[cfe-dev] [llvm-dev] problem (and fix) with -fms-extensions

[Marc Espie via cfe-dev]

On Wed, May 17, 2017 at 05:31:45PM -0700, Richard Smith wrote:
> 
>    On 17 May 2017 at 16:24, Marc Espie via cfe-dev
>    <[1]cfe-dev at lists.llvm.org> wrote:
> 
>      On Wed, May 17, 2017 at 03:55:06PM -0700, Reid Kleckner wrote:
>      >
>      >Â  Â  On Wed, May 17, 2017 at 3:22 PM, Marc Espie
>      <[1][2]espie at nerim.net> wrote:
>      >
>      >Â  Â  Â  On Wed, May 17, 2017 at 01:01:07PM -0700, Reid Kleckner
>      wrote:
>      >Â  Â  Â  >
>      >Â  Â  Â  >ÃÂ  ÃÂ  I'd rather not add new pre-defined macros if we
>      can avoid it.
>      >Â  Â  Â  There are
>      >Â  Â  Â  >ÃÂ  ÃÂ  already too many. You can probably detect this
>      situation
>      >Â  Â  Â  with:
>      >Â  Â  Â  >ÃÂ  ÃÂ  #if defined(_MSC_EXTENSIONS) &&
>      !defined(_WCHAR_T_DEFINED)
>      >Â  Â  Â  >
>      >Â  Â  Â  Was this added recently ?
>      >Â  Â  Â  There is no _MSC_EXTENSIONS in the clang I'm using...
>      >
>      >Â  Â  Looks like we only define it for Windows targets.
>      >Â  Â  How about __is_identifier?
>      >Â  Â  $ cat t.cpp
>      >Â  Â  #if !__is_identifier(__wchar_t)
>      >Â  Â  #error "have __wchar_t"
>      >Â  Â  #else
>      >Â  Â  #error "no __wchar_t"
>      >Â  Â  #endif
>      Nope, that will trigger errors on non clang compilers.
> 
>    You can wrap it in #ifdef __clang__ to fix that.
>    Â

And that leads to more or less the same shitty set of conditionals I already
posted about... as opposed to using _MSC_EXTENSIONS, which by the way is
already defined on windows, is unlikely to ever collide with anything, and
would only be on with -fms-extensions, which is not the most common option
ever...

In every single case, __is_identifier will replace a one-liner
with several lines of nested conditionals...  not a good thing.