Stream: t-compiler

Topic: pre-meeting triage 2020-02-20 #54818


pnkfelix (Feb 20 2020 at 12:54, on Zulip):

I will be doing pre-triage in this channel.

pnkfelix (Feb 20 2020 at 13:04, on Zulip):

I'm going to try to get through the agenda as best I can, but I had a rough night last night (very sick kid) and will need to go AFK for 40min or so very soon

pnkfelix (Feb 20 2020 at 13:04, on Zulip):

anyway I'll do what I can now, and resume again when I can

pnkfelix (Feb 20 2020 at 13:07, on Zulip):

First up: unprioritized nominated issues

pnkfelix (Feb 20 2020 at 13:07, on Zulip):

there are nine

pnkfelix (Feb 20 2020 at 13:08, on Zulip):

unpri nom 1/9: "internal compiler error: src/librustc_mir_build/hair/constant.rs:60: impossible case reached" #69310

pnkfelix (Feb 20 2020 at 13:09, on Zulip):

seems bad. doesn't include any useful diagnostic prior to ICE'ing. Its probably because the HAIR builder assumes you cannot have floating point numbers as array lengths. (and maybe that is supposed to be impossible in the HAIR.)

pnkfelix (Feb 20 2020 at 13:09, on Zulip):

its also a stable-to-beta regression

pnkfelix (Feb 20 2020 at 13:10, on Zulip):

P-high, assigning to self. But also: @Nell Shamrell-Harrington , would you be interesting in partnering up on this?

pnkfelix (Feb 20 2020 at 13:11, on Zulip):

(removing from nominations list)

pnkfelix (Feb 20 2020 at 13:11, on Zulip):

unpri nom 2/9: "ICE: src/librustc/middle/region.rs:1037: Encountered greater count 28" #69307

pnkfelix (Feb 20 2020 at 13:12, on Zulip):

happens on stable. not clear to be if this ever worked.

pnkfelix (Feb 20 2020 at 13:13, on Zulip):

its already tagged with A-async-await; @nikomatsakis , am I right that the combination of that tag and I-nominated means that it will be triaged by the async-await WG?

LeSeulArtichaut (Feb 20 2020 at 13:13, on Zulip):

(I'm currently downloading 1.39 to see if it already ICE's)

pnkfelix (Feb 20 2020 at 13:15, on Zulip):

(if I'm correct about this being in the WG-async-await queue, we probably need to figure out some system for marking things as "in the triage queue for team X". Or more specifically, some way for T-compiler to tag something as "not actually nominated for us...")

pnkfelix (Feb 20 2020 at 13:16, on Zulip):

(but I'm also not clear on whether this bug is best handled by WG-async-await or T-compiler, given that its arising from the region system.)

pnkfelix (Feb 20 2020 at 13:17, on Zulip):

#69307 triage: P-high. Leaving I-nominated as to not disrupt WG-async-await workflow.

pnkfelix (Feb 20 2020 at 13:18, on Zulip):

unpri nom 3/9: "Inconsistent Self behavior with generic structs" #69306

pnkfelix (Feb 20 2020 at 13:24, on Zulip):

(I'm leaving some comments)

pnkfelix (Feb 20 2020 at 13:29, on Zulip):

triage: P-high. Leaving nominated for T-lang to discuss.

pnkfelix (Feb 20 2020 at 13:30, on Zulip):

unpri nom 4/9: "Compiler bug: type parameter T/#0 (T/0) out of range when substituting" #69296

pnkfelix (Feb 20 2020 at 13:31, on Zulip):

doesn't include any useful diagnostic prior to ICE'ing.

pnkfelix (Feb 20 2020 at 13:32, on Zulip):

triage: P-high. Assigning to self. Removing nomination.

pnkfelix (Feb 20 2020 at 13:33, on Zulip):

unpri nom 5/9 "rls no longer builds after rust-lang/rust#69265" #69286

pnkfelix (Feb 20 2020 at 13:34, on Zulip):

historically we've treated tool failures like this as P-medium

pnkfelix (Feb 20 2020 at 13:35, on Zulip):

(and its auto-assigned to Xanewok)

pnkfelix (Feb 20 2020 at 13:35, on Zulip):

I'm sort of surprised we don't have a tool-failure label, BTW

pnkfelix (Feb 20 2020 at 13:35, on Zulip):

triage: P-medium. Removing nomination label.

pnkfelix (Feb 20 2020 at 13:37, on Zulip):

unpri nom 6/9: "const generic ICE on stable" #69239

pnkfelix (Feb 20 2020 at 13:43, on Zulip):

left some comnents but gotta go AFK

pnkfelix (Feb 20 2020 at 13:54, on Zulip):

okay back

pnkfelix (Feb 20 2020 at 13:55, on Zulip):

Hmm I wonder if this is related to #69296 ...

pnkfelix (Feb 20 2020 at 13:56, on Zulip):

anyway this isn't as bad as the title makes it sound; we do emit the expected diagnostic alerting the user to the fact that they are making use of an unstable feature. We just don't include the info about #![feature(...)] in this context.

pnkfelix (Feb 20 2020 at 13:58, on Zulip):

triage: P-medium, removing nomination label. Assigning to self. (Since this is leveraging a nightly-only feature, and we do emit the diagnostic saying so, this gets relatively lower priority.)

pnkfelix (Feb 20 2020 at 13:59, on Zulip):

unpri nom 7/9: "Memory unsafety problem in safe Rust" #69225

pnkfelix (Feb 20 2020 at 14:01, on Zulip):

lots of discussion.

pnkfelix (Feb 20 2020 at 14:02, on Zulip):

seems like an LLVM misoptimziation, which can be worked-around (papered-over ) by reverting a Rust stdlib change (done in #69241) that should be valid in principle.

pnkfelix (Feb 20 2020 at 14:04, on Zulip):

ooh awesome @comex made an LLVM IR reproducer!

pnkfelix (Feb 20 2020 at 14:05, on Zulip):

comex comment with LLVM IR reproducer . That comment relays a claim that LLVM 10 is not affected by this.

pnkfelix (Feb 20 2020 at 14:06, on Zulip):

and in fact points out a specific reversion commit that we may be able to apply to our version of LLVM

pnkfelix (Feb 20 2020 at 14:06, on Zulip):

(and then presumably readd the change from PR #67174 )

pnkfelix (Feb 20 2020 at 14:07, on Zulip):

SO: What priority to give this...

simulacrum (Feb 20 2020 at 14:07, on Zulip):

we are almost in place to land llvm 10, fwiw

simulacrum (Feb 20 2020 at 14:07, on Zulip):

though it's blocked on sort of "performance"

simulacrum (Feb 20 2020 at 14:08, on Zulip):

I'd personally say p-med as we're not actively planning on taking action here, I would not personally aim to bother with the llvm backport, but I guess we could

pnkfelix (Feb 20 2020 at 14:08, on Zulip):

It seems like the remaining work items for #69225 are 1. fix LLVM (either by cherry-picking their 58e8c793d0e43150a6452e971a32d7407a8a7401 or by upgrading to LLVM 10) and then 2. readd PR #67174.

simulacrum (Feb 20 2020 at 14:08, on Zulip):

it's not clear to me that this affects more cases

pnkfelix (Feb 20 2020 at 14:08, on Zulip):

however neither of these strike me as high-priority items.

pnkfelix (Feb 20 2020 at 14:09, on Zulip):

At least, this LLVM bug does not seem any better or worse than other LLVM codegen bugs. Which I guess is what @simulacrum just said.

pnkfelix (Feb 20 2020 at 14:11, on Zulip):

triage: P-medium. Removing nomination.

pnkfelix (Feb 20 2020 at 14:11, on Zulip):

leaving unassigned.

pnkfelix (Feb 20 2020 at 14:11, on Zulip):

@simulacrum is there an issue tracking the effort to upgrade to LLVM 10 ?

Wesley Wiser (Feb 20 2020 at 14:12, on Zulip):

The PR is #67759

Wesley Wiser (Feb 20 2020 at 14:13, on Zulip):

I don't know of a tracking issue

pnkfelix (Feb 20 2020 at 14:13, on Zulip):

PR is good enough for me.

simulacrum (Feb 20 2020 at 14:14, on Zulip):

yes, I would also note that I wouldn't feel comfortable just backporting the LLVM thing

simulacrum (Feb 20 2020 at 14:14, on Zulip):

it could have other implications

pnkfelix (Feb 20 2020 at 14:15, on Zulip):

hmm. I had impression that we cherry-picked the commit being reverted

pnkfelix (Feb 20 2020 at 14:15, on Zulip):

but I guess your point is that there must have been a reason for that

pnkfelix (Feb 20 2020 at 14:15, on Zulip):

and those would be the biggest potential downstream effects

pnkfelix (Feb 20 2020 at 14:15, on Zulip):

unpri nom 8/9: "_ in associated types causes nightly and beta compiler to panic" #69204

pnkfelix (Feb 20 2020 at 14:16, on Zulip):

triage: P-high. Has PR. Removing nomination tag.

pnkfelix (Feb 20 2020 at 14:16, on Zulip):

(awessome work @Esteban Küber !)

simulacrum (Feb 20 2020 at 14:18, on Zulip):

yes, correct

pnkfelix (Feb 20 2020 at 14:19, on Zulip):

unpri nom 9/9: "Worsened debug build codegen in beta" #68855

pnkfelix (Feb 20 2020 at 14:20, on Zulip):

glad to see there was discussion of the matter on the ticket

pnkfelix (Feb 20 2020 at 14:20, on Zulip):

but I cannot imagine this to be P-high priority.

pnkfelix (Feb 20 2020 at 14:21, on Zulip):

triage: P-medium. I'd like to assign it to someone. Who is best to take lead here: @oli ? @eddyb ? @Wesley Wiser ?

Wesley Wiser (Feb 20 2020 at 14:22, on Zulip):

I'll take point on that

pnkfelix (Feb 20 2020 at 14:22, on Zulip):

assigning to @Wesley Wiser and removing nomination.

pnkfelix (Feb 20 2020 at 14:23, on Zulip):

okay that's all the un-prioritized nominations

pnkfelix (Feb 20 2020 at 14:23, on Zulip):

next: unprioritized beta regressions

pnkfelix (Feb 20 2020 at 14:23, on Zulip):

there are 2

pnkfelix (Feb 20 2020 at 14:24, on Zulip):

unpri β regr 1/2: "beta regression: ICE on Tried to access field 0 of union Layout" #69191

pnkfelix (Feb 20 2020 at 14:26, on Zulip):

it would be good to identify which PR from #68078 injected the regression

pnkfelix (Feb 20 2020 at 14:26, on Zulip):

@simulacrum hey, the builds that the CI keeps around that we use for cargo-bisect-rustc ... does that include the CI builds for all PR's ?

simulacrum (Feb 20 2020 at 14:27, on Zulip):

if I follow you, that's the only builds we have?

pnkfelix (Feb 20 2020 at 14:27, on Zulip):

namely, is there some dumb way we could have it traverse rollup PR's ?

simulacrum (Feb 20 2020 at 14:27, on Zulip):

no, because rollup PRs don't produce artifacts (unless someone ran try)

pnkfelix (Feb 20 2020 at 14:27, on Zulip):

what I mean is, do we keep around build artifacts for PR's that weren't directly merged

pnkfelix (Feb 20 2020 at 14:27, on Zulip):

okay

simulacrum (Feb 20 2020 at 14:27, on Zulip):

only stuff that has a bors merge commit will have artifacts

pnkfelix (Feb 20 2020 at 14:27, on Zulip):

is it possible for someone to run try on a closed PR?

simulacrum (Feb 20 2020 at 14:27, on Zulip):

(incl. try builds and auto builds)

simulacrum (Feb 20 2020 at 14:27, on Zulip):

not today, no

pnkfelix (Feb 20 2020 at 14:28, on Zulip):

I'm basically wondering out loud if we can get these build artifacts for rollup PR's after the fact

pnkfelix (Feb 20 2020 at 14:28, on Zulip):

without forcing people to do it locally

simulacrum (Feb 20 2020 at 14:28, on Zulip):

I have historically wanted to make rollups produce artifacts for just linux, I suspect we have capacity for that

simulacrum (Feb 20 2020 at 14:28, on Zulip):

but it's somewhat painful to do architecturally today

pnkfelix (Feb 20 2020 at 14:29, on Zulip):

anyway

Santiago Pastorino (Feb 20 2020 at 14:29, on Zulip):

another thing we could do is have cargo-bisect-rustc figuring out it found a rollup and going down automatically with a bisect on the rust repo

Santiago Pastorino (Feb 20 2020 at 14:30, on Zulip):

it will take a lot of time because one would need to build rustc but at least is automated

pnkfelix (Feb 20 2020 at 14:30, on Zulip):

@Santiago Pastorino but cargo-bisect-rustc doesn't do local builds, does it?

pnkfelix (Feb 20 2020 at 14:30, on Zulip):

are you suggesting you we would add that?

simulacrum (Feb 20 2020 at 14:30, on Zulip):

I am against adding that support, I think

Santiago Pastorino (Feb 20 2020 at 14:30, on Zulip):

it does not, I'm suggesting that we could implement that

simulacrum (Feb 20 2020 at 14:30, on Zulip):

seems quite complicated

pnkfelix (Feb 20 2020 at 14:30, on Zulip):

maybe there

pnkfelix (Feb 20 2020 at 14:31, on Zulip):

maybe there is a way to abstract it out

Santiago Pastorino (Feb 20 2020 at 14:31, on Zulip):

another maybe easier thing is to have the tool pointing out what between which shas is the problem so one can manually bisect over rustc repo

pnkfelix (Feb 20 2020 at 14:31, on Zulip):

i.e. user provides driver script to do the builds, and cargo-bisect-rustc handles the mechanical stuff.

pnkfelix (Feb 20 2020 at 14:31, on Zulip):

anyway

pnkfelix (Feb 20 2020 at 14:32, on Zulip):

back to triage

pnkfelix (Feb 20 2020 at 14:32, on Zulip):

this strikes me as very much P-high

pnkfelix (Feb 20 2020 at 14:32, on Zulip):

triage: P-high, removing nomination (non-existant), assigning to self.

pnkfelix (Feb 20 2020 at 14:33, on Zulip):

unpri β regr 2/2: "beta regression: $crate may not be imported" #69190

pnkfelix (Feb 20 2020 at 14:34, on Zulip):

@simulacrum do you know if the root crate (piston2d-gfx_graphics?) has an updated version that fixes the issue?

simulacrum (Feb 20 2020 at 14:35, on Zulip):

I do not, no

simulacrum (Feb 20 2020 at 14:35, on Zulip):

I have not had time to revisit since I did initial triage

pnkfelix (Feb 20 2020 at 14:36, on Zulip):

no problem, just figured I'd ask.

pnkfelix (Feb 20 2020 at 14:36, on Zulip):

I personally think ensuring that such an updated version exists is all we can expect to do here for now.

pnkfelix (Feb 20 2020 at 14:36, on Zulip):

/me really wants to get moving on rfc#2834

simulacrum (Feb 20 2020 at 14:37, on Zulip):

certainly looks like it

simulacrum (Feb 20 2020 at 14:37, on Zulip):

the crater logs have v0.27, and the most recent is .68

simulacrum (Feb 20 2020 at 14:37, on Zulip):

including a release 10 days ago

simulacrum (Feb 20 2020 at 14:37, on Zulip):

otoh, there may not be a .27.1 so to speak

simulacrum (Feb 20 2020 at 14:37, on Zulip):

it is ~4 years old at this point

pnkfelix (Feb 20 2020 at 14:39, on Zulip):

okay, triaged as P-high and assigned to self. (I strongly suspect I'll just look, confirm, and close the issue as expected breakage.)

pnkfelix (Feb 20 2020 at 14:39, on Zulip):

next: unprioritized nightly regressions

pnkfelix (Feb 20 2020 at 14:40, on Zulip):

there are two

pnkfelix (Feb 20 2020 at 14:40, on Zulip):

unpri ν regr 1/2: "Significant performance regression on the encoding benchmark" #69197

pnkfelix (Feb 20 2020 at 14:41, on Zulip):

hmm. Not sure if I can justify making this P-high...

pnkfelix (Feb 20 2020 at 14:41, on Zulip):

lets nominate for discussion at meeting.

pnkfelix (Feb 20 2020 at 14:42, on Zulip):

unpri ν regr 2/2: " is_x86_feature_detected!("avx512f") fails to build on nightly" #68905

pnkfelix (Feb 20 2020 at 14:43, on Zulip):

hmm this does not seem like a T-compiler Q

pnkfelix (Feb 20 2020 at 14:44, on Zulip):

oh indeed, it is already tagged as T-libs, not T-compiler

pnkfelix (Feb 20 2020 at 14:44, on Zulip):

I'll let the libs team make decisions about how to handle this.

pnkfelix (Feb 20 2020 at 14:44, on Zulip):

but I'll also nominate it to draw it to their attention

pnkfelix (Feb 20 2020 at 14:46, on Zulip):

normally at this point in pre-triage I would walk through the I-nominated issues to try to identify ones that do not need the tag any more.

pnkfelix (Feb 20 2020 at 14:46, on Zulip):

but given that the meeting is starting in 14 minutes, I don't have time. I'll focus on setting up the beta-nomination and stable-nomination lists, and then on trying to cherry-pick which nominations we might try to focus on at todays' meeting.

pnkfelix (Feb 20 2020 at 14:47, on Zulip):

there are 5 beta nominations

pnkfelix (Feb 20 2020 at 14:49, on Zulip):

oh you know, this process doesn't need to be spelled out in this topic. I've already created a hackmd for the meeting today ; I'll just put them in there now.

centril (Feb 20 2020 at 14:58, on Zulip):

@pnkfelix I'm pretty surprised https://github.com/rust-lang/rust/issues/69306 is still a thing

centril (Feb 20 2020 at 14:58, on Zulip):

I don't think T-Lang needs to discuss; it's an obvious bug (it shouldn't compile)

pnkfelix (Feb 20 2020 at 15:03, on Zulip):

okay we can probably just remove the nomination then. Or maybe I'll see if someone in the meeting thinks they can resolve it.

centril (Feb 20 2020 at 15:03, on Zulip):

I've cced @eddyb who made a related fix; I'll remove t-lang under "clearly bug"

centril (Feb 20 2020 at 15:04, on Zulip):

(but it is important that we fix it sooner rather than later due to forward compat)

pnkfelix (Feb 20 2020 at 15:07, on Zulip):

centril said:

(but it is important that we fix it sooner rather than later due to forward compat)

yes. it might be quite hard/impossible to warning cycle such a fix

centril (Feb 20 2020 at 15:08, on Zulip):

Maybe; let's see what crater says

Nell Shamrell-Harrington (Feb 20 2020 at 15:49, on Zulip):

Would love to!

Nell Shamrell-Harrington (Feb 20 2020 at 15:51, on Zulip):

pnkfelix said:

P-high, assigning to self. But also: Nell Shamrell-Harrington , would you be interesting in partnering up on this?

Would love to!

pnkfelix (Feb 20 2020 at 17:07, on Zulip):

pnkfelix said:

unpri β regr 2/2: "beta regression: $crate may not be imported" #69190

hey @centril , am I correct that generally the PR's that upgrade future-compat lints to hard errors are not tagged with relnotes ? I only did a quick sampling.

centril (Feb 20 2020 at 17:18, on Zulip):

@pnkfelix generally speaking that's correct

centril (Feb 20 2020 at 17:18, on Zulip):

subject to exceptions for e.g. NLL migrate mode and whatnot

pnkfelix (Feb 20 2020 at 17:26, on Zulip):

would it be overload to start including them? I wonder if there's some easy way to summarize them.

centril (Feb 20 2020 at 17:27, on Zulip):

@pnkfelix it's probably OK to include them under "compatibility section"; we don't have that many each time

centril (Feb 20 2020 at 17:27, on Zulip):

want to file a policy question for release team and nominate?

pnkfelix (Feb 20 2020 at 17:34, on Zulip):

yeah will do

Last update: Jun 04 2020 at 17:30UTC