Oliver Bandel <oliver@first.in-berlin.de> writes:

> If he only would be cynic about "well I tried it, but it has proven
> it's a crap language... but I give you (experienced OCaml programmers)
> the chance to show me that it isn't bad (and *I* have o learn the language)"
> then this woul be ok.

I agree.

> Maybe I'm wrong here and he's a good guy. But IMHO it seems he's looking
> for proved Ocaml-failure (and not for "well, I tried it, it seems OCaml sucks,
> but please tell me it does not!", which I could accept).

I think his intention were good (but he is surely a bad guy too).

> Do you think *I'm the bad guy*?!

no :) of course not :)

> > I am not sure that making the function tail-recursive would have been the big
> > hit in this example.
> 
> But as long as nobody tried it / analzed it, your assumption is
> only an assumption, as my assumption is only an assumotion too!

Well at least my assumption about "use Map instead of assoc list is a big hit" 
has been proven. The running time from the program go from "more than 16 minutes" to just
50 seconds (this is what I call a big hit).
It would be interesting to make his function tail-rec (and keeping the assoc list)
and see if it is a big hit but I am too lazy for this and I hate 
those tail-rec transformation and I think it would
not be a big hit cos using Map and Array is the big hit.
It's your turn oliver bandel to do the job :)


> > I often transform my functions to make them tail-recursive because of stack overflow pb, not
> > that much because of speed pbs
> > (and many functions in the standard library are not tail-recursive, such as map)
> 
> 
> Well, maybe I have overestimated the tailrec-point,
> but as long as there is not a proven counter-example,
> the opposite of what I stated seems to be only an assumption too.

True.

> 
> !!! To see different results it would be nice to have different implementations
> !!! to compare them all!

I agree.

> Would be interesting to have different variations of the
> code, using different ways of coding some special tasks
> in different ways.... and maybe oe implementation,
> that uses *all* suggestions.
> 
> To do it in the language shootout, as on Jon stated it, seems
> to be a veryg good idea.

I agree.
I must confess that I have very few intuition about what is more important
in optimization, but perhaps because it depends on the program.
Sometimes X is a better optimization than Y on this program Z.
Sometimes Y is a better optimization than X on this program Z2.

> > This tail-recursion stuff is one of the thing I hate the most with fp because it forces
> > you to change your code to adapt to the machine whereas it should be the 
> > inverse.
> 
> But only because you hate it does not mean that changing the code will not result
> in better performance.

true :)


-- 
Yoann  Padioleau,  INSA de Rennes, France   www.irisa.fr/prive/padiolea/
Opinions expressed here are only mine. Je n'écris qu'à titre personnel.
**____   Get Free. Be Smart.  Simply use Linux and Free Software.   ____**