Use an internal random number source rather than math/rand directly about miniredis HOT 6 CLOSED

from miniredis.

Hi @zikaeroh,

I've implemented it in #87. I choose to do this in NewMiniRedis():

    source := rand.NewSource(rand.Int63())
    m := Miniredis{
        ....
        rand:        rand.New(source),
    }

This way seeding the global source still has some effect in Miniredis, while it give the option to override the seed. It's not 100% the same as before, but it's the simplest thing to do.

from miniredis.

I'll be happy to look at it more closely later. My first impression (skimming on my phone) that it'd definitely lead to different results between miniredis versions (since you're using a seed from Rand), and might also require some locking, but I haven't tested to see. I had actually been planning on submitting my own PR for consideration (but hadn't had time since asking).

from miniredis.

I opened a slightly different PR (#89) for your consideration; this one instead calls the global rand as it would have before, but will switch to an internal version when requested. I think this is better than always using an internal copy with rand.Int63() as a seed because:

• Using rand.Int63() changes the seed used (as it wouldn't output 1, Go's default seed), so those who had used the old advice of rand.Seed would have broke those users.
• The library would still rely on global state; every miniredis instance created would call rand.Int63() and would all behave differently (in case a test creates more than one server, only has one of them use randomness, etc).

from miniredis.

from miniredis.

Sure, random is random, except that existing users may have already taken the advice of using the global seed in non parallel tests, and would otherwise be broken. (The fact that we can seed at all is really because it's only psuedorandom anyway, which is intended.)

from miniredis.