[Bug middle-end/101600] [12 Regression] Spurious -Warray-bounds downcasting a polymorphic pointer

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101600

Martin Sebor <msebor at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Blocks|                            |56456
   Target Milestone|---                         |12.0
     Ever confirmed|0                           |1
                 CC|                            |msebor at gcc dot gnu.org
   Last reconfirmed|                            |2021-07-23
            Summary|Spurious -Warray-bounds     |[12 Regression] Spurious
                   |                            |-Warray-bounds downcasting
                   |                            |a polymorphic pointer
      Known to work|                            |11.1.0
           Assignee|unassigned at gcc dot gnu.org      |msebor at gcc dot gnu.org
             Status|UNCONFIRMED                 |ASSIGNED
      Known to fail|                            |12.0

--- Comment #1 from Martin Sebor <msebor at gcc dot gnu.org> ---
Thanks for the small test case!  I can confirm the false positive.  The IL
looks (close to) the following:

  <bb 5> [local count: 80404472]:
  # p_1 = PHI <p_9(D)(4)>
  if (p_1 != 0B)
    goto <bb 6>; [100.00%]
  else
    goto <bb 7>; [0.00%]

  <bb 6> [local count: 80404472]:
  iftmp.1_11 = &MEM[(struct S4 *)p_1 + -8B].D.2419;   <<< -Warray-bounds

except that p_1 is:

  p_1 = PHI <0B(3), p_9(D)(4)>

The bug is in access_ref::get_ref () in not clearing the base0 flag, making
-Warray-bounds think the pointer points to the first byte of a declared object.


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56456
[Bug 56456] [meta-bug] bogus/missing -Warray-bounds