Stream: t-compiler/meetings

Topic: [weekly meeting] 2020-10-22 #54818


Santiago Pastorino (Oct 22 2020 at 00:18, on Zulip):

Hi @T-compiler/meeting; the triage meeting will happen tomorrow at

Santiago Pastorino (Oct 22 2020 at 00:18, on Zulip):

WG-prioritization has done pre-triage in #t-compiler/wg-prioritization/alerts

Santiago Pastorino (Oct 22 2020 at 00:19, on Zulip):

@WG-prioritization has prepared the meeting agenda

Santiago Pastorino (Oct 22 2020 at 00:19, on Zulip):

We will have checkins from WG-rustc-dev-guide and WG-llvm

Santiago Pastorino (Oct 22 2020 at 00:20, on Zulip):

@Joshua Nelson have already filled @WG-rustc-dev-guide checkin, I'll copy that one to the agenda tomorrow

Joshua Nelson (Oct 22 2020 at 00:20, on Zulip):

/me is still filling out a few things

Santiago Pastorino (Oct 22 2020 at 00:20, on Zulip):

@nagisa do you have something you want to share about @WG-llvm?

Santiago Pastorino (Oct 22 2020 at 00:21, on Zulip):

Santiago Pastorino said:

nagisa do you have something you want to share about @WG-llvm?

if so, please feel free to fill the checkin directly in the agenda

Nikita Popov (Oct 22 2020 at 07:33, on Zulip):

There's a few things worth mentioning on the LLVM side:

Nikita Popov (Oct 22 2020 at 07:36, on Zulip):

The first parts of the mustprogress attribute have landed (https://reviews.llvm.org/D86233, https://reviews.llvm.org/D85393). This means that forward progress is now opt-in. Once this work is complete, this will resolve current bugs around infinite loops being optimized away. There isn't a patch for that yet, but I dare say the hardest part (convincing people that this is the right thing to do) is done now...

Nikita Popov (Oct 22 2020 at 07:41, on Zulip):

MemorySSA-based DSE has been enabled by default (https://reviews.llvm.org/D87163), which means dead stores are eliminated across based blocks. Expect some compile-time regression... More importantly for rust, I've ported MemCpyOpt to MSSA (https://reviews.llvm.org/D89207), which will allow memcpy's to be eliminated across blocks as well, a long-standing optimization problem. It's not the first attempt to do this, but I'm optimistic :)

Nikita Popov (Oct 22 2020 at 07:45, on Zulip):

Some older changes that I probably have not mentioned before. I've finally managed to make correlated value propagation optimize conditions based on local value ranges (https://reviews.llvm.org/D69686), which will resolve various missing bounds check optimizations where constant operands are involved. There is also a new constraint elimination pass (https://reviews.llvm.org/D84547) which is able to reason about complex implications symbolically. However, it is not enabled by default, and there's no roadmap to enable it at present.

Nikita Popov (Oct 22 2020 at 07:51, on Zulip):

Also been a while ago, but someone took over my work for introducing saturating float to int casts (https://reviews.llvm.org/D54749). There's been some slow progress, and maybe there's a chance that this will land before the heat death of the universe... Rust has worked around this in the meantime, but once/if it lands we can improve codegen.

Nikita Popov (Oct 22 2020 at 07:54, on Zulip):

That's all I can think of for now.

apiraino (Oct 22 2020 at 10:18, on Zulip):

thanks - added to the agenda

Santiago Pastorino (Oct 22 2020 at 13:01, on Zulip):

Hi @T-compiler/meeting, triage meeting will be starting in ~ 1 hour

Santiago Pastorino (Oct 22 2020 at 13:01, on Zulip):

Check out the meeting agenda

simulacrum (Oct 22 2020 at 13:05, on Zulip):

Zulip seems to have fixed the timezone viewer to work on mobile!

pnkfelix (Oct 22 2020 at 14:00, on Zulip):

Hi @T-compiler/meeting! Add a :wave: emoji to show you're here :)

pnkfelix (Oct 22 2020 at 14:00, on Zulip):

we will start off with 5 minutes for ...

Announcements

pnkfelix (Oct 22 2020 at 14:00, on Zulip):
pnkfelix (Oct 22 2020 at 14:01, on Zulip):

I mention these for two reasons: 1. Some of you may be interested in applying to such roles (and I suspect all of our regular contributors are strong candidates)

pnkfelix (Oct 22 2020 at 14:02, on Zulip):

but also 2. This is a sign that we're really hitting the big time. Regardless of your opinion of any of these big companies, this is good news for Rust.

pnkfelix (Oct 22 2020 at 14:02, on Zulip):
pnkfelix (Oct 22 2020 at 14:03, on Zulip):
pnkfelix (Oct 22 2020 at 14:03, on Zulip):
pnkfelix (Oct 22 2020 at 14:03, on Zulip):
pnkfelix (Oct 22 2020 at 14:03, on Zulip):
pnkfelix (Oct 22 2020 at 14:03, on Zulip):
Santiago Pastorino (Oct 22 2020 at 14:03, on Zulip):

note that compiler-team#359 has been in FCP for 21 days but people raised concerns

Santiago Pastorino (Oct 22 2020 at 14:04, on Zulip):

so unsure how we should proceed with that one

pnkfelix (Oct 22 2020 at 14:04, on Zulip):

hmm

Santiago Pastorino (Oct 22 2020 at 14:04, on Zulip):

maybe we should just removed the FCP label and have people seconding it again?

pnkfelix (Oct 22 2020 at 14:04, on Zulip):

@nikomatsakis ^

nikomatsakis (Oct 22 2020 at 14:05, on Zulip):

I plan to close it

nikomatsakis (Oct 22 2020 at 14:05, on Zulip):

We discussed this in the traits meeting, we have some different ideas

nikomatsakis (Oct 22 2020 at 14:05, on Zulip):

I'll just go do that now

pnkfelix (Oct 22 2020 at 14:05, on Zulip):
pnkfelix (Oct 22 2020 at 14:05, on Zulip):

WG checkins

pnkfelix (Oct 22 2020 at 14:06, on Zulip):

So, before we start with the actual WG checkin's

pnkfelix (Oct 22 2020 at 14:06, on Zulip):

I wanted to make a little note: I have noticed, at least for me (as a reader) that some of the check-in text each week can be pretty overwhelming

pnkfelix (Oct 22 2020 at 14:07, on Zulip):

And this week's check-in's, which I am about to post, led me and @Santiago Pastorino to thinking that when preparing the agenda, we should strive to condense the info a bit

Joshua Nelson (Oct 22 2020 at 14:08, on Zulip):

I might have gone a little overboard, sorry

pnkfelix (Oct 22 2020 at 14:08, on Zulip):

At the same time, I definitely do not 1. want to undermine the work that each WG does, nor 2. make life more difficult for the people preparing the agenda

pnkfelix (Oct 22 2020 at 14:08, on Zulip):

so my plan is this: If I or others decide to condense any check-in text, we will first copy the existing text to a fresh separate hackmd

pnkfelix (Oct 22 2020 at 14:08, on Zulip):

which we will link

pnkfelix (Oct 22 2020 at 14:08, on Zulip):

we won't necessarily inform the check-in authors of plans to do this, because I want freedom to do so at last minute. :wink:

pnkfelix (Oct 22 2020 at 14:09, on Zulip):

Joshua Nelson said:

I might have gone a little overboard, sorry

don't be sorry; Its hard to know what to post!

pnkfelix (Oct 22 2020 at 14:09, on Zulip):

and Some of the stuff on here I was not even aware existed, like the new way to not build LLVM

pnkfelix (Oct 22 2020 at 14:09, on Zulip):

anwyay

pnkfelix (Oct 22 2020 at 14:10, on Zulip):

On with the show. (This system of condensing the text was not applied this week. We'll do it for future meetings.)

pnkfelix (Oct 22 2020 at 14:10, on Zulip):

@WG-rustc-dev-guide checkin by @Santiago Pastorino and @Joshua Nelson:

pnkfelix (Oct 22 2020 at 14:10, on Zulip):

Authors

@1c3t3a, @arora-aman, @bugadani, @caass, @camelid, @ecstatic-morse, @follower, @guswynn, @JakobDegen, @JOE1994, @JohnTitor, @Julian-Wollersberger, @jyn514, @LeSeulArtichaut, @macdonaldo, @mark-i-m, @matklad, @mautamu, @mightyiam, @Monadic-Cat, @Nadrieril, @pickfire, @r-52, @richkadel, @spastorino, @Stupremee

pnkfelix (Oct 22 2020 at 14:10, on Zulip):

Major Changes

Some minor changes are omitted for brevity.

pnkfelix (Oct 22 2020 at 14:11, on Zulip):

Contributor-facing changes

pnkfelix (Oct 22 2020 at 14:11, on Zulip):
pnkfelix (Oct 22 2020 at 14:11, on Zulip):

Internal changes

pnkfelix (Oct 22 2020 at 14:11, on Zulip):

Changes in progress

pnkfelix (Oct 22 2020 at 14:11, on Zulip):

@WG-llvm checkin by @Nikita Popov :

The first parts of the mustprogress attribute have landed (https://reviews.llvm.org/D86233, https://reviews.llvm.org/D85393). This means that forward progress is now opt-in. Once this work is complete, this will resolve current bugs around infinite loops being optimized away. There isn't a patch for that yet, but I dare say the hardest part (convincing people that this is the right thing to do) is done now...

pnkfelix (Oct 22 2020 at 14:12, on Zulip):

MemorySSA-based DSE has been enabled by default (https://reviews.llvm.org/D87163), which means dead stores are eliminated across based blocks. Expect some compile-time regression... More importantly for rust, I've ported MemCpyOpt to MSSA (https://reviews.llvm.org/D89207), which will allow memcpy's to be eliminated across blocks as well, a long-standing optimization problem. It's not the first attempt to do this, but I'm optimistic :)

pnkfelix (Oct 22 2020 at 14:12, on Zulip):

Some older changes that I probably have not mentioned before. I've finally managed to make correlated value propagation optimize conditions based on local value ranges (https://reviews.llvm.org/D69686), which will resolve various missing bounds check optimizations where constant operands are involved. There is also a new constraint elimination pass (https://reviews.llvm.org/D84547) which is able to reason about complex implications symbolically. However, it is not enabled by default, and there's no roadmap to enable it at present.

pnkfelix (Oct 22 2020 at 14:12, on Zulip):

Also been a while ago, but someone took over my work for introducing saturating float to int casts (https://reviews.llvm.org/D54749). There's been some slow progress, and maybe there's a chance that this will land before the heat death of the universe... Rust has worked around this in the meantime, but once/if it lands we can improve codegen.

pnkfelix (Oct 22 2020 at 14:12, on Zulip):

Okay!

nikomatsakis (Oct 22 2020 at 14:13, on Zulip):

Those are both super exciting

pnkfelix (Oct 22 2020 at 14:13, on Zulip):

I am so happy, by the way, to see this work reported

nikomatsakis (Oct 22 2020 at 14:13, on Zulip):

and thanks to @Joshua Nelson and @Nikita Popov for preparing them!

pnkfelix (Oct 22 2020 at 14:13, on Zulip):

I mean WG-rustc-dev-guide has been doing epic work for a while

pnkfelix (Oct 22 2020 at 14:13, on Zulip):

but I am also very happy to see WG-llvm awaken from a status that I might call ... dormant? :smile:

nikomatsakis (Oct 22 2020 at 14:13, on Zulip):

Recommend debug-logging instead of debug #917

I'm thinking I gotta read and find out what this even means :P

pnkfelix (Oct 22 2020 at 14:14, on Zulip):

Okay moving along (because last week's meeting definitely ran over time)

pnkfelix (Oct 22 2020 at 14:14, on Zulip):

Beta-nominations

T-compiler

pnkfelix (Oct 22 2020 at 14:14, on Zulip):
pnkfelix (Oct 22 2020 at 14:15, on Zulip):

(this has not yet landed in nightly nor has it been r+'ed)

pnkfelix (Oct 22 2020 at 14:15, on Zulip):

but @simulacrum has given some feedback

pnkfelix (Oct 22 2020 at 14:16, on Zulip):

@simulacrum would you prefer to see a smaller patch put up as a backport candidate?

simulacrum (Oct 22 2020 at 14:16, on Zulip):

Hm I was not aware of beta-nom here

pnkfelix (Oct 22 2020 at 14:16, on Zulip):

the beta-nom only happened recently

pnkfelix (Oct 22 2020 at 14:16, on Zulip):

I can imagine being nervous about the self.can_coerce argument change

pnkfelix (Oct 22 2020 at 14:16, on Zulip):

even if its correct

simulacrum (Oct 22 2020 at 14:16, on Zulip):

Ah, I see the comment. I do not have particular feedback. The patch felt fairly small, and only touched error code I think anyway

pnkfelix (Oct 22 2020 at 14:17, on Zulip):

@Esteban Küber was that important for fixing the ICE?

simulacrum (Oct 22 2020 at 14:17, on Zulip):

I was hoping to get @nikomatsakis or someone else more familiar with coercion to double check the coercion ordering change

nikomatsakis (Oct 22 2020 at 14:17, on Zulip):

I can review, I just read it through

pnkfelix (Oct 22 2020 at 14:17, on Zulip):

okay

nikomatsakis (Oct 22 2020 at 14:17, on Zulip):

I think the coercion ordering change should be harmless

simulacrum (Oct 22 2020 at 14:17, on Zulip):

error -> ICE seems fine to backport

nikomatsakis (Oct 22 2020 at 14:17, on Zulip):

(and I see why @Esteban Küber did it)

pnkfelix (Oct 22 2020 at 14:18, on Zulip):

okay

pnkfelix (Oct 22 2020 at 14:18, on Zulip):

I think the PR overall seems safe to backport then

pnkfelix (Oct 22 2020 at 14:18, on Zulip):

I'm okay approving the backport if niko approves landing it on nightly

pnkfelix (Oct 22 2020 at 14:18, on Zulip):

next

pnkfelix (Oct 22 2020 at 14:18, on Zulip):
pnkfelix (Oct 22 2020 at 14:19, on Zulip):

I think we should decline to backport this, based on the comment thread

pnkfelix (Oct 22 2020 at 14:19, on Zulip):

and its possible it should just be closed

pnkfelix (Oct 22 2020 at 14:19, on Zulip):

but we don't have to decide about closing it right now

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

I mostly just want to decline to backport and move along, unless someone wants to champion backporting it.

nikomatsakis (Oct 22 2020 at 14:20, on Zulip):

one thing I would want to note:

nikomatsakis (Oct 22 2020 at 14:20, on Zulip):

In the future, would it be possible for me to be pinged when iOS-related PRs are made? I work for a company that intends on using Rust on iOS in production, so I'd gladly provide testing.

nikomatsakis (Oct 22 2020 at 14:20, on Zulip):

possible notification group?

pnkfelix (Oct 22 2020 at 14:21, on Zulip):

(the basic summary, as I understand from comments, is that bug filer had their Cargo.toml set up to build cdylib, which was silently ignored on iOS/tvOS target prior to the PR that allowed dynamic linking.)

pnkfelix (Oct 22 2020 at 14:21, on Zulip):

I'm inferring from the silence that no one wants to champion a backport here.

nikomatsakis (Oct 22 2020 at 14:22, on Zulip):

I'm still reading the thread

pnkfelix (Oct 22 2020 at 14:22, on Zulip):

(Oh, I forgot to point out that PR is contentious (based on comment thread), hasn't landed on nightly, and hasn't been r+'ed. :wink: )

nikomatsakis (Oct 22 2020 at 14:22, on Zulip):

I'm not sure I would champion the backport

nikomatsakis (Oct 22 2020 at 14:22, on Zulip):

but I do think the correct course of action is complicated

nikomatsakis (Oct 22 2020 at 14:22, on Zulip):

I can't tell yet if there's a known workaround?

pnkfelix (Oct 22 2020 at 14:22, on Zulip):

I think there are cargo changes that were proposed

pnkfelix (Oct 22 2020 at 14:23, on Zulip):

that at least fixed the problem for the bug filer

nikomatsakis (Oct 22 2020 at 14:23, on Zulip):

yeah, it seems like "no", but there are cargo changes

nikomatsakis (Oct 22 2020 at 14:23, on Zulip):

but I don't know if the cargo team would accept those changes

pnkfelix (Oct 22 2020 at 14:23, on Zulip):

I can certianly believe that we can do better here

pnkfelix (Oct 22 2020 at 14:23, on Zulip):

Oh, did the cargo changes not land ?}

nikomatsakis (Oct 22 2020 at 14:23, on Zulip):

I guess I think this PR is not the right fix, but I would like to help this person get their code working

pnkfelix (Oct 22 2020 at 14:23, on Zulip):

I didn't follow through to see what the status was of the linked cargo PR

pnkfelix (Oct 22 2020 at 14:23, on Zulip):

Yeah okay

nikomatsakis (Oct 22 2020 at 14:23, on Zulip):

https://github.com/rust-lang/cargo/pull/8789

pnkfelix (Oct 22 2020 at 14:24, on Zulip):

I suggest we open an issue

pnkfelix (Oct 22 2020 at 14:24, on Zulip):

and try to migrate discussion of the problem there

nikomatsakis (Oct 22 2020 at 14:24, on Zulip):

@Alex Crichton sounds dubious :)

pnkfelix (Oct 22 2020 at 14:24, on Zulip):

rather than try to judge the merits of this PR itself

pnkfelix (Oct 22 2020 at 14:24, on Zulip):

anyway

nikomatsakis (Oct 22 2020 at 14:25, on Zulip):

OK, I couldn't tell if the cfg_attr approach that @Jonas Schievink suggested would work ...

pnkfelix (Oct 22 2020 at 14:25, on Zulip):

declining to beta-backport

pnkfelix (Oct 22 2020 at 14:25, on Zulip):

nikomatsakis said:

OK, I couldn't tell if the cfg_attr approach that Jonas Schievink suggested would work ...

(I think its something where you need the relevant settings in the Cargo.toml; they don't work in the .rs file for some reason)

pnkfelix (Oct 22 2020 at 14:25, on Zulip):

(and I don't know if Cargo.toml's support conditional settings. I guess one could encode it with features...?)

pnkfelix (Oct 22 2020 at 14:26, on Zulip):

next

pnkfelix (Oct 22 2020 at 14:26, on Zulip):
pnkfelix (Oct 22 2020 at 14:27, on Zulip):

this is r+'ed and merged into nightly

pnkfelix (Oct 22 2020 at 14:28, on Zulip):

unfortunately disabling the optimization does cause observable regression, according to @simulacrum 's comment at end of this PR

simulacrum (Oct 22 2020 at 14:28, on Zulip):

I am increasingly concerned about the seemingly frequent disabling/reverts of mir opts that we're backporting - should we be considering a blanket removal of mir opts on beta or something like that?

simulacrum (Oct 22 2020 at 14:28, on Zulip):

(I am concerned with the need in the first place, but the backports make me even more concerned that we might miss one)

pnkfelix (Oct 22 2020 at 14:28, on Zulip):

mmm

tm (Oct 22 2020 at 14:29, on Zulip):

disabling it is observable improvement

pnkfelix (Oct 22 2020 at 14:29, on Zulip):

I dunno, a blanket disable poses more serious riskes w.r.t. performance regressions

pnkfelix (Oct 22 2020 at 14:29, on Zulip):

tm said:

disabling it is observable improvment

what? That's not how I interpreted the comment

pnkfelix (Oct 22 2020 at 14:29, on Zulip):

/me looks

pnkfelix (Oct 22 2020 at 14:29, on Zulip):

whoa

simulacrum (Oct 22 2020 at 14:29, on Zulip):

Oh yeah the comment is misleading, I missed that this is not a new optimization

simulacrum (Oct 22 2020 at 14:30, on Zulip):

anyway, I'm interested in hearing from mir-opt folks if this is just expected or if maybe we need more vetting or something on opts

pnkfelix (Oct 22 2020 at 14:30, on Zulip):

I don't even understand

pnkfelix (Oct 22 2020 at 14:30, on Zulip):

I guess the analysis itself is costly

nikomatsakis (Oct 22 2020 at 14:30, on Zulip):

sounds like this optimization may not have been very effective to begin with?

nikomatsakis (Oct 22 2020 at 14:30, on Zulip):

ah, that makes sense

nikomatsakis (Oct 22 2020 at 14:30, on Zulip):

(I guess this comes back to the need to benchmark generated code directly)

pnkfelix (Oct 22 2020 at 14:30, on Zulip):

yeah

nikomatsakis (Oct 22 2020 at 14:30, on Zulip):

How often have we been backporting "disable opt X" PRs?

pnkfelix (Oct 22 2020 at 14:31, on Zulip):

its good that we try to focus on compiler performance

simulacrum (Oct 22 2020 at 14:31, on Zulip):

I feel like I have seen 3-4 mir opt disable backports in the last cycle

simulacrum (Oct 22 2020 at 14:31, on Zulip):

maybe I'm overcounting though

pnkfelix (Oct 22 2020 at 14:31, on Zulip):

but doing it to the exclusion of looking at overall generated code performance is definitely subpar

pnkfelix (Oct 22 2020 at 14:31, on Zulip):

Lets beta backport this

nikomatsakis (Oct 22 2020 at 14:31, on Zulip):

It seems like disabling this optimization is the right move in this moment

nikomatsakis (Oct 22 2020 at 14:31, on Zulip):

but I'm glad you're raising the pattern

pnkfelix (Oct 22 2020 at 14:31, on Zulip):

and plan to investigate the question of mir-opts in general

pnkfelix (Oct 22 2020 at 14:32, on Zulip):

but we don't have to decide that right now

pnkfelix (Oct 22 2020 at 14:32, on Zulip):

(maybe make it an item to discuss tomorrow...)

pnkfelix (Oct 22 2020 at 14:32, on Zulip):

beta backport approved

simulacrum (Oct 22 2020 at 14:32, on Zulip):

FWIW, quick scroll through:

pnkfelix (Oct 22 2020 at 14:32, on Zulip):
nikomatsakis (Oct 22 2020 at 14:33, on Zulip):

case in point? :)

pnkfelix (Oct 22 2020 at 14:33, on Zulip):

hmm? I was assuming this is LLVM opt?

simulacrum (Oct 22 2020 at 14:33, on Zulip):

No, looks like mir opt.

pnkfelix (Oct 22 2020 at 14:33, on Zulip):

oh

nikomatsakis (Oct 22 2020 at 14:33, on Zulip):

looks like a MIR opt to me

pnkfelix (Oct 22 2020 at 14:33, on Zulip):

never mind

pnkfelix (Oct 22 2020 at 14:33, on Zulip):

/me thought there was an LLVM opt called inst-combine ...

nikomatsakis (Oct 22 2020 at 14:33, on Zulip):

I think there is

pnkfelix (Oct 22 2020 at 14:33, on Zulip):

oi

pnkfelix (Oct 22 2020 at 14:34, on Zulip):

okay, well, backport approved, I think

pnkfelix (Oct 22 2020 at 14:35, on Zulip):

pnkfelix said:

is there an issue filed for the bug in the mir-opt itself here?

pnkfelix (Oct 22 2020 at 14:35, on Zulip):

(I saw a bug filed for PR #78195, which was good; just wanted to know if there was a simllar bug here....)

pnkfelix (Oct 22 2020 at 14:35, on Zulip):

anyway we can move forward

pnkfelix (Oct 22 2020 at 14:36, on Zulip):

libs-impl

T-rustdoc

pnkfelix (Oct 22 2020 at 14:36, on Zulip):

Stable-nominations

T-compiler

pnkfelix (Oct 22 2020 at 14:36, on Zulip):
pnkfelix (Oct 22 2020 at 14:36, on Zulip):

this is not getting stable approved, skipping

pnkfelix (Oct 22 2020 at 14:36, on Zulip):
simulacrum (Oct 22 2020 at 14:37, on Zulip):

Hm, this might mean a point release is warranted given it's a soundness fix. Annoying that we didn't catch it earlier :/

pnkfelix (Oct 22 2020 at 14:37, on Zulip):

... this is an unsoundness... and the PR is pretty trivial, right? So it seems like a clear approval

simulacrum (Oct 22 2020 at 14:37, on Zulip):

(But regardless seems like a good candidate for stable backport)

pnkfelix (Oct 22 2020 at 14:37, on Zulip):

yeah. Approved for stable backport.

pnkfelix (Oct 22 2020 at 14:37, on Zulip):

libs-impl

T-rustdoc

pnkfelix (Oct 22 2020 at 14:38, on Zulip):

PRs S-waiting-on-team

T-compiler

pnkfelix (Oct 22 2020 at 14:38, on Zulip):
pnkfelix (Oct 22 2020 at 14:38, on Zulip):

I'm personally in favor of the keeping the status quo here.

pnkfelix (Oct 22 2020 at 14:39, on Zulip):

mostly out of conservatism

pnkfelix (Oct 22 2020 at 14:39, on Zulip):

i..e, the points in favor, if I understand them correctly, largely only affect use statements

pnkfelix (Oct 22 2020 at 14:40, on Zulip):

which I don't see as something we should be investing time optimizing

pnkfelix (Oct 22 2020 at 14:40, on Zulip):

the one bullet that does worry me is this: "We observed small performance degradation by having the longer paths, and this might help to recoup some of that lost perf."

pnkfelix (Oct 22 2020 at 14:41, on Zulip):

but how small? If we're observing it, it cannot be that small. And that in turn makes me wonder: This amount of difference matters? Why?

pnkfelix (Oct 22 2020 at 14:41, on Zulip):

(E.g.: Is it something where a certain length makes us cross a threshold in the memory allocator path? Or in the hashtables?)

pnkfelix (Oct 22 2020 at 14:41, on Zulip):

So it seems like there may be interesting questions to explore here

Santiago Pastorino (Oct 22 2020 at 14:42, on Zulip):

one alternative that I've read somewhere, unsure if it was considered was rustc/foo for dirs and rustc_foo to use

pnkfelix (Oct 22 2020 at 14:42, on Zulip):

yeah, that's listed as a secondary alternative in the linked hackmd I think.

Santiago Pastorino (Oct 22 2020 at 14:42, on Zulip):

the second one says compiler not rustc

pnkfelix (Oct 22 2020 at 14:42, on Zulip):

oh. But we've already moved everything into compiler/

pnkfelix (Oct 22 2020 at 14:42, on Zulip):

(right?)

Joshua Nelson (Oct 22 2020 at 14:43, on Zulip):

I would prefer rustc/ though

pnkfelix (Oct 22 2020 at 14:43, on Zulip):

We're not going to move everything somewhere else, are we?

Joshua Nelson (Oct 22 2020 at 14:43, on Zulip):

I don't feel strongly about any of this to be clear

Santiago Pastorino (Oct 22 2020 at 14:43, on Zulip):

pnkfelix said:

oh. But we've already moved everything into compiler/

yeah but some people complained about that

pnkfelix (Oct 22 2020 at 14:43, on Zulip):

I don't think we can resolve this in a triage meeting

Joshua Nelson (Oct 22 2020 at 14:43, on Zulip):

But I do kind of wish we'd picked different ones directories to move things to

pnkfelix (Oct 22 2020 at 14:43, on Zulip):

maybe we need a design meeting. :sad:

pnkfelix (Oct 22 2020 at 14:44, on Zulip):

anyway I think we're better off moving along. I don't want to run out of time again today

pnkfelix (Oct 22 2020 at 14:44, on Zulip):

libs-impl

pnkfelix (Oct 22 2020 at 14:44, on Zulip):

Issues of Note

Short Summary

pnkfelix (Oct 22 2020 at 14:44, on Zulip):

P-critical

T-compiler

pnkfelix (Oct 22 2020 at 14:44, on Zulip):
pnkfelix (Oct 22 2020 at 14:44, on Zulip):

hmm that sounds scary

LeSeulArtichaut (Oct 22 2020 at 14:45, on Zulip):

It’s a bug that has prevented quite a few people to use Rust 1.46+

nikomatsakis (Oct 22 2020 at 14:45, on Zulip):

hmm

pnkfelix (Oct 22 2020 at 14:46, on Zulip):

so we still need an MCVE here

nikomatsakis (Oct 22 2020 at 14:46, on Zulip):

that does seem like the first step

pnkfelix (Oct 22 2020 at 14:47, on Zulip):

(or maybe one is linked; see Davester47's comment near end. Not sure.)

pnkfelix (Oct 22 2020 at 14:47, on Zulip):

Also, @Joshua Nelson suggested reverting PR #72796

pnkfelix (Oct 22 2020 at 14:47, on Zulip):

but @lcnr said to revert anything we'd have to revert PR #75443

pnkfelix (Oct 22 2020 at 14:48, on Zulip):

these two comments seem in conflict; @Joshua Nelson , do you know what @lcnr meant?

nikomatsakis (Oct 22 2020 at 14:48, on Zulip):

#75443 is a fix for a problem introduced by #72796

LeSeulArtichaut (Oct 22 2020 at 14:48, on Zulip):

I think one of those PRs was a fix to another critical bug

nikomatsakis (Oct 22 2020 at 14:48, on Zulip):

so I think we'd probably have to revert both

pnkfelix (Oct 22 2020 at 14:48, on Zulip):

oh oh I see

nikomatsakis (Oct 22 2020 at 14:49, on Zulip):

#75443 is itself the right thing to do...I think :) I'm trying to page back in the discussion around this

pnkfelix (Oct 22 2020 at 14:49, on Zulip):

But if we remove #72796, then we can remove #75443, at least in short term?

pnkfelix (Oct 22 2020 at 14:50, on Zulip):

I'm basically wondering what, if anything, we can realistically do here that would be targeted for stable backport

nikomatsakis (Oct 22 2020 at 14:50, on Zulip):

yeah, I think so

nikomatsakis (Oct 22 2020 at 14:50, on Zulip):

I think it would be reasonable to back those out and try again

nikomatsakis (Oct 22 2020 at 14:50, on Zulip):

(though it'd be great to have a MCVE to work with)

pnkfelix (Oct 22 2020 at 14:50, on Zulip):

I think we should investigate doing that

pnkfelix (Oct 22 2020 at 14:50, on Zulip):

and yeah, getting an MCVE is important

pnkfelix (Oct 22 2020 at 14:50, on Zulip):

I can allocate time today to look into that, I think/hope.

pnkfelix (Oct 22 2020 at 14:51, on Zulip):

lets move along under assumption that we will investigate reverts as a short term fix

pnkfelix (Oct 22 2020 at 14:51, on Zulip):

while also looking into caching strategies etc for long term

pnkfelix (Oct 22 2020 at 14:51, on Zulip):
pnkfelix (Oct 22 2020 at 14:51, on Zulip):

where are we on this, we talked about it last week ...

pnkfelix (Oct 22 2020 at 14:52, on Zulip):

@WG-llvm do you all want to be in charge of making a decision here?

nikomatsakis (Oct 22 2020 at 14:52, on Zulip):

nikomatsakis said:

#75443 is itself the right thing to do...I think :) I'm trying to page back in the discussion around this

(yeah, I can't really see a problem with doing it)

pnkfelix (Oct 22 2020 at 14:52, on Zulip):

My personal inclination regarding rust#76387 is to backport LLVM's patch

pnkfelix (Oct 22 2020 at 14:53, on Zulip):

@Aaron Hill do you have time to look into doing that?

Aaron Hill (Oct 22 2020 at 14:53, on Zulip):

sure!

pnkfelix (Oct 22 2020 at 14:53, on Zulip):

fantastic!

pnkfelix (Oct 22 2020 at 14:53, on Zulip):

okay

pnkfelix (Oct 22 2020 at 14:53, on Zulip):

next

Aaron Hill (Oct 22 2020 at 14:53, on Zulip):

I haven't backported an llvm patch before, so I'm not sure about the procedure

Aaron Hill (Oct 22 2020 at 14:53, on Zulip):

do I just open a PR?

pnkfelix (Oct 22 2020 at 14:53, on Zulip):

lets talk about it after meeting

pnkfelix (Oct 22 2020 at 14:54, on Zulip):

(maybe in whatever zulip stream is allocated for that issue)

pnkfelix (Oct 22 2020 at 14:54, on Zulip):
pnkfelix (Oct 22 2020 at 14:54, on Zulip):

"but with issues to go through"

pnkfelix (Oct 22 2020 at 14:54, on Zulip):

/me doesn't understand that phrase

lqd (Oct 22 2020 at 14:55, on Zulip):

bors had issues testing it, but is doing so as we speak

Santiago Pastorino (Oct 22 2020 at 14:55, on Zulip):

sorry for my wrong english :)

Santiago Pastorino (Oct 22 2020 at 14:55, on Zulip):

exactly bots don't like that PR :P

pnkfelix (Oct 22 2020 at 14:55, on Zulip):

its not necessarily wrong english

pnkfelix (Oct 22 2020 at 14:55, on Zulip):

just some of those words, like "issues", have certain connotations (i.e. github issues) in this meeting that lead to my own confusion

pnkfelix (Oct 22 2020 at 14:56, on Zulip):

okay well it seems like its moving along

pnkfelix (Oct 22 2020 at 14:56, on Zulip):

and so shall we

pnkfelix (Oct 22 2020 at 14:56, on Zulip):
pnkfelix (Oct 22 2020 at 14:56, on Zulip):

we discussed this already as well, phew

pnkfelix (Oct 22 2020 at 14:56, on Zulip):

lets move along then

pnkfelix (Oct 22 2020 at 14:56, on Zulip):

libs-impl

pnkfelix (Oct 22 2020 at 14:56, on Zulip):

hmm

pnkfelix (Oct 22 2020 at 14:57, on Zulip):

unassigned beta regression

pnkfelix (Oct 22 2020 at 14:57, on Zulip):

some investigation on thread from use named the8472

simulacrum (Oct 22 2020 at 14:57, on Zulip):

Seems reasonable to revert the PR on beta perhaps

pnkfelix (Oct 22 2020 at 14:58, on Zulip):

yeah lets do that

simulacrum (Oct 22 2020 at 14:58, on Zulip):

(as a first step)

pnkfelix (Oct 22 2020 at 14:58, on Zulip):

okay

simulacrum (Oct 22 2020 at 14:58, on Zulip):

I can do that

pnkfelix (Oct 22 2020 at 14:58, on Zulip):

next

pnkfelix (Oct 22 2020 at 14:58, on Zulip):

T-rustdoc

pnkfelix (Oct 22 2020 at 14:58, on Zulip):

I'm going to skip the P-high's

pnkfelix (Oct 22 2020 at 14:58, on Zulip):

and jump to perforance logs

pnkfelix (Oct 22 2020 at 14:58, on Zulip):

because we didn't cover perf last week

pnkfelix (Oct 22 2020 at 14:58, on Zulip):

Performance logs

A variety of changes, nothing particularly notable from a performance
perspective. #77792 is an
interesting win on migrating to tracing values rather than the older log
formatting. #76859 is also
interesting, starting out as a functional fix but appears to have been a
performance win for incremental benchmarks in some cases as well.

Triage done by @simulacrum.
Revision range: 06a079c43efb062e335e6e6c9dabd3c750619980..22e6b9c68941996daa45786b4145e6196e51f0f4

pnkfelix (Oct 22 2020 at 14:59, on Zulip):

4 Regressions, 7 Improvements, 0 Mixed
2 of them in rollups

pnkfelix (Oct 22 2020 at 14:59, on Zulip):

Regressions

#77755

#77873

#78060

#77250

pnkfelix (Oct 22 2020 at 14:59, on Zulip):

Improvements

#76859

#77792

#77796

#77947

#77373

#77908

#78151

pnkfelix (Oct 22 2020 at 14:59, on Zulip):

Nags requiring follow up

pnkfelix (Oct 22 2020 at 15:00, on Zulip):

Okay so the one with the nag is I guess the main one we want to follow-up on

pnkfelix (Oct 22 2020 at 15:00, on Zulip):

Nominated Issues

T-compiler

pnkfelix (Oct 22 2020 at 15:01, on Zulip):

I'm skipping ahead to the RFC's seciton here

pnkfelix (Oct 22 2020 at 15:01, on Zulip):

because I think those seem most important to highlght, at least this week

pnkfelix (Oct 22 2020 at 15:01, on Zulip):

RFC

pnkfelix (Oct 22 2020 at 15:01, on Zulip):
nikomatsakis (Oct 22 2020 at 15:01, on Zulip):

(gotta run)

pnkfelix (Oct 22 2020 at 15:02, on Zulip):

this is also old

pnkfelix (Oct 22 2020 at 15:02, on Zulip):

not gonna be resolved in today's meeting

pnkfelix (Oct 22 2020 at 15:02, on Zulip):

so I'll just point it out

pnkfelix (Oct 22 2020 at 15:02, on Zulip):

as well as the other one

pnkfelix (Oct 22 2020 at 15:02, on Zulip):
pnkfelix (Oct 22 2020 at 15:02, on Zulip):

So there's stuff that we skipped on today's agenda

pnkfelix (Oct 22 2020 at 15:02, on Zulip):

namely the P-high bugs

pnkfelix (Oct 22 2020 at 15:03, on Zulip):

but still, good meeting everyone!

pnkfelix (Oct 22 2020 at 15:03, on Zulip):

thanks to everyone in @T-compiler/meeting for attending!!!

Santiago Pastorino (Oct 22 2020 at 15:03, on Zulip):

should we leave those RFC nominations up for next week?

Santiago Pastorino (Oct 22 2020 at 15:11, on Zulip):

I've just created the agenda for next meeting https://hackmd.io/DEgK3x4SRJCQlqV0McB0xw?both

Santiago Pastorino (Oct 22 2020 at 15:11, on Zulip):

we will have checkins from @T-compiler/WG-meta and @WG-mir-opt

Santiago Pastorino (Oct 22 2020 at 15:12, on Zulip):

@nikomatsakis about meta-wg, I don't think we have stuff to share or if there's some active work there

Santiago Pastorino (Oct 22 2020 at 15:12, on Zulip):

about mir-opts, cc @oli so you're aware for next week :)

oli (Oct 22 2020 at 15:16, on Zulip):

thanks!

pnkfelix (Oct 22 2020 at 15:26, on Zulip):

pnkfelix said:

pnkfelix said:

is there an issue filed for the bug in the mir-opt itself here?

I created one, filed as issue #78239

Jonas Schievink (Oct 23 2020 at 10:30, on Zulip):

fyi, I've approved the iOS revert due to https://github.com/rust-lang/rust/pull/77716#issuecomment-714788059

Last update: Nov 25 2020 at 02:15UTC