video_frames 1.1.0

video_frames: ^1.1.0 copied to clipboard

• VideoReader.seek(time) moves nextFrame()'s position: the next frame is the first at or after time. Decoding restarts at the previous sync frame and skips the frames before time without converting them, so playback from the middle of a video no longer decodes everything before it. frameAt is unaffected.

First release on pub.flutter-io.cn.

• iOS/macOS: a writer whose encoder stops accepting frames now fails with encoderStalled after 5 seconds (or 60 seconds to finish the file) instead of hanging forever, like Android.
• The example app now picks a video, applies a filter to every frame, exports it with the original sound and plays it. CI runs on GitHub Actions. The README has install steps, a before/after image, a comparison with other options, and error-code and memory tables.
• macOS support (12+). iOS and macOS share one Swift implementation (sharedDarwinSource), which moved from ios/ to darwin/.
• A writer whose finish() fails now frees the encoder and deletes the broken file (Android leaked the hardware encoder; both platforms left the file). An Android writer that fails while being created no longer leaks it either.
• Android: a stuck encoder now fails with "video encoder stalled" after 5 seconds instead of hanging addFrame or finish forever.
• iOS: a frame whose pixel memory can't be read or allocated now returns an error to Dart instead of crashing the app.
• Readers and writers that are never closed are freed when garbage-collected (a forgotten writer is cancelled). iOS now frees everything when the Flutter engine shuts down, as Android already did.
• Faster frame conversion. Android now decodes and encodes through OpenGL ES (the GPU converts, rotates and scales), about 3x faster at 1080p on an emulator; iOS uses Accelerate (vImage) for full-size frames.
• Android: the first frameAt on a reader no longer sometimes fails with "video decoder stalled".
• Written videos are tagged BT.709 (both platforms). Players no longer guess the color space, which shifted greens and skin tones in HD output.
• VideoReader.open(..., filter:) chooses how maxDimension scales frames: ScaleFilter.nearest (default), bilinear or area. frameAt now scales exactly like nextFrame on iOS. Nearest sampling picks the pixel at each output pixel's center on both platforms (iOS used to pick its top-left).
• Breaking: the host API is split into VideoReaderHostApi and VideoWriterHostApi, each on its own native queue, so decoding can overlap encoding (see the pipelined loop in the README). The test seam VideoFrames.api is replaced by VideoFrames.readerApi and VideoFrames.writerApi.
• VideoWriter.create(..., codec: VideoCodec.hevc) writes HEVC; it fails with a clear error on devices without an HEVC encoder.
• HDR input: iOS frameAt now decodes with AVAssetReader like nextFrame, so HDR frames get the same tone mapping (the image generator path was darker and differently colored). Android 13+ asks decoders to tone-map HDR to SDR, plays Dolby Vision 8/9 (e.g. iPhone videos) through the HEVC/AVC base layer when there is no Dolby Vision decoder, and picks a decoder that supports the track's profile (e.g. 10-bit) instead of the default one.
• Less copying per frame: Android reads GPU pixels straight into one reused buffer per decoder instead of allocating a new array (plus a copy) per frame; iOS builds frames as Data, dropping a full-frame copy in the plugin. The README has a per-resolution memory table.
• Android: a decoder that fails mid-stream now reports which codec and video it was ("c2.x.hevc.decoder (video/hevc 640x360) can't decode this video") instead of only "Invalid to call at Released state".
• Android: the HDR-to-SDR tone-mapping request is only sent for HDR video. On SDR video it made Qualcomm decoders fail ("not a supported pixel format"), so videos the plugin itself wrote couldn't be read on those phones.
• Android: encoders and decoders are tried in the device's preference order until one starts, ending with Android's software codecs. Vendor codecs can claim a format and still fail to start (Qualcomm's H.264 encoder at 64x64 and 256x64); those sizes now fall back instead of failing.
• Breaking: the generated host API and message types (VideoReaderHostApi, VideoWriterHostApi, VideoFrameMessage, VideoInfoMessage) moved from package:video_frames/video_frames.dart to package:video_frames/testing.dart, and VideoFrames.readerApi / writerApi are @visibleForTesting. Every public member is documented.
• Breaking: failures throw VideoFramesException with a VideoFramesErrorCode (fileNotFound, unsupportedFormat, codecUnsupported, encoderStalled, readFailed, writeFailed, unknown) instead of a raw PlatformException.

• Swift Package Manager support. The iOS sources moved to ios/video_frames/Sources/video_frames/, and CocoaPods still works.
• iOS 15 minimum, matching Flutter 3.47.
• Android build moves to AGP 9.1, Kotlin 2.4 and the kotlin { compilerOptions } DSL.

• VideoInfo.audioCompatible reports whether a video's audio can be copied into an MP4.
• VideoWriter.includesAudio: audio that can't be copied is now dropped, and the video is still written, instead of failing.

• VideoReader: sequential and random-access RGBA frames, with rotation applied and optional downscaling.
• VideoWriter: H.264 MP4 encoding from RGBA, with optional audio passthrough.