feat: offer timed opening and ending skips

2026-10-22 13:24:37 +02:00
parent 7b53b00b15
commit 3d949154a2
8 changed files with 165 additions and 10 deletions
@@ -31,7 +31,7 @@ The native resolver requests watch-route data and safely follows bounded same-or

 Titles come from returned series/selected-episode metadata. Only `op`/`mixed-op` and `ed`/`mixed-ed` become opening/ending bands. Bands retain fractional seconds, require finite positive player-reported duration, and reject nonfinite timestamps, originally reversed/zero-length intervals and wrong-episode timings. Display intervals are clipped to `[0, duration]`; intervals with no positive-length overlap are hidden. Raw timestamps remain unchanged, and duration changes recompute clipping without shifting or scaling. Optional provenance/votes/reference-length checks remain in force; a finite positive reference length differing from actual duration by more than 5% rejects the interval, and missing fields are not invented. Highest-vote eligible opening and ending win, with response order breaking ties. Overlaps are not merged.

-The controls mark interval boundaries with small rounded gaps in the slider track, without visible labels or boxes. Buffer rendering shares the segmented track; accessibility descriptions retain opening/ending names and times. Neither automatic seeking nor skip buttons are added. Bounds do not prove synchronization across cuts. Metadata tasks are bound to identity and lifecycle generation; replacement clears intervals and rejects stale results, including A → B → A.
+The controls mark interval boundaries with small rounded gaps in the slider track, without visible labels or boxes. Buffer rendering shares the segmented track; accessibility descriptions retain opening/ending names and times. While acknowledged, seekable video playback is inside a band, a **Skip Opening** or **Skip Ending** button offers a seek to its end for three seconds of monotonic time, including while paused. Each kind is offered once until close or media-identity replacement; pointer movement and seeking back do not reoffer ignored segments. Pending seeks and slider previews cannot expose a stale offer. The same action routes to local or Cast playback and retains playback intent. No automatic skipping is performed. Bounds do not prove synchronization across cuts. Metadata tasks are bound to identity and lifecycle generation; replacement clears intervals and rejects stale results, including A → B → A.

 ### External resolved-source compatibility

@@ -169,6 +169,8 @@ Metadata implementation verification (2026-10-22): protocol/security/bounds chec

 Controls alignment correction: removed the obsolete label-height reservation after switching to gapped segments. A native offscreen SwiftUI/AppKit layout smoke reproduced a 9-point top-row shift when intervals appeared; after restoring the standard 14-point slider frame, volume and timeline geometry match exactly with and without intervals. Production build passed; visible confirmation remains a human check.

+Segment-skip verification (2026-10-22): local HLS/VTT playback exercised late metadata, the controlled-clock three-second boundary while paused, persistent dismissal, pending-seek suppression, opening/ending seeks, exact media-end seeking, close/reopen reset and playing/paused intent. Exact-end HLS seeks initially failed with `Operation not permitted`; the shared native seek path now finishes network inputs at EOF instead of asking the demuxer to seek beyond its final packet. Network playback regression checks, metadata checks, production build and TLS loopback Cast segment skipping/restoration passed. Rapid play-then-seek probes encountered Core Audio startup errors on the active Bluetooth output; the settled-playback smoke waits for actual advancing playback and passed. Visible button placement, rapid audio transitions and physical-TV behavior remain unverified.
+
 ## Remaining limits

 - Confirm title readability and gapped slider segments in normal/compact controls, buffer/playhead visibility and subtitle alignment in the visible app using Tubular's real launch path. Offscreen native rasterization at compact/normal widths confirmed all four interval-boundary gaps remain clear even with full buffer coverage.