Stream: t-compiler/wg-meta

Topic: LLVM WG


nikomatsakis (Feb 27 2019 at 20:37, on Zulip):

Hey @nagisa @Nikita Popov -- we raised the idea of a WG oriented around

Still interested in pursuing this? If so, there are instructions in the compiler-team repository for how to setup the basic template, and I can create the Zulip stuff.

nikomatsakis (Feb 27 2019 at 20:37, on Zulip):

@Nikita Popov, this may be the first you're hearing of this =) so we can definitely talk it over

nagisa (Feb 27 2019 at 20:38, on Zulip):

I’m currently very busy with large changes in my (work) life, so I’m not going to pursue that for quite a while.

nagisa (Feb 27 2019 at 20:39, on Zulip):

OTOH if @Nikita Popov expresses their interest, then maybe? Also cc @rkruppe.

nagisa (Feb 27 2019 at 20:40, on Zulip):

(zulip said something about @rkruppe not being subscribed to the stream and therefore not being notified... and then presented me a button to subscribe them on my own volition?)

nikomatsakis (Feb 27 2019 at 20:55, on Zulip):

@nagisa Yes, that's part of how Zulip works. If @rkruppe is not subscribed to the stream, they wouldn't see your message. I just subscribed them. =)

nikomatsakis (Feb 27 2019 at 20:55, on Zulip):

I think the intention is that you will subscribe people if you are asking them a question, but not if you are just mentioning them in passing.

nikomatsakis (Feb 27 2019 at 20:55, on Zulip):

That reminds me that I want to schedule some sort of Zulip Team meeting and try to write-up these practices...

nikomatsakis (Feb 27 2019 at 20:55, on Zulip):

...but that's for another stream/topic ;)

nikomatsakis (Feb 27 2019 at 20:56, on Zulip):

I’m currently very busy with large changes in my (work) life, so I’m not going to pursue that for quite a while.

I'm happy to have it on a kind of "shortlist" of ideas for the future -- I was thinking it'd be great to make such a list.

rkruppe (Feb 27 2019 at 21:40, on Zulip):

weird that I can't even get a notification without being subscribed to the whole stream, did not expect that when I unsubscribed

rkruppe (Feb 27 2019 at 21:43, on Zulip):

I, too, am too tied up by other things to pitch in. But in my biased opinion, this is indeed an area where it would make sense to focus more contributions, so I'd be very happy to see such a WG happen and would gladly offer braindumping/mentoring/etc. as time permits

nikomatsakis (Feb 28 2019 at 16:28, on Zulip):

I think @rkruppe / @nagisa the "minimal initial investment" that probably makes sense for a WG is a set of leaders / mentors who are willing to do the following:

I don't think it has to be a super high investment, but that first step is sort of key. It seems ok to me if we create a WG, do some work to setup "mentorable bugs", and then step back to steep if a few folks show up. We can avoid creating too many bugs in order to avoid a flood. =) If nobody shows up, that's ok, we're no worse off -- better off, even, as we have some mentorable bugs in the queue for later.

But if writing decent mentoring instructions on 1-3 bugs is too much work, it's probably not worth it.

nagisa (Feb 28 2019 at 16:32, on Zulip):

I guess first order of business (if we want a list of mentorable issues) is then to find issues that stem from LLVM we want to fix the most. loop {} is an obvious candidate I had in mind when proposing the WG, but that is something I already started working on... I guess I could write up some instructions mayhaps?

nikomatsakis (Feb 28 2019 at 16:46, on Zulip):

loop { } would be great, yeah. I don't know how detailed the mentoring instructions have to be -- even outlining some high-level steps and asking people to find you for more in depth comments would be better than nothing

davidtwco (Feb 28 2019 at 16:52, on Zulip):

I know that I'd definitely be interested in helping out with this working group - I have zero experience with LLVM but with some high level instructions and a pointer or two about which areas of the code I should be looking at then I'd be happy to dive in and see what I can come up with.

nagisa (Feb 28 2019 at 16:59, on Zulip):

Either way, please give me like a month to get my contracting to lift off

nagisa (Feb 28 2019 at 17:00, on Zulip):

so that I don’t have to deal with mentoring people in addition to having to figure out taxes, laws, bookkeeping and all that other stuff that comes along with the pleasure of being own’s boss.

Nikita Popov (Feb 28 2019 at 18:20, on Zulip):

I'd be happy to help with the LLVM WG. We have quite a lot of open LLVM related bugs, and I think many should be approachable to newcomers with some triage.

Nikita Popov (Feb 28 2019 at 18:21, on Zulip):

@nagisa For loop { }, do you have a workaroud on the rust side in mind, or fixes on the LLVM side?

nagisa (Feb 28 2019 at 18:22, on Zulip):

LLVM already has @llvm.sideeffect so it would be possible to fix it on the Rust side, but it seems to me that fixing it on LLVM side in the first place is easier than making Rust emit @llvm.sideeffect.

nagisa (Feb 28 2019 at 18:23, on Zulip):

specifically the problem appears to be caused specifically by LLVM functionattr pass adding readnone, readonly or writeonly onto non-terminating functions...

nagisa (Feb 28 2019 at 18:24, on Zulip):

that is the part that needs to be fixed and it feels like a fairly easy thing to do, except, I am having a hard time figuring out how to prevent LLVM from doing that.

nagisa (Feb 28 2019 at 18:24, on Zulip):

(admittedly I didn’t look too hard into it since I did the initial investigation)

Nikita Popov (Feb 28 2019 at 18:30, on Zulip):

And once those attributes are there other memory accesses get reordered around calls?

nagisa (Feb 28 2019 at 18:32, on Zulip):

we should split it out into a separate stream.

dlrobertson (Feb 28 2019 at 18:41, on Zulip):

Seems like I'm a bit late to the party, but I'd also be interested in this WG. I'm new to rustc, but do have some experience with LLVM.

dlrobertson (Feb 28 2019 at 18:42, on Zulip):

What would the difference be between this WG and WG-codegen?

nagisa (Feb 28 2019 at 18:43, on Zulip):

LLVM WG is specifically aimed at work upstream, whereas WG-codegen deals with lowering-to-llvm.

nagisa (Feb 28 2019 at 18:44, on Zulip):

An example of such work would be working with upstream to get features we need developed and landed.

dlrobertson (Feb 28 2019 at 18:56, on Zulip):

:+1: thanks for the clarification. I'm just a casual contributor to rustc, so that plus my newbie-ness probably means I will not be much help, but I'm definitely interested in this WG

nikomatsakis (Feb 28 2019 at 19:01, on Zulip):

OK seems like there's definitely enough interest that this should happen -- who wants to open the PR to create the group?

nikomatsakis (Feb 28 2019 at 19:01, on Zulip):

(@nagisa, perhaps?)

nagisa (Feb 28 2019 at 19:01, on Zulip):

(nagisa, perhaps?)

okay, lemme see if I can find the repository that has instructions :slight_smile:

Nikita Popov (Feb 28 2019 at 20:21, on Zulip):

I think WG-codegen is dead (and wasn't really particularly alive to start with...)

nikomatsakis (Feb 28 2019 at 21:02, on Zulip):

WG-codegen is dead, love live WG-llvm

nikomatsakis (Feb 28 2019 at 23:02, on Zulip):

Created #t-compiler/wg-llvm, although it's not the most creative name :)

Last update: Nov 11 2019 at 22:50UTC