Re: Re : "Pointer arithmetic using Null pointer in gawk-3.1.0 testsuite"

[Paul Eggert]

Olatunji Oluwabukunmi Ruwase <address@hidden> writes:

> I am working on a bounds checking patch to gcc. While testing my
> patch on gawk-3.1.0

Gawk 3.1.2 doesn't have the code that you mentioned, so perhaps the
problem has been fixed.  You might try Gawk 3.1.2 with your bounds
checker to see how it fares.  It's in:
<ftp://ftp.gnu.org/gnu/gawk/gawk-3.1.2.tar.gz>

> <                     char *oldsplit = iop->buf + iop->secsiz;

Just out of curiosity, is iop->secsiz equal to zero when the problem
occurs?

If so, it is a very common C extension that NULL + 0 yields NULL.
I don't know of any architecture of interest to Gawk where this
extension does not hold.

Optionally, I suppose a bounds checker might be especially pedantic
and consider NULL + 0 to be an error.  However, I've used NULL + 0 ==
NULL myself in code that is quite portable in practice.  So I would
prefer a bounds checker that allows NULL + 0 by default, though I
suppose it might be nice to have a --pedantic option that disallows
NULL + 0.