> - > It needs to be tested on Windows (mingw and cygwin) > > The patch is supposed to change nothing on Cygwin, right? Cygwin has > perfectly fine POSIX socket support already.
I added a #error to verify that. That's why I needed testing on Cygwin. > - The transformation of WSAGetLastError() values to errno values is a nice > trick, but the mapping e -> e - 10000 will not work out because of some > clashes. Looking at mingw's <errno.h> and <winsock2.h> the first ones are > ok but not the later ones: > > I would propose to leave the higher WSA* values as is, though a > mapping e -> (e < 10025 ? e - 10000 : e) Good. But then, let's "randomly" pick 10060 for Yoann's other patch. > - It would be good to extend the gnulib <errno.h>, <string.h>, <stdio.h>, > strerror, perror replacements to handle these values (when the sys_socket > module is in use). (errno values are supposed to be all defined in > <errno.h>, > strerror is supposed to be declared in <string.h>, and perror is supposed > to be declared in <stdio.h>.) Yes... but can we do it later? :-) > - Can we call the replacement functions "rpl_getsockopt" etc. please, not > "win_getsockopt"? To make it clear that it's the gnulib replacement, and > because the prefix "win" is subtle brainwashing. Yes, of course. > - What about the need to call WSAInit()? Is it still needed at all? That's provided by the separate sockets module. > - The doc should document that while most fd-based POSIX functions are > correctly replaced, things like > int fd = connect (...); > FILE *fp = fdopen (fd, "w"); > will not work - because the fwrite() function will call write(), not send(), > under the hood. ReadFile and WriteFile work under win32 with sockets (learning that is what prompted me to submit this finally). Paolo
