Bug #22533: Peephole optimizer infinite loop

• Status: Open
• Assignee: -
• Target version: 4.0
• Backport: 3.3: UNKNOWN, 3.4: UNKNOWN, 4.0: UNKNOWN

[tekknolagi (Maxwell Bernstein)]

The following code compiles a small loop with and without the peephole optimizer. It spawns a subprocess and waits 5 seconds before killing it to demonstrate the hang.

SRC = "while true; if true then 1 end; end"
pid = fork do
  RubyVM::InstructionSequence.compile(SRC)
  exit!(0)
end
deadline = Time.now + 5
status = nil
while Time.now < deadline
  break if (status = Process.waitpid2(pid, Process::WNOHANG))
  sleep 0.05
end
if status
  puts "compiled (#{status[1].inspect}): not reproduced"
else
  Process.kill(:KILL, pid)
  Process.wait(pid)
  puts "compile_parsey(#{SRC.inspect}) still running after 5s: killed (hang)"
end
ok = RubyVM::InstructionSequence.compile(SRC, nil, nil, 1, peephole_optimization: false)
puts "without the peephole pass it compiles: #{ok.to_a[13].size} body entries"

It also happens with .compile_parsey. This affects at least 3.4+ through master.

[byroot (Jean Boussier)]

• Related to Bug #20572: Compilation hangs with nested until added

[byroot (Jean Boussier)]

I have a patch for this: https://github.com/ruby/ruby/pull/19844, but I'm not 100% certain it doesn't regress this optimization, and based on the handful of failing ZJIT tests, it looks like it does generate different code sometimes.

So I'd appreciate if someone with more knowledge of the peephole optimizer could review.