site_archiver@lists.apple.com Delivered-To: darwin-dev@lists.apple.com On Mar 22, 2005, at 10:32 AM, Tian-Jian Barabbas Jiang wrote: Hi all, mkfifo -m 0666 /tmp/pipe tail -f /tmp/pipe and then cat > /tmp/pipe > a > b > c cat > /tmp/pipe > d > e > f this time the 'tail -f' will show 'd', 'e', 'f' one by one immediately. Thank you for your patience to read this mail. :) Best Regards, Tian-Jian Jiang * Server.cpp #include <iostream> #include <fstream> #include <string> // for umask() and mkfifo() #include <sys/types.h> #include <sys/stat.h> using namespace std; int main() { umask(0); int returnValue = mkfifo("/tmp/pipe", 0666); previousString = outputString; in.close(); } } else cout << "mkfifo failed." << endl; return 0; } * Client.cpp #include <iostream> #include <fstream> #include <string> using namespace std; int main() { string inputString(""); ofstream out("/tmp/pipe"); while(cin >> inputString) { out << inputString << endl; if(inputString == "/quit") break; } out.close(); return 0; } _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev@lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/paulf%40aphrodite.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev@lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.appl... The behavior you are seeing is normal for any posix system. As far as your shell example goes, that's the effects of buffering. The reader won't read until a buffer is written with an underlying write () call to the pipe. When that happens is entirely dependent on the shell. The SIGPIPE's in your C code are normal. It means you tried to read from the read end of a pipe after the writer closed it. You can set SIGPIPE to SIG_IGN and have the read() operation fail with EPIPE instead. Note what you are trying to do will only work with one client at a time, and unless you arrange for some sort of handshaking, nothing will stop more than one client from trying and garbling your input. Normally two pipes are set up to handshake both ways between the client and the server. I've just wrote two small programs of client/server mechanism based on mkfifo, and I've encountered two strange situations. 1. The named pipe was unstable for my programs. I've experienced "Broken pipe." frequently while sending messages from the client. 2. Reading pipe was not that atomic. For example, If I just test by commands rather than handcraft programs, for the first try in another screen, 'tail -f' won't show anything until I leave 'cat'. But if I give it a try again, For my programs, if I want the server to read and to print the client's messages immediately, I have to repeat opening the file of named pipe. For more detail, please refer to comments in my programs below. And FYI, I also have tested these programs on FreeBSD 4.10 and FreeBSD 5.3, they have worked fine. I'm wondering if these are Darwin's problems, so I post this mail here. if(returnValue == 0) { string previousString(""); while(previousString != "/quit") { // It's easy to see the strange behavior in the following codes. I supposed the normal usage is to open file only once outside of the while-loop. // Unfortunately, I have to open the file again and again to make this program work on Darwin. But I don't need to do this on FreeBSD. ifstream in("/tmp/pipe"); string outputString(""); in >> outputString; if(outputString.length() > 0 && outputString != previousString) cout << "> " + outputString << endl; This email sent to paulf@aphrodite.com This email sent to site_archiver@lists.apple.com