Re: Gain performances and fix some issues using XS

unsubscribe

On Thu, 5 Feb 2026 at 23:11, Marc Chantreux <[email protected]> wrote:

> Hello porters,
>
> I would like to improve a module and have the feeling there is nothing
> I can do in pure perl so I post here to get advices. I apologize if
> I missused the list: please let me know.
>
> Thanks for any help and regards.
> Marc
>
> __DATA__
>
> I discovered [haskell](https://fr.wikipedia.org/wiki/Haskell) in the
> mid 2000 but never had the chance to put it at work. I wanted to
> get the similar logic in Perl so I wrote https://metacpan.org/pod/Perlude,
> presented it at French Perl Workshop ('10?) and YAPC::Europe'11.
>
> Book and Dolmen helped me to polish the thing and experimented a lazy
> version of Perlude (which is only "on demand", like unix pipes) then it
> remains untouched for many years. I tought I was the last one to use it but
> I noticed someone made a debian package out of it.
>
> This module was good enough for me to use it without feeling the need to
> improve it since then but now I have larger and larger datasets the
> known performance issue of perlude hurts. This issue is known and
> I declared I was unable to fix it at the time because Perlude is
> based on anonymous subroutines returning scalars in an list context (so
> the end of a stream is () which is a very functionnal programming
> thing). I don't know how to mesure what costs the most. I remember I
> tried to use some profilers but I was unable to understand the result
> because everything happened in __ANON__ (AFAIR).
>
> Since then I discovered C and try to give XS a chance (also because I
> would like to bind some fortran libraries at some point) so I would like
> to investigate the only 3 anoying things I have with Perlude and I
> wonder if XS could be a solution (as I found none in PP):
>
> 1. make it faster at execution
> 2. make it even faster/easy to write with expressions instead blocks
>    with filter/apply (which are "on demand" conterparts of grep/map).
> 3. put a value back to the stream in case of failure
>
> eg.
>
>         use Perlude;
>         sub fibo {
>                 my @seed = @_;
>                 sub {
>                         push @seed, $seed[0] + $seed[1];
>                         shift @seed;
>                 }
>         }
>         my $fib = fibo 1, 1;
>         print join ",", fold takeWhile {$_ < 50} $fib;
>         print " then ...", $fib->();
>
> problem 3:
>
>         # fibo: 1,1,2,3,5,8,13,21,34,55,89
>         # current output
>         # 1,1,2,3,5,8,13,21,34 then ...89
>         # ideal output
>         # 1,1,2,3,5,8,13,21,34 then ...55
>
> problem 2:
>
> current version:
>
>         # with expressions intead blocks
>         print join ",", fold takeWhile {$_ < 50} $fib;
>
>         # I wonder if there is a trick to make fold useless
>         print join ",", fold takeWhile $_ < 50, $fib;
>
>         # or overload an operator for subroutines to fold
>         print join ",", @< takeWhile $_ < 50, $fib;
>
> --
> Marc Chantreux
> Pôle CESAR (Calcul et services avancés à la recherche)
> Université de Strasbourg
> ☎  03.68.85.60.79
>