698ba36ae452d4f0e2768410374d68fed4a161ef
331
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Matthew Jackson
|
698ba36ae4 |
Document the pack_language and detect_rate equivalent mutants
Three pack_language mutants and two detect_rate boundary mutants survive mutation testing with no test able to close them, and it's not for lack of trying: they're equivalent by construction. Recording the proofs next to the code so nobody re-chases them: - (b[0] - 0x60) as u16, shifted << 10 then truncated to u16, is congruent mod 65536 to (b[0] + 0x60) as u16 shifted the same way, because 0x60 * 2 * 1024 is an exact multiple of 65536. The same swap on the second letter (shifted only << 5) is NOT equivalent, which is why only the first letter's mutant survives. - The two | with ^ mutations that OR the three packed fields together are equivalent because the fields (a lowercase letter minus 0x60, so 1..=26) always fit in 5 bits and never share a set bit once shifted into their 0/5/10 positions. - detect_rate's tolerance and tie-break comparisons only diverge from their <= mutants on an exact 0.5 fps distance or an exact tie, and a brute-force search over every achievable integer-nanosecond median found no case that lands on either boundary bit-exactly. |
||
|
Matthew Jackson
|
1eb8bdc9c7 | Merge branch 'mux-mp4' into dev | ||
|
Matthew Jackson
|
90a7fe2ff1 |
Assert the MP4 timing arithmetic, and name the faststart slack rule
MP4 track timing was numerically unasserted. Every operator in the PTS-to-ticks, duration, tkhd_dur and ctts chain could be flipped and the whole suite stayed green, because no test decoded an output file and checked a concrete number — the existing tests assert box presence and gross container shape only. That is the crate's worst failure mode: a title muxes "successfully" with silently wrong A/V sync or total duration, and nothing above can tell. The new tests build tracks with known PTS deltas and compare the emitted stts, ctts and tkhd.duration against computed values. Also lifted the faststart slack rule out of the match guard into faststart_fits(). A leftover hole of 1-7 bytes cannot be expressed as any ISO-BMFF box, since a box header is 8 bytes, so finish() must fall back to moov-at-end rather than write a free box that lies about its own size. The condition now has a name and a test instead of being an unexplained `g == 0 || g >= 8` inside a pattern guard. |
||
|
Matthew Jackson
|
1b95d346bb |
Cite the H.264 spec directly, not a reference implementation
The escape-stripper comments named a third-party decoder as the authority for the cumulative-zero rule. This repo is public and does not cite other implementations; the rule is specified in ITU-T H.264 §7.3.1, which is the citation that belongs here anyway. No behaviour change — comments only. The scan-secrets gate caught it. |
||
|
Matthew Jackson
|
48663c6a2f | Merge branch 'mux-codec' into dev | ||
|
Matthew Jackson
|
a05f1d4498 | Merge branch 'mux-mkv' into dev | ||
|
Matthew Jackson
|
3ecb2510e8 |
Assert what the MKV mux and demux actually produce
A mutation run over mux/mkv.rs and mux/mkvstream.rs left 148 survivors.
Reading them turned up no wrong code, but a lot of code whose output
nothing ever looked at. Most of that is on the read side: parse_track
had a dedicated arm for Language, TrackName, FlagForced, Video and
Channels and not one of them was checked, so a re-mux could have lost
the audio language, the subtitle forced flag, every track label, the
resolution and the channel layout with the suite still green. Ten of
the eleven CodecID comparisons were unasserted too — only HEVC was
pinned — so any of them could have been mis-wired and the stream would
have gone to the wrong parser. The round-trip test now writes a real
three-track title through the muxer and reads it back through the
reader, and a separate test walks every registered CodecID.
The BPS statistics tag was the worst of the write side. Its test
asserted `file_bytes.contains("800")`, which a wrong bitrate passes
trivially — 80000 contains "800". Both the tag and the back-patched
Segment duration are now decoded and compared to a computed number, on
a title that declares no duration so the whole max_block_ticks →
seconds → bits chain is exercised. Cue points get the same treatment:
their CueTrack and CueClusterPosition were never read back, on either
the keyframe path or the i16-forced-split path, which are two hand-
written copies of the same three fields.
The rest closes arithmetic that only a bad disc reaches: a zero frame
rate or zero display-aspect denominator (both divisions), a
TimestampScale that does not fit an i64, a cluster timestamp of exactly
i64::MAX, a TrackNumber of 65537 that truncates onto the valid track 1,
and the shortest legal Block at both VINT widths. Two tests separate a
clean end of stream from a device failure: swallowing the second one
truncates the output at a bad sector and reports the rip complete.
Also pinned: only the first video and first audio track may be default
(the de-duplication lives in MkvStream::create and had no test at all),
the activation trigger is the first VIDEO track rather than track 0,
the measured field order reaches the file rather than just the helper
that computes it, and a Blu-ray 3D base/dependent pair builds the merge
instead of shipping two unrelated H.264 tracks.
96 of the 148 mutants verified killed by hand. Of the remainder, most
are equivalent — disjoint-bit `|` that `^` cannot change, delete-arm
mutants whose fallback is the same constant, guards on tracing calls —
and the write_frame branch at 1452 is unreachable: a cluster is always
open by the time it can be entered.
|
||
|
Matthew Jackson
|
048f125879 |
Kill codec mutation survivors and unify H.264's duplicated escape stripper
The mux/codec parsers (startcode, h264, hevc, dts) had 300 surviving mutants between them, and it turned out to be for the reason you'd fear: the exp-Golomb readers and the AU-boundary bitstream scanners had essentially no direct unit coverage, only indirect exercise through full-frame parse() calls that never touched the actual edge cases. Direct fixes to test gaps: - The shared BitReader's read_ue truncation guard (`leading_zeros > 31`) and skip_start_code's 4-byte-vs-3-byte boundary check had no test at their exact boundary. Added tests that hit the boundary precisely; a `>=`/`==`/`<=` typo either rejects a legal 31-leading- zero code or reads one byte past the buffer. - H.264's private SpsReader duplicates the same read_bits/read_ue shapes with no tests of its own at all (only reached through multi-field SPS parsing, several fields deep). Added direct tests. - HEVC's per-AU trailing-zero strip after the last NAL (no start code following) walks `end` down to trim padding; a wrong-direction typo there walks off the end of the buffer instead of terminating - exactly the "loop must make positive progress on malformed input" class. Added a test with a zero-padded trailing NAL. - HEVC's SEI match guards (`sei_mastering.is_none()` / `sei_content_light.is_none()`) implement "first HDR10 value in the title wins" - untested, and a naive test using both-messages-per-AU can't even exercise the guards because the whole-scan early return above them already handles that case. Split into single-message- per-AU tests that actually reach the arms. - parse_mastering_display/parse_content_light_level's length guards were `< N` with no boundary test; one-byte-short input now confirmed to return None instead of indexing out of bounds. - DTS's drain_front collapses duplicate offset-0 PTS markers after rebasing; untested, and the visible effect (front_pts()) can't tell a working collapse from a broken one since it already returns the right marker either way - the actual defect is unbounded growth of pts_marks over a long recording, so the new test asserts the bound directly across repeated drains. - DTS's dts_core_samples/dts_core_sample_rate header-length guard and next_core_boundary's syncword-length guard got exact-boundary tests the same way; also caught a nblks `<<`/`>>` direction bug candidate in the mutant (confirmed the real code is correct, just untested). Real bug found and fixed, not just a test gap: H.264's parse_sps_high_profile_ext re-implemented emulation-prevention byte stripping inline (a window scan: match `00 00 03` at position i, advance 3, else advance 1) instead of calling the existing unescape_ebsp_prefix used by slice-header parsing. On a run of 3+ real zero bytes ahead of an 0x03 - non-conformant, but this is disc bytes, not a spec-clean encoder - the two disagreed: unescape_ebsp_prefix's cumulative zero counter (matching the H.264 reference decode process and libavcodec's RBSP extractor) treats it as an escape and drops the 0x03; the window scan treats it as real payload and keeps it, corrupting the SPS bits read after it. Extracted the shared rule into `unescape_ebsp` (parameterized on output length so both the 16-byte slice-header prefix and the unbounded SPS case can share it) and pointed both call sites at the one implementation. Added a regression test pinning the shared function's behaviour on the input that used to separate them. All new tests hand-verified against the actual mutation (operator flipped or guard replaced by hand, confirmed red, then restored) per the mutation-testing brief, not just written and trusted. |
||
|
Matthew Jackson
|
65dbcb1ca6 |
Close mutation-testing gaps in the TS/PS mux (ts.rs, ps.rs, tsmux.rs)
A 12,330-mutant run left 159 survivors across these three files, all from missing assertions rather than wrong code — every gap here is a test, no production logic changed. Two shapes accounted for most of them: - Buffer-cap constants (MAX_PES_BUFFER_TOTAL, MAX_PS_BUFFER, MAX_BD_PES_PAYLOAD, PES_BUFFER_INIT_CAP) were only ever read by tests through their own symbol, so a mutated `*`/`-` in the constant's definition changes what the symbol itself evaluates to and every self-referential assertion still passes. Pinned each against a literal computed independently in the test. - Several `>`/`==` boundary checks on framing lengths (MPEG-2 pack header, system header, BD-TS adaptation field) were only ever exercised with slack in the buffer, never at the exact byte the check exists for. Added exact-fit cases for the pack header, system header, and psi_payload_base's AF-consumes-everything boundary. Real, higher-value gaps closed along the way: - ts.rs's per-PID discontinuity_flag and the NULL-TS concealment marker both require adaptation_field_length > 0 before trusting the AF flags byte; neither branch had a test proving af_len == 0 (no flags byte at all, ordinary payload underneath) is left alone. - header_remaining (PES header spillover across TS packets) only had single-continuation-packet coverage, which can't distinguish `-=` from `+=`/`*=` because the corrupted value never gets read again. Added a case spanning two continuations. - ps.rs's parse_stream_id_extension (used for HD-DVD 0xFD routing) walks nine optional PES-header/extension fields with a `pos +=` each; only the PTS/DTS pair had ever been exercised. One test now arms every field and checks the walk lands on the right byte. - find_ps_boundary's `sc + 3 >= len` guard had no test at sc == 0 with a bare 3-byte start code, the case a `+` -> `-` mutation turns into a debug-mode subtract-overflow panic on ordinary tail-of-buffer input. - tsmux.rs: an oversized video access unit must go out as a single unbounded-length PES; the `is_video || small-enough` guard that enforces this had no test with a video frame actually over the bounded-PES threshold, so a `||` -> `&&` mutant survived (it would silently split a keyframe across several look-alike-independent PES units). Also pinned the PES-length and PTS big-endian encodes at values above 255 / with bit 29+ set, where a `>>`/`<<` swap first becomes observable. Every test above was verified by hand: applied the exact mutation, confirmed the test fails (or the specific panic fires), then reverted. Left unclosed, all confirmed equivalent by hand-tracing rather than just left alone: - Every `<<8 | byte` PID/length bit-combine (ts.rs pid/PAT/PMT parsing, ps.rs dvd_audio_pid/hddvd_extended_pid/parse_pts): the two halves never share a bit, so `|` and `^` produce identical output for every input - no test can tell them apart. - ts.rs's `af_len > 183` check in process_packet: fully subsumed by the `payload_start >= TS_PACKET_BYTES` check three lines later for every af_len that could trip it. - ts.rs's out-of-range `pid_index` sentinel (-1 vs 1): unreachable, since a TS PID is masked to 13 bits (max 8191) and the table is always sized to at least 8192. - A cluster of "push an empty slice on an exact boundary" mutants in tsmux.rs's write_pes_chain (offset < hdr_len, af_bytes stuffing guards): the guarded write becomes a length-0 write_all, a no-op either way. Not reached this pass, for lack of a clean seam within the time available - ps.rs's extract_packets bounded-PES-length exact-fit checks (lines 278/282/303, the `sc+6>len` / `sc+6+pes_len>len` / force-flush cap arithmetic). The first two need a scenario where "proceed vs. wait one more byte" is observable in the packet list, and the third only shows up at a start-code offset (sc) that survives to the moment the cap check runs - in this code path sc is always 0 once an unbounded PES buffer starts accumulating, since nothing before it ever drains. Didn't find a construction in the time available; flagged rather than papered over with a self-referential assert. |
||
|
Matthew Jackson
|
528a6b7345 |
Walk past empty extents iteratively instead of recursing
fill_extents skipped an exhausted or zero-sector extent by calling itself, which costs a stack frame per skipped extent. Nothing filters sector_count == 0 out of a UDF or MPLS extent list, so a malformed disc declaring a long run of empty extents recursed once per extent before reading a single sector. Rust does not guarantee tail-call elimination, so that overflows the stack — which aborts the process rather than returning an io::Error, taking a long-running service down with it. The skip is now a loop. The regression test runs on a 256 KiB stack, where the recursive version dies and the loop finishes immediately. |
||
|
Matthew Jackson
|
71686f1407 |
Lint the test code, and fix the 74 findings it had been hiding
Every other repo's CI now runs clippy with --all-targets. libfreemkv, the crate the other seven build against and the one held up as the reference workflow, was the last one still linting the library only — so its ~3,000 tests, by far the largest body of test code in the project, had never been linted at all. Turning the flag on surfaced 74 findings. Most were mechanical and applied with clippy --fix. The rest, by hand: - Four discarded Results in decrypt.rs. css::descramble_region returns a Result and four CSS tests threw it away, so a descramble that FAILED would have surfaced as a confusing buffer-comparison mismatch instead of the actual error. They expect() now. - A dead `kp` field on the PlantedWalk fixture. The test deliberately asserts Kp as the explicit AES-G3(dk, 1) relation from [C] §3.2.4 rather than against a stored value — its doc comment says so — which makes the field not just unused but a trap: the obvious "fix" of asserting against it would quietly weaken the test to comparing the fixture with itself. Removed. - Two hand-rolled ICB counters in the HD-DVD fixtures, a needless mut, three vec!s that only ever needed arrays, a filter_map whose every arm was Some, and a Vec::new()+push chain. - Doc list indentation in mkv.rs and mp4/read.rs, which was mis-rendering in the generated docs. - A five-[u8; 16]-tuple return type named FourLevelParts. Three lints are allowed at the specific sites, with reasons, because they are wrong for this domain: the underscores in the bitstream-header literals mark BITFIELD boundaries, not digit groups, so regrouping them uniformly would satisfy the lint by destroying the only thing they encode; and in three table-validation loops the loop variable is the domain value under test (a DTS SFREQ code, an AMODE value, a palette entry number), which is what the assertion messages name. |
||
|
Matthew Jackson
|
42591c77fc |
test: constrain the AC-3/E-AC-3/DTS header decode and the boxes it emits
All 59 measured survivors in mp4/audio.rs: 50 killed, 9 proven equivalent, none left unaddressed. No production change — every extraction reads correct against ETSI TS 102 366 (5.3.2, 5.4.2, Annex E.1.3, F.4, F.6.1) and TS 102 114 5.3.1. It was a fixture gap, not a code defect, and a specific one: the existing fixtures gave several fields the SAME value (fscod=0, bsmod=0, lfeon=1, acmod=7) and asserted only derived channel counts. Nothing asserted the emitted dac3/dec3/ddts payload BYTES at all, so the packer's shifts and masks were entirely unconstrained. A wrong mask there does not crash — it writes a box declaring the wrong channel configuration, and a player believes it. Several kills needed fixtures designed to discriminate rather than merely exercise: the reduced-rate branch needed fscod == 3, which no test in the file reached — and `== -> !=` survived on an fscod=0 fixture only because the reduced table happens to return 48000 there too the DTS LFF mask needed a value where XOR and AND differ in MEANING: on the 5.1 fixtures LFF flips 1 -> 2 and BOTH codes mean "LFE present" the acmod mix-level skips needed three different acmods, because `^` is a no-op unless acmod == 4 exactly the ddts bitrate needed 44100/512, which does not divide evenly — at 48000 the rounding is invisible and `/ -> %` on den/2 survives The 9 equivalents are one pattern: OR-ing a shifted high part with a masked low part on disjoint bit lanes, where the mask is on the same line as the OR. Since cargo-mutants applies one mutation at a time, no single mutant can break both. Each was applied and observed green. FILED, not fixed: reserved sample-rate codes are silently guessed as 48 kHz (audio.rs:133, :20/:153) while a reserved DTS AMODE is refused with a comment explaining why. Same silent-wrong-metadata class, opposite answer. Refusing would make such a stream unmuxable, which is a product call. |
||
|
Matthew Jackson
|
5559987325 |
test(mux): replace a false-green discontinuity test with the property it named
a_signalled_discontinuity_survives_a_backstop_discard asserted that a SOURCE-signalled discontinuity on discarded bytes still reaches the AU that follows. It did not test that. Deleting the disc_marks push, or the mark-retirement loop inside discard_gap_before, left it passing. The mechanism: the `discontinuity = true` rode the FIRST over-cap push, which still has the next AU's delimiter at buf[0] — so it force-flushes as an over-long AU rather than discarding, and THAT AU consumes the mark. The assertion's `.find(|x| x.data.contains(&0x22))` then filters it out, and the flag it reads comes entirely from `pending_gap`, set by the second push's backstop. Behaviourally identical to the test 40 lines above it, under a name promising something else. I wrote it this morning, in the same commit that fixed a different test for having a fixture that never reached the code it named, while cataloguing that exact shape. Third instance today of writing the bug I was hunting. The two mechanisms cannot be isolated in one fixture — a fragment that trips the backstop sets pending_gap regardless — so they now get one test each. The replacement drives disc_marks end to end with no backstop involved: a flagged fragment that carries a complete AU and is emitted, not discarded. Nothing else pinned that path. Removing the disc_marks push reds it. Found by the round-9 opus escalation over test quality, dispatched because the sonnet pass over the same 17,000 lines of new test code returned zero findings. |
||
|
Matthew Jackson
|
8c0de5711e |
fix(mux): resync-gate drops reach errors(), including after the gap resolves
ResyncGate::dropped is zeroed the moment a keyframe disarms the gate,
and the only EOF warning fires for gates STILL armed. So a mid-title gap
that resolves left no trace anywhere — and most gaps do resolve. A rip
with several concealed gaps reported 0 errors and 0 lost bytes while
whole GOPs had been discarded, which disc.rs's own test comment calls
the ONLY channel through which loss is reported.
This is the other half of
|
||
|
Matthew Jackson
|
9f25a4c454 |
fix(mp4): refuse a video track with no resolved dimensions
Resolution::pixels() returned (0, 0) for Unknown, and the MP4 sink wrote it verbatim into tkhd (ISO/IEC 14496-12 8.3.2) and VisualSampleEntry (12.1.3). Both fields are MANDATORY there, so unlike Matroska — which omits the optional PixelWidth/PixelHeight elements — MP4 has nothing to leave out. The result was a structurally complete file that passes every container check, declares a 0x0 video track, cannot be rendered, and is written with no error anywhere. WHY IT WAS POSSIBLE, which is the part worth keeping: pixels() previously fabricated 1920x1080 for Unknown. That was wrong but playable, so this sink never needed a guard and the absence of one was invisible. Changing the sentinel to (0, 0) moved the defect instead of removing it — a zero PAIR still reads as a usable value, so the sink stored it and serialised it. The accessor's doc comment then ENUMERATED the callers it believed were safe: "the Matroska sink omits the optional elements, the VobSub writer omits its size: line, and no caller divides by either dimension." Two of those three are true. MP4 was not on the list because MP4 has no guard at all, and a prose list cannot enforce itself. mkv.rs's own comment even states the principle — "the check belongs in the one accessor rather than in each caller that remembered to write it" — and labels/mod.rs still carried its own duplicate Unknown test long after the accessor took that job over. So: pixels() now returns Option. Not because Option is tidier, but because every caller genuinely needs a DIFFERENT answer and the compiler is the only thing that reliably makes them choose one. Matroska and the metadata sinks take unwrap_or((0, 0)) with the reason stated at each site; the VobSub path degrades to a palette-only .idx; MP4 fails with E_MP4_UNKNOWN_RESOLUTION (9055). Six call sites, not the five my first grep showed — I piped it through `head` and acted on a truncated list. The compiler caught the sixth. That is the same mistake as trusting a lens that reported silence. |
||
|
Matthew Jackson
|
944e6a8b09 |
fix: align the Linux fsync error with macOS, and clear three stale docs
Round 9 findings, triaged and verified against the pinned tree.
writeback_file: a bounded-fsync WorkerLost returned bare ErrorKind::Other
on Linux where macOS returns EIO. Round 8 fixed the Linux arm to return
Err at all — the right fix — but stopped short of matching the value, so
a consumer distinguishing timeout / halt / lost-worker had nothing to
branch on for the third case on one platform. Now EIO on both.
Three doc comments described the pre-fix behaviour, one of them for
longer than the bug existed:
linux.rs durable_sync still said "all three fallbacks return Ok(())"
mod.rs sync_all still said Linux silently swallows fsync failures and
callers must not treat Ok(()) as a durability barrier
mod.rs SequentialSink::finish repeated the same caveat
All three now say what the code does: a bounded-fsync failure is an Err
on every platform, so Ok(()) IS a durability barrier. A doc that
describes a fixed bug is worse than no doc — it tells a caller to write
a workaround for something that no longer exists.
au_assembly: discard_gap_before duplicated drop_marks_before's
mark-retirement body verbatim and added one statement. Mine, from
earlier today. It now calls it. Two copies of the same retirement loop
is exactly how the two call sites would drift back together.
clpi: ClpiStream's audio_format / audio_rate / video_format / video_rate
are decoded from untrusted on-disc bytes on every parse and read by
nothing. The identically-named fields consumed in disc/bluray.rs belong
to mpls::StreamEntry, not to this struct — checked, because an earlier
round wrongly called a live function dead. Deleted, along with the seven
test assertions that pinned them; the tests that pin pid, coding_type
and language remain. Also removed a section-header comment orphaned by
the get_extents deletion, describing a fixture that no longer exists.
|
||
|
Matthew Jackson
|
0bc8d7af9c |
test: constrain the AACS key-map gap fill, the PSI walk, and MP4 field offsets
Third pass over src/mux/. 40 survivors killed, no production change. resolve.rs — the deleted-statement cluster is now fully constrained. All 14 deletable statements probed; 9 were already caught, 5 survived: c.sort_unstable() in fill_base_key_gaps. Every existing case handed it cuts already in LBA order, but IndividualSegment.tbl is a record list. Verified on HEAD: deleting the sort passes all 54 resolve tests. The mutant lays a base-key fill straight over a forensic segment. last_idx = idx (FMTS gap fill) and last_idx = hit (multi-CPS cache hit). An extent with nothing to sample must inherit its neighbour's CPS unit; the mutants fall back to the first unit's key. Exactly the shape this file's own comments name — wrong key, no error, lost_bytes == 0. Both check_halt()? polls in probe_fmts_index_keys. These cannot be killed by outcome, since a later poll returns Halted too. The tests count reads instead, which is what the don't-hammer-a-struggling- drive rule actually says: after a Stop the drive is asked for zero content sectors. The four unresolved += 1 arms each got a test, and deleting each fails exactly one — one-to-one, so no fixture passes for the wrong reason. A control test pins that the baseline table resolves, so an expect_err cannot succeed for an unrelated reason. ts.rs::scan_streams was never entered. Six killed, two of which return wrong answers that look right: reading the PAT/PMT CRC as a table entry invents a stream on PID 546 out of CRC bytes, and dropping the ES_info_length skip decodes a descriptor as an entry and loses the one after it. Every existing PMT fixture declares ES_info_length = 0; a real BD PMT carries a registration descriptor on essentially every entry. Also ISO/IEC 13818-1 2.4.4.3 program_number == 0 is the network PID, not a program. mp4/read.rs — 13. Height read as the width beside it; channelcount; the 4-byte base-128 descriptor varint (every existing esds fixture uses a single byte); all three optional ES_Descriptor fields, whose loss is silent (an AAC track just loses its CodecPrivate); first-vs-last media edit, which is A/V desync of the difference; and the version-1 mvhd timescale offset, emitted by any writer whose duration exceeds 32 bits. dts.rs — 7, from a real cargo-mutants run over the file rather than guesswork. Including a buffer that IS the syncword, which is the state a sync split across PES packets lands in the moment its last byte arrives. Equivalents proven by application, not argued: the sample_encrypted_units guard pair is mutually redundant by construction (total*p/9 < total for p <= 8), so either alone is equivalent and both together are not; the PMT section_len guard is dead code where its PAT twin panics; three of the seven EXSS_HEADER_MIN_BYTES arithmetic mutants still sum to 10. |
||
|
Matthew Jackson
|
e99b634635 |
fix(mux): a backstop discard is a discontinuity; a stream-start trim is not
drop_marks_before retired discontinuity marks alongside timing marks at both of its call sites. At stream start that is right. At the MAX_AU_BUFFER backstop it is not, and the two are now separate. The backstop fires when 8 MiB accumulate with no AU start code in them — corrupt or hostile input — and throws the run away. There IS a prior AU in that case, and whatever emits next definitively does not continue it: a decoder handed that picture resolves its references against frames separated from it by megabytes of discarded data. Retiring the flag meant the resync gate (resync.rs, driven from mux/disc.rs) never armed, so the broken picture went out looking sound. Silent corruption is the one class of loss this crate refuses to have. At stream start the opposite holds. Bytes ahead of the first access-unit delimiter are the tail of an AU that began before sync, and there is no prior AU to be discontinuous from. Marking it would arm the gate at the head of every title and drop its opening GOP. That risk is why this was a decision rather than a fix, and splitting the call sites is what avoids paying it. Recorded as a sticky flag, not an offset mark. A mark placed at the new base is retired moments later by the pre-sync trim that follows resync — the gap has to outlive the bytes that caused it. I found that by writing the test first and watching it fail with the mark approach. The discard is a discontinuity whether or not the source signalled one, and a signalled one on discarded bytes still reaches the AU that follows; both directions are tested. Note the first over-cap run is NOT a discard: the next AU's delimiter is still at buf[0], so it force-flushes as an over-long access unit and loses nothing. Only a run with no opener at all reaches the backstop. The tests push twice for that reason — the single-push version passes without the fix. Swapping either call site for the other fails: reverting the backstop reds the two gap tests, and arming the gate at stream start reds the third. |
||
|
Matthew Jackson
|
b2e1982051 |
fix(clpi): resolve out_time past the last EP entry to the end of the clip
ClipInfo::get_extents fell back to `last EP SPN + 1` whenever out_time lay past the last entry-point. EP entries mark I-frames (BD-ROM Part 3, CPI / EP map) and a clip's final GOP lies after the last one, so a PlayItem covering a whole clip — whose OUT_time is the presentation end — always lands in that arm. The extent then stopped one source packet after the last I-frame. Measured on a fixture with 200,000 source packets and the last EP at SPN 131,072: sector_count came back 12,289 where covering the clip needs 18,750. Everything from the last entry point to EOF is outside the returned extent. Scope, stated plainly: get_extents has NO callers anywhere in the ecosystem today — it is #[allow(dead_code)] and documented as reserved for the timestamp-range read path. Nothing ships this loss. It is fixed now because a latent truncation in extent arithmetic is far cheaper to correct before it has callers than after. The SPN at-or-after an out-of-range out_time is the end of the clip, source_packet_count, with .max(last + 1) so a disc that under-declares its own packet count against its own EP map still yields a sane bound. Also 174 mutants killed across clpi, mpls, ifo and ebml — the first time any of these four files has been examined. And ebml's 8-byte VINT back-patch was duplicated verbatim in end_master and end_master_buf with its top four payload octets unreachable through either (they need a 16 MiB..256 TiB buffer); extracted to fixed_width_vint8 and tested across the full 56-bit payload, no behaviour change. 38 of ebml's 46 survivors are one equivalence cluster: every | in write_size / read_id / read_size / read_uint_val ORs into disjoint bit lanes, where ^ is the identical operation. Applied all 38 at once — green — then spot-checked four individually. |
||
|
Matthew Jackson
|
9de88969ca |
test: constrain the DiscStream loss surface and the empty-title guards
Second mutation pass over src/mux/. 26 survivors killed, no production change. Verified on HEAD before landing: each mutation below passes all 1,237 mux tests unmutated-suite. The priority item was the honest-loss-reporting surface. Both DiscStream::errors and DiscStream::lost_bytes could return a constant with nothing failing — a rip that lost sectors would report zero loss to the caller. This project has already shipped one defect of that shape (a total decryption failure reported as an empty title, exit 0). Driven now through two short-read fills so both land on values that are neither 0 nor 1 and differ from each other; no constant and no field swap survives. MkvStream::finish -> Ok(()) also survived. MkvMuxer::finish has the zero-frame MkvInvalid guard and two tests cover it, but the Stream wrapper above it could return Ok unconditionally and bypass the guard entirely — the empty-title defence was one layer thinner than it looked. au_assembly: pinned au_opener_from behaviourally to the normative byte values for all four modes, with negative cases for codes that are explicitly not openers (MPEG-2 slice 0x01..0xAF, user data 0xB2, extension 0xB5, sequence end 0xB7 per 13818-2 Table 6-1; VC-1 0x0A/0x0B/0x0C; H.264 SPS/PPS/IDR-slice). au_assembly and codec/ hold independent copies of these constants; they agree today, and comparing constants would not catch logic drifting apart, so both sides are now pinned to the spec instead of to each other. demux_sink::sanitize: every filename component demux:// writes comes from disc-controlled text, so the path-separator arm is a traversal guard. Deleting it now fails, including an end-to-end case where base = "../evil/Title" must produce exactly one file inside the chosen directory. stts_and_ctts_expand renamed to stts_expands_runs_to_per_sample_deltas_in_order and given runs with distinct deltas AND distinct lengths. Its old name claimed ctts coverage it never had, which is why the composition-time chain went unconstrained for eight rounds; the doc comment now points at the tests that do cover ctts. Correction to the previous pass: codec/truehd.rs flush -> vec![] IS equivalent. Applied it, full mux suite green. TrueHD buffers across PES but parse emits every complete unit immediately, so a residual buffer at EOF is a truncated access unit and is correctly discarded. The vec![Default::default()] variants are genuinely different and are killed. Deliberately not constrained: mkv::set_opening_capture (diagnostics behind a process-global tracing check, flaky under the parallel runner), and the three stdio.rs header paths (StdioStream holds concrete io::Stdin/Stdout and cannot be driven without a production refactor to injectable Read/Write). |
||
|
Matthew Jackson
|
170fd0c064 |
test: constrain MP4 composition timing, MLP substream directory, and codec-private absence
Mutation testing over src/mux/. No production change — 49 survivors killed, all proven red before green. The MP4 composition-time chain was entirely unconstrained: VideoTiming::ctts, build_ctts and parse_ctts could each return a constant and the suite stayed green. Confirmed on HEAD: build_ctts -> vec![] passes all 1,220 mux tests. A demuxed B-frame title presenting in decode order would have shipped. The cause is a test whose name asserts coverage its body does not deliver — stts_and_ctts_expand builds an stts box and never touches ctts, and write_then_read_round_trip asserts sample sizes and keyframe flags but not one PTS. Same shape as the set_speed forwarding finding, different disguise. mlp_num_substreams / mlp_substr_header_size: every TrueHD fixture in the crate uses one substream and no extraword, so both could return a constant and agree with all of them. These position mlp_parity_ok's window over the AU header, so a constant mis-windows the parity check on exactly the multi-substream AUs that carry 7.1 and Atmos. CodecPrivate absent vs empty: mkv.rs writes Some(bytes) verbatim and omits the element on None (RFC 9559 5.1.4.1.24), so a zero-length Some emits a track header asserting the config IS empty. Four parsers could return Some(vec![]) before any frame. Also: mandatory ISO/IEC 14496-12 boxes (tkhd, vmhd, smhd, dinf, mdhd) could each build empty; HEVC num_extra_slice_header_bits (H.265 7.3.2.3) was never non-zero in any fixture, so the slice-type offset skip was unexercised; chapter names from the disc go straight into <ChapterString> and the & escape must run first; a stray 0x47 in a payload must not latch a TS resync. Documented as equivalent rather than killed: CodecParser::flush and the three parser flush bodies that differ from the mutant only by a tracing call, and DropTally::log_summary. |
||
|
Matthew Jackson
|
c63dafcf1a |
Key each FMTS extent from its own CPS unit, and stop a bad ICB tag reading
as an empty directory On an AACS 2.1 disc the non-forensic gap fill hardcoded pool slot 0 as "the" base Unit Key, so every content LBA outside a forensic segment was keyed with CPS unit 1's key even on a disc carrying several CPS units. It does not fail loudly — it produces garbage plaintext. The gap fill now resolves each extent's own base key from its ciphertext, sharing the sampling and slot-picking the multi-CPS path already had rather than adding a second copy, and memoised in the existing per-disc cache. A disc with one base key still short-circuits with zero extra reads, which the existing probe-cost test pins. Forcing the slot back to a constant fails four tests, so the choice is load-bearing rather than incidental. A directory ICB whose descriptor tag is neither File Entry nor Extended File Entry (ECMA-167 4/14.9, 4/14.17) was turned into a successfully-read EMPTY directory, indistinguishable from a genuinely empty one, while the same tag on a file ICB was already a hard error. Fifth instance in this audit of a failure converted into a plausible success value, and the second in this very function — round 5 fixed a read error becoming a file size of zero here. An unrecorded extent (ECMA-167 4/14.14.1.1: allocated but not recorded, logically zeros that still occupy file space) was dropped entirely rather than contributing its length, so every later extent landed at the wrong file offset. Silent corruption, not an error. Extents now carry a recorded flag and the hole emits zeros without touching the media. The Volume Descriptor Sequence was swept at hardcoded sectors 32..64 while the anchor's own Main VDS Extent pointer was parsed into a comment and ignored; ECMA-167 3/10.2.1 defines that extent by the field, not by position, so a conformant volume placing it elsewhere failed to mount. drive_status decoded byte 5 as Media Status without checking the event header's NEA bit or notification class (MMC-6 §6.7), so a reply carrying no media event descriptor decoded as "no disc". The drive is untrusted input here, and this is the works-on-my-drive class. Two more tests were found asserting the defects they sit next to — one requiring unrecorded extents to be dropped, one that four drive fixtures built non-conformant replies the corrected decoder rightly rejects. Both rewritten. Combined with the DTS one in the previous commit that makes three tests this round that locked a bug in as intended behaviour, which is a different and worse failure than the tautological tests found so far: a tautology fails to catch a regression, these actively defend the defect. Not fixed, adjacent: extract_one_file streams extents sequentially and will now READ an unrecorded extent's sectors rather than writing guaranteed zeros. Offsets are right, and pressed media reads as zeros there, but it is not zero-guaranteed the way read_file now is; that needs a recorded flag through PlannedFile. |
||
|
Matthew Jackson
|
46eb88c51f |
Feed the CSS crack the canonical extent order, and stop Resolution faking 1080p
Seven defects in the code the test suite executes least — 913 lines of disc/mod.rs alone are run by no test at all, which is why this round scoped from coverage rather than from what previous rounds said they had read. Disc::scan_image kept its own copy of the crack's extent ordering and fed crack_key_outcome largest-cell-first. That is the fifth instance in this audit of a local reimplementation drifting from the canonical one, and the cost here is a key that does not descramble the feature: picking by sector count bypasses the capacity gate and can select a different VTS entirely. The copy is gone — which title comes from the canonical order the scan already applied, and the extents are handed over in playback order, exactly as decrypt_keys_for_title does. Its doc records why the duplicate existed so it cannot grow back. Resolution::pixels returned 1920x1080 for Unknown. That is the FOURTH instance of one trap and the other three were in this same file, two of them fixed hours earlier — without sweeping for siblings, which is the whole reason this one survived. It now returns (0, 0), and the sweep was done properly this time: every remaining Unknown arm across the crate is honest, and the two ColorSpace sites that look like fabrication are emitting H.273 code point 2, which is the spec's own "unspecified". Two callers carried local Unknown-to-zero workarounds — precisely the cost of making callers responsible for a lie — and one is now redundant. BD-ROM Part 3 code 0xA2 is the lossy secondary DTS stream, not lossless Master Audio. A test asserted the wrong mapping as intended behaviour, so correcting the code failed it; the test is deleted with a note pointing at its replacement. That is a NEW failure mode for this audit: not a test that cannot fail, but one that locks the defect in. There is no DtsExpress variant to map to, so it takes the lossy DTS-HD member and the approximation is documented. Also: DiscSession::identify could panic through drive_mut once the public API allows an absent drive — two siblings were converted in an earlier round and this one was missed; an extent end that added without saturating where the rest of the crate saturates; a diag reason string restating the comparator's sort keys and drifting from them, now derived from them; and a short read that advanced the offset by the full request, silently skipping the gap. That last one existed twice, in two reads with the same shape, now merged so they cannot drift apart. The short-read policy is a judgement call I could not derive from a spec: no skip_errors is a hard error, with skip_errors zero-fills and charges the loss. It deliberately does not retry mid-unit, because resuming inside an AACS aligned unit would trade a silent gap for a silent decrypt desync — the worse of the two. |
||
|
Matthew Jackson
|
dea968f32b |
Stop AudioChannels and SampleRate fabricating a value for Unknown
Three copies of the same two mappings existed. The canonical accessors returned 6 channels and 48000 Hz for Unknown; a third copy in diag.rs returned 0. The honest one was the copy. A plausible wrong answer is worse than an obvious one. Six channels at 48 kHz is indistinguishable from a real 5.1 track, so every caller became responsible for remembering to check the variant first — and this crate walked into exactly that: the json:// sink reported a confident 5.1 for audio whose neighbouring fields said "unknown". That was fixed at the call site earlier in this audit; this fixes it at the source. The accessors now return 0, which is what both in-crate call sites already coerced Unknown to by hand, so their guards are gone and the behaviour is unchanged. Zero is also obviously wrong if it ever reaches output, where six is not. The diag.rs duplicates are deleted rather than corrected — a fourth copy would have drifted too. Their only caller was a trace line in the same file, now on the canonical accessors. Their tests moved across and gained the Unknown case, which is the point: restoring either fabricated value fails both. Found by the round-7 correctness agent while fixing the json:// sink; it flagged the third copy as out of its scope rather than touching it. |
||
|
Matthew Jackson
|
079c9b1327 |
Add a seeded robustness harness for the untrusted-input parsers
Five parsers that take bytes straight off a disc are now swept with generated input asserting one property: they return Ok or Err and never panic. That is this crate's own hard rule, and the class seven rounds of reading is worst at. Written in-crate rather than with cargo-fuzz, which needs a nightly toolchain this project does not use, and without proptest or arbitrary, because one dev-dependency is a deliberate posture and the parsers take plain byte slices. What is given up is coverage-guided mutation, which is the real loss. What is gained is determinism: the same seed replays the same cases anywhere, so a CI failure reproduces locally verbatim. Three generators, and the second is the one that matters. Pure random bytes die at the magic check and exercise the entry guards only; prefixing valid magic is what reaches the parser body; mutating a mostly-zero record is what reaches the offset and count arithmetic a hostile image would lie about. That claim is MEASURED, not asserted. A harness whose cases all bounce off the entry guards is the fuzzing equivalent of a test that cannot fail, so one test counts how many generated cases parse to completion: 15,606 of 60,000, about 26%. If a future change to a guard drops that to zero, the test fails rather than continuing to report a meaningless pass. Two further tests pin that the three generators produce different bytes and that a seed replays identically. 1.2M cases across all five targets found nothing. On this evidence that is a real negative rather than an empty one. The first version of this file was itself broken in the way this audit keeps finding: its two meta-tests set FREEMKV_HARNESS_CASES and raced, because the test harness runs them in parallel and env mutation is unsound there. The budget is a parameter now, and the environment is read once at the call site. Two crate-internal parsers widened from private to pub(crate) so the harness can reach them. No public API change. |
||
|
Matthew Jackson
|
327087c70e |
Make five tests capable of failing, and stop the presence probe unmounting the disc
The worst of the five was a regression suite that never touched the code it guarded: nine batch-count tests called `safe_batch_count` and `buggy_batch_count`, both defined in the test file itself. The u16 truncation they exist to prevent could be reintroduced in sector/prefetched.rs with every one of them green. They now drive the real producer through the public API, and reinstating the truncation fails five of the nine. Worth recording that the symptom has changed since the original fix: the unit-alignment clamp below floors a zero batch at three sectors, so the bug is now a twenty-fold throughput cliff rather than the stall it once was. The MP4 reserve test's only numeric case was dominated by the floor and the buffer, so BYTES_PER_SAMPLE could be zeroed without failing it. It now has a case where the per-sample term dominates. The zero-count guard in FileSectorSource was likewise unfalsifiable — seek-past-EOF and a zero-length read both succeed — so the test now observes the file cursor. The AACS media-key ambiguity guard had no test at all; the pool scan is extracted so the verifier can be injected, because a genuine two-key collision needs one ciphertext decrypting under two AES-128 keys to plaintexts sharing a 64-bit magic, which is a 2^64 search and not a fixture. macOS implemented the documented cheap, side-effect-free presence probe by building a full exclusive transport — which force-unmounts the disc. Linux and Windows issue one TEST UNIT READY with no unmount; macOS was the outlier. It now walks the IOKit registry for the media object instead. The C shim's registry reads assumed CoreFoundation types the registry does not guarantee, so a driver publishing a CFNumber where a CFString was expected aborted the process from inside public API. Types are checked and a wrong type treated as absent. The unbounded waitpid on the unmount child is now a polled deadline, and the last-resort match gained the NULL check its two siblings already had. The empty-CDB guard existed only on Linux while a shared helper's comment claimed all three backends had it. Moved into the helper, so the comment is now true and macOS and Windows are covered. One finding was REJECTED with evidence rather than fixed. The TrueHD buffer-cap test was indeed bogus, but MAX_TRUEHD_BUF turns out to be unreachable by any input: the parser only retains data when the buffer is shorter than the declared AU, and that declaration is twelve bits, so the worst case is 8189 bytes against a 256 KiB cap. An exhaustive sweep over all 65536 AU headers confirmed it. The fixture now sits at the reachable ceiling and asserts that instead. The cap itself is left in place as defence, unreachable by construction, matching how the AC-3 resync guard was handled earlier in this audit. Two behaviour changes worth naming: Linux's empty-CDB error becomes InvalidCdbLength rather than a transport failure, and an unknown device now reports absent media rather than a not-found error, because the registry cannot tell an empty drive from a missing one. The latter is a conflation of the kind this audit has fixed three times; it is recorded for the next round rather than left silent. |
||
|
Matthew Jackson
|
b8fa5e74dc |
Stop the live rip path muxing Blu-ray 3D differently from the ISO path
Five defects, four of them the same shape: a local reimplementation of logic the crate already had, which had drifted from it. Each is now fixed by calling the canonical version rather than by patching the copy. DiscStream::new — the live disc:// path — built every parser through the plain codec lookup and never asked whether a video stream was an MVC dependent view, though resolve::build_demux_state does. The same 3D disc therefore muxed correctly from an ISO and incorrectly ripped live. The open-coded loop is gone; both paths now call build_demux_state. collect_psi_section reimplemented the continuity-counter gap test and disagreed with process_packet in the same file: it tolerated neither a duplicate packet nor an adaptation-field-only packet, which per ISO/IEC 13818-1 §2.4.3.3 does not increment the counter. A spec-legal PMT continuation was read as desync and the title's stream list came back empty. Both callers now share one `cc_is_gap`, and a duplicate packet's payload is no longer appended twice — doing so would have corrupted the section the check exists to protect. The json:// sink called the channel-count and sample-rate accessors unconditionally, and both fabricate a concrete value for Unknown, so it reported a confident 5.1 at 48 kHz for audio whose format was unknown while its own neighbouring string fields said "unknown". The keys are now omitted, matching mkv.rs. This matters more than it did: a sample-rate ladder fixed earlier in this audit means Unknown now reaches consumers that used to receive a wrong-but-concrete value. For an audio:// or sub:// sink the reference video track's output is filtered out, so its first PTS was never recorded and every delay was computed against zero — baking a wrong DELAY into the filename. The reference is now recorded whenever a frame is on the reference track, independent of whether that track has an output, so a normal title gets a correct delay; where no reference is ever observed the tag is omitted rather than guessed. A third copy of the channel/sample-rate mapping exists in src/diag.rs and was left alone as outside the confirmed set. It is the same drift shape and is recorded for the next round. |
||
|
Matthew Jackson
|
3f7d7af472 |
Bound three allocations an untrusted disc can drive without limit
The Program Stream demuxer appended every fed byte and enforced its 4 MiB cap only inside a branch reached once a start code had been found. Input containing no start code anywhere therefore hit no cap at all, and since a whole title is fed through this demuxer, a zero-filled or ciphertext VOB extent buffered the entire title — up to ~90 GB. When the buffer holds no start code, only a two-byte `00 00` prefix can begin a PS unit on the next feed, so that is kept and the rest dropped. The bound is exact rather than a heuristic: a start code can straddle a feed boundary by at most its first two bytes, so no real byte is discarded, and a test feeding `FF FF 00 00` then `01 E0 ...` pins that. The existing test named for this case fed a real start code first, so the cap it exercised was the in-PES one. Renamed to say what it covers. The BD-J label path had a different shape to anything found so far: the cap is on the COMPRESSED size of a disc file while the allocation scales with the decompressed size. A `.class` gated only by a path prefix inflates to the 64 MiB ceiling, yielding ~33M retained strings from `ldc` operands or ~67M pushes onto a symbolic stack whose depth was unbounded despite the Code attribute's own `max_stack` being parsed and then ignored. Bounded both, the stack by `max_stack` itself (JVMS §4.7.3). The VMG TT_SRPT title count is an untrusted u16 with no de-duplication, so ~800 KB of crafted IFO re-parsed one PGC 65535 times. Capped at 99, the DVD-Video maximum, so no conformant disc is clipped. Every cap carries stated headroom against real media, and each has a test locking that real media still passes. I rewrote three of the new assertions before landing them. They compared the result against the very constant under test — `total <= MAX_TT_SRPT_TITLES` — which passes vacuously the moment someone raises the constant, the most likely future regression and the seventh instance of this tautology shape in this audit. They now assert literals derived from the spec. The TT_SRPT fixture also had to change: with 65535 identical entries the de-duplication collapsed them on its own and the cap was never what bounded the result, so the test passed with the cap removed entirely. Distinct entries defeat dedup and leave the cap as the only guard; de-duplication now has its own fixture. Verified by raising each of the three constants and confirming all three tests fail. |
||
|
Matthew Jackson
|
fdd473d7e9 |
Remove emulation-prevention bytes before reading the H.264 slice header
The bytes after a NAL header are EBSP, not RBSP: ISO/IEC 14496-10 §7.4.1 has the encoder insert 0x03 after any 0x00 0x00, and §7.3.1 removes it before parsing. The measured-picture-type parse read the raw NAL instead, on the stated reasoning that slice_type is too early for an escape to intervene. That holds only up to a point. first_mb_in_slice is ue(v), so a value of 65535 or more needs sixteen leading zero bits and opens the payload with 0x00 0x00, which an encoder must then escape. A UHD frame is ~32,400 macroblocks, so a conforming Blu-ray never reaches it — but 8K does, and the disc is untrusted input. Such a stream decoded slice_type against a byte the encoder had inserted and reported the wrong picture type: a wrong result rather than an error, which is the class this lens exists for. The prefix is un-escaped into a 16-octet buffer rather than the whole NAL: the two ue(v) fields are at most 32 bits each, so nothing longer can be needed, and it keeps a per-frame allocation proportional to the frame off the path. The test pins both directions. It asserts the un-escaped prefix decodes to first_mb_in_slice = 65535 and slice_type = 2, AND that the raw EBSP does NOT — without that second assertion the test would pass whether or not the fix were present, which is the failure mode this audit has now found four times. The bit string was derived independently rather than by hand: my first attempt at the fixture was wrong by one nibble and the test caught it. Also covers the cases that must NOT be unescaped: a 0x03 not preceded by 00 00 is ordinary payload, and 00 00 03 03 keeps its second 0x03 because the escape resets the zero run. |
||
|
Matthew Jackson
|
d444afbdfc |
Turn a release-only slice panic into an error, and stop calling 32 kHz 48 kHz
Four round-6 findings. FileSectorSource::read_sectors guarded its output buffer with a debug_assert, which is compiled out in release — so an undersized buffer panicked with 'range end index out of range' instead of returning an error, out of a public SectorSource impl where the length is caller input. Drive::read_fua already carries this exact guard, with a comment recording the same panic being fixed there, and PrefetchedSectorSource has a regression test for the same case; this impl had been given neither. The new test is red in release for precisely the predicted reason: 'range end index 8192 out of range for slice of length 2049'. parse_track's sample-rate ladder ended in an unconditional S48, so any SamplingFrequency below 44100 was recorded as 48 kHz. A 32000 Hz AC-3 or DTS track is legal and common in broadcast-sourced content, and the wrong rate then propagated into the reconstructed AudioStream. Anything below the lowest mapped rate is now Unknown, which is what the crate's canonical SampleRate::from_hz already returned — the ladder disagreed with it. The ladder itself stays, because the MKV element is a float and wants tolerance rather than exact equality. shim_open_exclusive used the mach port from IOMainPort without checking the return; on failure the port is left untouched and every IOKit call below ran against an uninitialised value. shim_list_drives in the same file does check it. build.rs treated cc and ar as successful if the process merely SPAWNED, so a genuine compile error in the macOS C shim produced no object file and surfaced later as an unexplained link failure against a missing symbol. The shim is macOS-only and is neither linted nor compiled on the other two platforms, so a mistake in it has exactly one chance to be noticed. The last two have no test: one needs IOMainPort to fail, the other needs a deliberately broken C shim, and neither is reachable from the test harness. Both mirror a correct sibling in the same file, which is the evidence available. |
||
|
Matthew Jackson
|
399c3d2769 |
Reject an over-length CDB on every transport, splice H.264 param sets in place
Two fixes from round 5. The Linux and Windows backends truncated a CDB longer than 16 bytes (`cdb.len().min(16)`) where macOS returned InvalidCdbLength. Under SPC-4 a command's length is fixed by its opcode group code, so a shortened CDB is not a shorter form of the same command — it is a DIFFERENT command, and the drive executes it and answers GOOD with data for a request nobody made. A silently wrong result on the layer everything else sits on. Rather than mirror the guard a third time it now lives in scsi::mod as checked_cdb_len, with all three backends routed through it, so it cannot drift per platform again. That also makes it testable everywhere: each platform module is cfg-gated to its own host, so a guard inlined into linux.rs and windows.rs would have had no test coverage on any single machine. The shared helper is the only place the behaviour can be asserted on every platform's CI. The two existing macOS tests were tautological — they replicated the guard's logic inline instead of calling it, so they would have passed with the guard deleted. They now call the real helper. Separately, the H.264 keyframe parameter-set re-assert grew a few-hundred-byte prefix buffer to the full access-unit size, copied the whole frame into it, and dropped the presized buffer: one extra whole-frame allocation and copy per keyframe. A UHD title is ~200,000 frames of 150-400 KB with a keyframe every second or two, so that is thousands of avoidable multi-hundred-KB copies per title, each large enough to go through mmap. It now splices into the reserved headroom in place. This mirrors the identical fix already made in hevc.rs, which the H.264 path had drifted from. Byte-for-byte equivalence is pinned by a test whose expected literals were captured from the pre-change implementation, and which I confirmed still passes when the old build-and-copy code is restored. The no-reallocation claim is measured rather than argued: a counter over 30 bare keyframes, which reports 30 of 30 against the old path and 0 with the splice. The reallocation test initially passed even with PARAM_REASSERT_HEADROOM set to zero, because a small parameter set fits in the presize's incidental slack — it proved the fixture did not reallocate, not that the headroom prevented it. Its SPS is now large enough that the constant is load-bearing, so zeroing it fails the test. Not verified: no runtime behaviour on Linux or Windows: no drive, no ioctl. Both files were confirmed to compile for their own targets. |
||
|
Matthew Jackson
|
5c6a6d0785 |
Round 5: reject a degenerate fixed lace, bound the pending buffer by bytes
Five fixes. Three are real defects with regression tests; two are bounds that were expressible but not expressed. A fixed-size lace (RFC 9559 §10.3.4) whose body is empty declared n frames and carried none. The divisibility check passed, because 0 % n is 0, and `chunks` yields nothing on an empty slice whatever width it is given — so the clamp that existed to avoid chunks(0) returned zero frames where the Lacing Head said n. The whole lace vanished with no error raised and the caller saw a clean short block. A zero-size frame cannot be a valid frame, so it is now malformed. A disc read failure while fetching a directory entry's ICB became a file size of zero rather than an error. Zero is indistinguishable from a genuinely empty file, so an unreadable ICB on a damaged disc silently changed which titles a caller saw as present — read_directory already fails hard on its entry-budget guard, so propagating is also what the surrounding code does. read_file_size still returns Ok(0) for an ICB whose tag is neither File Entry nor Extended File Entry, which is a real zero and not a failure. The pending-frame buffer was capped at 4096 frames, which does not bound memory: frames are arbitrarily large and a UHD video frame runs to a few hundred KB, so the existing cap permitted over a gigabyte. Now bounded by bytes as well, at 64 MiB. round_up_grain overflowed for inputs within one grain of u64::MAX — div_ceil then multiply — and the wrapped product is small, turning the largest possible estimate into a negligible reserve. It saturates, and the reserve is clamped to what a `free` box's 32-bit size field can actually hold, since writing a larger one truncated the size and left mdat beyond a box claiming to be far shorter. No real title comes close; a 90 GB UHD title estimates a few MiB. The AC-3 resync guard now advances the PTS cadence like both of its sibling branches, so the three paths out of that block cannot disagree. This one is defensive and has NO test: reaching it needs input that both parses frames and leaves a megabyte of residue, and the parser's own carry rules drop pre-sync junk and cap a partial frame at 8192 bytes, so no such input was found. Stated here rather than covered by a test that would pass either way. Two findings from this round were rejected on inspection. A reported panic in the .mpls suffix check does not exist: the `.get(..)` on the line above returns None off a char boundary and `filter` never runs its closure, so the byte index is unreachable. A test written for it passed against the unfixed code, which is what surfaced the error. |
||
|
Matthew Jackson
|
f5e169efb3 |
Stop reporting a corrupt mkv:// source as an empty title
E6008 meant two unrelated things: "this title produced no muxable frames", which is a benign stub worth skipping, and "the source file is malformed", which is not. Because a single code carried both, is_skippable_title_stub answered yes to the second one — so feeding a truncated or corrupt mkv:// input made the engine classify it SkippableStub, print a notice saying the title was empty, and exit 0. Silent data loss reported as success. Split into E9053 MkvSourceInvalid for the read path (25 raise sites across mkvstream.rs and ebml.rs's read primitives) and E9054 MkvUnencodable for the four write-side sites, which are the encoder refusing to emit a body at or above the 56-bit VINT limit — an output limit with no input involved, so calling it a corrupt source would be wrong in the other direction. E6008 keeps only the zero-frame guard it was documented to mean. Kept one code for the whole read path rather than one per raise site: nothing a consumer does differs between a bad VINT, a non-UTF-8 string element, a truncated body and a child overrunning its parent. E9052 is the model for when a carve-out earns its keep — laced blocks name one specific RFC 9559 §10.3 feature with its own diagnosis. Also fixed meta_sink.rs raising MkvInvalid for a serde_json encode failure in the json:// sink, where no MKV is involved at all; it now matches the identical guard in mux/meta.rs. Reverting the split at the single code() arm reproduces the old classification: 22 tests fail, including both new assertions. The opposite direction is pinned too — dropping E6008 from the predicate fails the genuine-stub test, which drives the real muxer end to end. |
||
|
Matthew Jackson
|
0bbceed985 |
Round 4: fix 26 defects across crypto, resource use and codec paths
Twenty-six confirmed findings from the fourth audit round, landed as one cluster because they were found by agents working over disjoint file sets. The one worth calling out is a pair of AACS tests that could not fail. Both asserted CBC behaviour against a hand-rolled expectation that happened to be IV-independent, so replacing AACS_IV with sixteen zero bytes left them passing — they were pinning the code's own arithmetic, not the published constant. Replaced with a literal witness of the published IV plus the NIST SP 800-38A F.2.2 CBC-AES128 vector, and verified the other way round: zeroing AACS_IV now fails three tests. The rest are allocation and correctness work on hot paths: the Annex-B writer in demux_sink allocated and freed a whole-frame Vec per frame, which for a UHD title is ~200,000 allocations over the mmap threshold plus the page faults to first-touch each one; it now reuses a buffer on the writer, and still takes the NAL prefix width from the configuration record rather than assuming four. Six findings whose real fix lives in a consumer crate are recorded for re-filing rather than patched here. |
||
|
Matthew Jackson
|
4fcd28b487 |
Parse MKV lacing, route by real TrackNumber, honour NAL length size and edit lists
Four conformance defects in the read paths, two of them silent corruption. **Lacing was ignored entirely.** RFC 9559 §10.2 defines Xiph, EBML and fixed-size lacing, where one Block carries several frames; the reader took the Block payload verbatim, so a laced Block became a single "frame" consisting of a lacing header followed by concatenated frames — garbage to the codec parser, no error. Audio tracks from other muxers commonly use lacing, so an ordinary foreign MKV was silently mangled. All three modes are now parsed: Xiph 255-run sizes including the trailing-zero rule for exact multiples of 255, EBML unsigned first size plus SIGNED VINT deltas with the 2^((7*n)-1)-1 bias of §10.3.3, and fixed-size even division, with the last frame's size deduced from the remainder. Laced timestamps follow §10.3.5: the first frame takes the Block timestamp and the rest are spaced by the track's DefaultDuration, else BlockDuration/count, else shared with a warn. Parsing was chosen over refusing because refusal would leave freemkv unable to remux common foreign audio at all, and each mode is about fifteen lines. A malformed lacing header now raises a NEW code, E_MKV_LACING_INVALID = 9052, deliberately NOT MkvInvalid — because is_skippable_title_stub classifies MkvInvalid as a skippable nav stub, so reusing it would have recreated the exact conflation that is still open as a separate finding. A test asserts the new code is not skippable. **TrackNumber was assumed to be 1..N in TrackEntry order.** RFC 9559 §5.1.4.1.1 only requires it to be non-zero and unique, so sparse or unordered numbers are legal. Block routing and codec_private both computed track + 1. A real TrackNumber map is now built, recorded only for TrackEntries that yield a stream so dropped track types no longer shift the mapping. Verified red here independently, and the failure mode is worse than mis-routing: with track + 1 restored, a buttons track's payload was attributed to the AUDIO stream — wrong payload into the wrong codec parser. **The NAL length prefix was hardcoded to 4 bytes.** lengthSizeMinusOne lives in avcC byte 4 and hvcC byte 21 (ISO/IEC 14496-15 §5.3.3.1.2, §8.3.3.1.2) and was never read, so a source declaring 1- or 2-byte prefixes had its raw prefixed bytes emitted verbatim with no start codes. All four conversion sites now derive the width from the track's own configuration record. **Edit lists were ignored.** No edts/elst was parsed, so the presentation timeline an edit list defines (ISO/IEC 14496-12 §8.6.5/§8.6.6) was dropped — which is how encoder delay is normally expressed. Leading empty edits and the first media edit's media_time are now applied to both dts and pts, with the movie vs media timescale distinction respected. A list needing more than a constant shift applies the leading edit and warns rather than presenting the result as faithful. 17 tests. I reproduced the lacing mutant independently: returning the body whole kills five of them, including the exact-payload and malformed-header cases. Still open and deliberately untouched: the MkvInvalid / is_skippable_title_stub conflation across ~20 reader raise sites. It is a cross-cutting error.rs change and E_MKV_LACING_INVALID is the template for it. |
||
|
Matthew Jackson
|
9527bc1e13 |
Anchor forensic segments to the forensic clip, and stop CPS branching on resolve order
Two correctness defects in AACS 2.1 / FMTS key-map resolution, both order- or anchor-dependent, and both able to abort a whole disc or silently garble it. **The single-CPS short-circuit depended on which title resolved first.** `pool_len` counted the WHOLE unit-key pool, and resolve_fmts_key_map appends the disc's forensic index keys to that same caller-owned pool. The count was captured before THIS call's FMTS branch but not before earlier titles', so once any forensic title resolved, every later title saw a pool larger than one and fell into multi-CPS sampling — 8 random reads per extent, and a whole-disc DecryptFailed if no pooled key opened a menu extent's samples. A disc that ripped fine when a non-forensic playlist sorted first failed when a forensic one did. Forensic keys are now tagged FMTS_POOL_TAG_BASE = 1 << 24 and the short-circuit asks single_base_key_slot(), which excludes them. The old 1000 tag was NOT kept, and the reasoning is worth recording: base CPS ids are Unit_Key_RO.inf position + 1 and that count is a BE16, so 1000 sits inside a genuinely reachable id space. 1<<24 cannot collide. The id field is cosmetic — decrypt.rs indexes the pool by slot and reads only the key — so widening the tag is safe. **Forensic segment SPNs were anchored to the wrong clip.** They live in the forensic feature clip's byte space, but were mapped through clip_byte_to_lba(&title.extents, ..), which treats byte 0 as the start of the title's FIRST extent. Any playlist not beginning with the forensic clip mapped every segment to the wrong LBA: either the anchor probe sampled the wrong clip and the whole-disc resolve aborted with FmtsKeyMissing, or — worse — a forensic index key was applied to non-forensic sectors while the real forensic units kept the base key, giving silently garbled output with no error at all. The correct anchor turns out to be a DISC fact, not title data: an AACS 2.1 disc names its forensic feature BDMV/STREAM/<clip>.fmts, and carries one IndividualSegment.tbl, so the SPNs are in that one clip's byte space. A new forensic_clip_extents() finds the unique .fmts in the already-walked UDF tree, and those extents now drive the segment arithmetic, the addressability filter, and the index probe — whose title parameter is gone, since its reads were mis-anchored too. "Does this title carry forensic content" is now "does it read the forensic clip's sectors" rather than "do the segment bytes land somewhere in the concatenation". Where the clip is NOT identifiable — no .fmts, or several, making the SPN space ambiguous — on a disc that does carry a non-empty table, the resolve now fails loud with FmtsKeyMissing rather than guessing an anchor. That is a deliberate behaviour change: a hypothetical disc with two .fmts clips hard-fails where it previously produced a possibly-wrong map. Failing loud beats silently garbled output, and inventing an anchor was not acceptable. This site had been flagged independently three times — by the agent that added the FMTS per-disc memo, by the round-4 correctness lens with a concrete scenario, and by the round-4 conformance pass. Verified red here independently: reverting single_base_key_slot to count the whole pool fails both new tests. The agent's own evidence was probe_reads 48 vs 40 (the 8 extra sampling reads) and an E7013 DecryptFailed whole-disc abort, and for the anchor an E7026 FmtsKeyMissing on a [trailer, forensic] extent list. All six pre-existing FMTS tests pass unchanged through the new anchor, including the exact-cost assertions (40 probe reads, one key-service call per disc, one UDF walk for 60 titles), so the round-3 memoisation wins are intact. |
||
|
Matthew Jackson
|
58bdb42f8e |
Group whole E-AC-3 frame sets, and keep short reads unit-aligned
Two defects in fixes landed the same day, both found by round 4 auditing round 3's work rather than trusting it. **The E-AC-3 grouping ignored substreamid, so the timeline still doubled.** Confirmed independently by the correctness and conformance lenses and verified by hand: substreamid appeared only in test helpers, never in the production path. Per ETSI TS 102 366 (A/52) Annex E a frame set is independent substream 0 — mandatory, always first — with its dependents, then the OPTIONAL additional independent substreams 1..7 with theirs, all covering the same time period. Treating an additional independent substream as a new access unit advanced the clock a second time for the same 32 ms, which is exactly the doubling the grouping fix existed to prevent. No fixture caught it because every fixture used substreamid 0. is_dependent_substream becomes substream_role -> Starts | Extends: strmtyp 1 extends; strmtyp 0/2 with substreamid != 0 now extends (this was the bug); strmtyp 0/2 with substreamid == 0, legacy AC-3, and reserved strmtyp 3 start. Reserved 3 starts regardless of its id bits, because its BSI layout is undefined so those bits cannot be trusted — an unknown frame is neither merged into an unrelated programme nor silently discarded. The frame set stays ONE sample rather than being split into a separate track for the associated service, and the reasoning is in the module doc: a substream numbered 1..7 with no substream 0 is not conforming, so extracting one would mean renumbering ids and rebuilding frame sets — a transcode, not a remux. Programme selection is the player's job. A stream joined mid-frame-set (first sync is substreamid 3) is skipped with a debug and resyncs at the next id-0, mirroring the orphan-dependent rule: its mandatory id-0 substream was never seen, so it is neither decodable alone nor timeable. MAX_AC3_BUF 128 KiB -> 1 MiB, because an AU is now a whole frame set: worst case 8 independent x 9 substreams x 8192 B = 576 KiB, which the old cap could have dropped mid-hold. **The forced probe's two round-3 fixes cancelled each other.** CHUNK_SECTORS = 1023 exists (with a const assert) so every read starts on a 3-sector AACS aligned-unit boundary; the short-read fix advanced by actual bytes, making the advance a non-multiple of 3. Every later read was then misaligned, DecryptingSectorSource refused it before reading, the stop became ReadFailed, and no verdict was asserted — so content-based forced detection silently fell back to the vendor label on exactly the encrypted discs the 1023 change was written for. A partially-satisfied read now advances only by whole aligned units and re-reads the <=2 residue sectors from the next boundary, feeding only the aligned prefix so nothing is double-fed and no partial unit reaches the parsers. A read that fully satisfies its request still advances by all of it. When less than one aligned unit comes back the bytes are fed and the same LBA is retried twice before stopping, so a starved source cannot spin — verified by raising the retry limit and watching the test hang. Verified red independently here: reverting substream_role to strmtyp-only fails eac3_additional_independent_substream_stays_in_the_frame_set (6 access units where 3 are correct — the doubling, literally) and the mid-frame-set resync test. Reported, not fixed: dec3_box still hardcodes num_ind_sub - 1 = 0 and num_dep_sub = 0, so it under-declares any stream carrying additional independent or dependent substreams now that frame sets arrive whole. DolbyConfig has no fields for either; a real fix needs the parser to surface observed substream counts. That file is another lens's this round. Unverified: no real multi-programme DD+ stream exists here, so defect 1 rests on synthetic Annex-E fixtures. The retail DD+ check (No Time to Die, all substreamid 0) confirms single-programme discs are unaffected. |
||
|
Matthew Jackson
|
013881ac06 |
Restore the MP4 conformance fixes I clobbered while landing another agent's work
detect_rate's nearest-match fix, the colr HLG/BT.470 fix and their four tests were silently reverted. Cause: the r3fix-silent worktree was cut BEFORE the conformance commit landed, and I landed its work by copying whole files into the main tree. mp4/mod.rs was in both agents' file sets, so silent's copy — built on the older base — overwrote the conformance changes wholesale. The gate stayed green throughout, because reverting a fix and its tests together is perfectly consistent. Re-applied the conformance commit's diff for that file with a three-way merge; both agents' changes to mp4/mod.rs now coexist (final_report, UndescribableAudio and the max(track_id) id fix are all still present alongside RATE_TOLERANCE_FPS and the colr resolver delegation). Only mp4/mod.rs was affected. mp4/audio.rs and mkv.rs were in no other agent's file set and were intact. Process lesson, recorded because I would otherwise repeat it: NEVER land a parallel agent's work by copying whole files, when its worktree was cut at an older base than HEAD. Apply its DIFF (git apply -3), or rebase its worktree first. Copying files silently discards anything committed to those files in the interim, and no test can catch it because the tests disappear with the code. |
||
|
Matthew Jackson
|
5f8dc392c0 |
Sweep the pinned toolchain to Rust 1.97
The Windows UI needs current winsafe, whose real minimum is 1.89 (its manifest under-declares 1.87 while it uses NonNull::from_ref). Rather than stop at the minimum, this goes to current stable and fixes what that costs. The counter-intuitive result: 1.97 is CHEAPER than 1.89. libfreemkv had 54 clippy errors at 1.89 and 6 at 1.97, because clippy tightened the noisy collapsible_if lint in between. Stopping at the minimum would have been the most expensive choice available. Roughly 47 lints across the eight repos, the large majority auto-fixed: libfreemkv 6, freemkv-engine 14, bdemu 8, freemkv-keysources 7, autorip 6, freemkv-unlock 3, freemkv-i18n 3. The hand-fixed ones are a descending sort to sort_by_key(Reverse), four manual checked-division sites, a loop counter replaced by enumerate, and a loop whose first let-else became a while-let. Worth recording for whoever bumps next: clippy is MSRV-AWARE. Those 54 lints only appear once the crate DECLARES 1.89 or later, because let-chains become available. A bare `cargo +1.89 clippy` against a manifest still pinned at 1.87 reports clean and is meaningless — gate with the real precommit script, which is also the only thing that covers build scripts. The pin still sits below the Mac default, so it keeps doing its job: catching lint drift locally before CI sees it. |
||
|
Matthew Jackson
|
a32373ff40 |
Fix fifteen defects across perf, resource, panics and key hygiene
All 21 findings held up under verification; 15 fixed here, 6 deferred to files another agent held this round, 0 rejected. **A defect in my own round-2 probe fix.** CHUNK_SECTORS was 1024, and 1024 % 3 == 1 — verified — so every chunk after the first was misaligned against the 6144-byte AACS aligned unit and would be REJECTED by DecryptingSectorSource's alignment gate. On an encrypted disc the forced-subtitle probe I added last round would have read almost nothing past its first chunk. Now 1023 sectors (341 aligned units) with a const assertion that fails the build if it stops dividing, plus set_unit_base per extent so the source's gate is anchored where the extent actually starts. **The same probe skipped sectors on a short read**, advancing by the REQUESTED count rather than the bytes actually returned, so a partial read silently left a gap in the middle of the evidence. It now advances by n/SECTOR_BYTES and clamps n to the buffer. **Its cache key omitted the PGS PID set**, so a playlist declaring an extra subtitle PID got another playlist's verdict for a track that had never been probed. And the key was the whole extent list, so partial clip sharing missed entirely. Both fixed by keying (start_lba, sector_count, pid) — and per-extent keying was shown SOUND rather than assumed: ForcedTracker is two monotone booleans, so per-extent evidence composes by field-wise OR, order- and grouping-independently. Making that honest required per-extent demux state, so an extent's evidence comes only from its own bytes, and memoising only extents whose read reached a designed stop. **A reachable panic in the timeline.** mkvstream::parse_block accepts a TimestampScale up to i64::MAX, so a video frame can set high_ns = i64::MAX and the next passive frame panicked adding the backstep. In release it wrapped negative instead, firing the straggler clamp for essentially every passive frame — audio and subtitles rewritten onto the wrong point of the output timeline. All four sites saturate. **A public constructor divided by zero**: PrefetchedSectorSource::new_with_events with unit_align == 0. Now InvalidInput, matching its batch_sectors sibling. **Two Debug impls printed key material.** DiscInputs (volume_id, mkb, unit_key_ro, samples) and UnitKeyFile both derived Debug. Nothing logs them today — fixed as prevention, because the next tracing::debug! someone adds is the leak. A doc claim that DiscInputs "contains no secrets" was false and is corrected. **An env-var multiply could overflow** in file_sector_source; now bounded at 64 GiB like its writeback sibling, with the parse split out so the bound is testable without touching process env. **The mp4 demuxer allowed one sample per file byte** — ~64x RAM amplification. Now file_len/16, since only vide/soun tracks are indexed and the shortest legal AC-3 frame is 128 bytes. **Two pipeline concurrency defects**: a consumer apply() error was invisible to the producer, and abandon/finalise had a TOCTOU where a caller could report an unfinalised output. Both fixed with compare-exchange state rather than a bool. **Two per-frame copies removed**, both MEASURED rather than reasoned: the AU assembler now hands its allocation to the frame (same pointer, unchanged capacity, proven by asserting the pointer) and tsmux reuses one Annex-B buffer across frames. Both keep capacity deliberately — a naive split_off would have cost more than it saved. **A comment pointed at the wrong file** for a mirrored constant; the mirror is now compiler-enforced with a const assertion converting 90 kHz ticks to ns, so drift fails the build. Deferred to another agent's files, all confirmed: detect_rate's fractional-twin snap, the mp4 reserve's u32 truncation, round_up_grain's overflow, the quadratic base-key gap fill, and MkvStream's frame cap counting frames rather than bytes. Every fix verified red by reverting it. Also noted for later: DecodeSampleSet still derives Debug over multi-MB of on-disc ciphertext. |
||
|
Matthew Jackson
|
3efa6211f3 |
Make six silent mux failures observable
All six confirmed against the code. The governing rule this cluster serves: a
lossy or degraded outcome is never silent, because a corrupt rip the user does
not know about is the worst failure available.
**A 3D MKV re-mux silently lost one eye.** The BlockGroup read path had arms for
BLOCK / BLOCK_DURATION / REFERENCE_BLOCK only, so BLOCK_ADDITIONS fell into the
skip arm — while the writer does emit BlockAdditions > BlockMore > BlockAdditional
for the MVC dependent view. Reconstruction was judged out of scope and the
reasoning is recorded: PesFrame has no side-payload field and the header parser
never reads BlockAdditionMapping, so there is no dependent-view track to route the
AU to. Instead the loss is now LOUD — counted in bytes and events, warned once,
and surfaced through MkvStream's errors()/lost_bytes(), which the driver already
samples into MuxOutcome. One detail in the finding was wrong and is corrected: the
re-mux does NOT still advertise the mvcC mapping, because the header parser
ignores that element, so the output is a plain 2D H.264 track.
**An all-titles rip silently skipped real titles.** The header-buffer-cap
overflow returned Error::MkvInvalid, and is_skippable_title_stub matches exactly
E_MKV_INVALID | E_CSS_KEY_MISSING — verified here — so a 512 MiB-of-frames title
was classified as an empty nav/menu PGC stub and dropped. It now has its own
E9051 / MuxHeaderBufferExceeded { bytes }, outside the skippable set.
**The public pre-mux report contradicted the file.** Mp4Sink::finish() drops an
audio track it cannot describe, which I chose last round over failing an export
whose video is fine — but mp4_fit_report still listed that stream as included, so
the application's plan and the actual output disagreed. Fixed at both levels:
Mp4SkipReason is now non_exhaustive with NoSamples and UndescribableAudio,
Mp4Sink::final_report() describes the FILE rather than the plan, and for the
boxed dyn Stream path a defaulted Stream::undelivered_streams() carries the
information out to MuxOutcome::undelivered_streams with a driver-side warn.
**MP4 track ids could collide.** ids were assigned before the retain that drops
sample-less tracks, while next_id came from the post-retain count, so [1,3]
yielded next_id 3. Now max(track_id) + 1, saturating.
**Stream selection silently skipped its codec_privates prune** when the lists were
not the same length — but codec_privates is consumed POSITIONALLY and trailing
extras are documented as benign, so the length-equality guard was itself the bug.
The prune now runs unconditionally by index.
**The m2ts_mux scaffolding armed params_written on both the absent and the
unparseable codec_private arms** — the same defect already fixed in tsmux.rs.
Split into params_attempted (a latch, since retrying identical bytes cannot help)
and params_emitted, with a warn on each failure arm and an accessor so the
eventual wiring and its test can observe it.
Each fix verified red by mutating back to the prior behaviour: errors() 0 vs 1,
E6008 vs E9051, final_report [0,1] vs [0], next_track_id 3 vs [1,3], and the
selection prune resolving index 1 to the wrong track's record.
API surface deliberately widened: MuxOutcome gains a public field and Mp4Sink
becomes public. Nothing in-repo breaks. Note a behaviour change on the mkv://
input path — a 3D re-mux now reports non-zero loss, so a consumer treating
errors > 0 as disc damage will trip on it. That is intended: the outcome IS
degraded.
|
||
|
Matthew Jackson
|
9ad68dd092 |
Fix four MP4 conformance defects against the standards
**dec3 declared a 0 kbit/s AC-3 substream.** parse_dolby routes bsid < 11 to parse_ac3, which leaves data_rate_kbps = 0 and keeps the AC-3 bsid, yet dolby_sample_entry wrapped that config in ec-3/dec3 for any Codec::Ac3Plus track. ETSI TS 102 366 Annex F.4 assigns ac-3/dac3 to an AC-3 bitstream and F.6 assigns ec-3/dec3 to an Enhanced AC-3 one, so the entry now follows the SYNCFRAME that was actually parsed, not the playlist's codec label. Computing an AC-3 data rate and keeping ec-3 was rejected: it fixes one field while bit_stream_identification and num_dep_sub keep misdescribing the stream. The bsid threshold is hoisted into one constant so parser and entry-chooser cannot drift. Adjacent defect fixed in the same box: data_rate is 13 bits from a u16 source, so push's mask WRAPPED anything above 8191 (9000 became 808); it now saturates. **colr tagged HLG as PQ, and PAL as BT.601.** video_colr carried a second, drifted copy of the ColorSpace-to-CICP map: transfer 16 (PQ) for every BT.2020 stream with no HdrFormat override, and 6 (BT.601) for Bt470bg. Per ITU-T H.273 Table 3, HLG is 18 and BT.470-6 System B/G is 5 — and mkv::cicp_for_video already got both right. video_colr now delegates to that shared resolver, so the duplicated table is gone and cannot drift again. It keeps only its own decision about WHETHER to emit the box, since an absent colr and an all-unspecified colr mean the same thing per ISO/IEC 14496-12. **Exact 24.000 / 30.000 / 60.000 fps was declared 23.976 / 29.97 / 59.94.** detect_rate took the FIRST STD_RATES entry within 0.5 fps, and every 1000/1001 entry precedes its integer twin 0.024 fps away — a 0.1% error across the whole track's mdhd and stts. Fixed as nearest-wins rather than by reordering the table: reordering fixes today's table and re-breaks the moment someone appends a rate, while nearest-wins is order-independent. Verified red here independently by reverting to first-match, which fails exactly the two timing tests. **ddts MultiAssetFlag was set from has_extension.** In the DTSSpecificBox (ETSI TS 102 114) that flag signals more than one audio ASSET. A DTS-HD MA/HRA track is one asset whose extension substream carries the XLL/XBR component, so setting it from "an EXSS sync follows the core" sent a parser looking for a second asset descriptor while StreamConstruction simultaneously said there was no extension — the box contradicting itself. This module parses the core header only and never reads the EXSS asset table, so 0 is the only honest declaration. A StreamConstruction index for core+EXSS was deliberately NOT invented: that field is a table lookup that could not be confirmed against the standard, and a wrong index is worse than an under-declaration. DTS_AMODE_LAYOUT's masks were independently re-derived against ETSI TS 102 114 §5.3.1 and all 16 are CORRECT — only three adjacent comments were wrong (the AMODE 2/3/4 annotations were rotated by one, and AMODE 9's said "5.1 with LFE" when 0x0007 is the 5.0 mask and LFE is OR'd in separately). Comments corrected. Every one of the eight new tests decodes the field back OUT of the emitted bytes — data_rate from the dec3 body's leading 13 bits, MultiAssetFlag from bit 48 of the ddts tail, colr from the nclx payload inside a real stsd, and the frame rate from mdhd.timescale plus stts.sample_delta of a fully muxed MP4 — rather than restating arithmetic. This audit has already caught one of my own tests doing the latter. NOT fixed, root-caused and recorded at the site instead: an A_PCM/INT/BIG track ships with no BitDepth, which the Matroska Codec Specifications make a MUST. The width exists on disc (BD LPCM signals it in the ES header byte 3, DVD in the IFO audio attribute byte 1) but neither source reaches MkvTrack::audio, and the fix needs a new AudioStream member plus a deferred setter in files another agent held this round. Guessing 16 was rejected — it would confidently misdecode every 24-bit disc — as was refusing the track, which would regress the 16-bit majority that currently plays by accident. |
||
|
Matthew Jackson
|
13897e14f0 |
Resolve the forensic key map once per disc, not once per playlist
resolve_content_key_map loops every title into resolve_mux_key_map, which called resolve_fmts_key_map FIRST — before the CpsUnitCache — and on an FMTS disc returned immediately. So every playlist re-derived facts that belong to the DISC: a full UDF walk plus /AACS/IndividualSegment.tbl, and on an FMTS disc the anchor probe, the 32-index phase probe, and a fetch.fmts_indexes round trip. On a 60-playlist disc, measured on a synthetic fixture: 840 -> 14 metadata reads, 2,400 -> 40 probe reads, and 60 -> 1 key-service calls. Worst case before was up to 256 probe reads and 32 key-service calls per title. The 60 redundant key-service round trips are a strong candidate for the keyserver storm seen in the field. Two memos behind a pub(crate) DiscKeyCache. The table memo (UDF walk + tbl parse) is disc-invariant outright — nothing in that path mentions the title — and runs on EVERY disc, so a plain BD benefits too. Only the deterministic negatives are memoised as "not FMTS"; a DiscRead fault propagates uncached so a later title retries. A blind once-per-disc hoist of the PROBES was rejected as unsafe, and this is the load-bearing reasoning: the title enters through clip_byte_to_lba, which decides which segments are addressable and which LBA every probed clip byte reads from, so two titles with different extent lists probe different physical bytes. A hoist would serve title B an answer derived from title A's media and could silently turn a per-title FmtsKeyMissing into a success. The extent list is the ONLY per-title input, so keying on it is exactly sufficient — matching the ForcedProbeCache precedent. Result-identity was proved, not assumed: NEITHER probe reads the key pool. Verified here independently — probe_fmts_index_keys takes no keys parameter at all; index keys come from `fetch`, and the anchor's reply feeds the phase probe. So the pool's growth across titles, the one thing that does change between calls, cannot move a memoised value, and the result is order-independent. A test resolves three titles through a shared memo and through fresh memos and asserts both the per-title ranges and the final key pool (keys, slots, order) are identical. Not memoised, deliberately: fail-loud FmtsKeyMissing, and any run where an index hit a read fault — that is a property of a transient drive fault, not of the extents, and caching it would spread one bad read across 59 playlists. A fully-memoised title now does zero I/O, which made the old in-loop halt polls unreachable for it, so a check_halt on entry was added with a test that cancels after warming the memos. Also corrects my own overstatement from last round: the CpsUnitCache doc now says plainly that on an FMTS disc it removes NO reads, because this function returns before the extent loop ever runs. Five mutations, all verified red. Pre-existing bug flagged but not fixed: filter_addressable_segments only checks that a segment's START byte maps to some LBA in the title, so a play-all playlist can pass the filter while mapping segment bytes into the wrong clip, whose anchor then returns empty and aborts the sweep. |
||
|
Matthew Jackson
|
b4bf0daa82 |
Group E-AC-3 dependent substreams into one access unit
The AC-3 parser's own module doc stated the assumption: "AC3 frames are self-contained and always start with syncword 0x0B77". True for legacy AC-3, false for E-AC-3 above 5.1. Per ETSI TS 102 366 (A/52) Annex E, byte 2 of an E-AC-3 syncframe is strmtyp(2) | substreamid(3) | frmsiz[10:8], and an access unit is one INDEPENDENT substream plus every DEPENDENT substream that follows it until the next independent one. The parser emitted one PES frame per syncframe, so a decoder saw each dependent substream as a standalone frame with no parent — including the AC-3-core + E-AC-3-dependent form Blu-ray uses for Dolby Digital Plus. The extra channels were lost and the timeline ran at 2x. The bit position is cross-checked against code already in the tree: the existing frmsiz parse takes byte2 & 0x07 as its high bits, which is only consistent with strmtyp occupying byte2's top two bits. Legacy AC-3 is excluded by bsid < 11, where byte 2 is crc1 and reading strmtyp there would be nonsense. Reserved strmtyp 3 is treated as INDEPENDENT so an unknown type starts a fresh AU rather than merging into an unrelated one. The AU carries the INDEPENDENT substream's PTS, and only the independent substream advances the clock — dependents cover the same time period and add zero duration. That is what removes the doubled timeline. A trailing AU that can still grow is HELD across the PES boundary, because the boundary is unknowable until the next independent sync; the whole AU is re-scanned next call, so there is no shift and no double-count in the loss tally. Plain AC-3 is never held, which keeps DVD/AC-3 latency and behaviour unchanged. A latent pre-existing bug surfaced while testing this: a new PES's PTS was re-stamping an AU that began in an earlier PES, a constant one-frame shift. Fixed with a PtsAnchor so a PES timestamp applies to the first AU that STARTS in that PES's own bytes, while a genuine PTS jump is still adopted. Nine tests. Verified red against five mutations, each killing a specific set: reverting to the pre-fix behaviour kills 8 while plain_ac3_frames_are_not_grouped_or_delayed SURVIVES as the no-regression guard — reproduced independently here. Stamping the dependent's PTS kills 6; not holding across PES kills 4; holding plain AC-3 too kills 15; neutering the PTS anchor kills exactly the 2 split-across-PES timing tests. Three sibling defects found and deliberately NOT fixed, all in mp4/audio.rs: dec3 hardcodes num_dep_sub = 0 (and a nonzero value changes the box LAYOUT, not just a field, per Annex F/G); parse_eac3 ignores strmtyp/substreamid entirely; and a 7.1 DD+ track is still labelled 5.1 because the channel count comes from the independent substream while the extra channels are described by the dependent's chanmap, which nothing parses. Not verified: no real E-AC-3-with-dependents sample exists here, so all evidence is synthetic frames plus the spec layout. The multi-independent-substream case (num_ind_sub > 1, main + associated audio in one PID) is deliberately treated as one AU per independent substream and is untested. |
||
|
Matthew Jackson
|
ef36b452ad |
Fix three defects in last round's own fixes
Round 3 audited the round-1/2 fix commits rather than trusting them, and found three defects in that new code. This is why the pin moves each round. 1. LICENCE REGRESSION, and it was mine. Reverting the "distinguish a failed key source" commit also restored a verbatim reference-decoder table citation in src/mux/codec/dts.rs, because both changes were in that one commit. The MIT licence cleanup was silently undone at HEAD and nothing caught it. The citation is replaced with ETSI TS 102 114 §5.3.1 again, and — more importantly — the rule now lives in the leak gate instead of in my memory. scan-secrets.sh gains LICENCE_RE, which flags ff_dca*, dcadec, l-smash, libav*, and bare ffmpeg/FFmpeg as REF-IMPL-CITATION. `no ffmpeg` is explicitly allowed via negative lookbehind: stating what this project does NOT depend on carries no risk and is a genuine selling point. Verified by re-introducing the citation (gate fails) and removing it (gate clean). 2. TsMuxer armed params_written even when the avcC/hvcC parser returned None, so a track whose codec_private exists but will not parse was muxed to BD-TS with no VPS/SPS/PPS ever emitted — undecodable video, reported as success, with no log line. Last round's fix corrected WHICH parser is used and left this half untouched. Arming the flag is still right (retrying identical bytes cannot succeed) but it is no longer silent: it now warns with the track, codec and codec_private length. 3. test_aes_cbc_roundtrip defined a LOCAL fn aes_cbc_encrypt that SHADOWED the production primitive, so it round-tripped a copy of the algorithm against itself and never touched crypto::aes_cbc_encrypt — the function this cycle added. Any mutation to the shipped code passed it. The shadow is deleted and the test now calls the real primitive; verified by mutating crypto::aes_cbc_encrypt, which now fails it and previously would not have. |
||
|
Matthew Jackson
|
38aa895038 |
Memoise multi-CPS key-map sampling per extent, not per title
resolve_content_key_map calls resolve_mux_key_map once per title, and on a
multi-CPS disc that path issues 8 random single-unit reads per extent. A disc's
playlists overwhelmingly reference the same few clips — main feature, play-all,
per-chapter and seamless-branch variants — so the same physical extents were
re-sampled from the drive once per playlist. On a 60-playlist / 15-clip disc
that is ~2,400 non-sequential 6144-byte reads, roughly 8 minutes of pure seeking
at 200 ms per seek, before the mux starts. Now ~600 reads.
Keyed per EXTENT — (format, start_lba, sector_count) — rather than per title's
whole extent list, which is finer-grained than the forced-subtitle probe's cache
and strictly better here: a play-all playlist sharing 4 of 5 extents with the
main feature still hits on those 4.
Why a cached pool index is provably identical to a recomputed one, verified
rather than assumed:
* `pick` iterates the pool IN ORDER and returns the FIRST index whose key
decrypts a sample to clean.
* The pool is APPEND-ONLY. Checked across the whole crate: only `push`, with no
insert/remove/clear/retain/sort/dedup/truncate/drain/swap/reverse anywhere.
So appended keys can only land AFTER a matched index, and the first match for
the same samples cannot shift.
* The samples are a pure function of the three values in the key, read from
read-only optical media.
Two outcomes are deliberately NOT cached, which is what makes this safe rather
than merely faster:
* the inherited index (`None if samples.is_empty() => last_idx`) is per-TITLE
state, not a property of the extent — caching it would let one title's
carry-in index leak into another title's clear extent, i.e. a WRONG key;
* the fail-loud DecryptFailed verdict, so a retry after a key source banks the
missing key re-samples instead of inheriting a stale answer.
Halt is still polled before the cache lookup, so cancellation is unchanged.
resolve_mux_key_map keeps its exact signature and delegates with a fresh cache,
so there is no public API change. ContentFormat gains Eq + Hash (additive).
Four tests, and the two mutants that matter both verified red: disabling the
cache short-circuit fails the hit and recompute-equivalence tests, and wrongly
caching the inherited index fails multi_cps_inherited_index_is_not_cached.
|
||
|
Matthew Jackson
|
e3676e7cdf |
Build BlockGroups in memory so the hot path never seeks
Every BlockGroup frame back-patched its element size via ebml::end_master, which
does two stream_position() calls and two real seeks. BufWriter does not override
Seek::stream_position, so each position query is seek(Current(0)) = flush_buf +
lseek — and each of those flushed the 4 MiB BufWriter while every position-moving
seek reset WritebackPipeline::last_flush_pos. The buffer never got to do its job.
An MPEG-2 title takes this path for EVERY frame (the parser stamps a per-frame
duration, so I, P and B all become BlockGroups) — roughly 350,000 per feature.
A BlockGroup's size is knowable before writing, so there is no need to back-patch
at all. New seek-free twins start_master_buf / end_master_buf patch a placeholder
by buffer INDEX instead of file offset, and build_block_group assembles the whole
element into a persistent buffer that write_block_group and its MVC sibling take,
fill, write once, and hand back — including on the error path — so the allocation
is made once rather than per frame.
Measured with a counting writer that, like BufWriter, does not override
stream_position, over 200 frames:
seek calls position-moving
plain before 912 451
plain after 112 51
MVC before 2516 1253
MVC after 116 53
Per-frame cost is now zero; the residual is the header, the per-cluster
back-patch and Cues. For 350k BlockGroups that is 1.4M seek calls removed on the
plain path, 4.2M on an MVC title.
end_master is deliberately NOT changed for its other callers. The Cluster master
genuinely streams — frames are appended to an open cluster over time, so its body
cannot be buffered without holding a whole cluster in memory — and the rest
(EBML header, Tracks, Info, Chapters, Cues) run once, not per frame.
Byte-identity is the safety property, and it is structural: end_master always
patches a FIXED-width 8-byte VINT, and the buffered pair writes and patches
exactly that same placeholder, so the encodings cannot differ. Verified by
capture-then-compare over 200 frames (17 keyframes / 183 non-keyframes, multiple
clusters, BlockDuration present and absent, both reference branches) — output
byte-for-byte identical.
Three tests now pin it permanently: buffered vs seeking output byte-for-byte for
empty/tiny/multi-byte bodies, the same for NESTED masters (the MVC path nests
BlockAdditions > BlockMore, where an index-arithmetic slip would surface), and
end_master_buf erroring rather than panicking on a position outside the buffer.
All three verified red against a deliberately divergent placeholder width.
A note on verifying this kind of change: comparing emitted MKV across two
worktrees at different commits shows a spurious 14-byte difference, because
MuxingApp/WritingApp embed the build's git SHA twice. Compare at the same base.
|
||
|
Matthew Jackson
|
4ed245868e |
Revert "Distinguish a failed key source from one with no entry"
This reverts commit
|
||
|
Matthew Jackson
|
50f37462db |
Remove reference-decoder citations from a public MIT-licensed repo
This crate is MIT licensed. Comments citing another decoder's internal symbols and reproducing its tables verbatim create licence risk that no engineering benefit justifies, so every such citation is replaced with the primary source: ETSI TS 102 114 §5.3.1. Eleven sites across src/mux/codec/dts.rs and src/mux/mp4/audio.rs. The technical substance is unchanged in every case — the deficit-sample-count semantics, the reserved-field skips, the invalid LFF value, and the 16 legal AMODE codes are all spec facts and are now attributed as such. Four CHANGELOG entries that named a validator are reworded; the "no a reference decoder" dependency claim stays, since stating what this project does NOT depend on carries no risk. I initially argued this was a false positive on the grounds that the project's hygiene rules name internal infrastructure and reverse-engineering material, not open-source citations, and that a channel-count table from a standard is fact rather than expression. That reasoning missed the point: the exposure is MIT distributing text derived from GPL/LGPL sources, and that is the maintainer's risk to weigh, not mine. Reversed in full. |
||
|
Matthew Jackson
|
22a3e3fd01 |
Distinguish a failed key source from one with no entry
resolve_and_apply_traced collapsed `Ok(_) | Err(_)` into a single KeyNode::NoEntry step, so a key source that FAILED — server unreachable, keydb unreadable, malformed entry — was recorded identically to one that simply had no entry for this disc. The front-end renders that trace, so it told the operator their disc is not in the database when the real cause was a fixable infrastructure problem. drive_unit_keys and drive_fmts_indexes were refactored this cycle to preserve exactly this distinction; this path had not been. KeyNode gains a SourceFailed variant and the two arms are split. freemkv's trace renderer matches KeyNode exhaustively with no catch-all, so its arm is added in the same change — otherwise the consumer would not build. Also made ETSI TS 102 114 the primary authority for DTS_AMODE_COUNT's comment rather than a reference decoder internal symbol, and pointed it at this crate's own cross-checked DTS_AMODE_LAYOUT / DTS_AMODE_CH tables. A round-2 finding asked for every a reference decoder and a reference decoder citation in the DTS parser to be stripped as a public-repo hygiene violation. Rejected: the project's rules (scan-secrets.sh, CLAUDE.md) prohibit internal infrastructure references and reverse-engineering material, and a reference-decoder citation is neither. The AMODE channel-count table is a factual table from the standard, not expression copied from an implementation. Citing the spec plus a corroborating implementation is how a decodability gate should be justified. |