I did a quick check of this, and the two CLs seemed conflicting with each other (fs@opera.com also pointed out in r419384).

avi@, what results do you expect Findit to surface on Sheriff-o-Matic in this case?


Heuristic analysis caught both CLs as shown in https://findit-for-me.appspot.com/waterfall/build-failure?url=https%3A%2F%2Fbuild.chromium.org%2Fp%2Fchromium%2Fbuilders%2FAndroid%2Fbuilds%2F62334
But try-job rerun of the compile could only catch one CL in the current design, and it is the later commit (in this case r419384 instead of r419382).
Currently, try-job results take priority and show on Sheriff-o-Matic, because they are more reliable than heuristic results.

lijeffrey@, assign to you to follow up, as you were on findit rotation when the compile failures occurred.

> avi@, what results do you expect Findit to surface on Sheriff-o-Matic in this case?

Honestly, given those two CLs, I completely don't blame Findit, and I didn't have a problem working out the real culprit even without it. But you wanted examples of Findit failing, so I sent this to you.

I don't know if this is actionable; I wouldn't mind if you closed this if you couldn't easily improve things.

A quick glance at this one and something seems not right. There are 3 CLs in the regression range, the first of them was the culprit (call it r1), the middle a deps roll (call it r2), then the last the similar CL that Findit thinks was the culprit (call it r3).

Because both r1  and r3 showed up as hints in Findit's heuristic analysis, the try job was guided to test r2 (expecting it to pass) and r3 (expecting it to fail), which is exactly what happened.

However what I'm curious about is why r2 passed in the first place, since r1 would already have been in there causing problems.

I don't have bandwidth to look into this any further for this week, but for now I think it's worth keeping open since it seems like something else might be wrong.

The scenario is conflicting CLs landed around the same time.
This is not actionable for Findit, so closed.