On Wednesday 27 June 2007 14:18:07 Quôc Peyrot wrote:
> On Jun 27, 2007, at 2:14 PM, Jon Harrop wrote:
> > Provided performance is unimportant...
>
> Could you be more specifics about the performance hit?

Actually, no. I have no quantitative results but I remember that it is 
significant.

> I was quite happy with this solution, but maybe there is something
> more elegant to do?

Nice. :-)

> (I'm still in the process of learning good optimization patterns in
> ocaml which preserve readability)

I was going to mention symbol tables to Raj B for his Python JIT compiler but 
he hasn't gotten back to me yet.

Basically, you memoize strings:

# let symbol =
    let m = Hashtbl.create 1 in
    fun string ->
      try Hashtbl.find m string with Not_found ->
      Hashtbl.add m string string;
      string;;
val symbol : '_a -> '_a = <fun>

This is another trick I learned whilst writing my Mathematica interpreter (so 
many tricks, so little time). This function looks totally pointless, like a 
no-op, but if you pipe your identifiers through it (e.g. when building the 
AST during parsing) then all structurally-equal strings are the same physical 
string. If you're careful, this lets you use physical equality for string 
comparison and that is a lot faster.

You might do something similar by memoizing temporary strings of different 
lengths.

-- 
Dr Jon D Harrop, Flying Frog Consultancy Ltd.
The OCaml Journal
http://www.ffconsultancy.com/products/ocaml_journal/?e