[Bug tree-optimization/106922] [12/13 Regression] Bogus uninitialized warning on boost::optional<<std::vector<std::string>>>

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106922

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2
   Last reconfirmed|                            |2022-09-13
             Status|UNCONFIRMED                 |ASSIGNED
   Target Milestone|---                         |12.3
             Blocks|                            |24639
     Ever confirmed|0                           |1
           Keywords|                            |diagnostic,
                   |                            |missed-optimization
            Summary|[12 Regression] Bogus       |[12/13 Regression] Bogus
                   |uninitialized warning on    |uninitialized warning on
                   |boost::optional<<std::vecto |boost::optional<<std::vecto
                   |r<std::string>>>            |r<std::string>>>
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed also on trunk and also with -O1 and the usual suspects -fno-ivopts
-fno-thread-jumps.

The question is whether

        external = internal;

is problematic.  For example

struct MyStruct { };

std::vector<MyStruct> getMyStructs();

void test()
{
    Optional<std::vector<std::string>> external;
    Optional<std::vector<std::string>> internal;
    external = internal;
}

is diagnosed with -std=c++20 -O -Wuninitialized -fno-tree-fre
-fno-tree-dominator-opts

that's because we then no longer forward the uninit state of the Optionals
and thus diagnose the conditional (on initialized)

  <bb 6> [local count: 118702158]:
  _34 = MEM[(struct vector *)&external + 8B].D.88471._M_impl.D.87778._M_finish;
  _35 = MEM[(struct vector *)&external + 8B].D.88471._M_impl.D.87778._M_start;
  if (_34 != _35)
    goto <bb 41>; [89.00%]
...

the same happens with the unreduced testcase, there we have similar code
conditional on external.m_initialized where we are not able to forward
the m_initialized state.

It might be the bisected rev. causes a missed optimization here, I will have
to check.  At least I don't see a good reason why we don't forward it here.


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=24639
[Bug 24639] [meta-bug] bug to track all Wuninitialized issues