From: Jeff King <peff@peff.net>
Date: Thu, 21 May 2009 04:18:12 GMT
Subject: Re: [PATCH] setup_revisions(): do not access outside argv
Message-ID: <20090521041812.GE8091@sigill.intra.peff.net>
In-Reply-To: <7v7i0btdwu.fsf@alter.siamese.dyndns.org>

On Wed, May 20, 2009 at 06:58:41PM -0700, Junio C Hamano wrote:

> If a function takes (int ac, char **av), then people should be able to
> depend on the usual convention of
> 
>  (1) for any i < ac, av[i] is not NULL; and
>  (2) av[ac] is NULL.
> 
> With your patch, a broken caller's wish is simply discarded and nobody
> will notice.  Without your patch, at least you will know that the caller
> passed an inconsistent pair of ac and av to this function by seeing a
> coalmine canary segfault.
> 
> I would not mind a patch that adds an assertion that protects this
> function from broken callers, so that we can find them, but your patch
> makes me feel very uneasy.

I agree.

Having just fixed a segfault in the GIT_TRACE code caused by a
non-terminated argv generated by the alias code, I think I would prefer
that we just consistently do the NULL-termination. You are otherwise
creating a maintenance pitfall when somebody later passes the value to
unsuspecting code.

-Peff