feat: add bounded Quantum cabinet extraction

This commit is contained in:
2026-07-22 19:23:00 +02:00
parent 531612829d
commit a7b1cd7d16
20 changed files with 1099 additions and 66 deletions

View File

@@ -11,7 +11,7 @@
| CAB LZX `0x1303` | Supported extraction | Byte-for-byte rust-cab MIT unit fixture, two text entries |
| CAB without compression | Supported extraction | Byte-for-byte rust-cab MIT unit fixture |
| CAB MSZIP | Supported extraction | rust-cab single-block fixture plus a generated two-block dictionary-chain fixture |
| CAB Quantum | Unsupported | Metadata is recognized; extraction fails with `cab-compression-unsupported` |
| CAB Quantum | Supported extraction | Pinned compcol qtm.txt payload, generated one-folder CAB, and two-frame state/sentinel tests |
| Multi-cabinet CAB / spanning files | Unsupported | Header flags and spanning folder indexes are rejected |
| `.onetoc2` package entries | Listed only | Files are identified, but notebook hierarchy and section order are not interpreted |
| Standalone `.onetoc2` | Unsupported in the app | The picker accepts only `.one` and `.onepkg` |
@@ -30,6 +30,14 @@ and carries at most 32 KiB of history between blocks in one folder. The MSZIP
tests establish CAB extraction behavior; no redistributable MSZIP-compressed
OneNote export is currently present, so they do not establish producer-specific
`.onepkg` compatibility.
Quantum validates level/window metadata, bounds its window to 2 MiB by default,
requires full 32 KiB intermediate frames, preserves its arithmetic models and
LZ window across frames, and injects exactly one `0xFF` CAB block sentinel.
Trailer padding is limited to the documented 0..4 zero bytes; match, frame,
output, operation, and cancellation bounds are enforced. Its tests likewise
establish CAB extraction behavior only: no redistributable Quantum-compressed
OneNote export is currently present.
Trailing CAB data is ignored with a warning. Reserved CAB areas are bounded and
skipped. CAB names are decoded as strict UTF-8; unmarked valid UTF-8 receives a
warning, while invalid non-ASCII legacy-code-page names are rejected rather