Repeatable rand() and keys() for debugging

ibm1620 has asked for the wisdom of the Perl Monks concerning the following question:

Hello Perlmonks,

I'm having trouble debugging a heavily random program because I'm unable to force it to repeat its behavior.

After reading perlrun and web-searching a bit I concluded that the following snippet at the top of the program might do the trick:

#!/usr/bin/env perl
use v5.36;
BEGIN {
    srand(1);                  # rand generator is repeatable
    $ENV{PERL_PERTURB_KEYS}=0; # traversing keys is repeatable
    $ENV{PERL_HASH_SEED}=1;    # hashing is repeatable?
}
[download]

In addition to calling rand() to choose an array index, it also traverses lists of key(%hash). As far as I know, those are the only places where randomness can occur. (No hashes are modified after they're constructed at initialization.)

I gather from perldoc perlrun ("this mode is as close to pre 5.18 behavior as you can get.") that perhaps there are no guarantees anymore?

Can anyone suggest what I might be doing wrong or overlooking?

Comment on Repeatable rand() and keys() for debugging Download Code

Replies are listed 'Best First'.
Re: Repeatable rand() and keys() for debugging by choroba (Cardinal) on Dec 09, 2022 at 17:30 UTC
It's too late to set the env vars in the begin block, they must be set in the process that runs Perl itself. This works: `PERL_PERTURB_KEYS=0 PERL_HASH_SEED=1 perl -lE 'say for keys %{ { qw( a + 1 b 2 c 3 d 4 ) } }' c d a b` [download] If you really insist on doing it from Perl itself, the following seems to work: `#! /usr/bin/perl use warnings; use strict; use feature qw{ say }; if (($ENV{PERL_PERTURB_KEYS} // 1) != 0 \|\| ($ENV{PERL_HASH_SEED} // 0) + != 1) { $ENV{PERL_PERTURB_KEYS} = 0; $ENV{PERL_HASH_SEED} = 1; exec $^X, $0, @ARGV } say for keys %{ { qw( a 1 b 2 c 3 d 4 ) } }, @ARGV;` [download] `map{substr$_->[0],$_->[1]\|\|0,1}[\\|\|{},3],[[]],[ref qr-1,-,-1],[{}],[sub{}^ARGV,3]`	[reply] [d/l] [select]
Re^2: Repeatable rand() and keys() for debugging by ibm1620 (Hermit) on Dec 09, 2022 at 19:31 UTC
Thanks - that appears to be working consistently now. I can see now why it's more practical to just write a wrapper shell script to set the env vars, and then invoke perl.	[reply]