[RFC] middle-end/21743: Enable __builtin_clog

[Paolo Carlini <pcarlini at suse dot de>]

Hi,

I'm trying to figure out whether we can resolve this annoying issue one
way or another... A short summary. Some time ago Kaveh setup everything
related to the C99 builtins and now we have a complete set available,
besides (builtins.def):

/* The C99 clog function conflicts with C++ iostreams clog, see
   http://gcc.gnu.org/ml/gcc-patches/2003-09/msg00510.html  */
...
...
/*DEF_C99_BUILTIN        (BUILT_IN_CLOG, "clog",
BT_FN_COMPLEX_DOUBLE_COMPLEX_DOUBLE, ATTR_MATHFN_FPROUNDING)*/
/*DEF_C99_BUILTIN        (BUILT_IN_CLOGF, "clogf",
BT_FN_COMPLEX_FLOAT_COMPLEX_FLOAT, ATTR_MATHFN_FPROUNDING)*/
/*DEF_C99_BUILTIN        (BUILT_IN_CLOGL, "clogl",
BT_FN_COMPLEX_LONGDOUBLE_COMPLEX_LONGDOUBLE, ATTR_MATHFN_FPROUNDING)*/

The problem is that this kind of builtin injects in the global namespace
also the version without __builtin_* in front and we get a conflict with
the C++ runtime library as soon as the user does, in the gloobal namespace:

    using std::clog;

Now, as you easily see, it would be great to enable that remaining
builtin, both for consistency and, in practice, for use in the C++
runtime library itself: the implementation of std::complex::log could
nicely exploit it, the same that is already happening for the other C99
complex functions.

What can we realistically do? Some considerations:

- Andrew suggested enabling only the __builtin_* version of those
builtins. From my selfish C++ runtime point of view, would be ok, but
then we have to invent a new kind of DEF_*_BUILTIN macro, having both
BOTH_P == false and FALLBACK_P == true. Maybe, short-term is better than
what we currently have.

Now, there is something that I don't understand *at all* in the C99
builtins machinery itself... If I try to enable __builtin_clog, the
following doesn't compile:

    #include <iostream>
    using std::clog;

clog1.cc:2: error: 'clog' is already declared in this scope

And *also* this one doesn't compile:

    int main()
    {
      _Complex double a;
      clog(a);
    }

clog2.cc:4: error: 'clog' was not declared in this scope

???

Is clog - the complex mathematical function - declared or not?!?

I mean, in general, the C99 builtins machinery does *not* inject names
without __builtin_* in the global namespace (second example). So far, so
very good. But then, why I cannot bring in the global namespace itself
my std::clog (first example)?!?

Anyone willing to clarify or suggest solutions at any level?!?

Thanks for now, back to the shell...
Paolo.