+ * Otherwise, if "sigset()" is available, it probably has BSD semantics
+ * while "signal()" has traditional semantics, so we use "sigset()"; it
+ * might cause system calls to be restarted for the signal, however.
+ * I don't know whether, in any systems where it did cause system calls to
+ * be restarted, there was a way to ask it not to do so; there may no
+ * longer be any interesting systems without "sigaction()", however,
+ * and, if there are, they might have "sigvec()" with SV_INTERRUPT
+ * (which I think first appeared in 4.3BSD).
+ *
+ * Otherwise, we use "signal()" - which means we might get traditional
+ * semantics, wherein system calls don't get restarted *but* the
+ * signal handler is reset to SIG_DFL and the signal is not blocked,
+ * so that a subsequent signal would kill the process immediately.
+ *
+ * Did I mention that signals suck? At least in POSIX-compliant systems
+ * they suck far less, as those systems have "sigaction()".