Stream: t-compiler

Topic: planning meeting 2020.02.14


nikomatsakis (Feb 14 2020 at 14:45, on Zulip):

Hey @T-compiler/meeting -- planning meeting starts in about 15 minutes! The goal here (as a reminder) is to schedule steering meetings (aka design meetings) over the next few weeks. Now's your chance to submit last minute proposals. =)

You can view the list of pending proposals here.

(There are also a few proposals where we still need to write up the minutes...)

nikomatsakis (Feb 14 2020 at 14:45, on Zulip):

/poll Which topics would you most like to discuss in the next few weeks?

pnkfelix (Feb 14 2020 at 15:03, on Zulip):

heh, looking at the description for compiler-team#247, I actually feel better about how I've been running the triage meeting (at least recently)

pnkfelix (Feb 14 2020 at 15:04, on Zulip):

in terms of, I feel like the existing meetings have been focused on "avoiding disaster", and I've just been feeling crappy that we weren't doing more than that.

nikomatsakis (Feb 14 2020 at 15:04, on Zulip):

Hey @T-compiler/meeting -- planning meeting starting now-ish! See poll above.

nikomatsakis (Feb 14 2020 at 15:05, on Zulip):

Announcements

nikomatsakis (Feb 14 2020 at 15:05, on Zulip):

Whilst people persue the above, feel free to leave other announcements here

nikomatsakis (Feb 14 2020 at 15:05, on Zulip):
centril (Feb 14 2020 at 15:06, on Zulip):

/me needs to go out more... :cold_sweat:

nikomatsakis (Feb 14 2020 at 15:06, on Zulip):

pasted image

What's up with that.

nikomatsakis (Feb 14 2020 at 15:06, on Zulip):

I'm hoping you see a heart. :)

centril (Feb 14 2020 at 15:06, on Zulip):

There's a heart in the web-app

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

yours is almost better

pnkfelix (Feb 14 2020 at 15:07, on Zulip):

in terms of my increased brain activity trying to make sense of what that even is

Wesley Wiser (Feb 14 2020 at 15:08, on Zulip):

There's a blog post up for Inside-Rust which explains how to profile rustc using the -Zself-profiler tools. It's probably the most complete tutorial we have at the moment so if that interests you, please check it out and give feedback! https://github.com/rust-lang/blog.rust-lang.org/pull/516

nikomatsakis (Feb 14 2020 at 15:09, on Zulip):

pnkfelix said:

in terms of my increased brain activity trying to make sense of what that even is

it seems to me that there is some large bitmap of emoji images, and the app has somehow gotten the mapping of which emoji corresponds to what range of pixels in that bitmap all messed up

nikomatsakis (Feb 14 2020 at 15:09, on Zulip):

e.g. pasted image

nikomatsakis (Feb 14 2020 at 15:09, on Zulip):

@Wesley Wiser that's awesome!

nikomatsakis (Feb 14 2020 at 15:10, on Zulip):

OK, let's get started

nikomatsakis (Feb 14 2020 at 15:10, on Zulip):

First off, let's discuss which weeks we have available to us

pnkfelix (Feb 14 2020 at 15:11, on Zulip):

oh good point

nikomatsakis (Feb 14 2020 at 15:11, on Zulip):

/poll Click on a week if you will be ABSENT OR UNAVAILABLE

nikomatsakis (Feb 14 2020 at 15:11, on Zulip):

I made a poll, but the main thing is that I won't be here Feb 21 :)

nikomatsakis (Feb 14 2020 at 15:12, on Zulip):

I'm visiting family that day

nikomatsakis (Feb 14 2020 at 15:12, on Zulip):

But if others want to register their (un)availability, please do so

nikomatsakis (Feb 14 2020 at 15:12, on Zulip):

(Of course, I'm fine with people discussing things on Feb 21, I just can't lead or participate, so some topics may be better than others)

nikomatsakis (Feb 14 2020 at 15:13, on Zulip):

Next thing I would say is are these topics in the above list we can quickly rule out?

pnkfelix (Feb 14 2020 at 15:13, on Zulip):

I'm closing on the purchase of a house on March 6th. I suspect I won't be available during this time slot on that day.

nikomatsakis (Feb 14 2020 at 15:14, on Zulip):

Possible candidates

nikomatsakis (Feb 14 2020 at 15:14, on Zulip):

The llvm.side-effect change has had some recent deveopments

simulacrum (Feb 14 2020 at 15:14, on Zulip):

Well "recent" - we knew of them before, but didn't know we did

centril (Feb 14 2020 at 15:14, on Zulip):

the sideeffect change seems to be going nowhere; I was hoping the soundness hole would be fully and irrevocably fixed by now.

nikomatsakis (Feb 14 2020 at 15:15, on Zulip):

Right, we knew, but we forgot:)

nikomatsakis (Feb 14 2020 at 15:15, on Zulip):

Or something

centril (Feb 14 2020 at 15:15, on Zulip):

I can't say I agree with "forgot" personally.

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

centril said:

the sideeffect change seems to be going nowhere; I was hoping the soundness hole would be fully and irrevocably fixed by now.

what? At least data has been gathered

nikomatsakis (Feb 14 2020 at 15:15, on Zulip):

I do think we're making progress, indeed

centril (Feb 14 2020 at 15:15, on Zulip):

@pnkfelix well we were going to merge it, but then we didn't

nikomatsakis (Feb 14 2020 at 15:15, on Zulip):

The question is: would it be fruitful to have a discussion about this? What are the next steps otherwise?

pnkfelix (Feb 14 2020 at 15:16, on Zulip):

we merged it under a flag, as planned

pnkfelix (Feb 14 2020 at 15:16, on Zulip):

okay right, this is not venue for this discussion, sorry.

nikomatsakis (Feb 14 2020 at 15:16, on Zulip):

Yeah, let's just try to decide do we want to talk about this in a design meeting

simulacrum (Feb 14 2020 at 15:16, on Zulip):

I think there's not much that needs sync discussion

simulacrum (Feb 14 2020 at 15:16, on Zulip):

There are certainly thorny questions to answer and design work though

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

Well, can I ask what would be under discussion?

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

I.e. the question of "what are the goals to meet" seems like a question that different people have different answers to

simulacrum (Feb 14 2020 at 15:17, on Zulip):

That's true.

nikomatsakis (Feb 14 2020 at 15:17, on Zulip):

@simulacrum wrote this

I think the design meeting, or whatever process we choose, is going to primarily have to answer these questions:

nikomatsakis (Feb 14 2020 at 15:18, on Zulip):

Meta note, I'm regretting starting with this topic -- maybe we should have started with triage? That had the most votes, and is perhaps easier to figure out

nikomatsakis (Feb 14 2020 at 15:18, on Zulip):

Maybe we should table this and schedule triage? :)

simulacrum (Feb 14 2020 at 15:18, on Zulip):

Yes, seems good

nikomatsakis (Feb 14 2020 at 15:18, on Zulip):

And then return to side-effect after that

centril (Feb 14 2020 at 15:18, on Zulip):

we have 3 slots though; so we can plan all 3 meetings?

pnkfelix (Feb 14 2020 at 15:19, on Zulip):

(it might be premature to plan some of these meetings.)

centril (Feb 14 2020 at 15:19, on Zulip):

(well 4...)

Wesley Wiser (Feb 14 2020 at 15:19, on Zulip):

Two of the slots have one person unavailable

nikomatsakis (Feb 14 2020 at 15:19, on Zulip):

@centril I meant that we first place the triage meeting, and then return to the question of whether to use one of the 2 remaining slots for side-effect

nikomatsakis (Feb 14 2020 at 15:19, on Zulip):

I think it's clear that we want to do triage, in particular

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

aha, ok, sgtm

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

so the main question is: should we do it on the day when everyone's around, or prefer the week when I am absent :)

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

I think @pnkfelix should definitely be here

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

28th seems like a good day

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

I agree.

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

To clarify the proposal, this is about meeting right? Not triage of incoming issues? I felt a bit uncertain

nikomatsakis (Feb 14 2020 at 15:21, on Zulip):

correct

nikomatsakis (Feb 14 2020 at 15:21, on Zulip):

specifically

nikomatsakis (Feb 14 2020 at 15:21, on Zulip):

Proposal

At the broadest level, discuss the following questions:

centril (Feb 14 2020 at 15:21, on Zulip):

wait... I think I said something wrong here; I think the type system thing would be best on the 28th (in particular to have all the essential people there)

nikomatsakis (Feb 14 2020 at 15:21, on Zulip):

Hypothesis

More specifically, we would like feedback on the following hypothesis:

nikomatsakis (Feb 14 2020 at 15:22, on Zulip):

(there are more details in the hackmd)

nikomatsakis (Feb 14 2020 at 15:22, on Zulip):

centril said:

wait... I think I said something wrong here; I think the type system thing would be best on the 28th

ok yes I misunderstood what you meant :)

centril (Feb 14 2020 at 15:22, on Zulip):

(fwiw, I think I misspoke :P)

nikomatsakis (Feb 14 2020 at 15:23, on Zulip):

I don't necessarily have to be here for triage meeting, even though I proposed it :)

centril (Feb 14 2020 at 15:23, on Zulip):

My suggestion: 1st slot: Triage, 2nd slot: Type system

nikomatsakis (Feb 14 2020 at 15:23, on Zulip):

(to me it feels pressing to figure this kind of stuff out though)

centril (Feb 14 2020 at 15:23, on Zulip):

3rd: TBD

centril (Feb 14 2020 at 15:24, on Zulip):

@nikomatsakis I got the impression from @pnkfelix's note that things seem to be going well?

centril (Feb 14 2020 at 15:24, on Zulip):

or at least we're doing what we're supposed to

nikomatsakis (Feb 14 2020 at 15:24, on Zulip):

Well, I think we're expending more effort than we need to, if these are our goals.

nikomatsakis (Feb 14 2020 at 15:25, on Zulip):

but I certainly agree that the meeting has trended towards doing what I wrote (which is part of the motivation here)

nikomatsakis (Feb 14 2020 at 15:25, on Zulip):

I am also open to being told I'm mistaken

pnkfelix (Feb 14 2020 at 15:25, on Zulip):

There is a question of whether a design meeting is the right venue to confirm the hypotheses

pnkfelix (Feb 14 2020 at 15:26, on Zulip):

We maybe should do that over email first... ?

nikomatsakis (Feb 14 2020 at 15:26, on Zulip):

In what way?

nikomatsakis (Feb 14 2020 at 15:26, on Zulip):

But yeah I guess there is question of whether to discuss the hypotheses or skip to the conclusions :)

nikomatsakis (Feb 14 2020 at 15:27, on Zulip):

i.e., if we agree with those hypotheses, are there implications?

pnkfelix (Feb 14 2020 at 15:27, on Zulip):

its sort of a question of whether we expect to get involvement from all of, or the right members of, the team

pnkfelix (Feb 14 2020 at 15:27, on Zulip):

at a synchronous meeting

pnkfelix (Feb 14 2020 at 15:27, on Zulip):

the more targeted design meetings, people can sort of self-select

pnkfelix (Feb 14 2020 at 15:28, on Zulip):

but this question of "what are the goals of the triage meeting" seems both amorphous, but also the kind of question where we may want/need to get input from everyone

pnkfelix (Feb 14 2020 at 15:28, on Zulip):

("go around the room", so to speak...)

nikomatsakis (Feb 14 2020 at 15:28, on Zulip):

Makes sense. I'm open to suggestions =) I think a fair share of the buden falls on you here (but by no means all), so I'm interested to follow your lead.

pnkfelix (Feb 14 2020 at 15:28, on Zulip):

and while i do think that the design meetings are well attended (perhaps even better attended than the triage meeting?)

nikomatsakis (Feb 14 2020 at 15:28, on Zulip):

I was definitely specifically hoping to iterate on these questions:

nikomatsakis (Feb 14 2020 at 15:29, on Zulip):
nikomatsakis (Feb 14 2020 at 15:29, on Zulip):

(copied from the hackmd)

nikomatsakis (Feb 14 2020 at 15:30, on Zulip):

anyway, if you'd rather hold off until next planning cycle, we can do that. or schedule for later in the cycle (e.g., feb 28) to give time for an e-mail conversation first.

nikomatsakis (Feb 14 2020 at 15:30, on Zulip):

I'd prefer we host the discussion somewhere public with a record though, like internals

nikomatsakis (Feb 14 2020 at 15:30, on Zulip):

or zulip :)

pnkfelix (Feb 14 2020 at 15:30, on Zulip):

nikomatsakis said:

I'd prefer we host the discussion somewhere public with a record though, like internals

ah this is true

centril (Feb 14 2020 at 15:31, on Zulip):

(btw, "catastrophic", as phrased, is an exceedingly high bar to meet)

pnkfelix (Feb 14 2020 at 15:31, on Zulip):

Or maybe low bar

pnkfelix (Feb 14 2020 at 15:31, on Zulip):

(i.e. is it "easy" to avoid catastrophe?)

pnkfelix (Feb 14 2020 at 15:31, on Zulip):

or maybe that's your point

centril (Feb 14 2020 at 15:32, on Zulip):

I'm just saying it sounds very very serious

pnkfelix (Feb 14 2020 at 15:32, on Zulip):

its a high bar for categorization and thus a low bar for avoidance

pnkfelix (Feb 14 2020 at 15:32, on Zulip):

(hopefully)

nikomatsakis (Feb 14 2020 at 15:32, on Zulip):

OK, I'd like to reach a decision :)

pnkfelix (Feb 14 2020 at 15:32, on Zulip):

So maybe I can just ask directly: I think there are T-compiler members missing right now

nikomatsakis (Feb 14 2020 at 15:32, on Zulip):

(not about which word, I'm open to picking other words, catastrophic may indeed be too strong)

pnkfelix (Feb 14 2020 at 15:32, on Zulip):

e.g. I don't think @eddyb is here

pnkfelix (Feb 14 2020 at 15:33, on Zulip):

(and there are others like @nagisa that I think are absent and would be absent from any date we schedule)

pnkfelix (Feb 14 2020 at 15:33, on Zulip):

so the question is: Would it be a catastrophic mistake to schedule the triage meeting now, without their input, and likely at a time that they would not attend?

nikomatsakis (Feb 14 2020 at 15:34, on Zulip):

I don't think so

centril (Feb 14 2020 at 15:34, on Zulip):

(I see what you did there :P)

nikomatsakis (Feb 14 2020 at 15:34, on Zulip):

In part because I don't know that they specifically would be heavily participating in the process that results

pnkfelix (Feb 14 2020 at 15:34, on Zulip):

(i suspect it would not be catastrophic. but that's just me setting low bars for our decision making here)

nikomatsakis (Feb 14 2020 at 15:34, on Zulip):

like I think @eddyb tends to be doing more of the "fixing bugs" than "Tracking which bugs need to be fixed" sort of work :)

nikomatsakis (Feb 14 2020 at 15:35, on Zulip):

but I mean part of this is just "sync discussions are hard", idk.

pnkfelix (Feb 14 2020 at 15:35, on Zulip):

right. okay. given that viewpoint, I withdraw my suggestion that we delay or use email to get broader input, etc.

pnkfelix (Feb 14 2020 at 15:35, on Zulip):

"it will be fine"

nikomatsakis (Feb 14 2020 at 15:36, on Zulip):

So at this point we have two suggestions:

nikomatsakis (Feb 14 2020 at 15:37, on Zulip):

Maybe the question here hinges a bit on type system integration and what that meeting would actually be about

nikomatsakis (Feb 14 2020 at 15:37, on Zulip):

I was tempted to close it before this meeting so as to re-open with more details

pnkfelix (Feb 14 2020 at 15:38, on Zulip):

I suspect I could miss the type system integration meeting

nikomatsakis (Feb 14 2020 at 15:38, on Zulip):

Yeah, so another option would be

pnkfelix (Feb 14 2020 at 15:38, on Zulip):

It would be nice to be there, but I'd prefer to ensure niko is at the triage discussion meeting

nikomatsakis (Feb 14 2020 at 15:38, on Zulip):

I think that Mar 6 would probably be "review the roadmap for traits/chalk work"

nikomatsakis (Feb 14 2020 at 15:39, on Zulip):

preparing that roadmap is precisely the sprint task I care the msot about

centril (Feb 14 2020 at 15:39, on Zulip):

What about the 21th?

nikomatsakis (Feb 14 2020 at 15:39, on Zulip):

so i'm sort of happy to have some incentive to do it, but it will also take a bit of time

nikomatsakis (Feb 14 2020 at 15:39, on Zulip):

and I expect to force @pnkfelix to rubber duck with me a bit anyway ;)

nikomatsakis (Feb 14 2020 at 15:39, on Zulip):

centril said:

What about the 21th?

that was the "XXX", so presently unassigned

nikomatsakis (Feb 14 2020 at 15:40, on Zulip):

Let's say this then (unless anyone strongly objects):

nikomatsakis (Feb 14 2020 at 15:40, on Zulip):

I'll give 2 minutes for folks to raise objections while Derek and the Dominoes complete this absurdly long drum solo I'm currently listening to (ah the 70s..or whenever they were playing..)

nikomatsakis (Feb 14 2020 at 15:42, on Zulip):

Remaining two topics are:

I think compiler-team#161 was kind of my baby and (as I said before) I think it's not quite ready, in fact I think I will close, so the real question is -- discuss compiler-team#177?

nikomatsakis (Feb 14 2020 at 15:42, on Zulip):

(Also, the drum solo did actually just end with a burst of the other musicians coming in, very epic timing)

nikomatsakis (Feb 14 2020 at 15:43, on Zulip):

I'm torn because I sort of like not scheduling every week to give ourselves some time to breathe :) I don't know how we will find answers to those questions that @simulacrum raised though.

simulacrum (Feb 14 2020 at 15:43, on Zulip):

I don't know that I personally have time before the 21st to prepare an agenda, but I don't know exactly that we need more than my couple questions

simulacrum (Feb 14 2020 at 15:44, on Zulip):

I feel like it's pretty clear that there's also a divergence of caring about this soundness hole

nikomatsakis (Feb 14 2020 at 15:44, on Zulip):

Yeah, the main thing left to do I would think is to experiment with trying to find ways to reduce the overhead -- and that brings back your question of how much energy do we wish to invest in that

nikomatsakis (Feb 14 2020 at 15:44, on Zulip):

simulacrum said:

I feel like it's pretty clear that there's also a divergence of caring about this soundness hole

agreed :)

simulacrum (Feb 14 2020 at 15:45, on Zulip):

(which feeds into the energy question)

nikomatsakis (Feb 14 2020 at 15:45, on Zulip):

I honestly don't know how much I care about this hole in particular.

pnkfelix (Feb 14 2020 at 15:46, on Zulip):

I think we'd all like it resolved; its just a question of how much we're each willing to pay (in compile-time, right? That's the main cost we've identified, not end performance of executable, right?)

centril (Feb 14 2020 at 15:46, on Zulip):

/me feels strongly that we have a global safety promise, and that should come before compile times (which we have promised nothing about).

simulacrum (Feb 14 2020 at 15:46, on Zulip):

I will say that I also don't know that I care much, beyond giving people a way to do infinite loops (e.g. for panic handling or whatever) that doesn't cause UB. But even there I don't care that much

nikomatsakis (Feb 14 2020 at 15:47, on Zulip):

pnkfelix said:

I think we'd all like it resolved; its just a question of how much we're each willing to pay (in compile-time, right? That's the main cost we've identified, not end performance of executable, right?)

correct -- I guess that's a question. I assumed this was making the compiler run more slowly, but is it just making LLVM take longer? (i.e., more LLVM bitcode?)

simulacrum (Feb 14 2020 at 15:47, on Zulip):

It's all llvm

simulacrum (Feb 14 2020 at 15:47, on Zulip):

Compiler barely does anything

pnkfelix (Feb 14 2020 at 15:47, on Zulip):

clearly we just need to move to cranelift.

nikomatsakis (Feb 14 2020 at 15:47, on Zulip):

ok ok that's interesting

pnkfelix (Feb 14 2020 at 15:48, on Zulip):

(next thing you know, @bjorn3 will be telling us that cranelift also defines infinite loops as UB)

bjorn3 (Feb 14 2020 at 15:48, on Zulip):

Nope, not that I know of

nikomatsakis (Feb 14 2020 at 15:49, on Zulip):

Well, let's try to reach a discussion. I won't be here if we do discuss it. It seems like there's something of a "values debate" at play here.

I wonder if one action we could take instead is:

i.e., is there someone interested in exploring more advanced ways to limit the "extra bitcode" we wind up making?

bjorn3 (Feb 14 2020 at 15:49, on Zulip):

https://cranelift.readthedocs.io/en/latest/compare-llvm.html#undefined-behavior

nikomatsakis (Feb 14 2020 at 15:49, on Zulip):

(but of course somebody would have to write it)

simulacrum (Feb 14 2020 at 15:51, on Zulip):

More advanced ways require either global analysis or something like it I think. Not sure. Pretty much doing anything via side effect is going to have a high cost I think, if we aim for a 100% solution

pnkfelix (Feb 14 2020 at 15:51, on Zulip):

I think a post like that is a better first step than jumping straight to a design meeting

simulacrum (Feb 14 2020 at 15:51, on Zulip):

I may have enough time to write such a post, though I do not have time to evaluate feedback much I think

pnkfelix (Feb 14 2020 at 15:51, on Zulip):

especially since I don't think anyone present has the resources to throw at this problem

simulacrum (Feb 14 2020 at 15:51, on Zulip):

So maybe asking for help is not quite right :)

pnkfelix (Feb 14 2020 at 15:52, on Zulip):

(and without resourcing, I think a design meeting is premature)

nikomatsakis (Feb 14 2020 at 15:52, on Zulip):

maybe we don't do that part quite

nikomatsakis (Feb 14 2020 at 15:52, on Zulip):

I think it'd be useful just document what's been tried

simulacrum (Feb 14 2020 at 15:52, on Zulip):

Anyway, let's make a goal of getting that post written and we can revisit next planning meeting

simulacrum (Feb 14 2020 at 15:52, on Zulip):

I will draft an initial version

nikomatsakis (Feb 14 2020 at 15:53, on Zulip):

that works for me.

simulacrum (Feb 14 2020 at 15:53, on Zulip):

Is anyone opposed?

centril (Feb 14 2020 at 15:54, on Zulip):

/me has to run; the post seems fine; ciao

pnkfelix (Feb 14 2020 at 15:55, on Zulip):

okay great so this means we won't have a friday meeting next week, right?

nikomatsakis (Feb 14 2020 at 15:55, on Zulip):

correct

nikomatsakis (Feb 14 2020 at 15:55, on Zulip):

thanks @simulacrum for volunteering to write the post

nikomatsakis (Feb 14 2020 at 15:55, on Zulip):

I wonder if we can prompt LLVM to offer us something

nikomatsakis (Feb 14 2020 at 15:55, on Zulip):

:)

nikomatsakis (Feb 14 2020 at 15:56, on Zulip):

This is likely due to generating tons of side effect calls (at least one per function entry), which LLVM today also does not collapse into one after inlining. That generates horribly bloated IR, likely slowing down LLVM passes: rust-lang/rust#68402 (comment).

if indeed that is true, you could imagine some kind of "collapsable side-effect marker" or something

nikomatsakis (Feb 14 2020 at 15:56, on Zulip):

that might not even be so horribly complex for them

simulacrum (Feb 14 2020 at 15:56, on Zulip):

Maybe though it's unclear that would solve the problem :)

Maybe it would. I know nothing of llvm myself

nikomatsakis (Feb 14 2020 at 15:57, on Zulip):

Yeah I don't know either :)

bjorn3 (Feb 14 2020 at 15:57, on Zulip):

How hard would it to convince the LLVM devs to not treat infinite loops as UB, but only when a certain setting is enables? That way LLVM can still optimize infinite loops in C code without problems.

nikomatsakis (Feb 14 2020 at 15:57, on Zulip):

very hard

nikomatsakis (Feb 14 2020 at 15:57, on Zulip):

but

simulacrum (Feb 14 2020 at 15:57, on Zulip):

That has been under discussion within llvm for years now I think and without major progress

nikomatsakis (Feb 14 2020 at 15:57, on Zulip):

well, to be clear, IANAL -- I am not an llvm hacker :) --

nikomatsakis (Feb 14 2020 at 15:57, on Zulip):

but I think that the idea of a global switch is hard for them beacuse, in short, they don't know where the assumptions are baked in, and it's a global audit, etc

nikomatsakis (Feb 14 2020 at 15:58, on Zulip):

but those optimizations do have to account for side-effects from inline asm already etc

simulacrum (Feb 14 2020 at 15:58, on Zulip):

Or at least that's my impression from comments by rkruppe(?)

pnkfelix (Feb 14 2020 at 15:58, on Zulip):

e.g. discussion amongst llvm devs

nikomatsakis (Feb 14 2020 at 15:58, on Zulip):

so I'm not sure if they've considered the idea of "compilers flip that switch not in global configuration but by adding some kind of lightweight side-effect thing"

nikomatsakis (Feb 14 2020 at 15:58, on Zulip):

but yeah I'm really just totally speculating here

nikomatsakis (Feb 14 2020 at 15:58, on Zulip):

@Hanna Kruppe would be a much more knowledable source

bjorn3 (Feb 14 2020 at 15:59, on Zulip):

Ok, thanks.

nikomatsakis (Feb 14 2020 at 15:59, on Zulip):

I guess in particular @simulacrum your post might enable us to go to the LLVM devs and say

nikomatsakis (Feb 14 2020 at 15:59, on Zulip):

"why do YOU think this is slower? do you think you have ideas on how we could do something like this that woudln't be slower?"

simulacrum (Feb 14 2020 at 15:59, on Zulip):

Makes sense yeah

nikomatsakis (Feb 14 2020 at 16:00, on Zulip):

i.e., not a "help us internet!" but a more targeted appeal :)

nikomatsakis (Feb 14 2020 at 16:00, on Zulip):

anyway great!

nikomatsakis (Feb 14 2020 at 16:00, on Zulip):

I'll go setup the calendar invites and things

simulacrum (Feb 14 2020 at 16:00, on Zulip):

Okay, well, we're at time, but I think we achieved what we wantee

nikomatsakis (Feb 14 2020 at 16:00, on Zulip):

thanks all for a productive meeting

nikomatsakis (Feb 14 2020 at 16:00, on Zulip):

plus Derek and the Dominoes just finished their set

nikomatsakis (Feb 14 2020 at 16:02, on Zulip):

FINAL SCHEDULE

nikomatsakis (Feb 14 2020 at 16:03, on Zulip):

ALSO RESOLVED

nagisa (Feb 14 2020 at 16:14, on Zulip):

@pnkfelix @nikomatsakis ~this time is usually when I begin preparing some dinner, walking dog etc. so I generally tend to miss the meetings by very slightly. But I do echo the sentiment that whatever time is picked for a sync meeting, _somebody_ is likely to be missing it routinely. I would generally be much better about making myself available for the less… optional… sync-ups.

nikomatsakis (Feb 14 2020 at 16:21, on Zulip):

I created calendar invites for the two meetings. I also invited various people that seemed appropriate -- the meetings are public, but it's sometimes helpful to have an invitation for your calendar. If you'd like me to invite you to any of the events, ping me.

Last update: Jun 07 2020 at 10:35UTC