StormByte-Multimedia 3.3.2
C++26 multimedia processing module of the StormByte suite
Loading...
Searching...
No Matches
files Directory Reference
Directory dependency graph for files:
test/files

Detailed Description

Synthetic media fixtures

Video and audio payloads are synthetic seeded noise, approximately two seconds per clip unless stated otherwise. Subtitle cues are synthetic text. Images are synthetic noise. The redistributed font has its own license notice. Fixed fixtures must not be regenerated by tests.

Video

  • video/ffv1_two_flac.mkv: three seconds of synthetic 320x180/25 fps FFV1 testsrc2 video and two independent 48 kHz mono FLAC sine tracks (440 Hz and 880 Hz). Reproduces concurrent mux input setup with one-packet queues, local byte limits and a shared budget. Tests compare every compressed packet payload by track across repeated Transcoder and manual remuxes; they never regenerate this fixture. SHA-256: 8d928777d129309a07adc0e86a7a2262e399ef085d7e83f4fbbdd82fa5166a42 (609,533 bytes). Authored once with:

    ffmpeg -hide_banner -loglevel error -n \
    -f lavfi -i 'testsrc2=size=320x180:rate=25:duration=3' \
    -f lavfi -i 'sine=frequency=440:sample_rate=48000:duration=3' \
    -f lavfi -i 'sine=frequency=880:sample_rate=48000:duration=3' \
    -map 0:v -map 1:a -map 2:a -c:v ffv1 -c:a flac \
    -threads "$(nproc)" test/files/video/ffv1_two_flac.mkv
  • video/bluray_like_hdr10.mkv: HEVC Main 10, BT.2020/PQ, mastering-display and content-light metadata; English AC-3 5.1 and Spanish AAC 5.1; English and Spanish PGS; attached synthetic cover.
  • video/bluray_like_hdr10.mp4: H.264 8-bit SDR video with a BT.709 color matrix, two 5.1 AAC tracks and English/Spanish mov_text subtitles. Primaries and transfer are unspecified; HDR metadata is absent.
  • video/bluray_like_vp9.mkv: VP9 video, the same two distinct 5.1 audio codecs and English/Spanish PGS subtitles.
  • video/anime_like.mkv: one HEVC video, one English Opus 5.1 track, three bitmap PGS tracks (spa, eng, jpn) and one ipag-mona.ttf attachment.
  • video/hdr10_metadata_source.mkv: HEVC Main 10 source with BT.2020/PQ, mastering-display and content-light metadata.
  • video/hdr_without_mastering_metadata.mkv: BT.2020/PQ HEVC Main 10 with mastering-display and content-light side-data intentionally absent. File is expected to classify it as HDR10-like and report HDR10::Source::Heuristics from the signaled primaries/transfer, rather than treating it as SDR.
  • video/xvid_mp3_stereo.avi: Xvid/MPEG-4 Part 2 video with stereo MP3 audio.
  • video/late_bframes_dx50.avi: three seconds of synthetic 96x64/25 fps MPEG-4 Part 2 video and AC-3 audio. A one-second Xvid segment without B-frames precedes a two-second packed B-frame segment. The later reordering makes at least one decoded raw PTS differ from FFmpeg's best-effort presentation timestamp. Regress the decoded and transcoded timelines and copied audio with SVT-AV1 and VP9. This fixture uses FFmpeg's decoded timestamps as its oracle; it does not independently validate coded VOP timing or reproduce the duplicated-reference transition. SHA-256: 41771e6f9abfffe53be576626e0140c154f7b0f26a1a5ccb15d73642f9285a88 (111,168 bytes). Authored once with FFmpeg 9.0.2 and libxvid:

    ffmpeg -v error -n -f lavfi -i 'testsrc2=size=96x64:rate=25:duration=1' \
    -c:v libxvid -bf 0 -g 25 -vtag DX50 no-b.avi
    ffmpeg -v error -n -f lavfi -i 'testsrc2=size=96x64:rate=25:duration=2' \
    -c:v libxvid -bf 2 -g 25 -vtag DX50 with-b.avi
    ffmpeg -v error -n -i 'concat:no-b.avi|with-b.avi' \
    -f lavfi -i 'sine=frequency=440:sample_rate=48000:duration=3' \
    -map 0:v -map 1:a -c:v copy -tag:v DX50 -c:a ac3 -b:a 96k \
    -fflags +bitexact -flags:a +bitexact -shortest late_bframes_dx50.avi
  • video/vfr_pts.mkv: synthetic 96x64 FFV1 with irregular 40 ms and 80 ms frame gaps on a 1/1000 container timebase. Tests compare every decoded presentation timestamp before and after H.264/MP4 transcoding. SHA-256: 9285341f19ae9a237710907cc5985e662e7edf85b3c1da0e37f55769b5fff5be (73,279 bytes). Authored once with:

    ffmpeg -hide_banner -loglevel error -n -f lavfi \
    -i 'testsrc2=size=96x64:rate=25:duration=2' \
    -vf "select='not(eq(mod(n,3),1))'" -fps_mode vfr -c:v ffv1 \
    -fflags +bitexact -flags:v +bitexact -map_metadata -1 vfr_pts.mkv
  • video/vfr_subframe_pts.mkv: synthetic 96x64 FFV1 with alternating 45 ms and 35 ms presentation intervals, not multiples of the nominal 25 fps frame duration. The H.264, SVT-AV1 and VP9 tests check every decoded timestamp through encoding, including H.264 after a filter replaces each frame. SHA-256: 92df3e91d5923756e0978f2056006471d903dbe2e550d9f4dee2ad57dfc6d043. Authored once with:

    ffmpeg -v error -f lavfi -i 'testsrc2=size=96x64:rate=25:duration=1.32' \
    -vf "settb=1/1000,setpts=N*40+mod(N\,2)*5" -fps_mode passthrough \
    -enc_time_base:v 1:1000 -c:v ffv1 -fflags +bitexact -flags:v +bitexact \
    -map_metadata -1 -n vfr_subframe_pts.mkv
  • video/vfr_90k_pts.mp4 and video/vfr_90k_unrepresentable.mp4: synthetic 96x64 H.264 with a 90 kHz clock. The first uses 45 ms and 35 ms intervals representable by MPEG-4 Part 2; the second adds a single 90 kHz tick that cannot be represented together with the other intervals by that codec's 16-bit VOP clock. MPEG-4 and Xvid tests check exact preservation; a separate MPEG-4 test checks explicit failure. SHA-256: 6e988a1bbfea679f769de49faeebbd346cc37166cab1f0dd210c22c7b6566a17 and d6d90290e64713e83660fef779312cba90fe4b7850d47669874f1a373d3ee9a5. Authored once with FFmpeg using the following commands:

    ffmpeg -v error -f lavfi -i 'testsrc2=size=96x64:rate=25:duration=1.32' \
    -vf "settb=1/90000,setpts=N*3600+mod(N\,2)*450" -fps_mode passthrough \
    -enc_time_base:v 1:90000 -c:v libx264 -preset ultrafast \
    -video_track_timescale 90000 -fflags +bitexact -flags:v +bitexact \
    -map_metadata -1 -n vfr_90k_pts.mp4
    ffmpeg -v error -f lavfi -i 'testsrc2=size=96x64:rate=25:duration=0.16' \
    -vf "settb=1/90000,setpts=N*3600+mod(N\,2)*451" -fps_mode passthrough \
    -enc_time_base:v 1:90000 -c:v libx264 -preset ultrafast \
    -video_track_timescale 90000 -fflags +bitexact -flags:v +bitexact \
    -map_metadata -1 -n vfr_90k_unrepresentable.mp4
  • video/cfr_24fps_microseconds.mp4: synthetic 96x64 H.264 at 24 fps with a 1/1000000 time base. MPEG-4 output is compared at the source clock precision, accounting for the rounding of 24 fps timestamps to microseconds. SHA-256: a34af2c063b692700cc9182eea9ca7ef845ca43f6fd72e2677a6f33d7170fd4f.

    ffmpeg -v error -f lavfi -i 'testsrc2=size=96x64:rate=24:duration=1' \
    -vf 'settb=1/1000000' -fps_mode passthrough -enc_time_base:v 1:1000000 \
    -c:v libx264 -preset ultrafast -video_track_timescale 1000000 \
    -fflags +bitexact -flags:v +bitexact -map_metadata -1 \
    -n cfr_24fps_microseconds.mp4
  • video/packed_ac3_seek.avi: eight seconds of 96x64/25 fps synthetic testsrc2 packed Xvid B-frames (libxvid -bf 2 -g 25) and six distinct sine tones, AC-3 5.1(side), 48 kHz, 448 kb/s. Audio is deliberately stored in 2,240-byte AVI chunks (40 ms), not aligned to its 1,792-byte AC-3 frames (32 ms). Its audio stream uses scale 1, rate 56,000, sample size/block align 1 and byte-count length; video keeps its 1/25 timebase. RIFF sizes and idx1 offsets/flags are rebuilt after rechunking, preserving video packet order and keyframe flags. The first audio packet after a 4-second seek is a 1,344-byte unsynchronized fragment; the next is a full frame. This reproduces the demux/parser seek-framing failure independently of any copyrighted movie. Tests check paused images, changed pixels, resumed position, source-native planar float surround samples, sample continuity and bounded audio lead. SHA-256: 7ade8119ae1a37fb564b122d4743cc024b4f154e18c50336b23410aa2d4f9216 (652,758 bytes).
  • video/unknown_duration.webm: 130 seconds of synthetic 96x64 VP8 at 5 fps, no audio, no container duration and no bitrate-derived duration estimate. Both ffprobe and SimplePlayer report unknown (N/A and 0, respectively). Tests seek to absolute 30, 120 and 10 seconds while paused and resume from each target; this is not a known-duration file with its value overridden. SHA-256: da9a6016fa7bb3febbe19c5d37bcd0b79eacf54e2d5602feab93ce7c06f75cf5 (651,299 bytes). Authored once with:

    ffmpeg -v error -f lavfi -i 'testsrc2=size=96x64:rate=5:duration=130' \
    -an -c:v libvpx -b:v 0 -crf 20 -g 5 -live 1 unknown_duration.webm
  • video/kvazaar_unaligned_704x396.avi: synthetic MPEG-4 Part 2, YUV420P, square pixels, 704x396, 24 frames at 24 fps and no audio. Its height deliberately violates Kvazaar's multiple-of-eight requirement. The regression requires successful HEVC encoding at 704x400; automatic padding adds two black rows above and below, without scaling. VMAF also processes all 24 output frames. Authored once with FFmpeg 9.0.2; tests never regenerate the fixture. SHA-256: 8b8b595938d061cf2f2a64c1fd07deee65c7a0609f49696d95c6c3e67560b37f (246,386 bytes).

    ffmpeg -hide_banner -loglevel error -n -f lavfi \
    -i 'testsrc2=size=704x396:rate=24:duration=1' -an \
    -c:v mpeg4 -q:v 5 -bf 0 -pix_fmt yuv420p -map_metadata -1 \
    -fflags +bitexact -flags:v +bitexact test/files/video/kvazaar_unaligned_704x396.avi
  • video/interleave_long.mkv: twelve seconds, longer than any encoder lookahead the interleave tests configure, so a video encoder lagging audio by seconds shows up in file order. H.264 High 320x180 at 24000/1001 fps from the testsrc2 pattern, GOP 24 with three B-frames; English AC-3 stereo pink noise, seed 20261026. Authored with FFmpeg 9.0.2 on 2026-10-26. SHA-256: 7ed06e6575e3794c94e4a8e1a69fe6b002962ccde00c7d2414b0df888f5e08af (327,078 bytes).

    ffmpeg -hide_banner -loglevel error -n \
    -f lavfi -i 'testsrc2=size=320x180:rate=24000/1001:duration=12' \
    -f lavfi -i 'anoisesrc=color=pink:amplitude=0.1:sample_rate=48000:duration=12:seed=20261026' \
    -map 0:v -map 1:a -c:v libx264 -preset veryslow -crf 30 -g 24 -bf 3 -pix_fmt yuv420p \
    -c:a ac3 -b:a 96k -ac 2 -metadata:s:a:0 language=eng \
    -fflags +bitexact -flags:v +bitexact -flags:a +bitexact interleave_long.mkv
  • video/live_audio_switch_long.mkv: approximately 96 seconds, eight stream-copy repetitions of the synthetic interleave_long.mkv fixture. The live audio switching regression needs input beyond decoder read-ahead throughout six track changes; a two-second clip can reach EOF before those operations complete under instrumentation. Low-resolution video keeps decode overhead from overwhelming the real-time sink assertions. Tests retain their epoch, monotonic timestamp and 500 ms inter-frame checks and never regenerate this fixture. SHA-256: caf35ac55c2ca5132d904ee2c236ab79457f009d1908e458b6687ad5e1ca8ca7 (2,612,036 bytes). Authored once with:

    ffmpeg -v error -nostdin -n -stream_loop 7 \
    -i test/files/video/interleave_long.mkv -map 0 -c copy \
    test/files/video/live_audio_switch_long.mkv
  • video/skew_audio_late.mkv and video/skew_video_late.mkv: the first 20 seconds of live_audio_switch_long.mkv, stream-copied (H.264 320x180 and AC-3 stereo) with unchanged timestamps but skewed physical interleave: each audio (respectively video) packet is stored where the other stream is already three seconds ahead. Software preview regressions retain strict A/V timing and delivery-gap limits to isolate the decoded read-ahead scheduler. Separate Auto cases cover decoding and live audio switching with an intentional 600 ms display stall, without assuming real-time hardware delivery. Both policies require consecutive video timestamps, bounded live audio re-enable silence and unchanged epochs. Phase logs and buffered per-picture timestamps, audio-clock/queue snapshots and reset counters are printed even when an assertion returns early; CTest normally hides passing test output and shows it with --output-on-failure. Every source packet below 20 seconds is kept, including the leading IDR and P frame that Matroska demuxing reports without a decode timestamp, so both files start at their IDR and decode without missing references. SHA-256: 00beb685a8c2b0af73f3cb746880bd977e1b5a287be7e4371067b7be2613d768 (550,278 bytes) and 879fde8b3af4ce66e8497a4fc57329f0573952b5d46b7e1029c9496ce9c0ec95 (550,298 bytes). Authored once with the bit-exact authoring tool tools/skew_interleave.cxx (system libavformat 63.1.102; not built by tests):

    c++ -std=c++20 test/files/tools/skew_interleave.cxx \
    $(pkg-config --cflags --libs libavformat libavcodec libavutil) -o skew_interleave
    ./skew_interleave test/files/video/live_audio_switch_long.mkv \
    test/files/video/skew_audio_late.mkv audio 3 20
    ./skew_interleave test/files/video/live_audio_switch_long.mkv \
    test/files/video/skew_video_late.mkv video 3 20
  • video/missing_pts_bframes.avi: synthetic two-second 96x64 MPEG-4/AC-3 fixture with reordered frames and missing video packet PTS. Its regression verifies all 50 decoded/output video timestamps, intermediate encode progress and a two-second output timeline, including the final delayed frame.

    ffmpeg -f lavfi -i testsrc2=size=96x64:rate=25:duration=2 \
    -f lavfi -i sine=frequency=440:sample_rate=48000:duration=2 \
    -c:v mpeg4 -bf 2 -q:v 5 -c:a ac3 -b:a 192k -shortest missing_pts_bframes.avi

Dolby Vision

  • video/dovi_only.mkv: synthetic Profile 5, compatibility ID 0; IPT-PQ-C2 matrix, full range, BT.2020 primaries and PQ transfer are explicitly signaled. It has no static HDR10 mastering-display/content-light SEI or HDR10-compatible VUI; DOVI RPU Level 6 metadata is present.
  • video/dovi_hdr10.mkv: synthetic Profile 8, compatibility ID 1; explicit HDR10 mastering-display/content-light metadata and BT.2020/PQ signaling.

Both fixtures contain one HEVC Main 10 stream, 320x192, 24 fps, 48 frames and two seconds of deterministic noise with seed 424242. Dolby Vision configuration: version 1.0, level 1, compression 0, RPU present, enhancement layer absent and base layer present. Every decoded frame carries raw RPU and decoded Dolby Vision metadata.

The combined fixture has mastering-display primaries G(13250,34500), B(7500,3000), R(34000,16000) and white point WP(15635,16450), with chromaticity denominator 50000; maximum/minimum luminance 4000/0.005 nits; MaxCLL 636 and MaxFALL 103. File exposes stream configuration; decoded pipeline frames additionally expose parsed metadata and owned RPU bytes.

HDR10+

  • video/hdr10plus_metadata_source.mkv: fixed synthetic HEVC Main 10 source, 1920x1080, yuv420p10le, 24 fps, 48 frames and two seconds, with BT.2020 primaries, BT.2020 NCL matrix and PQ transfer. It combines the existing hdr10_metadata_source.mkv picture source with registered SMPTE ST 2094-40 dynamic metadata; it is not calibrated HDR content.
  • video/hdr10plus_metadata.json: the retained authoring metadata, HDR10+ profile A, JSON version 1.0, application version 1, one window and 48 frame entries. Two distinct payloads alternate on successive frames: targeted display luminance 400/600 nits, different MaxSCL, average/percentile distributions, knee points and four Bezier anchors. Tests construct these expected values in-process, without a JSON reader or authoring-tool dependency.

Static mastering metadata: G(13250,34500), B(7500,3000), R(34000,16000), WP(15635,16450), denominator 50000; maximum/minimum luminance 1000/0.005 nits; MaxCLL 1000 and MaxFALL 400. A one-off ffprobe validation on 2026-10-12 found 48 decoded frames, each carrying mastering-display, content-light and HDR10+ side data. Strict hdr10plus_tool extraction checks the registered HEVC SEI against the retained JSON. These tools are fixture-authoring/validation tools only, never repository test dependencies; tests do not regenerate either file.

The in-process HDR10+ suite checks public File detection against the static-only fixture, registry capabilities, Matroska remux, x265 HEVC, libaom AV1 and both native/libvpx VP9 decoders. Every hop compares all 48 normalized T.35 payloads, monotonic 24 fps timestamps within Matroska's 1 ms rounding, color properties and static HDR10, allowing only AV1 static-metadata quantization. The HEVC -> AV1 -> HEVC case also remuxes final HEVC to an elementary stream and requires native decoded AVFrame HDR10+ side data on every frame, proving registered SEI preservation independently of Matroska BlockAdditions. Raw-stream decoding is compared in display order, without a container timestamp requirement. VP9 support is limited to the existing Matroska ITU-T T.35 BlockAddition route, not invented VP9 bitstream metadata.

Audio

  • audio/noise_tagged_covers.mp3: the placeholder tagged MP3 below, re-encoded once at 44.1 kHz with two different, 64x64 synthetic pictures (PNG front and JPEG back). Picture descriptions/filenames are deliberately empty, as with common ID3 covers. No real music or artwork. SHA-256: 45d94dc09e2495c20089636b5baef24bc8048fb1a442f287bdbcc147bd81ed28 (72,026 bytes).
  • audio/noise_tagged_covers.aiff: the first 0.1 seconds of that covered MP3, PCM S16 big-endian with ID3 tags and the same two image payloads. Used to exercise AIFF remux independently of PCM encoder discovery. SHA-256: b47e4a807d7832cdf80797fb19b613516ff81d8ff8522cc1c1368f4ba1909297. Authored once with FFmpeg 9.0.2; tests never regenerate either fixture:

    ffmpeg -hide_banner -loglevel error -n -i test/files/attachments/synthetic-cover.png \
    -vf scale=64:64 -frames:v 1 cover-front.png
    ffmpeg -hide_banner -loglevel error -n -i cover-front.png -frames:v 1 cover-back.jpg
    ffmpeg -hide_banner -loglevel error -n -i test/files/audio/noise_tagged.mp3 \
    -i cover-front.png -i cover-back.jpg -map 0:a -map 1:v -map 2:v \
    -map_metadata 0 -ar 44100 -c:a libmp3lame -b:a 192k -c:v copy \
    -disposition:v attached_pic -metadata:s:v:0 comment='Cover (front)' \
    -metadata:s:v:1 comment='Cover (back)' -id3v2_version 3 -write_id3v1 0 \
    test/files/audio/noise_tagged_covers.mp3
    ffmpeg -hide_banner -loglevel error -n -i test/files/audio/noise_tagged_covers.mp3 \
    -map 0 -t 0.1 -c:a pcm_s16be -c:v copy -write_id3v2 1 \
    test/files/audio/noise_tagged_covers.aiff

    Mux-policy regressions convert audio while automatically copying both pictures and all ten placeholder tags, then remux with explicit image selection and again without it. All hops require exactly one audio stream, exactly two byte-identical pictures and no ordinary video streams. They cover Opus, Vorbis, Ogg FLAC, native FLAC, MP3, AAC/ALAC M4A, AIFF and Matroska audio. A covered source can still become CAF audio without inventing image support; selecting a font for Opus fails explicitly.

  • audio/noise_stereo.wav and audio/noise_stereo.mp3: stereo PCM and MP3 sources.
  • audio/noise_tagged.mp3: independently generated white noise, seed 20261025, amplitude 0.1, two seconds before MP3 encoder padding, 48 kHz, two identical channels, LAME 192 kbit/s. Authored with FFmpeg 9.0.2 on 2026-10-25, with ID3v2.3 and no ID3v1, artwork, lyrics or copyright tag. All descriptive values are placeholders, not references to artists or recordings.
Public field ID3v2.3 frame Value
Title TIT2 Placeholder Track
Artist TPE1 Placeholder Artist
Album TALB Placeholder Album
AlbumArtist TPE2 Placeholder Album Artist
Composer TCOM Placeholder Composer
Genre TCON Test
Comment COMM Synthetic noise fixture
Date TYER 2026
Track TRCK 3
Disc TPOS 2

The one-off FFmpeg authoring step writes its encoder identifier (Lavf63.1.102) into this immutable input fixture. Conversion and remux tests require output File::Metadata().Encoder() to equal StormByte-Multimedia <current build version>, not this source identifier. Backend technical vendor strings are separate and are not rewritten. SHA-256: 142c0bb795d29bab666824d0ddb5b908fe602efa888d779d73ae632f98a72b9e (49,837 bytes).

One-off authoring command, run from the repository root; tests neither invoke FFmpeg nor regenerate this fixed fixture:

ffmpeg -hide_banner -loglevel error -n -f lavfi \
-i 'anoisesrc=color=white:amplitude=0.1:sample_rate=48000:duration=2:seed=20261025' \
-ac 2 -c:a libmp3lame -b:a 192k -id3v2_version 3 -write_id3v1 0 \
-map_metadata -1 \
-metadata title='Placeholder Track' -metadata artist='Placeholder Artist' \
-metadata album='Placeholder Album' -metadata album_artist='Placeholder Album Artist' \
-metadata composer='Placeholder Composer' -metadata genre='Test' \
-metadata comment='Synthetic noise fixture' -metadata date='2026' \
-metadata track='3' -metadata disc='2' test/files/audio/noise_tagged.mp3
  • audio/noise_51.m4a: 5.1 AAC source.
  • audio/noise_tagged.opus: independently generated white noise (not converted from the MP3 fixture), seed 20261026, amplitude 0.1, two seconds before encoder padding, 48 kHz, two identical channels, libopus at 128 kbit/s. Contains the same ten placeholder values listed for audio/noise_tagged.mp3, stored as OpusTags/Vorbis comments on the audio stream rather than ID3. No copyright, lyrics or artwork. Authored with FFmpeg 9.0.2 on 2026-10-25; its encoder identifier is Lavc63.1.102 libopus. SHA-256: 9cabff81136745bc4fd53082f6319354d221e3fdb71dd90b31419ec01a883394 (23,180 bytes).

    One-off authoring command:

    ffmpeg -hide_banner -loglevel error -n -f lavfi \
    -i 'anoisesrc=color=white:amplitude=0.1:sample_rate=48000:duration=2:seed=20261026' \
    -ac 2 -c:a libopus -b:a 128k -map_metadata -1 \
    -metadata title='Placeholder Track' -metadata artist='Placeholder Artist' \
    -metadata album='Placeholder Album' -metadata album_artist='Placeholder Album Artist' \
    -metadata composer='Placeholder Composer' -metadata genre='Test' \
    -metadata comment='Synthetic noise fixture' -metadata date='2026' \
    -metadata track='3' -metadata disc='2' test/files/audio/noise_tagged.opus
  • audio/noise_51.opus: 5.1 Opus source.
  • audio/noise_71.wav: deterministic seeded PCM S16 LE noise, 48 kHz, eight channels and an explicit WAVEFORMATEXTENSIBLE 7.1 speaker mask (0x63f). Covers permitted automatic 7.1-to-5.1 encoding to AC-3/E-AC3 without a manual downmix filter.
  • audio/noise_opus_source_51.opus: explicit 5.1 Opus decoder input.
  • audio/noise_opus_destination_51.opus: independently encoded 5.1 Opus reference for encode-destination tests.

Subtitles

  • subtitles/*.sup: genuine HDMV PGS bitmap tracks in SUP format; the anime-like set includes Spanish, English and Japanese.
  • subtitles/*.srt: matching synthetic text cues, except for the independent late-cue regression below.
  • subtitles/late_first_cue.srt: original synthetic SubRip text containing one cue from 01:00:00,000 to 01:00:02,000. It exercises a subtitle track whose first cue is timestamped one hour after the start. Reading the fixture does not wait in real time; the test encodes it as ASS in Matroska through Transcoder and inspects the output with File. The Japanese PGS was rendered with IPAMonaGothic at a larger size for OCR coverage. The test transcodes it to SubRip using the bundled jpn.traineddata and checks for the hardcoded Japanese tokens 字幕 and 出発 in the resulting Matroska subtitle packet.

Images

  • logo/watermark-noise-logo.png: synthetic 96x96 RGB PNG, seeded noise on a cyan background with a four-pixel yellow border; seed 20261010. Watermark tests use 30% opacity.
  • logo/watermark-alpha.png: synthetic 96x96 RGBA PNG generated with FFmpeg: ffmpeg -f lavfi -i nullsrc=s=96x96 -vf "format=rgba,geq=r='if(lt(X,48),255,64)':g=96:b=192:a='if(lt(Y,32),0,if(lt(Y,64),128,255))'" -frames:v 1 watermark-alpha.png. Two colors and transparent, half-alpha and opaque bands exercise opacity and exact preservation of uncovered samples.
  • attachments/synthetic-cover.png: synthetic noise PNG used as a Matroska attachment and MP4/M4A cover art.

Font Attachments

  • attachments/ipag-mona.ttf: IPA Mona font, also attached to video/anime_like.mkv.
  • attachments/IPAMona-LICENSE.txt: IPA Mona redistribution notice, including M+ and public-domain notices. It is not muxed as an attachment because plain-text attachments have no media codec parameters.

Invalid Inputs

  • invalid/random_garbage.mkv: random bytes, expected to fail parsing.
  • invalid/truncated_hevc.mkv: deliberately incomplete Matroska header, expected to fail parsing or decoding.

Test-generated outputs

Generated media is written under the configured STORMBYTE_TEST_OUTPUT_DIR, normally <build>/test-output, under <executable>/<test-case>/. These outputs are not additional redistributed fixtures. Codec roundtrips generate their intermediate inputs within the same case and do not depend on execution order.

  • pipeline/audio-codecs/: stereo PCM, surround AAC and Opus fixtures exercise FDK-AAC, Vorbis, Opus, LAME, FLAC, ALAC and AC-3/E-AC3. The cases inspect container and codec identities, channel count, 48 kHz sample rate and positive duration. FDK-AAC is required when FFmpeg is bundled with nonfree enabled.
  • Tagged-audio cases independently convert both audio/noise_tagged.mp3 and audio/noise_tagged.opus to Vorbis/Ogg, Opus, native FLAC, ALAC/M4A, AAC/M4A and MP3, then reopen each output through public File::Metadata() and assert all ten placeholder fields and the absence of copyright. MP3-to-MP3 and Opus-to-Opus explicitly select encoders and recode; Opus-to-MP3 explicitly selects libmp3lame. The MP3 fixture also roundtrips through Vorbis and Opus back to MP3, exercising metadata read from stream comments and written back as ID3. Separate MP3-to-MP3 and Opus-to-Opus remux cases call Remux() without an encoder and preserve the same fields. An untagged WAV-to-Opus case must not invent music tags. Tests compare semantic values, not ID3 bytes: Vorbis/Opus/FLAC use Vorbis comments and M4A uses MP4 metadata. Optional encoder availability follows the existing codec test skip policy.
  • pipeline/video-codecs/: the H.264 SDR MP4 fixture is scaled to 320x192 for SVT/libaom AV1, VP8/VP9, x264, OpenH264, Kvazaar and x265. AV1 decoder cases generate their own input before selecting dav1d or libaom. The scaled dimensions are multiples of eight for Kvazaar compatibility.
  • pipeline/mux-policy/: the stereo PCM fixture produces direct FLAC, FLAC in OGA, ALAC in CAF and direct E-AC3. Negative cases cover incompatible codec/container combinations and WebM attachment rejection.
  • pipeline/mp4-policy/: the five-track H.264/AAC/timed-text MP4 fixture exercises individual and mixed outputs, two audio or subtitle tracks, reversed audio order, all five tracks, .m4a/.m4b aliases and mixed remux/encode. The HEVC MKV supplies a HEVC/AAC remux case. Its synthetic PNG cover is preserved byte-for-byte in video MP4 and audio-only M4A as cover art, and in Matroska as a real attachment. File must expose each cover through Attachments() without adding a playable video stream. Matroska additionally preserves the original filename. The IPA Mona font exercises explicit rejection of resources that the MP4 backend cannot represent; this is an implementation limit, not a claim that ISO BMFF forbids embedded resources. Successful outputs are checked for track count, order, codec, language, dimensions, audio properties and duration.
  • pipeline/subtitle-header/: manual pipelines use the existing MKV and MP4 fixtures as audio sources while reserving ASS or MOV timed-text encoders. Input subtitles are decoded and passed through a test filter that removes their payload before reaching the encoder; all queues have consumers and EOF propagates normally, but no output cues are generated. These cases check that codec preparation does not wait for a first cue and that Matroska retains the selected empty subtitle track. MP4 may omit a track that finishes entirely empty, as libavformat does; the test does not require an artificial empty final track. late_first_cue.srt separately exercises encoding of a real cue at one hour, including decoder initialization after the short source reaches EOF.
  • pipeline/watermark/: path and embedded-byte variants use the same colored-noise logo and distinct output filenames. Both outputs are 640x360 VP9 WebM. File checks container, codec, resolution and positive duration. Synthetic gradient tests compare graphics-white-relative linear RGB across SDR BT.709 and HDR BT.2020 PQ/HLG at 0/45/50/70/100% opacity, with a maximum per-channel error below 0.002. Float RGB and 10-bit 4:2:0 tests compare uncovered and fully transparent samples byte-for-byte, allowing shared chroma samples only where the logo has coverage.
  • pipeline/dovi/: DOVI-only, HDR10-only and combined cases each produce separate Matroska remux, libx265 reencode, watermark and pixel-replacement outputs. The -verified.mkv copies reopen the written outputs through public remux/analytics APIs to decode them again. Tests compare configuration, color properties, static HDR10 fields, decoded DOVI fields, exact per-frame RPU bytes and frame counts. DOVI inputs must yield 48 frames. Copy/move and owned-RPU loader cases additionally check public value/frame preservation. VP9 and Kvazaar cases require explicit DOVI rejection, not successful metadata loss.

Dolby Vision fixtures are not visually calibrated or Dolby-certified content. DOVI/HDR watermark regressions retain source dimensions and color properties and check metadata preservation, not overlay appearance or Dolby calibration. RPU spatial assumptions may become inconsistent after cropping, aggressive scaling or color transformations.

Tests use StormByte public APIs; no external media programs or libdovi are required. Optional implementations may be skipped only when unavailable; actual encoding failures are not successful skips. Bundled GPL-disabled configurations omit the x265/Kvazaar cases.