I think carol needs to be added to the test suite.

I got a tip that may be dup is needed because of this:

Another reason for duplicating a file descriptor is using it with fdopen. fclose closes the file descriptor that was passed to fdopen, so if you don't want the original file descriptor to be closed, you have to duplicate it with dup first.

Does that seem right?

I don't think dup is strictly needed, but it is probably the best way to ensure that we don't accidentally close a file descriptor used by another object. It also gets a bit confusing when we throw marshaling and sharing streams between threads in the loop, so it's probably the most practical thing to do.

I wasn’t able to get this working on MinGW. Based on some cursory reading, I think this might be a limitation in how MinGW handles the use of Windows sockets via file descriptors. As a result, calling this function in a MinGW environment will raise an error.

That is strange to me, since the windows path should be used on windows and mingw

I wasn’t able to get this working on MinGW. Based on some cursory reading, I think this might be a limitation in how MinGW handles the use of Windows sockets via file descriptors. As a result, calling this function in a MinGW environment will raise an error.

That is strange to me, since the windows path should be used on windows and mingw

I don't have a MinGW environment that I can test to explore further. You can see the errors that were occurring for various approaches in my Actions history. A sample error:

Starting suite test/suite-ev.janet...
error: could not flush file
  in file/flush [src/core/io.c] on line 284
  in thunk [test/suite-ev.janet] (tail call) on line 425, column 1

and if I disable buffering:

Starting suite test/suite-ev.janet...
error: error writing to file
  in file/write [src/core/io.c] on line 251
  in thunk [test/suite-ev.janet] (tail call) on line 424, column 1

I've removed the special casing in e94e8dc and, while the test suite includes a check to avoid testing in MinGW, the current commit can be experimented with in MinGW.

Thanks for the work here, @pyrmont! Looks pretty solid to me, and we can investigate the mingw issue later, as I also don't have a mingw setup handy for testing right now.

Great! Thanks for merging it in! I’m looking forward to using it in my project with the next Janet release :D