the crater queue is empty atm ( :heart: to microsoft for the additional builder)
if you want to experiment with new runs this is the best time for that!
I feel like I shouldn't have seen those messages :P
just don't queue 20 runs pls
no promises, mkay?
the good news is that I forgot what I wanted crater for, other than perhaps an upcoming PR that has some chance of breaking something obscure
ohhhh riiight debuginfo. does crater compile in debug mode, with the full implication of
iirc crater does indeed use debug mode
you can add custom rustflags anyway
because I'm messing around with that, and I'm worried some sanity checks will get tripped in the wild
@eddyb maybe we want crater runs for something const / promotion related?
maybe we could run the old/new pass in a crater run
@centril we already did one of those recently, at least for my old/new :P
oh we did?
but yeah, we should've done it for @ecstatic-morse's validator too
yeah I mean morse's validator
(we should probably just setup monitoring to page @eddyb when crater is idle)
@centril my PR that's in the queue, about promotion, that waited for a while on a crater run
@Pietro Albini hmm did we ever do a crater run for -Zsymbol-mangling-version=v0? oh yeah we did cause we found some weird crates using the backtrace crate on themselves to reflect a module/function path
not sure :sweat_smile:
hmm; I looked through all the PRs but didn't feel I wanted to crater run anything in particular
@nikomatsakis maybe you want to do a run on your leak-check stuff?
@centril I was thinking of coming back to that today -- uh -- yeah, good ida!
I think the PR is in a state where it could be cratered
well, why don't I rebase it I guess
crater is for try runs which will rebase for you :)
(presuming no conflicts)
rebasing shouldn't take that long
Is it possible to do a crater run that compares the parent of #64470 with a combination of #64470 and #65389? (I'd need to create a try build that contains the latter two). I think the bug that was fixed by #65389 will come up quite often and maybe make it tough to find new bugs.
But yes! We should do this.
@ecstatic-morse it should be possible to make crater compare any two bors commits -- I can run the command if you ping me here or whereever
@simulacrum I will have to create a new bors commit that is #64470 with #65389 cherry picked on it. Will bors complain about merge conficts when I try to schedule a try build since these commits are weeks old?
uh, probably? you can -- if necessary -- try to revert all the needed commits on your PR or w/e
or we can generate 1-2 separate try builds now with the relevant changes
Okay, I'll try some stuff. I could just run it on everything between, but there might be some unrelated regressions in there.
I don't have time to deal with the generation of the try builds but I can run the final crater if you don't have the permissions (not sure, you might)
kk. I'll ping you (I think I'm still not on the crater list)
git diff --shortstat master 2032 files changed, 27656 insertions(+), 36897 deletions(-)
I now know how a 10x developer feels
@simulacrum Will broken links in updated submodules cause a try build to fail?
I guess there's some links directly to https://doc.rust-lang.org/nightly/ that have become outdated
Disregard the above. crater queue is no longer empty.
broken links should be fine
crater queue is empty enough
My crater experiment is progressing so fast that I fear something has gone terribly wrong :smile:
20% after 5 hours can't be right, even distributed across two agents
Looks like metrics show that we've slowed down since then
we're still averaging ~70 jobs/minute across the run, ~40 jobs/minute in the last hour
the one run I sampled seemed to go fine