Add missing reset calls after continue in layoutRunsAndFloatsInRange().

Source/core/rendering/RenderBlockLineLayout.cpp

Index: Source/core/rendering/RenderBlockLineLayout.cpp
diff --git a/Source/core/rendering/RenderBlockLineLayout.cpp b/Source/core/rendering/RenderBlockLineLayout.cpp
index 2423b4c5b89694fce7f9e32e6c07d7902913f199..7ffcbd4d388d19a79f507cbeef0dca8a9ab1a1ef 100644
--- a/Source/core/rendering/RenderBlockLineLayout.cpp
+++ b/Source/core/rendering/RenderBlockLineLayout.cpp
@@ -880,6 +880,8 @@ void RenderBlockFlow::layoutRunsAndFloatsInRange(LineLayoutState& layoutState,
                             // We have to delete this line, remove all floats that got added, and let line layout re-run.
                             lineBox->deleteLine();
                             endOfLine = restartLayoutRunsAndFloatsInRange(oldLogicalHeight, oldLogicalHeight + adjustment, lastFloatFromPreviousLine, resolver, previousEndofLine);
+                            lineMidpointState.reset();
+                            resolver.setPosition(endOfLine, numberOfIsolateAncestors(endOfLine));

[rhogan, 2014/11/03 18:58:12]

This looks right to me. The lineMidpointState.reset() looks superfluous but if
you want to remove it you need to look closely at the logic that breaks before
it gets called at the top of the function.

In the bug you note that there's another superfluous-looking call to
lineMidpointState.reset() at the end of the function - that is definitely worth
addressing in a follow-up patch. If I were you I would keep the change above as
it is and address the superfluous calls in a follow-up.

You also need to add a test here. Once you have that you should ping an OWNER
like leviw or jchaffraix.

                             continue;
                         }