Repository navigation
Conversation
…traight colour over video (plxnative#1347) Styled ASS subtitles now load their attached fonts, animate heavy karaoke at the video's frame rate on the TV, and keep translucent colour as authored over video. - Fonts: main's 32 MiB attached-font cap is gone. A file's attachments are detached from libavformat as it opens (`detach_fonts`), the source store holds one `Arc<[u8]>` per font, and every libass renderer borrows it through the new facade call `plx_ass_add_font_borrowed` (facade ABI 5) instead of copying it. - Composite (`src/ass.c`): zero-coverage groups are skipped, the row blend is NEON with an exact div255 (the scalar path stays), and regions are published premultiplied through `Painter::tex_premultiplied`. Stats split wall time from thread CPU time; `clock_gettime` links through `-lrt` for firmware 6.4 and older. - Colour over video: the TV compositor blends the UI plane over video as straight alpha, so every partial-alpha pixel came out darkened. While the video plane is bound, a final GPU pass (`gfx/straighten.rs`, `GL_ARM_shader_framebuffer_fetch`) converts the framebuffer to straight colour over the boxes drawn that frame. - Rate: the ASS mailbox pipelines animated frames ahead of the playhead. A pool of up to 3 libass renderers runs below the frame thread's priority (`task::lower_priority`), sized by a pure memory verdict (`ass/pool.rs`: a quarter of MemAvailable for the helpers, the open primary not charged). Dense event boundaries no longer abandon the plan (`Rendered::moves_on`). An adaptive render scale (`ass/scale.rs`, down to 1/2) draws heavy animated stretches smaller, with hysteresis; stills stay full size. - `tests/fixtures/make_ass_stress.py` places per-syllable karaoke effects using libass's real advances. Measured on the TV (webOS 4.10.2, release build, stress fixture): - Subtitle updates in the heavy opening: 7-9/s before this change. With a 24p video it is now a median of 26/s, about 4% late. With the same video at 30p it is a median of 29.7/s, 2.2% late. - Composite p95 fell from 90-107 ms to 15-25 ms at full size. - Peak RSS fell from 541 MB to 287-372 MB. - Placement matches upstream libass 0.17.5 on captured frames, and the phone-recorded subtitle timing matches the test card's frame counter to within about a frame. Closes plxnative#1328
…has one to go to (plxnative#1349) ## What The player's Previous button now always plays the previous episode, and a fresh play has one to go to. - **Previous is the queue step only.** It used to become Restart five seconds into an episode: it sought the playing episode to 0 (rewriting its playhead and viewOffset), and under five seconds it was disabled again. The restart face and its five-second rule are gone. From beginning stays on the Info tab, in the More menu of HUD packages that fold it, and on the detail hero. - **A fresh play gets the show's queue.** Continue Watching, the detail page's Play/Resume, Up Next and autoplay used to POST `uri=<episode>&continuous=1`, which PMS answers with the episode and its successors only, so the queue had no row before the playing one. An episode whose show is known now asks for `uri=<show>&key=<episode>` (no `continuous`), accepts it only when the answer selects that episode, and otherwise falls back to the old request. When rows precede the selection but the window opens on it, the existing window read runs once. - **Leaving an episode keeps its progress.** Previous and Next never seek; the outgoing episode's stop report carries its real position, and the handoff's window re-read waits (bounded, 1500 ms, on the worker leg) for that report, so stepping back to an episode just left resumes it where it was. - HUD plugins: the `restart` face of the `previous` command still parses (published packages keep loading) and is no longer selected. - `tests/mock_pms.py` serves the show-keyed queue. Skip and Up Next prompts are unchanged: while one occupies the row, Previous/Next are not on it (the owner asked for no redesign there). ## Verification - Regression `a_fresh_play_of_a_middle_episode_lands_the_episodes_before_it` failed against the old request and passes; hostsim rig tests: Previous mid-episode plays the prior episode, seeks nowhere and the outgoing episode reports its real position; Previous on the first episode is disabled; Next. - Crate suites: plx_media 2015, plx_appkit 953, plx_screens 2137, plx_plex 801, plx_data 1540, application hostsim 1529, plx_media hostsim 2295 passed; `make check-nodef`, `check-deps`, ledger ratchet, localization, module layers, `tests/test_harness.py` (725), mock_pms selftest green. - TV, debug build on a real server, owner-driven: the play's queue came back as the show's (`playqueue: … rows=30`), `queue transport: Previous` went to the episode before and `Next` came back; the owner checked it. Not verified: whether PMS lists season-0 specials before S1E1 (Previous from S1E1 could land on one); PMS's window size on a long show; `tests/playback_navigation_sim.py`'s Previous block (needs `make sim-macos`; CI). The frozen published schema descriptions `site/hud/schema/hud/1.0.json`/`1.1.json` still say Previous may be "a restart of this one". 🤖 Generated with [Claude Code](https://claude.com/claude-code)
…xnative#1350) ## What The v0.10.0 release note and audit, revised for the third candidate, which is promoted to the release. - **Note:** a Fixed line for the styled-ASS work ([plxnative#1347](plxnative#1347), issue plxnative#1328), measured with a test file on the development set and not the reporter's file; the Previous line from plxnative#1349 now says the show-built queue was seen working against one real server; the Next and Previous feature line no longer speaks of a "queue". - **Audit:** the third candidate's range (plxnative#1347, plxnative#1349); the 2026-11-12 `--fps --mock` runs on the development set with their failures and the interleaved parent comparison; the second candidate's package installed over the stable install and byte-checked; per-change rows for plxnative#1347 and plxnative#1349; a known-issue line for Previous from a show's first episode (specials order undetermined); plxnative#1347's build changes; the action-required check re-run (no section owed). `ci/check-package.py --lint-note` and `--lint-audit` pass. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
…lxnative#1352) ## What The v0.10.0 release note's title and Home line now say what a viewer sees (a redesigned Home, what you were watching one press away) instead of the design name "stills", and the styled-ASS fix leads with "ASS we can". Wording only; `ci/check-package.py --lint-note` passes. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
…ed in plxnative-huds (plxnative#1353) ## What The v0.10.0 release note's title now names yellow subtitles, which people asked for. Its Player HUD line says that HUDs for the gallery are made and submitted in the separate [plxnative-huds](https://github.com/GLinnik21/plxnative-huds) repository, with a link to its authoring guide. Wording only; `--lint-note` passes. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
…lxnative#1354) ## What The v0.10.0 release note's title leads with the most-requested items this release answers: automatic intro skipping (discussion 826), yellow subtitles (1193), Next and Previous episodes (829), Play Version (521), then the replaceable player HUD and the styled-ASS fix. Wording only; `--lint-note` passes. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
… Image[], item-menu trigger) (plxnative#1355) The Site video run after v0.10.0 failed, so the site kept the v0.9.0 film and stills. Three things broke it after the last green render on 2026-10-24: - The pinned FFmpeg was pruned: BtbN deleted autobuild-2026-10-23-13-04. The new pin is autobuild-2026-11-12-15-40 (n9.0.2-57-gd838f9f9ae, linux64 GPL); its sha256 matches BtbN's asset digest, and the build has libsvtav1, libx264, libaom and libvmaf. - The demo mock's rows carried no Image[], so a Continue Watching card fell back to the legacy /clearLogo lookup. For Metropolis, which has no logo, that returned a 404, and the frame dump fails closed on it. Demo rows now carry Image[] the way a modern PMS sends it, with a stamped clearLogo only where the title has one, and a season or episode names its show's art. The detail fixtures were regenerated. - The plxnative-itemmenu trigger's pre-check refused whenever the page was at the hero, which is the whole time the Continue Watching row holds focus (since plxnative#1290). It now asks HomeScreen::item_menu_reachable, the one rule the hold itself uses. A Site video dry run on the head passed every render gate. Regression tests: CatalogImageLists, the_demo_rows_carry_the_image_list_a_modern_pms_sends_with_a_stamped_logo, the_item_menu_trigger_is_answered_on_the_deck_row_while_the_page_rests_on_the_hero and the_item_menu_is_reachable_on_the_deck_row_at_the_hero_page_and_on_a_shelf_only_on_the_grid_page.
Rendered by the site-video workflow from main@22eb2a9c (22eb2a9), film inputs tree hash 138fd0413097, on the pinned Linux/llvmpipe stack: two renders agreed frame for frame, the hold-injection gate and the sentinel scan passed, and every quality gate passed (43 gates: 43 pass). The 35 simulator-made stills (documentation figures, the site's close-ups, the link card, their WebP and phone copies and glows) were re-rendered by the same run from main@22eb2a9c (22eb2a9), inputs tree hash 4d03c76607eb, on the pinned Linux/llvmpipe stack: every scene was captured twice and the pair agreed within its noise bound. Photographs taken on the television are not regenerated. Run: https://github.com/GLinnik21/plx-native/actions/runs/43272025772 The checks prove stability, not taste. If this looks wrong, revert this commit; nothing else depends on it.
…st (plxnative#1357) BtbN prunes dated autobuilds, and two FFmpeg pins have been lost that way. The pinned tarball is now mirrored, with the same sha256, as a prerelease asset of this repository (mirror/ffmpeg-btbn-autobuild-2026-11-12-15-40, never latest). The pin lists the mirror first, with BtbN's URL as the fallback, and its comment says a future re-pin mirrors its tarball the same way.
…ive#1363) Dependabot bump of the node-tools group: @github/copilot 1.0.105 -> 1.0.108 in tools/package.json and its lockfile.
…-11-13 # Conflicts: # ipkroot/ctl/control # pkg/appinfo.json # rust-modules/Cargo.lock # rust-modules/Cargo.toml # rust-modules/media/src/route/decision.rs # rust-modules/plex/src/plex/timeline.rs # rust-modules/session/src/auth/owner.rs # rust-modules/src/app/run.rs # tests/test_harness.py
The machine-shape gate pins every Cargo manifest and the lock by hash. Regenerated both certificates for this workspace (the jellyfin crate and FinNative's manifests) and listed the i18n source, which differs from upstream only in CONTRIBUTE_URL.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Merges upstream
mainthrough plxnative#1364, which includes the 0.10.0 release. This lands as a merge commit.Conflicts
ipkroot/ctl/control,pkg/appinfo.json,Cargo.toml,Cargo.lock): kept FinNative's 0.1.0 and app id.PLXNATIVE_BASEmoves to0.11.0-dev, because upstream main is now past its 0.10.0 tag.session/src/auth/owner.rs: upstream's sans-io rewrite (refactor(sans-io): complete machine input and effect boundaries plxnative/plx-native#1358) folded the owner's helpers into its event arms. FinNative's session rules are ported into that new structure:Command::SignOutUser) now goes through theStartSwitcharm. That arm writes the shortened roster with the picker's commit and puts the remaining users on the picker. When the last user signs out, it does a fullSignOut.roster_refresh_applies), and the picker's start follows the same rule.can_switch_profilesstill refuses a switch when there is no one to switch to.HomeRosterstill carries the session to the Jellyfin roster.start_switchmarker follows the arm's new pattern.plex/timeline.rs: the Jellyfin play-queue hook is kept ahead of upstream's new show queue.media/route/decision.rs: both sides kept (movie_row_playablenext to upstream'sepisode_show_rk).src/app/run.rs: both sides kept.tests/test_harness.py: ledger pins are upstream's plus FinNative's own (workers 5, globals 22, forks 63).Also
ci/ports.inilistssession/src/auth/finnative.rsas a worker file. Its Quick Connect code window reads the clock, and upstream's new wall-clock gate only allows that in listed port files.Verification
ci/check-deps.sh: all gates green.plx_jellyfin: 147 passed.plx_screens --features finnative: 2150 passed.plx_session --features finnative: 345 passed. The 7 failures are real-curl or TLS discovery tests that need network or loopback this container doesn't have.plx_plexlib tests fail only in the known container-only areas (persistence, servers, http, TLS self-test).cargo check --workspace --lib --tests: clean.Generated by Claude Code