Re: Window + fork + Crypt::Random = pain

[ikegami]

fork doesn't really have anything to do with it. It's about threading. Keep in mind that fork creates a new *thread* in Windows, not a new *process*. The problem can be triggered more directly using:

perl -MMath::Pari -Mthreads -e "threads->create(sub{})->join();"

It appears that Math::Pari is not thread-safe.

[bsb]

I'm using the Math::Pari ppm but it's build process seems to only use the unix source PariBuild.pm

sub download_pari {
  my ($srcfile, $force) = (shift, shift);
  my $host = 'megrez.math.u-bordeaux.fr';
  my $dir  = '/pub/pari/unix/';
  my($ftp, $ua, $base_url);
  ...

There is a windows build at ftp://megrez.math.u-bordeaux.fr/pub/pari/windows/ which may avoid the problem although I haven't yet tried to use it.

Thanks for your reply

[Anonymous Monk]

If the module isn't thread safe, a special windows build wouldn't avoid the problem.

[bsb]

Some C libraries can be compiled to be threadsafe but are not by default, eg. SQLite. (I admit that's probably not the case here)

[bsb]

That also segfaults in linux.