On 03/15/2010 12:57 AM, Paolo Bonzini wrote: > On Sun, Mar 14, 2010 at 23:33, Bruno Haible <br...@clisp.org> wrote: >> Hi Paolo, Eric, >> >> On Cygwin 1.5.x the 'poll' test fails like this: >> >> Unconnected socket test... passed >> Connected sockets test... passed >> General socket test with fork... passed >> Pipe test... failed (expecting POLLHUP after shutdown) >> FAIL: test-poll.exe >> >> What do you think? Is Cygwin broken? Or is the test too strict?
Cygwin has known bugs in 1.5 on pipes, and different set of known bugs in 1.7.1 on pipes. 1.7.2 will be released soon, which uses yet another implementation for pipes, but which may also have bugs. Until we get a stable cygwin solution for properly emulating Unix pipes on top of Windows, it's hard to say what to expect from the gnulib side of things. But yes, the cygwin developers are aware that gnulib provides a decent testbed for whether pipes are working. > On the other hand, poll/select are broken on Windows for pipes anyway > (and not fixable), so they probably shouldn't be tested in > test-poll/test-select. Maybe for native Windows pipes, but cygwin can implement Unix pipes using something other than native Windows pipes and get correct semantics, insofar as pipes are only used between cygwin processes (but the moment you cross between cygwin-to-non-cygwin, then you are relegated to using native Windows pipes with all their flaws). -- Eric Blake ebl...@redhat.com +1-801-349-2682 Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature
