Rarely failing nil assertion in signal_spec

[straight-shoota]

Nil assertion failed (Exception)
  from src/class.cr:148:0 in 'not_nil!'
  from spec/std/signal_spec.cr:0:23 in '->'
  from src/signal.cr:255:3 in '->'
  from src/signal.cr:255:3 in 'process'
  from src/signal.cr:198:9 in '->'
  from src/fiber.cr:255:3 in 'run'
  from src/fiber.cr:29:34 in '->'
  from ???
FATAL: uncaught exception while processing handler for CHLD, exiting

It seems to be a race condition in the Singal::CHLD handler and happens very rarely.

I' might have seen this error more often, but here are at least two instances where it failed in unrelated CI runs:

• circleci/test_linux
• travis-ci/linux64 (unfortunately, travis doesn't seem to keep the logs of a job after it has been restarted)

[straight-shoota]

Another occurrance: https://circleci.com/gh/crystal-lang/crystal/18298?utm_campaign=vcs-integration-link&utm_medium=referral&utm_source=github-build-link

[asterite]

I GOT IT!!! 💡

This is the spec:

it "CHLD.trap is called after default Crystal child handler" do
  called = false
  child = nil

  Signal::CHLD.trap do
    called = true
    Process.exists?(child.not_nil!.pid).should be_false
  end

  child = Process.new("true", shell: true)
  child.not_nil!.wait # doesn't block forever
  called.should be_true
end

TIL I learned about SIGCHLD: it's trapped when a child process ends.

The spec fails because of a not_nil! (the trace isn't entirely correct with the filenames and line numbers). But my guess was that child.not_nil! is raising. How could that happen?

Well, if Process.new("true", shell: true) runs and ends before it gets assigned to child, which is super rare, then the handler will be invoked and child will still be nil ⚠️

This race condition can be reproduced if you change the child assignment to:

child = begin
  p = Process.new("true", shell: true)
  sleep 1
  p
end

Now to fix it. I guess if child is nil I'll just Fiber.yield until child gets a non-nil value. It's a bit hard fixing this because something happens before an expression gets assigned to a variable 🤔