Random.javaUtilConcurrentThreadLocalRandom not thread-local

[durban]

I think that cats.effect.std.Random.javaUtilConcurrentThreadLocalRandom might call a ThreadLocalRandom from the incorrect thread. Its unlikely this would cause any problems on OpenJDK, since there every ThreadLocalRandom is the "same", but... it's still not allowed, and who knows what other JVM-s might do.

I think this could happen in the following way. Inlining a few things, what it does is like this:

def nextBoolean: F[Boolean] =
  for {
    r <- Sync[F].delay(new SRandom(java.util.concurrent.ThreadLocalRandom.current()))
    out <- Sync[F].delay(r.nextBoolean())
  } yield out

So there is a flatMap between the call to .current() and (for example) the call to .nextBoolean(). And if there is an "autocede" between them, the fiber could be rescheduled to another JVM thread. And then .nextBoolean() would be called on another thread's ThreadLocalRandom. At least I think this could happen.

[vasilmkd]

That is a valid point and should be fixed. Feel free to open a PR.

[djspiewak]

This is actually a great first issue. Fixable just by collapsing the two delay blocks together. Please target the series/3.3.x branch!

[PiotrBosak]

I allowed myself to try to fix the issue. I hope that's fine with @durban .

[durban]

@PiotrBosak Sure, absolutely :-)

[armanbilge]

Btw this would have been another possible application of #2610.

[durban]

This was fixed by #2784.