Re: When to emit C++ debug info?

["B. Kosnik" <bkoz at nabi dot net>]

I cannot resist replying to this...

namespace b_std
{
  class b_ostream {
  public:
    int foo;
    virtual void bar();
    b_ostream& operator << (const char *);
  };

  extern b_ostream b_cout;
}

int main()
{
  b_std::b_cout << "Hello world";
}

> The question arises, what debug information should be emitted for this? 

Well, I suspect that somebody debugging this would want:

1) main
2) b_std::b_ostream that can be examined with 
   (gdb) p b_cout
   (or, in some future, better world)
   (gdb) p b_std::b_cout
3) b_std::b_ostream::operator<<(const char*) definition, if it existed

As you noticed, mainline g++ doesn't do this. (No difference between g++
-g, g++ -ggdb3, g++ -g3 -O0 -fno-eliminate-unused-debug-types). The fact
that a used variable (b_cout) doesn't have a DW_TAG_variable is a bug, I
think, and should be filed in bugzilla. I would think the structure_type
tag would fall out correctly if the variable tag was emitted.

The Bugzilla entry somewhat associated with this kind of thing involves
nested types. It's debug/11325, and was fixed by Scott Synder.

> The debug information is keyed to the vtable at the moment.

This will result in problems for libstdc++ templatized classes, because
of the "extern template" suppression of explicitly instantiated
templates. 

> This means that, even now that GDB can handle trivial uses of namespace
> information, real-world examples are pretty much useless.

real world examples are "less than optimally" debugged.

:-)

-benjamin