Re: [libvirt] [PATCH libvirt 2/6] include winsock2.h before windows.h

On 01/25/2012 03:16 PM, Marc-André Lureau wrote: If it were POSIX, the answer is yes - all POSIX headers are required to be idempotent and self-contained, so any header that requires some other header to be included first, instead of doing the prereq includes under the hood itself, is broken. And gnulib already does a great job of fixing the (surprisingly large number of) systems that fail this POSIX header litmus test. But we're talking about Windows, so we have to follow Microsoft conventions. And if Microsoft requires <winsock2.h> to be included before <windows.h>, then mingw64 is doing the right thing in mimicking that convention; at which point, we can't insist on "fixing" <windows.h>. But we _can_ insist on fixing libvirt's use of Windows headers so that _only_ the windows-specific files have to know about the dependency ordering constraints, and all other libvirt files can use any libvirt header without having to think about inherited dependency constraints. threads-win32.h is our windows-specific header that was triggering the problem, so that's the point in the chain worth fixing, so that normal files, like virkeepalive.c don't have to think about "if I include both "threads.h" and <unistd.h>, am I introducing an ordering constraint on windows platforms?" Glad to hear it; I've pushed the alternate patch. What repo? The mingw64 compiler isn't in Fedora 16 proper. -- Eric Blake eblake(a)redhat.com +1-919-301-3266 Libvirt virtualization library http://libvirt.org