On Wed, Apr 23, 2025 at 05:42:36PM +0000, Listeria monocytogenes wrote: > I just tried it on linux and the connection is closed even if we don't > request it
I don't know if this is documented or not but Xlib uses xcb internally and xcb opens the connection with CLOEXEC: https://lists.freedesktop.org/archives/xcb/2013-March/008207.html > I think it makes more sense to use posix_spawn_file_actions_* if we're > using posix_spawn It really doesn't. As you can already see, posix_spawn interface is absolutely garbagely designed for anything non-trivial. The whole posix_spawn_file_actions_t thing should've been an array where the user puts his "commands" in, instead of some opaque struct that needs to be error checked for every single call. In any case, explicitly closing the xfd would be consistent with the previous state so I agree with doing it. But you should use CLOEXEC like your first patch instead of posix_spawn_file_actions_*. Suggestion (untested): pid_t pid; extern char **environ; int err = 0; int xfd = ConnectionNumber(dpy); int flags = fcntl(xfd, F_GETFD); if (flags < 0 || fcntl(xfd, F_SETFD, flags | FD_CLOEXEC) < 0) err = errno; if (!err) err = posix_spawnp(&pid, argv[0], NULL, NULL, argv, environ); if (err) { die("slock: failed to execute post-lock command: %s: %s\n", - NRK
