math/rand/v2: add ChaCha8.Read

[FiloSottile]

Currently, if an application needs to generate a reproducible sequence of random bytes, it can choose amongst

• math/rand.(*Rand).Read, which uses the predictable Go 1 random number generator; or
• making a zeroReader that returns a sequence of zero bytes and using cipher.StreamReader to combine it with a *chacha20.Cipher, which is pretty inconvenient.

Since we now have a pretty good secure, deterministic, and specified RNG as math/rand/v2.ChaCha8, it makes sense to add a Read method to it.

func (*ChaCha8) Read(p []byte) (n int, err error)

Since it's on the *ChaCha8 value, it's not going to be accidentally misused in place of crypto/rand.

/cc @golang/security @golang/proposal-review

[zephyrtronium]

Compare #65562 which proposes adding Read to Rand instead, apparently on the assumption that it will have a ChaCha8 inside.

[neild]

Another possibility:

// Bytes fills p with pseudo-random bytes from the default source.
func Bytes(p []byte)

// Bytes fills p with pseudo-random bytes.
func (*Rand) Bytes(p []byte)

Using the name Bytes avoids any possible confusion with crypto/rand.Read.

Also, it's simpler than implementing io.Reader; why return an error when there's no way the operation can fail?

[FiloSottile]

Ah, yes, this is pretty much a #65562 duplicate, thanks @zephyrtronium. /cc @lukechampine

Even if it solves the name collision, I am not convinced by a package-level Bytes: without Seed, what's the point of having both that and crypto/rand.Read?

Eliding the error return is nice, but I think that having *ChaCha8 implement io.Reader is convenient. For example it can be swapped in for crypto/rand.Reader.

[neild]

without Seed, what's the point of having both that and crypto/rand.Read?

Sometimes you want some random bytes for non-security purposes, and you don't want to juggle two rand packages or deal with a Read function that returns an error?

[mateusz834]

I think this proposal offers a performance advantage compared to #65562. The Read method on per-source RNGs can be optimized, whereas with #65562, Rand.Read would need to call each time the Source.Uint64 method, but for consistency with the rest of package the #65562 seems better.

[hagemt]

I'm in favor of #67059 (this proposal) over #65562 because I am often generating keys with crypto/ed25519 for tests.

I want the key to be from a potentially-random seed, but using math/rand/v2 for GenerateKey is awkward for the reasons that @FiloSottile mentions. For others who may stumble upon this, I found it nice that I can do this:

import (
    "crypto/ed25519"
    rand1 "math/rand"
    rand2 "math/rand/v2"
)

type RNG struct {
    rand2.Source
}

func newRNG(seed int64) *RNG {
    r := rand1.New(rand1.NewSource(seed))
    return &RNG{r}
}

But, in order use ed25519.GenerateKey(newRNG(seed)) I need to make my type implement Read.

[lukechampine]

I'm also in favor of this proposal over my own. I stand by my position regarding the security of ChaCha8, but ultimately what matters is exposing a Read method somewhere, and directly tying it to ChaCha8 ensures that callers will know what they're getting wrt security.

[rsc]

Based on the discussion above, this proposal seems like a likely accept. — rsc for the proposal review group

The proposal is to make the ChaCha8 type implement io.Reader. It always returns the requested number of bytes and never returns an error.

[rsc]

No change in consensus, so accepted. 🎉 This issue now tracks the work of implementing the proposal. — rsc for the proposal review group

The proposal is to make the ChaCha8 type implement io.Reader. It always returns the requested number of bytes and never returns an error.

[gopherbot]

Change https://go.dev/cl/586316 mentions this issue: math/rand/v2: implement io.Reader for ChaCha8

[gopherbot]

Change https://go.dev/cl/586615 mentions this issue: math/rand/v2: add Read method to ChaCha8 source

[gopherbot]

Change https://go.dev/cl/587280 mentions this issue: math/rand/v2: add ChaCha8.Read

[rolandshoemaker]

Re-opening for proposal acceptance.

[rolandshoemaker]

Uh, this is not the right issue.