[GStreamer] FillTimer/buffering logic is not correct

• Component: Platform
• Status: RESOLVED WORKSFORME

[Ihor Ivlev]

https://github.com/WebKit/webkit/blob/master/Source/WebCore/platform/graphics/gstreamer/MediaPlayerPrivateGStreamer.cpp#L1229

m_maxTimeLoaded is evaluated by comparing fillStatus value to *total track duration*, which does not make sense:
fillStatus corresponds to bufferization level which is evaluated [in queue element] by
(buffered level)/(max buffer size), but total track duration is not related to these values.
(buffered level) and (max buffer size) are internal values of the queue element.

[Enrique Ocaña]

Created attachment 297833 [details]
Pipeline dump showing the GstDownloadBuffer element used for on-disk caching and buffer measurement

[Enrique Ocaña]

There are two cases here: on-disk buffering enabled and on-disk buffering disabled. 

I've done several tests in WebKit with some code to force on-disk buffering. A ${HOME}/.cache/WebKitWebProcess-XXXXXX file is created by the GstDownloadBuffer element while the video is shown, and automatically deleted when the player is destroyed. See the pipeline dump in PAUSED state I'm attaching.

I can confirm that in such a scenario, the "buffer size" is actually the whole media size, so it makes sense to correlate "100% buffered" with "all the media has been downloaded". See some logs obtained for a video tag configured with https://download.blender.org/durian/trailer/sintel_trailer-1080p.mp4 (14621544 bytes). The {start,stop} values are in percentage with fixed comma (eg: 123456 is 12.3456%), and the b{start,stop,total} values are in bytes:

### void WebCore::MediaPlayerPrivateGStreamer::fillTimerFired(): Percent buffering query failed
...
### void WebCore::MediaPlayerPrivateGStreamer::fillTimerFired(): Percent buffering query failed
### void WebCore::MediaPlayerPrivateGStreamer::fillTimerFired(): start: 0, stop: 1120, fillStatus: 0.112000
### void WebCore::MediaPlayerPrivateGStreamer::fillTimerFired(): bstart: 0, bstop: 16384, btotal: 133988
### void WebCore::MediaPlayerPrivateGStreamer::fillTimerFired(): start: 0, stop: 4482, fillStatus: 0.448200
### void WebCore::MediaPlayerPrivateGStreamer::fillTimerFired(): bstart: 0, bstop: 65536, btotal: 135568
...
### void WebCore::MediaPlayerPrivateGStreamer::fillTimerFired(): start: 0, stop: 962542, fillStatus: 96.254200
### void WebCore::MediaPlayerPrivateGStreamer::fillTimerFired(): bstart: 0, bstop: 14073856, btotal: 579
### void WebCore::MediaPlayerPrivateGStreamer::fillTimerFired(): start: 0, stop: 984953, fillStatus: 98.495300
### void WebCore::MediaPlayerPrivateGStreamer::fillTimerFired(): bstart: 0, bstop: 14401536, btotal: 223
### void WebCore::MediaPlayerPrivateGStreamer::fillTimerFired(): start: 0, stop: 1000000, fillStatus: 100.000000
### void WebCore::MediaPlayerPrivateGStreamer::fillTimerFired(): bstart: 0, bstop: 14621544, btotal: 0

For this reason I think that the current code seems right.

If on-disk buffering is disabled, the fillTimer isn't ever started and nothing related to buffering is reported.

Can you clarify your use case and provide some data which contradicts this analysis?

[Ihor Ivlev]

in my case, GstDownloadBuffer was not added to the pipeline, but GstQueue was added instead.

[Ihor Ivlev]

Created attachment 297925 [details]
Pipeline dump with GstQueue

[Ihor Ivlev]

Created attachment 297926 [details]
Log collected when GstQueue was used

[Ihor Ivlev]

I have attached pipeline dump and the log.
It appeared for me as a mainstream issue (I found this using WPE port), but after looking at Enrique's pipeline dump, I am not sure if the issue is reproducible with the mainstream webkit.

[Philippe Normand]

Let's close this then. If you find a test showcasing the issue, please re-open!