https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65686
--- Comment #11 from Richard Biener <rguenth at gcc dot gnu.org> --- So we can simplify the fix for PR13962 by only considering pointer with decl compares which points-to should handle well enough (and not fall into the issue of bogus ptr1 vs. ptr2 simplifications because we don't properly track whether ptr1 or ptr2 may be NULL). ISTR such checks also happen in libstdc++ code. Not sure if appropriate at this stage, I would tend to a no here. I'll queue re-visiting PR13962 during next stage1.