Eli Zaretskii
eliz****@gnu*****
Sun Jul 29 04:21:26 JST 2018
> From: Keith Marshall <keith****@users*****> > Date: Sat, 28 Jul 2018 19:40:45 +0100 > > > #if !defined _WIN32_WINNT || _WIN32_WINNT < 0x501 > > # undef _WIN32_WINNT > > # define _WIN32_WINNT 0x501 > > #endif > > #include <winsock2.h> > > #include <ws2tcpip.h> > > > > This used to work with MinGW runtime 3.x, but now fails, evidently > > because winsock2.h no longer includes windows.h, > > Actually, it does, but indirectly via <winsock.h>, with which it now > shares code which is common to WinSock v1.1 and v2 protocols. Hmm... you are right. So the problem happens due to this chain of events: . winsock2.h includes winsock.h . winsock.h defines __WINSOCK_H_SOURCED__ . winsock.h then includes winerror.h . winerror.h sees that __WINSOCK_H_SOURCED__ is defined, and refrains from defining ERROR_* macros > > So I need to include winerror.h before winsock2.h, in order to have > > those error codes defined. > > It shouldn't matter which order you include them; <winsock2.h> first, > followed by <winerror.h> should work just as well. Right. But include it I must, yes? In which case, at least for a while, a solution such as the one below is the best I can use: #if defined __MINGW32_VERSION && __MINGW32_VERSION >= 5000002L # include <winerror.h> #endif #include <winsock2.h> Or is there a better/cleaner way out, which doesn't break older w32api versions (nor MinGW64, which doesn't define __MINGW32_VERSION)? > > Do these programs do something that isn't "kosher" nowadays? What > > should a program do if it wants to include winsock2.h, but doesn't > > want to include windows.h before that? Is that allowed? > > Yes. In fact, IIRC, it is required for strict MSVC compatibility. > > What is less clear cut is: should <winsock2.h>, (and indeed > <winsock.h>), unconditionally expose all of <winerror.h>, or just those > definitions which are relevant for the WinSock protocols? The problems happen in programs that include winsock2.h, but also need to handle Windows errors that have nothing to do with sockets. I think what w32api v5.0.2 does assumes that this combination cannot happen, but it does, of course. > On balance, it may be best to just expose all of <winerror.h>, whenever > it is included Agreed. Thanks.
Fork