Skip test on BSD systems.

[Math2]

They usually have bidirectional pipes, which makes the test fail.

Types of Changes

• Bug fix.

Contribution

• [x] I tested my changes locally.
• [x] I agree to the Developer's Certificate of Origin 1.1.

[ioquatix]

Instead of skipping the test, what about shutdown(WR) on it? i.e. IO#close_write.

[Math2]

Instead of skipping the test, what about shutdown(WR) on it? i.e. IO#close_write.

Nah, shutdown is only for sockets on BSD. Pipes don't support it at all. But now that you mention it, closing the read end of the pipe makes the write end unwritable and it makes the test work.

[Math2]

Oops actually scratch that, I made a mistake. If you close the "output" and then try to write on the "input", you'll get EPIPE (instead of EBADF on Linux).

So yeah that's something that could be tested instead of just skipping the test.

[Math2]

What do you think?

I just tested, on Linux you always get EBADF even if you close the output beforehand. So I don't think this test can be made to be the same on both Linux and BSD.

diff --git i/test/io/event/selector/buffered_io.rb w/test/io/event/selector/buffered_io.rb
index c6b0ac5..5119235 100644
--- i/test/io/event/selector/buffered_io.rb
+++ w/test/io/event/selector/buffered_io.rb
@@ -39,11 +39,14 @@ BufferedIO = Sus::Shared("buffered io") do

        it "can't write to the read end of a pipe" do
            skip "Windows is bonkers" if RUBY_PLATFORM =~ /mswin|mingw|cygwin/
+           bsd = RUBY_PLATFORM =~ /bsd/
+           
+           output.close if bsd # BSD traditionally has bidirectional pipes

            writer = Fiber.new do
                buffer = IO::Buffer.new(64)
                result = selector.io_write(Fiber.current, input, buffer, 64)
-               expect(result).to be == -Errno::EBADF::Errno
+               expect(result).to be == -(bsd ? Errno::EPIPE : Errno::EBADF)::Errno
            end

            writer.transfer

[ioquatix]

This looks okay to me. Actually, the test is just that it's an error, we probably don't care much which error it is. It's invalid behaviour - it's more a test that errors are reported correctly.