[Bug c++/100485] New: False positive in -Wmismatched-new-delete

[Bug c++/100485] New: False positive in -Wmismatched-new-delete

[fiesh at zefix dot tv]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100485

            Bug ID: 100485
           Summary: False positive in -Wmismatched-new-delete
           Product: gcc
           Version: 11.1.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: fiesh at zefix dot tv
  Target Milestone: ---

The following code triggers -Wmismatched-new-delete:

----------------------------------------

extern "C" void free(void *);
class Base
{
public:
        Base();
        void * operator new(unsigned long, const int &);
        void operator delete(void * ptr, const int &) { free(ptr); }
};
void f()
{
        new (0) Base;
}

----------------------------------------

Supplying `__attribute__((malloc (free, 1)))` to `operator new` also doesn't
help.

[Bug c++/100485] False positive in -Wmismatched-new-delete

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100485

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |msebor at gcc dot gnu.org
             Blocks|                            |100406
         Resolution|---                         |WONTFIX
             Status|UNCONFIRMED                 |RESOLVED

--- Comment #1 from Martin Sebor <msebor at gcc dot gnu.org> ---
The call to the inline operator delete is inlined but the call to new is not so
the warning sees a mismatch.  This can be suppressed either by inlining both
calls or by preventing it by declaring the operators with attribute noinline.  

The article below gives more insight:

https://developers.redhat.com/blog/2021/04/30/detecting-memory-management-bugs-with-gcc-11-part-1-understanding-dynamic-allocation/
https://developers.redhat.com/blog/2021/05/05/detecting-memory-management-bugs-with-gcc-11-part-2-deallocation-functions/


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100406
[Bug 100406] bogus/missing -Wmismatched-new-delete

[Bug c++/100485] False positive in -Wmismatched-new-delete

[fiesh at zefix dot tv]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100485

--- Comment #2 from fiesh at zefix dot tv ---
But this isn't really a solution since I can't inline new without moving a lot
of code into the header, and marking `operator new` noinline isn't what I want
either.  I read both articles prior to making this bug report (and asked on
IRC).  I don't understand why supplying `__attribute__((malloc (free, 1)))` to
operator new doesn't silence this since it should then expect a call to free.

[Bug c++/100485] False positive in -Wmismatched-new-delete

[fiesh at zefix dot tv]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100485

--- Comment #3 from fiesh at zefix dot tv ---
* marking operator delete noinline

[Bug c++/100485] False positive in -Wmismatched-new-delete

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100485

--- Comment #4 from Martin Sebor <msebor at gcc dot gnu.org> ---
-Wmismatched-new-delete considers only "mismatches between calls to operator
new or operator delete and the corresponding call to the allocation or
deallocation function."  It doesn't also consider attribute malloc, by design. 
It would be possible, even easy, to change that and have the warning also look
for attribute malloc to handle this case but it doesn't seem like a good design
feature to me to provide support for associating a member operator new() with
an unrelated deallocation function, whether it's free() or another operator
delete().  To do that, I would instead suggest suppressing the warning by using
#pragma GCC diagnostic ignored "-Wmismatched-new-delete":

extern "C" void free (void *);

class Base
{
public:
        Base();

        void * operator new(unsigned long, const int &);
        void operator delete(void * ptr, const int &) {
          #pragma GCC diagnostic ignored "-Wmismatched-new-delete"
          free (ptr);
        }
};

void f()
{
        new (0) Base;
}

[Bug c++/100485] False positive in -Wmismatched-new-delete

[fiesh at zefix dot tv]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100485

--- Comment #5 from fiesh at zefix dot tv ---

> extern "C" void free (void *);
>   
> class Base
> {
> public:
>         Base();
> 
>         void * operator new(unsigned long, const int &);
>         void operator delete(void * ptr, const int &) {
>           #pragma GCC diagnostic ignored "-Wmismatched-new-delete"
>           free (ptr);
>         }
> };
> 
> void f()
> {
>         new (0) Base;
> }

Alas, keeping the diagnostic ignore local by means of `diagnostic push` /
`diagnostic pop` makes it not work correctly, presumably since the warning is
generated at the call site.  Making it global to everything including the
header file seems like a bad idea to me.

[Bug c++/100485] False positive in -Wmismatched-new-delete

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100485

--- Comment #6 from Martin Sebor <msebor at gcc dot gnu.org> ---
That does sound frustrating, sorry.  The #pragma has a limitation that can make
it difficult to use in inlining contexts.  I submitted a patch for it for GCC
11 but it got pushed to GCC 12 (see pr98465 comment 32).  Until it's fixed, the
only other solution I can think of is to prevent the member operator delete
from inlining using attribute noinline.

[Bug c++/100485] False positive in -Wmismatched-new-delete

[fiesh at zefix dot tv]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100485

--- Comment #7 from fiesh at zefix dot tv ---
Thanks for the outline!  We'll turn off -Wmismatched-new-delete with GCC 11 and
try to switch to the selective opt-out with pragmas in 12.  That's a good
workaround for now.

On a random related note, the man page says -Wmismatched-new-delete is enabled
by default, but playing around with it, it seems it's only turned on by -Wall:
https://godbolt.org/z/n1s6Y8G85

[Bug c++/100485] False positive in -Wmismatched-new-delete

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100485

--- Comment #8 from Martin Sebor <msebor at gcc dot gnu.org> ---
(In reply to fiesh from comment #7)
> On a random related note, the man page says -Wmismatched-new-delete is
> enabled by default, but playing around with it, it seems it's only turned on
> by -Wall: https://godbolt.org/z/n1s6Y8G85

Thanks, I just fixed it in r12-1544.