Supplementary Feature Parity Review: Deep Dive¶

This document highlights critical functionality present in native FFmpeg that has been entirely omitted from the ffmpeg-wasi JSON spec and C implementation.

1. Missing Time Manipulations (Trimming and Seeking)¶

Native FFmpeg: Supports fast seeking (-ss before the input), output duration limits (-t), and end-time limits (-to). ffmpeg-wasi: The JSON jobSpec schema has no fields for start times or durations. The main loop in process.c processes the file unconditionally from byte 0 until EOF. - Impact: You cannot extract a 10-second clip from a 2-hour movie without decoding the entire 2-hour movie up to that point.

2. Inability to Read Raw Formats¶

Native FFmpeg: Allows forcing format and stream configurations using -f, -pixel_format, -video_size, and -framerate before an input. ffmpeg-wasi: The implementation relies entirely on avformat_open_input(..., NULL, NULL) and avformat_find_stream_info. - Impact: If an input file lacks headers (e.g., raw YUV video, raw PCM audio), avformat will fail to guess the format, and the process will abort. The JSON spec needs input_options to pass custom dictionaries to the demuxer.

3. Missing Metadata and Chapter Passthrough¶

Native FFmpeg: Automatically attempts to copy global metadata (title, artist, creation time) and chapters from the input to the output. It also supports -map_metadata. ffmpeg-wasi: Only allocates output streams and copies codec parameters.

avcodec_parameters_from_context(go->ost->codecpar, go->enc);

- Impact: All ID3 tags, MP4 atoms, rotation flags, and chapter markers present in the original inputs are completely stripped and lost in the resulting output files.

4. Forced Output Formats¶

Native FFmpeg: Allows overriding the guessed output format via -f format. ffmpeg-wasi: Relies entirely on avformat_alloc_output_context2(&o->ofmt, NULL, NULL, o->path). - Impact: It guesses the muxer entirely based on the file extension of o->path. If a user wants to output a raw H.264 stream but names the file output.vid, the implementation will fail to guess the format and abort.

5. PTS (Presentation Time Stamp) and A/V Sync Handling¶

Native FFmpeg: Contains complex logic for Audio/Video synchronization (-vsync, -async, -fps_mode). It knows how to drop or duplicate frames to maintain constant frame rates (CFR). ffmpeg-wasi: Blindly forwards the pts provided by the buffersink directly into the encoder. While libavfilter attempts to maintain sync, without explicit options to dictate how desyncs are handled (e.g., inserting silent audio or duplicating video frames), output files generated from corrupted or variable-framerate inputs may suffer from heavy audio drift.