touch: more fixes in the message pump.

touch: more fixes in the message pump.

Remove the signal handler when the message-pump terminates. This wouldn't
normally be necessary, but it can happen that the message-pump is terminated,
and then additional GTK+ operations happen (e.g. in unit_tests).

Also check for non-null dispatcher before dispatching events. This can happen
for the first-run dialog, and also in the unit_tests.

BUG=none
TEST=unit_tests

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=76643

[sadrul, 2011-03-02 22:26:17]

Opinions?

[Evan Martin, 2011-03-02 22:31:20]

LGTM, though I admit I don't understand it too well

[sadrul, 2011-03-02 23:06:43]

On 2011/03/02 22:31:20, Evan Martin wrote:
> LGTM, though I admit I don't understand it too well

MessagePumpGlibX sets up a callback for the "realize" signal on creation. So
RemoveGtkWidgetRealizeNotifier is used to clean it up when the message-pump is
destroyed. This wouldn't normally be necessary for chrome, since not much is
done after the UI message-pump is terminated. But unit_tests creates multiple
message-pumps for the different tests, and the "realize" callbacks from
destroyed pumps get invoked, which causes it to crash.

The check for non-null dispatcher is because in some cases (e.g. for the
first-run dialog), MessagePumpForUI::Run is called, which runs the pump without
any dispatcher. So the check helps avoid a crash in such cases.

I hope that makes a bit more sense?