The current SourceBuffer::canPlayThroughRange() implementation is based on average buffering rate. While this approach makes sense in a context of continuous playback, where the JS app is always trying to append data, this isn't always the case in some real life apps. For instance, an app may append a lot of data on page load (enough for sustained playback), then decide to wait for whatever reason, and then start playback. In those circumstances, wait would cause the average buffering rate to be artificially low. There are more examples of the kind of problems that a time-based/average buffering rate-based approach may cause.

See: https://github.com/WebPlatformForEmbedded/WPEWebKit/issues/928

One solution would be to use Firefox strategy to this problem[1]: if the ranges to be checked have 3 seconds or more buffered after the current position, we consider that  sustained playback is possible. This solves the issues seen in some real world apps.

If this approach causes any problems on non-WPE/WebKitGTK ports, a possible improvement might be to let the platform implementation decide the desired strategy in some way. For the mentioned platforms it's important to get rid of the problems that the time-based implementation causes.

[1] https://github.com/mozilla/gecko-dev/blob/master/dom/media/mediasource/MediaSourceDecoder.cpp#L320