On Wed, 13 May 2020 at 09:44, Martin Oberzalek <kingleo@xxxxxx> wrote:
>
> Hello everybody,
>
> gcc 4.4.7 warns me if I want to pass a C++ Object through a variadic function
>
> eg:
>
> #include <stdio.h>
> #include <iostream>
>
> int print( const char *fmt, ... )
> {
> va_list argptr;
>
> va_start( argptr, fmt );
>
> char* text = va_arg(argptr, char*);
>
> std::cout << text << std::endl;
>
> va_end( argptr );
>
> return 0;
> }
>
> int main()
> {
> std::string xx = "hello";
>
> print( "%s\n", xx ); //<= wrong, gcc 4.4.7 warns me here
> print( "%s\n", "123" ); //< ok, no warning
> }
>
> g++ -o p -O2 -Wall
> printf.cc: In function ‘int main()’:
> printf.cc:35:21: warning: cannot pass objects of non-POD type ‘std::string’
> {aka ‘class std::__cxx11::basic_string<char>’} through ‘...’; call will abort
> at runtime
> 35 | print( "%s\n", xx );
>
> This warning (I call it feature) disappeared in later versions. And as I
> detected: gcc can now pass a std::string object through a variadic function.
>
> I've many variadic functions, and these are all implemented in C. They cannot
> handle C++ objects. So in my use case this is always wrong.
>
> I'm wondering why this warning disapeared, and if somebody else is missing it?
It used to be undefined to pass non-POD objects through ... but the
C++ standard changed to make it conditionally-supported, and G++
chooses to support it, with well-defined behaviour.
This was changed in
https://gcc.gnu.org/g:d2f2e467c03f108a407473366c950bc2371ca00a
> Is anybody interested in getting such a warning back?
> Is there another way (pragma?) getting such a warning?
Use -Wconditionally-supported or add:
#pragma GCC diagnostic warning "-Wconditionally-supported"
> I've created a patch, to get this warning back. Is anybody interested in such
> a patch, and how are the chances to get this in g++ as an opt in feature?
I doubt it, because there's a new way to get the warning, which
correctly matches the G++ semantics.
