Map IO::TimeoutError to Net::OpenTimeout

[jbourassa]

Since Ruby 4.0, Ruby raises an IO::TimeoutError when the open_timeout is reached.

The fix: Similar to Net::HTTP in ruby/net-http#263, map those errors to Net::TimeoutError. This case is simpler though because net-imap only supports Ruby 3.2, and IO::TimeoutError exists in 3.2: https://docs.ruby-lang.org/en/3.2/IO/TimeoutError.html

Before -- notice the exception class at the end of the line:

mise exec ruby@3.4.10 -- ruby -Ilib -rnet/imap -e 'Net::IMAP.new("imap.gmail.com", open_timeout: 0)'
./lib/net/imap.rb:3706:in 'Net::IMAP#tcp_socket': Timeout to open TCP connection to imap.gmail.com:143 (exceeds 0 seconds) (Net::OpenTimeout)

mise exec ruby@4.0.7 -- ruby -Ilib -rnet/imap -e 'Net::IMAP.new("imap.gmail.com", open_timeout: 0)'
~/.local/share/mise/installs/ruby/4.0.7/lib/ruby/4.0.0/socket.rb:930:in 'block in Socket.tcp_with_fast_fallback': user specified timeout for imap.gmail.com:143 (IO::TimeoutError)

After

mise exec ruby@4.0.7 -- ruby -Ilib -rnet/imap -e 'Net::IMAP.new("imap.gmail.com", open_timeout: 0)'
./lib/net/imap.rb:3706:in 'Net::IMAP#tcp_socket': Timeout to open TCP connection to imap.gmail.com:143 (exceeds 0 seconds) (Net::OpenTimeout)

---

For tests, the only reasonable way I found requires using a non-loopback ip; I don't know if we want that in CI; so I left it out.

[nevans]

@jbourassa Thanks! I wasn't quite sure if we want to move forward with IO::TimeoutError in new ruby versions, or maintain backward compatibility by mapping that to Net::OpenTimeout. But, I'm happy to follow net-http's lead on this.

For testing... what do you think of this test?

  def test_timeout
    host = ENV.fetch("TEST_CONNECT_TIMEOUT_HOST", "10.255.255.1")
    pend host if host.start_with? "skip"
    assert_raise(Net::OpenTimeout) do
      Net::IMAP.new(host, open_timeout: 0)
    end
  end

I used 10.255.255.1 IP address because that's what's used in the tests in https://github.com/ankane/the-ultimate-guide-to-ruby-timeouts, and I expect we can use that in most dev/test/CI environments. But I do wonder if it'd be better to use an IP address from one of the officially designated TEST subnets [RFC5737].

In other environments, we can set the environment variable to something appropriate, e.g: TEST_CONNECT_TIMEOUT_HOST=192.0.2.1 or TEST_CONNECT_TIMEOUT_HOST="skipped on Windows in CI".

[jbourassa]

For testing... what do you think of this test?

Sounds great, well done.

Thanks for looking at the PR and more generally your work on net-imap.