Stream: t-compiler/meetings

Topic: [weekly meeting] 2020-08-05 #54818


Santiago Pastorino (Aug 05 2020 at 19:28, on Zulip):

Hi @T-compiler/meeting; the triage meeting will happen tomorrow at 2pm UTC

Santiago Pastorino (Aug 05 2020 at 19:28, on Zulip):

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

Santiago Pastorino (Aug 05 2020 at 19:28, on Zulip):

@WG-prioritization have prepared the meeting agenda

Santiago Pastorino (Aug 05 2020 at 19:29, on Zulip):

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

Santiago Pastorino (Aug 05 2020 at 19:30, on Zulip):

I'll be sharing some stuff about @WG-prioritization

Santiago Pastorino (Aug 05 2020 at 19:30, on Zulip):

@blitzerr and/or @nikomatsakis do you have something you want to share about @WG-rfc-2229?

nikomatsakis (Aug 05 2020 at 19:40, on Zulip):

I won't be at the meeting

nikomatsakis (Aug 05 2020 at 19:41, on Zulip):

but also, blitzerr isn't involved anymore

nikomatsakis (Aug 05 2020 at 19:41, on Zulip):

we should update the web-site

nikomatsakis (Aug 05 2020 at 19:41, on Zulip):

that said, I can leave a note :)

nikomatsakis (Aug 05 2020 at 19:49, on Zulip):

done

Santiago Pastorino (Aug 06 2020 at 13:20, on Zulip):

Hi @T-compiler/meeting, triage meeting will be starting in ~ 40 minutes

Santiago Pastorino (Aug 06 2020 at 13:20, on Zulip):

Check out the meeting agenda

Joshua Nelson (Aug 06 2020 at 13:27, on Zulip):

Hi @Santiago Pastorino, I don't see https://github.com/rust-lang/rust/issues/75100 on the list, could it be added?

Joshua Nelson (Aug 06 2020 at 13:27, on Zulip):

It has I-nominated

Joshua Nelson (Aug 06 2020 at 13:28, on Zulip):

I could add P-critical if it helps you find it

simulacrum (Aug 06 2020 at 13:29, on Zulip):

it needs T-compiler for compiler team to discuss

Joshua Nelson (Aug 06 2020 at 13:29, on Zulip):

Hmm ... It needs help from the compiler team but it only affects rustdoc

Joshua Nelson (Aug 06 2020 at 13:29, on Zulip):

Or in other words no one in rustdoc knows how to fix it

Joshua Nelson (Aug 06 2020 at 13:32, on Zulip):

I added T-compiler but you can remove it if it doesn't fit

pnkfelix (Aug 06 2020 at 14:02, on Zulip):

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

pnkfelix (Aug 06 2020 at 14:03, on Zulip):

we will start off with 5 minutes for ...

Announcements

Santiago Pastorino (Aug 06 2020 at 14:03, on Zulip):

Joshua Nelson said:

Hi Santiago Pastorino, I don't see https://github.com/rust-lang/rust/issues/75100 on the list, could it be added?

adding it to the agenda

simulacrum (Aug 06 2020 at 14:03, on Zulip):
pnkfelix (Aug 06 2020 at 14:04, on Zulip):
pnkfelix (Aug 06 2020 at 14:04, on Zulip):
pnkfelix (Aug 06 2020 at 14:04, on Zulip):
pnkfelix (Aug 06 2020 at 14:04, on Zulip):
pnkfelix (Aug 06 2020 at 14:05, on Zulip):
pnkfelix (Aug 06 2020 at 14:05, on Zulip):
pnkfelix (Aug 06 2020 at 14:05, on Zulip):

WG checkins

pnkfelix (Aug 06 2020 at 14:05, on Zulip):

@WG-prioritization checkin by @spastorino:

pnkfelix (Aug 06 2020 at 14:05, on Zulip):

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

Work is currently on a temporary hiatus because of exam period.

We have been working on various refactorings to allow captures
to capture not just individual variables but arbitrary places.

Some of the refactorings we are working on:

Santiago Pastorino (Aug 06 2020 at 14:07, on Zulip):

about prioritization checkin, it probably doesn't make sense anymore to checkin, given that things are settling, but we can discuss that after the meeting if you want @pnkfelix

pnkfelix (Aug 06 2020 at 14:08, on Zulip):

regarding tracing: this is an area that I'm not very familiar with, but it looks interesting.

pnkfelix (Aug 06 2020 at 14:09, on Zulip):

The docs for the tracing crate indicate that its aimed at making sense of asynchronous systems; but I wonder if the idea here is that those same characteristics will be useful to make sense of our query system?

oli (Aug 06 2020 at 14:09, on Zulip):

pretty much, yea. We don't know when and where queries execute

pnkfelix (Aug 06 2020 at 14:09, on Zulip):

or I guess the tracing being added is just for miri?

oli (Aug 06 2020 at 14:09, on Zulip):

this will only get "worse" with parallelism

oli (Aug 06 2020 at 14:09, on Zulip):

nope

oli (Aug 06 2020 at 14:09, on Zulip):

that's just the demo

pnkfelix (Aug 06 2020 at 14:09, on Zulip):

ah, your intent is indeed to have it work with everything, not just miri?

oli (Aug 06 2020 at 14:10, on Zulip):

jup

pnkfelix (Aug 06 2020 at 14:10, on Zulip):

I wasn't sure how to interpret the phrase "entire MIR interp stack"

pnkfelix (Aug 06 2020 at 14:10, on Zulip):

anyway I'll look into this more later

oli (Aug 06 2020 at 14:10, on Zulip):

I want to be able to annotate every query in a way that you can see the query stack in the logs if you want to

oli (Aug 06 2020 at 14:10, on Zulip):

or that you can only log things that happen within a specific query

pnkfelix (Aug 06 2020 at 14:10, on Zulip):

thanks @oli !

pnkfelix (Aug 06 2020 at 14:11, on Zulip):

Beta-nominations

T-compiler

pnkfelix (Aug 06 2020 at 14:11, on Zulip):
pnkfelix (Aug 06 2020 at 14:13, on Zulip):

yeah this seems like a pretty clear beta-backport approve situation, even for a P-medium bug

pnkfelix (Aug 06 2020 at 14:13, on Zulip):

beta-approved

pnkfelix (Aug 06 2020 at 14:13, on Zulip):
pnkfelix (Aug 06 2020 at 14:13, on Zulip):

so I have to ask: Who the eack was running around adding #[track_caller] to their fn main ?

pnkfelix (Aug 06 2020 at 14:14, on Zulip):

anyway, I guess I'm fine with backporting here

pnkfelix (Aug 06 2020 at 14:15, on Zulip):

beta backport approved

pnkfelix (Aug 06 2020 at 14:15, on Zulip):
pnkfelix (Aug 06 2020 at 14:17, on Zulip):

skimming over the diff, specifically the test case

pnkfelix (Aug 06 2020 at 14:17, on Zulip):

can someone explain to me why this is emitting that note about feature(const_fn) ?

pnkfelix (Aug 06 2020 at 14:17, on Zulip):

https://github.com/rust-lang/rust/pull/75136/files#diff-06753d66edbf473847020e3a89ccc160

pnkfelix (Aug 06 2020 at 14:18, on Zulip):

seems ... like a bug in the diagnostic, right?

pnkfelix (Aug 06 2020 at 14:18, on Zulip):

maybe because of how this particular error is being signalled, by returning Err from the check_rvalue method?

oli (Aug 06 2020 at 14:18, on Zulip):

if you turn that flag on it'll actually compile, because under the const_fn feature gate a bunch of things work that don't work stably in const fn

pnkfelix (Aug 06 2020 at 14:18, on Zulip):

or will adding feature(const_fn) actually make it work?

oli (Aug 06 2020 at 14:18, on Zulip):

the ICE was only in the stable const fn checks

pnkfelix (Aug 06 2020 at 14:18, on Zulip):

oh okay

pnkfelix (Aug 06 2020 at 14:19, on Zulip):

okay then

pnkfelix (Aug 06 2020 at 14:19, on Zulip):

(arguably the diagnostic could say that as part of its message, i.e. "unsizing casts are allowed solely for references in stable Rust")

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

but I'm fine with this, people will figure it out from the linked issue, presumably

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

beta backport approved

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

libs-impl

T-rustdoc

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

Stable-nominations

T-compiler

pnkfelix (Aug 06 2020 at 14:20, on Zulip):
simulacrum (Aug 06 2020 at 14:21, on Zulip):

Note though that I am personally against this (and we probably won't have a .3 anyway)

pnkfelix (Aug 06 2020 at 14:21, on Zulip):

it seems like a somewhat trivial patch?

pnkfelix (Aug 06 2020 at 14:21, on Zulip):

What risk do you identify with it, @simulacrum ?

simulacrum (Aug 06 2020 at 14:21, on Zulip):

mostly just "touched LLVM"

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

I don't mind letting it just ride the trains

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

"touched LLVM" ?

simulacrum (Aug 06 2020 at 14:22, on Zulip):

it does seem true that if LLVM just renamed, then we're fine

simulacrum (Aug 06 2020 at 14:22, on Zulip):

but I am always scared of backporting changes that affect codegen like this, where it's not really a "one line" change

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

this PR itself does not touch LLVM; but you just mean its dangerous to play games with our interface to LLVM?

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

yeah okay

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

that does sound like a reasonable fear

pnkfelix (Aug 06 2020 at 14:23, on Zulip):

so, what are risks associated with not backporting this, then

Wesley Wiser (Aug 06 2020 at 14:23, on Zulip):

Is there a reason to back port this to stable beyond just "LLVM has done this"?

pnkfelix (Aug 06 2020 at 14:23, on Zulip):

you cannot do cargo build --target=aarch64-apple-ios on stable now?

Wesley Wiser (Aug 06 2020 at 14:23, on Zulip):

This isn't breaking anyone on stable currently right?

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

pnkfelix said:

you cannot do cargo build --target=aarch64-apple-ios on stable now?

oh, you can do the build, you just get a ton of warning diagnostic output

simulacrum (Aug 06 2020 at 14:24, on Zulip):

yes, and someone alleged that the performance or whatever could be worse

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

yeah okay, that doesn't sound "that bad" compared to the risk of landing this too soon and having unexpected fallout

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

have you confirmed @simulacrum that this is indeed in the beta?

simulacrum (Aug 06 2020 at 14:25, on Zulip):

as best as I can tell, it is

simulacrum (Aug 06 2020 at 14:25, on Zulip):

I can double check

Wesley Wiser (Aug 06 2020 at 14:25, on Zulip):

I think I would prefer to let this ride the beta train.

pnkfelix (Aug 06 2020 at 14:25, on Zulip):

okay; your message from 7 days ago said you didn't understand how PR #73086 was in the beta

pnkfelix (Aug 06 2020 at 14:26, on Zulip):

lets just quickly establish now: if this isn't in the current beta, we would approve a beta backport, right?

simulacrum (Aug 06 2020 at 14:26, on Zulip):

I was confused by this comment "This is fixed in beta and nightly via #73086. Should probably be nominated for backport to a point release, if there is one."

simulacrum (Aug 06 2020 at 14:26, on Zulip):

which implied to me that we had backported or something (since it mentioned both nightly and beta)

simulacrum (Aug 06 2020 at 14:26, on Zulip):

but I just re-checked and it is indeed in beta

pnkfelix (Aug 06 2020 at 14:26, on Zulip):

i see

pnkfelix (Aug 06 2020 at 14:26, on Zulip):

okay

simulacrum (Aug 06 2020 at 14:26, on Zulip):

just naturally, by riding the trains

pnkfelix (Aug 06 2020 at 14:26, on Zulip):

sounds great. declining to do stable backport.

pnkfelix (Aug 06 2020 at 14:27, on Zulip):

libs-impl

T-rustdoc

pnkfelix (Aug 06 2020 at 14:27, on Zulip):

PR's S-waiting-on-team

T-compiler

pnkfelix (Aug 06 2020 at 14:28, on Zulip):

here is zulip discussion that niko referenced in a github comment: #t-compiler > android change #74860

pnkfelix (Aug 06 2020 at 14:29, on Zulip):

I think I agree with niko that an FCP is probably warranted here

pnkfelix (Aug 06 2020 at 14:30, on Zulip):

but before firing off an FCP

pnkfelix (Aug 06 2020 at 14:30, on Zulip):

should this be: T-compiler alone, or T-compiler+T-libs, or T-libs alone?

simulacrum (Aug 06 2020 at 14:31, on Zulip):

I think T-libs definitely, given that it is a breaking change, right?

Santiago Pastorino (Aug 06 2020 at 14:31, on Zulip):

this was one of my main questions, @nikomatsakis ended telling me that at this point everyone may want to weigh in

simulacrum (Aug 06 2020 at 14:31, on Zulip):

(though perhaps only on broken code)

pnkfelix (Aug 06 2020 at 14:31, on Zulip):

right. I'm just curious whether we want to delegate the decision entirely over to T-libs

Santiago Pastorino (Aug 06 2020 at 14:32, on Zulip):

at least a brief discussion about it is good :)

pnkfelix (Aug 06 2020 at 14:32, on Zulip):

(it strikes me as a somewhat trivial piece of code; the heart of the matter, to me, is about the public facing API change.)

pnkfelix (Aug 06 2020 at 14:32, on Zulip):

The author claims that one should only observe breakage in "weird" scenarios, but I'm not yet certain how true tha tis

pnkfelix (Aug 06 2020 at 14:33, on Zulip):

after all, if it really only broke things in weird scenarios, then why would you add it? (I suppose teh true argument is that it breaks buggy programs

simulacrum (Aug 06 2020 at 14:33, on Zulip):

imo, T-compiler doesn't need to weigh in -- a single reviewer can confirm it to do what it says

pnkfelix (Aug 06 2020 at 14:33, on Zulip):

(Okay to be fair, the author did say "either buggy or very weird", not just "weird")

simulacrum (Aug 06 2020 at 14:33, on Zulip):

(like we don't weigh in on T-lang decisions "just because they touch compiler code")

pnkfelix (Aug 06 2020 at 14:33, on Zulip):

exactly.

pnkfelix (Aug 06 2020 at 14:34, on Zulip):

does anyone object to me removing T-compiler, leaving tagged as T-libs, and calling for an FCP?

Santiago Pastorino (Aug 06 2020 at 14:34, on Zulip):

:+1:, should we also remove T-compiler label then? ... right :)

pnkfelix (Aug 06 2020 at 14:34, on Zulip):

(I'm not even sure I will have the authority to call for an FCP if its solely tagged as T-libs; I guess we'll see)

pnkfelix (Aug 06 2020 at 14:35, on Zulip):

basically: does anyone here think that we should require a majority of T-compiler to explicitly approve of this change before it lands?

simulacrum (Aug 06 2020 at 14:35, on Zulip):

you don't have authority to call fcp for t-libs

pnkfelix (Aug 06 2020 at 14:35, on Zulip):

(also, the PR author may need help figuring out where to add tests. :smile: )

pnkfelix (Aug 06 2020 at 14:36, on Zulip):

@Wesley Wiser this is unfortunately a scenario where emoji feedback is ambiguous

Wesley Wiser (Aug 06 2020 at 14:36, on Zulip):

Sigh

pnkfelix (Aug 06 2020 at 14:36, on Zulip):

I cannot tell if a :thumbs_down: is a way to agree with my (negative) statement or disagreement.

Wesley Wiser (Aug 06 2020 at 14:36, on Zulip):

I think we should not require a majority of T-compiler to sign off.

pnkfelix (Aug 06 2020 at 14:37, on Zulip):

okay

pnkfelix (Aug 06 2020 at 14:37, on Zulip):

I think all present members of T-compiler agree on that point then.

pnkfelix (Aug 06 2020 at 14:37, on Zulip):

I'll remove T-compiler and try to get someone on T-libs to call for FCP

pnkfelix (Aug 06 2020 at 14:37, on Zulip):

libs-impl

pnkfelix (Aug 06 2020 at 14:37, on Zulip):

Issues of Note

Short Summary

pnkfelix (Aug 06 2020 at 14:38, on Zulip):

P-critical

T-compiler

pnkfelix (Aug 06 2020 at 14:38, on Zulip):
pnkfelix (Aug 06 2020 at 14:39, on Zulip):

What do people think of revising the range.rs code in the short term, potentially solely on Windows?

pnkfelix (Aug 06 2020 at 14:40, on Zulip):

I have to admit, I have a slight prefernce for investing a bit more time in fixing the bug in LLVM upstream

pnkfelix (Aug 06 2020 at 14:40, on Zulip):

I spent a little while on it, got my feet wet with LoopStrengthReduction code there

Wesley Wiser (Aug 06 2020 at 14:40, on Zulip):

pnkfelix said:

What do people think of revising the range.rs code in the short term, potentially solely on Windows?

This seems like a reasonable work around in the short-term to me. But fixing the root issue would be best.

pnkfelix (Aug 06 2020 at 14:40, on Zulip):

(asked some embarrassing questions and/or made some embarrassing hypotheses)

pnkfelix (Aug 06 2020 at 14:41, on Zulip):

I'm not even certain, to be honest, that switching to wrapping_add in the way I suggest will fix ths

pnkfelix (Aug 06 2020 at 14:41, on Zulip):

(that is, I don't yet know if the add nuw is coming from our own codegen, or from some other LLVM optimization pass...)

pnkfelix (Aug 06 2020 at 14:42, on Zulip):

anyway its a stable-to-stable regression

pnkfelix (Aug 06 2020 at 14:42, on Zulip):

and a long standing one, for that matter

pnkfelix (Aug 06 2020 at 14:43, on Zulip):

so its not like we're up against a big deadline for a fix

Wesley Wiser (Aug 06 2020 at 14:43, on Zulip):

Ah ok. Well it seems like if it's as easy as switching to wrapping_add, we should perhaps do that now but if it's not, fixing the LLVM issue is probably a better use of time than trying to work around it.

pnkfelix (Aug 06 2020 at 14:43, on Zulip):

its just ... emabarassing

pnkfelix (Aug 06 2020 at 14:43, on Zulip):

moving along

pnkfelix (Aug 06 2020 at 14:43, on Zulip):
pnkfelix (Aug 06 2020 at 14:44, on Zulip):

yeah I briefly tried to reduce to MCVE but didn't get too far

pnkfelix (Aug 06 2020 at 14:44, on Zulip):

@Matthew Jasper are you around? You said you were planning to make a PR that only includes relevant commit from PR #73905, but maybe if @Matthew Jasper is short on time, someone else should try to identify the commit to cherry-pick?

pnkfelix (Aug 06 2020 at 14:45, on Zulip):

this is s a stable-to-beta regression, so we are up against a deadline here, in that we'd like something backported to beta before ...

pnkfelix (Aug 06 2020 at 14:45, on Zulip):

/me checks forge

Wesley Wiser (Aug 06 2020 at 14:46, on Zulip):

~3 weeks from now I believe

pnkfelix (Aug 06 2020 at 14:46, on Zulip):

right next release is august 27th

pnkfelix (Aug 06 2020 at 14:46, on Zulip):

but we will start testing a given beta some number of days before that point

pnkfelix (Aug 06 2020 at 14:47, on Zulip):

so, you know, it'd be nice to have this taken care of in the next two weeks

pnkfelix (Aug 06 2020 at 14:47, on Zulip):

its already assigned to me, I guess I can switch from MCVE generation to instead evaluating matthew jaspers commits from PR #73905

pnkfelix (Aug 06 2020 at 14:48, on Zulip):
pnkfelix (Aug 06 2020 at 14:49, on Zulip):

hmm, the patch in PR #74960 is a little more complicated then what I like to have to look at for beta backports

pnkfelix (Aug 06 2020 at 14:49, on Zulip):

/me tries to remember when niko is on vacation over next few weeks

simulacrum (Aug 06 2020 at 14:50, on Zulip):

until monday, inclusively, at least

pnkfelix (Aug 06 2020 at 14:50, on Zulip):

thanks @simulacrum

Santiago Pastorino (Aug 06 2020 at 14:50, on Zulip):

pnkfelix said:

hmm, the patch in PR #74960 is a little more complicated then what I like to have to look at for beta backports

this needs first to be merged on master and the we will need a backport

pnkfelix (Aug 06 2020 at 14:51, on Zulip):

tuesday is soon enough that I'm okay letting this continue to be on niko's plate for now

Santiago Pastorino (Aug 06 2020 at 14:51, on Zulip):

so following your previous reasoning about Matthew Jasper's PR, there's some rush here too

pnkfelix (Aug 06 2020 at 14:51, on Zulip):

(its true, there is rush here too; but I had that in mind in my thinking above)

pnkfelix (Aug 06 2020 at 14:51, on Zulip):

libs-impl

T-rustdoc

pnkfelix (Aug 06 2020 at 14:51, on Zulip):

Unassigned P-high regressions

Beta regressions

pnkfelix (Aug 06 2020 at 14:51, on Zulip):
pnkfelix (Aug 06 2020 at 14:52, on Zulip):

well its not like anyone uses arm chips, amirite? :wink:

pnkfelix (Aug 06 2020 at 14:53, on Zulip):

does the switch to gimli sidestep this entirely?

pnkfelix (Aug 06 2020 at 14:54, on Zulip):

or wait, does the switch to gimli predate rust#72746

pnkfelix (Aug 06 2020 at 14:54, on Zulip):

/me doesn't actually understand our unwind/backtrace story anymore

pnkfelix (Aug 06 2020 at 14:55, on Zulip):

Especially since unwind and backtrace are different things

pnkfelix (Aug 06 2020 at 14:55, on Zulip):

(sigh, sorry for the noise)

Joshua Nelson (Aug 06 2020 at 14:56, on Zulip):

/me has to go at 11 and can't answer too many questions about rustdoc afterwards

pnkfelix (Aug 06 2020 at 14:56, on Zulip):

okay lets see then

pnkfelix (Aug 06 2020 at 14:56, on Zulip):

rust#74820 remains unassigned

pnkfelix (Aug 06 2020 at 14:56, on Zulip):

it remains a beta regression

pnkfelix (Aug 06 2020 at 14:57, on Zulip):

so it would probably be good to have someone assigned to it

pnkfelix (Aug 06 2020 at 14:57, on Zulip):

but I'm not yet willing to add it to my own pile

pnkfelix (Aug 06 2020 at 14:57, on Zulip):

Nightly regressions

pnkfelix (Aug 06 2020 at 14:57, on Zulip):

lets quickly skip ahead to the nominated issue

pnkfelix (Aug 06 2020 at 14:58, on Zulip):

since @Joshua Nelson wont be able to hang out after the hour is up

pnkfelix (Aug 06 2020 at 14:58, on Zulip):

Nominated Issues

T-compiler

pnkfelix (Aug 06 2020 at 14:59, on Zulip):

do we have confirmation that PR #73566 is indeed the cause of this regression?

Joshua Nelson (Aug 06 2020 at 14:59, on Zulip):

I don't know what else it could be

Joshua Nelson (Aug 06 2020 at 14:59, on Zulip):

it's one of the known drawbacks of the pr, it treats async fn as impl Trait

Joshua Nelson (Aug 06 2020 at 14:59, on Zulip):

crater missed that it broke code

pnkfelix (Aug 06 2020 at 14:59, on Zulip):

hmm

Joshua Nelson (Aug 06 2020 at 14:59, on Zulip):

https://github.com/rust-lang/crater/issues/532

pnkfelix (Aug 06 2020 at 15:00, on Zulip):

okay I have some passing understanding of PR #73566

pnkfelix (Aug 06 2020 at 15:00, on Zulip):

namely since it was replacing the use of everybody_loops, which I was always horrified in how rustdoc was using it

pnkfelix (Aug 06 2020 at 15:01, on Zulip):

@Joshua Nelson I'll self-assign this and take a look

Joshua Nelson (Aug 06 2020 at 15:01, on Zulip):

more info at https://github.com/rust-lang/rust/issues/75100#issuecomment-668335125

Joshua Nelson (Aug 06 2020 at 15:01, on Zulip):

have to go, sorry

pnkfelix (Aug 06 2020 at 15:01, on Zulip):

last

pnkfelix (Aug 06 2020 at 15:01, on Zulip):

is

pnkfelix (Aug 06 2020 at 15:01, on Zulip):

performance report

pnkfelix (Aug 06 2020 at 15:01, on Zulip):

Performance logs

Triage done by simulacrum.
Revision range: efc02b03d18b0cbaa55b1e421d792f70a39230b2..19cefa68640843956eedd86227ddc1d35dbc6754

8 regressions, 2 improvements, 1 of them on rollups.
1 outstanding nag from last week.

pnkfelix (Aug 06 2020 at 15:02, on Zulip):

Regressions

pnkfelix (Aug 06 2020 at 15:02, on Zulip):

Improvements

pnkfelix (Aug 06 2020 at 15:02, on Zulip):

Nags:

simulacrum (Aug 06 2020 at 15:02, on Zulip):

The nag has since been resolved

pnkfelix (Aug 06 2020 at 15:02, on Zulip):

Who was going to investigate that last one? Was it @davidtwco ?

davidtwco (Aug 06 2020 at 15:02, on Zulip):

Yeah, we've fixed that.

pnkfelix (Aug 06 2020 at 15:02, on Zulip):

oh great

pnkfelix (Aug 06 2020 at 15:03, on Zulip):

okay then

davidtwco (Aug 06 2020 at 15:03, on Zulip):

#75155 made some changes to polymorphisation and saw some decent improvements

pnkfelix (Aug 06 2020 at 15:04, on Zulip):

okay well then I'm not sure if any of the items on the performance report deserve further discussion

pnkfelix (Aug 06 2020 at 15:04, on Zulip):

the regression on #65989 sounds potentially worrisome

pnkfelix (Aug 06 2020 at 15:05, on Zulip):

at least "up to 4%" is a scary phrase

pnkfelix (Aug 06 2020 at 15:05, on Zulip):

but since @njn said it was a "slight performance regression", maybe the 4% is only in rare scenarios

pnkfelix (Aug 06 2020 at 15:05, on Zulip):

we probably should explore other ways to present this data

lcnr (Aug 06 2020 at 15:05, on Zulip):

looks like it's only in stress tests

pnkfelix (Aug 06 2020 at 15:05, on Zulip):

@Wesley Wiser @eddyb there was something you all were looking at recently, right?

pnkfelix (Aug 06 2020 at 15:05, on Zulip):

other perf presentations, that is?

pnkfelix (Aug 06 2020 at 15:06, on Zulip):

/me tries to remember where that was

Wesley Wiser (Aug 06 2020 at 15:06, on Zulip):

@eddyb is working on collecting instruction counts instead of wall-clock time in the self-profiler.

Wesley Wiser (Aug 06 2020 at 15:06, on Zulip):

Is that what you're thinking of?

pnkfelix (Aug 06 2020 at 15:07, on Zulip):

yeah, and there was some other kind of graph that @eddyb was demoing as part of that

pnkfelix (Aug 06 2020 at 15:08, on Zulip):

I cannot find it though

pnkfelix (Aug 06 2020 at 15:08, on Zulip):

well maybe I'll find it later and if so, add it to agenda for another meeting. :)

pnkfelix (Aug 06 2020 at 15:08, on Zulip):

okay that's all everyone, sorry for going a bit over time

pnkfelix (Aug 06 2020 at 15:09, on Zulip):

bye to everyone in @T-compiler/meeting , thank you all so much for attending!

njn (Aug 06 2020 at 21:04, on Zulip):

pnkfelix said:

but since njn said it was a "slight performance regression", maybe the 4% is only in rare scenarios

@simulacrum did the triage this week! :grinning:

njn (Aug 06 2020 at 21:05, on Zulip):

And the 4% regression was on ctfe-stress-4, which is an artificial stress test for compile-time function evaluation, so it doesn't matter much. Apart from that benchmark, #65989 was actually a small win overall

Last update: Nov 25 2020 at 02:15UTC