Skip to content

fix: ignore dead-end ancillary headers in short free-format streams - #58

Merged
choskeli merged 1 commit into
mainfrom
fix/short-free-format-dead-end
Oct 28, 2026
Merged

choskeli merged 1 commit into
mainfrom
fix/short-free-format-dead-end

Conversation

@choskeli

@choskeli choskeli commented Oct 28, 2026 •

Copy link
Copy Markdown
Contributor

A complete 864-byte free-format frame at 48 kHz was rejected when an ancillary header at byte 100 could not form a complete alternative. Reconsider EOF-closed one-/two-frame layouts only when ordinary root selection found no candidate and the probe contains one free-format configuration. Alternate headers contribute veto evidence only through complete walks whose every frame passes the existing necessary prefix, side-information and main-data capacity checks.

This also fixes the independently reviewed complete-but-impossible alternative: headers at 0/100/432 with byte 442 set to 0x10 must retain the valid singleton because the 432-byte interpretation has invalid side fields. The analogous two-frame stream with a supported footer-bearing ID3 tag is covered. Exact-capacity and genuine crossing alternatives remain protected.

Current head: d383a0b, directly based on merged #57 at 2e4b81c. Only the reviewed ten-file short-frame delta was replayed onto actual main. Parser, tests, fuzz inputs and all 32 proof bindings are identical to reviewed head 90e39b0; the only differences from that reviewed tree are four incoming cache files from main. No CI/queue or counting-default changes.

Each candidate is followed once. The fallback retains at most two frame records per deferred candidate and builds one bounded cumulative array at the root. Failed suffixes do not allocate fallback arrays. Existing successful selections, ties, three-header inference, plausible metadata and indexed/changed-rate/suffix policies retain their behavior. These are necessary validity checks, not audio decoding or a universal uniqueness rule. Indexed maximum-rate padding and broader ambiguity policies remain separate follow-ups.

Current integration validation: 1872 parser/API tests passed with one existing optional skip, 162 policy tests passed, build and 43 corpus replays passed. Independent integration source review is clear. Current-head CI run 40630646271 passed all 34 jobs, including the complete fresh parser audit: 1360 mutants, 100% adjusted. Both required app-bound checks passed; the tested merge tree exactly equals reviewed tree 21947c48835e48a4490fe331a09acab912a2d41e. Historical exact-head evidence below applies to reviewed head 90e39b0, not the new integration:

  • 1885 parser/API tests passed, two existing optional skips; 202 policy/shard tests passed, along with formatting, lint, TypeScript, build and all 32 exact proof bindings.
  • 43 corpus replays and the 8/32/128 MiB memory benchmark passed.
  • Red→green: the original ancillary-byte-100 regression failed at chunks 1/7/4096 before the first correction. Six independently supplied invalid-alternative cases also failed before the follow-up and now pass. Tests preserve genuine crossing ambiguity, metadata/configuration conservatism and exact main-data capacity.
  • The first hosted run exposed the inherited release-smoke mismatch and 17 unreviewed mutation survivors. The rate fix resolves the inherited issues. Duplicate short-candidate checks were removed; meaningful survivors gained regressions. A real crossing layout distinguishes a two-byte prefix-marker shift, and an existing failed-suffix probe distinguishes seven prefix allocations from the correct five.
  • All 31 existing proof identities and assumptions remain. One new independently reviewed exact equivalence covers the prefix-loop < length → <= length mutation: it adds only an ignored out-of-range typed-array write. The source-bound proof is explicit in mutation-equivalents.json and docs/mutation-review.md. Neither the meaningful marker shift nor the per-suffix allocation mutation is exempted.
  • Hosted CI run 40623443519: all 34 jobs passed on this head, including the complete fresh parser audit: 1360 mutants, 100% adjusted, packaged-runtime smoke, coverage, compatibility and bounded-resource checks. Earlier full local checks had six Windows socket failures reproduced on untouched main; the final focused suites above and hosted Linux aggregate pass. No additional full local mutation campaign was run.
  • Independent specification review cleared this exact head: the two previously failing impossible-alternate witnesses returned 1 and 2 respectively at chunks 1/7/4096; 157 focused tests and all 32 proof bindings passed review. This clearance applies to the reported correction, not all ambiguity policies.

Merged normally after independent review and all current-head required checks at e8b2dd3. No release publication.

Base automatically changed from fix/iso-free-format-ceiling to main October 28, 2026 04:06
@choskeli
choskeli force-pushed the fix/short-free-format-dead-end branch from 90e39b0 to d383a0b Compare October 28, 2026 04:08
@choskeli
choskeli marked this pull request as ready for review October 28, 2026 04:09
@choskeli
choskeli merged commit e8b2dd3 into main Oct 28, 2026
40 checks passed
@choskeli
choskeli deleted the fix/short-free-format-dead-end branch October 28, 2026 04:18
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant