Ruby 3.0 support

[voxik]

There is test failure testing on Ruby 3.0:

  1) ChildProcess can write to stdin interactively if duplex = true
     Failure/Error: raise msg
     
     RuntimeError:
       timed out after 10 seconds:
       expected "hello\ncat: -: Resource temporarily unavailable\n" to match /\Ahello\r?\n\z/m
       Diff:
       @@ -1,2 +1,3 @@
       -/\Ahello\r?\n\z/m
       +hello
       +cat: -: Resource temporarily unavailable
       
     # ./spec/spec_helper.rb:197:in `wait_until'
     # ./spec/io_spec.rb:121:in `block (2 levels) in <top (required)>'

Any chance to get this fixed please?

[voxik]

It seems that the problem might be some race condition, since this works:

$ rspec ./spec/io_spec.rb
..........

Finished in 0.79139 seconds (files took 0.07509 seconds to load)
10 examples, 0 failures

[voxik]

Ah, forgot to include the CHILDPROCESS_POSIX_SPAWN=true, so it doesn't work indeed.

[voxik]

This was the last successful build:

https://travis-ci.org/github/enkessler/childprocess/jobs/687662868

and this is where the issue first appeared:

https://travis-ci.org/github/enkessler/childprocess/jobs/698756797

[voxik]

So the cat: -: Resource temporarily unavailable is reported from here:

childprocess/lib/childprocess/unix/posix_spawn_process.rb

Lines 51 to 58 in c626073

	ret = Lib.posix_spawnp(
	pid_ptr,
	@args.first, # TODO: not sure this matches exec() behaviour
	actions,
	attrs,
	argv,
	envp
	)

called from here:

childprocess/spec/io_spec.rb

Lines 114 to 116 in c626073

	process.start

Checking what was changed in Ruby, this is susceptible commit:

ruby/ruby@0e3b0fc

But that might be completely false lead 😇

[voxik]

This seems to be the root cause, the change from RUBY_PIPE_NONBLOCK_DEFAULT, which was 0, to O_NONBLOCK. I have locally replaced the O_NONBLOCK by 0 and the test suite passes.

Unfortunately, other test case fails then:

  1) ChildProcess lets a detached child live on
     Failure/Error: expect(alive?(c_pid)).to be true
     
       expected true
            got false
     # ./spec/childprocess_spec.rb:214:in `block (2 levels) in <top (required)>'

[voxik]

I have reported this against Ruby:

https://bugs.ruby-lang.org/issues/17607

[eregon]

I wonder if #172 / #175 might solve this. At least CHILDPROCESS_UNSET=should-be-unset bundle exec rake passes for me on that PR with Ruby 3.0.0.

[eregon]

Confirmed that #175 (comment) actually fixes the CI on ruby-head (the only ruby-head CI failing is then due to installing FFI, which is unrelated and no longer needed with that PR).

[voxik]

Thx, the test suite passes with #175. Nevertheless, I'd still like to also understand the reasons why it started to fail, although it will have smaller priority 😇

[sds]

Closed by #175.