Re: MemoryTracking internal adjustments
Date: Tue, 16 Aug 2011 18:58:04 +0200
Yang Tse <yangsita_at_gmail.com> wrote:
> 2011/8/16 Dan Fandrich wrote:
> > valgrind will do these checks already, and better, too, but valgrind is
> > not available on all platforms that curl supports. This feature could
> > catch problems that only show up on those platforms, so I'd hate to see
> > it be unceremoniously dropped.
> I admit that 'magic filling' might to some point serve its purpose on
> those builds which do not use valgrind and when the compiler being
> used is not inserting/using extra code to perform runtime checks to
> detect uninitialized pointers and vars.
> Main point in wanting to drop the 'magic filling' is that it defeats
> valgrind and compiler runtime checks of uninitialized pointers and
> Given that valgrind is generally available on linux systems would it
> be ok for you if the 'magic filling' got #ifdef'ed out when __linux__
> is defined at compile time?
valgrind is also available for FreeBSD on i386 and amd64
and at least on amd64 it's properly picked up by curl's test
- application/pgp-signature attachment: signature.asc