When we get a IME_CANCEL_COMPOSITION event on OSX, clear the

When we get a IME_CANCEL_COMPOSITION event on OSX, clear the
ime state.

BUG=9883
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=48017

[tony, 2010-05-21 07:21:48]

This depends on https://bugs.webkit.org/show_bug.cgi?id=39475 .  With the two
patched in, a preventDefault() in a keydown event clears all IME state.

[Hironori Bono, 2010-05-21 09:05:14]

Thank you for your change.
As just for clarification, it seems your WebKit change cleans up the composition
node before sending an IME_CANCEL_COMPOSITION event. So, the browser process
doesn't have to send a WebCompositionCommandDiscard command to a renderer,
right?

Regards,

Hironori Bono

[tony, 2010-05-21 09:43:34]

On 2010/05/21 09:05:14, hbono wrote:
> Thank you for your change.
> As just for clarification, it seems your WebKit change cleans up the
composition
> node before sending an IME_CANCEL_COMPOSITION event. So, the browser process
> doesn't have to send a WebCompositionCommandDiscard command to a renderer,
> right?

Yes, that's right, although it is harmless to do so.  Should I add a bool to
avoid the extra WebCompositionCommandDiscard from being sent?

I originally didn't have the WebKit change and what happened was:
1) Renderer sends a cancel composition event
2) Browser cancels the composition and sends a WebCompositionCommandDiscard back
to the renderer
3) Renderer ignores the WebCompositionCommandDiscard event because
m_suppressNextKeypressEvent is still true.  This makes it look like there's a
composition even though it's been canceled.

[Hironori Bono, 2010-05-21 10:01:45]

LGTM.
This description makes sense. Thank you for your description.

Regards,

Hironori Bono