Media, segments, and the index

Part of SABR in the extractor. A SABR response carries media as separate UMP parts. The extractor correlates them by a one-byte header id, validates their length, decompresses them when needed, and exposes completed SabrMediaSegment objects.

Media parts

• MEDIA_HEADER (20) carries a SabrMediaHeader and opens a segment.
• MEDIA (21) starts with the header id byte; the remaining bytes are appended to that segment.
• MEDIA_END (22) starts with the header id byte and closes the segment.

Audio and video can be interleaved freely because each open header has its own accumulator.

SabrMediaHeader

The current decoder reads these fields:

#	Field
1	header id
2	video id
3–5	itag, last-modified value, xtags
6	start byte range
7	compression (0 none, 1 gzip, 2 brotli)
8	initialization-segment flag
9	sequence number
10	bitrate in bits per second
11–12	start and duration in milliseconds
13	nested fallback FormatId
14	expected on-wire content length
15	nested time range (ticks and timescale)
16	sequence last-modified value

When millisecond values are absent but the nested time range has a positive timescale, the decoder derives them as ticks * 1000 / timescale.

Assembly and decompression

SabrMediaSegmentCollector.collect(response) replays a buffered response. The streaming path uses Incremental, which receives parts as they arrive and emits a segment at onMediaEnd. Media for an unknown or already closed header is discarded; a header without MEDIA_END is not emitted.

Before decompression, the collector checks contentLength against the number of bytes received on the wire. It then applies gzip or brotli according to the header. Unsupported algorithms, overflows, truncated payloads and decompression failures are reported as protocol or recoverable errors.

SabrMediaSegment can hold decompressed bytes in memory or use a spool file. For large uncompressed segments the incremental collector can expose a progressive file-backed segment; callers should use openStream() instead of copying it back with getData().

Initialization and segment timelines

FORMAT_INITIALIZATION_METADATA (part 42) describes the initialization and index ranges for a format. The client fetches the initialization range with YoutubeSabrRequestHelper.fetchInitializationData(...), then parses those bytes with YoutubeSabrFormatTimeline.parse(...):

• SabrMp4SegmentIndexParser reads an ISO-BMFF sidx box and converts each subsegment duration to milliseconds;
• SabrWebmSegmentIndexParser reads Matroska/EBML Cues and derives each segment's duration from the next cue or the format duration.

SabrSegmentIndex is a 1-based list of entries containing sequence number, start time, duration and computed end time. YoutubeSabrFormatTimeline maps a sequence to its time range and maps a requested time to the first segment ending after that time.

The current extractor does not contain the former SabrSegmentRequest or YoutubeSabrStreamState classes. Segment selection and buffered-range policy are responsibilities of the caller that builds YoutubeSabrRequest.

Next: The session driver.