bug#9960: Compiling Emacs trunk with MSVC

[Fabrice Popineau]

> Christoph Scholtes <cschol2112@googlemail.com> writes:
>
> > Then we get to emacs.c:
> > emacs.c(537) : error C2059: syntax error : '<<'
> > emacs.c(537) : error C2059: syntax error : 'constant'
> > emacs.c(537) : error C2059: syntax error : ')'
>
> cl.exe does not like the shift operator in this statement:
>
> static int __declspec(align(1 << 3)) test = 1;

That comes from Fabrice, assuming that removing `static' doesn't help.

Does it fail even if you take (1 << 3) in one more level of
parentheses?

What about an intermediate macro, as in

 #define FOO (1 << GCTYPEBITS)
 static int __declspec(align(FOO)) test = 1;

Ok. I have to admit that it does not work this way with msvc. Actually, I can't find a way to make

cl.exe compute the (1<< GCTYPEBITS) before giving it to the align declaration. At least not with the current release of cl.exe (16.00.40219.01). I can't check about the previous ones I may have used.

Our best bet is to use a construction like :

#define GCTYPEBITS 4

#if (1<<GCTYPEBITS) == 2

#define ALIGN_GCTYPEBITS 2

#elif (1<<GCTYPEBITS) == 4

#define ALIGN_GCTYPEBITS 4

#elif (1<<GCTYPEBITS) == 8

#define ALIGN_GCTYPEBITS 8

#else

#error Define ALIGN_GCTYPEBITS

#endif

static int __declspec(align(ALIGN_GCTYPEBITS)) test = 1;

void main (int argc, char *argv[]) { }

Sorry but I don't see a shorter answer for the moment.

(Looking into other issues ... and yes I compiled it, I'm using my own version every day :-) )

Fabrice