Created attachment 545610 [details] [diff] [review]
Kill content processes that cause processing errors in the chrome process
Mark the tracking flag when this lands: we'll want release drivers to watch the crash stats for Fennec to see if anything bad shows up.
test_process_error.xul (which tests crashing the content process) is failing with this patch, because on windows the content process tries to set up D3D10 layers but they're not supported yet. Something causes a protocol error, and then BAM the content process is killed. It's good that this patch is working, but this reveals a fundamental problem: the kill-on-protocol-error approach isn't going to be able to trigger breakpad on windows. Nor on Mac. So we have to decide between killing a possibly-pwned process but getting crash reports from kills on plain-jane errors, or keeping things the way they are. Will think on't.
Darnit, I forget we never landed this.
Can this land?
I've been sorting out test failures since Monday.