Benoit Daloze

Merged. Do you need a release?

Given #986, it seems important to communicate very clearly which classes are internal, e.g., using `private_constant`. Apparently "a class is only public if present in the documentation/README" is not clear...

Fixing this would mean adding overhead for Concurrent::Array on CRuby. Not ideal, but it might be the right thing to do for semantics, unsure. Holding a lock during a block...

@eviljoel Right, this is indeed inconsistent. The documentation is the intended behavior but is indeed not the case on CRuby. I plan to fix this issue and so make it...

As a note https://bugs.ruby-lang.org/issues/19078 is a solution from Ruby 3.2 for this issue. It probably doesn't just work for thread pools though where threads are created long before the task....

Just for info, autoload is also still unsafe on many implementations, so I would not recommend it for concurrent-ruby.

Well, recent MRI has a couple autoloading bugs, there is a RubySpec which fails sporadically for years, a few bug/deadlock reports for JRuby, JRuby+Truffle and new implementations most likely do...

I solved some of the cycles for https://github.com/ruby-concurrency/concurrent-ruby/issues/934, but this specific cycle remains. Yeah, it sounds like we should avoid using Delay there, seems pretty straightforward. I would avoid using...

Related to #594

This exists in many gems as `peach`/`pmap`, but I think it would make total sense in c-r. It sounds like [Scala's `.par`](http://docs.scala-lang.org/overviews/parallel-collections/overview.html) to me, that might be an interesting direction...