Re: [merd-devel] doc on syntax

[Pixel <pi...@ma...>]

Yoann Padioleau <Yoa...@ir...> writes:

[...]

> gcd(10,) for partial app can be good also cos it EXHIBIT the fact that we do a partial app
>  sometimes we can return a func that is a partial app and we dont see it clearly
> ex: let f1 x = ...... in (f2 x)
>  and f2 is in fact of arity 2 !! so by using f2(x,) we can help the reader of the prog.

i meant this, adding it.

> 
> +++ WHYISHIIP; excellent principle 
>  quite related to the principle of least surprised

added

> 
> +++ associate a type with a name is excellent, but can you generalise??
>  i would like to say
>  vartype "%.*" = Hash 
>  vartype "@.*" = Array
>  vartype "p_.*" = Ref 
> I would like to do as in ruby
>  vartype "@@.*" = instance var

for this, you need to change merd's lexer for the variable names. But the idea
is good.

> can it be applied to function name too ??
> would like to do as in scheme
>  vartype ".*?" = func -> Bool
>  vartype ".*!" = func -> Unit

wow, that's nice! great idea.

> 
> merd can do even more typing than ml :)
> you say it is complicated but dont really see why.

i meant
- complicated in C
- complicated for the type error reporting

> in caml you can explicitly type some var/func as in
>  let f (i:int) .... and the inference engine use this
>  here you just have a pre processing phase where you add that
>  if you have a list of (regexp * type)
>  for every identifier matching regexp, you assoc its corresponding type.

agreed, adding it to existing MLs should be very easy, but you may get
incomprehensible error messages. I may be pessimist on this though.

[...]

> have you got an interpreter ??

nope. But i've solved a lot of pb. I'm gonna base the inference on o'haskell's.

there are still things to fix (mainly abstract values, if i don't want
existential types)