Invalidate first line display item clients when first line style changes

third_party/WebKit/Source/core/dom/Element.cpp

Index: third_party/WebKit/Source/core/dom/Element.cpp
diff --git a/third_party/WebKit/Source/core/dom/Element.cpp b/third_party/WebKit/Source/core/dom/Element.cpp
index e785c06dfc53d2423487e8af18cf389fe9095aa8..bfe6476ed60a517b55de336e312b86abf4a445ac 100644
--- a/third_party/WebKit/Source/core/dom/Element.cpp
+++ b/third_party/WebKit/Source/core/dom/Element.cpp
@@ -1791,7 +1791,8 @@ StyleRecalcChange Element::recalcOwnStyle(StyleRecalcChange change)
         updateCallbackSelectors(oldStyle.get(), newStyle.get());
 
     if (LayoutObject* layoutObject = this->layoutObject()) {
-        if (localChange != NoChange || pseudoStyleCacheIsInvalid(oldStyle.get(), newStyle.get()) || svgFilterNeedsLayerUpdate()) {
+        bool pseudoStyleCacheIsInvalid = this->pseudoStyleCacheIsInvalid(oldStyle.get(), newStyle.get());
+        if (localChange != NoChange || pseudoStyleCacheIsInvalid || svgFilterNeedsLayerUpdate()) {

[rune, 2015/12/01 11:09:37]

I'm a bit confused about the role of diffPseudoStyles(), which affects
localChange here, and pseudoStyleCacheIsInvalid(). They seem to do similar
things?

I guess this is the reason for the increased invalidation numbers in the
scrollbar tests?

I suspect this change also fixes crbug.com/563943 which I just reported.

Another thing - the early returns from the for-loop in
pseudoStyleCacheIsInvalid(). If we have both first-line and a different pseudo
element, and we encounter the other pseudo element first in the for-loop, it
seems firstLineStyleDidChange() will not be called because of the early return.
We can handle that in another CL.

[Xianzhu, 2015/12/01 20:06:25]

I see diffPseudoStyles() compares cached pseudo styles, but
pseudoStyleCacheIsInvalid() compares old cached pseudo style and new unchached
pseudo style. I'm not sure when a new style will have cached pseudo style, but
the two functions seem to do different things.
The numbers in the test is the numbers of calls to
StyleResolver::styleForElement(), pseudoStyleForElement() and styleForPage().
There are one styleForElement() call for the element's own style and one
pseudoStyleForElement call for each pseudo style.

diffPseudoStyles() doesn't increase the number because it accesses cached pseudo
style only.
Thanks for reporting the bug. Unfortunately this CL doesn't fix the bug. Will
investigate later.
Done in this CL.

[Xianzhu, 2015/12/01 20:19:09]

I reverted the changes about early returns in the latest patch set. It was
incorrect and further change for crbug.com/563943 would need to change that
again.

             layoutObject->setStyle(newStyle.get());
         } else {
             // Although no change occurred, we use the new style so that the cousin style sharing code won't get