Came across an unusual case while doing security triage. Unless I'm missing something, this looks like a pretty serious bug.

See https://cluster-fuzz.appspot.com/v2/testcase-detail/5918821207769088?noredirect=1 for the case in question.

It looks like we're not correctly identifying the crashing line in the older revisions we're diffing against. The CL only made comment changes, but it's identified as landing on the crashing line exactly. While it did modify the line number we had identified as crashing, that was the line number in the older CL. The actual crashing line is elsewhere, and can be computed based on the diff.