changeset 24121:5395b18f9f08

Revert my revision 849d4f7265598a9f0340411c4c0c0401d488ec3b, which removed the select() code in child DNS processes. Stu pointed out that this code is what allowed our child DNS processes to hang around for 40 seconds waiting for additional requests, then die a natural death. But that wasn't happening even WITH the select code because the parent was killing the DNS children when it was done with them. So I made another change to set the resolver to NULL so that it isn't killed by purple_dnsquery_destroy(). I'm assuming that we still want our DNS lookup children to hang around for a little while after they're done. I reduced the timeout from 40 seconds to 20 seconds. An arguably better way to do this is to go back to having the child block on read() instead of calling select(), then have the parent set a timer that kills the child after a certain about of time. But I don't see an advantage to doing it either way, and this is simpler.

--- a/libpurple/dnsquery.c	Tue Sep 16 17:30:23 2008 +0000
+++ b/libpurple/dnsquery.c	Tue Sep 16 17:56:01 2008 +0000
@@ -108,6 +108,13 @@
 		}
 	}
 
+	/*
+	 * Set the resolver to NULL so that it doesn't get killed so that
+	 * it sits around waiting for additional DNS requests for a few
+	 * seconds longer.
+	 */
+	query_data->resolver = NULL;
+
 	purple_dnsquery_destroy(query_data);
 }
 
@@ -209,6 +216,16 @@
 	 * the result back to our parent, when finished.
 	 */
 	while (1) {
+		fd_set fds;
+		struct timeval tv = { .tv_sec = 20, .tv_usec = 0 };
+		FD_ZERO(&fds);
+		FD_SET(child_in, &fds);
+		rc = select(child_in + 1, &fds, NULL, NULL, &tv);
+		if (!rc) {
+			if (show_debug)
+				printf("dns[%d]: nobody needs me... =(\n", getpid());
+			break;
+		}
 		rc = read(child_in, &dns_params, sizeof(dns_params_t));
 		if (rc < 0) {
 			fprintf(stderr, "dns[%d]: Error: Could not read dns_params: "