Re: [committed] libstdc++: Define std::numeric_limits<_FloatNN> before C++23

[Pranav Kant <prka@google.com>]

[-- Attachment #1: Type: text/plain, Size: 4056 bytes --]

Thanks for bringing this to my attention. I am working on a fix. Will keep
this thread posted.

Clang *does* define this macro only when float128 type is available. But
the problem seems to be that clang doesn't define _Float128 alias type
which is what's being used here. It only defines __float128 type. Should be
easy to fix.

On Wed, Oct 4, 2023 at 8:56 AM Jonathan Wakely <jwakely@redhat.com> wrote:

> On Wed, 4 Oct 2023 at 16:54, Stephan Bergmann <sbergman@redhat.com> wrote:
> >
> > On 8/17/23 22:32, Jonathan Wakely via Libstdc++ wrote:
> > > Tested x86_64-linux. Pushed to trunk.
> > >
> > > -- >8 --
> > >
> > > The extended floating-point types such as _Float32 are supported by GCC
> > > prior to C++23, you just can't use the standard-conforming names from
> > > <stdfloat> to refer to them. This change defines the specializations of
> > > std::numeric_limits for those types for older dialects, not only for
> > > C++23.
> > >
> > > libstdc++-v3/ChangeLog:
> > >
> > >       * include/bits/c++config (__gnu_cxx::__bfloat16_t): Define
> > >       whenever __BFLT16_DIG__ is defined, not only for C++23.
> > >       * include/std/limits (numeric_limits<bfloat16_t>): Likewise.
> > >       (numeric_limits<_Float16>, numeric_limits<_Float32>)
> > >       (numeric_limits<_Float64>): Likewise for other extended
> > >       floating-point types.
> > > ---
> > >   libstdc++-v3/include/bits/c++config |   4 +-
> > >   libstdc++-v3/include/std/limits     | 194
> +++++++++++++++-------------
> > >   2 files changed, 103 insertions(+), 95 deletions(-)
> > >
> > [...]
> > > diff --git a/libstdc++-v3/include/std/limits
> b/libstdc++-v3/include/std/limits
> > > index 52b19ef8264..7a59e7520eb 100644
> > > --- a/libstdc++-v3/include/std/limits
> > > +++ b/libstdc++-v3/include/std/limits
> > > @@ -1890,189 +1890,197 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
> > [...]
> > >   __glibcxx_float_n(64)
> > >   #endif
> > > -#ifdef __STDCPP_FLOAT128_T__
> > > +#ifdef __FLT128_DIG__
> > >   __glibcxx_float_n(128)
> > >   #endif
> > >   #undef __glibcxx_float_n
> > [...]
> >
> > The above change (from __STDCPP_FLOAT128_T__ to __FLT128_DIG__) now
> > started to cause issues with Clang on Clang 18 trunk:
> >
> > * Clang does not support a _Float128 type.
> >
> > * Clang does not predefine __STDCPP_FLOAT128_T__.
> >
> > * But since
> > <
> https://github.com/llvm/llvm-project/commit/457f582ffe23e951380bc345c4c96ec053c09681
> >
> > "[clang] Predefined macros for float128 support (#67196)", Clang 18
> > trunk does predefine __FLT128_DIG__ now.  Which causes
> >
> > > $ cat test.cc
> > > #include <limits>
> >
> > > $ clang++ -fsyntax-only test.cc
> > > In file included from test.cc:1:
> > >
> /home/sbergman/gcc/trunk/inst/lib/gcc/x86_64-pc-linux-gnu/14.0.0/../../../../include/c++/14.0.0/limits:1995:1:
> error: use of undeclared identifier '_Float128'
> > >  1995 | __glibcxx_float_n(128)
> > >       | ^
> > >
> /home/sbergman/gcc/trunk/inst/lib/gcc/x86_64-pc-linux-gnu/14.0.0/../../../../include/c++/14.0.0/limits:1903:27:
> note: expanded from macro '__glibcxx_float_n'
> > >  1903 |     struct numeric_limits<_Float##BITSIZE>
>           \
> > >       |                           ^
> > > <scratch space>:36:1: note: expanded from here
> > >    36 | _Float128
> > >       | ^
> > > 1 error generated.
> >
> > (I don't know whether or not it is useful for Clang to predefine
> > __FLT128_DIG__ when not providing a _Float128 type.  I assume
> > <https://www.iso.org/standard/65615.html> "ISO/IEC TS 18661-3:2015", as
> > referenced by the C++ standard, might be relevant here, but don't know
> > that document.  I added Pranav, the author of the relevant Clang commit,
> > in cc here.)
>
>
> It's completely wrong or Clang to define a macro describing properties
> of a non-existent type.
>
> This was reported as
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111687 and closed as
> INVALID, it needs to be fixed in Clang.
>
>
> >
>
>