The compression codec

Technical appendix · Home · Français

Used by .HUF streams and by G3.CON record 0x27 subs 9 and 10. Those two are the only .CON payloads that pass try_decompress. All of this is implemented by maup_parse.py.

Routine	EA	Role
`Lz_DecompressFar`	`0x19D92`	entry: AX = length low, DX = length high; clears state, sets limit
`Lz_Decompress`	`0x19C6A`	main loop, owns the ring window
`Lz_ReadToken`	`0x19BD6`	tree walk to a symbol
`Lz_ReadDistance`	`0x19C14`	distance decode
`Lz_ReadByte` / `Lz_ReadBit`	`0x19855` / `0x197F8`	input
`Lz_BitInit`	`0x198AF`	tree and bit-reader init
`Lz_UpdateTree` / `Lz_RebuildTree`	`0x19ACE` / `0x19958`	adaptive update

Framing

A stream starts with a u32 output length, skipped at 0x19C70. The tree is built at runtime, so there is no static table to find. Decoding stops when the produced count reaches the length.

Bits

MSB-first through a 16-bit register (DS:303E, count at DS:3040).

Tokens and matches (verified)

tree: 314 symbols; token < 0x100 is a literal
match length = token - 0xFD (add ax,0FF03h at 0x19CF6)
c = ReadByte()
high = LzDist_ExtraBits[c] << 6        ; DS:2E3E
n = LzDist_BaseBits[c] - 2             ; DS:2F3E
acc = c; repeat n times: acc = acc*2 + ReadBit()
distance = high | (acc & 0x3F)

Window

A 4096-byte ring at DS:6754, with its first 0xFC4 bytes filled with spaces and writing starting at 0xFC4. Match source = (pos - distance - 1) & 0xFFF.

The `0x400` counter is not a chunk format

0x19D4F-0x19D69 only raises wLzProgressLimit by 0x400 when the output passes it. Nothing reads that for decoding: it doesn't change input position, bit state, tree or window. maup_parse.py decodes every stream correctly without chunk handling.

Verification

selftest decodes all 29 MUS and 28 SND streams; header lengths agree and tail padding is at most 7 bits.

Previous: File formats, scenes and palettes · Next: Video · Technical appendix · Home

Hbtvl