From: Jeff King <peff@peff.net>
Date: Sat, 08 Feb 2014 21:21:54 GMT
Subject: Re: [PATCH] builtin/blame.c::prepare_lines: fix allocation size of sb->lineno
Message-ID: <20140208212154.GA4283@sigill.intra.peff.net>
In-Reply-To: <87lhxmc4sr.fsf@fencepost.gnu.org>

On Sat, Feb 08, 2014 at 10:49:40AM +0100, David Kastrup wrote:

> But please note that since sb->lineno originally comes from a zeroed
> memory area and is passed to xrealloc, this requires that after
> 
> int *p;
> memset(&p, 0, sizeof(p));
> 
> the equivalence
> 
> ((void *)p == NULL)
> 
> will hold.  While this is true on most platforms, and while the C
> standard guarantees the slightly different
> ((void *)0 == NULL)
> is true, it makes no statement concerning the memory representation of
> the NULL pointer.
> 
> I have not bothered addressing this non-compliance with the C standard
> as it would be polishing a turd.  A wholesale replacement has already
> been proposed, and it's likely that this assumption is prevalent in the
> Git codebase elsewhere anyway.

Yes, we explicitly break this part of the standard in the name of
practicality (it simplifies frequently-used code, and machines on which
it matters are rare enough that nobody has ever complained about it).

So I do not think this is a problem.

However, is there a reason not to use:

  sizeof(*sb->lineno)

rather than

  sizeof(int)

to avoid type-mismatch errors entirely (this applies both to this patch,
and to any proposed rewrites using malloc).

-Peff