Disable input events when a JS dialog is showing.

Disable input events when a JS dialog is showing.

Prior to this patch, you could still send input events and they
would get queued in the renderer to be handled after the JS
dialog was dismissed.  This makes our behavior more like other
browsers.

This also works around a linux/mac bug where the hung renderer dialog would appear when an alert is showing.

BUG=12176

[tony, 2009-08-01 01:38:07]

I'm sending this out mainly for feedback.

Currently, if an alert is showing, you can still send input events to other tabs
in the same process, they just don't get run until the alert is dismissed. 
Here's an example:
  http://www/~tc/tests/chrome/popup-no-location.html

If you open the popup, click the alert, click the "hello world" link in the
popup, then dismiss the alert, the page will navigate after the alert is
dismissed.  This patch prevents this type of behavior (which seems undesireable)
from happening.

Thoughts?  Is this a bad idea?

[tony, 2009-08-03 20:51:58]

ping

On 2009/08/01 01:38:07, tony wrote:
> I'm sending this out mainly for feedback.
> 
> Currently, if an alert is showing, you can still send input events to other
tabs
> in the same process, they just don't get run until the alert is dismissed. 
> Here's an example:
>   http://www/~tc/tests/chrome/popup-no-location.html
> 
> If you open the popup, click the alert, click the "hello world" link in the
> popup, then dismiss the alert, the page will navigate after the alert is
> dismissed.  This patch prevents this type of behavior (which seems
undesireable)
> from happening.
> 
> Thoughts?  Is this a bad idea?

[darin (slow to review), 2009-08-03 20:58:21]

LGTM

But, with this change won't we end up in situations where we only get a
mouse/key up event without a preceding mouse/key down event?  Could that screw
things up?

[jam, 2009-08-03 21:01:51]

lgtm, I don't see any issues with the approach.

[tony, 2009-08-04 20:39:12]

On 2009/08/03 20:58:21, darin wrote:
> LGTM
> 
> But, with this change won't we end up in situations where we only get a
> mouse/key up event without a preceding mouse/key down event?  Could that screw
> things up?

It doesn't seem to cause any problems based on some manual testing.  You can
actually already do this (e.g., mouse/key down outside of chrome then mouse/key
up inside chrome).

I also added this to the onbeforeunload dialog.

[darin (slow to review), 2009-08-04 21:22:27]

LGTM