[Bug c++/94768] New: Wreturn-type should be error, not warning

[Bug c++/94768] New: Wreturn-type should be error, not warning

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94768

            Bug ID: 94768
           Summary: Wreturn-type should be error, not warning
           Product: gcc
           Version: 10.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: dcb314 at hotmail dot com
  Target Milestone: ---

For this C++ code:

void g( int);

struct S
{
        int a;
        int b;

        S & operator = ( const S & that)
        {
                g( that.a);
        }
};

G++ produces only a warning, although it does give us a clue
what it is expecting:

$ /home/dcb/gcc/results/bin/gcc -c -g -O2 -Wall apr26d.cc
apr26d.cc: In member function ‘S& S::operator=(const S&)’:
apr26d.cc:12:2: warning: no return statement in function returning non-void
[-Wreturn-type]
   11 |   g( that.a);
  +++ |+  return *this;
   12 |  }
      |  ^
$ 

I have to add -Werror=return-type to get compilation to stop.

$ /home/dcb/gcc/results/bin/gcc -c -Werror=return-type apr26d.cc
apr26d.cc: In member function ‘S& S::operator=(const S&)’:
apr26d.cc:12:2: error: no return statement in function returning non-void
[-Werror=return-type]
   11 |   g( that.a);
  +++ |+  return *this;
   12 |  }
      |  ^
cc1plus: some warnings being treated as errors
$ 

I'll have a look at switching on Werror=return-type permanently locally.

IMHO, for a C++ function returning non-void, a complete absence of any return
statement in the function really should produce an error.

Yes a warning is a diagnostic and so conforms with ISO C++ rules,
but finding obvious bugs at compile time is preferable to runtime.

[Bug c++/94768] Wreturn-type should be error, not warning

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94768

Jonathan Wakely <redi at gcc dot gnu.org> changed:

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

--- Comment #1 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to David Binderman from comment #0)
> IMHO, for a C++ function returning non-void, a complete absence of any
> return statement in the function really should produce an error.

This is a valid C++ program and it would be non-conforming to reject it. The
compiler can't even assume the program has undefined behaviour, because g(int)
could throw an exception, or do something that terminates, or simply never
return (e.g. go into an infinite loops with side effects like I/O).

[Bug c++/94768] Wreturn-type should be error, not warning

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94768

--- Comment #2 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to Jonathan Wakely from comment #1)
> This is a valid C++ program

s/program/translation unit/

[Bug c++/94768] Wreturn-type should be error, not warning

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94768

--- Comment #3 from David Binderman <dcb314 at hotmail dot com> ---
(In reply to Jonathan Wakely from comment #1)
> This is a valid C++ program and it would be non-conforming to reject it.

Surprising. The standard looks broken to me. Standards conformance
only really matters if GNU C++ is in standards conformance mode.

Most of the time, it is in its own GNU mode and so could do a more
useful job here rather than IMHO blindly following non-useful standards.

I checked the source code of the popular Fedora Linux distribution.
There are 32 examples of this problem in the C++ code, so they will
need fixing.

Interestingly, there are a whopping 258 examples in the C code, but that's a
separate issue.

[Bug c++/94768] Wreturn-type should be error, not warning

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94768

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Jonathan Wakely from comment #1)
> (In reply to David Binderman from comment #0)
> > IMHO, for a C++ function returning non-void, a complete absence of any
> > return statement in the function really should produce an error.
> 
> This is a valid C++ program and it would be non-conforming to reject it. The
> compiler can't even assume the program has undefined behaviour, because
> g(int) could throw an exception, or do something that terminates, or simply
> never return (e.g. go into an infinite loops with side effects like I/O).

Just to expand on this.  undefined at runtime behavior is not supposed to stop
compilation of a translation unit.

[Bug c++/94768] Wreturn-type should be error, not warning

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94768

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to David Binderman from comment #3)
> I checked the source code of the popular Fedora Linux distribution.
> There are 32 examples of this problem in the C++ code, so they will
> need fixing.
> 
> Interestingly, there are a whopping 258 examples in the C code, but that's a
> separate issue.

Note C and C++ have slightly different notice about "no return statement in
function returning non-void"; In C it is only undefined if the return value is
used while in C++ it is undefined the moment that the end of the function
happens.

[Bug c++/94768] Wreturn-type should be error, not warning

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94768

--- Comment #6 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to David Binderman from comment #3)
> Most of the time, it is in its own GNU mode and so could do a more
> useful job here rather than IMHO blindly following non-useful standards.

Nobody is "blindly following" anything. It's been discussed in detail by people
who know what they're talking about, over many years (including again when
-Wreturn-type was enabled by default for C++). Rejecting valid C++ code by
default is not OK.

[Bug c++/94768] Wreturn-type should be error, not warning

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94768

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=55189
         Resolution|INVALID                     |DUPLICATE

--- Comment #7 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Dup of PR 43943 (and others)

*** This bug has been marked as a duplicate of bug 43943 ***