Hi, Am Mittwoch, den 09.10.2013, 15:21 -0400 schrieb Richard Eisenberg:
Wait! I have an idea! The way I've been describing GND all along has been an abbreviation. GHC does not coerce a dictionary from, say, Ord Int to Ord Age. Instead, GHC mints a fresh dictionary for Ord Age where all the methods are implemented as coerced versions of the methods for Ord Int. (I'm not sure why it's implemented this way, which is why I've elided this detail in just about every conversation on the topic.) With this in mind, I have a proposal:
1) All parameters of all classes have nominal role. 2) Classes also store one extra bit per parameter, saying whether all uses of that parameter are representational. Essentially, this bit says whether that parameter is suitable for GND. (Currently, we could just store for the last parameter, but we can imagine extensions to the GND mechanism for other parameters.)
Because GND is implemented using coercions on each piece instead of wholesale, the nominal roles on classes won't get in the way of proper use of GND. An experiment (see below for details) also confirms that even superclasses work well with this idea -- the superclasses aren't coerced.
what do you need the extra bit for? During GHD, can’t you just create the new dictionary (using "method = coerce original_method) and then see if it typechecks, i.e. if the method types can be coerced. (If not, the error messages might need massaging, though.) 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