Re: Narrower (per-method) GND

No, I don't think that would be adequate, but maybe there's a way to work that in. It's inadequate because MINIMAL doesn't carry any assertion of efficiency. If I indicate I want a class derived by GND, and it works, then I expect its implementation to be, at worst, very very slightly slower than the underlying implementation. If the class author doesn't make such a claim, I want users to have to be explicit about the methods derived by GND. On Jan 12, 2017 8:01 AM, "Reid Barton" wrote:

On Mon, Jan 9, 2017 at 5:11 PM, David Feuer wrote:

...

On Mon, Jan 9, 2017 at 1:32 PM, Richard Eisenberg wrote:

...

2. Defaulting to the implementation written in the class (or `error "undefined method"` in the absence of a default. This is essentially the default default.)

I want to be able to specify that a certain default definition is good enough not to worry about.

Is this the same as the purpose of the MINIMAL pragma? http://ghc.readthedocs.io/en/latest/glasgow_exts.html#minimal-pragma

Imagine GND provides implementations for those methods whose types are amenable to `coerce`ion and leaves the other methods without definitions. Then, taking into account the MINIMAL pragma, GHC either does or does not produce a warning/error about missing class methods, maybe customized to mention the failure to `coerce` a method in GND. Would that be adequate?

Regards, Reid Barton