Repository navigation
Releases
FastPlay v0.5.0
FastPlay 0.5.0 adds a compact native menu and fast marked-range MP4 export,
while preserving the single-player architecture and GPU-resident playback path.
Highlights
Native menu without permanent chrome
- Media, Playback, View, and Help commands are available from a compact native
menu row. - Press
Ctrl+Shift+Mto show or hide the menu. The choice is remembered for
new instances, andAlttemporarily reveals a hidden menu. - Switching menu visibility preserves the client size and screen position.
Export a marked clip
- Set In and Out points, then press
Ctrl+Shift+Eor choose
Media → Export Marked Range. - MP4-family inputs use stream copy when the In point is keyframe-safe;
otherwise FastPlay transcodes to H.264/AAC for accurate boundaries. - Export runs asynchronously, refuses overwrites, cleans temporary files after
failure, and reports completion in the player. - The MSI and portable ZIP include
ffmpeg.exeandffprobe.exe, so export
works without a separate FFmpeg installation.
Additional polish
- Screenshots are saved as PNG.
- Frameless hit testing and middle-click input behavior are corrected.
- Borderless fullscreen covers the whole monitor with the menu docked, and
leaving fullscreen restores the exact previous window size. - The local benchmark harness can run reproducible open/seek baselines over
multiple corpus directories.
Upgrade notes
- Existing installs upgrade in place through the MSI major-upgrade path.
- No settings migration is required.
- Packaged FFmpeg tools increase the installer and portable archive sizes.
Validation
cargo fmt --checkcargo check --all-targetscargo test --all-targets(281 unit tests and 4 integration tests)cargo clippy --all-targets -- -D warningscargo build --releasecargo wix- portable ZIP contents and MSI administrative extraction verified, including
ffmpeg.exeandffprobe.exe - packaged FFmpeg tools produced and reprobed a 2.500-second H.264/AAC clip
- portable software-fallback and MSI-extracted hardware-path launches reached
responsive playback windows with the expected media title
FastPlay v0.4.6 (Stable)
Note: FastPlay is now stable enough that I use it as my default media player. If there’s a feature you’d like to see, please open an issue. Otherwise, I’ll be shifting my focus to other projects while continuing to maintain FastPlay as needed.
FastPlay 0.4.6 improves window behavior, portrait video playback, and Windows app icons.
What’s new
- Frameless mode: Press
Ctrl+Shift+Sto hide or restore the normal window frame. FastPlay remembers your preference. - Better portrait video handling: Phone videos now open in a properly sized portrait window instead of leaving large empty areas on the sides.
- Improved window controls: Frameless windows can still be moved, resized, maximized, and used fullscreen normally.
- Updated FastPlay icon: Improved icons for the app, installer, Start menu, and video file associations.
- Portable ZIP improvements: The portable Windows build is now packaged directly from the FastPlay repository for more consistent releases.
Existing settings and installations continue to work normally. No account, migration, or internet connection is required.
FastPlay v0.4.5
FastPlay 0.4.5 is a reliability release, with two fixes worth calling out.
A leak that only bit people who keep many players open at once: with a dozen
instances running, one would eventually freeze, refuse to close, and sometimes
take other applications — including Windows Explorer — down with it.
And a scrubbing bug that could silently kill video for the rest of a file,
reachable on any clip that falls back to software decode.
Fixes
Many instances open → a frozen, unclosable window
Release builds never freed a single GDI object. Every cleanup call in the
overlay rasterizers was written debug_assert!(DeleteObject(..)), and
debug_assert! does not merely skip the check when assertions are off — it
does not evaluate the expression at all. All 36 delete sites across the seven
rasterizers (subtitle, timeline, timeline label, idle, help, recent, volume)
were affected: roughly 8.5 handles leaked per timeline-overlay rebuild,
measured 25 → 709 over 40 seek pairs.
Windows allows 10,000 GDI handles per process but only 65,536 across a
desktop session, so a dozen instances exhausted the shared pool at ~5,400
each — before any single player hit its own limit. Every process in the
session then began failing GDI allocations, which is why unrelated
applications crashed alongside FastPlay.
Six CreateDIBSection sites also leaked the font and device context on their
error path. All deletes now route through helpers that run in every build,
with a test that fails if a raw delete reappears outside them.
Scrubbing could kill video for the rest of the file
Seeking while the video decoder was reopening left the picture frozen for the
remainder of that file — audio kept playing, the timeline kept moving, and only
reopening the file brought video back. Reachable by scrubbing any clip that
falls back to software decode.
The decoder's control channel is deliberately shared and outlives the worker
thread, so holding one was never evidence that a worker was alive. The
coordinator asked exactly that question, so a worker that had exited on a
cancelled open left its channel registered forever, and every subsequent seek
went to a channel nobody was reading.
Liveness alone was too blunt a fix: "no worker is running" has two causes that
need opposite responses. A worker that died on a cancelled open must be
respawned; one that exited because the file has no video stream at all must
not, or every seek reopens and re-demuxes the file to rediscover the same
absence. Seek delivery is now three-way, and both cases are handled. Measured
on an audio-only .m4a over 8 seeks: 9 video-worker spawns before, 1 after.
A fatal error left the window hung instead of reporting
Closing FastPlay deliberately stops its workers and exits without releasing
the D3D11 device in-process — that release intermittently faults inside the
graphics driver, and Windows Error Reporting then freezes the still-visible
window for seconds while it writes a crash dump.
Every error path bypassed that discipline. A fatal error returned straight
out of the event loop, dropping the session on the way — releasing the device
while the decode and audio workers were still running inside those very
drivers. So the GDI failure above surfaced not as an error message but as a
hung window parked in WER. Clean close and fatal error now take the same
route: persist progress, stop workers, flush the trace, exit.
Diagnostics survived only one instance
Every player wrote the same session.log and crash.log, truncating on each
write, so concurrent instances erased each other's traces — the evidence for
the freeze above had to be reconstructed from the Windows event log instead.
Each run now writes session-<utc-stamp>-<pid>.log and its crash counterpart,
swept after 7 days.
Also in this release
Work that landed since 0.4.4 and ships here for the first time:
- a new app icon: the old one was the full tile artwork, whose dark plate
dominated at small sizes — the 16px entry was 97% opaque dark pixels, reading
as a black square with an illegible glyph in the title bar and taskbar. It is
now the glyph alone on transparency - audio-only files reach the end-of-file state on seek, the error overlay
displays rather than failing silently, and demuxer packet leaks are closed - playback follows the default audio endpoint when Windows switches devices,
with deduplicated recovery that preserves play/pause state - COM notification teardown ordering fixed; recent-file writes are atomic
- the pure overlay rasterizer and the keyboard keymap moved out of the unsafe
FFI seams into tested modules
Validation
cargo fmt --check,cargo clippy --all-targets -D warnings,cargo test
(260 passing, including new UTC-stamp, retention, GDI source-invariant, and
run-log resolution tests),cargo build --release,cargo wix- GDI growth over 40 seek pairs: +708 → 0, flat across two runs
- Fault injection at the real GDI failure site: before, no worker-exit lines
in the trace and the device released under live workers; after, workers
stop first, exit code 1, no Windows Error Reporting event - Development build, twelve concurrent instances tiled so none is occluded,
40 seek pairs each: GDI handles 300 → 606 and climbing on0.4.4, flat at
192 on0.4.5; all twelve closed from the title-bar X, and each wrote its
own intact log - Installed MSI, same twelve-instance test against
C:\Program Files\fastplay\bin\fastplay.exe, plus in-place upgrade over
0.4.4, Start Menu shortcut,.mp4association, and uninstall/reinstall
Upgrade notes
- Existing installs upgrade in place through the MSI major-upgrade path.
- No account, network, or cloud storage is required.
- Old
session.log/crash.logfiles are not read any more and can be
deleted; new per-run logs are swept automatically after a week.
FastPlay v0.4.4 - Native HDR Output, Full-Range PQ, and 120fps Pacing
FastPlay 0.4.4 caps off a same-day pair with 0.4.3. Together they bring native HDR output, and this release adds two follow-up fixes found immediately afterward: full-range PQ video, notably Topaz Video AI "HDR Enhanced" exports, now plays, and high-frame-rate content at 120 fps now presents every frame.
HDR, since v0.4.3 (same day)
On a display with Windows HDR enabled, HDR10 (PQ) and HLG video, including iPhone recordings, presents natively in HDR on a 10-bit PQ swapchain, with the stream's HDR10 static metadata passed to the display and HDR-aware overlays and screenshots. On SDR displays, the same content is tone-mapped to SDR in a pixel shader. The decision is made per file open, based on the monitor the window is on. Details are in the [v0.4.3 notes](https://github.com/CalvinSturm/FastPlay/releases/tag/v0.4.3).
Fixes in v0.4.4
Full-range PQ video failed to open
Topaz Video AI's "HDR Enhanced" mode exports genuinely PQ-encoded video in 8-bit full-range H.264, a combination FastPlay's HDR path previously declined at open as unsupported. It now plays like any other HDR content: natively on an HDR-active display and tone-mapped on SDR. All safety checks are unchanged. Non-standard matrices still decline, SDR content never enters the HDR path, and the shader output is pixel-verified exact against spec math.
One frame in six dropped at 120 fps
The clock that video frames schedule against advanced in approximately 10 ms steps because Windows only updates the audio position once per engine period. At 120 fps, that made two frames come due at once on most steps, causing the scheduler to drop one. This created a structural loss of approximately 17% that never appeared at 60 fps or below.
The clock now advances smoothly between updates. Measured on a 240 Hz display, a synthetic 120 fps clip went from 405 dropped frames to 6 of 2400. Real 120 fps files improved from approximately 83% of frames presented to approximately 99%. A/V sync is unaffected because the smoothing is bounded to 12 ms and re-converges with every update. Presenting 120 fps still requires a display faster than 120 Hz.
Validation
cargo fmt --check,cargo clippy --all-targets,cargo test
(199 passing, including new clock-smoothing and signal-resolution tests),
cargo build --release,cargo wix- Full-range PQ shader output: 0/1023 versus full-range BT.2020 spec math
(bench/verify-colors-pq.ps1 -Mode shader-pq -FullRange); all other
pixel benches green; scrub → pause → resume audio behavior re-verified.
Upgrade notes
- Existing installs upgrade in place through the MSI major-upgrade path.
- No account, network, or cloud storage is required.
FastPlay v0.4.3 - Native HDR Output
FastPlay 0.4.3 is an HDR release. On a display with Windows HDR enabled, HDR10 (PQ) and HLG video, including iPhone recordings, now present natively in HDR instead of being tone-mapped down to SDR. On SDR displays, the v0.4.2 tone-map path is unchanged.
Highlights
- Native HDR output. HDR10 PQ and HLG present on a 10-bit PQ swapchain when the display is HDR-active; SDR displays keep the tone-map path.
- HDR10 static metadata (mastering display, MaxCLL/MaxFALL) is read from the stream and handed to the display pipeline.
- Overlays and screenshots are HDR-aware: subtitles and OSD render at the correct brightness on the HDR chain, and screenshots of HDR playback come out as normal SDR images.
- Fixed audio occasionally staying silent after scrubbing plus pause/play.
- Fixed the app icon losing its transparency.
Notes
- The HDR-or-SDR decision is made per file open, for the monitor the window is actually on. Moving a playing HDR window onto a non-HDR monitor mid-playback looks washed out until the file is reopened.
- The conversion is FastPlay's own pixel shader on both paths because the GPU video processor offers no usable HDR conversions on real hardware. PQ passes through bit-transparently; HLG is completed to display light and PQ-encoded. The tone map stays fixed (203 cd/m² diffuse white, knee 0.75) and is not metadata-driven.
- The audio fix: scrub seeks pause the audio sink without clearing it, and resuming afterward could leave it permanently wedged if its buffer was full. The resume path now resets the sink, which also stops a brief burst of old-position audio.
- The SDR path is byte-identical to v0.4.2 (±2/255 versus an ffmpeg reference).
Validation
cargo fmt --check,cargo clippy --all-targets,cargo test
(198 passing),cargo build --release,cargo wix- Shader output pixel-verified against CPU models and spec math: PQ
passthrough 0/1023, HLG→PQ ≤1/1023, overlays 0/1023, end-to-end playback
≤1/255, and static metadata round-trip exact, each with negative controls;
full bench suite inbench/. - Real 4K60 HEVC HLG plays sustained on hardware decode through the HDR
path; multi-monitor behavior validated manually.
Upgrade notes
- Existing installs upgrade in place through the MSI major-upgrade path.
- No account, network, or cloud storage is required.
FastPlay v0.4.2 - HDR Playback, Audio-Only Files, and Crash/Correctness Fixes
FastPlay 0.4.2 is a media-compatibility and correctness release. HDR video
(HDR10 PQ and HLG — including iPhone recordings) now plays, audio-only files
play, and three long-standing defects are fixed: audio going silent after a
scrub, a stretched picture when zooming a rotated video, and a hard crash when
switching files rapidly through a play queue. No charter changes:
PlaybackSession remains the single, concrete coordinator.
Highlights
- HDR video plays. HDR10 PQ and HLG are tone-mapped to SDR in a pixel shader.
- Audio-only files play (
.mp3,.flac,.wav,.ogg,.aac,.m4a,.opus). - Fixed audio going permanently silent after scrubbing early in a file.
- Fixed the picture stretching when zooming a rotated video.
- Fixed an access-violation crash when spamming PageUp/PageDown through a queue.
Features
HDR playback (HDR10 PQ and HLG)
HDR-tagged video previously failed to open. It is now tone-mapped to SDR in
FastPlay's own pixel shader: the decoded NV12/P010 frame is sampled through
per-plane shader-resource views, then converted BT.2020 NCL YCbCr → R'G'B' →
PQ or HLG EOTF (plus the HLG OOTF) → normalized to diffuse white (203 cd/m²,
per BT.2408) → a knee/shoulder tone curve that rolls off highlights → BT.2020 →
BT.709 → sRGB, into the existing SDR swapchain.
This deliberately does not use the GPU video processor's HDR→SDR conversion,
which was the previous design. That approach is unreachable on real hardware:
on an NVIDIA RTX 3080 Ti, the video processor advertises no HLG (GHLG)
input conversion at all — no format, no output space — and accepts PQ only to
linear-scRGB or HDR10 outputs, never to the gamma-2.2 sRGB an 8-bit SDR
backbuffer scans out. Doing the transfer math ourselves is both portable and
exact.
Fixes
Audio silent after an early scrub
The audio worker decodes independently of video, and its cancellation predicate
watches the command sequence — which any seek bumps. A seek arriving while the
worker was still opening the file cancelled that open, and the worker exited
permanently. Nothing brought it back: the coordinator gates its respawn on the
existence of the worker's control channel, which is an Arc that outlives the
thread. Video played on; audio was gone for the rest of the file. Intermittent,
because it only fires when the seek lands inside the audio-open window — which is
why it favoured large files.
The audio open now distinguishes "this file has no audio stream" (permanent —
exit) from "a seek superseded the open" (transient — serve the seek and reopen).
Stretched picture when zooming a rotated video
Zoom/pan clips the destination rect and maps the visible region back to a source
rect to sample. That mapping was done on the source's own axes, while the base
rect was already built from rotated display dimensions. At 90°/270° the
destination's horizontal axis is the source's vertical one, so the crop came
off the wrong axis with the wrong shape and the rotate-then-fit smeared it. At
180° the axes line up, but the crop was taken from the opposite side, so a
panned zoom framed the wrong region.
Crash when switching files rapidly (access violation)
Spamming PageUp/PageDown through a play queue killed the process with an access
violation, usually after 30–50 presses.
FFmpeg's D3D11VA hardware-device context (AVD3D11VADeviceContext) takes
ownership of the ID3D11Device it is given: it releases the device when the
decoder is torn down, and does not AddRef it on init. FastPlay was handing it a
borrowed pointer, so every decoder teardown released the shared device against an
AddRef that never happened. A single open survives that — the device carries
dozens of other references — but queue navigation is a full open+teardown per
press, so each one nets one release. The refcount walks to zero, the device is
freed out from under the swapchain, the decode workers and the video processor,
and the next call faults inside d3d11.dll.
Present since before this release line; not introduced by the HDR work.
Technical notes
- HDR is tone-mapped to SDR, never passed through. On an HDR display it is
still shown as SDR: native HDR passthrough is not implemented, so HDR content
will not use the display's full brightness or gamut. - Tone mapping is fixed (no exposure or curve controls) and ignores
mastering-display / content-light metadata, so grading choices in the source
are not honoured exactly. - HDR is declined cleanly at open on a device that cannot sample the decoder's
output format, rather than erroring at the first draw (which device recovery
misreads as device-lost and retries forever). - The SDR path is unchanged: it still blits through the D3D11 video processor and
measures within ±2/255 of an ffmpeg reference decode.
Validation
cargo fmt --checkcargo clippy --all-targetscargo test(187 passing)cargo build --release,cargo wixbench\verify-colors.ps1— SDR backbuffer within ±2/255 of an ffmpeg reference.bench\verify-subtitles-hdr.ps1(new) — subtitle overlays composite correctly
on top of HDR video, with the picture beneath them untouched.- HDR tone-map verified by pixel against a CPU model of the shader, on
synthetic HLG and PQ clips with known values: exact match, 0/255 delta on every
channel, including seven unclipped PQ midtones. - Rotated zoom verified by rendering a clip containing a perfect circle: a 90°
display matrix rendered it at 720×280 (aspect 2.571) before, 448×450 (aspect
0.996) after; unrotated and 180° clips measure identical before and after. - Queue-switch crash: previously crashed after 32–48 presses on mixed, SDR-only
and HDR-only queues; now survives 150 presses at 35 ms and 120 at 55 ms on all
three. - Media-shape regressions: A/V, audio-only
.mp3, video-only.mp4, and 4K60
HLG all play and seek.
Upgrade notes
- Existing installs upgrade in place through the MSI major-upgrade path.
- No account, network, or cloud storage is required.
FastPlay v0.4.1
Stability and playback-robustness release.
Highlights
- Fixed an intermittent close crash/freeze caused by fragile D3D11 GPU teardown on exit.
- Fixed choppy audio on heavy 4K60 files when video falls back to software decoding.
- Audio playback is now protected from slow video decode paths using an independent audio decode worker.
- Improved shutdown reliability while preserving recent/resume state.
- Added env-gated audio diagnostics for underruns, queue depth, and WASAPI behavior.
Technical notes
- Some high-resolution H.264 profile/format combinations may be rejected by D3D11VA/NVDEC and fall back to software decode.
- FastPlay now keeps audio realtime even when video decode cannot sustain full framerate.
- Under overload, video may degrade first (late frames dropped) instead of starving audio.
Includes PR #16 (close-crash fix) and PR #17 (4K60 software-decode audio fix). Full notes: docs/release-notes-v0.4.1.md.
Installer: fastplay-0.4.1-x86_64.msi
SHA-256: 66D153DB67EBF130F3FD048235D136C49C9EC9310B9AFFFFAEA303287498F7AD
FastPlay v0.4.0 - Lightweight Play Queue and Folder Playback
FastPlay 0.4.0 adds a lightweight, in-memory play queue: drop several files or
a folder, step through them with PageUp / PageDown, and have FastPlay
auto-advance to the next file when the current one reaches its natural end. It
stays true to the charter — PlaybackSession remains the single, concrete,
single-file coordinator, and the queue is owned by the event loop next to the
recent-files state. No playlists, no media library.
Highlights
- Added a lightweight play queue from multi-file and folder drops
- Added manual previous/next navigation with
PageUp/PageDown - Added auto-advance to the next queued file at the natural end of playback
- Preserved single-file behavior: a normal open never scans the parent folder
- Kept
PlaybackSessiona single-file coordinator; the queue lives in the event loop
New Features
Play queue
- Drop multiple files onto the window to build a multi-item queue
- Drop a folder to queue its supported media files (non-recursive)
- Entries are filtered to supported media (subtitles excluded), de-duplicated
case-insensitively, and ordered naturally (Episode 2beforeEpisode 10) - Opening a single file (CLI, file dialog, drag/drop, or Recent Files) creates a
single-file queue and does not scan the parent folder
Manual navigation
PageDownopens the next queued file;PageUpopens the previous one- Navigation only acts when the queue has more than one item and never wraps at
the first or last entry - A brief status overlay reports the result (
Next: …,Previous: …,
No next file,No previous file,Open failed)
Auto-advance
- When a file in a multi-item queue reaches its natural end, FastPlay opens the
next queued file automatically - Auto-advance does not run for a single-file queue and does not wrap at the end
of the queue (it showsEnd of queueinstead) - Looping, auto-replay, and in/out range stops are respected: auto-advance does
not fire when you are intentionally looping or replaying a range
Architecture Notes
PlayQueueis a pure, unit-tested data structure owned by the event loop;
PlaybackSessionholds no queue state- All opens — manual and automatic — flow through the existing
open_media -> session.open_atpath, so resume, recent files, subtitles,
metrics, and in/out reset behave identically to a manual open PlaybackSessionexposes a single new, minimal, edge-triggered
take_ended_signal()that fires once per natural end-of-file transition and is
consumed exactly once, so auto-advance cannot fire every frame while ended- Candidate/commit discipline: next/previous lookups never move the cursor; the
cursor commits only after the target file opens, so a failed open leaves the
queue position unchanged
Controls
| Key | Action |
|---|---|
PageDown |
Next file in the play queue |
PageUp |
Previous file in the play queue |
Validation
cargo fmt --checkcargo clippy --all-targets -- -D warningscargo test --all-targets(140 passing)git diff --check- Manual playback smoke:
- drop multiple videos; let the first end; confirm the second opens
- let the last queued video end; confirm it does not wrap
- use
PageUp/PageDown; confirm manual navigation works - open one normal file; let it end; confirm no parent-folder auto-advance
- confirm recent files and resume still work
- confirm loop / auto-replay / in-out range behavior is unaffected
Upgrade Notes
- Existing installs upgrade in place through the MSI major-upgrade path
- The play queue is in-memory and session-only; nothing new is written to disk
- No account, network, or cloud storage is required
Deferred (intentionally out of scope)
- Persistent playlist files, shuffle, repeat modes, recursive folder scanning
- A multi-select open dialog (multi-item queues are built via drag-and-drop)
- An in-window queue list / overlay, preloading, and gapless playback
- Media-library behavior
FastPlay v0.3.0 - Recent Files, Resume Playback, and Seek Reliability
FastPlay 0.3.0 adds the first daily-use product basics: recent files and resume playback position. It also fixes important seek/scrub reliability issues and completes a major internal cleanup of the playback coordinator.
Highlights
- Added Recent Files overlay with
Ctrl+Shift+O - Added resume playback position across CLI, dialog, drag/drop, and recent-file opens
- Fixed A/V desync after backward seeks
- Fixed resumed-file scrubbing mapping that could make playback stop near EOF
- Added a local benchmark harness for open, seek, pause/resume, drops, underruns, and hardware fallback metrics
- Refactored the large playback coordinator into focused tested helpers
New Features
Recent Files
- Press
Ctrl+Shift+Oto open the Recent Files overlay - Use
Up/Downto select a file - Press
Enterto reopen - Press
Deleteto remove an entry - Recent files are capped at 20 and deduplicated
Resume Playback Position
FastPlay now remembers the last playback position for opened files and resumes when appropriate.
Resume applies to:
- CLI path open
- File dialog open
- Drag/drop open
- Recent Files overlay open
FastPlay avoids resuming near the end of a video.
Fixes and Improvements
Seeking and Scrubbing
- Fixed backward-seek A/V desync caused by stale pre-seek audio batch timestamps
- Fixed resumed-file seek/scrub mapping where resume-origin offset could push seeks past EOF
- Improved repeated seek/scrub reliability
- Preserved A/V sync after backward seeks
Architecture and Maintainability
- Enforced
cargo fmt --checkin CI - Added technical debt and roadmap documentation
- Extracted focused modules for:
- viewport state
- clip-range state
- overlay management
- audio coordination
- video frame queue
- input dispatch
- decode-thread handle lifecycle
Benchmarking
- Added local benchmark harness under
bench/ - Captures p50/p95 latency for open, seek, pause/resume
- Captures frame drops, audio underruns, and hardware fallback counts
- Benchmark output supports JSON and CSV
Validation
cargo fmt --checkcargo clippy --all-targets -- -D warningscargo test --all-targetsgit diff --check- Manual playback smoke:
- open
- play/pause/resume
- seek forward/backward
- repeated scrub
- recent-file reopen
- resume playback
- close while playing/paused
- A/V sync after backward seek
Upgrade Notes
- Existing installs should upgrade in place through the MSI major-upgrade path
- Recent/resume data is stored under
%APPDATA%\FastPlay - No account, network, or cloud storage is required
SHA-256
EB329E2C4B761E8E42D6B6A6AD6C2EBDAF88196BB8D81A6B4E943E65F745FF24 fastplay-0.3.0-x86_64.msi
FastPlay v0.2.2
FastPlay 0.2.2 is a clip-range playback stability hotfix over 0.2.1. It fixes freezes, crashes, lag, and inconsistent Space behavior when timeline seeks land before an I marker or after an O marker.
Highlights
- fixed a D3D11 access violation caused by ending an I/O clip while the presenter retained a decoded frame
- fixed lag and jitter when clicking the timeline to the right of the O marker during playback
- fixed Space being ignored or producing inconsistent playback while a timeline seek was still in flight
- resuming from outside the active I/O range now restarts at the I marker, or at the beginning when no I marker is set
Fixes and Improvements
Clip ranges and timeline seeking
- clip boundaries now stop playback logically while keeping the persistent decoder parked for safe, low-latency replay
- audio and video submission remain stopped in the ended state instead of consuming work beyond the clip boundary
- outside-range timeline clicks land paused for inspection rather than issuing a second resume seek that immediately collides with out-point enforcement
- pause/resume intent is preserved while a seek is in
Seekingor seek-relatedPriming - replay and resume reuse the persistent decoder through an in-place seek instead of asynchronously tearing down D3D11VA
Validation
- repeated timeline clicks before I and after O
- Space pause/resume during and after timeline seeks
- replay from the I marker after landing outside the active range
- 27 automated tests pass
- release build and WiX MSI packaging pass
Upgrade Notes
- existing MSI installs upgrade in place through the WiX
MajorUpgradepath - no settings or configuration changes are required