46d152dfc5
start() unconditionally spawned a fresh stremio-runtime every time it was called. Today only the GUI thread calls it, but the crash-restart path can re-fire start() while the previous teardown is still in progress, racing two runtimes for port 11470. Track a `running: Arc<AtomicBool>` on the struct: swap(true) on entry and bail with a log if the previous server thread has not yet flipped it back to false. Reset to false right before sender.notice() so that the GUI's crash handler can legitimately spawn the next instance. Closes #57 |
||
|---|---|---|
| .github/workflows | ||
| bin | ||
| bin-arm64 | ||
| images | ||
| mpv-arm64 | ||
| mpv-x64 | ||
| setup | ||
| src | ||
| .gitignore | ||
| build-arm64.ps1 | ||
| build.ps1 | ||
| build.rs | ||
| Cargo.lock | ||
| Cargo.toml | ||
| generate_descriptor.js | ||
| libmpv-2_arm64.zip | ||
| libmpv-2_x64.zip | ||
| README.md | ||
| rust-toolchain.toml | ||
| server.js | ||
| stremiover.js | ||
| upload.ps1 | ||
Stremio shell: new gen
A Windows-only shell using WebView2 and MPV
Goals:
- Performance
- Reliability
- Easy to ship
In all three, this architecture excels the Qt-based shell: it is about 2-5x more efficient depending on the use case, as it allows MPV to render directly in the window through it's optimal video output rather than using libmpv to integrate with Qt.
This is due to Qt having a complex rendering pipeline involving ANGLE and multiple levels of composing and drawing to textures, which inhibits full HW acceleration.
Meanwhile in this setup MPV uses whichever pipeline it considers to be optimal (like the mpv desktop app), which is normally d3d11, allowing full HW acceleration.
For web rendering, we use the native WebView2, which is Chromium based but shipped as a part of Windows 10: therefore we do not need to ship our own "distribution" of Chromium.
Finally, this should be a lot more reliable as it uses a much simpler and more native overall architecture.