Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83; site wlbr.UUCP
Path: utzoo!watmath!clyde!akgua!sdcsvax!bmcg!cepu!trwrba!trwrb!scgvaxd!wlbr!jm
From: jm@wlbr.UUCP
Newsgroups: net.unix-wizards
Subject: Re: problems with alarm(0)
Message-ID: <173@wlbr.UUCP>
Date: Fri, 25-May-84 17:33:24 EDT
Article-I.D.: wlbr.173
Posted: Fri May 25 17:33:24 1984
Date-Received: Fri, 1-Jun-84 21:07:32 EDT
References: <179@philmds.UUCP>
Organization: Eaton ES, Westlake Village, CA
Lines: 35

x
johan@philmds complains that bracketing a read/write with alarm()s
may fail because the SIGALRM may come in even after an alarm(0) call.

My customary way of preventing this is to ignore the SIGALRM before
calling alarm(0).

	onalarm()
	{
		signal(SIGALRM, onalarm);
		alarm(2);
	A:
	}

	{
		...
		signal(SIGALRM, onalarm);
		alarm(2);
	B:
		i = read(fd, buf, sizeof(buf));
ADDED LINE>>>>	signal(SIGALRM, SIG_IGN);
		alarm(0);
		if (i == -1 && errno == EINTR) {
			...  /* handle timeout */
		} else {
			...  /* handle data read */
		}
		...
	}

In this manner, even if the SIGALRM is given to the process, it will do
no harm.

		Jim Macropol
		{scgvaxd,ihnp4,trwrb,vortex}!wlbr!jm