[Bug c++/122780] Unnecessary stack allocation and copy for copy/move-assignment of trivial type

[m.cencora at gmail dot com via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122780

--- Comment #4 from m.cencora at gmail dot com ---
I see, thanks for explanation.

Is there a func attribute that would say "this func doesn't use globals" which
would allow for the mentioned optimization to apply?

[Bug c++/122780] Unnecessary stack allocation and copy for copy/move-assignment of trivial type

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122780

--- Comment #3 from Andrew Pinski  ---
Simple as:
```
struct bar a;

bar create() noexcept
{
  bar t; // NVR will transform t into the copy passed in
  t = a;
  t.data[10]++;
  t.data[20] = a.data[10];
  return t;
}

void g()
{
  foo_not_ok(a);
}
```

If the copy is removed in foo_not_ok, then NVR would be invalid. But the C++17
standard allows (requires?) for NVR here. So not having a copy will get the
wrong value into t.data[20].

[Bug c++/122780] Unnecessary stack allocation and copy for copy/move-assignment of trivial type

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122780

Andrew Pinski  changed:

   What|Removed |Added

 Resolution|--- |INVALID
 Status|UNCONFIRMED |RESOLVED

--- Comment #2 from Andrew Pinski  ---
Yes this is invalid. as create can access obj_on_heap via a different way
before it is finished so there needs to be a copy.

[Bug c++/122780] Unnecessary stack allocation and copy for copy/move-assignment of trivial type

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122780

--- Comment #1 from Andrew Pinski  ---
I think this is invalid. obj_on_heap could be accessed in create is finished so
there needs to copied.