Fix level extents read as bytes, and extract the level geometry

A defect in the Phase 1 extractor, found by reading the data extent's contents:
level extents are declared in sector counts, the generic G/ pair table uses byte
counts, and level containers were being routed through the generic path. So
LEVEL0.WAD's data extent -- 6,484 sectors -- was taken as 6,484 bytes. Every
test still passed, because a truncated file is still a file.

The extent's own header gave it away: it declares an uncompressed total of
13,277,920 bytes, and 13,277,920 / 6,484 is 2047.7. A ratio landing on 2048 is
a unit error, not a compression ratio. Across 27 levels this had been discarding
roughly 300 MB of level content.

Level containers now go through the typed LevelHeader instead. Level 0 yields:

  data.bin         13,279,232   the data extent
  core_data.wad     9,738,678   the geometry, compressed
  core_data.bin    16,969,296   the geometry, decompressed
  core_index.bin       31,600   the geometry directory
  gameplay.bin       659,456
  sound_bank.bin   1,503,232

core_data is a WAD LZ77 stream -- the format already implemented for the game --
and it decompresses cleanly. The first real level content this project has
recovered.

New: LevelDataHeader parses the data extent's own header, a table of byte ranges
naming the level overlay executable, the geometry directory, the GS RAM image,
the HUD banks and the geometry. Note this is a third meaning for a "size" field
on the same disc, after sector counts in the level header and byte counts in the
generic table; all three are documented together in level-data.md so the next
parser does not rediscover this the hard way.

The geometry directory's offsets were checked against the geometry rather than
taken on trust: all twelve land inside the 16.9 MB of decompressed core_data,
which is what confirms the header is being read correctly. The class tables
themselves do not yet decode under the documented layout, and that is recorded
as open work in plan/level-assets.md rather than papered over.

75 unit tests, 9 integration tests against the real disc. Hash table regenerated
at 5,159 files and 5,671,915,799 bytes -- hashes only, no content. Extraction
now writes to ~/Documents/gc-extracted rather than /tmp, which is 16 GB and did
not have room for a 5.3 GB output.
This commit is contained in:
2026-10-09 16:28:26 +02:00
parent a9e43e39c1
commit 16c04db559
9 changed files with 868 additions and 147 deletions
+108
View File
@@ -0,0 +1,108 @@
# The level data extent
Target: `SCUS_972.68` v1.01. Implementation: `crates/gc-formats/src/level_data.rs`.
The `data` extent of each `LEVEL<n>.WAD` is the level's own container. It holds
the level's executable, the geometry directory, the Graphics Synthesizer RAM
image, the HUD banks and the geometry itself, each at a byte offset declared by
a table at the start of the extent.
## Header
```text
offset size field
0x00 8 overlay the level's executable
0x08 8 core_index the geometry directory
0x10 8 gs_ram Graphics Synthesizer RAM image
0x18 8 hud_header
0x20 40 hud_banks[5]
0x48 8 core_data the compressed geometry
0x50 8 transition_textures
```
Every field is `{offset, size}` in **bytes** relative to the start of the
extent. The table is 0x58 bytes and the first extent starts at 0x80, leaving 40
bytes unused.
Level 0, measured:
| Extent | Offset | Size |
|---|---:|---:|
| overlay | 128 | 2,751,268 |
| core_index | 2,751,424 | 31,600 |
| gs_ram | 2,783,040 | 518,656 |
| hud_header | 3,301,696 | 4,060 |
| hud_banks[0..4] | 3,305,792 … | 44,935 / 138,675 / 16 / 16 / 34,286 |
| core_data | 3,523,904 | 9,738,678 |
| transition_textures | 13,262,592 | 15,328 |
Extents are ordered and non-overlapping but not gap-free: 28 bytes separate the
end of the overlay from the start of `core_index`. The test asserts ordering
rather than tiling for that reason.
## A third convention for sizes
Three different meanings for a "size" field appear on this disc. All three are
byte offsets; the size unit differs:
| Where | Size unit |
|---|---|
| `LEVEL<n>.WAD` named extents | **sectors** |
| `G/*.WAD` generic pair table | **bytes** |
| `data` extent header | **bytes** |
The first of those was a real bug: the extractor read level extents through the
generic path, taking a 6,484-sector extent as 6,484 bytes and losing 13 MB of
every level. See `../plan/level-assets.md`.
## `core_data`: the geometry
The largest extent, and a WAD LZ77 stream — the format in
[`wad.md`](wad.md), with a 16-byte header whose declared size equals the whole
extent, i.e. the stream spans it exactly.
Level 0 decompresses from 9,738,678 to **16,969,296 bytes**, a 1.74× ratio. It
contains the four geometry subsystems (`tfrag` terrain, `tie` environment
instances, `shrub` foliage, `moby` actors), sky shells, collision and
particles.
## `core_index`: the geometry directory
A 31,600-byte structure at the head of which sits a header of offsets into
`core_data`. On level 0:
| Field | Value | Lands inside core_data |
|---|---:|---|
| gs_ram | 404, count 18,704 | yes |
| occlusion | 2,117,888 | yes |
| sky | 2,192,576 | yes |
| collision | 2,615,232 | yes |
| moby_classes | 238 | yes |
| tie_classes | 74 | yes |
| shrub_classes | 7 | yes |
| tfrag_textures | 93 | yes |
| moby_textures | 168 | yes |
| textures_base | 3,518,848 | yes |
| part_bank | 8,141,696 | yes |
| fx_bank | 8,332,160 | yes |
Every offset falls inside the 16.9 MB `core_data`, which is the evidence that
the header is being read correctly.
`sun` and `collision` at their offsets begin with plausible structure — the
collision block opens `40 00 00 00 50 c8 0d 00`, a plausible count followed by a
size.
**Not yet resolved:** the `ArrayRange` fields (`moby_classes` and friends) do
not decode to a usable class table under the documented layout. Offsets and
counts read as plausible numbers but the entries they point at are not valid
`{offset, o_class}` records. The GC core header layout differs from the
documented R&C1 one in a way not yet determined. Recorded in
`../plan/level-assets.md`; this is the next thing to resolve.
## Provenance
The structure matches `GcUyaLevelDataHeader` in the Wrench project
(<https://github.com/chaoticgd/wrench>, GPLv3). **No Wrench code is copied** —
see `PROVENANCE.md`. Everything here is written from the specification and
verified against all 27 levels on this disc.