Expand description
Block codecs with UnityPy’s semantics (helpers/CompressionHelper.py:238-244).
- None returns the stored bytes as they are, whatever blocks-info says.
- LZMA (
:17-44) is a raw LZMA1 decoder over the 5 property bytes and the rest, run to the end of the input: the declared size is ignored, a truncated stream yields what decoded, an over-long one yields all of it. Output is capped atcrate::limits::Limits::max_block_output(aPARITY.mdsafety divergence). Fewer than 5 input bytes raise (struct.unpack("<BI", ...)). liblzma static, asdocs/spikes/lzma.mddecided. - LZ4/LZ4HC (
:83-94) islz4.block.decompress(data, uncompressed_size): the declared size is the maximum output. A shorter output is accepted; a longer one raises (python-lz4:LZ4BlockError), and so does an empty input, whatever the declared size (LZ4_decompress_safereturns -1). - LZHAM raises (
:234-235), and so does an unknown type (files/BundleFile.py:515-521;CompressionFlagsis anIntFlag, so the map lookup is what raises).
A block’s decoded length can differ from blocks-info; the bundle always lays entries out by the actual lengths, as UnityPy’s join does (plan section 3.6), so a codec returns what it decoded and never pads or cuts.
LZ4 acceptance is liblz4’s. UnityPy’s python-lz4 4.4.4 runs liblz4
1.9.4’s LZ4_decompress_safe, which refuses some blocks lz4_flex decodes:
its end-of-block conditions (a literal run that ends within the last 12
bytes of the declared size or the last 8 bytes of input must be the last
sequence and end exactly at the end of input; a match must end at least 5
bytes before the declared size; the length-extension guards). Lz4
is a port of liblz4’s decoder (lz4.rs), so it refuses what liblz4
refuses and decodes what it decodes, and the oracle fixture tests/fixtures/scaffold/lz4_blocks.json (python-lz4’s verdict and
output for a grid of blocks around each boundary) is the acceptance test.
Output. A codec writes into a DecodeBuf, which checks the block’s
cap and charges the input’s meter before it grows (plan section 3.6), or
names a range of its input (the stored codec: Decoded::Input), which
the caller slices. It cannot hand back bytes of its own: a buffer a
registered codec allocated (a Python decompressor’s bytes) would be
input-sized memory no meter charged (round 5), so its output goes through
out like any other. Every block is decoded through decode_block,
which starts the block’s segment and checks what the codec returned. A
codec is caller code when a user registers it: the core calls it with no
lock held.
Owner: package C. The registry type and decode_block are implemented
by the scaffold.
Modules§
- encode
- Bundle packers following
helpers/CompressionHelper.py:158-231. LZ4 and LZ4HC both use pristine liblz4 1.9.4 HC at level 9. - unitycn
- UnityCN’s AES vectors and in-place LZ4-token walk.
Structs§
- Codecs
- Codecs by compression type number (0-63).
- Lz4
- Compression types 2 and 3: an LZ4 block,
uncompressedas the maximum. - Lzma
- Compression type 1: liblzma’s raw LZMA1 decoder, to the end of the input.
- Stored
- Compression type 0: the stored bytes.
Enums§
- Codec
- A block’s compression,
flags & 0x3F. - Decoded
- Where a codec put a block’s output (
BlockCodec::decode).
Traits§
- Block
Codec - Decodes one block.
- Block
Decryptor - Decrypts one block in place before it is decoded (UnityCN: P3). UnityPy
decrypts a block whose flags carry 0x100 and whose codec is not
None, of a bundle whose header’s encryption flag built a decryptor (files/BundleFile.py:123-124,files/BundleFile.py:512-513); a bundle whose flag is clear is never decrypted, whatever the context holds.
Functions§
- decode_
block - Decodes one block of a bundle with
codec(the codec ofwhich), as every caller does (package C’s bundle; implemented by the scaffold).