[Bug tree-optimization/83312] [8 regression] bogus -Warray-bounds warning

[law at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83312

Jeffrey A. Law <law at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at redhat dot com
           Assignee|unassigned at gcc dot gnu.org      |law at redhat dot com

--- Comment #4 from Jeffrey A. Law <law at redhat dot com> ---
It's actually not jump threading that figures things out here.   VRP uses SCEV
to identify J as an induction variable.  So during substitute_and_fold it
transforms the controlling conditional from:

Folding statement: if (j_9 < 0)
Folding predicate j_9 < 0 to 0
Folded into: if (0 != 0)


Which makes the path with the problem code unreachable before we even look at
jump threading.  But the problem block still in the IL and it's considered
reachable.

I think we can avoid the warning by simply clearing EDGE_EXECUTABLE on the
appropriate edge when we simplify the conditional.  That won't capture
secondary effects though.  We could go further and have a step that propagates
EDGE_EXECUTABLE just before we do the array bounds warnings.  I'm not sure if
it's worth the effort.  Thoughts?