2013-04-30.md

Agenda

• Goal sheet (Azita)
• Landing borrow checker (nmatsakis)
• Closure recursion (nmatsakis)
• typedef reform (pcwalton)
• removal of <-> (pcwalton)
• @ in patterns (pcwalton)
• servo build (Jack)

Goal sheet

• meta discussion about how to manage ourselves

Landing borrow checker

• niko: how should I land my new borrow checker?
• patrick: I usually patch bomb. Because of bors, things bounce many times. Little patches get in the way and cause breakage. Usually I end up with a set of patches then a bunch of smaller 'fix test' commits because of conflicts.
• graydon: Do you mean one big patch or one pull with lots of commits?
• niko: That's what I'm wondering. ....
• niko: What I'm asking is can I just push this and deal with the fallout? < missed>
• niko: I usually go commit by commit, make a comment, then find it's corrected in a later commit.
• graydon: We're still in the stages of reworking major subsystems, and everyone is putting trust in you.
• niko: Ok, I'll just get it landed. What I've got is passing all the tests and building itself.

Closure recursion

• niko: Wrote a blog post. &fn is unsound. I don't think it's too important but does require a fix. Bottom line: we have to prevent closures from calling themselves.
• niko: It's very hard to actually call a closure, but it can be done. It's incompatible with a lot of things we're trying to do. e.g. 'for' loops need to borrow stack variables without worrying about creating hidden aliases.
• niko: I think it won't break any code
• niko: Until recently closures were limited to argument position... we expanded scope of closures without reasoning about what it meant.
• jclements: This will be a static error? You can fence out the Y combinator?
• niko: Yes. It amounts to making &fn affine.
• brson: Does this affect borrowed objects?
• niko: No, because their state is always in the self pointer which is already subject to the normal rules.
• niko: restructuring check so that upvars are treated like everything else.
• graydon: Reifying the stack and ensuring those upvars are treated like everything else.
• niko: Yes.
• niko: I won't get to that for a while but be aware of it.

Changing typedefs

• patrick: Calling methods on typedefs, but you would not be able to use from them. Niko described a way to make use work as well.
• patrick: Easy for monomorphic paths, but with type parameters is pretty hard
• N: I think it is something we can scope in over time
• brson: Would be nice to do the types of things that bjz is trying to do on the pull request
• niko: Putting the type parameters near the place where they are declared is common.
• graydon: I don't understand from the comments whether bjz was satisfied with the workarounds

Removing swap

• patrick: Saw a comment that we can't remove swap, for some reason I can't remember.
• patrick: I do know swapping array elements will not borrow check if we remove it.
• niko: We can have a function for it
• patrick: Often swap isn't what you want. You want n-ary swap, 'shuffle', taking a few elements and reorder them
• patrick: Treemap wants this.
• tjc: Reason before had to do with purity, so not an issue now - rust-lang/rust#3466
• niko: We can probably get rid of swap
• niko: One thing that would be useful here and elsewhere is a set of operations that can divvy up a &mut [T] slice into disjoint subslices or into pointers to disjoint elements.
• pnkfelix: I think I needed swap at some point
• niko: I think due to bugs in the borrow checker
• pnkfelix: let's wait till that lands then
• niko: that'd be helpful, it's hard to deal with the overlap between old and new
• graydon: I would very much favor doing anything to get borrow checker changes done soon, since they really affect everyone
• graydon: kind of like resolve
• pcwalton: new checker should make people's lives easier, particular around flow sensitivity

@ in patterns

• P: I know we discussed in Vancouver but don't recall the outcome
• P: Can we remove x @ Pattern?
• pnkfelix: What about x = Pattern?
• jc: right now expressions can appear in grammars
• N: didn't we say we just want constants?

- jc: maybe I should make a small proposal on email and people can agree or bikeshed - graydon: I think in any case, `=` is not going to be used elsewhere in the pattern language - niko: It's bad enough that we have to deal with guards that mutate state

Servo breakage

• jack: Servo doesn't build because of an ICE
• patrick: Tim may want to look into this. ICE in pattern matching in trans. #6117
• patrick: match forgot about one of the patterns (or something) and tried to index into an array
• patrick: It only happens on one platform in monomorphised code.
• graydon: Which platform?
• patrick: Mac Mountain Lion
• jack: So it doesn't affect any rust developers but affects people trying to use rust ...
• patrick: Call to resolve::iter.
• niko: Looks like the bind_irrefutable_patterns code? Anyway let's talk about this offline.