http://qs1969.pair.com?node_id=1167789


in reply to setting PERL_PERTURB_KEYS & PERL_HASH_SEED in a perl file

Hi gravid,

As far as I can tell from a glance at the source, Perl only checks those environment variables once on startup (the function Perl_get_hash_seed in util.c sets the variables PL_hash_rand_bits_enabled and PL_hash_rand_bits). It might be possible to change the values at runtime from XS code, but that's not my area of expertise.

But I think it's important to point out that hash ordering has always been random, even though older Perls used to return hash keys with some consistency to the ordering, leading to code sometimes relying on this fact. But at least since v5.18.0 that has changed, see Hash overhaul. The bottom line is, one shouldn't rely on the order hash keys are returned in, in any version of Perl.

I find the easiest solution is usually something like for my $key (sort keys %hash) ..., or in some cases modules like Tie::IxHash can help. See also How do I sort a hash (optionally by value instead of key)? This might be the "better way" to fix your code than to rely on Perl's hash ordering.

Hope this helps,
-- Hauke D

Replies are listed 'Best First'.
Re^2: setting PERL_PERTURB_KEYS & PERL_HASH_SEED in a perl file
by bart (Canon) on Jul 17, 2016 at 08:13 UTC
    But I think it's important to point out that hash ordering has always been random, even though older Perls used to return hash keys with some consistency to the ordering, leading to code sometimes relying on this fact. But at least since v5.18.0 that has changed, see Hash overhaul. The bottom line is, one shouldn't rely on the order hash keys are returned in, in any version of Perl.
    Actually it's much older than that. In order to protect perl from some form of DOS attacks, a randomization of key ordering was implemented in perls as old as 5.8.1, so that the next time you run it, you'd most likely get a different keys order for the same hash keys inserted in the same order.

    The reason for that is a possible DOS attack over hash collisions (from 2003) which was fixed that same year in October.

      Hi bart,

      Thanks very much for that, I dimly remembered there had been a change in the past (hence the "at least"), I just didn't think to check perl581delta ;-)

      Regards,
      -- Hauke D