We gathered timing information for the parts of the benchmark that it did complete
Which parts didn't complete? Did you try the beta or aurora?
Code that can consistently crash firefox would probably be elgible for a security bounty.
Which parts didn't complete? Did you try the beta or aurora?
Code that can consistently crash firefox would probably be elgible for a security bounty.
The only thing I could even measure that would lead to a crash is memory usage. I just re-ran the test and watched the process's memory usage. It hovered in the 300-400MB range during the first third of the test. Then, during the test I describe, memory usage rocketed to around 1GB, at which point the browser crashed hard.
I'm not sure why Firefox would exhibit this behavior when other browsers don't, but we know a few people at Mozilla (they're customers of ours) so we'll probably reach out.
Seeing "We're sorry. Firefox had a problem and crashed." as well as its exclusion from the performance summary made it feel like there was a bit of a bias against Firefox, where even in your conclusion it's stated that it performs "quite well" - that sentence does not reflect on the rest of the article at all.
Maybe it was hardware acceleration in the particular machine with the particular video driver you use?
Can you try the test on an entirely different video setup just to prove it can get through, even if the numbers cannot be compared?