Bug #20917: redo/next in nested begin block causes wrong order of execution

• Status: Open
• Assignee: -
• Target version: -
• ruby -v: ruby 3.3.5 (2024-09-03 revision ef084cc8f4) [x86_64-darwin24]
• Backport: 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: UNKNOWN

[hoshiumiarata (Arata Hoshiumi)]

It seems that redo/next in a nested begin block can cause the wrong order of execution.
For example:

for _ in [0]
  puts 0
  begin
    puts 1
    begin
      puts 2
      redo
    ensure
      puts 3
    end
  ensure
    puts 4
    break
  end
end

It prints:

0
1
2
3
4
3
4
=> nil

But I think it should print:

0
1
2
3
4
=> nil

Because execution order should be:

1. puts 0
2. puts 1
3. puts 2
4. redo
5. unwind to nested ensure block
6. puts 3
7. unwind to outer ensure block
8. puts 4
9. break
10. end of loop

Interestingly enough, if we add an empty rescue block before any of the ensure blocks, then the execution order is correct.

[Eregon (Benoit Daloze)]

Indeed, it seems like a bug, I think as well the "redo jump/unwind/exception" should run ensure's and get to the "break jump/unwind/exception" which should override the redo like when a Ruby exception overrides another.

FWIW, this is 0 1 2 3 4 on both TruffleRuby and JRuby.

[kddnewton (Kevin Newton)]

Agreed this is a bug. For additional context, this is the same on parse.y and prism.