You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
StreamGenerator in our tests assumes mvhd and mdhd boxes contain the same timescale. This is true for the content we're templating on right now, but is not generally true. The mdhd box seems to override the mvhd box, so we should look at mdhd instead.
See #1191 (comment) for an example of content with different timescales in these two places.
This is not causing any issues right now, and is limited to a class used in our integration tests. But we should fix this to use the mdhd box to guard against breakage if/when we update or add to the segments used by StreamGenerator today.
The text was updated successfully, but these errors were encountered:
StreamGenerator over our multidrm clip is now generating the wrong timestamps, which causes a buffering state in some percentage (maybe 2-5%?) of the offline integration test runs.
I'm not sure why the manifestation of this bug is so incredibly specific.
It's incredibly specific because it is the combination of this StreamGenerator bug, a very specific test expectation, and a very specific buffer starvation threshold that are causing the test to fail.
Commit 1284dd4, Change-Id: I8cad9bfde3309de7c2b8301b90aa8c40b6e4d247
introduced a change to the buffering logic. Instead of going into a
buffering state when less than 0.5 seconds of content was buffered,
the thredhold changed to one half of the configured rebufferingGoal.
Now the rule is one half of the rebufferingGoal or 0.5 seconds,
whichever is smaller. This accounts for unusually small
rebufferingGoal settings, while keeping a saner default behavior.
This change in behavior was one part of the test flake described
in internal bug report b/128923302. The other component to the test
flake was issue #1206, which still has a separate fix.
Issue #1206
Fixes b/128923302
Change-Id: I5f133959acac6fd4ebe7fed37d2df88892e6399e
StreamGenerator
in our tests assumes mvhd and mdhd boxes contain the same timescale. This is true for the content we're templating on right now, but is not generally true. The mdhd box seems to override the mvhd box, so we should look at mdhd instead.See #1191 (comment) for an example of content with different timescales in these two places.
This is not causing any issues right now, and is limited to a class used in our integration tests. But we should fix this to use the mdhd box to guard against breakage if/when we update or add to the segments used by
StreamGenerator
today.The text was updated successfully, but these errors were encountered: