WebVTT cue settings, teletext grid-row placement, and the blank-row fold

[superuser404notfound]

Follow-up to #233 from tresby's retest. Three separate findings, one of which contradicts a note that shipped with 5.26.0.

WebVTT cue settings arrive after all

5.26.0 reported line / position / align as upstream-blocked. That was a conclusion about the wrong layer. libavcodec's WebVTT decoder does drop them (webvttdec.c says so as a @todo in its file header), so the ASS event line carries no trace. The demuxer keeps them: libavformat/webvttdec.c:166 attaches the verbatim settings string to every packet as AV_PKT_DATA_WEBVTT_SETTINGS, and matroskadec.c:3973 propagates the same side data for WebVTT in Matroska (verified by round-tripping a settings-carrying .vtt through MKV with -c:s copy).

Both decoders now read that side data and map it onto SubtitleCue.placement, and the #112 packet store carries the string through a packet rebuild, since side data does not live in the payload.

Mapping rules and their reasons are in WebVTTCueSettings: a percentage line anchors to the frame edge it is nearer (the spec's default line alignment pins the box top, so line:90% would hang a two-line cue off the frame), a line number keeps only the half of the frame it names (line boxes need a rendered line height the engine does not have), size and vertical have no equivalent in the placement model, and a position without a line keeps only the alignment column.

Teletext keeps its row placement on non-subtitle-flagged pages

The report claimed libzvbi does not emit \an. It does: gen_sub_ass derives the vertical anchor from the grid row and writes {\anN}, in thirds. But that block sits behind is_subtitle_page, which comes from the row-0 header flags (NEWSFLASH clear AND SUBTITLE set AND SUPPRESS_HEADER set, libzvbi-teletextdec.c:617). Off that path it writes the whole page instead, one " \N" per grid row with the empty rows included, and the ordinal of the first non-blank row becomes the only carrier of the position. edgeTrimmed was removing it.

teletextBody now applies ffmpeg's own formula to that ordinal, gated on no \an having arrived and on teletext only. Note that on a flagged page the leading \Ns are NOT the grid row: they are the empty_lines / 2 filler that nudges the block inside its anchor, which is why the gate matters more than the counting.

Two consistency fixes in the same area

• The styled path edge-trimmed with a literal space/tab/newline test while the plain path used .whitespacesAndNewlines, so a Unicode space survived on a styled cue and not on an unstyled one. Not reachable from teletext (IS_TXT_SPACE maps U+00A0 to a space, written as \h), reachable from SRT, WebVTT and ASS.
• The AU live tv #107 blank-line fold worked per run plus one run boundary, which misses the shape the source produces most easily: the padding of an empty row carries the spacing attribute that changes colour, so the blank row lands in a whitespace-only run of its own and breaks the chain. It now folds the flattened sequence and re-splits along the run boundaries.

Blocked on FFmpegBuild

The sidecar half needs the WebVTT demuxer, which was never compiled in: only the decoder was, so avformat_open_input rejected a .vtt file with AVERROR_INVALIDDATA and an external WebVTT subtitle never loaded at all. Same shape as the raw-PGS sup demuxer gap. FFmpegBuild main carries the flag now (superuser404notfound/FFmpegBuild@ac70e2c); this branch still needs the pin bump once that is tagged.

Tests

Issue233WebVTTCueSettingsTests (10, including an end-to-end sidecar decode through the real demuxer), Issue233TeletextRowPlacementTests (9). Full suite green against the local FFmpegBuild build: 1229 Swift Testing, 392 XCTest.