Mailing-List: contact dev-help@subversion.apache.org; run by ezmlm
Precedence: bulk
Date: Thu, 13 Aug 2015 23:25:17 +0000
From: Daniel Shahaf <d.s@daniel.shahaf.name>
To: dev@subversion.apache.org
Subject: Re: Review of sizeof usage
Message-ID: <20150813232517.GH1998@tarsus.local2>
References: 
 <CA+t0gk0QV_2mwEmK-OPjoP9BzXCPPiKccW3Wg3tBwPmQcLUgcQ@mail.gmail.com>
 <87pp2tvpfc.fsf@wandisco.com>
 <55CA5344.8090005@wandisco.com>
 <20150811223111.GH1859@tarsus.local2>
 <55CAFED5.4030109@wandisco.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <55CAFED5.4030109@wandisco.com>
User-Agent: Mutt/1.5.21 (2010-09-15)

Branko Čibej wrote on Wed, Aug 12, 2015 at 10:07:49 +0200:
> On 12.08.2015 00:31, Daniel Shahaf wrote:
> >
> >>> We have had problems with both styles in the past, so neither is immune
> >>> to bugs.  I prefer the explicit type as it is easier to grep.
> >> The explicit type form is more accident-prone than the variable form
> >> because any change requires two modifications in the same statement
> >> instead of one.
> > Why doesn't the compiler or buildbot catch accidents?
> 
> I can't imagine a way for the compiler to emit warnings for such
> constructs without getting a far too large percentage of false
> positives. It's perfectly valid, and in many cases required by some
> object-like architecture, to allocate a buffer that has a different size
> than the one implied by the pointer that stores the return value. This
> is C, after all.
> 

Okay, so from the compiler authors' perspective, "allocation size mismatches
pointed-to-object size" warnings should not be on by default.  Fair enough.
But from our perspective as Subversion maintainers, we never *intentionally*
allocate a buffer smaller than the pointed-to object, so the warnings would be
useful to us.  We should therefore opt-in to them.

(via compiler flags, or have buildbot run static analysis, or…)

Cheers,

Daniel

> -- Brane