Rust Programming Language Language Design Team, 6 Jun 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Lang Team Meeting 2019.06.06

Description

Main topic: std-aware cargo

A

I'm gonna start the recording, I guess. Are we.

B

Not to make any decisions but like I, think we're maybe about what we're gonna get um I think both is coming and I. Think Justin Josh said: they're traveling I I thought they were troubling to soon be able to travel.

A

Let's do this, let's start going through things more people show.

B

So, just just to be a quick recap from last week's action I, please. We have some reports that we need to do.

A

Before we go so for those of you from the cargo team, usually we start with some light triage, which probably will be boring for you. Oh yes, we show that and then we'll get to the standard aware of cargo should I switch the order of things. We could do that that won't work for me. If we want to start with discussing standard or cargo and then switch back I, don't expect well, I, don't know, let's try that all right, I think.

C

I'm a bit in order to sure we actually get to the yet you.

A

Know I was going to say, let's try through 30 minutes and then we'll see where we are I mean.

A

But maybe one of you would like to talk about the possible I.

B

Don't know how many people I read it, but it was a while ago yeah, maybe like.

D

I can also give a very dated summary at this point. Eric I think you're reading it very recently, though, you probably have the most up to date in that case, but if you don't want to give us some wreckage, I could slowly tried.

E

D

Right, so this RFC is quite old at this point and I believe pretty eated to the point that a lot of things have changed, just the general background and kind of where it wants to go from what I understand the very high level motivation. Is it kind of sucks that cargo doesn't really understand that every crate in the world depends on stood or court or whatever sister said? This is your case.

D

We currently have things like proc macro as well, and so this is particularly problematic in the sense that we cannot recompile the center library for things like debug assertions for more debug info for Less debug info for I want to turn off the battery support.

D

I want to have a completely do a texture, because I'm, but I want to tweak some cogent flags like I, want to enable I'm, webassembly and I want to enable the Atomics feature in the snare library there's a whole slew of reasons why we want to recompose I can go into more details with what exactly.

C

D

There, and so the goal of this RFC is to basically enable that in one form or another, and so it's largely a carbon related thing as far as I know, as I suspected with my touch language stuff as well in terms of crates and sisters and whatnot, but that's sort of the general thrust and then the actual details of the RFC I think have changed a lot over time. It's sort of unclear like what's in cargo tamil, what exactly is inferred where? What exactly? When does cargo recomposed any library?

D

When is it not recomposed in your library and so I? Think, at least from my personal opinion, we probably want to keep today's system by default, which is if I type Karl go to Linux I'm, not going to recompile the standard library just keep using whatever I previously downloaded. But somehow, when something reasonable happens, maybe we do switch to the mode compiling things is.

B

That we recently like.

B

D

I think that so number one it would be a big compile time we hit it's, not huge. The standard library actually compiles pretty quickly. It is noticeable for like a hello world, but I also think it's an issue in terms of just functionality. Sometimes we have like building Lib batteries for standard stuff like having a C compiler one way or another.

D

We don't typically need that, but users don't always have a fully configured environment that is guaranteed to compile those kinds of things and so for those use cases, it's often very useful for us to build it ourselves and then ship, something that we know works to users or something like that. So that's a much more esoteric, the in the build time which is and.

F

Then back trace is a point specifically like the ultimate plan is to remove that right and replace it with arrests only in dependency. It's.

D

True, that's probably like in the far future going to happen, but we still have the thing with compiler, artigue and but again.

G

D

We could certainly say for the time being and I would guess the next year. So we're gonna have us some C code, which is either more performant or just make.

A

Sense things so, there's the first of all I think earlier you were saying there was fair amount of lis interaction. You want to outline some already had in mind, but I did have a few other questions too about features and stuff I'm, not sure how off to date, this cuz I was looking over my pronounce like there's some notion of the features like basically Pargo features applied to live standard, great.

D

So the idea here is that it's there's nothing stable if initially and we as a lips team would decide what features are stable, but the idea is. This would eventually allow for tweaking the function a little. Everyone really other, so removing binary blue via a number of mechanisms, switching implementations for whatever reason: okay,.

A

One thing I was wondering about was like.

A

Since you're here to answer my questions, they put some things like this. You know in the RC, but the one concern I had with this was long. If we take something that used to not require a feature and make it feature a pole, then that's the freakin change.

A

D

So this is a general cargo problem where, if you it's technically a breaking change too dearly, but anyway, the features is kind of a big change. For this exact reason where you could turn.

G

D

And so this is one where I think this is sort of an end goal really want to get to a system where we can do this, but we're not gonna stabilize this. Unless we actually are confident, we have a non-breaking thing. It's like. Maybe we just say that you can't turn off default features of core or, like a feature, has no default feature.

D

So something like that I'm like we that's a guarantee we provided, and so this is where, like there, this artist, he specifically was proposing a couple of phases where, like you, can't do this in the first stable phase, because we it.

A

Seemed like there we go like it's a problem in the ecosystem, but it's a huge problem Lipstein because we don't have our 2.02 to I. I would put it in for ice. Like that part constraints, I guess should be resolved. Some.

B

Basis, it seems to me that removing features isn't that problematic, because it's most like from a testing point of view, you simply comment out code, whereas with switching implementations, then you need to test those implementations and consume ibly have like specific builders for that? How can you keep up with the quality in terms of having.

B

Switching implementations and mostly possibly having like end times and great.

D

This is something the lives team we haven't barely discussed either for liqueur or the state library like we talked about this platform, abstraction layer idea or set of like something with features there, but nothing here was gotten create to the point that it's an actual proposal that we want to move forward with, and so like another example of why we would not want default features at all, especially for, like literally everyone.

D

That's another reason we're like we have a lot more in terms of features to talk about for the loop state perspective of what exactly do you want to stabilize.

B

To stabilize the target specification format to you, Jason fire thingy.

D

Yes, and from what I understand, I I think I'm, also I'm speaking I, like slightly mixed up, because I actually thought this was mostly about the really really old are sees. This is actually far more recent, but Green is back into mind. This is from discussions at the All Hands. So this is I. Haven't mentioned this in the library a lot, but this is also sort of equally applicable decor.

D

I think the main use case of this sort of see is specifically for cord specifically for the infinity core system, where those are the cases that we don't have targets in receipt yet, but LLVM has all the support, for example like there's some processor with some subset of features, and we don't have an official name for it, so the idea is sort of way to stabilize that so I don't have a ton of opinions of Jason target specification crazy, too so I, don't know how we're gonna quite stabilize that it's kind of like a compiler team thing as well.

D

B

The thing I'm wondering about is whether we are like tying ourselves too much. You know if we consider things like cream lift and if this indications.

B

D

And that's like, basically than one of the number one concerns about these target specifications, but I think it also goes further here in terms of not so much LOV em, but just the current day and age of the compiler see where the target specifications have been changing all the time. As we add random options for remove random options, we just stop interpreting some options, and so it's not so much for me, it's not so much an LVN. This is the compiler to question. So there's not really a library.

D

But for me this is a question of how like stabilization in general, as opposed to anything related to Olivia myself. Cuz I am actually is pretty pretty well divorced from this, and in most cases or some cases at least yeah.

D

But, and also to reiterate, this RFC was explicit explicitly written would especially when I talk with him about this was to propose various phases and so I think like it gets much more questionable as time goes on in terms of like whether we want to stabilize all of that. But the intention was that there should be some kernel subset, which is very useful to everyone. Where, maybe you don't you don't touch features at all? You don't touch about new targets. All you can do is on the standard library and liqueur on the stable channel.

D

You can recompile the standard library, at least with different cogent options, so like Panik, aboard optimizations, debug info and then paving forward for these later features like features so.

B

One question I have I'm, not really sure I understand is how does this interact with strapping and.

D

This is where every single compiler is paired, with exactly one copy of the standard library that it can compile, and so we can guarantee on stable that it can compile its own, stable and so cargo will basically just set the receive stripe, flag and say, and but cargo will do that, knowing that it's guaranteed to work, because it's the exact source code and so every channel is a company with its own source code, and so they can all compiled its own source code. So.

B

B

Carter would just assume.

D

That, whatever is at that location, is the right state of library is.

C

That in the RFC, some more.

D

Well, I should also say this is my that's my own personal opinion. So I don't want to start hashing slavery, and so that's.

A

The attack vector that's hypothesize, heard somebody edits that perfect part of those internal directories, so I I guess the way you would compile something with my seafood. Well.

B

505 folks already uses roughly will see bootstrap so basically to do that. If you, if you have something but I, feel like it's. Okay for us.

A

To just declare them you already so probably.

B

Right, basically, if you, if you do this, then we don't offer you. You have voided your wires, a difference to.

C

Be clear, my question was more: it wasn't about the the mechanism we used to enforce the consistent variant. That describe is one of the environment in the RFC and I now see the invariant he described, but the source code, the 1:1 mapping from the compiler version, the source code. It is actually the RFC itself. I shall make sure that was spelled out there.

A

So I guess this is just the matter of phase this Alex, but I was just thinking about how there's a lot of like laying items and so on they're, not in Lib core. Oh I have an idea about that. Actually, I.

B

Think some of the line items could be moved out from the core to to like really really tiny crate. That only concern himself with line and line items. So that would require introduction of some new line. Items filled with the four ends, but reduce them to be.

D

Clear the attention here is to not mess with blank items at all and so like, for example, today I could create an hour loop, unstable linking against core that's missing line items I just can't link that like, if you try and produce a dynamic library and executable your errors, saying like this Lang I'm, mystified or whatever, and so the idea is.

D

We don't want to touch that at all, or we can still enable that, like on, like some of this I think is also empowering the nightly users to do more or the idea is if we, if this eventually rides this table train.

D

This is sort of like the fundamental kernel of all nightly users and things like that, and so they might still be doing crazy, Netley things where they're touching my items are doing various aspects but they're building on the stable baseline of where recompile assign a library and then linking things in later and then otherwise. If you're just recap on the state of library plus liqueur, and you make it executable, although I'm sort of provided for you need a lot starch, the middle right.

A

I guess I mostly wanted to ask whether they're in like had been given a lot of thought or to like work or not. But it sounds like one.

B

Thing that I'm actually trying is to it appears to have long items work, basically in here, lowering all it's okay, if that's one.

D

Example, where I think it's maybe the best story today, is that I think with enough support and the embedded ecosystem. At this point, you, if you put if you have a crate with a panic, Handler and you have lib core, compile with pinnacle board, then you can link executables that only have low core that don't leave the standard library when you probably also grain, allocate or somewhere at some point.

D

But um if you compile lib core with panic on whines and you tried to create an executable, you just can't do that on stable, and so it's like only in this one compile mode. It's kind of secretive that now you can produce a stable executable, but if you compile it on a different mode, you can't produce an executable on stable and so I think.

D

The idea is there's some like atoms, which are defined by stood which are primarily allocator and the panic Handler and it's probably gonna, be a couple of the random ones, but um it's kind of like we try to work without it if possible. But it's not really well documented in what situations you do work without it or how you have to compile things of that point. If that makes sense, so.

A

That's actually yeah that's kind of what I was that's the concern I had, but I didn't know it. I mean that does mean this work may be more quality in some sense, but not really that like.

A

If we currently don't require a I guess, I would say: I currently consider it sort of a best effort that we try not to require more laying I understand necessary. But if it's really inconvenient, we don't try that hard. You know you're supposed to have them around, and this sounds like it would definitely be on mild stability. Hazard, query. It used to work without this link item and now doesn't it.

B

Speak like that I figured like the feelings. It would be good to write down like an example work where things might go wrong, so you know, like I, mean I'm using like a case where this happened. In the past, I mean it's central like no.

C

B

Like an example of just a situation where, where laziness could become instability hazards like the code example or something.

D

In terms of compelling implementation, this is it's basically only effects for as far as I know, laying items that are in a locker stood just anything above quarter because you may not have a locker stood and so like there is the ability to have an inherent like methods on stir, if you like Alec, and so the compiler just has to be ready used for like oh well. Now, maybe some crates don't like to Ella, but I can never sort of crates do, and so there could be like some interesting details in the compiler.

D

Like the other larger examples, though, I mean the panic and the panic handle is probably the biggest one where it has like random symbol, references and or compiler intrinsic for. If you have Pentagon line mood and that's why cuz core itself has to reference, those somehow have to be pulled in, but a lot of this isn't necessarily laying out on business more so like. How do we know? What's in library both on the compiler side flavor, so I was.

A

More concerned about things like you, don't have a company which you could do if you can supply your own court. Oh.

D

Ok, so that is like one of the final phases of this RFC, where you can like, basically edit the chord edit, lift standard and I'd like that. As far as I know, we literally always require a nightly compiler and so I I think I mean.

A

D

Stable version of this, you will always be compelling the exact source code and you'll, probably literally always require a core I. Think there's cereal use cases we're going to be stable, II producing a full restaurant that doesn't have look or right.

A

That's basically like itself, but another example would be things like f/32, runtime or whatever, which is in those dangers but I'm not so concerned. Examples I see at least don't seem that hard for the compiler to deal with I'm, not being there, because it's not like we just there are no methods. There are no methods.

B

A

It's more more, the core stuff, that's the biggest concern one.

B

One thing I'm wondering is whether it would be a good idea to spit this RFC to like target specification was one thing and like feature gating and sorry having features for the core ends. It's not library, there's a different thing are those like atomic things. We can ship.

D

In my opinion, I think that's what I recommend it as well, but Jesus was more of a fan of putting it all in one having explicit phases now.

H

Have the future work part of an RFC, so there might be a dedicated spot, for this is phase one and there are future pages in a way. There was something that Ric was latch to look at right.

B

I suppose I suppose one could merge the RFC and then split them into different practical issues. After doing, though I mean.

A

I said we would talk for about 30. Let me talk for about 22 minutes, I'm wondering in the last little bit is there it's just like. There hasn't been a lot of length een like really correlate team concerns here. Is there something we're missing anyone? Think of things like splitting? The RFC feels like a maybe it's good. Maybe it's not that it's.

B

Mostly most of the question of target features at Target specifications, I think the might impact us and the knife and bits perhaps.

D

It does sound like the the sense that I'm getting is that it free for now it is. It should be required that if you were on stable, you were linking and slick, or literally all rest programs for line items and that's like a baseline requirement that will always be required. I've.

A

Always assumed that was the point of the court.

E

I had a question about that. There was an RFC and Aaron had a blog post about how he would like to see sort of core an STD, merge and then used features to select exactly which things are enabled. Is that still any desire to have that or is it? Is it like core and STD are always going to be separate.

D

Right now, the intention is to not block us from that future, but we're not sort of actively pursuing it, because there's thorny questions that we are my.

E

Concern here is that this would sort of solidify that core is a dependency in cargo, and that would you know kind of cement. That distinction isn't.

B

It possible to have that you know how core is a dependency in cargo of them removed things from this done like it's. It uses the SVM, so it doesn't use those things when they see it chase it out.

F

Sure, sorry I'm not sure if we were just talking over each other, but I think that it's already, you know pretty clearly observable via, like the actual linking arcs and things that you give to rusty. When you compile programs that there is a distinction between those as crates and as like units of compilation right.

D

So this is one of the really thorny questions about. How are we actually gonna ship a feature like this in terms and not bring anyone, or this.

F

D

F

Sorry go ahead, Alex there's.

D

A lot of ideas about how, like maybe core, will just react, sport all stood, and so it dependency will go the other way, but you just can't really tell and so like. Maybe if you depend on corn cargo Tommy, we actually depend on student, but with less features or whatever it's all sort of magical and hand wavy and like I'm, not personally convinced the future will ever come about, but just for a lot of these reasons.

D

But I don't think that I wouldn't I easily say that this is not making it harder to have that feature come available, even though it's so.

F

So today, to give an example of the thing that I was talking about in know stood programs. We don't pass a linker org that allows rusty to find Lipsy, and so, if you, if you try and compile something for fuchsia with without that is that is no stood today it. It would fail the compile and in this world, where we it, where we had like merged the crates into one.

A

Presumably my building and elaborate like by having it this things are in there, but there are CFG doubters because most of the iron somehow I think.

E

I think like the idea was not to get rid of core or anything like that. It was actually to get rid of STD and make STD a sort of facade for everything. Underneath.

F

Right but we can't have multiple compilations of stood right in in the same program or in the same look like linked artifacts, hmm but we need to be able to link some objects without the standard library and without Lib C and not other objects.

B

B

In terms of who, which things should deal with this I, get the sense that it's mostly Fargo and then also the comparison.

B

That's me my impression.

B

I think the language team can probably remain that more possibly so we won't like actively work on this, and then you can seems like with obviously leaves effects happy in some way, but should I should I add another team I.

A

Don't know I, don't care lips, you can decide if they want to cook them still longer I feel like it. I feel happy and I be in this discussion. I, don't see a lot of yeah that doesn't where the point of a discussion was to feel make sure yeah all right should we do it switch to triage and, like these four people actually chewed out all right? No, you can stay or not thanks. Everybody Thanks.

B

A

B

So reviewing lost back to triage, reviewing lost with action items, so we are going to Nico is going to write a report about the balancing points. I haven't yet and like time vision, we also need to to figure out what the rules are and to make sense of this. For.

A

Just a minute I put this I would call the rebalancing coherence, I plan to do it, but I, don't this very particularly high priority, so it felt like I'd rather do sooner.

B

Yeah I agree the I'm working on the top there's imbalance. I almost have a cure, he's a ripe imports and do something about it today.

B

So moving on to, we can jump directly to infiltrate.

A

Yeah, we could talk about that, um so you had this on the radar for a long time. I think both was going to write her some responses. Nothing yet I also have been looking over it and I feel like I. Have my own personal. Take, which is I would like to see us be kind of on the overall plan and like to work out some examples. A little more I think we probably do know the overall plan. I just won't feel I.

A

Have it like all in my head in cash I actually thought the RFC had a pretty good summary. Don't when I read it reread it last night, so it seems like there could be grounds for some kind of meeting to talk it over in depth, but that's probably separate from quotes and specific concern. Although we could talk about that.

A

So basically I'd like to see this like more there's an opportunity for us to advertise to everyone. Here's what here's all the moving parts around infiltrate, kind of give the context that we saw off and feel the kids I think.

B

What we need to do is break up the tracking issue and close the old one and make a new one and then sort of like have a master tracking issue that links to those places. So we can track it better. Like a constant thing,.

B

I was going to do after merging, hopefully merging this artsy. It felt premature that so both.

A

Sigh have you had time to look at this? No ya know. One of the concerns that arose. To my mind was this that I don't at least I hadn't considered.

A

Currently we support well I guess this turns into X, which I feel like that's big things like this, where you have unused type parameters in this one and I guess this is what that would map to. But it didn't feel like that.

A

Not sure why I made a difference to me, but it wasn't like obviously that's should say they had a strange feeling to me and I'm trying to my finger on exactly mine, but um I know. We should at least include some examples in that claim. I.

B

Mean I'm sure of our court would be happy to.

A

I mainly try to decide if I'm like, but why. The thought was then that if we.

A

B

B

Both do you have any thoughts about. My overall know that we are discussing.

G

I still want to take time to think about it and just have not had time because I'm working on basically and it takes up all of my time.

G

A

Could ask this question.

A

A

Where do we put this relative to its weight? Do we want to do we want to the guest one question I had was whether you feel like the before having any group discussion. You would rather take time to think about on your own or whether you rather have some group discussion about it.

A

Life like that's a way to make time yeah.

G

I would like to think, but on my own I also feel like maybe like, like.

G

Basically, we could not block in putting it on resolve like.

G

Is the reason people keeping about this is because they want to make implementation changes to the nightly feature, and we could possibly look like do that ahead of making a final decision on this index. I think.

A

B

I close out this index, especially I, think Barker is concerned about too much work without like having a commitment from us, but absolutely it's like Alex. That's.

G

How much yeah it's not that hard to implement the students? It's a syntactic change right. So this is like mostly I.

F

Actually, disagree with that I think that there's meaningful change here that that is being undervalued because of the generics capture behavior, is different. It's any more.

F

F

In scope automatically rate but but this wonder, would it.

B

F

You to have multiple nested invocation, it uses of infiltrate Cree, which is a different thing and it D sugars to multiple existential types underneath right.

H

F

It doesn't along, though,.

A

Do you mean if this is specifically disallowed in northern France? No, it is yes, it is I just realized. Okay, let's look at it I believe it happy to be wrong. It.

F

Definitely wasn't in at least a previous version of the RFC.

A

This version, I think said: it's left for future work. Okay did I bridge in that I wrote, but.

B

Though it seems like the RFC has some things: I think there was one one section should have been removed, but wasn't I.

B

Think so it says the following: are forbidden: I just read the text, yeah I'm doing the same, but it says before it says: restricting confound infiltrate essence, but before that it says that it's love. So it's saying to two contradictory things: okay,.

B

A

Okay, in that case, if that's where you, if that, if that's not correct than I agree, it is like not just a syntactic change but a fairly stronger change. I didn't quite understand your point about capturing things in school.

F

I'm happy to write up some some more examples and and clarify okay, I.

C

Was in the impression that this this thing where it says right now, our she was indeed SOI, both selling syntax and not about it.

F

It is about settling syntax, but the two different syntaxes allow you to express different, the actual behaviors, but.

C

Then that is an expressiveness concern.

B

Actually doesn't because because I mean you can.

B

F

To express all the things either way the night, neither of them allows you to do a thing that wouldn't be expressible under the other, but one of them would require you to write multiple other. If you wrote the explicit existential type syntax today on nightly, you would have to use multiple existential types, and some of them would have to have generic parameters that they don't have today. Sure that.

B

Seems mostly like a like the curse of recent expansion. I'm doing things is mostly a sugar.

F

It's not I, think I'll write up some examples on the RFC thread to clarify this, but it's not as trivial as you would think, because when you're declaring it, for example, inside of a treat, it needs to capture the self type and other generic parameters and scope in the trait. But you can't actually declare that extension type as it as a trait part of the trait. So you have to declare it outside and then just pass in all those parameters.

F

So it is different, like it's not hugely different, but it's like they're real, like expressiveness differences. In terms of like writing it and one invocation versus multiple and and how the user interacts. You're saying something about a situation like this.

F

Yeah and then also sorry, maybe I'm, not looking at the rate. Yes, so there's that there's also that allows you to capture the self type.

F

So if you were to, if you remove it out into an X essential, a separate existential type say if you had like multiple implicates and then you'd have to pass around the self type, as well as any type parameters in scope, it's also allowing you to have infiltrate that's not top level and in an Associated type, which isn't currently possible the thing the syntax, without declaring it outside of the trait imple.

F

A

I think we probably need to move on, but I feel like we do over cur like this has been stalled for a while, and we need to come up with I.

B

Do think I do think.

B

A

Want to do that might be an untrue, but the other. That is one possible way. I mean one Avenue for, but it seems like we well. Both should do some thinking, but we should also is, if we I mean I, think I I feel like it wouldn't hurt us to like I, think both original concern about the teaching and so on, or the sort.

G

A

Model that is here.

F

I think that's still applicable is that is that we were saying like well.

A

That's still a nice I, I sort of fundamentally feel like people are going to want to write infiltrate anywhere that they can I should work out what all the semantics of where you are, and that and that and work on the teaching from separately. I mean like we have that's what they're gonna want to write, whatever you give them, but I also I still think. That's a was a important insightful point and I'm kind of wondering um I, don't know I'm, basically feeling like I'm, not satisfied. If you haven't clear plan forward and ask.

B

Do we think we can? We can say that if you want to implement this unlikely, go ahead, I think we can through well yeah. We could say that okay, can someone write that.

F

It's it's worth doing that that is like quite a bit of work if we just kind of throwing it away so.

C

It's gonna be the whole question of there's a reason we may be are expressed in the first place, to avoid exactly I, probably shouldn't say that yeah.

F

And that's also increasing the like emotional and social investment in that particular like right is going down the road yeah.

B

Okay, let's not someone shouldn't either come in since some or today, I think I will leave a comment today and I will.

A

Summarize that we talked about, and so.

B

The rest of us call thingy sailor, we'll leave a comment to clear up the confusion. I, don't think it's a message here.

B

Moving on, we have up here from mentoring, go to allow CFG fans, eg attributes on generic parameters. We already allow attributes on generic parameters, but not the edging, specifically because a little bug that we actually not fix and that we forgot to England and I have actually this. It should be like very trivial, and you knew it.

A

Question is: does anyone technically it's a language change? Does anyone feel object or feel we need an MCP? It seems like a pretty obvious bug to me that it, but it's already in that's okay, then everybody check their boxes are done. I'm checking mine right now. You can see it.

A

B

I will unknown eight. This.

A

Next one was this closure, so we have an our existing RFC that allows us to convert closures to X turn to function, pointers, but it doesn't coerce to unsay function, pointers or those with an API.

A

The latter, in particular, might be useful from writing against C callbacks point of view. Wait.

B

Existing oh I see well.

A

We we had an RFC that allowed us to do this closure to draw up the function pointer, coercion in the very first place.

A

I, don't recall what the orison said about these cases. Maybe it just left them to work or maybe, if they were explicitly disallowed, someone wrote a PR, they kind of implements it and the question was: what should they do? It shouldn't ask for an artesian Landover. This feature gate for what it's worth. I already left a.

F

Comment on the issue: what did your comments me? Oh, it's saying that I thought that there were some unclear design points and that they should get pulled out into a short RFC, yeah I. Don't think you can actually.

A

Have this case of a closure? That's called from two different places because of the current design, I I'm. Sorry Nico. Can you clarify that, because we're doing the coercion based on the expected type, that's propagated down? That basically means we know that it's being assigned to a location that has the type extra CFM. So there's no melius with another adi.

F

Right but but it could be assigned to that location and then also assigned to another location, because it's a function pointer. So it's copyright, yeah, but then the type but.

A

If it's assigned from like you could write code like this, which I think is where you're saying, but now the type is extra in CFM, so that this ends. This must be an extra in CFM as well. No.

F

I'm saying you what you you put nothing on on that top line, and then you assign some value of a struct that is of type extra and CFN, and then you also call the like this. Like.

A

This would be an error. Basically I'm saying is this closure expression itself I think has to have the type. The way we do these collisions has to have it touch. The coercion has to be known at that exact spot, so the type of the we coerce, the actual r-value like it, never gets. There's no alias Oh, interesting. Okay.

A

However, I would say that in general I think we could it's kind of a side comment, but I think we could be more liberal about maybe ice and basically, whenever you coerce from the zero size type, who represents a function to a function pointer, we presently require it to have the same API as you grow up, but you don't need to have that requirement. We could make wrappers um right.

A

We would then it would. Then you would be saying there's a unique function, pointer per function for you.

F

Well, because we already have we are, you can already do this with function items right where you can pass the same like function, item name into multiple things that expect like a function pointer or an unsafe function, pointer or you know something like that right and it'll coerce differently in each location. In that case, it's a different like it sort of a different like value expression or place. I, don't know what will work. It's an arm value whatever those things are. I could be wrong about the where the coercion takes. My memory.

A

Okay, yes, it seems like a nothing yeah. In any case, the real question at hand is how come I think we wanted stabilization. We don't want to go instance. People.

B

So how do you feel about on merging this with the feature gate unlikely and that currently doing an arcing or like me after the PRS, merge? Whatever.

A

I feel okay with an artsy if somebody wants to try beautifully. Maybe this wants to be part of the effort. Fine yeah I think like.

F

I feel like this is actually like I've, given my opposition, we're seeming opposition to it. This is actually a thing that I would use quite frequently, and so I would be interested in following up with this. Okay and I would be happy to see it man than that I.

B

B

Okay, great thank you and.

F

H

Think the bigger question.

F

There was the my improper c-types concern, because that would be a concern for my usages of it. Yeah yeah.

B

Yeah I think that's an interesting so.

F

B

Taylor, could you up because you coming from come on from us and like.

F

I'm sure you just want me to comment additionally in response to what I already said, noting that we discussed in the line team meeting yeah. Okay I will do that. Thank you. Thanks.

B

Thirteen minutes first time.

B

There was a plan, we have implements determination for optional units.

B

B

That Scott doesn't think this is a good idea and doesn't think this is a good idea, because well I think it is somewhat ambiguous. What does man thinks of some, and none are maybe y'all can say what you.

A

Think so this allows you to write function may narrow option and it gives currently an error code, a return code of one in the case of amendment yeah I personally, don't feel a strong desire doing that. I, don't I, don't know what I would expect in terms of error codes, I, probably just wouldn't I.

F

Mean I almost feel like less strongly about this, because, like exactly the reason, I feel, this way is just why I would never type this.

A

Bug output like it's just someone program that I'm here what it does I.

B

A

B

Main use case for this for those that this that you would want to use it for dr. Collins and doc tests, probably not elsewhere, so.

A

That's a pretty compelling use case, but except for the fact that I don't know what an error like it's not obvious at all, but I guess it's like! Well, then don't use it. If this long it makes you want your something. I think that we should wait.

C

Why wouldn't unbearable our treats non like an error right, I'm going to submit and I'm surrendering that.

A

Pressure markup treats them like, don't keep going.

F

Yeah increase it in treats it like I, like return, none as the.

F

A

Went all the way truly wanted to check for none. You would like a certain age number when you're writing a test, and you use question mark. That's like the same applies to error. You can write tests that look for.

B

I guess my I guess my thoughts here is that it's not clear to me that question mark and determination it directly links and that way, I wonder.

A

If this says the doctor s1, no I I find that pretty compelling. Actually it sort of changed my opinion they went from week week opposed to week approves and.

G

A

Terms of determining.

G

What the error, what the none should, what status got its report? I? Think we should let the lid team decide that, because, when we did, this was a result and we decided to use the debug output that was ultimately like users. Similar thing I was wrong. Mistake: nineteen that kind of happened because we decided when maybe it's marvelous issue and maybe lives. Experience would be more relevant here. I, don't.

F

Feel strongly either way. I'm help me no I, think yeah I mean III. Guess I would have more opposition to this. If the return code was anything other than the one that seemed obvious to me, but but I mean mostly I feel like this is just a thing: I wouldn't use em. It seems weird, it seems surprising in a way that, like I.

A

Mean the test cases, that's the main one I.

F

Will check my box of somebody FCPS, but otherwise I'm not gonna start FCP and I. Don't really like it. Yeah I.

B

Think we can pass this so I. Think. A reason here is that we passed this to Lipstein. We keep lying on the lake on the line label on the PR and it flips once we have CP and either way just for one.

A

Thing I would notice that Scott seemed mildly opposed. Nine years old I prefer to I give.

B

Him a chance because he's looking for me I think store, can do it. Whether concern yeah.

A

I agree I'm just saying that's a good reason to stay on. Yes,.

B

Okay, I think Owens about this.

B

um So we are skipping, looks like ID, because it's not relevant right now, so moving on so I would actually prefer to talk about the slice thing for concerts. Just what's the slice of this, the slice is this no in denominator, Darcy, first to go ahead.

A

I feel these are not version.

B

So we had an RC a while back about finalizing this, in fact, or slice patterns with slices.

B

And which petition converts and then not much has happened, but they made some adjustments to the RFC, not semantic changes, but basically making more detail, and then they are sponsored Nader, and so the there was a bit of debate about whether dot should be a pattern syntactically, but not semantically.

B

They also made a follow-up for Part. C were becomes a patterns, in fact P. So my point of view I would actually should merge them and then they can be like implemented together because it seems like part.

A

Can you summarize I this is this thing has been going on for a hundred years, you read like a nerd.

B

How many at the end, it kind of summarizes.

A

The final state I remember, I, wrote way, though,.

B

So the just looking at the summary in the RC itself. Basically, you can advise button that is like. First, we have an element like X, comma Y, the intersection pattern.

B

And that would buy salary.

B

Your section pattern, this.

A

Is the ad thing yeah.

B

And then you, if you don't want to use apps, that you can just use dot dot. So it's something like this.

B

It's a binding to the left. It's not and it's not an actual in such an afternoon. Actress but I just went through thinking of that way. So it's like items or ref items or yeah sorry I'm in I'm, in slightly.

A

This this would be a slicing with right and.

B

And you can also write.

A

And then there's a companion rx, you can make dot dot, sort of usable and more contacts like.

B

It basically turns some semantic, some syntactic errors into semantic errors and I think that's a good idea, because it's some device, the grammar and yes that for better macros and diagnostics as well, so I think we should devote yeah I'm in favor of these things.

A

I would approve an S with you.

B

What do y'all think.

B

She keeps fine with me.

B

Alright I will make an FTP.

A

All right three minutes I think we can probably there anything. That's super small.

B

Well, I need to talk about kinetics or, and the.

A

Stare I don't feel like there's, probably not a good time, but maybe I don't know what Taylor did you went to was that do nominated these I did.

F

Yeah I did want to talk about it, but we're running close. Sometimes so we can. We could let people go they long, but I say: let's leave it for next week. That sounds good to me. I think I, better last week for.

A

The focused part.

B

Discussion I had that ability, I, think, doesn't wait. I think boats had mentioned a seafood, waiters and everything might discuss right. For my part, I haven't felt super well in those days, so I haven't been to do anything about the tests and.

A

Let me discuss it. A bit. I could be happy with either this or other things yeah. So you.

G

A

Just been traveling a lot, which is why I keep pushing his back, but I think things happen. Indistinct um you don't have a lot scheduled passing should.

C

We say vehicles for discussions or when boats like, if boats have time to be able to talk about it. So.

G

I think so, for you thinking like I've written a stabilization report, I'm gonna email, you a gist version of it. I think it doesn't have his links. Previous discussions didn't need to be added. However, I was going to do that into us today, but in this situation involving send has occurred and so I think, depending on how that conversation is going yeah either we should talk about that or we should talk about or I will or next Thursday I will post a stabilization and we can talk about other.

G

Whatever other blockers people have so I think we should I wait.

G

A

I think it's great chatting on the stand description depending on where it is but I don't know. If you all those who didn't see it already, there's a thread when.

B

We're discussing what's the impact, so what's the link impact on stable or not stable for traits I.

F

Think that the feeling was that, because it might change the syntax for traits and we wouldn't want to have a syntax or traits that was different than the syntax, we used in free functions, and so we might want to adopt something different for free functions which are part of what we're stabilizing right. Yes,.

G

The that we want to apply the additional send bound to all a sycophant ring.

A

G

We might have.

A

Some I have to run out of.

B

Discussion about that before the meeting, if I think it would be a good idea to discuss this on the next posting.

F

I think people with thoughts about this should also make sure to participate in the internals thread because yeah, it's pretty active I.

G

Think that we should be really active on this, and possibly it will like it seems like this we're going to talk about the other alternative. Is that like before then it becomes obvious that we're not going to do this, in which case I will just post stabilization support and then that's what we'll talk about all.

B

Right I mean so it's been. For my part, it's mainly about just making sure that everything is okay like before we amp up the excitement for 1.337, because I'm not sure about I, feel ready for for 1.37 twins.

A

Maybe we cannot, if you're busy, I, think I think having a good suite of tests is a great idea. Young.

B

I will I will try to do my best, but yes.

A

I'm also still working on the lifetime stuff got distracted by the fact that we had Willian regression. Alright, thanks.

F