v2 - #10

Draft

mihar-22 wants to merge 119 commits into

mainfrom

t3code/improve-caption-rendering

Commits on Sep 2, 2026

chore: upgrade toolchain to rollup 4, vitest 3, typescript 5.9, prettier 3

mihar-22
and
claude
committed
677c875
fix(vtt)!: escape rendered cue HTML and harden tokenizer nesting

mihar-22
and
claude
committed
ab55dea
fix(vtt): report reversed cue ranges, keep decimal percentages, parse long hours

mihar-22
and
claude
committed
67f6aa1
feat(vtt): add HLS X-TIMESTAMP-MAP helpers

mihar-22
and
claude
committed
2c705a5
feat(srt): strip coordinates, map font colours, apply numpad alignment tags

mihar-22
and
claude
committed
ee70060
feat(ssa)!: rewrite parser with PlayRes scaling, override tags, and embedded fonts

mihar-22
and
claude
committed
d55c63c
feat: add TTML/IMSC1, SCC (CEA-608), LRC, and SBV parsers

mihar-22
and
claude
committed
f3e5c2f
feat(renderer): sorted cue index, identity diffing, frame-accurate sync, faster collisions

mihar-22
and
claude
committed
c998c04
feat(styles): balanced wrapping, container units, text-stroke outlines, edge-style presets

mihar-22
and
claude
committed
21e240e
docs: document new formats, renderer APIs, styling, and security model

mihar-22
and
claude
committed
7db9948
fix(renderer): apply cue position horizontally and size fonts from the cue box

mihar-22
and
claude
committed
6ff0657
test: add Vitest 4 browser mode with Playwright and real-layout positioning tests

mihar-22
and
claude
committed
9140499
chore: drop misplaced pure annotations

mihar-22
and
claude
committed
c0a9899
build: replace rollup with tsdown

mihar-22
and
claude
committed
0b060e2
fix(renderer): transform-aware collision boxes, fixed cues, reading-order stacking

mihar-22
and
claude
committed
6cc4878
feat(ssa): size cue boxes to their text and pin positioned cues

mihar-22
and
claude
committed
e447158
test(browser): cover transforms, fixed cues, and SSA layout in Chromium

mihar-22
and
claude
committed
88e269b
docs: regenerate README screenshots from sandbox scenarios

mihar-22
and
claude
committed
7e8967f
feat(vtt): support STYLE blocks with scoped, sanitized ::cue styling

mihar-22
and
claude
committed
85869ae
fix(vtt)!: spec-conformant file parsing with explicit real-world tolerances

mihar-22
and
claude
committed
fe52207
feat(scc): decode CC2 via the channel option

mihar-22
and
claude
committed
f026f88
feat(ttml): cell units, font sizes, vertical writing modes, sibling end inference

mihar-22
and
claude
committed
a27e262
fix(renderer): spec-conformant line placement for vertical, negative, percentage, and RTL cues

mihar-22
and
claude
committed
26acd00
feat(renderer): event-driven sync mode driven by native TextTrack cue boundaries

mihar-22
and
claude
committed
98ce53d
fix(ssa): honour \1c colour tags and keep toggled formatting properly nested

mihar-22
and
claude
committed
b5d486d
docs: document STYLE blocks, strict vs tolerant parsing, sync modes, and development workflow

mihar-22
and
claude
committed
a3d9de1
test(browser): screenshot-based visual regression for representative scenarios

mihar-22
and
claude
committed
c700e99
chore: ignore vitest attachments

mihar-22
and
claude
committed
55f871d
feat(cea): stream-oriented CEA-608 decoder with field 2 (CC3/CC4) and cc_data parsing

mihar-22
and
claude
committed
34ed96f
feat(cea): CEA-708 (DTVCC) decoder

mihar-22
and
claude
committed
e22e32a
docs: document CEA-608/708 stream decoding and visual regression testing

mihar-22
and
claude
committed
4ed8285
refactor(renderer): split positioning into measure, layout, and write phases

mihar-22
and
claude
committed
ddc9b8b
feat(cue)!: structured layout and text style model on VTTCue with JSON round-tripping

mihar-22
and
claude
committed
91b5a52
feat(element): <media-captions> custom element

mihar-22
and
claude
committed
9cf1f9c
feat(track): CueTrack with incremental sorted index, mutable end times, eviction, and events

mihar-22
and
claude
committed
e8c39b3
feat(renderer): render from a CueTrack with live updates and retention-based eviction

mihar-22
and
claude
committed
dd362ee
feat(renderer): build cue DOM directly, dispatch enter/exit events, optional screen reader announcer

mihar-22
and
claude
committed
a9f94b2
feat(styles)!: cascade layer, registered custom properties, and per-format parser entries

mihar-22
and
claude
committed
9613e63
feat(cea): live mode emitting open-ended cues that are updated in place

mihar-22
and
claude
committed
b7b3b71
docs: layout model, CueTrack, custom element, cue events, announcer, layers, parser entries

mihar-22
and
claude
committed
5f41c74
test(wpt): vendor the web-platform-tests WebVTT parsing suites

mihar-22
and
claude
committed
671bd0e
fix(vtt)!: resolve every parser bug surfaced by the WPT suite

mihar-22
and
claude
committed
8ddfc26
docs: describe the WPT conformance harness

mihar-22
and
claude
committed
d186c31
feat(cue): layout height, background images, animations, class names, and an entity registry hook

mihar-22
and
claude
committed
77062e1
ci: test, type-check, and build on pull requests; manual workflow to record Linux visual baselines

mihar-22
and
claude
committed
c21d1b1
fix(cue): support open-ended (Infinity) end times on the native VTTCue

mihar-22
and
claude
committed
917998a
docs(sandbox): layout model, live track, and custom element scenarios with screenshots

mihar-22
and
claude
committed
453f30e
feat(entities): optional full HTML named character reference table

mihar-22
and
claude
committed
6baa190
docs: entities entry, open-ended cues, sandbox scenarios, and CI

mihar-22
and
claude
committed
c3c7d6b
fix(cue): keep Infinity end times intact on the fallback cue class

mihar-22
and
claude
committed
a346195
feat(cea): CEA-708 pen sizes, edges, window fill and borders, print and scroll directions, display effects, word wrap

mihar-22
and
claude
committed
67bf627
feat(ttml): set animations, image cues, wall-clock time base, drop-frame SMPTE times

mihar-22
and
claude
committed
1bb296a
docs: TTML animations, images, clock time base; CEA-708 fidelity

mihar-22
and
claude
committed
b36c973
style: format workflows and add .prettierignore for vendored and generated files

mihar-22
and
claude
committed
c98cc90
style: format remaining files

mihar-22
and
claude
committed
55085d8

Commits on Sep 3, 2026

build: adopt Vite+ (oxfmt, oxlint, vitest, tsdown) as the toolchain

mihar-22
and
claude
committed
38e6c56

Commits on Sep 7, 2026

feat(cue): per-span styles, inline drawings, clip paths, and an animation model

mihar-22
and
claude
committed
24cf263
feat(renderer): media-synced animations, transform-aware collisions, stacking/lineStep/safeArea options

mihar-22
and
claude
committed
a50b1f2
docs: renderer options, span styles, drawings, and animation model; CSS hooks for small caps and forced cues

mihar-22
and
claude
committed
4274edb
feat(cea): 708 pen backgrounds, font tags, horizontal scroll, wipe directions, ticker windows

mihar-22
and
claude
committed
838850d
feat(ssa): typesetting fidelity: per-span tags, fades, move, \t, karaoke sweeps, drawings, clips, effects

mihar-22
and
claude
committed
ee8b606
docs: SSA/ASS tag coverage after the typesetting work

mihar-22
and
claude
committed
26a4fe3
feat(ttml): container animations, span styles, dropPAL, forced display, padding, and IMSC fixes

mihar-22
and
claude
committed
213f4b0
test(imsc): vendor 62 W3C IMSC test documents and check parsing conformance

mihar-22
and
claude
committed
e93a4e7
docs: TTML/IMSC coverage and test suites

mihar-22
and
claude
committed
a5e8005
feat(track): dedupe option for cues repeated across HLS/DASH segments

mihar-22
and
claude
committed
12192d9
perf(tokenizer): cache tokens per cue, invalidated on text or span changes

mihar-22
and
claude
committed
5c29485
feat(renderer): reduced-motion support and user preset attributes

mihar-22
and
claude
committed
98232a1
ci: size budgets, publint and attw package checks, coverage, API docs, release workflow

mihar-22
and
claude
committed
1204074
docs: reduced motion, caption presets, track dedupe, and the expanded CI/tooling scripts

mihar-22
and
claude
committed
e1073fb
refactor(renderer): split ordering, animation sync, announcer, and style injection into modules

mihar-22
and
claude
committed
fa0b88e
feat: add SAMI and MicroDVD parsers

mihar-22
and
claude
committed
f8be5ea
feat(mp4): demux fMP4 wvtt and stpp subtitle tracks into cues

mihar-22
and
claude
committed
af43fe5
test(bench): parse, tokenize, track, layout, and CEA decode benchmarks with a recorded baseline

mihar-22
and
claude
committed
ddb95ff
perf: O(1) CueTrack membership and 708 commits only after a completed packet

mihar-22
and
claude
committed
918cee1
test(corpus): real-world style fixtures for every format with reviewable JSON snapshots

mihar-22
and
claude
committed
4cbeaa7
fix(parsers)!: harden against hostile input found by fuzzing

mihar-22
and
claude
committed
f0f35c0
chore: exclude corpus fixtures and snapshots from formatting and linting

mihar-22
and
claude
committed
98fe852
chore: enable noImplicitAny across the codebase

mihar-22
and
claude
committed
30e2e0c
feat(playground): interactive visual playground for every parser and renderer feature

mihar-22
and
claude
committed
8231415
fix(vtt)!: drop the non-spec settings-on-a-separate-line tolerance

mihar-22
and
claude
committed
e3ca69a
fix(ssa): lazy tag emission and screen-fixed \clip rectangles on layout-positioned cues

mihar-22
and
claude
committed
ed315f7
fix(styles): do not paint-contain cue boxes so rotated, scaled, and animated text is not clipped

mihar-22
and
claude
committed
a0ca8f0
fix(ssa): rotate and scale around the alignment anchor or \org

mihar-22
and
claude
committed
5f47863
fix(playground): make the SSA \clip sample visibly slice the glyphs

mihar-22
and
claude
committed
548a680
fix(vtt): assigning a region no longer throws in Firefox and WebKit

mihar-22
and
claude
committed
11782f6
test(browser): run the layout suites in Firefox and WebKit too

mihar-22
and
claude
committed
4b40269
feat(renderer): composable core with pluggable features and a media-captions/renderer entry

mihar-22
and
claude
committed
179f1ee
build: size budgets from tree-shaken usage probes

mihar-22
and
claude
committed
081f242
feat(playground): renderer feature presets

mihar-22
and
claude
committed
202e35b
fix(styles): fit image cues to their box instead of tiling

mihar-22
and
claude
committed
2c0ab18
feat(canvas): headless measurement and a canvas writer (media-captions/canvas)

mihar-22
and
claude
committed
0ba6707
docs: canvas exploration findings and a typed style model proposal

mihar-22
and
claude
committed
adfc2e2
feat(playground): canvas view

mihar-22
and
claude
committed
3689a80
feat(model)!: typed cue style model, writers serialise it

mihar-22
and
claude
committed
100fab4
docs: record the typed style model outcome

mihar-22
and
claude
committed
d8b9cd9
fix(renderer): keep vertical cues in place when the overlay is re-laid out

mihar-22
and
claude
committed
c347499
feat(canvas): vertical writing modes

mihar-22
and
claude
committed
910519c
feat(playground): picture-in-picture and fullscreen through a composited captureStream

mihar-22
and
claude
committed
522f12f
docs: migration guide from 1.x, PiP compositing, and vertical text notes

mihar-22
and
claude
committed
bc04599
ci: declare the pnpm version so pnpm/action-setup can install it

mihar-22
and
claude
committed
cb90180
ci: record the pnpm overrides in the lockfile

mihar-22
and
claude
committed
d8e7fbd
chore: format package.json

mihar-22
and
claude
committed
73d0898
test(fuzz): loosen the parse budget on CI runners

mihar-22
and
claude
committed
ec9df1b
ci: run the docs script explicitly and document the new entries

mihar-22
and
claude
committed
36965e9
feat(canvas): ruby annotations and orthographic 3D rotations

mihar-22
and
claude
committed
46dfe50
fix(vtt): ignore mismatched end tags as the spec does

mihar-22
and
claude
committed
793a1c9
feat(parse): add lenient option and run the WPT suite spec-exact

mihar-22
and
claude
committed
f1d88ea
ci: record linux visual baselines as an artifact

mihar-22
and
claude
committed
637ab62
test(browser): add linux visual baselines and run the suite in CI

mihar-22
and
claude
committed
5a4e797
docs: lenient parsing, spec end tags, canvas ruby and 3D, CI visual suite

mihar-22
and
claude
committed
0da8071
fix(canvas): let ruby annotations overflow the line box like browsers

mihar-22
and
claude
committed
af00e53
test(browser): ruby box agreement is engine-aware

mihar-22
and
claude
committed
eb613e7
fix(playground): drive the canvas view on every time change

mihar-22
and
claude
committed
c601b68
fix(canvas): composite cue opacity as a group, like CSS opacity

mihar-22
and
claude
committed
2437af2

Commits on Sep 9, 2026

fix(renderer): compute region boxes from their anchors

mihar-22
and
claude
committed
f47c25c
feat(canvas): match the DOM on region scroll, inline sizes, and line breaks

mihar-22
and
claude
committed
93a9ce2
test(browser): DOM/canvas parity sweep over every playground sample

mihar-22
and
claude
committed
8a984b7
test(browser): compare the alignment edge in the parity sweep

mihar-22
and
claude
committed
659bb05