Regression when IO.wait_* or rb_io_wait is interrupted by Thread#kill

[larskanis]

Since Truffleruby-24.0.0 there is a regression regarding interruptions while waiting for IO.

This rose up in ruby-pg here: https://github.com/ged/ruby-pg/actions/runs/8393741842/job/22989402958#step:12:1202 but it is reproducible with pure ruby code like this:

require "io/wait"
describe "wait_readable" do
  it "should stop waiting on Thread#kill" do
    rd, _wr = IO.pipe
    start = Time.now
    t = Thread.new do
      rd.wait_readable(1)
    end
    sleep 0.1

    t.kill
    sleep 0.1
    puts t.backtrace
    t.join

    expect( Time.now - start ).to be < 0.9
  end
end

In Truffleruby-23.x and MRI this looks like so:

rspec test-wait_io.rb 

.

Finished in 0.20932 seconds (files took 0.1265 seconds to load)
1 example, 0 failures

But in Truffleruby-24.0.0 this fails, because the poll is not interrupted:

rspec test-wait_io.rb 
<internal:core> core/truffle/polyglot.rb:334:in `execute'
<internal:core> core/truffle/polyglot.rb:334:in `call'
<internal:core> core/posix.rb:142:in `truffleposix_poll_single_fd'
<internal:core> core/posix.rb:95:in `truffleposix_poll_single_fd'
<internal:core> core/truffle/io_operations.rb:262:in `poll'
/home/lars/.rvm/rubies/truffleruby-24.0.0/lib/truffle/io/wait.rb:25:in `wait_readable'
/home/lars/comcard/ruby-pg/test-wait_io.rb:8:in `block (3 levels) in <top (required)>'
F

Failures:

  1) wait_readable should stop waiting on Thread#kill
     Failure/Error: expect( Time.now - start ).to be < 0.9
     
       expected: < 0.9
            got:   1.017792
     # ./test-wait_io.rb:17:in `block (2 levels) in <top (required)>'

Finished in 1.1 seconds (files took 0.42542 seconds to load)
1 example, 1 failure

Failed examples:

rspec ./test-wait_io.rb:4 # wait_readable should stop waiting on Thread#kill

[andrykonchin]

Thank you for the report, we'll look into it