* When building with CMake, dynamic model lists are now generated
* Factored common code out of generate_*_list.js files to deduplicate
* Better separated generated model lists from main game code
`dynamic_asset_data.c` and `dynamic_animated_asset_data.c`
previously #included the corresponding generated headers, and
`dynamic_asset_loader.c` declared externs for the contents.
This messiness was likely done so the generated code would be built
automatically (the Makefile globs all C files under `src/`).
Now, model list data is output to C source files which are built
explicitly. This, with some refactoring, allows the previously
mentioned source files and externs to be removed.
This is a bit hacky in the Makefile but will be automatic under
CMake by using target properties.
* Reorganized some files under `tools/`
This value is set when compiling anyway, but this change makes the value
match for consistency.
Now that the default is the actual intended value, it doesn't need to be
overridden in the Makefile, but since scene scale needs to be passed to
skeletool regardless I'm going to keep the preprocessor define too so
they don't get out of sync.
Libultra include directory and library path are now found dynamically and
saved in a CMake cache variable so they can be overridden by users.
The game now links to libultra. Most linker errors are now due to lack of
integration with generated code dependencies.
With CMake, the game version is stored in a generated header file so
dependencies on it are kept track of automatically.
The game now compiles when using CMake but does not link.
Generated code dependencies are not hooked up yet and corresponding
targets need to be built manually.
Incidental changes along the way:
* Move `export_level.lua` to `tools/level_scripts/`
* Small reorganizations/fixes to existing `CMakeLists.txt` files
* Support passing absolute paths to level export
Some parts need to be cleaned up/removed/streamlined after the CMake
build can build the game on its own and compatibility with what the
old makefile does is no longer needed.
It is still necessary to special case extracting the raw WAV from the
video, but now conversion is handled using a .sox file just like the
other sounds.
This simplifies sound processing for the CMake transition.
To better organize generated files, portal_pak_modified/ will be used
for files generated by modifying portal_pak_dir/ contents.
Build/ will be for conversion to a final form needed for compilation,
without content modification.
Useful for room shapes which cannot be handled by standard visibility
checks.
Normally this can be addressed with additional rooms and doorways, but
sometimes a room is required to be a complex shape (for example, to
accomodate animated static geometry).
These instances should be few and far between, so leaving this as an
opt-in feature that can be used during level authoring.
* Doorway normals now always point toward room A. This was intended
and code was written for it, but typos prevented it from working.
* Consistent doorway directions allowed resurrection of previous
collisionSceneRaycastDoorways() implementation
Standard culling is quite coarse, for performance reasons:
* Only uses axis-aligned bounding boxes
* One visible object in a BVH node will make all others in the same node visible
These properties allow the bounding boxes of rotated objects to clip through
the back of portals to be considered visible, causing improper rendering.
This commit introduces a new optional argument for `@static` level
objects: `precise_culling`. Static geometry marked for precise culling
will be culled based on its exact (i.e., rotated) bounding box, in the
same way dynamic/animated geometry is.
Precise culling is more expensive than standard culling so it is opt-in.
It is intended only for the few situations that can hit the edge cases
described above.
The player's bounding box was not reinitialized on save file load,
causing the collision detection area to include the space between the
level's starting point and their current position.
In large levels (i.e., chamber 15) when far away from the start, this
caused the internal collider limit to be exceeded resulting in a buffer
overflow and crash.
