[Bug c++/124834] Devirtualization should not attempt to call pure virtual functions

[luigighiron at gmail dot com via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124834

--- Comment #7 from Halalaluyafail3  ---
(In reply to Drea Pinski from comment #6)
> >Note that the call to foo is never actually reached, so the undefined 
> >behavior does not happen.
> 
> Yes but isn't that is the same as doing:
> ```
> int f();
> 
> int g()
> {
>   f();
> }
> 
> int main()
> {
>   return 0;
> }
> ```
> 
> The undefined behavior is not reached here either.
f is odr-used here, so it will be IFNDR. But since foo in the original example
is a pure virtual function and the call is not using an explicitly qualified
name, A::foo is not named by the expression foo(). If it was not like this
then:

struct C{
virtual void bar()=0;
void baz(){
bar();
}
};

Would be invalid without a definition of C::bar. The original example is doing
the same thing as this, just inside of the constructor.

[Bug c++/124834] Devirtualization should not attempt to call pure virtual functions

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124834

--- Comment #4 from Drea Pinski  ---
r0-61719-g585b44d301a56e changed it from an error to a warning and it was an
error since 1996.

[Bug c++/124834] Devirtualization should not attempt to call pure virtual functions

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124834

--- Comment #6 from Drea Pinski  ---
>Note that the call to foo is never actually reached, so the undefined behavior 
>does not happen.

Yes but isn't that is the same as doing:
```
int f();

int g()
{
  f();
}

int main()
{
  return 0;
}
```

The undefined behavior is not reached here either.

[Bug c++/124834] Devirtualization should not attempt to call pure virtual functions

[luigighiron at gmail dot com via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124834

--- Comment #8 from Halalaluyafail3  ---
Additionally, the undefined behavior in the original example comes from
[class.abstract] saying that using the pure virtual function in this context is
undefined behavior.

[Bug c++/124834] Devirtualization should not attempt to call pure virtual functions

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124834

--- Comment #5 from Drea Pinski  ---
(In reply to Drea Pinski from comment #4)
> r0-61719-g585b44d301a56e changed it from an error to a warning and it was an
> error since 1996.

https://inbox.sourceware.org/gcc-patches/414946e5.2080...@codesourcery.com/ was
when the request to turn the error into an warning happened.

[Bug c++/124834] Devirtualization should not attempt to call pure virtual functions

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124834

--- Comment #3 from Drea Pinski  ---
Now GCC 3.4.6 used to error out and not just warn though:
:4: error: abstract virtual `virtual void A::foo()' called from
constructor

[Bug c++/124834] Devirtualization should not attempt to call pure virtual functions

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124834

--- Comment #2 from Drea Pinski  ---
(In reply to Halalaluyafail3 from comment #1)
> It seems to be accepted in GCC 4, so I suppose it is a regression albeit a
> very old one.

It was not accepted in GCC 4. godbolt does not support linking with GCC 4.

[Bug c++/124834] Devirtualization should not attempt to call pure virtual functions

[luigighiron at gmail dot com via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124834

--- Comment #1 from Halalaluyafail3  ---
It seems to be accepted in GCC 4, so I suppose it is a regression albeit a very
old one.