Bug 82406 – [chromium] Compositor visibility setting must be updated even if not actively compositing

• Component: WebCore Misc.
• Status: RESOLVED FIXED

[Vangelis Kokkevis]

This is a problem surfaced by the following bug:
http://code.google.com/p/chromium/issues/detail?id=119812

WebViewImpl::setVisibilityState() only updates the visibility state in the compositor if it's in accelerated compositing mode. However, if a page falls out of compositing mode while it's hidden (e.g. a html5 video stops playing while the tab is not visible) then switching back to the tab never notifies the compositor that it has become visible again.

[James Robinson]

Created attachment 134189 [details]
Patch

[James Robinson]

Vangelis, do you know any way to confirm whether this patch fixes that bug? This is a blind fix (didn't even compile check) but seems like the behavior we want.

[Vangelis Kokkevis]

Comment on attachment 134189 [details]
Patch

(unofficial) R+ .  This will do the trick.

[Nat Duca]

Comment on attachment 134189 [details]
Patch

Mmm I like this! We should do this regardless.

Which draws my attention to another thing: if we create a layer tree view in !visible, we get the visibility wrong too, e.g. setIsAcceleratedCompositingActive. Can we fold a fix into there too?

[Nat Duca]

Comment on attachment 134189 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=134189&action=review

> Source/WebKit/chromium/src/WebViewImpl.cpp:3459
> +        if (!visible && isAcceleratedCompositingActive())

This feels like a wart incidentally. I never knew this was here until just now. :(

[Adrienne Walker]

Comment on attachment 134189 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=134189&action=review

> Source/WebKit/chromium/src/WebViewImpl.cpp:3460
> -        if (!visible)
> -            m_nonCompositedContentHost->protectVisibleTileTextures();
>          m_layerTreeView.setVisible(visible);
> +        if (!visible && isAcceleratedCompositingActive())
> +            m_nonCompositedContentHost->protectVisibleTileTextures();

Er, why is the call to protectVisibleTileTextures reordered? Don't we need to protect before setVisible which may drop tiles?

[James Robinson]

(In reply to comment #6)
> (From update of attachment 134189 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=134189&action=review
> 
> > Source/WebKit/chromium/src/WebViewImpl.cpp:3460
> > -        if (!visible)
> > -            m_nonCompositedContentHost->protectVisibleTileTextures();
> >          m_layerTreeView.setVisible(visible);
> > +        if (!visible && isAcceleratedCompositingActive())
> > +            m_nonCompositedContentHost->protectVisibleTileTextures();
> 
> Er, why is the call to protectVisibleTileTextures reordered? Don't we need to protect before setVisible which may drop tiles?

Whoops, I did not mean to change that order. Will fix.

[James Robinson]

Created attachment 134212 [details]
Patch

[Adrienne Walker]

Comment on attachment 134212 [details]
Patch

R=me.

[WebKit Review Bot]

Comment on attachment 134212 [details]
Patch

Clearing flags on attachment: 134212

Committed r112417: <http://trac.webkit.org/changeset/112417>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.