Welcome to the main channel on the development of MoarVM, a virtual machine for NQP and Rakudo (moarvm.org). This channel is being logged for historical purposes.
Set by lizmat on 24 May 2021.
Nicholas good *, * 07:27
timo good **/* 07:49
jnthn o/ 08:11
Nicholas \o 08:16
timo good * to all my * and \* and \\* 09:07
lizmat twitter.com/SpaceX/status/1545035306960687105 # on the live video stream, it looked like the engine burn cut out too soon 13:28
wonder if "landed" is a euphemism for "forcefully deconstructed" in this case :-)
vrurg_ A general question: is it possible to use separate RNGs for Moar's internal needs and for HLL? I have encountered a case where (likely, but not sure) hash operations broke Rakudo's t/09-moar/00-misc.t. 14:00
jnthn If a test is dependent on hash ordering that'd be a bug in the test 14:07
I'm also not sure if the RNG for hashing would count as internal need or HLL, given hash randomization is a HLL-promised behavior (for security reasons) 14:08
vrurg_ jnthn: It's a bit different case. (nqp::srand(1) and then nqp::rand_I twice) x 2 – to make sure the first two produced numbers are the same, i.e. RNG is seeded as expected. 19:35
After I did some coercion/return typecheck work in dispatcher some more core code got touched in between the two nqp::rand_I. Somehow RNG got involved and the second pair of numbers produced got different from the first pair. 19:37
japhb vrurg: Are you saying that the second srand is getting ignored? Because no matter how far apart, I would expect nqp::srand(constant) followed *immediately* by nqp::rand_I should produce the same sequence. 19:39
vrurg japhb: No, it doesn't. The key word is 'immediately. And because the first and the second rand_I are used as arguments to `is-deeply`, a lot of events happens in between. 19:40
japhb Ah! So it's not really testing what it thinks its testing. Sounds like a test bug? 19:41
vrurg The point is that Test module doesn't use RNG. Not for is-deeply. So, for a user these two rands do look as immediate. And I would like it to remain this way. 19:42
japhb (Though TBH being able to spin up a separate PRNG sequence is something I've wanted for quite a long time, just not quite strong enough to go and make it so.)
vrurg I know there are cases out there where being able to seed and get predictable outcomes from RNG is crucial for testing. So, it's not a whim. 19:43
japhb vrurg: Where's the test? I think I might be misunderstanding your description of the test code. 19:44
Oh, you linked it earlier
NM
OK, I did misunderstand your description, but in a way that strengthens my belief that the test as written is wrong (just lucky previously). 19:46
That said, I 100% agree that being able to test against a known sequence of random numbers is very useful.
And that it would be nice to separate the hash sequence from the "user" sequence. 19:48
vrurg I have fixed the test, as a part of github.com/rakudo/rakudo/pull/4977. But, anyway. 19:50