chacha20: rand_core v0.10.0-rc.4 + set_stream()/get_block_pos() change

[nstilt1]

It does not build because it requires one change to rand_core#44, and the Cargo.toml needs to be updated.

Changes

• Update rand_core to a preview from Revise index / reset / skip methods rust-random/rand_core#44
  • I need to change the Cargo.toml once that PR makes it through
• Update set_stream() to reset the block_pos + index to 0. Before, calling set_stream() would basically behave like this:

fn set_stream(stream_id: S) {
  self.stream = S.into();
  if self.index != 64 {
    // generate_and_set is called, resulting in
    self.block_pos += 4;
    self.index = self.index; // no change to the index, but change to the block_pos
  }
}

• • I had a test that had repeated set_stream() calls and the only way to replicate it was to manually find the word_pos that resulted from each call. We can remove this test or at the very least reduce the test, since I had to add more assertions to figure out what was going on in the previous implementation.
• Update get_block_pos() to return the current block_pos instead of returning block_pos + 4 due to generate() calls incrementing the pos by 4.
  • This is not a breaking change since get_block_pos() was not implemented for rand_chacha, but it is a breaking change for anyone who was using get_block_pos() from the prerelease versions.
• Update docs to explain that set_stream() nullifies any calls made to set_word_pos() or set_block_pos().

Regarding #438

[baloo]

rand_core 0.10.0-rc-4 released this morning, any chance to get this out of draft status?

[tarcieri]

I can look shortly

[dhardy]

@baloo you mean this evening!

See rust-random/rand#1712. I'll try to resolve the rand_chacha test failure in the morning, though I think you already did for chacha20.

[baloo]

That was more directed towards @nstilt1

I'd be happy to take over to adjust the rand_core pin, otherwise I think this is ready to go.
I'd prefer to preserve @nstilt1 authorship and I can't do that easily. You can probably push over to his branch as a collaborator and then merge.

[baloo]

If you take the last commit from #502, and push it over their branch, I think it's ready to merge (and that could probably use an rc-release after that)

[nstilt1]

I am here. But I see there were some changes involving a TryRngCore trait

[tarcieri]

@nstilt1 I believe you just need to impl TryRngCore<Error = Infallible>, but maybe @dhardy can inform better

[nstilt1]

I still want to clean up that test where I did a bunch of manual debugging when troubleshooting set_stream. It looks gross and should be able to have a lot of assertions to be safely removed

[baloo]

@nstilt1 nstilt1#9 should have everything you need for the trait update.

[nstilt1]

Thanks. I was working on it but found that one of the tests had a (very minor) breaking change. I've never used trait objects in that manner and it has an easy fix, but I'm also pretty sure that the trait objects test is pretty new anyway

[tarcieri]

It'd be great if we can get something out quickly as this blocks updating everything else.

(I also have a green branch locally as it were, based on @nstilt1's, but I feel like I'm duplicating effort here)

[tarcieri]

Gonna release this since we use chacha20 as a dependency in several places

[nstilt1]

Okay. Gonna get started on #476

[tarcieri]

@nstilt1 sweet, thanks!

[tarcieri]

It's released: https://crates.io/crates/chacha20/0.10.0-rc.7