[Omnibox] Elide omnibox text

ui/gfx/render_text.cc

Index: ui/gfx/render_text.cc
diff --git a/ui/gfx/render_text.cc b/ui/gfx/render_text.cc
index 97b6c8fb82f36751df3c94ebca4864cf392b05b7..776b315af1e93fd142fc6ce94b4fa4bdaf55f7a7 100644
--- a/ui/gfx/render_text.cc
+++ b/ui/gfx/render_text.cc
@@ -563,6 +563,7 @@ void RenderText::SetElideBehavior(ElideBehavior elide_behavior) {
   // TODO(skanuj) : Add a test for triggering layout change.
   if (elide_behavior_ != elide_behavior) {
     elide_behavior_ = elide_behavior;
+    lines_.clear();

[msw, 2017/04/17 22:39:10]

Hmm, I'm a little surprised this isn't already handled [indirectly] via
OnDisplayTextAttributeChanged(). (presumably, when display text properties
change, we can't trust the result of the previous line breaking pass). Can you
check if removing this |lines_.clear();| statement and calling UpdateDisplayText
(perhaps indirectly, via EnsureLayout()) is sufficient? I'm guessing that you
don't need any RenderTextChanges, and the user is just somehow skipping a layout
pass after changing the elide behavior.

Otherwise, it might make sense to do |lines_.clear()| in
OnDisplayTextAttributeChanged. (Theoretically we can avoid clearing the lines if
we are switching between no eliding and fade eliding, but that's not a big
deal.)

[simonhong, 2017/04/17 23:21:43]

UpdateDisplayText() didn't handled it properly.
In RenderTextHarfBuzz::EnsureLayoutRunList(), UpdateDisplayText() is called.
Then, update_display_text_ and update_display_run_list_ are set to false.
After that, we lost the chance of clearing previous |lines_|.
Moved.

     OnDisplayTextAttributeChanged();
   }
 }