Remove extra calls of didMeaningfulLayout() during the same navigation

third_party/WebKit/Source/web/WebViewImpl.cpp

Index: third_party/WebKit/Source/web/WebViewImpl.cpp
diff --git a/third_party/WebKit/Source/web/WebViewImpl.cpp b/third_party/WebKit/Source/web/WebViewImpl.cpp
index 7be5f1006676c7800988794aeae343f3fe494fa6..dbc95cea10b156ae6382562a2ec2922b3bffd873 100644
--- a/third_party/WebKit/Source/web/WebViewImpl.cpp
+++ b/third_party/WebKit/Source/web/WebViewImpl.cpp
@@ -3990,14 +3990,17 @@ void WebViewImpl::setSelectionColors(unsigned activeBackgroundColor,
 #endif
 }
 
-void WebViewImpl::didCommitLoad(bool isNewNavigation, bool isNavigationWithinPage)
+void WebViewImpl::didCommitLoad(HistoryCommitType commitType, bool isNavigationWithinPage)
 {
     if (!isNavigationWithinPage) {
-        m_shouldDispatchFirstVisuallyNonEmptyLayout = true;
-        m_shouldDispatchFirstLayoutAfterFinishedParsing = true;
-        m_shouldDispatchFirstLayoutAfterFinishedLoading = true;
+        if (commitType == StandardCommit || commitType == BackForwardCommit) {
+            m_shouldDispatchFirstVisuallyNonEmptyLayout = true;
+            m_shouldDispatchFirstLayoutAfterFinishedParsing = true;
+            m_shouldDispatchFirstLayoutAfterFinishedLoading = true;
+        }
 
-        if (isNewNavigation) {
+        if (commitType == StandardCommit) {
+            // A new session history item should be created for this load.
             pageScaleConstraintsSet().setNeedsReset(true);
             m_pageImportanceSignals.onCommitLoad();
         }
@@ -4298,6 +4301,12 @@ void WebViewImpl::setRootGraphicsLayer(GraphicsLayer* layer)
         // attempt to paint too early in the next page load.
         m_layerTreeView->setDeferCommits(true);
         m_layerTreeView->clearRootLayer();
+
+        // didMeaningfulLayout() should be dispatched again.
+        m_shouldDispatchFirstVisuallyNonEmptyLayout = true;

[dglazkov, 2016/01/15 19:41:48]

Interesting! This is a partial revert of
https://codereview.chromium.org/1380363005/, which indeed caused this problem.
We shouldn't need it though if we ignore didCommitLoad from child frames, right?

[wychen, 2016/01/15 22:31:02]

This is still needed, for cases like refreshing, or redirection.

[dglazkov, 2016/01/15 22:36:43]

Both refresh and redirect will result in a new didCommitLoad call: they both
signal destroying and re-creating a new main frame doc, which will trigger the
load and the didCommitLoad. Are there any others we could be forgetting?

[wychen, 2016/01/21 01:47:52]

Ah. I filtered out HistoryInertCommit in this patch set (#1), so I needed these.
They are removed now.

+        m_shouldDispatchFirstLayoutAfterFinishedParsing = true;
+        m_shouldDispatchFirstLayoutAfterFinishedLoading = true;
+
         visualViewport.clearLayersForTreeView(m_layerTreeView);
     }
 }