Agenda

composable for loop protocol (nmatsakis)
remove rule that requires unit type for top-level do, if, etc and replace with warning (nmatsakis)
limit destructors to owned type (nmatsakis)
syntax for multiple type bounds (nmatsakis)

For loop protocol

N: http://smallcultfollowing.com/babysteps/blog/2013/01/16/revised-for-loop-protocol/
N: Short vers: argument returns bool, iterator itself returns (). hard to compose
N: Proposed change: the iterator passes on the bool so that they can be composed
N: Additionally want 'for' to always return (). reason: top level statements without semicolons must return ()
(agreement)
N: Might also want a newtype boolean or enum to indicate break/continue
(niko: explains 'do')
B: Newtype easier now with lang items

    if cond { ... } else { ... }
    foo();

    let x = if cond { ... } else { ... } + 10;

N: We could insist on semicolons but it looks funny.
N: Today: if the statement begins with if, do, for, a few other such keywords
N: Parse as a statement, that is, do not continue past closing brace
N: But this could be wrong:

    fn foo() -> int {
        if cond { ... } else { ... } - 10
    }

N: We won't parse this as intended. Today, though, we would report an error if the if does not have a unit type.
N: But there is one important case where this breaks down:

     do task::spawn { ... }

N: spawn returns the task id but you sometimes don't want it.
N: ok, maybe not anymore, but it used to
N: another option: keep current parser but issue a warning is there is a binary operator on the same line as the closing right brace
B: So we're going to look at the spans?
N: Yeah, maybe.
G: not in favour of EOL-dependent algorithm
N: Graydon, even if it's just a lint warning?
N: Graydon, what about about if we just look at the next token and don't care about the spans?
G: more reasonable, yeah. "}" followed by binop.
N: OK. That's fine with me.
G: I think that's more-or-less emulating an ASI rule as a lint mode, which seems fair
N: Yes, that's the idea. Not actually affecting the parse, but still using ASI-like heuristics to issue a warning when you're probably not getting what you expect
P: Can we just do warn-unused-result?
N: Yes
G: \o/
N: I do not know how to interpret this emoticon, never have :) oh it means happy
N: let's discuss this on irc---too hard here!

N: Graydon said this is what he wanted from the beginning
B: Includes borrowed pointers?
N: No borrowed or managed pointers except &static
G: well, back when the distinctions were drawn along different lines (cyclic vs. acyclic and such), but that dtors have always been limited to a small-ish set of types, those that can safely be destructed top-down w/o potentially referencing nonsense memory. just a question of figuring out what the safe subset is.
N: Any disagreement?
crickets