Subframes (and resource documents) in content should just use the refresh driver of the root content document (which is the same thing as the root document in e10s-land).

We need this so that the external resource docs won't keep holding on to the main doc refresh driver after the main doc presshell goes away.

Comment on attachment 456538 [details] [diff] [review] Part 2 with a tweak to Thaw/Freeze No, this is wrong.

Comment on attachment 456542 [details] [diff] [review] Part 2 for real This is still failing some tests. Need to sort that out.

And the reason it's wrong is that the refresh driver only flushes the prescontext it's initialized with.

The patches in bug 569520 depend on this bug at the moment. It could be detangled, with some pain...

Comment on attachment 456666 [details] [diff] [review] Part 2. Make it possible for refresh driver to flush more than one presshell + PRPackedBool mReflowScheduled; // If true, we have a reflow scheduled. + // Guaranteed to be false if + // mReflowContinueTimer is non-null. + + PRPackedBool mSuppressInterruptibleReflows; Fix indent

Comment on attachment 456667 [details] [diff] [review] Part 3. Make subframes use the parent's refresh driver + ourItem->GetSameTypeRootTreeItem(getter_AddRefs(root)); + if (root != ourItem) { + mRefreshDriver = parent->GetShell()->GetPresContext()->RefreshDriver(); Couldn't you just use GetSameTypeParent here and check for null? I think we really need an HasSameTypeParent method on nsIDocument for situations like this...

> Fix indent Done. > Couldn't you just use GetSameTypeParent here and check for null? Yes. Done. > I think we really need an HasSameTypeParent method on nsIDocument for > situations like this... Or perhaps IsRootOfType or something... Bug 586178 filed.

Risk: Moderate. I'm pretty sure things should work fine, and I've done a bunch of testing on sites with subframes, and our tests pass, but there's a bunch of rejiggering of how we trigger reflow and restyling in here. Should be high-quality rejiggering, of course! Benefit: lays the groundwork for some performance-related improvements we want to make. Should somewhat improve performance on pages with subframes, and should definitely reduce the number of wakeups on such pages (which is good for mobile).

Comment on attachment 464706 [details] [diff] [review] Roll-up patch for approval. DO NOT CHECK IN Approved with a short leash.

Part 1 got pushed as http://hg.mozilla.org/mozilla-central/rev/d07f35b7a38f The others got pushed as well, but then backed out for now.

Part 2: http://hg.mozilla.org/mozilla-central/rev/7d133a2a7d20 Part 3: http://hg.mozilla.org/mozilla-central/rev/4e326cd7bac6