Re: stdbool.h does not define _Bool when included by C++ code

[Rich Felker <dalias@libc.org>]

On Tue, Aug 01, 2017 at 09:06:01AM +1200, Michael Clark wrote:
> 
> > On 31 Jul 2017, at 9:46 PM, Jens Gustedt <jens.gustedt@inria.fr> wrote:
> > 
> > Hello,
> > 
> > On Mon, 31 Jul 2017 11:18:28 +0200 Szabolcs Nagy <nsz@port70.net> wrote:
> > 
> >> iow, this is either a minor gcc bug or a big fat c++ defect
> >> depending on how you look at it, the libc cannot fix this
> >> properly, just emulate the broken nonsense in gcc stdbool.h
> >> that nobody should rely on.
> > 
> > Basically stdbool.h is already a header to accomodate C++ usage of
> > bool, false and true to C. I makes not much sense to include this
> > in C++ code.
> 
> It’s useful exactly for the case where you have C code in your C++,
> which is why I suspect GCC has the condition in its version of
> stdbool.h
> 
> > I think that applications that want to be shure that their code
> > compiles for both should use "bool" and should do
> > 
> > #ifndef __cplusplus
> > # include <stdbool.h>
> > #endif
> 
> I think you’re missing the point. C code can use _Bool and this can
> be put through the C++ front-end. Compiler explorer does this by
> default, and the GCC and glibc headers already have the fix, which
> is #define _Bool bool (in the patch) for the case where C code is
> fed in to the C++ front end.

Compiling C code with a C++ compiler is utterly wrong, and will fail
in more subtle ways; erroring out early seems much more desirable. C
is not a subset of C++, and even the syntax that's in the intersection
of both languages is semantically different between them.

> Have a look at gcc’s stdbool.h
> 
> Notice that compiler has all the signs that it is actually C,
> including using gcc and clang in the compiler labels instead of g++
> and clang++
> 
> - https://godbolt.org/g/rhEZCx
> 
> There are both C and C++ users who use compiler explorer, and it
> seems the C users will have to add casts to malloc return values and
> what not. Notice the dissasembly has de-mangled type-safe linkage.
> It’s useful because you can see the arity of the function in
> compiler explorer disassembly.

Adding -xc to command line fixes this. I've also requested Matt
Godbolt add a first-class language selection choice somewhere and I
thought he was going to do it, but I don't see it yet.

> > We can put more comfort into this by lifting this #ifndef/#endif stuff
> > to the contents of the file, if people really want this. But code that
> > relies on this would not be guaranteed to be portable.
> > 
> > Using "bool" by means of the header is the intended usage, these
> > attrocities such as _Bool, _Noreturn etc are only there for backwards
> > compatibility. Hopefully at least some of them will be phased out in
> > the future.
> > 
> > I think we should not encourage the usage of these keywords in
> > application code unless this is needed for backwards compatibility,
> > nor should we try to impose their possible usage into other languages.
> 
> In any case the workaround is to rm PREFIX/include/stdbool.h so that
> the libgcc version gets included.

If this works, it's by chance, and there's no guarantee that it will
continue to work. There are definitely incompatibilities using gcc's
other std*.h headers with musl; certain include orders will yield
multiple or conflicting definitions or other issues. stdbool.h is
probably simple enough for now that it's not affected.

> It’s essentially discouraging C++ users who want more conformity
> with C.

I'm not entirely opposed to the proposed change if it eases adoption
and integration issues, but I'm skeptical that it really does, and I
do think it's something of a misfeature. Pretending a C++ compiler can
compile C hides bugs.

Rich