math/rand: convenience functions always use the same seed (1). at least use time.Now() #16486
Comments
|
Especially in tests, deterministic random numbers are very helpful. It
makes it possible to reproduce a test failure.
I think this is working as intended.
|
|
If you think this is helpful in tests, then you should seed. Not the other way round. |
|
Can you give specific examples of tests you think are negatively affected by this? |
|
@adg every one of them, as someone who writes and/or reads and runs the test - without manually checking the random data - most probably will think the test runs with different data each time, when in fact it's the same data each time. yes, the data otherwise would be random, and as such only at random problems will be highlighted, but like this in tests that’s never the case. this is especially true with out of bounds testing and shifting etc. especially with fuzzing. |
|
@ProhtMeyhet if this is your contention then it should be easy to cite one of those examples and explain how the test is bad. |
|
@adg i'm saying the behavior is totally unexpected and COULD be harmful for tests with the strong implication that go programmers didn't notice this either. nothing more, nothing less. i repeat myself: something that calls itself "random" should give back different data on each call - even beyond program begin/end. otherwise it's unexpected. this is true for every language i know of. @fdietze probably got it wrong too just 3 months ago: #15142 |
|
Sorry, this is documented behavior and we can't change it anyway for go1compat reasons. We actually have gone out of our way in the past to keep the random behavior fixed and generating the same numbers. |
|
In your report you say:
But the docs say:
This also matches my expectation with standard library pseudo-random number generators in various languages. I don't think I've worked in an environment that seeds the PRNG for me. We can't change this, anyway, because people may be depending on behavior. https://golang.org/doc/go1compat @ProhtMeyhet there's nothing wrong about @minux's comment in #15142. Can you please elaborate? |
|
well, then i should be waiting for go2. hopefully it'll get a major cleanup of fuckups. :-) @adg sorry, meant the reporter fdietze, but copy+paste got me. |
|
FWIW I think the decision to see math/rand's default source with 1 was the correct choice. |
Please answer these questions before submitting your issue. Thanks!
go version)?go version go1.6.2 linux/amd64
go env)?linux/amd64
globalRand in math/rand is always initialised with the constant seed 1. this is very unexpected. it should at least use time.Now().UnixNano() instead, else the random sequence is always the same.
even most of the go programmers didn't notice this. there are about 60 tests in the standard library that use "math/rand", but only 2 call rand.Seed() with time.Now(). as such they do not really test random, but the same sequence again and again.
pseudo-random numbers, different with every call.
the same pseudo-random numbers on every call.
The text was updated successfully, but these errors were encountered: