Hi, Am Freitag, den 06.09.2013, 16:47 +0000 schrieb Simon Peyton-Jones:
So
* for newtypes newtype N a = MkN <rep-ty> the coercion is between (N a) and its representation type
. The coercion is allowed if the data constructor MkN is in scope * for data types (T a), the coercion is between (T a) and (T b), The coercion is allowed if the roles allow it.
The two are handled quite differently.
OK? This is far from obvious (since I was very confused about it), so worth writing up on the design page. As well as implementing.
Sounds reasonable. Implementation is simple, I just remove the check that I had added: https://github.com/nomeata/ghc/commit/46c6c7 https://github.com/nomeata/ghc-testsuite/commit/d5c13 One question which is left open, which I cannot answer: Are there situations where the library author wants to prevent coercion, but needs a non-Nominal role for some other reasons? But we’ll see.
Roles are in HEAD so you can use them right now.
I know, my patch is against head and already makes use of roles: TcCoercibleFail.hs:16:8: No instance for (Coercible (Map Int ()) (Map Age ())) because the first type argument of ‛Map’ has role Nominal, but the arguments ‛Int’ and ‛Age’ differ arising from a use of ‛coerce’ In the expression: coerce In the expression: coerce $ Map one () :: Map Age () In an equation for ‛foo3’: foo3 = coerce $ Map one () :: Map Age () Greetings, Joachim -- Joachim “nomeata” Breitner mail@joachim-breitner.de • http://www.joachim-breitner.de/ Jabber: nomeata@joachim-breitner.de • GPG-Key: 0x4743206C Debian Developer: nomeata@debian.org