Bug #19407: 2 threads taking from current ractor will hang forever

• Status: Closed
• Assignee: ractor
• Target version: -
• Backport: 2.7: UNKNOWN, 3.0: UNKNOWN, 3.1: UNKNOWN, 3.2: UNKNOWN

[luke-gru (Luke Gruber)]

In the current implementation of Ractors, it's possible to take from the current ractor. This could be useful
when co-ordinating threads:

t = Thread.new do
  obj = Ractor.current.take
  p obj # do some work with obj
end
t0 = Thread.new do
  obj = Ractor.current.take
  p obj # do some work with obj
end
Ractor.yield :go

However it hangs forever:

t = Thread.new do
  obj = Ractor.current.take
  p obj
end
t0 = Thread.new do
  obj = Ractor.current.take
  p obj
end
sleep 0.5

Should "self-take" be disabled, or was it designed to allow it but this is just a bug?

[hsbt (Hiroshi SHIBATA)]

• Status changed from Open to Assigned
• Assignee set to ko1 (Koichi Sasada)

[jhawthorn (John Hawthorn)]

• Assignee changed from ko1 (Koichi Sasada) to ractor

[jhawthorn (John Hawthorn)]

• Status changed from Assigned to Closed

I believe the Ractor::Port interface doesn't suffer from this issue

ruby -e 'th = 2.times.map { |i| Thread.new { p(i => Ractor.receive) } }; %w[hello world].each { Ractor.current.send it }; th.each(&:join)'
{0 => "hello"}
{1 => "world"}

However I think it might make sense for us to forbid Ractor.current.join/Ractor.current.value

❯ ruby -e 'Thread.current.join'
-e:1:in 'Thread#join': Target thread must not be current thread (ThreadError)
        from -e:1:in '<main>'

❯ ruby -e 'Ractor.current.join'
(nothing, but exits cleanly with Ctrl-C)