Use GetDisplayText() instead of text() for rendering text.

ui/gfx/render_text_mac.cc

Index: ui/gfx/render_text_mac.cc
diff --git a/ui/gfx/render_text_mac.cc b/ui/gfx/render_text_mac.cc
index 1c046648dc63220fcca566f1ec8c85a106cd3d5e..4cfeb5b13126fedb5739f6e079fe8471a083a608 100644
--- a/ui/gfx/render_text_mac.cc
+++ b/ui/gfx/render_text_mac.cc
@@ -148,7 +148,7 @@ void RenderTextMac::EnsureLayout() {
                          &kCFTypeDictionaryValueCallBacks));
 
   base::ScopedCFTypeRef<CFStringRef> cf_text(
-      base::SysUTF16ToCFStringRef(text()));
+      base::SysUTF16ToCFStringRef(GetDisplayText()));

[tapted, 2015/03/02 22:36:38]

This is one of the things I tried... unfortunately it wasn't enough to fix the
eliding problem.

(note RenderTextMac's override of GetDisplayText() is quite different to the one
in RenderTextHarfBuzz).

E.g. GetDisplayText() changes its return when the value of text_elided_ changes.
That happens in RenderText::UpdateDisplayText(), but UpdateDisplayText() doesn't
trigger a call to RenderTextMac::OnLayoutTextAttributeChanged(), so the
previously calculated runs are not invalidated.

[Jun Mukai, 2015/03/02 23:10:37]

I don't think RenderTextMac is quite different here, if elide_behavior is set,
display_text() and text_elided() should be updated properly in render_text.cc,
and this file seems to invoke UpdateDisplayText() properly.  If it doesn't
return elided text, that's a regression of RenderText.
Also, the new test case should catch the failure by checking the number of
glyphs in |line_|.

The test seems succeeding on the case, do you know what's wrong with my test
case?

   base::ScopedCFTypeRef<CFAttributedStringRef> attr_text(
       CFAttributedStringCreate(NULL, cf_text, attributes));
   base::ScopedCFTypeRef<CFMutableAttributedStringRef> attr_text_mutable(