Stream: t-compiler

Topic: pre-meeting triage 2019-05-23 #54818


pnkfelix (May 23 2019 at 12:02, on Zulip):

I'll be doing pre-triage in this channel.

pnkfelix (May 23 2019 at 12:13, on Zulip):

first prepass: unprioritized nominated T-compiler

pnkfelix (May 23 2019 at 12:13, on Zulip):

first: "[LLVM] Segmentation fault on MacOS with > 1 codegen units and optimization > 0" #60925

pnkfelix (May 23 2019 at 12:18, on Zulip):

this is tough. It is such a nasty bug

pnkfelix (May 23 2019 at 12:19, on Zulip):

but if it is indeed an LLVM bug when handed symbol names with .llvm. in them ... then it seems easy to work around.

pnkfelix (May 23 2019 at 12:26, on Zulip):

I'm going to make this P-high, but only to resolve, as a team, how best to resolve this in very short term. I'm talking like, "do we add a lint for instances of using llvm as an identifier, warning about potential for hitting this."

pnkfelix (May 23 2019 at 12:27, on Zulip):

next: "Borrow checker on beta rejects code accepted on stable." #60927

pnkfelix (May 23 2019 at 12:28, on Zulip):

I don't think we're going to try to "fix" this regression; I believe its being lumped in as acceptable fallout from a soundness fix.

pnkfelix (May 23 2019 at 12:29, on Zulip):

I'm going to close this bug, but I'll also include a note about it on our main topic so everyone's aware

centril (May 23 2019 at 12:29, on Zulip):

I forgot to think about it for the blog post

pnkfelix (May 23 2019 at 12:29, on Zulip):

oh I guess I shouldn't close it until the blog post for the release is done

centril (May 23 2019 at 12:29, on Zulip):

thanks for the reminder

centril (May 23 2019 at 12:29, on Zulip):

@pnkfelix not sure if I should add something or not -- do we believe it is noteworthy?

centril (May 23 2019 at 12:30, on Zulip):

if so I need to do this really quickly cause it's release day today

pnkfelix (May 23 2019 at 12:30, on Zulip):

I am under the impression that the diagnostic does not give one much of a hint as to where the problem originates

Pietro Albini (May 23 2019 at 12:30, on Zulip):

and because I'm about to start the release process :sweat_smile:

Pietro Albini (May 23 2019 at 12:31, on Zulip):

(hi!)

centril (May 23 2019 at 12:31, on Zulip):

I guess my question is: do we think it will affect many users?

centril (May 23 2019 at 12:31, on Zulip):

otherwise it's not blog post noteworthy

Pietro Albini (May 23 2019 at 12:32, on Zulip):

I think it would be good to have a "known breakage" section either in the relnotes or the blog post for each release

centril (May 23 2019 at 12:32, on Zulip):

would prefer relnotes in that case

Pietro Albini (May 23 2019 at 12:32, on Zulip):

but that's more of a t-release thingy

pnkfelix (May 23 2019 at 12:37, on Zulip):

I don't think this is blog post worthy

pnkfelix (May 23 2019 at 12:37, on Zulip):

release notes seems fine to me.

centril (May 23 2019 at 12:37, on Zulip):

@pnkfelix yeah I agree :+1:

pnkfelix (May 23 2019 at 12:38, on Zulip):

@centril I'm going to remove the I-nominated tag (because I don't think we need to dedicate time to discussing #60927 during the triage meeting), but I will not close it unless you want to e.g. find some other way to track the work remaining regarding messaging.

pnkfelix (May 23 2019 at 12:41, on Zulip):

next: "ICE with incorrect turbofish" #60989

pnkfelix (May 23 2019 at 12:42, on Zulip):

hmm. bisected to a rollup of 37 PRs.

centril (May 23 2019 at 12:42, on Zulip):

fun :P

pnkfelix (May 23 2019 at 12:42, on Zulip):

@centril , would it make any sense to limit the number of PRs in a rollup?

centril (May 23 2019 at 12:42, on Zulip):

@pnkfelix only if you make infra give us more machines for bors

pnkfelix (May 23 2019 at 12:42, on Zulip):

(that, or invest time in CI infrastructure to provide e.g. prebuilt binaries of intermediate points within a rollup PR)

Pietro Albini (May 23 2019 at 12:43, on Zulip):

more machines for bors won't help that much btw

Pietro Albini (May 23 2019 at 12:43, on Zulip):

unless we get double or triple the number of builders to run builds in parallel

centril (May 23 2019 at 12:43, on Zulip):

@pnkfelix 37 PR rollups are rare and only happen when we have CI problems or if I have been on vacation or something

centril (May 23 2019 at 12:43, on Zulip):

unless we get double or triple the number of builders to run builds in parallel

Yes, that =P -- we should have that.

Pietro Albini (May 23 2019 at 12:44, on Zulip):

and at least with the current setup that's expensive

pnkfelix (May 23 2019 at 12:44, on Zulip):

but if they're rare, then there wouldn't be much cost to saying "only 10 PR's per rollup", right?

pnkfelix (May 23 2019 at 12:44, on Zulip):

because it would "just" quadruple the latency?

pnkfelix (May 23 2019 at 12:44, on Zulip):

(in those rare cases)

centril (May 23 2019 at 12:44, on Zulip):

@pnkfelix sometimes it is needed so I don't want any limits

centril (May 23 2019 at 12:45, on Zulip):

sometimes you have a bunch of trivial PRs and you get 10 PRs easily

pnkfelix (May 23 2019 at 12:45, on Zulip):

okay, yes

centril (May 23 2019 at 12:45, on Zulip):

over 10 PRs is common

pnkfelix (May 23 2019 at 12:45, on Zulip):

all I'm saying is that there is a pretty serious difference between trying to evaluate a 10 PR rollup in terms of trying to identify culprits post bisection

pnkfelix (May 23 2019 at 12:45, on Zulip):

vs evaluating a 37 PR rollup

pnkfelix (May 23 2019 at 12:46, on Zulip):

anyway

centril (May 23 2019 at 12:46, on Zulip):

@pnkfelix well yeah, you shouldn't put anything into a rollup -- we have a new system for that: @bors r+ rollup=never

centril (May 23 2019 at 12:46, on Zulip):

this tells me that you don't want the PR in a rollup

pnkfelix (May 23 2019 at 12:47, on Zulip):

anyway I've digressed from the main issue, which is how to prioritize #60989

pnkfelix (May 23 2019 at 12:47, on Zulip):

@centril you've added rollup=never because not enough people were utilizing @bors rollup ?

pnkfelix (May 23 2019 at 12:48, on Zulip):

(anyway I don't see how rollup=never addresses the point I was making about trying to do detective work on a 10 PR rollup vs a 37 PR rollup)

centril (May 23 2019 at 12:48, on Zulip):

@pnkfelix no; rollup=never is for big PRs, changes to bootstrap, rustbuild, perf sensitive stuff, etc.

centril (May 23 2019 at 12:49, on Zulip):

I rollup non-@bors rollup PRs all the time

pnkfelix (May 23 2019 at 12:49, on Zulip):

wait a minute

pnkfelix (May 23 2019 at 12:49, on Zulip):

is #60989 talking about a bug that affects the stable that is about to be released?

centril (May 23 2019 at 12:50, on Zulip):

@Pietro Albini ^----

pnkfelix (May 23 2019 at 12:50, on Zulip):

I rollup non-@bors rollup PRs all the time

and I'm saying: Is this sufficiently motivated

Pietro Albini (May 23 2019 at 12:50, on Zulip):

wait what

pnkfelix (May 23 2019 at 12:50, on Zulip):

blargh

Pietro Albini (May 23 2019 at 12:51, on Zulip):

let me check

centril (May 23 2019 at 12:51, on Zulip):

and I'm saying: Is this sufficiently motivated

Yes. Otherwise the number of PRs will accelerate and so will bors wait times

Pietro Albini (May 23 2019 at 12:51, on Zulip):

fuck

centril (May 23 2019 at 12:52, on Zulip):

@pnkfelix the best way to avoid big rollups that are hard to bisect is to just do more rollups more frequently

Pietro Albini (May 23 2019 at 12:52, on Zulip):

yes it affects the 1.35.0 which is about to be released

centril (May 23 2019 at 12:52, on Zulip):

other than that, it's the infra team's problem to fix the situation

Pietro Albini (May 23 2019 at 12:52, on Zulip):

I didn't start the process yet though

pnkfelix (May 23 2019 at 12:52, on Zulip):

pnkfelix the best way to avoid big rollups that are hard to bisect is to just do more rollups more frequently

Otherwise known as: Put an upper bound on the number of PRs in a rollup

centril (May 23 2019 at 12:53, on Zulip):

That's an artificial measure; I know better what to put in a rollup and what not to

centril (May 23 2019 at 12:53, on Zulip):

and I do rollups really frequently these days

centril (May 23 2019 at 12:53, on Zulip):

like... sometimes even 3 per day

pnkfelix (May 23 2019 at 12:53, on Zulip):

okay, but then

pnkfelix (May 23 2019 at 12:54, on Zulip):

break that 37 PR rollup into 4?

centril (May 23 2019 at 12:54, on Zulip):

37 is an extreme outlier due to an infra problem for a few days... it is not representative of normal operations

pnkfelix (May 23 2019 at 12:54, on Zulip):

I really am just having difficulty understanding the objection

Pietro Albini (May 23 2019 at 12:54, on Zulip):

can we have this discussion later?

centril (May 23 2019 at 12:54, on Zulip):

I think it might have been due to 259 problems

Pietro Albini (May 23 2019 at 12:54, on Zulip):

how bad is that regression

pnkfelix (May 23 2019 at 12:54, on Zulip):

yeah lets not talk about the rollup question right now

pnkfelix (May 23 2019 at 12:54, on Zulip):

(we can put that in a separate topic)

pnkfelix (May 23 2019 at 12:54, on Zulip):

I'm trying to figure out the impact of that regression, and whether it should make us delay the release

Pietro Albini (May 23 2019 at 12:54, on Zulip):

it seems like the code was erroring anyway and this is an error -> ICE

pnkfelix (May 23 2019 at 12:55, on Zulip):

I have to imagine it would be relatively easy to identify the cause

pnkfelix (May 23 2019 at 12:56, on Zulip):

but as @Pietro Albini notes, an error -> ICE is itself not as high-priority (right?)

centril (May 23 2019 at 12:56, on Zulip):

depends on how frequent the error occurs

Pietro Albini (May 23 2019 at 12:56, on Zulip):

well, we're not breaking any code

pnkfelix (May 23 2019 at 12:57, on Zulip):

of course its really painful for the person seeing the diagnostic

centril (May 23 2019 at 12:57, on Zulip):

if it regresses diagnostics experience substantially it should be a blocker

pnkfelix (May 23 2019 at 12:57, on Zulip):

how about this

Pietro Albini (May 23 2019 at 12:57, on Zulip):

it's a bad UX but I'm not sure if it's that bad to delay the release

pnkfelix (May 23 2019 at 12:57, on Zulip):

I'll leave it nominated

pnkfelix (May 23 2019 at 12:57, on Zulip):

and triage it as P-high

pnkfelix (May 23 2019 at 12:57, on Zulip):

let compiler team weigh in on whether its worth delaying the release or not

centril (May 23 2019 at 12:58, on Zulip):

maybe stable nominate also for the fix PR?

Pietro Albini (May 23 2019 at 12:58, on Zulip):

if we have a patch done in max a few hours we could do a stable rebuild and still be in time for releasing today

pnkfelix (May 23 2019 at 12:59, on Zulip):

@Pietro Albini are you going to be around for the compiler meeting itself? Would you be able to weigh in on questions about e.g. "if we have a beta-backportable fix in two days, how long would that delay release to incorporate it"

Pietro Albini (May 23 2019 at 12:59, on Zulip):

yep

pnkfelix (May 23 2019 at 12:59, on Zulip):

or I'm guessing its more like "we would not redo a beta promotion"

pnkfelix (May 23 2019 at 12:59, on Zulip):

(and we would instead stable-backport to the current beta-promotion)

pnkfelix (May 23 2019 at 12:59, on Zulip):

or maybe there's no effective difference between those two...

pnkfelix (May 23 2019 at 12:59, on Zulip):

okay, lets maybe just wait to have this conversation with the rest of the compiler team then.

Pietro Albini (May 23 2019 at 12:59, on Zulip):

honestly if y'all can manage to r+ a PR against stable in a few hours we can still make it for today

pnkfelix (May 23 2019 at 13:00, on Zulip):

triage #60989: P-high. Leaving I-nominated.

pnkfelix (May 23 2019 at 13:00, on Zulip):

(I'm not going to have anything ready in "a few hours" myself.

pnkfelix (May 23 2019 at 13:00, on Zulip):

especially if each pre-triage item continues to take this much time.)

pnkfelix (May 23 2019 at 13:01, on Zulip):

It does seem like bugs tagged with regression-from-stable-to-beta during the last week before a release should probably get some sort of eager treatment

pnkfelix (May 23 2019 at 13:02, on Zulip):

e.g. maybe I and other compiler team members should be scanning for such bugs each day

Pietro Albini (May 23 2019 at 13:02, on Zulip):

and ping release

pnkfelix (May 23 2019 at 13:03, on Zulip):

next: "rls no longer builds after rust-lang/rust#61027" #61031

pnkfelix (May 23 2019 at 13:07, on Zulip):

I don't think this is P-high

pnkfelix (May 23 2019 at 13:09, on Zulip):

Who maintains the toolstate checker that rust-highfive reports?

Pietro Albini (May 23 2019 at 13:09, on Zulip):

infra/kenny

pnkfelix (May 23 2019 at 13:09, on Zulip):

I'm wondering whether we could be providing links to the build logs that show the failures

pnkfelix (May 23 2019 at 13:09, on Zulip):

okay I'll go ask in Discord

pnkfelix (May 23 2019 at 13:14, on Zulip):

anyway, triage #61031: P-medium, removing I-nominated. Assigned to self to track, but also leaving assigned to Xanewok and Centril.

pnkfelix (May 23 2019 at 13:15, on Zulip):

its also possible that #61031 was in fact fixed, given that toolstate reported that rls broke again in #61071.

pnkfelix (May 23 2019 at 13:20, on Zulip):

next: "ICE with lifetime'd associated type on stable" #61064

pnkfelix (May 23 2019 at 13:21, on Zulip):

okay this is a well known problem

pnkfelix (May 23 2019 at 13:21, on Zulip):

I think I'm going to assign it to myself because I'm tired of seeing issues filed about it.

pnkfelix (May 23 2019 at 13:22, on Zulip):

maybe I shouldn't say "well known"

pnkfelix (May 23 2019 at 13:22, on Zulip):

I thought there was some other issue where someone pointed out that we were ICE'ing after reporting the error due to the unstable feature

pnkfelix (May 23 2019 at 13:22, on Zulip):

but none of the linked issues seem to be pointing that out

pnkfelix (May 23 2019 at 13:28, on Zulip):

anyway: triage #61064: P-medium. Removing I-nominated. Closing as dupe of #60654. Assigning #60654 to self.

pnkfelix (May 23 2019 at 13:35, on Zulip):

last: "rls no longer builds after rust-lang/rust#60740" #61071

pnkfelix (May 23 2019 at 13:36, on Zulip):

this seems to be well in hand thanks to @Philipp Hansch

pnkfelix (May 23 2019 at 13:37, on Zulip):

triage #61071: P-medium, removing nominated label.

pnkfelix (May 23 2019 at 13:42, on Zulip):

next prepass: nominated issues not tagged with T-compiler but may nonetheless be relevant to use

pnkfelix (May 23 2019 at 13:42, on Zulip):

(the list here may need to be updated to include -label:T-release)

Pietro Albini (May 23 2019 at 13:43, on Zulip):

that's a bit relevant for the compiler team

pnkfelix (May 23 2019 at 13:43, on Zulip):

and maybe #60459 is too

Pietro Albini (May 23 2019 at 13:43, on Zulip):

yeah, for that

pnkfelix (May 23 2019 at 13:43, on Zulip):

how important is this? Feels P-medium to me.

pnkfelix (May 23 2019 at 13:44, on Zulip):

but #60459 might warrant P-high ...

Pietro Albini (May 23 2019 at 13:44, on Zulip):

with a t-release hat I'd like that check to be in place again asap

pnkfelix (May 23 2019 at 13:45, on Zulip):

oh dear

pnkfelix (May 23 2019 at 13:45, on Zulip):

#60459 is a beta-regression too ?

pnkfelix (May 23 2019 at 13:45, on Zulip):

which means it also might be part of the pending release?

Pietro Albini (May 23 2019 at 13:46, on Zulip):

no

Pietro Albini (May 23 2019 at 13:46, on Zulip):

it's a beta-1.36-regression, not a beta-1.35-soon-to-be-stable-regression

pnkfelix (May 23 2019 at 13:47, on Zulip):

...

pnkfelix (May 23 2019 at 13:47, on Zulip):

what is the beta on the playpen?

Pietro Albini (May 23 2019 at 13:47, on Zulip):

1.36

Pietro Albini (May 23 2019 at 13:48, on Zulip):

I also checked locally with the pre-release and it errors fine

pnkfelix (May 23 2019 at 13:48, on Zulip):

So ... do we release the result of the nightly->beta rollover before the stable resulting from the beta->stable rollover?

pnkfelix (May 23 2019 at 13:48, on Zulip):

I'm mostly trying to figure out if I can use the playpen as a way to roughly observe the pending release behavior

pnkfelix (May 23 2019 at 13:49, on Zulip):

I also checked locally with the pre-release and it errors fine

okay, thank you for checking this.

Pietro Albini (May 23 2019 at 13:50, on Zulip):

monday: beta->stable, and that is released as the pre-release on dev-static
tuesday: master->beta, and that is released as the new beta on static (prod)
thursday: the pre-release is moved to static (prod)

Pietro Albini (May 23 2019 at 13:50, on Zulip):

so yes

pnkfelix (May 23 2019 at 13:50, on Zulip):

i see

pnkfelix (May 23 2019 at 13:51, on Zulip):

what's the motivation for doing master->beta and beta => static before pre-release => static ?

pnkfelix (May 23 2019 at 13:52, on Zulip):

in particular, I would think the methodology of referencing play.rust-lang.org to predict forthcoming results is ... worthwhile?

Pietro Albini (May 23 2019 at 13:52, on Zulip):

dunno, that process was made before I joined the project

pnkfelix (May 23 2019 at 13:52, on Zulip):

okay

Pietro Albini (May 23 2019 at 13:52, on Zulip):

probably not to have beta being released on a friday?

pnkfelix (May 23 2019 at 13:52, on Zulip):

true

pnkfelix (May 23 2019 at 13:52, on Zulip):

hmm

pnkfelix (May 23 2019 at 13:53, on Zulip):

still, seems slightly unfortunate

pnkfelix (May 23 2019 at 13:53, on Zulip):

but I won't dwell on that at the moment

pnkfelix (May 23 2019 at 13:54, on Zulip):

(the other, maybe silly question, is whether there's a way to ask play.r-l.o "what versions do you have deployed")

pnkfelix (May 23 2019 at 13:54, on Zulip):

(because I know there's been lots of times that I've wanted to know "the truth" there)

Pietro Albini (May 23 2019 at 13:54, on Zulip):

it's shown in the dropdown

pnkfelix (May 23 2019 at 13:54, on Zulip):

oh of course

Pietro Albini (May 23 2019 at 13:54, on Zulip):

totally unrelated, I bisected the error->ICE regression

Pietro Albini (May 23 2019 at 13:55, on Zulip):

https://github.com/rust-lang/rust/issues/60989#issuecomment-495228473

pnkfelix (May 23 2019 at 13:55, on Zulip):

yeah okay. so looking at that, one can clearly see that right now, stable is 1.34.2 and beta is 1.36.0-beta.1; and thus 1.35 behavior is not currently observable via play

Pietro Albini (May 23 2019 at 13:55, on Zulip):

yep

pnkfelix (May 23 2019 at 13:55, on Zulip):

shoot, that PR is also fixing an ICE

pnkfelix (May 23 2019 at 13:56, on Zulip):

but maybe will help point to a targetted fix

pnkfelix (May 23 2019 at 13:57, on Zulip):

okay, anyway

pnkfelix (May 23 2019 at 13:57, on Zulip):

regarding #60929, based on @Pietro Albini (w/ T-release hat)'s comments, I'll make it P-high and assign to self.

pnkfelix (May 23 2019 at 13:59, on Zulip):

next prepass: zero unprioritized beta regressions

pnkfelix (May 23 2019 at 14:00, on Zulip):

next prepass: one unprioritized nightly regression

pnkfelix (May 23 2019 at 14:00, on Zulip):

namely "String::as_ref() Into<String> doesn't compile on nightly" #60958

pnkfelix (May 23 2019 at 14:00, on Zulip):

but this is libs

pnkfelix (May 23 2019 at 14:00, on Zulip):

and already deemed acceptable breakage

pnkfelix (May 23 2019 at 14:00, on Zulip):

so not our problem

Last update: Nov 22 2019 at 05:10UTC