Meeting weekly 2011 10 11

Attending

Brian, Dave, Marijn, Niko, Patrick

Unsafe blocks

• Niko:
• can tag functions and blocks as unsafe
• unsafe functions only usable from other unsafe functions and blocks
• can't use unsafe functions as first-class values
• status:
• missing restriction to second-class
• need to audit all operations, see which need to be restricted
• currently all natives tagged as unsafe
• Marijn: maybe not the right default; forces us to write wrappers
• Niko: probably don't want to have to add a safe keyword
• Patrick: maybe come up with heuristics based on whether they take pointers?
• Brian: calling into LLVM could cause an assert to happen just about anywhere
• Marijn: can't always prove things are safe, but don't want to have to infect everything
• resolved - safe by default for native

Typestate

• Marijn:
• not working out in some places where we lose info due to mutation
• can we stop using it where it isn't appropriate?
• Niko: should look at Scala's approach to subtyping for cases like subsets of disjoint unions
• Dave: this is like ad-hoc unions
• Patrick: should visit when we revisit object system after v0.1
• Marijn: do we have subtyping?
• Patrick: kind of, only a very little bit due to mutable? (which we probably want to change)
• resolved - where they are proliferating, just change to assertions if we're hitting them and eliminate if we aren't

Miscellany

• Marijn:
• tutorial in Markdown? - fine with everyone
• destructuring assignment? - fine with everyone, but need to think about syntax (preferably no sigil/keyword)
• mix assignment and binding in a single term?
• Dave: I don't like mixing binding and reference/assignment in one form
• Marijn: but this saves you from destructuring twice
• Patrick: well, you can bind the names w/o initializing up front, and then just use assignment
• Marijn: that's uglier
• Dave: subjective; my instinct says it's not
• Marijn: I'll work on it and keep thinking about it