Re: What's wrong with C++?

[Georg Bauhaus <rm-host.bauhaus@maps.futureapps.de>]

On 2/17/11 10:08 PM, Yannick Duchêne (Hibou57) wrote:
> Le Thu, 17 Feb 2011 21:50:20 +0100, Georg Bauhaus <rm-host.bauhaus@maps.futureapps.de> a écrit:
>> Oddly enough, it is good practice to *not* play the inference
>> game when creating "models" in ML languages: when ML programmers
>> describe a structure or type, they use type names explicitly.
>> Every function is written in terms of its (typed) profile,
>> not just the names.
> Not sure I've understood. Could you give a short example ?

For example in OCaml's standard library everything has
an interface.

val copy : string -> string
(** Return a copy of the given string. *)

To me, this seems to represent the modeling part; the implementation
would be the real thing, not the model.
More importantly, programmers wishing to just use the string (model)
would look at the interface.  No type inference so far.

ML is similar. It has signatures for its structs.

Haskell definitions are frequently prefixed with function
type, every major (model) definition states its type explicitly.

Here is a quote from the prelude.

class  (Eq a, Show a) => Num a  where
     (+), (-), (*)    :: a -> a -> a

Thus if you want to know what (*) stands for in Haskell, you will find
it here.  "Generic" type /a/ is explicitly a Num and an instance of
Eq and Show.

Specifying types explicitly also helps early error detection:

let f1 = function x -> function y -> x +. y ;;
let f2 = function (x : int) -> function (y : int) -> x +. y ;;

let test1 = f1 3 4 ;;
let test2 = f2 3 4 ;;

Both f2 and test1 trigger a diagnostic message.
test1 finds the mistake in the definition of f1,
but arguably, that's too late!
f2's error is detected immediately when it has been read.

So, apparently, type inference is used by programmars when writing
small local things, but not to replace explicit type annotations.