On mingw, there was another test failure. Apparently the behaviour of
write() on a non-blocking pipe fd depends on whether some pipe reader
is currently blocked, read()ing from it.
- If yes, then write() of more than the pipe buffer's size yield a
partial write, no failure.
- If no, then wr
I've fixed the 'nonblocking' unit test for pipes now, so that it succeeds on
all Unix platforms except Cygwin.
On mingw, I've got two failures, though. The first one is because write() on a
non-blocking pipe sets errno to ENOSPC instead of EAGAIN.
$ ./test-nonblocking-pipe-main.exe ./test-nonbl
