It QIs to nsITextControlElement all the time. Could we not do that somehow?

That is a quick and simple approach to help with bug 1346723. Other code paths using nsITextControlElement aren't as hot.

See also http://searchfox.org/mozilla-central/search?q=public+nsITextControlElement&path=

And I couldn't prove that we don't need the strong reference, and the WeakFrame there hints that objects may die, so I decided to stick with strong references.

Comment on attachment 8905890 [details] [diff] [review] less_qiing_when_setting_textformcontrol_value.diff Review of attachment 8905890 [details] [diff] [review]: ----------------------------------------------------------------- ::: dom/html/HTMLInputElement.h @@ +342,5 @@ > { > return mHasPatternAttribute; > } > > + virtual already_AddRefed<nsITextControlElement> GetAsTextControlElement() override no virtual for final classes. ::: dom/html/HTMLTextAreaElement.h @@ +167,5 @@ > > NS_DECL_CYCLE_COLLECTION_CLASS_INHERITED(HTMLTextAreaElement, > nsGenericHTMLFormElementWithState) > > + virtual already_AddRefed<nsITextControlElement> GetAsTextControlElement() override here as well. ::: layout/forms/nsTextControlFrame.cpp @@ +1137,1 @@ > NS_ASSERTION(txtCtrl, "Content not a text control element"); MOZ_ASSERT here?

The existing coding style uses virtual and override, and IMO virtual is a great annotation to tell that the method call is slow. And yes, I guess I could change that to MOZ_ASSERT

Pushed by opettay@mozilla.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/560c9ee4d584 nsTextControlFrame::SetValueChanged is too QI heavy, r=baku