Module container 

Container awareness for ZIP-based formats (DOCX/XLSX/PPTX/EPUB/JAR/ODT/plain ZIP).

The container is parsed with a small, strict ZIP framing parser that keeps the exact bytes of every local header, central-directory record, data descriptor and the end-of-central-directory block. Member payloads are handled by kind:

• stored — the payload is the member’s plaintext;
• deflate — the raw deflate stream is inverted with preflate (Microsoft, Apache-2.0), which yields the plaintext plus a small “corrections” blob that recreates the original stream bit-exact whatever encoder produced it (zlib, zlib-ng, miniz, .NET, …);
• raw — anything else (unknown methods, encrypted members, streams preflate cannot model): the compressed bytes themselves are kept.

The plaintext/raw payloads are what T-saur chunks, deduplicates and compresses. Rebuilding is deterministic and is verified against the original hash at pack time; if the framing has gaps, ZIP64 records or anything else we do not model, the whole file is stored raw instead.

Constants

KIND_DEFLATE

KIND_RAW

KIND_STORED

LEPTON_MAX_INPUT

JPEG streams above this size are kept literal (bounded memory/time).

PREFLATE_PLAINTEXT_LIMIT

Deflate streams above this plaintext size are kept raw (bounded memory, anti-bomb).

SEG_JPEG

JPEG payload recompressed losslessly with Lepton (the chunked payload is the Lepton stream).

SEG_LITERAL

SEG_ZLIB

STREAM_EXTENSIONS

ZIP_EXTENSIONS

Functions

explode

Parse the ZIP framing and invert the member payloads. Returns None for anything we do not model exactly (the caller then stores the file raw).

explode_streams

Split a PDF (or similar) into literal segments, inverted zlib streams and, when jpeg is set and the build has Lepton, recompressed JPEG streams. A file that is itself a JPEG becomes a single JPEG segment (or, with jpeg off, no recipe at all: it is stored as it is, which every build can read).

is_stream_container

is_zip_like

jpeg_to_lepton

Lepton-encode a JPEG; the library verifies the round trip itself. Single-threaded so the output is deterministic whatever the machine.

kinds

Statistics helper for reports.

lepton_to_jpeg

rebuild

Rebuild the container from the recipe and the member payloads (same order as recipe.members).

rebuild_streams

try_bitexact

Explode and prove that the rebuild reproduces the exact input bytes.

try_bitexact_streams