If the original block of a non-orphan block becomes an orphan block, the copy object does not become an orphan block (incompatibility with CRuby).

In CRuby, after copying a block object from a directly given block object, when control is returned from that method, the copied object also switches to orphan block. However, mruby does not switch to orphan block, even going back at least to version 1.3.0.

def m(&b)
  b.dup
end

x = m { break 1 }
begin
  x.call

  # Even if mruby goes back at least to version 1.3.0, it gets here.
  raise "BAD!"
rescue LocalJumpError
  # Where CRuby gets here.
  p "GOOD!"
end

The way I see to solve this problem is either to follow the CRuby behaviour, or to compromise and make all copy objects orphan blocks.

(1) If we compromise, we just force the MRB_PROC_ORPHAN flag in mrb_proc_copy(). Perhaps additions to doc/limitations.md are needed.
(2) If the behaviour is to be similar to CRuby's behaviour, we believe it would be to add proc->body.origin so that we can track whether it is an orphan block or not. The workload is somewhat more involved.
(3) We will close this issue as if we didn't see it.

Which one looks better for mruby?

:black_cat: I'm in the process of retrying #6282, which was reverted in #6295, and ran into this issue when I thought the previous issue was resolved.

mruby / mruby

If the original block of a non-orphan block becomes an orphan block, the copy object does not become an orphan block (incompatibility with CRuby). #6345