sort: Use fastrand

[oech3]

3189432 -> 3179752 byte

Summary
  sort-nano -R /etc/pacman.conf ran
    1.05 ± 0.27 times faster than sort -R /etc/pacman.conf

non-crypto rng with getrandom should be safe.

[xtqqczze]

WyRand has 64 bit output size, which we use to generate 128 bit salt?

[oech3]

I guess it is just generating 64bit twice.

[anastygnome]

Hey, as a sidenote, you can use

git push <remotename> <commit SHA>:<remotebranchname>

to push only one commit to a remote.

[oech3]

I deliberately split 2 commit to run bench only and drop later.

[xtqqczze]

@oech3 Going forward, you could use NOMERGE as the commit message for clarity in these scenarios.

[oech3]

No need. This is draft.

[xtqqczze]

No need. This is draft.

I mean, in future.

I think like #11123, you should use fastrand in this PR, since uu_sort already it as a transitive dependency.

[oech3]

since uu_sort already it as a transitive dependency.

What crate? Also does shuf have similar dep (though shuf might require high-quality rng...)?

[xtqqczze]

fastrand is a dependency of tempfile.

[xtqqczze]

though shuf might require high-quality rng...)?

We can still get some improvement changing from default RNG (ChaCha20) to ChaCha8.

[oech3]

Since CodSpeed might not report anything:
3189320 -> 3179328 byte

Summary
  sort-fastrand -R /etc/pacman.conf ran
    1.00 ± 0.21 times faster than sort-main -R /etc/pacman.conf

[oech3]

This will be draft until #11123 was merged.

[xtqqczze]

There might be no statistical significant difference between all three implementations, but one less dependency is a win.

[oech3]

(btw, I was considering to remove rng dep from fuzzer by reusing ASLR provided pointer address...)

[xtqqczze]

(btw, I was considering to remove rng dep from fuzzer by reusing ASLR provided pointer address...)

I don't know, this approach might be limited by low entropy (especially on 32-bit systems)

[github-actions]

GNU testsuite comparison:

Skipping an intermittent issue tests/tail/symlink (passes in this run but fails in the 'main' branch)
Note: The gnu test tests/printf/printf-surprise is now being skipped but was previously passing.