Bugzilla

(In reply to Daniel Holbert [:dholbert] from comment #14)
> So I suspect the actual regression here (on Jan 23 or whenever) was in whatever code processes these logs, not a Gecko regression.

It's also possible that this has always been an issue, but we happened to be avoiding it because these expected-crash WPTs were previously  at the end of a typically-green WPT test "bucket" (and hence weren't being flagged as part of any orange test-runs); and the proximal "regression" may have been that somebody just added WPTs that reshuffled the WPT buckets such that these expected-crashes are now at the start of a bucket that happens to contain some (known) intermittent-failures, which means the crashes now show up (inadvertently, as noise) in occasionally-orange test logs.

(In reply to Daniel Holbert [:dholbert] from comment #14)
> So I suspect the actual regression here (on Jan 23 or whenever) was in whatever code processes these logs, not a Gecko regression.

It's also possible that this has always been a potential-issue, but we happened to be avoiding it because these expected-crash WPTs were previously  at the end of a typically-green WPT test "bucket" (and hence weren't being flagged as part of any orange test-runs); and the proximal "regression" may have been that somebody just added WPTs that reshuffled the WPT buckets such that these expected-crashes are now at the start of a bucket that happens to contain some (known) intermittent-failures, which means the crashes now show up (inadvertently, as noise) in occasionally-orange test logs.

(In reply to Daniel Holbert [:dholbert] from comment #14)
> So I suspect the actual regression here (on Jan 23 or whenever) was in whatever code processes these logs, not a Gecko regression.

It's also possible that this has always been a potential-issue, but we happened to be avoiding it because these expected-crash WPTs were previously  at the end of a typically-green WPT test "bucket" (and hence weren't being flagged as part of any orange test-runs); and the proximal "regression" may have been that somebody just added WPTs that reshuffled the WPT buckets, such that these expected-crashing tests are now at the start of a bucket that happens to contain some (known) intermittent-failures, which means the crashes now show up (inadvertently, as noise) in occasionally-orange test logs.

(In reply to Daniel Holbert [:dholbert] from comment #14)
> So I suspect the actual regression here (on Jan 23 or whenever) was in whatever code processes these logs, not a Gecko regression.

~It's also possible that this has always been a potential-issue, but we happened to be avoiding it because these expected-crash WPTs were previously  at the end of a typically-green WPT test "bucket" (and hence weren't being flagged as part of any orange test-runs); and the proximal "regression" may have been that somebody just added WPTs that reshuffled the WPT buckets, such that these expected-crashing tests are now at the start of a bucket that happens to contain some (known) intermittent-failures, which means the crashes now show up (inadvertently, as noise) in occasionally-orange test logs.~