On 20/03/20 17:45, Guy Harris wrote:
On Mar 20, 2020, at 8:09 AM, João Valverde <joao.valverde@xxxxxxxxxxxxxxxxxx> wrote:
On 19/03/20 18:38, Guy Harris wrote:
This isn't unique to Windows. It dates back to old BSD, in which struct in_addr contained a union of multiple different types for an IP address, with some types being structures breaking up the address into host and network bits, and even included bits for IMP numbers. s_addr was defined to be the member of the union that just defined an address as a 32-bit integer, so if you referred to the s_addr "field" of the structure it gave you the 32-bit integer value.
Because POSIX defines struct in_addr as an opaque structure with an s_addr element, some BSD Socket implementations get creative with the use of unions and use a macro definition for "s_addr", which is terribly bad practice and a tremendously ugly botch.
One such implementation was in an obscure OS called "4.2BSD":
https://www.tuhs.org/cgi-bin/utree.pl?file=4.2BSD/usr/src/sys/netinet/in.h
4.2BSD came out in 1982, and the first version of POSIX came out in 1988 - and, as I remember, it had no networking APIs in it - so you can't says "they did that because POSIX let them do that". It's more like "POSIX allows that because some UN*Xes didn't discard that 4.2BSDism".
And the general idea of using unions to overlay a 32-bit integer version of an IP address and various structure versions showing pre-CIDR divisions of IP addresses dates back to the BBN TCP/IP:
https://www.tuhs.org/cgi-bin/utree.pl?file=BBN-Vax-TCP/bbnnet/net.h
Interesting bit of archeology. You're right that the practice predates
POSIX with the BSDs. The point is that the inevitable name collisions
are an obvious botch.
