and scripts can play Horizon. `horizon-ctl`, shipped beside the game in every release folder, is a command line that reads the game as JSON and acts like a player; `horizon-ctl mcp` serves the same commands as an MCP server (one tool per command, screenshots returned as images).
```sh
horizon-ctl launch --world Survival --seed 5 --wait # finds the version the launcher installed
horizon-ctl state # position, health, food, target, screen...
horizon-ctl find-block birch_log # the nearest logs
horizon-ctl goto -0.5 65 -2.5 --near 2.5 # walks there along a path
horizon-ctl break -1 65 -3 # mines it; replies with the drops
horizon-ctl open-inventory && horizon-ctl click-slot 0# menus: slots, buttons by label, text fields
```
Commands block until the action is done and reply with what happened; errors are JSON with distinct exit codes, `horizon-ctl help` lists every command and `horizon-ctl COMMAND --help` explains one with examples. It talks to the game over a Unix socket on Linux and macOS and over a token-protected localhost port on Windows. See [Agent control in the native client guide](docs/native-client.md#agent-control-horizon-ctl-and-mcp).
## Build from source
For developers. Requirements: Rust 1.90 or newer, Python 3.10 or newer, JDK 25, and on Linux the ALSA development headers (`libasound2-dev` on Debian and Ubuntu). Development and the vanilla harnesses run on Linux.
For developers. Requirements: Rust 1.90 or newer, Python 3.10 or newer, JDK 25, and on Linux the ALSA and udev development headers (`libasound2-dev libudev-dev` on Debian and Ubuntu). Development and the vanilla harnesses run on Linux.
Horizon contains no game files and never downloads them. It builds its resources from **your own copy** of Minecraft: Java Edition 26.3, verified against the checksums in `research-lock.json`:
@@ -98,18 +100,29 @@ Horizon contains no game files and never downloads them. It builds its resources
cargo run --release -p horizon-prepare -- sounds# optional: sounds from your launcher install
cargo run --release -p horizon-client
```
Everything prepared lands in the gitignored `.local/`. It is derived from the game: never commit it or share it.
The client starts on the title screen. Command-line options (`--world`, `--seed`, `--creative`, `--render-distance`, `--dedicated`), controls, scripting through `horizon-ctl`, and the architecture are described in [the native client guide](docs/native-client.md). The rest of the documentation is indexed in [docs/README.md](docs/README.md), including [the parity workflow](docs/parity-workflow.md).
The client starts on the title screen. Command-line options (`--world`, `--seed`, `--creative`, `--render-distance`, `--dedicated`), controls, scripting through `horizon-ctl` and the architecture are described in [the native client guide](docs/native-client.md). The rest of the developer documentation is indexed in [docs/README.md](docs/README.md), including [the parity workflow](docs/parity-workflow.md).
## Contributing
Issues and pull requests are welcome: see [CONTRIBUTING.md](CONTRIBUTING.md) for how we work (vanilla first, small pull requests, labels), or [the contributing guide](https://playhorizon.dev/docs/contributing) on the site. Questions and ideas: [Discord](https://discord.gg/HcTeMZxe5E). Before opening a pull request, run the self-check and paste the block it prints:
```sh
git config core.hooksPath .githooks
python3 tools/pr_ready.py
```
It runs rustfmt, clippy with warnings denied, the tests of the crates you touched, the Python tests, the public-tree check, and the vanilla harnesses your change can affect. CI runs the same checks on GitHub's Linux runners; the vanilla harnesses need the game, which Horizon never downloads, so they run on a maintainer's computer holding their own copy. Never commit game files, decompiled source or prepared resources: `tools/check_public_tree.py` rejects them, and tracked Java is allowed only under the project's own `dev/horizonlab` namespace.
## Project status
The table below is the project's own audit of each subsystem. Every subsystem now matches vanilla in every harness that compares it. 100% means 100% of what the harnesses compare, not proof that nothing else differs. The remaining known differences and the final checks are listed in the roadmap.
The table below is the project's own audit of each subsystem. Every subsystem now matches vanilla in every harness that compares it. 100% means 100% of what the harnesses compare, not proof that nothing else differs. The remaining known differences are listed below.
### Java parity: 100%
@@ -138,39 +151,16 @@ them too (Rust uses Java's own float maths on each platform).
| Audio | 1 | None: the sound engine, sound events, music, situational music, ambience and subtitles match vanilla |
| Block behaviours and ticking | 1 | None: block updates, random ticks, redstone, dispensers (6,000 cases), crafters (4,000 crafts), block entity removal (containers spill, 4,000 cases), copper golem chest sorting and the sculk family (435 scenarios: shrieker warnings and warden summoning included) match vanilla |
### Roadmap
### Known issues
**Known differences.** Online-mode servers (Mojang's session servers) are not supported; skins served over https show the default skin unless already cached; pack icons show the unknown-pack texture in game; the creative hotbar keys (C/X) aren't wired to the hotbar loader; item tooltips lack attribute lines.
**Final checks.** A from-scratch audit (`tools/parity.py --all`) on Linux and on macOS, and the launch checks on the packaged Linux, macOS and Windows builds listed in [Rust rewrite status](docs/rust-parity.md#completion-criteria-and-playable-builds).
**Next.** With vanilla parity as the base:
- **Performance:** optimize the client and the server as far as they can go (rendering, chunk generation and meshing, memory, multithreaded ticking) while the differential harnesses keep gameplay identical to vanilla.
- **Modding support:** a stable API for writing mods against Horizon.
- **Agent control:** an MCP server and a CLI so AI agents and scripts can play and drive the game, growing out of today's `horizon-ctl` scripting socket.
- **Built-in recording:** capture gameplay video in a keypress, without external tools.
- **Automatic version updates:** a pipeline that ports each new Minecraft release on its own: compare the new version with the pinned one, regenerate the data, rewrite the changed parts and prove them with the same differential harnesses.
## Contributing
Issues and pull requests are welcome. Before opening a pull request, run the offline checks:
CI runs these checks on GitHub's Linux runners. The vanilla comparisons need the game, which Horizon never downloads, so they run only when started by hand on a self-hosted runner holding the maintainer's own copy. Never commit game files, decompiled source or prepared resources: `tools/check_public_tree.py` rejects them, and tracked Java is allowed only under the project's own `dev/horizonlab` namespace.
Horizon can't sign in to a Microsoft account yet, so it can't join online-mode servers (it joins offline-mode ones); skins served over https show the default skin unless already cached; pack icons show the unknown-pack texture in game; the creative hotbar keys (C/X) aren't wired to the hotbar loader; item tooltips lack attribute lines.
## Licence
[MIT](LICENSE), for the project's original work only. Dependencies keep their own licences. Minecraft is a trademark of Mojang Synergies AB; Minecraft code and assets belong to Mojang and Microsoft and are covered by the [Minecraft EULA](https://www.minecraft.net/en-us/eula). Decompiled game source and prepared resources stay on your machine and are never published.
Third-party credits and licenses: [Third-party notices](THIRD_PARTY_NOTICES.md).
## Author
Made by [Ludovic Marie](https://lu4ovic.com) ([@ludovic111](https://github.com/ludovic111)).
| 100 | Vanilla | stopped by its watchdog (a 60 s tick) | | | | | | | | 15.1 / 120 |
Both servers sent each walking bot the same number of distinct chunks over the whole run (680–690). Horizon's
"new chunks/s" inside the measured window is lower, so compare the totals, not that column.
### `explore` at full speed: unequal work
Bots fly outwards as fast as chunks arrive (`horizon-bots`' default, up to 22 blocks a second). Both servers now pull them back ("moved too quickly"), but Horizon sends
more new terrain, so this pass doesn't compare equal work. The server-machine load of Horizon's 1-bot run is
estimated from the time between the runs around it: its log was overwritten before it was read.
| Bots | Server | MSPT | P95 | TPS | RSS | CPU | Chunks/s | New chunks/s | Pulled back | Load (server / bots) |
"engine":"horizon-engine: the server build of #916 (horizon-client --no-default-features) at 519ec121, as install.sh --archive installed it",
"java":[
"openjdk 25.0.4.1 2026-08-18"
],
"settings":{
"servers":"horizon,vanilla",
"bots":"1,10,25,50",
"behaviour":"mixed",
"join_every":1.0,
"warmup":300,
"seconds":300,
"seed":12345,
"explore_speed":0.8,
"gamerule":"",
"heap":"4g",
"nice":19,
"bots_via":"ssh macmini",
"connect":"the ThinkPad's tailnet address",
"properties":{
"level-seed":"12345",
"view-distance":"10",
"simulation-distance":"10",
"online-mode":"false",
"white-list":"false",
"gamemode":"creative",
"force-gamemode":"true",
"spawn-protection":"0",
"max-players":"200",
"difficulty":"easy",
"sync-chunk-writes":"true"
}
},
"note":"server_bench.py was stopped after vanilla's 50-bot server crashed, before it wrote this file; the runs are the lines it printed, and the vanilla 50 entry is taken from that server's log.",
"runs":[
{
"server":"horizon",
"bots":1,
"behaviour":"mixed",
"start_seconds":19.73,
"mspt":{
"average":1.77,
"p50":1.34,
"p95":4.21,
"p99":13.04
},
"mspt_worst_p99":47.5,
"tps":20.0,
"rss_mb":{
"mean":686.54,
"max":686.7
},
"cpu_percent":27.92,
"chunks_per_second":9.94,
"unique_per_second":0.01,
"forgotten_per_second":9.9,
"waited_ticks_per_second":0.0,
"chunks_total":6092,
"chunks_per_bot":{
"walk":6092.0
},
"unique_chunks_per_bot":{
"walk":699.0
},
"resent":0,
"pulled_back":0,
"moved_too_quickly":0,
"moved_wrongly":0,
"bots_joined":1,
"bots_failed":[],
"tick_queries":58
},
{
"server":"vanilla",
"bots":1,
"behaviour":"mixed",
"start_seconds":20.45,
"mspt":{
"average":11.34,
"p50":10.6,
"p95":17.12,
"p99":33.22
},
"mspt_worst_p99":119.1,
"tps":20.0,
"rss_mb":{
"mean":3404.16,
"max":3416.3
},
"cpu_percent":35.93,
"chunks_per_second":9.98,
"unique_per_second":0.01,
"forgotten_per_second":9.93,
"waited_ticks_per_second":0.0,
"chunks_total":6084,
"chunks_per_bot":{
"walk":6084.0
},
"unique_chunks_per_bot":{
"walk":699.0
},
"resent":0,
"pulled_back":0,
"moved_too_quickly":0,
"moved_wrongly":0,
"bots_joined":1,
"bots_failed":[],
"tick_queries":59
},
{
"server":"horizon",
"bots":10,
"behaviour":"mixed",
"start_seconds":19.54,
"mspt":{
"average":4.74,
"p50":3.83,
"p95":10.48,
"p99":19.69
},
"mspt_worst_p99":120.1,
"tps":20.0,
"rss_mb":{
"mean":855.26,
"max":882.7
},
"cpu_percent":29.78,
"chunks_per_second":61.35,
"unique_per_second":1.79,
"forgotten_per_second":61.47,
"waited_ticks_per_second":0.0,
"chunks_total":40762,
"chunks_per_bot":{
"walk":6247.67,
"explore":1165.0,
"idle":473.0
},
"unique_chunks_per_bot":{
"walk":701.83,
"explore":1086.0,
"idle":473.0
},
"resent":0,
"pulled_back":4159,
"moved_too_quickly":0,
"moved_wrongly":0,
"bots_joined":10,
"bots_failed":[],
"tick_queries":58
},
{
"server":"vanilla",
"bots":10,
"behaviour":"mixed",
"start_seconds":19.95,
"mspt":{
"average":34.3,
"p50":32.2,
"p95":53.72,
"p99":89.08
},
"mspt_worst_p99":375.8,
"tps":20.0,
"rss_mb":{
"mean":3713.5,
"max":3754.9
},
"cpu_percent":128.07,
"chunks_per_second":61.66,
"unique_per_second":1.93,
"forgotten_per_second":61.71,
"waited_ticks_per_second":0.0,
"chunks_total":37568,
"chunks_per_bot":{
"walk":5729.0,
"explore":1124.0,
"idle":473.0
},
"unique_chunks_per_bot":{
"walk":701.17,
"explore":1055.0,
"idle":473.0
},
"resent":0,
"pulled_back":3976,
"moved_too_quickly":20,
"moved_wrongly":0,
"bots_joined":10,
"bots_failed":[],
"tick_queries":58
},
{
"server":"horizon",
"bots":25,
"behaviour":"mixed",
"start_seconds":9.61,
"mspt":{
"average":11.72,
"p50":9.88,
"p95":24.45,
"p99":43.79
},
"mspt_worst_p99":129.8,
"tps":20.0,
"rss_mb":{
"mean":947.35,
"max":1086.1
},
"cpu_percent":57.83,
"chunks_per_second":158.57,
"unique_per_second":5.78,
"forgotten_per_second":158.45,
"waited_ticks_per_second":0.0,
"chunks_total":108126,
"chunks_per_bot":{
"walk":6628.53,
"explore":1266.6,
"idle":473.0
},
"unique_chunks_per_bot":{
"walk":704.67,
"explore":1189.2,
"idle":473.0
},
"resent":0,
"pulled_back":3496,
"moved_too_quickly":0,
"moved_wrongly":0,
"bots_joined":25,
"bots_failed":[],
"tick_queries":58
},
{
"server":"vanilla",
"bots":25,
"behaviour":"mixed",
"start_seconds":37.78,
"mspt":{
"average":50.45,
"p50":46.39,
"p95":74.09,
"p99":191.21
},
"mspt_worst_p99":1388.1,
"tps":19.82,
"rss_mb":{
"mean":3866.18,
"max":4745.0
},
"cpu_percent":192.42,
"chunks_per_second":157.63,
"unique_per_second":8.25,
"forgotten_per_second":156.12,
"waited_ticks_per_second":0.0,
"chunks_total":88763,
"chunks_per_bot":{
"walk":5387.0,
"explore":1118.6,
"idle":473.0
},
"unique_chunks_per_bot":{
"walk":699.0,
"explore":1075.0,
"idle":473.0
},
"resent":2,
"pulled_back":3629,
"moved_too_quickly":296,
"moved_wrongly":0,
"bots_joined":25,
"bots_failed":[],
"tick_queries":58
},
{
"server":"horizon",
"bots":50,
"behaviour":"mixed",
"start_seconds":14.93,
"mspt":{
"average":23.82,
"p50":20.11,
"p95":50.98,
"p99":74.64
},
"mspt_worst_p99":195.9,
"tps":20.0,
"rss_mb":{
"mean":1031.41,
"max":1201.2
},
"cpu_percent":98.21,
"chunks_per_second":302.81,
"unique_per_second":10.96,
"forgotten_per_second":303.47,
"waited_ticks_per_second":0.0,
"chunks_total":212181,
"chunks_per_bot":{
"walk":6496.53,
"explore":1255.5,
"idle":473.0
},
"unique_chunks_per_bot":{
"walk":701.9,
"explore":1184.7,
"idle":473.0
},
"resent":0,
"pulled_back":7565,
"moved_too_quickly":0,
"moved_wrongly":0,
"bots_joined":50,
"bots_failed":[],
"tick_queries":58
},
{
"server":"vanilla",
"bots":50,
"behaviour":"mixed",
"start_seconds":11.16,
"error":"stopped by its watchdog 30 s after the last bot joined: \"A single server tick took 60.01 seconds (should be max 0.05)\" (Player.tick)"
| 100 | Vanilla | stopped by its watchdog | | | | | | |
Vanilla's watchdog stopped it at 50 bots (in `ClientboundLightUpdatePacketData`, copying light for a chunk
packet) and at 100 (in `Entity.move` from a move packet).
## Findings
- Horizon's RSS is 1.9 to 2.6 times smaller in every run, and it starts in 11 s against vanilla's 15 s.
Vanilla's heap is fixed at 4 GB (`-Xms4g`), so part of its RSS is that setting rather than need.
- With 1 bot Horizon's tick is 3.3 times shorter walking and 4.5 times shorter exploring at equal work;
at 10 bots it is still shorter in both.
- At equal exploring work Horizon's tick is about half of vanilla's at 10 and 25 bots.
- When the bots only fly around spawn (`walk`), vanilla's tick hardly grows with players (16 to 20 ms from
1 to 50 bots) while Horizon's grows by about 0.5 ms per bot, so vanilla is ahead from about 20 bots: 16.0
against 20.7 ms at 25 and 20.5 against 31.4 ms at 50. Both keep 20 TPS up to 50.
- With 100 bots vanilla's watchdog stops it in every pass, and Horizon keeps running at 9 to 11 TPS.
### Where Horizon's walk tick goes at 25 bots
Profiled on the ThinkPad with `perf` (the server thread, 20 s, self time):
| Share | What | Whose |
|---:|---|---|
| 18 % | `Streaming::update`: every tick it walks all loaded columns and proto chunks and tests each against every player's area (unloading, dropping protos, finishing and decoration candidates), so the cost grows with players times chunks | streaming / engine tick |
| 20 % | `PalettedContainer::get` and `World::block`, mostly from block entity ticks: on the Mac profile `tick_hoppers` looks up the block under every block entity in the loaded world each tick and compares names as strings, and spawners rebuild their display id | engine tick |
| 8 % | `light_data` when a chunk is sent: the light layers were built twice per chunk; a follow-up commit on #623 builds them once and keeps each section's packed layers | server (this work) |
| 7 % | encoding chunk sections and heightmaps for the chunks each bot newly sees | server |
| 4 % | `Hub::tick` (moves, tracking) | server |
| 2 % | the moved-wrongly collision check | server |
The first two grow with the number of players but aren't specific to the server: they are the
engine's tick and streaming, reported to perf-server-2. Before #623 and #649, encoding every chunk section
once per player and waking each connection's writer thread once per packet cost more than everything above
(at 50 walking bots the tick was 281 ms; see the earlier runs).
## Earlier runs (same machine and method)
-`run-1-before-chunk-cache.json`: before #623, with vanilla's locator bar on. Horizon's walk tick was
27.5, 40.7, 281 and 467 ms at 10, 25, 50 and 100 bots, and vanilla's 33.2 ms at 50 (the locator bar).
-`run-2-chunk-cache.json`: Horizon with #623 only: 13.2, 23.5, 31.4 and 84.4 ms walking at 10 to 100
bots.
## Files
-`horizon.json`, `vanilla.json`: the walk and explore passes (vanilla with `locator_bar=false`).
-`explore-equal-rate.json`: the explore pass at 0.8 blocks a second, both servers.
-`run-1-before-chunk-cache.json`, `run-2-chunk-cache.json`: the earlier runs.
To run it again (bots on another machine reached at `--connect`):
"experiment_note":"Experimental projection variant (reused sounds, Arc-keyed string pools and direct flags); reverted after mixed CPU/RSS evidence. Current in this file is not the final PR version."
"current":"release example built at cf837118 + current uncommitted delta in PR #607 worktree",
"historical":"release example built at exact cf837118 commit, with same Cargo release profile and shared dependency target"
},
"mode_commits":{
"baseline":"independent loader algorithm at e422bd13, executed in current binary",
"pr607":"shared early construction algorithm at a01d47f2, executed in current binary",
"cf837118":"exact historical binary at cf837118, deferred mode",
"current":"current PR worktree, deferred mode with interned strings"
},
"limitations":"Loader-only (Assets + BlockFacts + Behaviours), no full client startup or first-frame timing. Same release profile but different compilation contexts for historical binary; hardware activity and cache warmth vary. CPU and elapsed differences between cf837118 and current are mixed and may be noise. Peak RSS is whole-process Linux ru_maxrss."
"measurement":"same optimized binary, isolated processes in alternating forward/reverse four-mode runs",
"limitations":"Includes Assets::load, BlockFacts, Behaviours; excludes complete client startup, pack/mod root resolution, graphics and first frame. Different modes emulate the original independent, PR #607 at a01d47f2, paired and deferred startup sequences in the same release build. Machine was not idle. Elapsed times show high variance; no statistically established gain between PR, paired and deferred modes."
"experiment_note":"Exploratory variants (Arc-key pool, reused sounds, combined proposal) alongside cd61e11b; none gave a robust consistent improvement and all were reverted."
Loaded 100 of 1436 files, more files were not shown because too many files have changed in this diff.
Show more
Reference in new issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Ludovic Marie