And this is 100% Youtube’s fault, not Firefox’s fault, they created this issue:

This problem is triggered by bad muxed VP9 bytestream served by Youtube, so it’s not a regression on our side, this issue can also be reproduced on old versions Firefox. Usually when muxing a video bytestream, the video samples’ timestamp should be monotonizally increasing and no overlap between samples. But there are some bad video samples in YT’s bytesteam, they overlapped with the previous sample. Eg. [124416000, 125126000] and [125125000, 131382000]. The next one should start from 12516000 instead of starting from 125125000 causing an overlapping.

That overlapped sample triggers this and our WebM demuxer fails to calculate the next timestamp in that situation. The end time of video sample was set to the same as the sample’s start time, and that causes a gap being detected for the next sample, resulting in resetting append state. When doing so, mNeedRandomAccessPoint would be set to true and that triggers the sample skipping mechanism per the spec.

Therefore, there would be many sample being incorrectly skipped and won’t be added into the buffered range. When entering the buffering state, Firefox would be waiting those sample which has been skipped but Youtube thought that those samples were already appended. That makes the endless buffering happened.

Source: https://bugzilla.mozilla.org/show_bug.cgi?id=1878510#c113 (Alastor Wu [:alwu])

  • SGG
    link
    English
    345 months ago

    Honestly might be part of it, by going out of spec on the timestamps it probably let’s them more easily insert different length ads