On 10/03/2013 01:40 PM, Sergey Mironov wrote:
Oh, yes, Thanks! I should have read the manual more carefully. benignEffectful helped me. Still, the contract looks unusual (after IO of Haskell). Did you think about including part of the contract in the type signatures? I mean that we probably could have transaction monad parametrized with some type indicating 'effectfulness'.
I've thought about deducing such information automatically from FFI signatures, considering any [transaction]-typed function to be effectful. It would still be important to think about each function and whether its effect is "benign" (page-local) or not, but I'll plan to experiment with adding [transaction]al functions as non-benign effectful by default.
By the way, here is another idea of this kind, bothering me since I started to explore Ur: can (should) we actually parametrize the transaction monad with the 'side' type indicating server-side, client-side or 'any side' ? Currently, programmer should keep the side in mind, `alert' is the example (it has `transaction unit' type, still we should not call it from the server side context). If we lifted this information to the type annotations, we would then have type client_side type server_side val alert : string -> transaction client_side unit val getServerEnv : string -> transaction server_side (option string) (* should work on both server and client sides *) val debug : t ::: Type -> string -> transaction t unit AFAIK the compiler already keeps all distinctions of this kind internally! So what can go wrong with this?
I agree that it would be better to use types to track the sorts of effects that a [transaction] can have. I have a variety of other ideas to remove some of the other not-very-compositional aspects of the way Ur/Web works right now, and my plan is to tackle all of these some day in a complete rewrite.
_______________________________________________ Ur mailing list [email protected] http://www.impredicative.com/cgi-bin/mailman/listinfo/ur
