This package is a drop-in replacement for math/rand
, except it also allows you to serialize the random sources' internal state, and restore it at will.
That can come in handy when debugging flaky tests: if you know an initial seed for your whole suite that makes some test fail, you don't need to replay all your tests up to the one that fails any more: just dump the random generator's internal state at the beginning of the failing test once, and you can then load that state to only run the offending test.
It's a drop-in replacement for math/rand
with just a couple of methods on top:
import (
"github.com/wk8/go-rand"
)
source := rand.NewSource(time.Now().UnixNano())
state, err := source.Marshall()
if err != nil {
// ...
}
r := rand.New(source)
// say the next call to r.Int() yields 12
if err := source.Unmarshall(); err != nil {
// ...
}
// your source is now re-set to what it was when it was marshalled
// in particular the next call to r.Int() will yield 12 again
or if you prefer the global version:
state, err := rand.Marshall()
// ...
err = rand.Unmarshall(state)
It uses the exact same RNG as math/rand
; in fact most of this repo is copy-pasted directly from there.
That's also why given the same seed, it will give you the exact same pseudo-random values as math/rand
's generator.
And for this very same reason, you shouldn't trust this repo any more than math/rand
's generator, which is not crypto-secure