Xref: utzoo alt.msdos.programmer:1188 comp.sys.ibm.pc:44009
Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!uunet!lll-winken!bu.edu!bu-cs!husc6!ddl
From: ddl@husc6.harvard.edu (Dan Lanciani)
Newsgroups: alt.msdos.programmer,comp.sys.ibm.pc
Subject: new MSC 5.1 bug?
Keywords: bug MSC 5.1 loops
Message-ID: <1537@husc6.harvard.edu>
Date: 9 Feb 90 00:25:57 GMT
Organization: Harvard University, Cambridge MA
Lines: 32


	The following useless program illustrates what I think is
a bug in the loop optimization code.  It appears that the compiler
becomes confused and believes that p1 and p2 are equal even when
the code within the if statement is not executed.  Compiling with
-Od produces correct results.  In case it isn't obvious, this
loop abstracts a search (with intent to delete) from a linked list.
If this example is perhaps too far reduced in content, the original
bug tripper can be found in the Sun rpc portmapper daemon's
deregistration function.

				Dan Lanciani
				ddl@harvard.*

struct foo {
	int a, b;
	struct foo *next;
} protofoo = { 1, 2 }, *foo = &protofoo;

main()
{
	struct foo *p1, *p2;

	for (p1 = 0, p2 = foo;;) {
		if (p2->a != protofoo.a || p2->b != protofoo.b) {
			p1 = p2;
			p2 = p2->next;
			continue;
		}
		printf("p1 = %x, p2 = %x\n", p1, p2);
	}
}