The stride resolver's index.bdmv read, and the live UHD check, took a Stop for an unreadable
index and went on with a default stride. The capture and the scan's version read now end as
Halted, like their other reads.
Windows were planned per extent, and each small extent planned a read of itself, so a title of
thousands of interleaved units spent the 256 MiB budget on its first minutes. The windows are
now planned over the title's sectors end to end and mapped onto the extents they cross.
bdmt title and description text kept raw control characters from CDATA and plain bodies,
so a disc could put a terminal escape sequence into a shown title or a file name. It now
goes through the same filter as the Blu-ray di:name title.
The interleaved-unit walk polls halt before each read, passes a drive's Halted up instead of
taking it for an unreadable chain, gives the cell back whole on a short read rather than
reparsing the last pack, and cannot overflow on a cell spanning every sector.
A title's duration no longer subtracts the dropped leading angle pieces: the PGC time never
counted them, and the cell-sum fallback now counts an angle block once, as the chapter marks
do. Text-data chapter names are placed by the part-of-title table on the program each
chapter starts at; names that cannot be placed one per program are dropped. A text-data
name is looked for within its longest accepted length, not to the end of the table.
ContentFormat::MpegPs covered DVD VOBs, HD DVD EVOs and mpg:// files alike, so every stage
keyed on it ran on all three. The VOBU cell filter dropped every cell after the first of an
mpg:// VOB file (one extent, the whole file) and of an HD DVD .evo; the CSS crack, the forced
VobSub probe and the AC-3 diagnostic probe also ran on HD DVD titles.
The DVD scan now sets ContentFormat::DvdPs. The VOBU navigation, CSS key resolution, forced
VobSub probe and AC-3 probe run only on it; the demuxer, codec parsers, timeline and AACS PS
checks treat both as a program stream (ContentFormat::is_program_stream). mpg:// stays
MpegPs and gets no DVD stage.
- A genuine disc's VPLST*.XPL is an AACS-encapsulated ARF ([HD] §4.4.2,
Table 4-13: "AACS", type 12h, Nfs at bytes 7..11, a 272-byte file name field,
then the XML). The wrapper is stripped before parsing; rips ship it bare.
- An AACS directory no longer marks an HD DVD encrypted on its own: rips keep it
renamed (ANY!, AAC!) over clear EVOs. Identify and scan sample the largest
EVOs' packs and keep the AACS verdict only when a pack is scrambled.
- The clip-derived feature also joins PEVOB_1/PEVOB_2 splits.
- Comments: the VTI is the Advanced VTS information file; 0xFD ext 0x56 is the
VC-1 sub video, and TrueHD rides private_stream_1.
HD DVD encrypts each 2048-byte pack on its own ([HD] AACS HD DVD and DVD
Pre-recorded Book 0.952, §4.3), not in 6144-byte aligned units. Bytes 0..128 stay
clear; bytes 128..2048 are AES-128-CBC under the AACS IV with
Kc = AES-G(Kt, Dtk || CPI_lsb96), Dtk = pack bytes 84..88 (§4.3.2, Table 4-7).
Kt is the Title Key named by TITLE_KEY_PTR in the CPI (Table 4-2), carried in the
GCI packet of the NV_PCK that leads each EVOBU (CPI at GCI data offset 12).
- aacs::hddvd: pack classification (NV_PCK with its CPI, scrambled PES pack,
HL_PCK), the content key, per-pack decrypt, and a payload check past byte 128
(PES/padding start codes, AC-3/E-AC-3 sync at the first-access-unit pointer).
- The encrypted flag is PES_scrambling_control of every pack, not byte 20 of a
unit's first pack (an NV_PCK there hid the encrypted packs after it).
- is_clean for MpegPs no longer passes ciphertext: an encrypted pack keeps its
pack start code, so cleanliness is judged on the flags and the payload check.
- The decrypting reader carries the CPI between reads and looks back for the
NV_PCK when a read starts mid-EVOBU; a pack with no CPI is blanked and counted,
one under a Segment Key or an unheld Title Key fails loud.
- Key resolution holds every Title Key under its VTKF entry number (a source's
key index is its position among the present entries; genuine discs populate
all 64 slots) and proves them by decrypting packs from three 2048-sector
windows of the main title; keys that open none refuse (E7013).
- Spec registry: KS-30 quotes [HD] §4.3.2; KS-31 records the disc-measured CPI
location. The spec guards now pin the per-pack flag and the 128-byte head.
The parity_image_hddvd_clear golden changes because its clear fixture is now
judged clear (no key asked, every file extracted) instead of keyed best effort.
- HD DVD: the AACS directory is any root directory holding MKBROM.AACS, whatever
its name. Both encrypted "300" HD DVDs use a plain /AACS (mirror AACS_BAK), which
the old "name ends in !" rule missed, so they could not be keyed. A *_BAK
mirror is used only when no primary copy exists.
- Unit_Key_RO.inf stride: one resolver (content certificate, then a pre-recorded
MKB's type, then index.bdmv INDX0300, then UHD) for the scan, read_aacs_inputs
and resolve_candidate, which now takes the version instead of re-deriving it
from the MKB with a V10 default.
- A Class II MKB (0x000A1003) keeps its media key data in record 0x0c, which the
PK/DK derivation does not read: check_mkb_class reports it as MkbClassError
(E7109) and derivation stops with a warning instead of failing silently.
- STN frame-rate nibbles 5 and 8 are unassigned (libbluray, MediaInfo, tsMuxer);
map them to Unknown instead of 30 and 60 fps. Fight Club UHD's menu clip
carries 8 and was shown as 60fps.
- bdmt titles: Warner's non-English files carry "Blu-ray" in di:name; skip it and
fall through to the next candidate. One placeholder check is shared with the
META title read in bluray.rs.
- /BDMV/AACS no longer marks a disc encrypted: no disc has it, libaacs reads only
root AACS/, and our key readers never look there.
- Doc fixes: the bdmt namespaces are urn:BDA:bdmv;disclib + urn:BDA:bdmv;discinfo,
di:description is a container, and index versions include 0240.
- HEVC: the mastering-display SEI alone now yields HDR10 metadata; MaxCLL/MaxFALL
are optional and omitted from Colour when the stream has no content-light SEI
(Deadpool, Sicario, Lucy, Life of Pi and others carry 137 without 144).
- MPLS: read the HEVC hdr_plus_flag (bit 6 of the byte after color_space, as
libbluray mpls_parse.c) and label an HDR10 base with it HDR10+.
- MKV: the disc's DV track is the EL + RPU stream beside the HDR10 base track, so
its dvcC now says bl_present=0 with the base's compatibility id 6 (Dolby TS
spec 7.2.2; mkvmerge's profile 7 id).
- BD audio_format 6 ("multi-channel", no count) no longer claims 5.1 for
DTS-HD and DD+, which reach 7.1 and are not re-read from the stream.
Region: a DVD reports its VIDEO_TS.IFO region mask (VMGI_MAT 0x23, bit n
set = region n+1 prohibited) as DiscRegion::Dvd or Free; UHD/FMTS stay
Free; Blu-ray and HD DVD report the new DiscRegion::Unknown instead of a
false Region-free.
Labels: audio code extension 2 is Descriptive, 3/4 Commentary; subpicture
code extension 5-7 (captions) is SDH and 9 is forced. The attributes are
per logical stream, so each label follows AST_CTL/SPST_CTL routing.
Forced probe: with probe_forced_subtitles, sample each DVD title's VobSub
streams under the PGS probe's budget and windows and mark a stream forced
when every sampled displayed SPU opens with FSTA_DSP. Scrambled packs are
descrambled after the title's keyless CSS crack, or skipped.
An interleaved cell's sector range also holds another program's units. The scan now follows
the cell's navigation packs (each unit's end, then the next unit's start) so a title's extents,
and its size, are exactly what it plays. Discs write "no next unit" as 0x7FFFFFFF or
0xFFFFFFFF; both end the chain, and an unreadable chain falls back to the whole range.
An angle block plays one angle: extents keep its first cell wherever the block sits, and
chapter times count it once. labels::audio_codec_label exposes the library's friendly codec
name for front ends that show the codec apart from a track's variant tag.
VIDEO_TS.IFO can carry a text data manager (pointer at VMGI 0xD4) with
per-title chapter names. Parse its first language unit, map titles onto
TT_SRPT order and use the names as Chapter.name, keeping the numeric
ordinal for any chapter the disc does not name. A title whose name count
differs from its chapter count is ignored, and every read is bounded by
the table's last byte and the IFO length.
A cell in an interleaved block shares its sectors with another program's units (a second
language version), and each VOB runs its own clock. The mux read every sector of a cell and
inferred joins from timestamp jumps, so a title mixed in the other version's video, froze at
the jumps, and lost about half a second of audio at each VOB join.
Each VOBU's navigation pack names its cell and its start/end time: keep only the VOBUs of
the cell each extent opens with, and shift each new VOB's clock to start where the last
kept VOBU ended, on every track at once.
A PES PTS belongs to the first access unit commencing in that PES
(ISO/IEC 13818-1 2.4.3.7). The assembler gave every mark in an AU's byte
range to that AU, so in a program stream that packs pictures back to back
(ffmpeg's MPEG-PS) each picture took its successor's PTS. MPEG-2 GOPs whose
first displayed picture is a leading B then re-locked their origin a frame
off, read as a join, and dropped that open GOP's leading Bs: the synthetic
A1 fixture lost 3 of 180 frames.
A mark opened inside an AU that runs past it now times the next AU; one
running into the first AU after pre-sync bytes times that AU. Source
provenance is unchanged. DVD and HD DVD titles, which open a PES at each
timed picture, are byte-identical.
The plan trims video by each picture's own PTS in decode order, so an IN
mark inside a GOP dropped the I-picture before it and kept the pictures
predicted from it, and an OUT mark dropped an anchor displayed after it
while keeping the B-pictures displayed before it that decode from it.
Under a seam plan each video picture is now also judged by what the marks
took: once a picture that may be a reference is trimmed (any but an MPEG-2
B or a leading picture), the track drops up to the next keyframe, and that
keyframe's leading pictures with it. These drops count with the frames
outside the marks; kept frames keep their PTS. Sinks pass the picture's
keyframe flag and coding to the plan.
The MPEG-2 fix left its siblings emitting pictures whose reference is not in
the output, and every codec, MPEG-2 included, resumed after a gap on a
keyframe whose leading pictures predict from an anchor the resync gate
dropped.
H.264, HEVC and VC-1 now decide each picture in decode order against the
anchors the output holds, as MPEG-2 does: none before the first
random-access picture, at a join (the decode clock, else the PTS frontier,
moves at the random-access picture), at a VC-1 BROKEN_LINK, and from the
keyframe the gate resumes on after a discontinuity (MPEG-2 too); the gate
still drops and counts the gap itself. Open-GOP leading B-pictures (H.264
by PTS, VC-1 CLOSED_ENTRY 0 by decode order, interlaced field pairs from
FCM/FPTYPE) and HEVC RASL are dropped when that anchor is missing; a CRA
at any join, not only a 3 s step back, is written as BLA. Dropped pictures
keep their display slots, so every kept frame keeps its PTS, and an MVC
dependent unit goes with its base picture. Closed-GOP and IDR-only titles
are unchanged.
connection_condition 1 is the non-seamless join and 5/6 the seamless ones;
the PlayItem and mark_clip_boundary docs had them reversed.
Thirteen ignored tests, each failing on current code, cover H.264 (stream
start, join and gap resync onto an open GOP; inter pictures before the
first random-access picture), HEVC (a forward-PTS join and a gap resync
onto a CRA hand its RASL to the decoder; pictures before the first IRAP),
VC-1 (an open entry point at the start or after a gap; BROKEN_LINK), the
MPEG-2 gap resync onto an open GOP, and the seam plan keeping a picture
whose reference it dropped at a clip's IN mark. Two passing tests pin
the cases that are already right: a closed VC-1 entry point and RASL
after a seamless HEVC join.
A title that opens on an open GOP emitted the GOP's leading B-pictures,
whose forward reference precedes the title; players without ffmpeg's
open-GOP skip showed them as a green or corrupt opening frame.
The parser now decides each picture in decode order against the anchors
the output holds: pictures before the first I-picture and leading
B-pictures of an open GOP whose forward reference is missing (stream
start, broken_link, or a join where the PES timeline jumps) are dropped.
Dropped pictures keep their display slots, so every kept frame keeps its
PTS. Closed GOPs are untouched.
A multi-clip playlist's source PTS restarts at each clip join. The m2ts sink passed it
through, so one TS stepped back in time at every join, and the DTS deriver's order guard
then wrote every later reordered AU as PTS only (a 34-clip UHD title kept DTS on its first
clip alone). The sink now maps each frame through TimelineContinuity, driven by the base
video view, and fails a seam plan that dropped most of the title, as the other sinks do.
The image Read policy refused any extent tail shorter than one 3-sector AACS unit. Blu-ray
stream files always end on the unit grid, but an HD DVD .EVO need not, and a clear HD DVD
now rips through the keyed pipeline (unit_align 3), so the main title of a real disc failed
E9030 at the end of its first off-grid extent. The tail is now one short read, as the live
policy already did; the decrypt stage passes it when clear and refuses it (E7013) when it is
the head of an encrypted unit.
KeyRing::acquire probes the pieces in scope once, each unit judged by its format's detector
(BD/UHD: the CPI flag; HD DVD: flagged or not whole MPEG-PS). Content read in the clear asks
no source, whatever key files the disc declares; for FMTS the forensic segments must read clear
too. Encrypted content goes to the format's keying (BD/UHD per piece, HD DVD best effort) and is
refused E7022 / E7032 when no source has the key. Replaces 9e9ff21's HD DVD-only check.
DefaultDuration came from the IFO's video standard (29.97 for NTSC). A soft-telecined film
title carries 23.976 coded frames a second, so the header disagreed with the frame count
(11440 frames over 476 s on the dvd fixture's title 2). finish() now rewrites the value when
the kept frames' PTS span measures another standard rate.
62d56df warned from parse_pgcit for every split title on every scan; a CLI rip scans its disc
three times, so a DVD with four split titles logged the same warning twelve times. The scan
now logs one line per disc naming every split title, and a rescan of that disc logs it at
debug.
ed02bc1 named every disc layout's output title by its meta title, else its volume id. A
folder's volume id is its directory name, so a DVD title ripped from an image and from its
extracted folder got different segment titles and the two MKVs stopped being byte-identical.
A drive still names by meta title, else volume id.
e44ee04/1f5f5df refused every HD DVD declaring more than one CPS unit before reading any
content (E7022), so a decrypted image that kept its AACS directory stopped ripping. Acquisition
now samples each piece first: unflagged, structurally whole MPEG-PS packs make a clear set.
Matthew Jackson