[Bug libmudflap/53952] [4.8 Regression] FAIL: libmudflap.c++/pass55-frag.cxx ( -O[123]) execution test

["rguenther at suse dot de" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53952

--- Comment #3 from rguenther at suse dot de <rguenther at suse dot de> 2012-11-29 09:26:31 UTC ---
On Thu, 29 Nov 2012, jakub at gcc dot gnu.org wrote:

> 
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53952
> 
> Jakub Jelinek <jakub at gcc dot gnu.org> changed:
> 
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>                  CC|                            |jakub at gcc dot gnu.org
> 
> --- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-11-29 07:57:12 UTC ---
> I'd say it is the right thing, just should be done at a different spot, instead
> of doing it for everything just check it for MEM_REF.  I think only MEM_REF can
> have ADDR_EXPR of a non-TREE_ADDRESSABLE DECL_P as argument, look e.g. at
> mem_ref_refers_to_non_mem_p in expr.c (which checks more than that), or
> opf_non_addressable and opf_not_non_addressable handling in
> tree-ssa-operands.c.
> So, MEM_REF with ADDR_EXPR of a non-TREE_ADDRESSABLE DECL_P IMHO doesn't need
> to be instrumented (seems I should modify asan/tsan too).

Yes, that's true.  Only MEM_REF and TARGET_MEM_REF can have ADDR_EXRPs
of sth non-addressable.  And yes, no instrumentation necessary.

Richard.