Stream: t-compiler/meetings

Topic: [weekly meeting] 2020-09-24 #54818


Santiago Pastorino (Sep 23 2020 at 16:27, on Zulip):

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

Santiago Pastorino (Sep 23 2020 at 16:27, on Zulip):

The @WG-prioritization have done pre-triage in #t-compiler/wg-prioritization/alerts

Santiago Pastorino (Sep 23 2020 at 16:27, on Zulip):

@WG-prioritization have prepared the meeting agenda

Santiago Pastorino (Sep 23 2020 at 16:28, on Zulip):

We will have checkins from @WG-prioritization and @WG-rfc-2229

Santiago Pastorino (Sep 23 2020 at 16:28, on Zulip):

we've already filled our short @WG-prioritization checkin

Santiago Pastorino (Sep 23 2020 at 16:29, on Zulip):

@Aman Arora @nikomatsakis do you have something you want to share about @WG-rfc-2229?

Aman Arora (Sep 23 2020 at 16:32, on Zulip):

I can add some details I can't edit the note.

apiraino (Sep 24 2020 at 13:01, on Zulip):

I've added them (I think Santiago Pastorino can give you access to the file (for future use)

Santiago Pastorino (Sep 24 2020 at 13:46, on Zulip):

@Aman Arora send me you HackMD username on a PM please :)

Santiago Pastorino (Sep 24 2020 at 13:47, on Zulip):

Hi @T-compiler/meeting, triage meeting will be starting in ~ 13 minutes

Santiago Pastorino (Sep 24 2020 at 13:47, on Zulip):

Check out the meeting agenda

nikomatsakis (Sep 24 2020 at 13:54, on Zulip):

(fyi, I'm likely not available today)

pnkfelix (Sep 24 2020 at 14:03, on Zulip):

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

pnkfelix (Sep 24 2020 at 14:04, on Zulip):

we will start off with 5 minutes for ...

Announcements

pnkfelix (Sep 24 2020 at 14:04, on Zulip):
pnkfelix (Sep 24 2020 at 14:04, on Zulip):
pnkfelix (Sep 24 2020 at 14:05, on Zulip):
pnkfelix (Sep 24 2020 at 14:05, on Zulip):
pnkfelix (Sep 24 2020 at 14:05, on Zulip):
pnkfelix (Sep 24 2020 at 14:05, on Zulip):
pnkfelix (Sep 24 2020 at 14:05, on Zulip):
pnkfelix (Sep 24 2020 at 14:06, on Zulip):

WG checkins

@WG-prioritization checkin by @Santiago Pastorino:

pnkfelix (Sep 24 2020 at 14:06, on Zulip):

@WG-rfc-2229 checkin by @Aman Arora and @nikomatsakis:

Santiago Pastorino (Sep 24 2020 at 14:06, on Zulip):

I think we can remove ourselves (WG-prioritization) from the checkins by now, unsure what do you think about that @pnkfelix but we can discuss on a different topic

pnkfelix (Sep 24 2020 at 14:07, on Zulip):

(well certainly this last checkin seemed worth celebrating)

Santiago Pastorino (Sep 24 2020 at 14:07, on Zulip):

yes a ton of people have joined and this is why I think it was important to share but we can always add stuff in the announcements

pnkfelix (Sep 24 2020 at 14:08, on Zulip):

true

pnkfelix (Sep 24 2020 at 14:08, on Zulip):

(especially since the wg-prioritization team is the team that is preparing the announcements each week!)

Santiago Pastorino (Sep 24 2020 at 14:08, on Zulip):

:joy:

pnkfelix (Sep 24 2020 at 14:09, on Zulip):

let me just re-emphasize that first announcement: we have a planning meeting tomorrow, same time as this meeting.

pnkfelix (Sep 24 2020 at 14:09, on Zulip):

(this is my way of trying to force myself to remember it myself...)

pnkfelix (Sep 24 2020 at 14:10, on Zulip):

but okay, lets move along now

pnkfelix (Sep 24 2020 at 14:10, on Zulip):

Beta-nominations

T-compiler

pnkfelix (Sep 24 2020 at 14:10, on Zulip):
Santiago Pastorino (Sep 24 2020 at 14:11, on Zulip):

seems like a large change for a backport and I don't think it's urgent to fix the kind of issue it fixes

pnkfelix (Sep 24 2020 at 14:12, on Zulip):

is the exponential growth observable in typical programs? Or just on pathological microbenchmarks?

Wesley Wiser (Sep 24 2020 at 14:12, on Zulip):

I'm actually going to say we shouldn't backport because this isn't fixing a regression, it sounds like performance has always been bad in this case. I'd prefer to let this ride the trains.

Santiago Pastorino (Sep 24 2020 at 14:13, on Zulip):

I was going to say the same, is not even a regression

pnkfelix (Sep 24 2020 at 14:13, on Zulip):

yes I agree 100%

pnkfelix (Sep 24 2020 at 14:14, on Zulip):

(there are instances where a change is so good that we still will backport it even if it isn't a regression, but I think the risk associated with the size of this patch means we should let it ride the trains.)

pnkfelix (Sep 24 2020 at 14:14, on Zulip):

okay, declining to backport

pnkfelix (Sep 24 2020 at 14:14, on Zulip):
pnkfelix (Sep 24 2020 at 14:15, on Zulip):

regarding this dialog: https://github.com/rust-lang/rust/pull/76837/files#r490459679

pnkfelix (Sep 24 2020 at 14:15, on Zulip):

(where @RalfJ asked "so this is re-enabled via opt-level=3? What is plan to avoid accidentally opting into buggy optimizations?")

pnkfelix (Sep 24 2020 at 14:16, on Zulip):

@oli said "we've made a decision but haven't implemented it yet"

pnkfelix (Sep 24 2020 at 14:16, on Zulip):

what is the decision and/or design?

Wesley Wiser (Sep 24 2020 at 14:16, on Zulip):

I have a PR that needs to be rebased that adds the new -Zunsound-mir-opts flag https://github.com/rust-lang/rust/pull/76899

pnkfelix (Sep 24 2020 at 14:16, on Zulip):

okay

Wesley Wiser (Sep 24 2020 at 14:16, on Zulip):

After which no amount of -Zmir-opt-level=99999 will run unsound mir opts.

pnkfelix (Sep 24 2020 at 14:17, on Zulip):

unless you opt into the -Zunsound-mir-opts flag, of course

Wesley Wiser (Sep 24 2020 at 14:17, on Zulip):

Sure yeah, but you should hopefully know what you're getting into at that point :)

pnkfelix (Sep 24 2020 at 14:17, on Zulip):

anyway, okay. backport approved.

pnkfelix (Sep 24 2020 at 14:17, on Zulip):
pnkfelix (Sep 24 2020 at 14:18, on Zulip):

the decision here is probably coupled with our decision about whether to back out the LLVM 11 upgrade in the first place, no?

pnkfelix (Sep 24 2020 at 14:19, on Zulip):

still seems like a clear win to backport this under the likely assumption that we will be sticking with the LLVM 11 upgrade, I think.

Santiago Pastorino (Sep 24 2020 at 14:19, on Zulip):

I guess we can also vote as if we are sticking with LLVM 11 maybe?

pnkfelix (Sep 24 2020 at 14:19, on Zulip):

yeah I don't know if that discussion will happen today or tomorrow. Probably tomorrow I suspect.

pnkfelix (Sep 24 2020 at 14:19, on Zulip):

(at the aforementioned planning meeting)

Santiago Pastorino (Sep 24 2020 at 14:20, on Zulip):

we've said if I remember correctly that we were going to spend some time tomorrow

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

right. Okayh

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

anyway lets go ahead and backport approve rust#77063

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

I imagine its going to be trivial to land it,

Esteban K├╝ber (Sep 24 2020 at 14:20, on Zulip):

Famous last words

pnkfelix (Sep 24 2020 at 14:21, on Zulip):

at least compared to the overall effort of backing out the upgrade if we go down that path

pnkfelix (Sep 24 2020 at 14:21, on Zulip):

libs-impl

T-rustdoc

pnkfelix (Sep 24 2020 at 14:21, on Zulip):

Stable-nominations

T-compiler

libs-impl

T-rustdoc

pnkfelix (Sep 24 2020 at 14:21, on Zulip):

PRs S-waiting-on-team

T-compiler

pnkfelix (Sep 24 2020 at 14:21, on Zulip):
pnkfelix (Sep 24 2020 at 14:22, on Zulip):

yeah so the FCP period for compiler-team#347 is 10 days, so that is expected to be approved 3 days from now

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

and then I imagine we'll just land rust#75699

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

but we can leave it marked S-waiting-on-team

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

libs-impl

pnkfelix (Sep 24 2020 at 14:23, on Zulip):

Issues of Note

Short Summary

pnkfelix (Sep 24 2020 at 14:23, on Zulip):

P-critical

T-compiler

pnkfelix (Sep 24 2020 at 14:23, on Zulip):
pnkfelix (Sep 24 2020 at 14:24, on Zulip):

(so given that we just approved the backport, I guess this is under control.)

LeSeulArtichaut (Sep 24 2020 at 14:24, on Zulip):

This is fixed on nightly now

pnkfelix (Sep 24 2020 at 14:24, on Zulip):
Wesley Wiser (Sep 24 2020 at 14:24, on Zulip):

(I'm going to assign the unassigned P-critical issue to myself (#76803) since I created the PR to backport and I'm also reviewing the real fix)

pnkfelix (Sep 24 2020 at 14:25, on Zulip):

is the review upstream progressing on the fix for rust#76387 ?

pnkfelix (Sep 24 2020 at 14:25, on Zulip):

/me looks

Aaron Hill (Sep 24 2020 at 14:25, on Zulip):

I added some additional reviewers

Aaron Hill (Sep 24 2020 at 14:25, on Zulip):

so hopefully someone takes a look at it soon

pnkfelix (Sep 24 2020 at 14:25, on Zulip):

yes I see, and one of them asked someone else to look at it yesterday

pnkfelix (Sep 24 2020 at 14:26, on Zulip):

@Aaron Hill do you think its worth addressing the Pre-merge checks that are on the upstream code review tool?

pnkfelix (Sep 24 2020 at 14:26, on Zulip):

Or do you figure you're better off waiting for actual review feedback from a human before you do that?

Aaron Hill (Sep 24 2020 at 14:26, on Zulip):

I can address the variable name 'i' one

Aaron Hill (Sep 24 2020 at 14:26, on Zulip):

I'm not sure about the other one, since it's a pre-existing method name

Aaron Hill (Sep 24 2020 at 14:26, on Zulip):

I don't know if they'll want me to rename it

pnkfelix (Sep 24 2020 at 14:26, on Zulip):

hmm interesting

pnkfelix (Sep 24 2020 at 14:26, on Zulip):

yeah I see, okay.

pnkfelix (Sep 24 2020 at 14:27, on Zulip):

better to follow the precedent set in the code, I agree.

pnkfelix (Sep 24 2020 at 14:27, on Zulip):

well this is hopefully under control

pnkfelix (Sep 24 2020 at 14:27, on Zulip):

if its still unreviewed next week I guess we'll have to decide whether we want to land it ourselves even without upstream feedback

pnkfelix (Sep 24 2020 at 14:27, on Zulip):

but I think I'm willing to wait until then ...

pnkfelix (Sep 24 2020 at 14:28, on Zulip):

rust#76387 is only on nightly? Its not on beta?

pnkfelix (Sep 24 2020 at 14:28, on Zulip):

/me is a little surprised if so

pnkfelix (Sep 24 2020 at 14:28, on Zulip):

oh it never worked?

pnkfelix (Sep 24 2020 at 14:29, on Zulip):

yeah okay not a regression, it never worked.

pnkfelix (Sep 24 2020 at 14:29, on Zulip):

which means we'd be unlikely to beta backport it, I think...

pnkfelix (Sep 24 2020 at 14:29, on Zulip):

but it might get folded into a general LLVM upgrade, which may happen on beta ... or maybe the most recent upgrade is the last one we'll do.

pnkfelix (Sep 24 2020 at 14:29, on Zulip):

Okay I'm just talking to myself at this point.

pnkfelix (Sep 24 2020 at 14:30, on Zulip):

anyway I think its under control

nagisa (Sep 24 2020 at 14:30, on Zulip):

we might want to backport it into our llvm anyway

pnkfelix (Sep 24 2020 at 14:30, on Zulip):

cherry-pick, yes

nagisa (Sep 24 2020 at 14:30, on Zulip):

whenever it lands upstream

pnkfelix (Sep 24 2020 at 14:30, on Zulip):

I'm mostly sitting here debating about a beta backport

nagisa (Sep 24 2020 at 14:30, on Zulip):

I don't believe it made the llvm11 train

pnkfelix (Sep 24 2020 at 14:31, on Zulip):

e.g. if it doesn't get reviewed upstream, then I don't want to backport it to beta, but I might be willing to still cherry-pick it to our LLVM on our nightly after the next beta is cut.

pnkfelix (Sep 24 2020 at 14:31, on Zulip):

i.e. maximize its testing during its train ride.

pnkfelix (Sep 24 2020 at 14:32, on Zulip):

but this is all getting ahead of myself, since they may well review it upstream in the near future.

pnkfelix (Sep 24 2020 at 14:32, on Zulip):
pnkfelix (Sep 24 2020 at 14:32, on Zulip):

shoot I promised to write up mentorship instructions here and didn't follow through

pnkfelix (Sep 24 2020 at 14:33, on Zulip):

at this point time is tight, I probabl should reassign to someone who can take charge on just doing an alpha-rename

pnkfelix (Sep 24 2020 at 14:33, on Zulip):

@Santiago Pastorino would you be interested in working with me on this?

Santiago Pastorino (Sep 24 2020 at 14:33, on Zulip):

sure

pnkfelix (Sep 24 2020 at 14:33, on Zulip):

I suspect its largely mechanical; the biggest problems are just having the time to do some rebuilds. :)

nagisa (Sep 24 2020 at 14:33, on Zulip):

we could revert the pr from beta to get more time if necessary, it was entirely a perf optimisation IIRC

pnkfelix (Sep 24 2020 at 14:33, on Zulip):

okay lets touch base about it later

pnkfelix (Sep 24 2020 at 14:34, on Zulip):

I thought PR #73565 fixed a bug

pnkfelix (Sep 24 2020 at 14:34, on Zulip):

namely: "Fixed a bug where Debug would create aliasing references when using the specialized zip impl"

pnkfelix (Sep 24 2020 at 14:35, on Zulip):

yeah it fixed this bug: nth_back() for Zip returns wrong values #68536

nagisa (Sep 24 2020 at 14:35, on Zulip):

got it

pnkfelix (Sep 24 2020 at 14:35, on Zulip):

(I mean maybe @nagisa meant there is some other PR that injected #68536 and we could revert that PR?)

pnkfelix (Sep 24 2020 at 14:35, on Zulip):

oh okay

pnkfelix (Sep 24 2020 at 14:35, on Zulip):

anyway I think we can handle this.

pnkfelix (Sep 24 2020 at 14:35, on Zulip):

I just have been distracted with ... stuff...

pnkfelix (Sep 24 2020 at 14:35, on Zulip):

this is the wrong period of my life for me to default to self-assigning things

pnkfelix (Sep 24 2020 at 14:36, on Zulip):

anyway I'm confident @Santiago Pastorino and I will take care of it.

pnkfelix (Sep 24 2020 at 14:36, on Zulip):
pnkfelix (Sep 24 2020 at 14:36, on Zulip):

aforementioned fix is #76837

eddyb (Sep 24 2020 at 14:37, on Zulip):

oof I wish I could help a bit more with the mir-opt stuff, it's so hard to get them right :(

pnkfelix (Sep 24 2020 at 14:37, on Zulip):

oh and we beta-backport approved that earlier in the meeting.

pnkfelix (Sep 24 2020 at 14:37, on Zulip):

so that's under control too

pnkfelix (Sep 24 2020 at 14:37, on Zulip):

libs-impl

T-rustdoc

pnkfelix (Sep 24 2020 at 14:37, on Zulip):

Unassigned P-high regressions

Beta regressions

nagisa (Sep 24 2020 at 14:37, on Zulip):

yeah I feel like we're way too lax in our mir-opt efforts.

eddyb (Sep 24 2020 at 14:37, on Zulip):

I guess it's a good idea to throw new MIR optimizations at nightly and disable on beta - maybe we want to automate this?

eddyb (Sep 24 2020 at 14:38, on Zulip):

i.e. we could gate some optimizations on the release channel

nagisa (Sep 24 2020 at 14:38, on Zulip):

(and we still haven't the means to enable developers to write them in an obviously correct ways)

Wesley Wiser (Sep 24 2020 at 14:38, on Zulip):

We should take this out of band with this meeting but yeah, I would love to do better in this regard.

eddyb (Sep 24 2020 at 14:38, on Zulip):

(PM me if you have updates, at least on this one thing)

pnkfelix (Sep 24 2020 at 14:39, on Zulip):

I'm not 100% sure I'm on board for having something that's as a matter of course enabled on nightly and disabled on beta

pnkfelix (Sep 24 2020 at 14:39, on Zulip):

mostly because I would like nightly to be somewhat of a predictor of how the beta will behave...

pnkfelix (Sep 24 2020 at 14:40, on Zulip):

but at the same time I totally understand the idea, of giving the MIR optimizations a nice space to be deployed without pressure to get it perfect...

eddyb (Sep 24 2020 at 14:40, on Zulip):

maybe there could be a grace period, if nothing breaks, we let it graduate to beta. not the best though

pnkfelix (Sep 24 2020 at 14:40, on Zulip):

(of course another problem there is how many people are relying on nightly itself, for other reasons, and thus its not such a great idea to be lax in its soundness...)

Wesley Wiser (Sep 24 2020 at 14:40, on Zulip):

That's basically what we've been doing but we don't get enough feedback on nightly to catch everything

eddyb (Sep 24 2020 at 14:40, on Zulip):

more eyes, formal methods (hard), or better optimization infrastructure, would all be great but, I assume we're all spread thin atm

pnkfelix (Sep 24 2020 at 14:41, on Zulip):

:+1:

Wesley Wiser (Sep 24 2020 at 14:41, on Zulip):

We do generally try to make it easy to backport a revert that disables the optimization on beta.

pnkfelix (Sep 24 2020 at 14:41, on Zulip):

so, back to P-high beta regressions

pnkfelix (Sep 24 2020 at 14:41, on Zulip):
nagisa (Sep 24 2020 at 14:42, on Zulip):

there have been a couple of cycles in various places of DAGCombiner et al. in LLVM

nagisa (Sep 24 2020 at 14:42, on Zulip):

but for these to be at all actionable minimal reproducer is critical.

pnkfelix (Sep 24 2020 at 14:42, on Zulip):

okay. this hasn't had much time under the ice-breakers tag

pnkfelix (Sep 24 2020 at 14:43, on Zulip):

so maybe we'll see more progress here

pnkfelix (Sep 24 2020 at 14:43, on Zulip):

it also feeds into the question of whether to undo the LLVM 11 upgrade. :sad:

cuviper (Sep 24 2020 at 14:44, on Zulip):

FWIW, rc3 just landed and is known to fix some issues

pnkfelix (Sep 24 2020 at 14:44, on Zulip):

true, maybe this will get magically fixed.

pnkfelix (Sep 24 2020 at 14:44, on Zulip):

(by that, that is)

pnkfelix (Sep 24 2020 at 14:44, on Zulip):

anyway I'm willing to let this sit for now

LeSeulArtichaut (Sep 24 2020 at 14:44, on Zulip):

rc3 landed on nightly, right? So IIUC we could test this now, is that correct?

cuviper (Sep 24 2020 at 14:44, on Zulip):

Yes

pnkfelix (Sep 24 2020 at 14:44, on Zulip):

it will probably come up in tomorrow's (planned?) discussion of LLVM 11 upgrade before the we get into the actual planning of the planning meeting.

pnkfelix (Sep 24 2020 at 14:45, on Zulip):

@LeSeulArtichaut are you willing to take charge on testing this on nightly?

LeSeulArtichaut (Sep 24 2020 at 14:45, on Zulip):

pnkfelix said:

LeSeulArtichaut are you willing to take charge on testing this on nightly?

Sure, updating my nightly toolchain now

pnkfelix (Sep 24 2020 at 14:45, on Zulip):

(and double-checking that the bug reproduces for you in the first place, that is?)

pnkfelix (Sep 24 2020 at 14:45, on Zulip):

okay.

pnkfelix (Sep 24 2020 at 14:46, on Zulip):

next

pnkfelix (Sep 24 2020 at 14:46, on Zulip):

Nightly regressions

pnkfelix (Sep 24 2020 at 14:46, on Zulip):
pnkfelix (Sep 24 2020 at 14:46, on Zulip):

oh my

pnkfelix (Sep 24 2020 at 14:47, on Zulip):

does wasm only use LLD or something?

pnkfelix (Sep 24 2020 at 14:47, on Zulip):

(or is LLD the default on windows?)

pnkfelix (Sep 24 2020 at 14:47, on Zulip):

/me looks

nagisa (Sep 24 2020 at 14:47, on Zulip):

former.

pnkfelix (Sep 24 2020 at 14:47, on Zulip):

thanks

pnkfelix (Sep 24 2020 at 14:48, on Zulip):

hmm I wonder if we have an wasm notification group yet

pnkfelix (Sep 24 2020 at 14:48, on Zulip):

seems like a good thing to form potentially

LeSeulArtichaut (Sep 24 2020 at 14:48, on Zulip):

I believe we don't

LeSeulArtichaut (Sep 24 2020 at 14:48, on Zulip):

Wasn't there an MCP about this?

pnkfelix (Sep 24 2020 at 14:48, on Zulip):

about a wasm ping group?

LeSeulArtichaut (Sep 24 2020 at 14:48, on Zulip):

Yes

LeSeulArtichaut (Sep 24 2020 at 14:49, on Zulip):

https://github.com/rust-lang/compiler-team/issues/337

pnkfelix (Sep 24 2020 at 14:50, on Zulip):

okay so the group has been approved but has not been created yet?

LeSeulArtichaut (Sep 24 2020 at 14:50, on Zulip):

Seems so

pnkfelix (Sep 24 2020 at 14:50, on Zulip):

okay

pnkfelix (Sep 24 2020 at 14:50, on Zulip):

lets fix that. :)

pnkfelix (Sep 24 2020 at 14:50, on Zulip):

but not right now

pnkfelix (Sep 24 2020 at 14:51, on Zulip):

anyway this seems like something that we're unlikely to resolve ourselves

pnkfelix (Sep 24 2020 at 14:51, on Zulip):

the best we can do is properly isolate the bug to LLD itself and file a bug upstream

pnkfelix (Sep 24 2020 at 14:52, on Zulip):

I'm going to leave this unassigned for now

pnkfelix (Sep 24 2020 at 14:52, on Zulip):

Performance logs

This was the first week of semi-automated perf triage, and thank goodness:
There was a lot going on. Most regressions are either quite small or already
have a fix published.

pnkfelix (Sep 24 2020 at 14:52, on Zulip):

#72412 is probably the most
interesting case. It fixes a pathological problem involving nested closures by
adding cycle detection to what seems to be a relatively hot part of the code.
As a result, most users will see a slight compile-time regression
for their crates.

pnkfelix (Sep 24 2020 at 14:52, on Zulip):

Triage done by @ecstaticmorse.
Revision range: dbb73f8f79ab176a897d5a95e696adb71b957cbe..b01326ab033e41986d4a5c8b96ce4f40f3b38e30

2 Regressions, 5 Improvements, 4 Mixed,
1 of them in rollups

pnkfelix (Sep 24 2020 at 14:53, on Zulip):

Regressions

#76575 compare generic constants using AbstractConsts

#74040 fix unification of const variables

pnkfelix (Sep 24 2020 at 14:53, on Zulip):

Improvements

#76656 Don't query stability data when staged_api is off

#76311 Split core::slice to smaller mods

#76880 Update cc crate to 1.0.60 to understand aarch64-apple-darwin with clang

#76975 Rollup of 15 pull requests

#76680 Make ensure_sufficient_stack() non-generic, using cargo-llvm-lines

pnkfelix (Sep 24 2020 at 14:53, on Zulip):

that core::slice one #76311 is interesting

pnkfelix (Sep 24 2020 at 14:54, on Zulip):

maybe spitting file into mods caused change to LLVM inlining decisions?

pnkfelix (Sep 24 2020 at 14:54, on Zulip):

weird

Wesley Wiser (Sep 24 2020 at 14:54, on Zulip):

Yeah very

nagisa (Sep 24 2020 at 14:54, on Zulip):

cc one is similarly weird

eddyb (Sep 24 2020 at 14:54, on Zulip):

maybe the core::slice one interacts with CGU partitioning?

pnkfelix (Sep 24 2020 at 14:54, on Zulip):

that's another good theory

eddyb (Sep 24 2020 at 14:54, on Zulip):

though that's less of a problem in non-incremental builds I guess

pnkfelix (Sep 24 2020 at 14:54, on Zulip):

true

eddyb (Sep 24 2020 at 14:55, on Zulip):

still, dumping the CGUs before/after could be useful

Wesley Wiser (Sep 24 2020 at 14:55, on Zulip):

Maybe it has something to do with how much metadata we need to load when compiling against core::slice or something?

pnkfelix (Sep 24 2020 at 14:55, on Zulip):

(I can believe that random crate updates could have downstream effects on CGU partitioning, which could explain the CC one #76880)

eddyb (Sep 24 2020 at 14:55, on Zulip):

@Wesley Wiser no, all that is per-crate

eddyb (Sep 24 2020 at 14:55, on Zulip):

(I wish the CGU dump format was easier to deal with but it's workable I guess)

pnkfelix (Sep 24 2020 at 14:56, on Zulip):

anyway lets keep going

eddyb (Sep 24 2020 at 14:56, on Zulip):

now if only I knew what the difference between -Zprint-mono-items=lazy and -Zprint-mono-items=eager was...

pnkfelix (Sep 24 2020 at 14:56, on Zulip):

its interesting to ponder but probably not a good use of time right now

pnkfelix (Sep 24 2020 at 14:56, on Zulip):

Mixed

#76244 Removing the def_id field from hot ParamEnv to make it smaller

#72412 Issue 72408 nested closures exponential

#74949 Validate constants during const_eval_raw

#75119 New MIR optimization pass to reduce branches on match of tuples of enums

pnkfelix (Sep 24 2020 at 14:57, on Zulip):

hmm so regarding #75119

pnkfelix (Sep 24 2020 at 14:57, on Zulip):

the summary here at least is saying that it had max improvement on ... "real-world" code (depending on whether one regards style-servo-opt as "real-world")

pnkfelix (Sep 24 2020 at 14:58, on Zulip):

while the biggest regression was on a stress testing microbenchmawrk

pnkfelix (Sep 24 2020 at 14:58, on Zulip):

I'm inclined to weight the former as more important than the latter.

Santiago Pastorino (Sep 24 2020 at 14:58, on Zulip):

(up to 515.8% on incr-unchanged builds of ctfe-stress-4-check)

ugh

pnkfelix (Sep 24 2020 at 14:58, on Zulip):

but the real question is what did "been a regression overall" mean, in terms of how things looked overall

pnkfelix (Sep 24 2020 at 14:59, on Zulip):

Santiago Pastorino said:

(up to 515.8% on incr-unchanged builds of ctfe-stress-4-check)

ugh

this one has a fix though, according to the report?

oli (Sep 24 2020 at 14:59, on Zulip):

yes

oli (Sep 24 2020 at 14:59, on Zulip):

that has a fix

Santiago Pastorino (Sep 24 2020 at 14:59, on Zulip):

ahh I was checking it ok ok

pnkfelix (Sep 24 2020 at 15:00, on Zulip):

pnkfelix said:

I'm inclined to weight the former as more important than the latter.

oh, #75119 is still under active discussion by @Wesley Wiser and others, in terms of whether it should be guarded under mir-opt-level=2 or above

pnkfelix (Sep 24 2020 at 15:00, on Zulip):

so lets not worry to much about that here then

pnkfelix (Sep 24 2020 at 15:01, on Zulip):

(it does indeed look like a regression overall when I look at the report here: https://perf.rust-lang.org/compare.html?start=81e02708f1f4760244756548981277d5199baa9a&end=2e0edc0f28c5647141bedba02e7a222d3a5dc9c3&stat=instructions:u )

pnkfelix (Sep 24 2020 at 15:01, on Zulip):

Nags requiring follow up

pnkfelix (Sep 24 2020 at 15:01, on Zulip):

argh we are over the hour and still didn't get to nominated issues. :sad:

pnkfelix (Sep 24 2020 at 15:02, on Zulip):

let me skim the list quickly and see if there are any that are burning a hole in our ... pocket ...

pnkfelix (Sep 24 2020 at 15:02, on Zulip):

Nominated Issues

T-compiler

pnkfelix (Sep 24 2020 at 15:02, on Zulip):
pnkfelix (Sep 24 2020 at 15:02, on Zulip):

(didn't we talk about this last week?)

pnkfelix (Sep 24 2020 at 15:03, on Zulip):

(I cannot remember now. Maybe best to put it through FCP checkbox process at this point?)

Wesley Wiser (Sep 24 2020 at 15:03, on Zulip):

I think that was some other arm target

pnkfelix (Sep 24 2020 at 15:03, on Zulip):

ah thanks

Wesley Wiser (Sep 24 2020 at 15:03, on Zulip):

I can start the FCP process if that helps.

pnkfelix (Sep 24 2020 at 15:03, on Zulip):

everything else here is either not something we can resolve quickly, or is something we already discussed in the meeting

Pietro Albini (Sep 24 2020 at 15:03, on Zulip):

yeah, this is apple silicon

pnkfelix (Sep 24 2020 at 15:04, on Zulip):

(We may need to have a steering meeting regarding rust#76480 and macro stuff in general...)

pnkfelix (Sep 24 2020 at 15:04, on Zulip):

(at least I feel like we need to have a dedicated wg or something in that space...)

pnkfelix (Sep 24 2020 at 15:04, on Zulip):

libs-impl

pnkfelix (Sep 24 2020 at 15:05, on Zulip):

the list of nominated issues is interesting in general: https://hackmd.io/rtJNgRt3Sz2N5jX8c2-m3A?both#Nominated-Issues

pnkfelix (Sep 24 2020 at 15:05, on Zulip):

but we're out of time today

pnkfelix (Sep 24 2020 at 15:05, on Zulip):

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

Santiago Pastorino (Sep 24 2020 at 17:20, on Zulip):

next week we will have checkins from @WG-rls2.0 and @WG-self-profile

Santiago Pastorino (Sep 24 2020 at 17:20, on Zulip):

created next week's agenda from template https://hackmd.io/2HOfYckVTaOHBmLlLvvBXA?both

Santiago Pastorino (Sep 24 2020 at 17:21, on Zulip):

@matklad can you fill @WG-rls2.0 checkin for next week?

Santiago Pastorino (Sep 24 2020 at 17:22, on Zulip):

@Wesley Wiser can you fill it for @WG-self-profile?

Wesley Wiser (Sep 24 2020 at 17:38, on Zulip):

Yes!

Last update: Nov 25 2020 at 02:45UTC