Probes UNMASKED_RENDERER_WEBGL once per page via a throwaway WebGL context; defaults the lava subsystem off on a positive software-renderer match or total WebGL failure; releases the throwaway context via WEBGL_lose_context after reading the renderer string to avoid exhausting the browser's per-page context limit.
Byte cap (96MB) unchanged as the hard OOM bound; the wider time window only lets sparse Opus use existing memory headroom to ride out decode jitter. Diag logs pin whether the block is back-pressure or decode throughput.
Tracks whose total audio falls below the playback-start threshold (Opus <1s lead, WAV <6 buffers) silently hung loaded-but-not-playing. After MarkStreamCompleteAsync, call TryStartPlaybackAsync when _streamingPlaybackStarted is still false so the scheduler drains its buffers and fires onPlaybackEnded exactly once.
pause() clears underrun_ so setStreamComplete can't fire TrackEnded while paused; resetToStart() resets streamComplete. Prior fix: underrun_ park + streamComplete discriminator prevent the Opus-startup false-end. Tests: 18 PlaybackScheduler cases including pause-during-underrun and underrun->resume->genuine-end-once.
Opus resolved its 48kHz context lazily on the first chunk, close()ing and rebuilding the live graph mid-decode. Move the recreate into initializeStreaming so it runs before any bytes flow; the lazy call early-returns. WAV path unchanged.
A second LoadTrackStreaming for the same in-flight track (UI double-fire, queue re-entry, or JS false-end auto-advance) is now dropped; a different-track load still supersedes. Targets the Opus double-load; keeps load-gen diagnostics.
Finish the Settings "Apply" behavior so changing streaming quality mid-track
switches format immediately instead of only persisting the cookie for the next
play.
- SettingsMenu reads the AudioPlayerProvider cascade and threads the player into
StreamQualitySetting as an explicit parameter (the MudMenu panel portals to
MudPopoverProvider, outside the cascade scope, so a [CascadingParameter] there
lands null). StreamQualitySetting's Apply persists the cookie, then asks the
player to reload preserving position.
- Add a "load at timestamp" path to the player rather than restart-from-0-then-
seek (which audibly played the start and raced the just-started scheduler into
a crash). ReloadPreservingPositionAsync loads the track in the newly-resolved
format beginning DIRECTLY at the saved position:
* new JS resolveStreamOffset(position) resolves the file-absolute byte offset
with no playback/buffer state (Opus from its sidecar immediately; WAV after
a header probe),
* StartFromPositionAsync converges onto the existing seek/refill loop
(RunSegmentedStreamAsync with a non-null seekPosition) so the decoder
reinitializes for a header-less Range continuation and starts playback at
the target,
* ProbeHeaderAsync feeds the byte-0 segment to the decoder WITHOUT starting
playback until the WAV header parses (bounded by 256 KB); the probe buffers
are dropped by the continuation's clearForSeek, so nothing is audible.
- IStreamingPlayerService gains ReloadPreservingPositionAsync; the QueueService
test fake implements it.
Document the three-layer memory bound (raw queue, decoded queue, network) in the streaming seam after the HW-accel-off OOM fix landed on streaming-overhaul.
The inter-segment back-pressure gate matched WAV byte density but let a 4MB Opus segment (~100s at 320kbps) decode eagerly into main-process RAM, OOMing the tab with HW accel off. Drain per chunk past high-water, gated on playback start. Adds load-generation diagnostics for the double-load hypothesis.
RecoverFromFailedRefill now sets IsStreamingMode=true so the in-place
seek-retry route isn't wedged. The generic-catch unload path is gated on
the loadCts identity, so a superseded load no longer clobbers a newer
operation's state.
cursor>=totalLength is the sole forward-EOF test; a short non-final body is
a truncation error, not EOF. Mid-stream forward-load failures now invoke
RecoverFromFailedRefill so the scheduler halts instead of a silent false end.
Two regression tests pin both paths.
Replace the open-ended forward GET with sequential bounded bytes=start-end
segments, the next fetched only when the scheduler drains below low-water,
so the browser holds ~one segment regardless of file size. Seek converges
on the same loop. Strip BP-DIAG.
Temporary, grep-tagged diagnostics at the read-loop pause, the scheduler
latch, and the chunk-result path to show whether ProductionPaused latches,
reaches C#, and parks the loop. Strip once the cause is confirmed.
Without SetBrowserResponseStreamingEnabled the browser buffers the whole
body before yielding, so the Phase 21.2 read-loop pause backpressured an
already-downloaded payload. Set it on both the initial and seek/refill
requests; safe no-op on the SSR path.
C6/AC8: IsStillActiveSeek() predicate guards all three SeekBeyondBuffer
failure exits, so a superseded seek never recovers over a newer seek's
state. AC6: empty scheduler routes to seekBeyondBuffer so a same-target
retry (seek or play) refetches instead of no-oping.
Seek-back past the retained tail reuses the existing seek-beyond-buffer
Range path (per-path resolver). A failed refill now halts the scheduler
into a paused-but-loaded state (AC6) instead of a silent false end.
Skip the back-pressure interop poll while paused (UC5). Document complete()
draining the stash in full by design. Rename scheduler isProductionPaused to
evaluateProductionPause (latch-advancing); window exposure name unchanged.
Shared scheduler fill signal (forward water-marks + hard byte cap) pauses
the C# read loop above high-water and, for Opus, stops the demux/decode
feed so WebCodecs queues stay near-empty. Routes through the existing
cancellation discipline; releases the latch on clear/seek.
The inclusive <= bound is correct; comments now say 'at or behind'. New
test drives eviction through the real onended trigger with a live mid-array
source pinning the frontier.
Drop already-played buffers from the front while advancing the time
anchor so position/index bookkeeping stays exact. Shared by both decode
paths, no format branch. Back-retain is a config seam for 21.2.
Window both the WAV StreamDecoder and Opus WebCodecs paths feeding one PlaybackScheduler — shared eviction, per-path back-pressure; reuse the now-live index-driven Opus seek for refill. Drops stale approximate-seek language; adds OQ6/OQ7.
Surface the sidecar duration on the first Opus chunk instead of gating it on the first decoded buffers; C# locks UI Duration on chunk 1, and async WebCodecs decode left it at 0 — killing seek and the duration-gated visualizer.
Seek now trims the lead-in so playback lands at the requested time, not the page start; decoder drain polls decodeQueueSize (bounded) instead of a single timeout. Minor cleanups.
Previous probe sample had invalid Ogg page CRC32s, so Chrome/Firefox rejected it and the capability check always returned false. New 176-byte libopus sample has verified-correct CRCs. Adds structural-validity tests.
Both SettingsCookieService and DarkModeCookieService now call window.DeepDrftSettings.setCookie (new Interop/settings/settings.ts) instead of eval. New tests cover SettingsServiceBase parse/format round-trip and the PreferenceAwareStreamingPlayerService invariant (Lossless skips probe; LowData inherits base).
daniel-c-harvey