[erlang-questions] Module language for Dialyzer?

[Loïc Hoguin]

On 10/07/2013 10:13 AM, Dave Cottlehuber wrote:
> From: Anthony Ramine Anthony Ramine
> Reply: Anthony Ramine n.oxyde@REDACTED
> Date: 07. Oktober 2013 at 09:24:30
> To: Siraaj Khandkar siraaj@REDACTED
> Subject:  Re: [erlang-questions] Module language for Dialyzer?
>
>>
>> I am being serious and I don't see how behaviours alone relate more to ML functors than to ML signatures without parameters, as a behaviour can't be parameterized.
>>
>> I also don't see where is the inversion, nor what you can do with ML interfaces that you can't do with behaviours.
>>
>> An example would be great.
>>
>> Le 7 oct. 2013 à 05:51, Siraaj Khandkar a écrit :
>>
>>> I hope you're not being serious, because that would be abstraction inversion. Behaviour is a more specific concept.
>>
>> --
>> Anthony Ramine
>
> As a practical example, there are many (5? 8?) json encoders in the erlang community, and an additional one as part of R17 via maps (if I read Joe's new book correctly).
>
> How would you implement a consistent interface (a protocol) across those today, to allow users to switch between (say) native R17, erlang-only jsx, and nif-driven jiffy?
>
> Today I'd guess you would need to write a behaviour and then a bunch of shims between each existing encoder such that it matches the behaviour. Or have an hrl file full of macros.

I don't understand. Wasn't it said that the type is opaque? Not the case 
with JSON libs, you get a structure with a known type that you will want 
to and will manipulate directly.

You don't have a consistent interface for those libs specifically 
because of that. You could, but it would make manipulating the data too 
awkward to be any good.

-- 
Loïc Hoguin
Erlang Cowboy
Nine Nines
http://ninenines.eu