Files

30 lines
1.6 KiB
Markdown

# GAM level entities
Derived from all 35 original files. Implemented reader and lossless encoder:
`tools/levels/gam.py`; corpus regression tests: `tests/test_levels.py`.
All multibyte integers/floats below use big endian byte order.
```
file: char[4] level_code, u32 entity_count, entity[entity_count]
entity: char[4] class_code, u32 property_count, property[property_count], byte[4] trailer
property: u8 name_length, char[name_length] name, u32 type, u32 payload_length,
[u8 string_length if type == 1], byte[payload_length] payload
```
The trailer is opaque: its purpose is unverified. The string prefix must equal
payload_length but is excluded from payload_length. Type 1 is Latin-1-compatible
text (no terminator); type 3 is float32; type 6 is signed int32. Type 2 has four
opaque bytes in this corpus (RotateToDest) and type 4 remains opaque. Retain raw
payload bytes and property order to preserve float bit patterns and unknown types.
No compression or padding occurs between the fields. There is no file footer.
Reader rejects truncation, impossible counts, duplicate properties, scalar length
mismatches, string length mismatches, non-finite known scalars and trailing bytes.
Original corpus passes these constraints; they are not proof of every conceivable
GAM version. Never read a Windows binary using host-endian assumptions.
Examples: `VR01.gam` has header `56 52 30 31 00 00 00 10`; its first object at 8
is `3JIM` with 19 properties. `level1b.gam` starts `LV1B`, 129 objects, with the
first `3JIM` at 8. All 3,299 object trailers are retained without reinterpretation.