Files
ux/docs/architecture.md
agra 7243ef7de4 video: vendor FFmpeg software AVC renderer
Adds an LGPL FFmpeg-backed video renderer that slots ahead of Media3's
MediaCodecVideoRenderer via EXTENSION_RENDERER_MODE_PREFER. Resolves
playback failures on Huawei EMUI 11 (Mate 20, Kirin 980): the Codec2
HiSilicon AVC decoder initialises cleanly on iOS High@3.1 streams with
deep DPB + full-range yuvj420p, then errors on the first sample inside
MediaCodecVideoRenderer (init-failure fallback can't catch this).
Google's C2 SW AVC decoder hits its 8-frame output-delay cap on the
same shape and stalls on dequeueOutputBuffer.

Media3's own decoder-ffmpeg ships only an audio renderer;
ExperimentalFfmpegVideoRenderer has been a stub since 2020 (returns
FORMAT_UNSUPPORTED_TYPE, createDecoder returns null). NextLib is
GPL-3.0. So we vendor our own Apache-licensed JNI on top of LGPL
FFmpeg, dynamically linked at runtime.

Build flow:
  - android/ffmpeg/ holds the JNI source + CMakeLists + orchestrator
    script + LGPL notice. No native binaries in git.
  - :ux:buildFfmpegJni Gradle task (wired to preBuild) clones
    Media3 1.9.2 + FFmpeg release/6.0 into build/ffmpeg-work/ on
    first run, builds h264-only static libs per ABI, links
    libffmpegJNI.so per ABI into build/jniLibs/<abi>/. AGP picks
    them up via sourceSets.main.jniLibs.srcDirs +=. Gradle
    UP-TO-DATE skips the task when ffmpeg_jni.cc / CMakeLists /
    build_ffmpeg.sh are unchanged.

Renderer:
  - FfmpegVideoDecoder (SimpleDecoder) sends each packet with its
    inputBuffer.timeUs as pkt->pts; the JNI overwrites
    outputBuffer.timeUs with f->pts on receive so frames emitted in
    display order carry their true display PTS (input PTS in decode
    order scrambles ExoPlayer's drop logic and halves the render
    rate on B-frame streams).
  - FfmpegOutputSurface does YUV->RGB in one GLES2 pass against an
    EGL window surface sized to display orientation. Y plane uses
    GL_NEAREST (1:1 sized, sampling at exact texel centres
    preserves luma detail); chroma uses GL_LINEAR. Pre-rotated quad
    UVs (0/90/180/270) keep the YUV sampling correct when the
    coded frame needs rotation for display.
  - FfmpegVideoRenderer swaps the output buffer's width/height for
    90/270 streams before super.renderOutputBuffer notifies size,
    matching MediaCodecVideoRenderer's post-rotation reporting.

Decoder fallback:
  - Renderers.kt selects FfmpegVideoRenderer first when
    libffmpegJNI.so is loaded; falls through to the platform path
    for formats FFmpeg doesn't handle or ABIs without the .so.
  - MediaCodec selector deprioritises every HiSilicon decoder
    (OMX.hisi.* and c2.hisi.*) so the platform path picks
    c2.android.avc.decoder ahead of the C2 Hisi variant when FFmpeg
    isn't available. Required because the C2 Hisi failure is
    post-init, which Media3's setEnableDecoderFallback(true) can't
    intercept.

Compositor:
  - VideoCompositor.setInputSurfaceSize lets the renderer resize the
    codec-input SurfaceTexture before eglCreateWindowSurface so the
    EGL surface inherits matching buffer dimensions on creation
    (MediaCodec sizes natively; EGL doesn't).
  - VideoPlayerInstance wires Renderers.build with a sizer callback
    that calls into compositor.setInputSurfaceSize from the FFmpeg
    renderer thread.

Adds docs/architecture.md with the layered video pipeline diagram,
file map, renderer-selection rationale, build flow, and LGPL
boundary notes.
2026-05-28 19:24:17 +03:00

8.8 KiB

ux — architecture

This doc maps the load-bearing modules in the ux Flutter package: where each lives, what it owns, and which native pieces back it on each platform. Banlu is the primary consumer; everything here is meant to be reusable across other Flutter apps that depend on ux.

Conventions:

  • Source paths are anchored at the package root (e.g. lib/src/...).
  • Native paths use android/... and darwin/... (shared iOS+macOS sources) or ios/ / macos/ (platform-specific).
  • "Banned" means: don't reintroduce. Lints and tests exist to catch regressions where applicable.

Module map

Module Dart Android Apple
XKeyboard — frame-accurate keyboard tracking + interactive dismiss lib/src/keyboard/ android/src/main/jni/keyboard_bridge.c darwin/Keyboard/
XSensors — gyro / accelerometer over JNI/FFI lib/src/sensors/ android/src/main/jni/sensor_bridge.c darwin/Sensors/
XCamera — CameraX (Android) / AVCaptureSession (Apple) lib/src/camera/ android/src/main/kotlin/io/swipelab/ux/camera/ darwin/Camera/
XScanner — ZXing QR scanner (Android) / VNDetect (Apple) lib/src/scanner/ android/src/main/kotlin/io/swipelab/ux/scanner/ darwin/Scanner/
XVideoPlayer — ExoPlayer (Android) / AVPlayer (Apple) lib/src/video/ android/src/main/kotlin/io/swipelab/ux/video/ darwin/Video/
XFile, XNotifications, XWindow, navi — see source lib/src/ mixed mixed

Video pipeline (Android)

The Android video stack is the most involved piece in ux because it spans four cooperating layers (Dart controller → Kotlin player instance → Media3 ExoPlayer → GL compositor → Flutter texture). It also carries the most production-load-bearing fix: a vendored FFmpeg video decoder that fronts the platform MediaCodec path for files the OEM hardware decoder can't handle.

Layers

Dart  XVideoPlayerController (lib/src/video/x_video_player.dart)
       │   MethodChannel `ux/video`
       ▼
Kotlin VideoPlayerInstance (one per Dart controller)
       │   owns:
       │   ├── ExoPlayer  (Media3 1.9.2; renderers wired via Renderers.kt)
       │   ├── VideoCompositor (codec output → Flutter texture; GL blit)
       │   └── Flutter SurfaceTextureEntry (the texture id Dart Texture() samples)
       ▼
Native Media3 video renderer
       │   tier 1: FfmpegVideoRenderer  (libavcodec, SW decode, GL YUV→RGB)
       │   tier 2: MediaCodecVideoRenderer (HW path; HiSilicon C2 decoders
       │           deprioritised so EMUI 10/11 falls through to c2.android.avc.decoder)
       ▼
GL    SurfaceTexture (compositor input)
       │   compositor reads + applies SurfaceTexture transform
       │   blits to Flutter SurfaceTexture (output)
       ▼
Flutter Texture widget samples the SurfaceTexture, scales to widget size

Files

What Where
Dart controller lib/src/video/x_video_player.dart
MethodChannel binding lib/src/video/x_video_player_channel.dart
Texture widget lib/src/video/x_video_player_view.dart
Per-controller native player android/src/main/kotlin/io/swipelab/ux/video/VideoPlayerInstance.kt
ExoPlayer renderer factory android/src/main/kotlin/io/swipelab/ux/video/Renderers.kt
GL blit between codec and Flutter texture android/src/main/kotlin/io/swipelab/ux/video/VideoCompositor.kt
FFmpeg renderer (Java) android/src/main/java/io/swipelab/ux/video/ffmpeg/
FFmpeg JNI source android/ffmpeg/ffmpeg_jni.cc
FFmpeg build orchestrator android/ffmpeg/build_ffmpeg.sh

Renderer selection (Renderers.kt)

Renderers.build(context, surfaceSizer) returns a DefaultRenderersFactory subclass that builds the video renderer list with FFmpeg first and MediaCodec second:

  1. FfmpegVideoRenderer — picked when FfmpegLibrary.isAvailable() returns true (i.e. libffmpegJNI.so is loaded for the device's ABI) and the format is H.264. Software decode via libavcodec, no DPB cap, handles iOS yuvj420p (full-range) and deep B-frame reorder structures that defeat platform decoders on older Huawei silicon. YUV→RGB happens in a one-pass GLES2 shader inside FfmpegOutputSurface. Y plane uses GL_NEAREST filtering (1:1 sized to the EGL surface, sampling at exact texel centres preserves luma detail); chroma uses GL_LINEAR for clean 4:2:0 upsampling.
  2. MediaCodecVideoRenderer (Media3 default) — picked for any other format, or if FFmpeg is unavailable. Two non-default tweaks:
    • setEnableDecoderFallback(true) so init-time failures bounce to the next decoder.
    • Custom MediaCodecSelector deprioritises every HiSilicon decoder (OMX.hisi.* and c2.hisi.*). The EMUI 10 OMX variant fails codec start (caught by fallback); the EMUI 11 C2 variant initialises cleanly but errors on the first sample, which fallback can't catch — name-based deprioritisation is the only way past it when FFmpeg is unavailable for that ABI.

The two-tier design is intentional: FFmpeg adds a slight battery cost vs. hardware decoding, but it's the only path that reliably plays the full range of producer-side content (iOS recordings in particular). Hardware stays as the fallback for the formats and devices where it works.

Why we vendor an FFmpeg video renderer

Media3's published decoder-ffmpeg is audio-only. ExperimentalFfmpegVideoRenderer in the same library has been a stub since 2020 (createDecoder returns null, supportsFormat returns FORMAT_UNSUPPORTED_TYPE). The only well-maintained community alternative (NextLib) is GPL-3.0 — would impose copyleft on every consumer. So we ship our own Apache-licensed JNI on top of the LGPL FFmpeg, vendored under android/ffmpeg/.

How the FFmpeg .so gets built

There are no native binaries in the repo. The :ux:buildFfmpegJni Gradle task (wired to preBuild in android/build.gradle) handles everything:

  1. Clones Media3 (1.9.2) and FFmpeg (release/6.0) into the consumer's build/ffmpeg-work/ on first run.
  2. Drops the vendored ffmpeg_jni.cc + CMakeLists.txt over the upstream copies (so the produced JNI exposes the video entry points FfmpegVideoDecoder.java calls).
  3. Cross-builds FFmpeg static libs (H.264 only) for all 4 Android ABIs.
  4. Links libffmpegJNI.so per ABI and writes them into <ux-build>/jniLibs/<abi>/, which AGP picks up via sourceSets.main.jniLibs.srcDirs +=.

Up-to-date checking on ffmpeg_jni.cc + CMakeLists.txt + build_ffmpeg.sh means subsequent builds skip the task entirely. The slow part (FFmpeg static-lib build, ~30 min on an x86 host, ~2 min on Apple Silicon) only happens on the first build for a given checkout, or after a gradle clean.

License: FFmpeg is LGPL v2.1, linked dynamically at runtime via System.loadLibrary("ffmpegJNI") — the LGPL boundary stays intact. The build config omits --enable-gpl and --enable-nonfree. See android/ffmpeg/LICENSE-FFMPEG.txt and android/ffmpeg/README.md.

Looping and end-of-stream

XVideoPlayerController.setLooping(true) maps to ExoPlayer's REPEAT_MODE_ONE (the entire seek-to-0-and-resume cycle is handled in-engine). Non-looping playback emits a completed event when STATE_ENDED is reached without repeat — see VideoPlayerInstance.kt. Consumers that need tap-play-at-end-restarts behavior (Telegram parity) check position against duration on the Dart side and call seekTo(Duration.zero) before play() when within 2 seconds of the end.


Video pipeline (Apple)

AVPlayer + AVPlayerItem + AVPlayerItemVideoOutput. No FFmpeg fallback needed — AVFoundation handles iOS-produced H.264 (and HEVC) directly without the DPB-cap / full-range quirks the Android platform decoders trip over. See darwin/Video/VideoPlayerInstance.swift.