[
https://issues.apache.org/jira/browse/TIKA-5030?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18138825#comment-18138825
]
ASF GitHub Bot commented on TIKA-5030:
--------------------------------------
dschmidt commented on PR #3526:
URL: https://github.com/apache/tika/pull/3526#issuecomment-6660787306
My agent says:
- `SubtitleTracks.wanted()` (SubtitleTracks.java:93-101) asks
`shouldParseEmbedded` about a phantom `track-0.vtt` before any track is found,
and `checkEmbeddedLimits` sets the limit flags as a side effect: an MP4/MKV
without subtitles at the depth or count limit now reports
`EMBEDDED_DEPTH_LIMIT_REACHED`/`EMBEDDED_RESOURCE_LIMIT_REACHED` (e.g.
maxEmbeddedDepth=1, zip holding an .mkv), and with `throwOnMax*` the exception
is swallowed. Probing only once a text track exists would avoid the false flags.
- MKV: a SimpleBlock, or a BlockGroup without BlockDuration, becomes a
zero-length cue `T --> T` (MkvSubtitleExtractor.java:352, asserted in
MkvSubtitleExtractorTest.java:136). WebVTT wants end > start; the track's
DefaultDuration (not read today) or the next block's timestamp could supply it.
- SRT tracks lose every `{...}` (MkvSubtitleExtractor.java:79, 415), so
`Press {Enter}` loses `{Enter}`; `SubtitleParser` only strips `{\...}` override
blocks.
- The 64 MB cap also applies to content already in memory
(MP4Parser.java:207-213): a byte[]-backed MP4 over 64 MB gets no subtitles,
while the docs say "or a source that can hold it in memory".
- Mp4SubtitleExtractor.java:404-409 warns on every text/sbtl/subt track that
isn't tx3g/text, so CMAF files with `wvtt`/`stpp` tracks now carry a container
warning they didn't before. Intended?
- `extractSubtitles` only takes effect from `tika-config.json`; a
per-request `mp4-parser`/`matroska-parser` block is accepted and ignored (same
as `maxBoxSize`).
- Unverified: BOM-less QuickTime `text` samples are decoded as UTF-8
(Mp4SubtitleExtractor.java:563), so older Mac-encoded text would come out as
U+FFFD.
Tests:
- The `tryRetainInMemory()` branch (MP4Parser.java:212, re-openable
`TikaInputStream.get(IOSupplier, ...)`) has no test; forcing it to false fails
nothing.
- `testFileInputOverSpoolLimit` writes its 100 MB sparse file with
`Files.createTempFile` into the system temp dir; `@TempDir` is the convention
here.
- `testCorruptedBytes` (both classes) only reports the seed on a
RuntimeException, so a `@Timeout` hang or an Error loses it. The `catch
(IOException)` in the MP4 one is dead, the extractor records IOExceptions
itself.
Nits: SubtitleTracks.java:79 says a Matroska stream "is read to its end",
but the walk stops at the first Cluster without a text track; CHANGES.txt:73 is
whitespace-only; the PR title got truncated ("…embedded …") and would become
the squash subject.
> Extract MP4 text subtitle tracks as embedded WebVTT and mkv
> -----------------------------------------------------------
>
> Key: TIKA-5030
> URL: https://issues.apache.org/jira/browse/TIKA-5030
> Project: Tika
> Issue Type: Task
> Reporter: Tim Allison
> Priority: Minor
>
> First step is pure java on the file formats that are easy. Second step will
> be an alternative option that shells to ffmpeg.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
