[Bug c/99768] New: [11 Regression] Bogus -Wuninitialized diagnostic with type punning

[Bug c/99768] New: [11 Regression] Bogus -Wuninitialized diagnostic with type punning

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99768

            Bug ID: 99768
           Summary: [11 Regression] Bogus -Wuninitialized diagnostic with
                    type punning
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: acoplan at gcc dot gnu.org
  Target Milestone: ---

For the following code (which has a strict aliasing violation):

float foo(unsigned v)
{
    unsigned tmp = v;
    float *f = (float *)(&tmp);
    return *f;
}

at -O2 -Wall we warn that tmp is used uninitialized:

test.c: In function ‘foo’:
test.c:5:12: warning: ‘tmp’ is used uninitialized [-Wuninitialized]
    5 |     return *f;
      |            ^~
test.c:3:14: note: ‘tmp’ declared here
    3 |     unsigned tmp = v;
      |              ^~~

but this is clearly bogus. Not sure if this is strictly considered a bug given
the strict aliasing problem, but thought it might be worth checking in any
case.

[Bug c/99768] [11 Regression] Bogus -Wuninitialized diagnostic with type punning

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99768

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |11.0
             Blocks|                            |24639

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
I think the warning is correct - there is no valid initialization of a float
entity in 'tmp' (in fact GCC removed the initialization from 'v').  It might
be misleading but then what do you expect with invalid code?

We could, in theory, do another walk to find an initialization with strict
aliasing disabled and if we then find a candidate note that there are possible
initializations (we won't be able to prove they are definite inits) but
a violation of strict-aliasing might hide them.  OTOH this would pop up a lot
and likely confuse people to use -fno-strict-aliasing to simply hide the
warning but not fix the missing initialization.


Referenced Bugs:

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

[Bug middle-end/99768] [11 Regression] Bogus -Wuninitialized diagnostic with type punning

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99768

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2021-03-25
             Status|UNCONFIRMED                 |NEW
          Component|c                           |middle-end
     Ever confirmed|0                           |1
                 CC|                            |msebor at gcc dot gnu.org

--- Comment #2 from Martin Sebor <msebor at gcc dot gnu.org> ---
The -Wuninitialized was introduced in
g:b825a22890740f341eae566af27e18e528cd29a7.  I don't think I meant for this to
be diagnosed so I suspect it's a bug in the new code.  I see no basis in the IL
(below) to issue the warning.  GCC should issue -Wstrict-aliasing instead (it
does but only at level 2).

float foo (unsigned int v)
{
  float * f;
  unsigned int tmp;
  float _5;

  <bb 2> :
  tmp = v_2(D);
  f_4 = &tmp;
  _5 = *f_4;            << -Wuninitialized
  tmp ={v} {CLOBBER};
  return _5;

}

[Bug middle-end/99768] Unhelpful -Wuninitialized diagnostic with type punning

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99768

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement
   Target Milestone|11.0                        |---
            Summary|[11 Regression] Bogus       |Unhelpful -Wuninitialized
                   |-Wuninitialized diagnostic  |diagnostic with type
                   |with type punning           |punning

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Let's keep it open as enhancement since the diagnostic isn't very helpful to
the user - but it is still technically correct since GCC will simply remove
the initialization as it has no valid use.

[Bug middle-end/99768] Unhelpful -Wuninitialized diagnostic with type punning

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99768

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |vanyacpp at gmail dot com

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 110534 has been marked as a duplicate of this bug. ***