On Jun 27, 2007, at 2:31 PM, Brian Hurt wrote:

> Actually, what I'd like is a more powerful regular expression engine-

This is off the top of my head, but I've always thought it would be  
nice to have built-in regexps using a Scanf style.  Instead of using  
$1, $2 (or \1, \2, depending on your language), have each  
parenthesized argument be passed to a function, and the whole thing  
is a function of a string parameter.  If a scanf conversion is used,  
that is the type passed to the function, otherwise it's simply passed  
as a string.  Something like:

# let re = /(foo|bar|baz)/ (fun x -> match x with | "foo" -> ...  
etc ... );;
re : string -> 'a = <fun>

# let simple_phone_number = /(%d)-(%d)-(%d)/ (fun x y z -> (x,y,z) );;
simple_phone_number : string -> (int * int * int) = <fun>

I don't know how hard this would be to implement.  It would require  
some scanf style compiler magic, but it overcomes scanf's shortcoming  
that it doesn't allow character classes.  It might even be possible  
for use in pattern matching:

# let f str = match str with
| /foo/ -> (* It's a foo *)
| /bar/ -> (* It's a bar *)
| /baz/ -> (* It's a baz *)
| /(%d)-(%d)-(%d)/ -> (* It's a phone number *)
| _ -> (* Not matched *)
;;
f : string -> 'a = <fun>

Any idea how hard this would be?

--Jonathan