James Cheney

That seems possible, but it's such a simple interface/ADT that I'd be surprised, and the implementation should be completely hidden (aside from hacks that violate abstraction like using `==` or...

OK. I was also surprised that this seemed to affect correctness. I'm inclined to close this or re-categorize as a bug - the high-level behavior shouldn't depend on internal implementation...

Even weirder: with the library implementation in branch `issue945`, the following happens (running with an empty prelude): ``` links> fun foo(l) { ...... foo(l) ...... }; foo = fun :...

You've designed this such that `pow(_,_)` would desugar to something of type `(Int) -> (Int) -> Int`, which is nice because the desugaring is more compositional but a bit counterintuitive....

We already check that things are (generalized) values in other places, for type inference, so perhaps we could check that if underscores appear in a function call then all of...

+1. Perhaps an alternative could be that the argument to `startServer` is a list describing the routing table. This would be more restrictive than the current approach in which routes...

`orderby` uses `sortByBase`, but the Links code for `sortByBase` and `sortBy` is identical. So the problem is probably nothing to do with `orderby` per se, my guess would be that...

Ha. Yes, this cyclic import seems like it should be an error. I think that if we move to having explicit `import`, with dot-notation aligned with directory structure, then we...

Cool. Yes, `realpath` and `readlink` are not portable across Linux vs. BSD (MacOSX), or at least, the command-line tools that provide access to these are not portable. If you can...

Sounds good. I think we definitely want to avoid having `import`s be hidden away in non-toplevel declarations, so that the imports of a file are statically defined.