[Bug tree-optimization/106093] [12/13 Regression] False positive -Wstringop-overflow with -O3 when resizing std::vector

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106093

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[Regression] False positive |[12/13 Regression] False
                   |-Wstringop-overflow with    |positive
                   |-O3 when resizing           |-Wstringop-overflow with
                   |std::vector                 |-O3 when resizing
                   |                            |std::vector
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2023-01-16
             Status|UNCONFIRMED                 |NEW
                 CC|                            |jwakely.gcc at gmail dot com
      Known to work|                            |11.3.0
           Priority|P3                          |P2
   Target Milestone|---                         |12.3

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.

(gdb) p debug_gimple_stmt (stmt)
# .MEM_120 = VDEF <.MEM_81>
MEM <vector(8) char> [(char *)vectp.79_117] = vect__18.77_116;

it's possibly a missed optimization for the vectorizer introduced compute
of the number of iterations.  We have

<bb 7> [local count: 58465242]:
_23 = operator new (1);
*_23 = 0;
__cur_29 = _23 + 1;
if (_3 != _4)
  goto <bb 8>; [89.00%]

<bb 8> [local count: 52034065]:
_15 = (unsigned long) _3;
_38 = (unsigned long) _4;
_36 = _15 - _38;
_54 = _36 + 18446744073709551615;
_10 = _54 > 6;
if (_10 != 0)
  goto <bb 9>; [64.00%]

so we allocate 1 byte but then compute the iteration as difference from _4 and
_3 which are computed from

_3 = m_stream.D.31893._M_impl.D.31166._M_finish;
_4 = m_stream.D.31893._M_impl.D.31166._M_start;

in particular the new allocated storage is processed but the old size is used?!