Mac: make it even less likely that the renderer+Breakpad leaks Mach ports.

Mac: make it even less likely that the renderer+Breakpad leaks Mach ports.

SuicideOnChannelErrorFilter::OnChannelError() does get called under certain
(unknown) circumstances, with some regularity. This causes a leak of a Mach port
when used with Breakpad.

BUG=28547
TEST=See bug.

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=34534

[viettrungluu, 2009-12-15 01:34:32]

Sending this out, but let me go make a branded build and test it....

[John Grabowski, 2009-12-15 01:48:13]

LGTM

Be sure you add extra into to krisr's confirmation script to help be sure this
gets hit.

[Mark Mentovai, 2009-12-15 04:14:58]

LGTM on the change itself.  Additional thoughts:

1. 249 branch?  I think mal already did a build to use for the beta
update, you might want to talk to him.

2. Do we need to do anything for other process types, like the plugin process?

Mark

[viettrungluu, 2009-12-15 08:02:31]

On 2009/12/15 04:14:58, Mark Mentovai wrote:
> LGTM on the change itself.  Additional thoughts:
> 
> 1. 249 branch?  I think mal already did a build to use for the beta
> update, you might want to talk to him.

Yes, that's the plan. I'll will talk to mal.

> 2. Do we need to do anything for other process types, like the plugin process?

That's a good question. A very cursory skimming of the code suggests that the
plugin process may need it too, but it's hard to tell. I haven't seen a leak
which I can definitely (or even likely) pin on the plugin process, though it's
hard to tell when there are multiple leaks.

I think we should throw have the testers hammer on this, and be satisfied if it
leaks slowly enough.

- Trung