Mailing-List: contact dev-help@apr.apache.org; run by ezmlm
Precedence: bulk
Sender: trawick@bellsouth.net
To: dev@apr.apache.org
Subject: Re: cvs commit: apr/network_io/unix sendrecv.c
References: <20010503223802.28219.qmail@apache.org>
	<20010503162219.C899@waka.ebuilt.net>
From: Jeff Trawick <trawickj@bellsouth.net>
Date: 03 May 2001 21:31:34 -0400
In-Reply-To: <20010503162219.C899@waka.ebuilt.net>
Message-ID: <m3lmod7vd5.fsf@adsl-77-241-65.rdu.bellsouth.net>
Lines: 45
User-Agent: Gnus/5.0808 (Gnus v5.8.8) Emacs/20.3
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii

"Roy T. Fielding" <fielding@ebuilt.com> writes:

> >   Index: sendfile.c
> >   ===================================================================
> >   RCS file: /home/cvs/apr/test/sendfile.c,v
> >   retrieving revision 1.12
> >   retrieving revision 1.13
> >   diff -u -r1.12 -r1.13
> >   --- sendfile.c	2001/03/31 22:37:13	1.12
> >   +++ sendfile.c	2001/05/03 22:37:58	1.13
> >   @@ -373,6 +373,7 @@
> >                printf("apr_sendfile()->%d, sent %ld bytes\n", rv, (long)tmplen);
> >                if (rv) {
> >                    if (APR_STATUS_IS_EAGAIN(rv)) {
> >   +                    assert(tmplen == 0);
> >                        nsocks = 1;
> >                        tmprv = apr_poll(pfd, &nsocks, -1);
> >                        assert(!tmprv);
> 
> No assert should ever be present in server code, period.  I don't care if
> the only way it could be triggered is by a cosmic ray hittng a memory cell
> at just the wrong moment in time, it doesn't belong in the server
> code.

My first take was to suggest that you stop being excited about an
assert() in a test program.  But then I disagree 100% with your
comments on assertions :)  I practically always prefer that software
go out with a bang rather than fail silently, even when I don't have
time or don't want to complicate the logic by handling unanticipated
conditions more carefully than with an assertion.

When I used to work on mission critical code we had a fair number of
asserts, usually added when they could be done so with little extra
pathlength... They are infinitely more reliable than any comment and
they also acknowledge the fact that shit happens and you need to
figure out where things went south.

I like that Apache has asserts that only work with AP_DEBUG builds as
well as asserts that are always active.  Some AP_DEBUG asserts added
by Greg Ames and others have helped us get some nasty bugs out.

-- 
Jeff Trawick | trawickj@bellsouth.net | PGP public key at web site:
       http://www.geocities.com/SiliconValley/Park/9289/
             Born in Roswell... married an alien...