From owner-freebsd-hackers  Tue May 28 12:23:27 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id MAA22931
          for hackers-outgoing; Tue, 28 May 1996 12:23:27 -0700 (PDT)
Received: from palmer.demon.co.uk (palmer.demon.co.uk [158.152.50.150])
          by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id MAA22925
          for <hackers@FreeBSD.ORG>; Tue, 28 May 1996 12:23:22 -0700 (PDT)
Received: from palmer.demon.co.uk (localhost [127.0.0.1])
	  by palmer.demon.co.uk (sendmail/PALMER-1) with ESMTP id UAA13118
	  for <hackers@freebsd.org>; Tue, 28 May 1996 20:19:50 +0100 (BST)
To: hackers@FreeBSD.ORG
CC: scrappy@ki.net
From: "Gary Palmer" <gpalmer@FreeBSD.ORG>
Subject: I HATE OPTIMISING COMPILERS
Date: Tue, 28 May 1996 20:19:48 +0100
Message-ID: <13116.833311188@palmer.demon.co.uk>
Sender: owner-hackers@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk


Well, I certainly do at the minute.

Take the following innocuous piece of code:

    if (pdu->version == SNMP_DEFAULT_VERSION)
	pdu->version = session->version;

    if (pdu->version == SNMP_DEFAULT_VERSION){
	fprintf(stderr, "No version specified\n");
	snmp_errno = SNMPERR_BAD_ADDRESS;
	return 0;
    }

Looks fine, right?

Sorry, you lose (and so did I, well over an hour of time tracking down
this damned problem).

Every time you run that piece of code, you get "No version specified".
So I go and stick in a nice printf just above the first if statement
to check that it's not doing something funny. The code now becomes:

    printf("pdu->version = %d\nsession->version = %d\nSNMP_DEFAULT_VERSION = %d\n",
	   pdu->version, session->version, SNMP_DEFAULT_VERSION);

    if (pdu->version == SNMP_DEFAULT_VERSION)
	pdu->version = session->version;

    if (pdu->version == SNMP_DEFAULT_VERSION){
	fprintf(stderr, "No version specified\n");
	snmp_errno = SNMPERR_BAD_ADDRESS;
	return 0;
    }

Which produces:

pdu->version = -1
session->version = 2
SNMP_DEFAULT_VERSION = -1
No version specified

So, okay, session->version != SNMP_DEFAULT_VERSION, but the problem is
the aggregate of those two if statements disagrees. Hmm. Okay. Put in
a second printf:

    printf("pdu->version = %d\nsession->version = %d\nSNMP_DEFAULT_VERSION = %d\n",
	   pdu->version, session->version, SNMP_DEFAULT_VERSION);

    if (pdu->version == SNMP_DEFAULT_VERSION)
	pdu->version = session->version;
 
    printf("pdu->version = %d\nsession->version = %d\nSNMP_DEFAULT_VERSION = %d\n",
	   pdu->version, session->version, SNMP_DEFAULT_VERSION);

    if (pdu->version == SNMP_DEFAULT_VERSION){
	fprintf(stderr, "No version specified\n");
	snmp_errno = SNMPERR_BAD_ADDRESS;
	return 0;
    }

And all of a sudden IT STARTS WORKING PROPERLY. Take out the two
printfs, a nasty idea forming in my mind. The code now becomes:

    if (pdu->version == SNMP_DEFAULT_VERSION)
	pdu->version = session->version;

    session->version = pdu->version;

    if (pdu->version == SNMP_DEFAULT_VERSION){
	fprintf(stderr, "No version specified\n");
	snmp_errno = SNMPERR_BAD_ADDRESS;
	return 0;
    }

Which looks kinda daft, but again, the code worked as expected. Take
out the ``session->version = pdu->version;'' statement, and it starts
failing again. Now I'm really suspicious. Go to the Makefile. Change
``CFLAGS=-O -g -Dfreebsd2'' to be just ``CFLAGS=-Dfreebsd2''. make
clean all. Go and rebuild the client apps that depend on the
library.

BINGO. Much cheering and celebrating. I go to the fridge and get
another coke.

Anyone want to help me take GCC out the back kicking and screaming to
face the shooting squad? Admittedly that code isn't the most elegant
way of doing that assignment/test, but GCC should NOT produce such
bogus code from it :-( And here we (or at least I) thought that `-O'
was safe to use :-(

``Sigh.''

Gary
--
Gary Palmer                                          FreeBSD Core Team Member
FreeBSD: Turning PC's into workstations. See http://www.FreeBSD.ORG/ for info