mirror of
https://github.com/libretro/virtualjaguar-libretro.git
synced 2026-11-09 11:51:42 -10:00
Joe MattielloandClaude Opus 5.5
1dec1cb2fd
* test(app): no_threaded_limits keeps every feature on while the emulation thread runs (#967 G1a)
Fails when a threaded-only gate, its user-facing text or its limit test comes back, when the GDB stub is locked for any
reason but hardcore, when a file starts asking "am I threaded?" beyond its allowlisted count, or when tsan.supp grows.
Verified red on the pre-3a tree (abc9054e) and green on develop.
Refs #1079
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(app): Studio gate for the emulation thread with every feature on (#967 G1b)
--show-window tools opens all eight Tools windows, --perf-halt does F9's break and continue at fixed UI iterations, and
emu_thread_gate_check.sh asserts no owner violation, a window presenting at display rate while halted by F9 and by gdb
(ports from the OS at run time), a connected Neuro client answering presses, all with --emu-thread on.
Refs #1079
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(app): no_threaded_limits counts bare threaded() and fails on under-counts (#967 G1a fixup)
Check 5 now also counts the bare threaded() calls inside App and CoreHost, counts occurrences instead of lines, and fails on
an allowlist that is too loose; the over-broad "Off while" text pattern is anchored to the old Settings string; commented-out
add_test lines and ==/!= comparisons no longer confuse checks 3 and 4.
Refs #1079
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(app): build channel and the emulation-thread default rule (#967 3d)
VJ_BUILD_CHANNEL (release | nightly | dev, default dev) sets VJ_EMU_THREAD_DEFAULT (1 for nightly), next to VJ_TV_DEFAULT.
The default applies only while ui.emu_thread is unwritten: an explicit Off stays Off in nightlies and an explicit On stays
On in releases (emu_thread_default_unit). Nothing is built with the nightly channel yet.
Refs #1079
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(app): show the build channel and the emulation-thread default (#967 3d)
Settings > Interface says which way the default goes in this build, About names the channel, and the start-up log reports
channel, default, the stored choice and this run. The smokes derive their expected output from the configured channel, so
a nightly configure checks the On default and an explicit Off still winning; the inline hashes are untouched.
Refs #1079
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(app): the Studio gate wakes the display before each run and stays quiet after a kill (#967 3d)
A window created while the Studio display sleeps blocks forever in the GL swap, threaded or inline (soak-results.md), so the
G1c attempt-1 "hang" was the display. Before each of runs A, B and C the gate now wakes it (caffeinate -u -t 3), holds it
awake for the run (caffeinate -u -d -i, ended after the run) and re-checks system_profiler. Still asleep: the run is skipped and
reported "display asleep, run skipped" as an environment failure (exit 3 when nothing else failed), distinct from a hang.
A killed or skipped run no longer prints python tracebacks or "[: : integer expression expected": the CSV helpers answer
"no data" for a missing file, empty reads become 0, the per-run asserts that need logs are skipped after a watchdog kill,
and the shell's "Terminated" job notices are silenced.
Refs #1079
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(app): testServeDebugLane waits for the lane service to finish, not for a fixed sleep (#967 3b flake)
emu_thread_unit failed on GitHub's hosted macOS runner (run 39929091926) at "it waited about maxWait despite the frame post
(0.0 ms)". The test waited for the fake core to take its serveNow count, then slept 10 ms and read `waited`, which the
emulation thread stores only after the 4 ms lane wait returns; a thread descheduled in between left `waited` at its initial
0. Not the stale debug wake: nothing sets wakeDbg_ before that first wait (reopen() clears it, the frame post only sets wake_),
and a stale wake would make a wait of microseconds, not leave the value unset.
The test now counts finished rounds (stored after `served` and `waited`) and waits for the count, with a 5 s bound, and the
fake core sleeps 30 ms before serving so the old reading order fails every time. The assertion is kept; its upper bound is 2 s
instead of 0.5 s for a slow VM.
Refs #1079
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* ci(app): nightly builds default the emulation thread on (#967 3d)
app-package.yml derives CHANNEL from release_tag (nightly; v* -> release; else dev) beside TV_DEFAULT and passes
-DVJ_BUILD_CHANNEL to the macOS, Windows and Linux configures; the flatpak job rewrites the committed manifest's
-DVJ_BUILD_CHANNEL=release to the lane's channel. app.yml configures one Linux row with the nightly channel so the On
default is built and tested on every PR. Docs updated, including the design text the G1d review found out of date
(closed windows and Neuro's watch copier, frame advance while halted, debugCall waits, the unstated bounds).
Refs #1079
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>