Created attachment 373087 [details] [review]
composition: Output one single segment "per seek"

Instead of pushing a segment per new stack, output one single segment
that represents the whole duration of the composition. We now
retimestamp everything inside that segment using buffer PTS running
time.

The main goal of doing so is to avoid having the encoder drainning on
new segments which, with some encoders, leads to the introduction of
a few "silent" samples, meaning hicups/blips/blops.

Created attachment 373569 [details] [review]
composition: Output one single segment "per seek"

Instead of pushing a segment per new stack, output one single segment
that represents the whole duration of the composition. We now
retimestamp everything inside that segment using buffer PTS running
time.

The main goal of doing so is to avoid having the encoder drainning on
new segments which, with some encoders, leads to the introduction of
a few "silent" samples, meaning hicups/blips/blops.

Review of attachment 373569 [details] [review]:

While I'm not fan of the encoder behaviour with segments, I think this change do make sense. The composition endup making the upper layer look like a single stream, not leaking the upstream complexity into the downstream elements. This change basically makes the stream time match the rendering (the output), instead of jumping around. I think it will make the nlecomposition much less of a special case, and help prevent later regressions.

I have found a big regression in that patch and am working on it :-)

-- GitLab Migration Automatic Message --

This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/gstreamer/gst-editing-services/issues/40.