[Mesa-dev] question about container_of
emil.l.velikov at gmail.com
Mon Feb 27 13:26:11 UTC 2017
On 27 February 2017 at 12:08, Julien Isorce <julien.isorce at gmail.com> wrote:
> Since 2012 commit ccff74971203b533bf16b46b49a9e61753f75e6c it is said:
> "sample must be initialized, or else the result is undefined" in the
> description of mesa/src/util/list.h::container_of .
> But I can find a few places where it is used without initializing that
> second parameter, i.e. like:
> struct A a;
> container_of(ptr, a, member);
> Then I can add the "= NULL" but should it be just
> container_of(ptr, struct A, member);
> like in the kernel and some other places in mesa ?
Strictly peaking these are toolchain (ASAN iirc) bugs, since there is
no pointer deref, as we're doing pointer arithmetic.
Afaict the general decision was to to merge the patch(es) since they
will make actual bugs stand out amongst the noise. In the long run,
it's better to fix the tool (ASAN/other) than trying to "fix" all the
cases in mesa and dozens of other projects. But until then patches are
safe enough ;-)
That's my take on it, at least.
More information about the mesa-dev