Greetings So I've been researching on this particular topic for past few days and haven't yet reached a satisfactory conclusion - What would be a safe way to get/set sockaddr_in, sockaddr_un or other similar structures from the generic sockaddr_storage structure? I've read some textbooks including the recommended 'Unix Network Programming' by W.Richard Stevens and some others, and what I gathered from there looks something like - sockaddr_storage s; memset(&s, 0, sizeof(s)); sockaddr_in *p = (sockaddr_in *) &s; p->sin_family = AF_INET; p->port = htons(80); // now 's' is ready to be passed in socket api calls int res = bind(sockfd, (sockaddr *) &s, sizeof(s)); // and similar calls but recently I discovered that this method wasn't following strict aliasing rules and was in fact a violation. So I made some stackoverflow threads - [1] - http://stackoverflow.com/questions/42176962/fill-sockaddr-storage-struct-with-values-of-sockaddr-in [2] - http://stackoverflow.com/questions/42178179/will-casting-around-sockaddr-storage-and-sockaddr-in-break-strict-aliasing and read through official documentations, textbooks and other online resources but couldn't get my doubts clear. getaddrinfo only supports ipv4/v6 and therefore isn't a solution for me since I'm writing code that will support many other socket domains eg unix, netlink etc.. Another recommended solution was to first declare appropriate structure type, fill it - struct sockaddr_in sin; sin.sin_family = AF_INET; sin.sin_port= htons(port); and then use memcpy to transfer the data to original sockaddr_storage structure memcpy(reinterpret_cast<char *>(&s), reinterpret_cast<char *>(&sin), sizeof(sin)); and now pass s to socket api function calls which can deduce the structure themselves.. but can they? Would this also work in reverse, like sending sockaddr_storage s to recvfrom call and then checking its ss_family field, if it is AF_INET then create sockaddr_in and memcpy data back to it, if sockaddr_in6 do memcpy to it and similarly for other domains.. I need to do it using sockaddr_storage structure otherwise, I'll have to write tons of overloaded methods for each socket domain and maintenance will become more expensive compared to a single generic structure which was already designed to do so. Afaik Microsoft's compiler don't suffer from this problem since they explicitly allow these structures to be casted and no concept of strict violation is present. In case of clang, I'm not sure if they've strict aliasing violation or not, but they do tend to follow gcc so that's where I'll put my guess. Any sort of help would be beneficial to me, I'm open to suggestions/advice as well Abhinav Gauniyal
