Bug 78826 – [GTK] [AC] Events can starve accelerated compositing updates

• Status: RESOLVED
• Resolution: FIXED
• Component: WebKitGTK

[Martin Robinson]

WebCore timers can be starved by a deluge of GTK+ events. We want AC updates to run with the same priority as events so that dragging doesn't block rendering.

[Martin Robinson]

Created attachment 127404 [details]
Patch

[Gustavo Noronha (kov)]

Comment on attachment 127404 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=127404&action=review

> Source/WebKit/gtk/WebCoreSupport/AcceleratedCompositingContextGL.cpp:175
>      core(m_webView)->mainFrame()->view()->syncCompositingStateIncludingSubframes();

One thing I've noticed while working with the clutter AC stuff is that this call may some times not go through because of a pending layout, and the sync will in those cases not go through untill we are lucky to have another sync requested with no layout pending. I am unsure tbh whether that is caused by a bug in our implementation, but it looks like something that is expected, since it's a check instead of an ASSERT() inside FrameView::syncCompositingStateForThisFrame. Unrelated to this patch, but something to keep in mind.

> Source/WebKit/gtk/WebCoreSupport/AcceleratedCompositingContextGL.cpp:188
> +        m_syncTimerCallbackId = g_timeout_add_full(GDK_PRIORITY_EVENTS, 1000.0 / 60.0, reinterpret_cast<GSourceFunc>(syncLayersTimeoutCallback), this, 0);

OK, I guess. I was thinking here right now that I tend to consider layer synchronization more a relayout-like thing than a redraw-like thing. Using GTK_PRIORITY_RESIZE might make more sense if we ever figure out the appropriate balance among the existing priorities =(

[Martin Robinson]

Committed r108104: <http://trac.webkit.org/changeset/108104>