Use current_randstate() for seeding PRNGs used by Interfaces

[Erik Bray]

There is a method Interface.rand_seed() which returns a randomized seed that can be used to initialize different Interfaces RNGs, where applicable (it use called by individual interface implementations' set_seed() methods when the provided seed is None).

For some reason (itself not entirely clear to me) this does not use the current_randstate(), but rather initializes a new randstate() and uses that to obtain an arbitrary random seed. Maybe there's no good reason for that in the first place, in fact?

In any case, for the doctests we want consistent results, so with this patch I have this method use current_randstate().seed() so that results are consistent across test runs.

This came up in comment:341:ticket:22626 where it turned out some tests that use the GAP pexpect interface which depend on GAP's PRNG were different between test runs.

There are many other tests like this sprinkled about which manage this by manually calling current_randstate().set_seed_gap() before the test, but putting something like this in the documentation would seem rather distracting.

[Dima Pasechnik]

OK, reviewing this probably needs a test run of ptestlong and ptest just to make sure nothing breaks. Looking into this now.

[Erik Bray]

Replying to dimpase:

OK, reviewing this probably needs a test run of ptestlong and ptest just to make sure nothing breaks. Looking into this now.

Already did last night. Remarkably, nothing broke. There probably isn't actually all that much that depends on this, and most tests that could have been affected by it were already manually calling current_randstate().set_seed_<foo>() ahead of time...

[Dima Pasechnik]

over to bots...

[Erik Bray]

Marking this as critical since it's a dependency of #22626

[Erik Bray]

Closed 4 days ago so I'm assuming this didn't go in 8.5. Please don't close a ticket as fixed without updating the milestone to its target release.