Mark the final Ogg page as the end of the stream

[django23]

TL;DR: BCR's .oga recordings are missing the marker that says "this is where the file ends". Most players do not care, but stricter ones refuse to play the file at all, and in Firefox none of them work. This adds the marker when a recording finishes. Nothing about the audio itself changes, and existing recordings are unaffected.

Technical version

The OGG/Opus files BCR writes never terminate the logical bitstream: no page in the output has the end-of-stream bit set in its header type flags. The Ogg spec requires the last page of a stream to carry that bit, so every .oga recording BCR has produced is malformed.

The cause is in AOSP's OggWriter. It sets each packet's flag from its own member inside the write loop:

op.e_o_s = mReachedEOS ? 1 : 0;

but mReachedEOS only becomes true at the end of threadFunc(), after the loop has finished. It is false for every packet actually written, and neither stop() nor reset() writes a terminating page afterwards.

OggContainer wraps MediaMuxer and, once it has stopped, sets the bit in the last page's header and recomputes that page's CRC. Audio data is untouched. Like FlacContainer, this is a finalization step guarded on encoder EOF, so aborted recordings behave exactly as they do today.

The practical benefit is that demuxers which seek to the end of the file to find the last granule position will accept the recordings. Firefox is one, and currently refuses to play them with NS_ERROR_DOM_MEDIA_METADATA_ERR.

The page and CRC logic lives in OggPage, free of Android dependencies so it can be unit tested on the JVM. OggPageTest covers the backwards page search, the OggS magic appearing inside page data, the flag and CRC update, and that nothing outside the header type byte and CRC field is touched. The CRC is checked against a bit-by-bit reference implementation that shares no code with the table driven one.

assembleDebug and the unit tests pass.

[chenxiaolong]

Thanks for the PR! That's quite a dumb bug on the AOSP side. I'm able to reproduce this in libstagefright with the latest AOSP android-17.0.0_r1 release.

I will review this after work today.

[chenxiaolong]

Thanks! This looks good to me. I only added 7f7e7e3 to annotate OggContainer with @RequiresApi(Build.VERSION_CODES.Q).