The usual objection to this historically is that cryptographically secure random integers, floats, etc are rarely needed in real cryptographic constructs, and so providing them in the standard library would only encourage people to abuse crypto/rand to do unsound and insecure things.

There has been previous discussion about this but I can't locate the relevant issue(s) at the moment.

/cc @FiloSottile

I'm coming at this because I don't see an easy way to implement "give me a slice with $n random characters from $charset" (kubernetes-retired/bootkube#984). math/rand's Int31n per-character would be secure, if not particularly efficient. And for the Bootkube use-case, efficiency is not terribly important.

To be more efficient, you'd need an Intns(max int, count int) []int (with a better name?), but one thing at a time ;).

Read a slice of random bytes and translate them to the desired alphabet, using a base64-style approach? That is adaptable (and has been adapted) to all kinds of alphabets of different radices, including those not a power of two.

This is already very easy to implement using the standard library

func RandInt32() (int32, error) {
    var buf [4]byte
    _, err := rand.Read(buf[:])
    return int32(binary.LittleEndian.Uint32(buf[:])), err
}

int32(binary.LittleEndian.Uint32(buf[:])) 

This doesn't work as well with base 36 (or other bases that aren't powers of two). For those, as far as I can tell, you need to either throw out the remainder and re-draw (like this) or convert the base 256 input into your target base (e.g. via math/big, like this).

Re-drawing is inefficient, especially since the modulus based approach is already inefficient. You can end up drawing multiple bytes to get an unbiased draw from [0,36).

Converting to the target base is awkward too (unless there's some helper function that I'm missing?). See maintainer feedback on kubernetes-retired/bootkube#984 as evidence of that.

But in general, yeah, this could be handled outside of crypto/rand by a helper that converted bytes to your target base (which is basically what I have in kubernetes-retired/bootkube#984 now, although the current implementation swaps endian-ness). Where would that go in the stdlib? encoding/binary?

I've filed #25534 with a proposed Choose helper for random token generation in arbitrary bases. If this proposal is blocked on rarely needing a crypto source for the Rand methods, maybe we can table this until the Choose proposal settles?

Like in #25534, this seems fairly special-purpose. It is OK for functionality to exist in packages outside the standard library. Not every conceivable use has to be a function in the standard library.

Furthermore, I would push back on anything that further blurs the line between crypto/rand and math/rand. Adding math/rand.Reader was arguably already a mistake. Let's not make more.