Re: [PHP4BETA] PF_INET

Philippe Verdy <verdy_p@wanadoo.fr> wrote : [snipped explanation of packet families vs. address families] >But that's right, any known implementation of the socket layer #define >address families like protocol families. >To answer your question, I think that BeOS is not conforming to the socket >standard, because it lacks the abstraction of "address families". If you >don't have AF_XXX symbols in your include files, but if you've got >compatible system calls for sockets, you should better add custom >definitions which add missing definitions within your socket include files, >so that you will use AF_XXX wherever address nodes are designated, but not >PF_XXX which is used to select transport protocols. Making so, your source >will compile correctly on future socket implementations, and on other >systems. At least you should look at examples in "winsock.h" (for 16-bit >Windows 3.x) to know which AF_XXX constitutes the basic socket services. >The best thing then is to use code like: >#include <socket.h>... // any system includes for sockets... >#ifndef AF_INET >#define AF_INET PF_INET >#endif >within the C source that uses sockets, so that you won't modify the body of >C functions that SHOULD always use AF_XXX (not PF_XXX) where node addresses >are specified. >There is no guarantee that AF_INET and PF_INET will be kept the same in the >future, when IPv6 will become THE standard. There has been some misunderstanding : 1) What's lacking on BeOS is PF_xxx, not AF_xxx. Not much existing code would compile if AF_xxx did not exist ! 2) The two lonely instances of PF_INET in the PHP source are IMHO a case of being uselessly zealous because the rest of the code uses AF_INET (for example, fsock.c) so at best it is an inconsistency. Cheers, --Jonathan