[Bug middle-end/94992] New: gcc thinks a member variable is uninitialised

[Bug middle-end/94992] New: gcc thinks a member variable is uninitialised

[rafael at espindo dot la]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94992

            Bug ID: 94992
           Summary: gcc thinks a member variable is uninitialised
           Product: gcc
           Version: 10.1.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: rafael at espindo dot la
  Target Milestone: ---

Created attachment 48477
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=48477&action=edit
testcase

In the attached testcase gcc with -O1 (but not -O0 or -O2) produces the
following warning

futures_test.cc:57:22: warning: ‘f2.my_future::_promise’ is used uninitialized
in this function [-Wuninitialized]
   57 |         : _promise(x._promise)

The produced assembly also uses uninitialized memory.

The problem is that _promise is initialized to nullptr in the default
constructor.

[Bug middle-end/94992] gcc thinks a member variable is uninitialised

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94992

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Comment on attachment 48477
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=48477
testcase

>    my_future f2;
>    new (&f2) my_future(std::move(f2));

This is going to be problematic as f2 is being created but you are using the
old object which has the same address as the new one.

[Bug middle-end/94992] gcc thinks a member variable is uninitialised

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94992

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

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

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
AS mentioned.

[Bug middle-end/94992] gcc thinks a member variable is uninitialised

[rafael at espindo dot la]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94992

--- Comment #3 from Rafael Avila de Espindola <rafael at espindo dot la> ---
For completeness, this is a reduction of a std::swap(x,x). The placement new
was originally in the move assignment operator.

I was able to reproduce this with gcc 9 by moving a few functions out of line.

As far as I can tell, the relevant standard line is from basic.life:

The lifetime of an object \placeholder{o} of type \tcode{T} ends when:
...
\item the storage which the object occupies is released,
or is reused by an object that is not nested within
\placeholder{o}\iref{intro.object}.