Test2::Plugin::SRand breaks rand under fork.

[demerphq]

Test2::Plugin::SRand assumes that it can set srand() to get deterministic testing. This breaks the behavior of rand() and fork(). As of 5.37.x there is a new env var which should be used instead PERL_RAND_SEED, which behaves correctly with fork.

$ PERL_RAND_SEED=1234 ./perl -le'fork; print rand() for 1..3;' | sort
0.21453485716799
0.338195834114355
0.378420907809538
0.740876929442496
0.839252316250594
0.842944087068648
$ PERL_RAND_SEED=1234 ./perl -le'fork; print rand() for 1..3;' | sort
0.21453485716799
0.338195834114355
0.378420907809538
0.740876929442496
0.839252316250594
0.842944087068648

versus this:

$ ./perl -le'srand(1234); fork; print rand() for 1..3;' | sort
0.21453485716799
0.21453485716799
0.338195834114355
0.338195834114355
0.740876929442496
0.740876929442496
$ ./perl -le'srand(1234); fork; print rand() for 1..3;' | sort
0.21453485716799
0.21453485716799
0.338195834114355
0.338195834114355
0.740876929442496
0.740876929442496

notice that calling srand() explicitly results in the child processes getting the same srand() state, whereas using PERL_RAND_SEED is deterministic but ensures each child process has its own seed.

I'd say that Test2::Plugin::SRand breaks anything that uses fork and rand() which does not itself use srand() explicitly.

[exodist]

The "broken" behavior is exactly as expected and if I am not mistaken documented. It may not be ideal in 5.37.x and above, and that should be documented as well now.

Apart from documenting the "better" solution, what fix would you propose?