Entry 258

The placement groups move into the build

2026-09-27buildtooling

Filescrates/piney-gen/src/placementcrates/piney-data/src/area/mod.rscrates/piney-data/src/statics/mod.rscrates/piney-data/src/sound/mod.rscrates/piney-data/src/field/mod.rscrates/piney-data/src/dungeon/mod.rscrates/piney-audio/src/setbl.rstools/portdata.pyplans/build-data.md

Step 3 of plans/build-data.md is done. The last tables whose values were in the repository now come from the build:

• the area generator's words and event areas;
• the towns' statics;
• the sound tables and setbl.cpp's rows;
• the fields;
• the dungeons.

Before, the Python generators wrote these as Rust into piney-data, one file per volume, as 2.68 MB of generated Rust:

Now each is a file per volume, PINEY/TABLES/<group>.bin, 385-398 KB per volume in all. The largest is dungeon.bin: 289,416 bytes for INF, 301,966 for MUT and 301,958 for OUT and QUA. pack::DATA_VERSION is 8.

The Rust generators

The five generators are now Rust, in piney_gen::placement: area, statics, sound, field and dungeon. Each is a function from a volume to its file's bytes, listed in placement::GROUPS. piney-gen gen writes their output to work/data/<volume>/TABLES/, and piney-build's port_files adds it to each disc. --group <name> runs one of them; field and dungeon take under 0.3 s each for all four volumes.

These Python tools were deleted, per the owner's rule that a Python tool goes once Rust replaces it:

• tools/area_tables.py, statics.py (with test_statics.py), sound_tables.py, field_tables.py and dungeon_tables.py;
• the test_generated_file_is_current checks in test_field_rs.py and test_dungeon_rs.py.

The ports read the code exactly as the Python did. They find the same things:

• the li/branch pairs;
• the registers that move between compilers (this in $s0 or $s2, the object percentage in $f20 or $f1, the hill counters);
• the jump tables on the dungeon type;
• the fog blocks of the DUNGEON constructor.

Two parts of the Python ran game code in tools/eemu.py. Both now run in piney-eemu, with the machine sinit.rs already set up (every loaded segment, the overlay and the C library stand-ins):

• FIELD_MESH::SetMESH2 and FIELD::SetSmallMESH (INF gcmn.prg:0x005af780, 0x005ade80), over a FIELD whose every height and colour names its own cell;
• WORLD_MAN::SetDungeonTexClut for servers 0-4 and field types 0-11.

Some values are models or measurements, not reads. The Python generators held them as constants, and the Rust ones carry the same constants:

• WORLD::Init's draw counts;
• tools/dungeon.py's model of SetDungeonTypeFromField;
• the expected WORLD::Generate constants, which the generator checks the code against.

tools/test_field.py and tools/test_dungeon.py still check those models against the game.

The file shapes

Each module's types stay hand-written. Each gets a Load impl in field order, in the store's format from 256. Where the game's tables share an array, the file holds it once and the users name it by index:

• the fields' FOBJECT_INFO_TABLE rows (a lake field's LakeObjTABLE entry is its key rows);
• the dungeons' ROOM_INFO tables;
• the dungeons' dungeonFog* tables;
• the story dungeons' ROOMDATA and GIMMICKDATA arrays.

tables_of(volume) keeps its reference in a OnceLock per volume. INF became a LazyLock<&'static Tables>. Every user of field::INF, dungeon::INF and sound::INF compiled unchanged through deref coercion, except one in se3d.rs, which took a *.

setbl is made only for Infection and Mutation, since Outbreak and Quarantine have no spc0SeData. A group maker returns None for a volume without the table, and the build skips it. piney-audio reads Infection's rows, as setbl.cpp's port always did. So a build without Infection has no setbl for the port to read. Nothing else in the port depends on which discs are in a build.

Python that read the generated Rust

tools/iopemu.py's wave_bank and tools/test_sound3d_rs.py's Setbl had parsed the generated Rust. They now read the port's files through tools/portdata.py: sound(volume) and setbl(volume) decode work/data/<volume>/TABLES/sound.bin and setbl.bin, the same format the Rust Load impls read. tools/sound_ee.py's FIELD_VOICE, which it had imported from sound_tables.py, is now inline.

Checked

Each move was checked the same way. A dump_<group> example prints {:#?} of every volume's tables. It ran in a worktree at 70f4632, which still has the compiled-in values, and again after the move. The two outputs were compared byte for byte:

area, statics, sound and setbl were also identical in their commits (e39f305, 20cf8fa, 20d36af).

The whole workspace's tests pass (694), reading the tables from work/data, and clippy is clean.