Re: [Caml-list] Patterns that evaluate

[skaller <skaller@users.sourceforge.net>]

On Thu, 2007-02-15 at 08:41 -0500, Jacques Carette wrote:
> skaller wrote:
> > It is a common wish, but has many problems IMHO.
> > First, it isn't very general. 
> Fallacious argument: OCaml has records, so there is no need for tuples 
> which are less general.  Yet it has them. 

It isn't an argument, it's a bullet point.

> The point is to balance generality with convenience.  

Yes, I agree.

> > patterns from expressions, would be extremely fragile:

> Sure.  But that is the normal semantics of the rest of OCaml you're 
> complaining about here! 

That's only partly true. In much of ocaml, you can use local
construction such as 

	let x = expr in expr

which names the variable 'x' to be used in 'expr'. There are
issues of hijacking of course. However with the pattern thing,
the issue is not *which* definition of x you're referring to,
but whether you're referring to one at all -- or actually
introducing one.

Current patterns have two name classes: pattern match variables
(lower case first letter) and constructors (upper case first
letter or backtick for polymorphic variants, or perhaps
a #term for them).

Adding a third category suggests a new lexical mark, to keep
the 'kind' of the symbol lexically determinate.

[Yes, I know Ocaml implicitly introduces variables not
only in patterns .. but also type variables]

-- 
John Skaller <skaller at users dot sf dot net>
Felix, successor to C++: http://felix.sf.net