[erlang-questions] Parameterized module initialization

[Anthony Ramine]

Le 26 juin 2012 à 16:56, Joe Armstrong a écrit :

> I like this mechanism.  This way of hiding additional data is very flexible.
> 
> I've used this in several programs, and the resulting code is easy to
> understand. This nice thing is that using this you can make very nice
> client APIs. For example, for string processing you might say
> 
>             Str = {ansi_string, "abc"},         or
>             Str = {utf8_string, "u+1234"}
> 
> And then when you call
> 
>             Str:substr(1,4)
> 
> Get the correct behaviour depending upon the underlying string type ie
> either ansi_string:substr(1,4,{ansi_string,"abc"}) or
> utf8_sting:substr(1,4,{utf8_string, "u+1234"} ) will be called
> 
> If you write a program where strings might be in latin1, or utf8 or
> something else you'd still
> have to tag the string (as in say {latin1, "abc"}) so you knew what
> the encoding was and still
> have to call different modules depending upon the tag. This is what
> the above mechanism does
> so it's very nice for hiding the mess in polymorphic interfaces.
> 
> /Joe

I think it is quite ugly to have terms as module parameters; a more beautiful way in my opinion would be to have something like Caml functors, which are mappings from structures (modules in Erlang) to structures.

You would have a string functor which takes as a parameter a string implementation module, either ansi_string or utf8_string. You could then abstract the common stuff from ansi_ and utf8_ in string.

Regards,

--
Anthony Ramine