Cygwin seems to block on close() if a FIFO contains data. This test deadlocks because it expects the ability to read from the FIFO asynchronously later.
Applied ketas' micro patches, and updated an example in the docs.
C:\...\POE-1.366> nmake test "C:\opt\perl-5.20.1\bin\perl.exe" mylib/gen-tests.perl lib/POE.pm Can't open perl script "mylib": Permission denied NMAKE : fatal error U1077: 'C:\opt\perl-5.20.1\bin\perl.exe' : return code '0xd' Stop. This happens because perl ends up seeing mylib and /gen-tests.perl as two separate thingies. It seemed to me that the most straightforward solution which I do not think should effect any other platforms is to quote the filename arguments. Hope this makes sense.
… IPPROTO_UDP. CPAN testers were failing POE because Socket didn't provide these on the full range of test machines. The common thread seems to be Perl prior to 5.9. While quite old, there's no compelling reason not to support it. This change also caches the protocol numbers at startup, rather than looking them up every time. It should be slightly faster as a result.
…dler. The rt.cpan.org bug alleges that the empty string is more compatible. It doesn't seem to make a difference for contemporary versions of Perl, so why not?
Grinnz provided this test case on Freenode #perl. It shows that die() within a SIGDIE handler is not reported. Instead, the original error is thrown without any indication why it wasn't handled as intended.