[Bug c++/103375] New: [12 regression] -Werror=uninitialized false positive on unittest-cpp

[Bug c++/103375] New: [12 regression] -Werror=uninitialized false positive on unittest-cpp

[slyfox at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103375

            Bug ID: 103375
           Summary: [12 regression] -Werror=uninitialized false positive
                    on unittest-cpp
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: slyfox at gcc dot gnu.org
  Target Milestone: ---

Initially build failure noticed on unittest-cpp-2.0.0. I suspect it's a week
old regression.

Here is the minimal extracted example:

$ cat TestTestRunner.cpp
    struct TestReporter {};
    struct TestRunner
    {
        explicit TestRunner(TestReporter& reporter);
          // /*somewhere in .cpp:*/ : reporter_ (reporter) {}
        TestReporter & reporter_; // only store reference

        void RunAndReport();
    };
    struct FixtureBase
    {
        FixtureBase() : runner(reporter) {}
        TestRunner runner;
        TestReporter reporter;
    };
$ g++-12.0.0 -O2 -Wall -Werror -c TestTestRunner.cpp
    TestTestRunner.cpp: In constructor 'FixtureBase::FixtureBase()':
    TestTestRunner.cpp:12:32: error: member 'FixtureBase::reporter' is used
uninitialized [-Werror=uninitialized]
       12 |         FixtureBase() : runner(reporter) {}
          |                                ^~~~~~~~
    cc1plus: all warnings being treated as errors


I think progaram's intent is to only store an object's reference at
construction time. And use it only in the rest of methods after construction is
done.

$ LANG=C g++-12.0.0 -v
Using built-in specs.
COLLECT_GCC=/nix/store/fxn6ca5x1yb1xj8rskkvw5zajrnj8y79-gcc-12.0.0/bin/g++
COLLECT_LTO_WRAPPER=/nix/store/fxn6ca5x1yb1xj8rskkvw5zajrnj8y79-gcc-12.0.0/libexec/gcc/x86_64-unknown-linux-gnu/12.0.0/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with:
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 12.0.0 20211121 (experimental) (GCC)

[Bug c++/103375] [12 regression] -Werror=uninitialized false positive on unittest-cpp since r12-5391-g0790c8aacdfb4f

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103375

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |12.0
                 CC|                            |mpolacek at gcc dot gnu.org

[Bug c++/103375] [12 regression] -Werror=uninitialized false positive on unittest-cpp since r12-5391-g0790c8aacdfb4f

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103375

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2021-11-23
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Confirmed.

If you swap the order of runner and reporter you don't get the uninitialized
warning.


I do have to wonder if the warning should be improved in messaging here because
it is not obvious why this might be bad.
In this case maybe used before is initialized due to order of the variables or
something to that effect.

[Bug c++/103375] [12 regression] -Werror=uninitialized false positive on unittest-cpp since r12-5391-g0790c8aacdfb4f

[slyfox at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103375

--- Comment #2 from Sergei Trofimovich <slyfox at gcc dot gnu.org> ---
Created attachment 51855
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51855&action=edit
a.cpp.cpp

Also attaching unmodified preprocessed original file as I might have removed
too much. Original example uses inheritance with virtual functions.

gcc complains as:

$ g++-12.0.0 -std=c++14 -Werror=uninitialized -c a.cpp.cpp
a.cpp.cpp: In constructor '{anonymous}::FixtureBase::FixtureBase()':
a.cpp.cpp:23467:19: error: member '{anonymous}::FixtureBase::reporter' is used
uninitialized [-Werror=uninitialized]
23467 |          : runner(reporter)
      |                   ^~~~~~~~
cc1plus: some warnings being treated as errors

[Bug c++/103375] [12 regression] -Werror=uninitialized false positive on unittest-cpp since r12-5391-g0790c8aacdfb4f

[slyfox at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103375

--- Comment #3 from Sergei Trofimovich <slyfox at gcc dot gnu.org> ---
Proposed upstream workaround as
https://github.com/unittest-cpp/unittest-cpp/pull/185.

[Bug c++/103375] [12 regression] -Werror=uninitialized false positive on unittest-cpp since r12-5391-g0790c8aacdfb4f

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103375

Marek Polacek <mpolacek at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |INVALID
             Status|NEW                         |RESOLVED

--- Comment #4 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
The reduced test no longer warns.

The unreduced warns, but that seems to be a good thing, just switch the order
of these:

23480       TestRunner runner;
23481       RecordingReporter reporter;

Not a bug, IMHO.