Update module github.com/Eyevinn/mp4ff to v0.57.0

[renovate]

This PR contains the following updates:

Package	Change	Age	Confidence
github.com/Eyevinn/mp4ff	v0.56.0 → v0.57.0

---

Release Notes

Eyevinn/mp4ff (github.com/Eyevinn/mp4ff)

v0.57.0

Compare Source

Added

• saiz versions 1 and 2 (16- and 32-bit sizes, ISO/IEC 14496-12:2026) are
  decoded and encoded, and SaizBox.AddSampleInfo picks the lowest version
  that fits, so samples with more than 39 subsamples can be encrypted.
  GenerateFtyp adds the saie brand for them and CheckBrands checks it

• Constants for ftyp/styp brands from ISOBMFF, MP4, CMAF, DASH and codec
  specifications (mp4.BrandIso6, mp4.BrandCmfc, mp4.BrandLmsg, ...), and
  HasCompatibleBrand on FtypBox and StypBox

• InitSegment.GenerateFtyp, InitSegment.SetFtyp and
  MediaSegment.GenerateStyp set ftyp/styp brands from the content: lowest
  isoN brand, codec brands, CMAF brands, and DASH msdh/msix/lmsg

• File.CheckBrands reports ftyp/styp brands that the content contradicts or
  that miss spec recommendations, as BrandIssue errors and warnings, also
  available as mp4ff-info -brands

• Experimental paint-model subtitle boxes: stpc/wvtc sample entries, and
  ttmn/vttn (no change) and ttmb (TTML body only) samples, also handled
  by mp4ff-subslister. The 4CCs are unregistered and may change

• Fragment.AppendFullSamples, TrunBox.AppendFullSamples,
  SampleAccessor.AppendSamples and AppendSampleRange extract samples into
  caller-provided storage, so reused slices avoid allocation. Implementations
  of SampleAccessor outside mp4ff must add the two methods

• Fragment.Samples iterates over the full samples of a track without
  allocating

• hevc.ParseSPSNALUnitWithVPS parses an SPS with a map of the VPSs it may
  refer to, so that a multilayer extension SPS gets the chroma format, picture
  size, conformance window and bit depths that it does not signal itself
  inherited from the rep_format() of its VPS. hevc.ParseSPSNALUnit is
  unchanged and leaves those values unset, since the SPS payload can be parsed
  without them

• hevc.SPS exposes NuhLayerID, ExtOrMaxSubLayersMinus1,
  MultiLayerExtSpsFlag, UpdateRepFormatFlag, SpsRepFormatIdx,
  InferScalingListFlag and ScalingListRefLayerID, and hevc.RepFormat
  exposes SeparateColourPlaneFlag, ConformanceWindowFlag and
  ConformanceWindow

• RsotBox for the Redundant Sample Original Timing Box (rsot) of
  ISO/IEC 14496-12:2026 Section 8.8.18, a traf child that documents that the
  first sample of a track fragment is a copy of the previous sample and how long
  that sample has already run (ElapsedDuration), and that the fragment's last
  sample was truncated to fit and how long it was meant to last
  (OriginalDuration). Either duration may be signalled on its own, and
  CreateRsotBox sets the flags from the non-zero arguments. It is reachable as
  TrafBox.Rsot

• AudioSampleEntryBox.NormalizeQuickTime rewrites a QuickTime-shaped audio
  sample entry (sound sample description version 1 or 2, QuickTime residue in
  the version 0 reserved fields, or a wave-wrapped esds) to the plain ISO
  version 0 form with the esds as a direct child. Entries without a reachable
  esds, and wave boxes holding anything besides their recognized decoder-init
  atoms, are left untouched

• CloneBox returns a deep copy of a box (via an encode/decode round trip),
  so a box tree that shares parts with another structure can be mutated
  safely — for example handing InitProtect (which rewrites the stsd sample
  entry in place) an init segment built from a cached, shared trak

• Fragment.AddFullSamples adds many samples to a fragment at once. Adjacent
  sample slices (such as the output of GetFullSamples) are coalesced into
  runs that are added as mdat data parts, so the payload is never copied and
  contiguous input becomes a single part. Any data already in the mdat is
  closed into a part first, so samples can be added before or after with
  AddFullSample. Since the samples are referenced rather than copied, their
  buffers must be kept alive and unmodified until the fragment is encoded. The
  output is byte-identical to adding the samples one by one with
  AddFullSample

• mp4.Defragment and mp4.DefragmentTracks convert a fragmented file to a
  progressive one: sample tables (stts, ctts, stsc, stsz, stss, stco/co64)
  are synthesized from the fragment metadata while sample data and sample
  descriptions are preserved. Every track is rebased so that its first tfdt
  becomes media time zero, edit-list media times are shifted along, a final
  edit with zero segment duration (open-ended in a fragmented file, where the
  movie duration is unknown) is resolved to the remaining media duration so
  progressive readers do not apply it as a zero-length edit, and a track
  starting later than the earliest one keeps its presentation alignment
  through an empty edit; a trivial identity edit list is dropped.
  The output ftyp has major brand mp42, the lowest isoN brand the output
  needs, the codec brands, and the ISO/IEC 14496-15 brands of the input. Per-sample encryption auxiliary information (senc or
  saiz/saio; constant-IV full-sample encryption without such data passes
  through losslessly), unsupported edit lists, zero timescales, truncated
  byte ranges, and payloads larger than the input file are rejected

• mp4.Defragment and mp4.DefragmentTracks resolve overlapping fragments:
  when a fragment's tfdt re-declares an earlier decode time (a
  retransmission), the fragment appearing later in the file wins and the
  superseded samples are dropped at sample granularity, whether or not the
  re-sent bytes are identical. Every abandoned time range must be declared
  again by surviving fragments, so no declared content is ever silently
  dropped. Ambiguous overlaps fail closed: cuts inside a sample, abandoned
  time ranges that no surviving later fragment declares again, and
  overlapping files whose fragments use absolute base data offsets are
  rejected

• mp4.Defragment and mp4.DefragmentTracks accept hybrid files that carry
  progressive samples in the moov before the first fragment: the progressive
  samples come first with their chunk structure preserved (also when stco
  offsets are not monotone), and the fragment samples are appended.
  Fragments may supersede progressive samples under the same
  later-declarer-wins and coverage rules as fragment overlaps, and hostile
  sample tables are validated before any count-proportional allocation

• SttsBox.SampleDurations, CttsBox.CompositionTimeOffsets, and
  StssBox.SampleIsSync expand a whole sample table into one value per
  sample, validating the entries against the declared sample count

• mp4ff-defragment command line tool exposing the defragmentation with
  optional track selection

Changed

• SaizBox.DefaultSampleInfoSize is a uint32 and SaizBox.SampleInfo a
  []uint32, to hold the sizes of saiz versions 1 and 2
• CreateFtyp gives cmfc + [cmfc, iso6] and CreateStyp gives
  cmfs + [cmfs, cmff, cmfl, msdh]: the major brand is repeated and dash is dropped
• Minimum Go version bumped from 1.19 to 1.23, for slices and iter
• The commands and examples that create their own output file now buffer their
  writes, like WriteToFile already does. Encode makes about one write call
  per box, so writing straight to a file cost a syscall per box: affects
  mp4ff-encrypt, mp4ff-decrypt, mp4ff-crop, mp4ff-mvhevc and the
  add-sidx, combine-segs, initcreator, ivf-to-mp4, resegmenter and
  segmenter examples. Output bytes are unchanged
• New package documentation section "Writing files and segments efficiently"
  covering when to use a buffered Encode and when a reused
  bits.FixedSliceWriter with EncodeSW is better
• WriteToFile buffers its output instead of writing straight to the file.
  The many small boxes of a moof cost one write syscall each without it, while
  a large mdat payload is unaffected, since a write larger than the buffer goes
  directly to the file. Output bytes are unchanged
• SttsBox.GetDecodeTime returns an error instead of panicking with index out
  of range when the stts entries cover fewer samples than the requested sample
  number (a file whose stts and stsz disagree), or when called with sampleNr 0.
  Signature change: GetDecodeTime(sampleNr uint32) (decTime uint64, dur uint32)
  → GetDecodeTime(sampleNr uint32) (decTime uint64, dur uint32, err error).
  SttsBox.GetSampleNrAtTime returns an error instead of panicking for an
  stts box without entries
• TrunBox.GetFullSamples returns an error when the trun-declared sample
  sizes point outside the mdat data (a truncated or corrupt file), instead
  of panicking with index out of range. Signature change:
  GetFullSamples(...) []FullSample → GetFullSamples(...) ([]FullSample, error).
  Fragment.GetFullSamples (unchanged signature) propagates the error
• StscBox.GetChunk returns an error instead of panicking for chunk numbers
  that no stsc entry covers. Signature change:
  GetChunk(chunkNr uint32) Chunk → GetChunk(chunkNr uint32) (Chunk, error)
• Sample-table decode (stsz, stts, ctts, stco, co64, stss) bulk-reads
  each table with one slice-reader call instead of one call per entry, roughly
  halving moov parse time for long progressive files
• Fragment encryption and decryption expand the AES key once per run instead
  of once per sample (cenc) or per protected subsample range (cbcs), making
  fragment encryption roughly 40% (cenc) / 18% (cbcs) faster. A bad key
  length is now reported by NewFragmentEncryptor instead of by the first
  sample encryption

Fixed

• CreateEmptyTrak/AddEmptyTrack pick the media header from the handler
  type: subt, stpp and clcp tracks get sthd, and stpp gets handler subt
• The streaming SampleAccessor.GetSample, GetSampleRange and GetSamples
  allocated the sample size a trun claimed before checking it against the mdat
• Invalid NALU length sizes are rejected: the 3-byte size in lhvC and vvcC,
  anything but 1, 2 or 4 bytes when encoding hvcC, lhvC and vvcC
  (ErrInvalidLengthSize), and differing lhvC and hvcC sizes in a sample
  entry. mp4ff-nallister rejects VVC tracks without 4-byte lengths
• hevc.SplitNalusByLayerID panicked on 32-bit platforms for a length field of
  2^31 or more
• DecodeSgpd/DecodeSgpdSR did not bound the per-entry description_length
  against the bytes left in the sgpd box, and the sample group entry
  decoders size their slices from it. A 32-byte box declaring
  description_length = 0xfffffff0 for an alst entry allocated 4GB in
  DecodeAlstSampleGroupEntry. The length is now checked against what is left
  of the box before the entry is decoded
• avc.ParsePPSNALUnit skipped the six 4x4 picture scaling lists when
  pic_scaling_matrix_present_flag was set but transform_8x8_mode_flag was
  not, since the read loop was nested inside the 8x8 mode check. The lists are
  always present when a picture scaling matrix is signalled; only the two or
  six extra 8x8 lists depend on 8x8 mode, so such a PPS misparsed
  second_chroma_qp_index_offset and then failed the trailing-bits check. As a
  consequence, chroma_format_idc from the SPS is now only looked up when 8x8
  mode makes the list count depend on it, so a PPS whose SPS is not in the map
  no longer fails in the cases where the SPS is not needed
• HEVC slice header and PPS parsing accepted num_ref_idx_l0_active_minus1,
  num_ref_idx_l1_active_minus1 and their PPS defaults outside the range 0 to
  14 that the spec allows. The values were stored in a uint8 without a check,
  so a stream declaring 255 made parsePredWeightTable allocate a
  zero-length slice while its loop guard still ran, and parsing panicked with
  an index out of range. All four read sites now check the bound and return an
  error naming the syntax element, its value and the legal range
• hevc.ParseSPSNALUnit misparsed a non-base-layer SPS that uses the
  multilayer extension form, as MV-HEVC and SHVC enhancement layers do. Such an
  SPS signals sps_ext_or_max_sub_layers_minus1 instead of
  sps_max_sub_layers_minus1 and then omits profile_tier_level(), the chroma
  format, the picture size, the conformance window and the bit depths, and it
  carries sps_infer_scaling_list_flag. The parser always took the base-layer
  branch, so it read those absent fields from the following bits and typically
  failed with EOF. nuh_layer_id is now taken from the NAL unit header and
  the two forms are parsed accordingly
• rep_format() in the VPS extension dropped separate_colour_plane_vps_flag
  and the conformance window, and left the chroma format and bit depths unset
  for a rep_format() with chroma_and_bit_depth_vps_present_flag equal to
  zero instead of inferring them from the preceding one
• DecodeContainerChildren and DecodeContainerChildrenSR reported a wrapped
  parent size for a box too small to hold the fixed fields before its children,
  as dref, stsd, trep and meta can be. Such a box is now rejected up
  front, naming the declared and the minimum size
• DecodeEsdsSR derived its descriptor size from hdr.Size-12 without a length
  check, wrapping for a shorter box. Both the check and the size now measure the
  payload, which is also correct for an extended-size header
• DecodeContainerChildren now checks the child position against the
  container end before decoding each child, matching
  DecodeContainerChildrenSR. A container truncated exactly on a child
  boundary is reported instead of silently decoding short, and an empty
  container box no longer consumes its next sibling
• ftyp and styp boxes with a payload shorter than 8 bytes are rejected at
  decode instead of panicking. MajorBrand and MinorVersion slice the payload
  at fixed offsets, so a short box made Info panic, including in mp4ff-info
• IsSyncSampleFlags only inspected sample_depends_on and ignored
  sample_is_non_sync_sample, so flags that mark a sample non-sync while also
  saying sample_depends_on = 2 were reported as a sync sample. It now
  requires both indications to agree, like Sample.IsSync always has, and no
  longer accepts the reserved sample_depends_on = 3. Sample.IsSync is
  unchanged in behaviour and now delegates to it, so the two cannot drift
  apart again
• The mdat payload is now the concatenation of its data parts and its
  monolithic data, so the two can be combined in any order and are written in
  the order they were added. Adding a part on top of monolithic data no longer
  panics with "cannot mix sample parts with monolithic sample data", and data
  added with AddSampleData on top of data parts is no longer silently
  dropped at encode time, which produced a fragment whose trun declared
  samples that were never written. MdatBox.SetData replaces the whole
  payload and so now also drops any data parts
• NewSdtpEntry packed the sampleDependedOn argument into both two-bit
  dependency fields and dropped sampleDependsOn, so every entry it built
  carried a wrong sample_depends_on value. Note that entries built with the
  broken constructor re-encode byte-identically; only newly constructed
  entries change
• TrakBox.GetSampleData indexed its result slice with the absolute sample
  number instead of the offset within the requested interval, panicking with
  index out of range for any interval not starting at sample 1
• TrunBox.GetSampleInterval (and thereby Fragment.GetSampleInterval)
  returns an error instead of panicking when the requested interval points
  outside the mdat data
• File.CopySampleData returns an error instead of panicking when a chunk
  offset or sample size in the sample tables points outside the mdat data.
  This is the progressive-file counterpart of the fragmented trun case
• MdatBox.ReadData and MdatBox.CopyData rejected valid ranges ending at
  the very end of the mdat payload, so reading the last sample (or the full
  payload) failed with "invalid range provided". They now also reject a start
  before the mdat payload, which previously underflowed and panicked
• mp4ff-nallister, mp4ff-pslister and mp4ff-subslister panicked instead
  of reporting an error on a progressive file with sample tables pointing
  outside the mdat data
• CttsBox.GetCompositionTimeOffset panicked for a sample beyond the entries
  of the box, which a file with a ctts and stsz that disagree on the sample
  count triggers. Such a sample now gets composition time offset 0
• TrakBox.GetSampleData panicked for an sdtp with fewer entries than the
  track has samples. The sample flags from a missing sdtp entry are now left
  at their default
• StscBox.GetContainingChunks and StscBox.ChunkNrFromSampleNr panicked
  with index out of range for an stsc without entries (which is valid, and
  what init segments have) and divided by zero for an entry with
  samplesPerChunk == 0. Both now return an error
• StscBox.GetSampleDescriptionID and StszBox.GetSampleSize indexed outside
  their tables for out-of-range chunk and sample numbers
• StszBox.GetTotalSampleSize validated the sample interval against
  SampleNumber but indexed SampleSize, so a box where the two disagree
  panicked
• mp4ff-crop panicked on a file whose ctts, stts, stsc or stsz does not
  cover all samples of the track, and reported only the last of several
  cropping errors
• CryptSampleCenc, DecryptSampleCbcs and EncryptSampleCbcs panicked for
  subsample patterns whose byte counts reach outside the sample. Nothing ties
  the senc byte counts to the actual sample sizes, and big counts could also
  wrap in uint32. They now return an error
• DecryptFragment and DecryptFragmentWithKeys panicked when the senc box
  held subsample information for a different number of samples than the trun
  declares
• The nalu scanning functions in the avc and hevc packages trusted the
  4-byte nalu length fields of a sample, so a bad length field made them slice
  outside the sample and panic. A length field close to 2^32 also wrapped when
  added to the position, which defeated the check in avc.GetNalusFromSample
  and in GetAVCProtectRanges/GetHEVCProtectRanges. Affected are
  FindNaluTypes, FindNaluTypesUpToFirstVideoNALU/...Nalu,
  ContainsNaluType, IsIDRSample, IsRAPSample, HasParameterSets,
  GetParameterSets, GetNalusFromSample and ConvertSampleToByteStream
• avc.ContainsNaluType (and thereby avc.IsIDRSample) lacked the guard
  against samples shorter than 4 bytes that the other scanning functions have,
  so the loop limit underflowed and the first read went outside the sample
• DecodeStssSR and DecodeSttsSR returned a zero-filled table and no error
  when the reader held fewer bytes than the declared entry count; all
  sample-table decoders now report truncated input as an error, before
  allocating the table

---

Configuration

📅 Schedule: (UTC)

• Branch creation
  • At any time (no schedule defined)
• Automerge
  • At any time (no schedule defined)

🚦 Automerge: Enabled.

♻ Rebasing: Whenever PR is behind base branch, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.

---

• If you want to rebase/retry this PR, check this box

---

This PR was generated by Mend Renovate. View the repository job log.