Bruno Haible <[EMAIL PROTECTED]> writes: > Simon Josefsson wrote: >> It would replace the >> inet_ntop module, and lib/inet_ntop.h would be removed. >> >> With this, files only need to #include <arpa/inet.h> for inet_ntop or >> inet_pton, just like it should be. > > How would this work on platforms which have an <arpa/inet.h> already > but don't have and don't declare the inet_ntop() function? Such as > Solaris 7 and many other platforms.
Ah, good catch. So perhaps we should have two modules here: "arpa_inet": Provides an arpa/inet.h on systems that lack it. The header is needed on mingw to provide declarations for ntohl, ntohs, etc (by depending on and including sys/socket, which includes winsock2.h). It would also help remove #if HAVE_ARPA_INET_H tests. "inet": Implements inet_?to? and has a header file inet.h that applications should include. Thoughts? One question here is whether arpa_inet should depend on inet? Then the arpa_inet module would provide the inet_ntop/inet_pton functions when necessary, which arguable (?) should be provided when arpa/inet.h is provided. The disadvantage is, of course, that applications that depend on arpa_inet and need inet_?to? will have to include both arpa/inet.h and inet.h to make sure they get the inet_?to? declarations. So there isn't really a need for "arpa_inet" to explicitly depend on "inet", the user of these modules will have to make sure he's using both manually anyway. I think I'll start working on this approach. Whether the dependency is there explicitly or not can be decided later. /Simon
