sigset botch

[utzoo!decvax!duke!mcnc!unc!dopey.smb]

If a sighold is done, a subsequent sigset call will return SIG_HOLD
rather than a routine address.  An attempt to use this value will store
a bad address -- 3! -- so that any occurrence of the signal will send
the program flaming down the tubes.  That is, a sequence like this will
fail (miserably):


	sighold(SIGINT);
	...
	s = sigset(SIGINT, myint);
	....
	sigset(SIGINT, s);


It's a problem because closed subroutine often like to take their own
actions with knowing the current state of affairs.  I've tripped over
this in some versions of ucbmail.