You're right that can't post to a Racket semaphore in a signal handler. The way to allocate a global and make it available to Racket code is to create a new kind of event; the 2006 post is still accurate. In particular, the new event's polling function (which is called by the scheduler) can check the global. Calling scheme_signal_received() from a signal handler ensures that the event's polling function is called after the global is changed.
A self-pipe is also a fine approach, since the event wrapper for a pipe is already available via scheme_make_fd_input_port(). At Sat, 20 Dec 2014 14:05:32 -0500, Tony Garnock-Jones wrote: > Hi all, > > If I wanted to catch a Unix signal (say, SIGUSR1) within my Racket > program, how should I go about it? > > My current best guess is: > - load a C extension, which > - creates a semaphore and > - hooks the signal with a handler (MZ_SIGSET looks relevant), which > - posts to the semaphore when the signal arrives. > > However this seems like a recipe for problems. Is it even permitted to > post to a semaphore from a signal handler? Will the runtime be upset > about unscheduled intrusions into its demesne? Is it possible for a C > extension to allocate a global value like a semaphore and make it > available to Racket code? > > Is there a better way? > > - Perhaps a self-pipe would be better than a semaphore? > - A thread from 2006 [1] points at scheme_signal_received() but > I don't understand why creating an event type is the right thing > to do. > > Tony > > [1] http://lists.racket-lang.org/users/archive/2006-July/013999.html > > ____________________ > Racket Users list: > http://lists.racket-lang.org/users ____________________ Racket Users list: http://lists.racket-lang.org/users
