Stream: t-compiler

Topic: #54818 weekly meeting 2018-12-06


pnkfelix (Dec 06 2018 at 09:43, on Zulip):

Hi @T-compiler ; our meeting isn't for several hours, but I figured I'd start doing the triage early today.

pnkfelix (Dec 06 2018 at 09:46, on Zulip):

first up: "problems with trait objects with a trait projection value that contains Self" #56288

pnkfelix (Dec 06 2018 at 09:46, on Zulip):

ariel has written up a detailed analysis

pnkfelix (Dec 06 2018 at 09:48, on Zulip):

and has nominated it for lang team discussion

pnkfelix (Dec 06 2018 at 09:48, on Zulip):

I'm certainly not going to take unilateral action there right now

pnkfelix (Dec 06 2018 at 09:49, on Zulip):

next: "prohibit "two-phase borrows" with existing borrows?" #56254

pnkfelix (Dec 06 2018 at 09:50, on Zulip):

i think the lang team summary from https://github.com/rust-lang/rust/issues/56254#issuecomment-442980457 still applies at least in terms of our short term plans for issue #56254

pnkfelix (Dec 06 2018 at 09:51, on Zulip):

the main point of progress is that @RalfJ has posted a response to joshtriplett's request for an explanation of why the model prohibits 2PB with existing borrow

pnkfelix (Dec 06 2018 at 09:52, on Zulip):

in any case we (and/or the NLL team) should figure out the actual plan here, in terms of what we're doing in the short term. I again will not take unilateral action here.

pnkfelix (Dec 06 2018 at 09:53, on Zulip):

next: "NLL error on closure, but not on equivalent function" #55526

pnkfelix (Dec 06 2018 at 09:53, on Zulip):

I looked into this and identified two distinct issues to address

pnkfelix (Dec 06 2018 at 09:55, on Zulip):

left a comment and am demoting #55526 itself to P-medium

pnkfelix (Dec 06 2018 at 09:56, on Zulip):

next: "regression: stack overflow on macosx with xcode 6.4" #55471

pnkfelix (Dec 06 2018 at 09:57, on Zulip):

@Oli has indicated that we have a fix but it won't be backported. Leaving this bug at P-high.

pnkfelix (Dec 06 2018 at 09:58, on Zulip):

Oddly we tried to work around the stack overflow in short term by bumping the stack size (in PR #56467) but it appears that did not work for gnzlbg...

oli (Dec 06 2018 at 09:59, on Zulip):

Yea, the issue is that the stack size I changed was only for threads, not for the main stack

pnkfelix (Dec 06 2018 at 09:59, on Zulip):

Ah, and we still have that other bug where we used to create a separate thread to run the compiler, but have stopped doing that.

oli (Dec 06 2018 at 10:00, on Zulip):

https://rust-lang.zulipchat.com/#narrow/stream/131828-t-compiler/subject/stack.20growing/near/150929391

pnkfelix (Dec 06 2018 at 11:21, on Zulip):

(i spawned off the topic "#48575 potential reversion / guard under -Z flag" for further discussion of that)

pnkfelix (Dec 06 2018 at 11:22, on Zulip):

moving right along

pnkfelix (Dec 06 2018 at 11:22, on Zulip):

next is "1.30.0 fails to build for target powerpc-unknown-netbsd" #55465

pnkfelix (Dec 06 2018 at 11:22, on Zulip):

I assigned this to myself but then got distracted by NLL stuff

pnkfelix (Dec 06 2018 at 11:23, on Zulip):

but since we still have a good amount of time before the meeting, I might be able to look into it before the meeting starts

pnkfelix (Dec 06 2018 at 11:23, on Zulip):

otherwise, no progress.

pnkfelix (Dec 06 2018 at 11:24, on Zulip):

next: "rustc 1.30.0's linker flavor inference is a non-backwards compat change to -Clinker" #55396

pnkfelix (Dec 06 2018 at 11:25, on Zulip):

I see that two follow-up PR's have been posted for this, PR #56349 (which has landed) and PR #56351 (which is being put through FCP among T-compiler)

pnkfelix (Dec 06 2018 at 11:25, on Zulip):

only check boxes left are for @nikomatsakis and @Vadim Petrochenkov

pnkfelix (Dec 06 2018 at 11:25, on Zulip):

anyway it seems like its chugging along nicely. Thank you very much @davidtwco and @nagisa !

pnkfelix (Dec 06 2018 at 11:26, on Zulip):

final P-high issue on the list is: "Breaking macro name resolution change in 1.29" #54783

pnkfelix (Dec 06 2018 at 11:26, on Zulip):

petrochenkov has tentatively tagged this as a wontfix

pnkfelix (Dec 06 2018 at 11:27, on Zulip):

So, we could close this immediately, or leave it open because we still don't have an answer as to why this is being reported as an error and not a warning. The latter seems like it could be a symptom of a broader potential problem?

pnkfelix (Dec 06 2018 at 11:28, on Zulip):

still, the P-high label and the assignment of petrochenkov all seem appropriate.

pnkfelix (Dec 06 2018 at 11:29, on Zulip):

@nagisa do you object if I go in and edit your comment on PR #56351 to use the american "-Clinker-flavor` rather than the british "-Clinker-flavour", at least to match my assumption/understanding that the flag itself uses the american english form?

pnkfelix (Dec 06 2018 at 11:30, on Zulip):

(probably would have been less annoyance overall if I had just silently done it) ... ===> and so I just did it.

pnkfelix (Dec 06 2018 at 11:35, on Zulip):

so, done with P-high issues

pnkfelix (Dec 06 2018 at 11:36, on Zulip):

there's nothing on the Release milestone link

pnkfelix (Dec 06 2018 at 11:36, on Zulip):

no beta-noms, no stable-noms.

pnkfelix (Dec 06 2018 at 11:37, on Zulip):

we already discussed the single stable-to-beta regression (#55471)

pnkfelix (Dec 06 2018 at 11:37, on Zulip):

so that brings us to the stable-to-nightly regressions

pnkfelix (Dec 06 2018 at 11:38, on Zulip):

first: "ICE: cannot relate region: LUB(ReErased, ReEmpty)" #56350

pnkfelix (Dec 06 2018 at 11:39, on Zulip):

left comment on #56350 saying P-high and assigned it to myself.

pnkfelix (Dec 06 2018 at 11:39, on Zulip):

next is: "problems with trait objects with a trait projection value that contains Self" #56288 which we already discussed above.

pnkfelix (Dec 06 2018 at 11:40, on Zulip):

(my take on that bug is that it, at the very least, is waiting for the lang team to weigh in. but compiler experts may well want to weigh in.)

pnkfelix (Dec 06 2018 at 11:40, on Zulip):

next is "rayon_1_0_0::sort perf regression" #56283

pnkfelix (Dec 06 2018 at 11:41, on Zulip):

I looked into this and confirmed that regression goes away if one turns jemalloc back on. I subsequently downgraded it to P-medium.

pnkfelix (Dec 06 2018 at 11:41, on Zulip):

(I also nominated it for discussion at the meeting, in terms of whether it should be closed as "not-a-bug/wont-fix")

pnkfelix (Dec 06 2018 at 11:42, on Zulip):

next: "Inference regression in nightly affecting Pin::set and futures" #56237

pnkfelix (Dec 06 2018 at 11:43, on Zulip):

this is now a stable-to-beta regression, at least according to play (using the minimized version from the issue)

nagisa (Dec 06 2018 at 11:44, on Zulip):

No, feel free.

nagisa (Dec 06 2018 at 11:44, on Zulip):

I see you did it already (linker-flavour -> linker-flavor)

pnkfelix (Dec 06 2018 at 11:45, on Zulip):

yeah after reflection it seemed silly to even ask ...

pnkfelix (Dec 06 2018 at 11:45, on Zulip):

if someone wants to volunteer to take point on resolving #56237, feel free to assign yourself.

Vadim Petrochenkov (Dec 06 2018 at 11:46, on Zulip):

PR #56351 (which is being put through FCP among T-compiler)

Hmm, how alive is rfcbot.rs? I regularly check it for FCPs, but I've never seen #56351 there.

pnkfelix (Dec 06 2018 at 11:46, on Zulip):

(it sounds like qnighy is willing to put in effort to try to fix this, under the assumption that it might have been injected by PR #56045 )

pnkfelix (Dec 06 2018 at 11:47, on Zulip):

@Vadim Petrochenkov I'm not sure how alive it is, but #56132 only went into RFC 15 days ago

pnkfelix (Dec 06 2018 at 11:47, on Zulip):

so if the discrepancy is due to rfcbot.rs dying, it happened sometime between 15 and 7 days ago

pnkfelix (Dec 06 2018 at 11:48, on Zulip):

last stable-to-nightly regression is "ICE with pyo3 on 1.31.0 nightly" #55380

pnkfelix (Dec 06 2018 at 11:49, on Zulip):

ariel points out that this is relying on an unstable feature and thus should not be tagged as stable-to-nightly regression.

pnkfelix (Dec 06 2018 at 11:50, on Zulip):

sounds like ariel has a fix posted in PR #56446

pnkfelix (Dec 06 2018 at 11:52, on Zulip):

Next section ... well, there's nothing tagged as Waiting for our team

pnkfelix (Dec 06 2018 at 11:52, on Zulip):

there are two nominated issues; the've already been mentioned above. We'll talk about them at the meeting itself.

pnkfelix (Dec 06 2018 at 11:54, on Zulip):

(well, now there's three nominated issues. You'll all see.)

pnkfelix (Dec 06 2018 at 11:56, on Zulip):

there are 45 open stable-to-stable regression bugs, but the newest is 26 days old.

pnkfelix (Dec 06 2018 at 11:57, on Zulip):

and they are all either already tagged with a P-label, or they explicitly belong to a different team.

pnkfelix (Dec 06 2018 at 11:58, on Zulip):

So, great. That's all the pre-meeting work I plan to do.

pnkfelix (Dec 06 2018 at 11:58, on Zulip):

Now I can jump back and see about looking at "1.30.0 fails to build for target powerpc-unknown-netbsd" #55465 before the meeting itself...

oli (Dec 06 2018 at 13:41, on Zulip):

I'm gonna be 15-20 min late for the meeting

pnkfelix (Dec 06 2018 at 14:04, on Zulip):

oh right, here's a ping to @T-compiler reminding everyone that the meeting will start in about 55 minutes.

oli (Dec 06 2018 at 15:01, on Zulip):

by some accident I managed to be on time.

oli (Dec 06 2018 at 15:01, on Zulip):

@T-compiler mtg?

pnkfelix (Dec 06 2018 at 15:02, on Zulip):

OMW

pnkfelix (Dec 06 2018 at 15:02, on Zulip):

okay so lets see

pnkfelix (Dec 06 2018 at 15:03, on Zulip):

I already did a pre-pass through the P-high issues above

pnkfelix (Dec 06 2018 at 15:03, on Zulip):

"ICE: cannot relate region: LUB(ReErased, ReEmpty)" #56350 is newly added to the list and assigned to me; I haven't looked at it yet. :)

pnkfelix (Dec 06 2018 at 15:04, on Zulip):

Oh, here's a link to the list I'm looking at: P-high issues

pnkfelix (Dec 06 2018 at 15:04, on Zulip):

"problems with trait objects with a trait projection value that contains Self" #56288 is waiting for the T-lang team to weigh in

pnkfelix (Dec 06 2018 at 15:05, on Zulip):

"prohibit "two-phase borrows" with existing borrows?" #56254 got weigh-in from the lang team. I think the next step ... is to decide whether we are going to move forward with prohibiting the 2PB's described, or if we're going to try to push @RalfJ to use a more general model.

pnkfelix (Dec 06 2018 at 15:06, on Zulip):

caveat: I have not yet actually read @RalfJ 's blog post nor the summary note.

pnkfelix (Dec 06 2018 at 15:06, on Zulip):

but in any case ... its an issue that ... is semi-important to figure out, but probably does not take precedence over any edition bugs that might arise in the very near term...?

pnkfelix (Dec 06 2018 at 15:07, on Zulip):

that's my current guess. that is, I'd leave it as P-high.

pnkfelix (Dec 06 2018 at 15:07, on Zulip):

@nikomatsakis do you want to keep it assigned to you, or should someone else take it? Given that its in a sort of fuzzy abstract design state, and you are probably the person who has the best overlap between NLL and unsafe-code-guidelines work ...

nikomatsakis (Dec 06 2018 at 15:07, on Zulip):

yep, I figured we'd circle back next week on that issue

pnkfelix (Dec 06 2018 at 15:08, on Zulip):

okay we'll come back to it next week, great

pnkfelix (Dec 06 2018 at 15:08, on Zulip):

next; "Inference regression in nightly affecting Pin::set and futures" #56237

oli (Dec 06 2018 at 15:08, on Zulip):

has a fix in https://github.com/rust-lang/rust/pull/56282

nikomatsakis (Dec 06 2018 at 15:08, on Zulip):

I think there is a pending PR I've failed to review

pnkfelix (Dec 06 2018 at 15:08, on Zulip):

oh great

nikomatsakis (Dec 06 2018 at 15:08, on Zulip):

I was just looking at it

pnkfelix (Dec 06 2018 at 15:08, on Zulip):

I'll assign this to niko so that he either r+'es the PR or comes up with a new fix.

pnkfelix (Dec 06 2018 at 15:10, on Zulip):

I already summarized state for "regression: stack overflow on macosx with xcode 6.4"#55471 up above and I don't think we need to discuss it further here

pnkfelix (Dec 06 2018 at 15:10, on Zulip):

(well it was either here or somewhere else, heh)

pnkfelix (Dec 06 2018 at 15:11, on Zulip):

I've posted a PR for "1.30.0 fails to build for target powerpc-unknown-netbsd"#55465

pnkfelix (Dec 06 2018 at 15:11, on Zulip):

but if @eddyb is around, I'd like them to weigh in on the Question I posed in a comment at end of #55465

pnkfelix (Dec 06 2018 at 15:11, on Zulip):

next: "rustc 1.30.0's linker flavor inference is a non-backwards compat change to -Clinker" #55396, progress seems to be going well in @davidtwco and @nagisa 's hands.

pnkfelix (Dec 06 2018 at 15:12, on Zulip):

but oh there was one thing

pnkfelix (Dec 06 2018 at 15:13, on Zulip):

@Vadim Petrochenkov asked why rfcbot didn't have PR #56351 listed for them

pnkfelix (Dec 06 2018 at 15:13, on Zulip):

I guess that's an infrastructure issue and not something for us to deal with here

pnkfelix (Dec 06 2018 at 15:13, on Zulip):

but I just wanted to basically warn people that if you're waiting on checkboxes, you probably shouldn't be afraid to ping people about it

davidtwco (Dec 06 2018 at 15:14, on Zulip):

(rfcbot.rs didn't list it at all, not just for @Vadim Petrochenkov)

pnkfelix (Dec 06 2018 at 15:14, on Zulip):

and likewise, people should not get offended about getting pings about check boxes.

pnkfelix (Dec 06 2018 at 15:14, on Zulip):

because our tools don't always measure up to the needs of our fellow humans

pnkfelix (Dec 06 2018 at 15:14, on Zulip):

okay that's enough preaching

pnkfelix (Dec 06 2018 at 15:15, on Zulip):

next: "ICE with pyo3 on 1.31.0 nightly" #55380 has a fix from ariel

pnkfelix (Dec 06 2018 at 15:15, on Zulip):

and finally "Breaking macro name resolution change in 1.29" #54783 ... @Vadim Petrochenkov argues effectively that we should just say "wontfix" for this

pnkfelix (Dec 06 2018 at 15:16, on Zulip):

since the basic issue is that somehow we didn't properly implement the warning cycle here

pnkfelix (Dec 06 2018 at 15:16, on Zulip):

but at this point the error is out there

Pietro Albini (Dec 06 2018 at 15:16, on Zulip):

next: "ICE with pyo3 on 1.31.0 nightly" #55380 has a fix from ariel

this just got merged btw

pnkfelix (Dec 06 2018 at 15:16, on Zulip):

and going from error -> warning-cycle -> error

pnkfelix (Dec 06 2018 at 15:16, on Zulip):

that path seems not-worth-it

pnkfelix (Dec 06 2018 at 15:16, on Zulip):

the only misgiving I have about just closing #54783 as a wontfix

pnkfelix (Dec 06 2018 at 15:17, on Zulip):

is that I, like @Vadim Petrochenkov , am a little concerned that no one has identified why the warning cycle didn't work? @Vadim Petrochenkov said they had implemented one.

nikomatsakis (Dec 06 2018 at 15:17, on Zulip):

do we have some idea of how widespread the impact is here?

nikomatsakis (Dec 06 2018 at 15:17, on Zulip):

that was a bit unclear to me

pnkfelix (Dec 06 2018 at 15:19, on Zulip):

dunno. @Vadim Petrochenkov says its something unexpected being done by #[derive(DbEnum)]

nikomatsakis (Dec 06 2018 at 15:19, on Zulip):

yes ok

pnkfelix (Dec 06 2018 at 15:19, on Zulip):

so I guess anyone using that, or any other proc_macro that does something similarly unexpected, could be affected.

nikomatsakis (Dec 06 2018 at 15:20, on Zulip):

that seemed like a bit of a guess

pnkfelix (Dec 06 2018 at 15:20, on Zulip):

but at least the error message you get seems like enough to get you over the hurdle?

nikomatsakis (Dec 06 2018 at 15:20, on Zulip):

but yeah without knowing what DbEnum is doing that is unexpected...

eddyb (Dec 06 2018 at 15:20, on Zulip):

@pnkfelix I already replied to <https://github.com/rust-lang/rust/issues/55465#issuecomment-444878134>

pnkfelix (Dec 06 2018 at 15:20, on Zulip):

as in, its a message that was meant to be a warning alone, no error. but I'm assuming if you address the warning, then the error will go away...

pnkfelix (Dec 06 2018 at 15:20, on Zulip):

Oh okay thanks @eddyb , soorry I didn't check there before pinging you here.

pnkfelix (Dec 06 2018 at 15:21, on Zulip):

@eddyb (this means I actually have no idea what your ideal solution to #55465 actually is.)

pnkfelix (Dec 06 2018 at 15:21, on Zulip):

(cut-and-paste the code? Use crates.io crates and only that?)

eddyb (Dec 06 2018 at 15:21, on Zulip):

the latter

eddyb (Dec 06 2018 at 15:22, on Zulip):

@Alex Crichton already wants to go in that direction anyway

eddyb (Dec 06 2018 at 15:22, on Zulip):

I just don't know the issue that tracks that

pnkfelix (Dec 06 2018 at 15:22, on Zulip):

Okay, that seems then orthogonal to this issue then?

pnkfelix (Dec 06 2018 at 15:22, on Zulip):

since the problem that arose here was with code that is using a crates.io crate?

eddyb (Dec 06 2018 at 15:22, on Zulip):

idk I thought someone figured out that the submodule was in between versions and missing the fix

pnkfelix (Dec 06 2018 at 15:23, on Zulip):

okay maybe i'll follow up with you about this after the meeting.

eddyb (Dec 06 2018 at 15:23, on Zulip):

the context is probably in the messages I was replying to

pnkfelix (Dec 06 2018 at 15:23, on Zulip):

anyway

pnkfelix (Dec 06 2018 at 15:23, on Zulip):

I'm going to close #54783 as "wontfix"

pnkfelix (Dec 06 2018 at 15:24, on Zulip):

if anyone feels inspired by the conversation to try to find out why the warning-cycle code was subverted by DbEnum

pnkfelix (Dec 06 2018 at 15:24, on Zulip):

then by all means, channel that inspiration.

pnkfelix (Dec 06 2018 at 15:26, on Zulip):

okay that's the P-highs.

pnkfelix (Dec 06 2018 at 15:26, on Zulip):

there's nothing on the Release milestone (at least not in the link's current form)

pnkfelix (Dec 06 2018 at 15:26, on Zulip):

there's one beta-nominated PR

pnkfelix (Dec 06 2018 at 15:27, on Zulip):

"Fix #56237: normalize type before deferred sizedness checking." #56282

oli (Dec 06 2018 at 15:27, on Zulip):

that's the fix for the Pin issue

pnkfelix (Dec 06 2018 at 15:27, on Zulip):

yeah

pnkfelix (Dec 06 2018 at 15:28, on Zulip):

i"m just skimming it now

pnkfelix (Dec 06 2018 at 15:28, on Zulip):

to try to gauge risk

nagisa (Dec 06 2018 at 15:28, on Zulip):

since the beta was cut…

nagisa (Dec 06 2018 at 15:28, on Zulip):

I mean beta was uplifted to stable

nagisa (Dec 06 2018 at 15:28, on Zulip):

I’m not sure nightly was uplifted to beta

pnkfelix (Dec 06 2018 at 15:28, on Zulip):

ah

oli (Dec 06 2018 at 15:28, on Zulip):

I nominated after the beta cut

pnkfelix (Dec 06 2018 at 15:29, on Zulip):

so maybe its moot whether we attempt a backport

nagisa (Dec 06 2018 at 15:29, on Zulip):

so there might not be a need to backport anything at all?

pnkfelix (Dec 06 2018 at 15:29, on Zulip):

if it lands in master before nightly uplift to beta

pnkfelix (Dec 06 2018 at 15:29, on Zulip):

okay

nagisa (Dec 06 2018 at 15:29, on Zulip):

@Pietro Albini has nightly to beta uplift happened yet?

oli (Dec 06 2018 at 15:29, on Zulip):

oh, that hasn't happened yet?

pnkfelix (Dec 06 2018 at 15:29, on Zulip):

lets ... leave it nominated

pnkfelix (Dec 06 2018 at 15:29, on Zulip):

and we can discuss next week

oli (Dec 06 2018 at 15:29, on Zulip):

why can we break submodules again then?

Pietro Albini (Dec 06 2018 at 15:29, on Zulip):

yep, it happened yesterday

pnkfelix (Dec 06 2018 at 15:29, on Zulip):

that is, I don't see a need to rush a decision here

Pietro Albini (Dec 06 2018 at 15:30, on Zulip):

so any PR needs to be backported

pnkfelix (Dec 06 2018 at 15:30, on Zulip):

theres no stable-nominated issues.

pnkfelix (Dec 06 2018 at 15:30, on Zulip):

we've already discussed the two issues on the regression-from-stable-to-beta list

pnkfelix (Dec 06 2018 at 15:31, on Zulip):

for regression-from-stable-to-nightly, there are three issues, two of which we've discussed.

pnkfelix (Dec 06 2018 at 15:31, on Zulip):

the third is "rayon_1_0_0::sort perf regression" #56283

pnkfelix (Dec 06 2018 at 15:31, on Zulip):

which I nominated because I want to close as wontfix

pnkfelix (Dec 06 2018 at 15:32, on Zulip):

and now that I've stated that in front of all of you

pnkfelix (Dec 06 2018 at 15:32, on Zulip):

I'm going to just do it

nikomatsakis (Dec 06 2018 at 15:32, on Zulip):

I think closing is good

pnkfelix (Dec 06 2018 at 15:32, on Zulip):

and let you all voice your objections on ticket. :)

pnkfelix (Dec 06 2018 at 15:33, on Zulip):

there's no PR's waiting on our team

pnkfelix (Dec 06 2018 at 15:34, on Zulip):

there are two I-nominated issues, but the first (#56288) we already touched on, and is nominated for the lang team.

pnkfelix (Dec 06 2018 at 15:34, on Zulip):

(of course anyone who wants to discuss should feel free to do so, especially on the issue comment thread where the lang team members will see since they dont hang out on zulip)

pnkfelix (Dec 06 2018 at 15:35, on Zulip):

the second issue is one I tagged as T-compiler in order to force niko to do something with its nomination tag or choose a better T-tag

pnkfelix (Dec 06 2018 at 15:35, on Zulip):

namely "edition lint: migrating extern crate with #[macro_use]" #52043

pnkfelix (Dec 06 2018 at 15:35, on Zulip):

is there a good team to figure out these matters? Are we that team?

pnkfelix (Dec 06 2018 at 15:36, on Zulip):

for context, nomination was from this comment

nikomatsakis (Dec 06 2018 at 15:36, on Zulip):

er

nikomatsakis (Dec 06 2018 at 15:36, on Zulip):

I think we can remove nomination

nikomatsakis (Dec 06 2018 at 15:36, on Zulip):

I think T-compiler is probably a fine team for this though

pnkfelix (Dec 06 2018 at 15:36, on Zulip):

I can't tell what the status is here. We have a lint (landed in #52275) but as far as I can tell ([playground]) it does not fire. In fact, I get no warnings at all when using extern crate in Rust 2018, even with idiom lints enabled. Is that expected?

(We do appear to have stabilized use_extern_macros in https://github.com/rust-lang/rust/pull/50911.)

[playground]: https://play.rust-lang.org/?gist=8c58ddde2528c4a3767e449ef58b998b&version=nightly&mode=debug&edition=2018

pnkfelix (Dec 06 2018 at 15:36, on Zulip):

aw that almost worked.

nikomatsakis (Dec 06 2018 at 15:37, on Zulip):

idiom lints are basically "deferred"

nikomatsakis (Dec 06 2018 at 15:37, on Zulip):

I imagine the right thing is to have a working group focused on them

pnkfelix (Dec 06 2018 at 15:37, on Zulip):

well, it doesn't have a P-label nor an assignment

nikomatsakis (Dec 06 2018 at 15:37, on Zulip):

but who would lead and populate it, I have no idea

pnkfelix (Dec 06 2018 at 15:37, on Zulip):

so its likely to get lost if we denominate it.

nikomatsakis (Dec 06 2018 at 15:37, on Zulip):

I think the A- tag

nikomatsakis (Dec 06 2018 at 15:37, on Zulip):

is how we would find it

pnkfelix (Dec 06 2018 at 15:37, on Zulip):

you figure we or some WG will come back to A-edition-2018-lints eventually?

nikomatsakis (Dec 06 2018 at 15:37, on Zulip):

maybe we want a "idiom lint" thing

pnkfelix (Dec 06 2018 at 15:37, on Zulip):

okay.[

pnkfelix (Dec 06 2018 at 15:38, on Zulip):

I'll remove nomination tag then

nikomatsakis (Dec 06 2018 at 15:38, on Zulip):

yes, that's what I'm assuming

pnkfelix (Dec 06 2018 at 15:40, on Zulip):

I'll leave it as T-compiler until a WG is formed, if that ever happens.

pnkfelix (Dec 06 2018 at 15:40, on Zulip):

oaky so ...

pnkfelix (Dec 06 2018 at 15:40, on Zulip):

that's basically all of the rote stuff

pnkfelix (Dec 06 2018 at 15:40, on Zulip):

(I already discussed the stable-to-stable regressions during my earlier prepass)

pnkfelix (Dec 06 2018 at 15:41, on Zulip):

Are there any topics people want to spend ~19 minutes on?

pnkfelix (Dec 06 2018 at 15:41, on Zulip):

or shall we adjourn?

nikomatsakis (Dec 06 2018 at 15:41, on Zulip):

I want to mention that

Zoxc (Dec 06 2018 at 15:41, on Zulip):

I'd like make the compiler be able to compile multiple independent crates, so we can make progress on this front without making everything on-demand.

nikomatsakis (Dec 06 2018 at 15:41, on Zulip):

the steering meeting for tomorrow was postponed 1 week

nikomatsakis (Dec 06 2018 at 15:41, on Zulip):

I'd like make the compiler be able to compile multiple independent crates, so we can make progress on this front without making everything on-demand.

that's on the agenda of possible improvements for the next year

nikomatsakis (Dec 06 2018 at 15:42, on Zulip):

it'd be good to map out what steps make sense

nikomatsakis (Dec 06 2018 at 15:42, on Zulip):

Actually, along those lines, @eddyb are you here?

nagisa (Dec 06 2018 at 15:42, on Zulip):

We should discuss that main thread PR revertion thing

nikomatsakis (Dec 06 2018 at 15:42, on Zulip):

I was wondering about trying to organize a meeting next week to try and discuss that a bit synchronously (it was one of the work items that came up at least steering meeting)

nagisa (Dec 06 2018 at 15:42, on Zulip):

at least get a general consensus that, yes, we want to do it.

nikomatsakis (Dec 06 2018 at 15:42, on Zulip):

can you summarize what's up there?

nikomatsakis (Dec 06 2018 at 15:43, on Zulip):

I wasn't following but saw there were a bunch of comments

eddyb (Dec 06 2018 at 15:43, on Zulip):

I don't have a lot of opinions on where to start

nagisa (Dec 06 2018 at 15:43, on Zulip):

Basically there was a PR at some point in the past that made the compiler utilize the main thread (details are fuzzy to me at this moment). I strongly suspect it breaks illumos and now it appears that it causes problems with stack sizes.

pnkfelix (Dec 06 2018 at 15:44, on Zulip):

note: "up there" == #48575 potential reversion / guard under -Z flag

nikomatsakis (Dec 06 2018 at 15:44, on Zulip):

(breaks illumos how?)

nagisa (Dec 06 2018 at 15:44, on Zulip):

(as in, we are not in control of stack sizes for the main thread, that’s controlled by the OS)

nikomatsakis (Dec 06 2018 at 15:44, on Zulip):

seems clear though that if it is breaking things we should prob revert at least for now

Zoxc (Dec 06 2018 at 15:44, on Zulip):

I'm all for reverting that PR btw. It's probably broken on Windows too

nikomatsakis (Dec 06 2018 at 15:44, on Zulip):

ah, ok

nikomatsakis (Dec 06 2018 at 15:44, on Zulip):

what was the motivation for doing it?

nagisa (Dec 06 2018 at 15:44, on Zulip):

Illumos is https://github.com/rust-lang/rust/issues/52577

eddyb (Dec 06 2018 at 15:44, on Zulip):

independent crates in the same process but separate threads seems ok

eddyb (Dec 06 2018 at 15:44, on Zulip):

that PR should just be reverted

eddyb (Dec 06 2018 at 15:44, on Zulip):

that PR should just be reverted

pnkfelix (Dec 06 2018 at 15:44, on Zulip):

motivation was a developer wanted bt to work in the debugger

pnkfelix (Dec 06 2018 at 15:44, on Zulip):

without having to look at backtrace for all threads

pnkfelix (Dec 06 2018 at 15:45, on Zulip):

just the main one

eddyb (Dec 06 2018 at 15:45, on Zulip):

ths motivation is flimsy

nikomatsakis (Dec 06 2018 at 15:45, on Zulip):

ok, that's pretty weak

eddyb (Dec 06 2018 at 15:45, on Zulip):

I've never used rustc with a debugger

oli (Dec 06 2018 at 15:45, on Zulip):

isn't that gonna fail with parallel rustc anyway?

pnkfelix (Dec 06 2018 at 15:45, on Zulip):

I have

oli (Dec 06 2018 at 15:45, on Zulip):

me2

eddyb (Dec 06 2018 at 15:45, on Zulip):

except.when LLVM crashed the entire process

pnkfelix (Dec 06 2018 at 15:45, on Zulip):

its very fun with rr

nikomatsakis (Dec 06 2018 at 15:45, on Zulip):

I use the debugger, but I don't think it should be the determining factor for this sort of decision =)

davidtwco (Dec 06 2018 at 15:45, on Zulip):

its very fun with rr

Oh how I wish rr worked on Ryzen.

Zoxc (Dec 06 2018 at 15:46, on Zulip):

I want to start by passing a list of crates to rustc. Then running each pass on all of them in parallel. So parsing on all, type check on all, etc.

eddyb (Dec 06 2018 at 15:46, on Zulip):

it's much less useful than our backtraces and debug logging

nagisa (Dec 06 2018 at 15:46, on Zulip):

I used rr on rustc as well, doing stuff in main thread helps in no way.

eddyb (Dec 06 2018 at 15:46, on Zulip):

rr is.sometjing I haven't considered

pnkfelix (Dec 06 2018 at 15:46, on Zulip):

So my Q is: Why did this person find it useful?

eddyb (Dec 06 2018 at 15:46, on Zulip):

also Zulip is so broken on mobile wow

pnkfelix (Dec 06 2018 at 15:46, on Zulip):

they said something about signal handlers

eddyb (Dec 06 2018 at 15:47, on Zulip):

anyway let's just revert

nagisa (Dec 06 2018 at 15:47, on Zulip):

Note that this causing problems with illumos is just my educated guess, I haven’t actually verified

nagisa (Dec 06 2018 at 15:48, on Zulip):

(namely this PR is the only PR to change relevant files in the whole regression range)

nikomatsakis (Dec 06 2018 at 15:48, on Zulip):

seems like we should revert

eddyb (Dec 06 2018 at 15:48, on Zulip):

maybe we should name our threads? if we're not doing it already

pnkfelix (Dec 06 2018 at 15:48, on Zulip):

Our options as I see them are: 1. do nothing (which I think none of us want, at least in long term), 2. revert PR #48575, or 3. add a -Z flag that causes rustc to run solely on main thread.

pnkfelix (Dec 06 2018 at 15:48, on Zulip):

is (3) that bad of a maintenance burden for us?

eddyb (Dec 06 2018 at 15:49, on Zulip):

but that's my only concession to that usecase

nagisa (Dec 06 2018 at 15:49, on Zulip):

@eddyb I believe we do, and our threads are called rustc as opposed to main.

nagisa (Dec 06 2018 at 15:49, on Zulip):

unless something changed at some point

eddyb (Dec 06 2018 at 15:49, on Zulip):

then I find the motivation almost pointless

eddyb (Dec 06 2018 at 15:49, on Zulip):

we need several threads anyway

oli (Dec 06 2018 at 15:49, on Zulip):

3) is useless if we go for parallel rustc

pnkfelix (Dec 06 2018 at 15:50, on Zulip):

my assumption is that the -Z flag would disable parallel rustc

eddyb (Dec 06 2018 at 15:50, on Zulip):

we already have worker threads don't we

eddyb (Dec 06 2018 at 15:50, on Zulip):

for incremental

eddyb (Dec 06 2018 at 15:50, on Zulip):

and profiling

eddyb (Dec 06 2018 at 15:50, on Zulip):

I don't see how main thread helps unless you have exactly 1 thread

nagisa (Dec 06 2018 at 15:51, on Zulip):

Somewhat a tangent, but parallel-rustc is awesome and we should make it work properly (tools do not work with that atm and fail the build)

nikomatsakis (Dec 06 2018 at 15:51, on Zulip):

@pnkfelix wait, why do none of us want "do nothing"?

pnkfelix (Dec 06 2018 at 15:51, on Zulip):

@pnkfelix wait, why do none of us want "do nothing"?

okay maybe that was presumptuous

nikomatsakis (Dec 06 2018 at 15:51, on Zulip):

@nagisa yes @mw and I have been investigating how well it works and thinking about next steps last day or two, totally agreed

pnkfelix (Dec 06 2018 at 15:51, on Zulip):

1. status quo is presumed to have broken some targets

nikomatsakis (Dec 06 2018 at 15:51, on Zulip):

oh

nikomatsakis (Dec 06 2018 at 15:51, on Zulip):

I thought you meant relative to "before the PR landed"

pnkfelix (Dec 06 2018 at 15:52, on Zulip):

2. we cannot ensure soundness

nikomatsakis (Dec 06 2018 at 15:52, on Zulip):

I definitely don't want to leave in the current state

pnkfelix (Dec 06 2018 at 15:52, on Zulip):

no, I meant "do nothing" relative to current state

pnkfelix (Dec 06 2018 at 15:52, on Zulip):

sorry that was not clear.

nikomatsakis (Dec 06 2018 at 15:52, on Zulip):

a -Z flag seems...ok, but it's one more annoying knob. I think we should not do it personally

Zoxc (Dec 06 2018 at 15:52, on Zulip):

Parallel rustc needs some CI support also

pnkfelix (Dec 06 2018 at 15:53, on Zulip):

I would like for us to at least try to engage with ishitatsuyuki to try to better understand their motivation

pnkfelix (Dec 06 2018 at 15:53, on Zulip):

before blindly reverting

pnkfelix (Dec 06 2018 at 15:53, on Zulip):

it seems obvious to me that they were highly motivated to do this

nagisa (Dec 06 2018 at 15:54, on Zulip):

They’re on IRC, I can try probing

nagisa (Dec 06 2018 at 15:54, on Zulip):

/me pinged them on #rust-internals

nikomatsakis (Dec 06 2018 at 15:54, on Zulip):

I believe @Tatsuyuki Ishi is also on Zulip

pnkfelix (Dec 06 2018 at 15:54, on Zulip):

which means either they are overlooking some change they could make to their own workflow, or we are overlooking some concern they have

Zoxc (Dec 06 2018 at 15:54, on Zulip):

The PR was just a convenience thing for gdb debugging though

nikomatsakis (Dec 06 2018 at 15:54, on Zulip):

though maybe not actively monitoring

pnkfelix (Dec 06 2018 at 15:55, on Zulip):

The PR was just a convenience thing for gdb debugging though

okay? I don't think we should discount that, depending on how bad the debugging experience was for their workflow

nagisa (Dec 06 2018 at 15:55, on Zulip):

… its past midnight in Japan anyway, so unlikely we’ll hear back anytime soon.

pnkfelix (Dec 06 2018 at 15:55, on Zulip):

I don't want to get into a pattern of saying "so few of us use debuggers to debug rustc, we don't need to worry about breaking such usage."

pnkfelix (Dec 06 2018 at 15:56, on Zulip):

but I readily admit there is a huge difference between breaking use of gdb

Zoxc (Dec 06 2018 at 15:56, on Zulip):

You'd have to select the correct thread manually. Probably is something gdb should fix

pnkfelix (Dec 06 2018 at 15:56, on Zulip):

vs making it slightly less convenient to get the backtrace of interest

nikomatsakis (Dec 06 2018 at 15:56, on Zulip):

I don't think "breaking" is a good summary

nikomatsakis (Dec 06 2018 at 15:56, on Zulip):

I feel like this is a gdb bug :P

nikomatsakis (Dec 06 2018 at 15:57, on Zulip):

I like the idea of naming threads though

pnkfelix (Dec 06 2018 at 15:57, on Zulip):

the fact that ishitatsuyuki said "when interupting with signals" leads me to assume that they are using Ctrl-C

nikomatsakis (Dec 06 2018 at 15:59, on Zulip):

plausible

nikomatsakis (Dec 06 2018 at 15:59, on Zulip):

I think I don't really care :) most of all, I just want to unbreak stuff

nikomatsakis (Dec 06 2018 at 16:00, on Zulip):

I am amenable to "revert and consider an alternative approach to making gdb more convenient", though I feel like the need to find your thread is just "part of using gdb" for me

nikomatsakis (Dec 06 2018 at 16:00, on Zulip):

maybe i'm just too accustomed to how annoying it is :)

pnkfelix (Dec 06 2018 at 16:02, on Zulip):

okay well I'm glad we talked about this

pnkfelix (Dec 06 2018 at 16:03, on Zulip):

(I won't veto a PR that reverts PR #48575 ; I just wanted to state that I would prefer to get more info first.)

pnkfelix (Dec 06 2018 at 16:03, on Zulip):

having said that, I gotta take off now!

Zoxc (Dec 06 2018 at 16:03, on Zulip):

Anyone object to replacing tcx.hir with tcx.hir()? I want to land https://github.com/rust-lang/rust/pull/56502 before it rots =P

nikomatsakis (Dec 06 2018 at 16:04, on Zulip):

@Zoxc seems good

Zoxc (Dec 06 2018 at 16:06, on Zulip):

Or should I say, rots more =P

Tatsuyuki Ishi (Dec 06 2018 at 22:29, on Zulip):

I'm completely fine with reverting that PR, as getting stack guard correct is almost impossible

Tatsuyuki Ishi (Dec 06 2018 at 22:31, on Zulip):

And yeah, I prefer Ctrl+C plus bt to investigate performance issues, so it's a little more convenient if it's kept as a flag

Vadim Petrochenkov (Dec 06 2018 at 22:41, on Zulip):

I also use gdb because backtraces from libbacktrace used by RUST_BACKTRACE don't work on mingw, but I haven't noticed any improvements to debugging experience from using the main thread.

Last update: Nov 16 2019 at 01:05UTC