[Bug tree-optimization/110091] [13/14/15/16 Regression] bogus -Wdangling-pointer on non-pointer values

[siddhesh at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110091

--- Comment #12 from Siddhesh Poyarekar <siddhesh at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #11)
> (In reply to Andrew Pinski from comment #10)
> > because we dont have a marker that starts the storage .
> > The storage is started when the address is taken or a store happens to it.
> > The above ir can happen via unrolling even.

OK, that explains the decision to run this before unrolling, from the
implementation:

/* Check for dangling pointers in the earliest runs of the pass.
   The latest point -Wdangling-pointer should run is just before
   loop unrolling which introduces uses after clobbers.  Most cases
   can be detected without optimization; cases where the address of
   the local variable is passed to and then returned from a user-
   defined function before its lifetime ends and the returned pointer
   becomes dangling depend on inlining.  */

Is it possible for uses to appear after CLOBBERs before unrolling too?

> Checkout cfgexpand.cc and add_scope_conflicts* functions for how the part of
> GCC that is used for correctness implements that.

Thanks, I'll take a look.