Hi, Am Dienstag, den 02.07.2013, 14:11 +0000 schrieb Simon Peyton-Jones:
I'm not sure that a plugin is the way to go here, though it's a good start.
as I said, it is just to host the prototype that allows for quick experimentation and allows people to test it without having to compile GHC.
· I’m not sure how you expect to deal with essential NT arguments:
newtype T a = MkT a a
tNT :: NT a b -> NT (T a) (T b)
tNT n = createNT ???
I planned to use tNT = error "magic message to GHC.NT.Plugin" that is then replaced by GHC.NT.Plugin (and if for some reason it is missed by it it stays type safe)
To your questions:
· To do these kind of things, CoreM will need more reader stuff. In particular:
o The global TypeEnv o The GlobalRdrEnv
Ok, I guess this means that I’ll need to leave the easy land of Plugin experimentation very soon... I also noticed a problem with my logic for creating the NT-lifting function: Just having the constructors of C in scope is not sufficient to safely provide NT a b -> NT (C a) (C b) as the parameters of the constructor might wrap a in another type constructor, i.e. data Foo a = Foo (Set a) then we certainly don’t want the user to be able to write deriving fooNT :: NT a b -> NT (Foo a) (Foo b) as he can easily get the unwanted (Set a) -> (Set b) coercion for arbitrary a b from that. So in this example, the compiler should at most accept deriving fooNT :: NT (Set a) (Set b) -> NT (Foo a) (Foo b) but it feels a bit weird that the „internals“ of Foo are relevant here. Following this line of thought it means that for data Proxy a = Proxy we would allow deriving proxyNT :: NT (Proxy a) (Proxy b) without a "NT a b" parameter, as long as the Proxy data constructor is in scope. Greetings, Joachim -- Joachim Breitner e-Mail: mail@joachim-breitner.de Homepage: http://www.joachim-breitner.de ICQ#: 74513189 Jabber-ID: nomeata@joachim-breitner.de