On 22/11/14 20:53, Gisle Vanem wrote:
> I'm trying to understand the logic behind the generation
> of some statements in math.h (from math.in.h). The MSVC
> compiler has cosf() defined as a macro in it's <math.h>:
> #define cosf(x) ((float)cos((double)(x)))
>
> (for x86,C). Then how is Gnulib supposed to replace that?
> Since no matter what's put in Gnulib's "math.h", the
> preprocessed output of cosf.c contains this rubbish:
>
> extern int gl_signbitl (long double arg);
> float
> ((float)cos((double)(float x))) << MSVC's macro for cosf()
> {
> return (float) cos ((double) x);
> }
>
> Isn't there supposed to be an "#undef cosf" in "math.h"
> in this case? All I see is:
> # if !@HAVE_COSF@ << line 448
> # undef cosf
>
> I assume if some vendor have 'cosf' it should be undefined
> before Gnulib is trying to override or replace it. Not vice
> versa. So shouldn't this be:
> # if @HAVE_COSF@ || @REPLACE_COSF@
> # undef cosf
Well all the ...f() variants seem to be treated like this:
acosf() coshf() asinf() atanf() atan2f() ceilf() expf() ...
In the m4, the corresponding check is with: AC_CHECK_FUNCS([cosf])
I.E. the linker is used to find the function in libm.
If it's only a macro and not in libm, then the code in math.h.in
above makes sense.
So are you saying that HAVE_COSF is 1 for you (i.e. it's in your libm),
and you're also replacing it?
> Also, in test-cosf.c, the
> SIGNATURE_CHECK (cosf, float, (float));
>
> doesn't compile/link as it is here now (this macro cannot
> contain another macro). Or maybe test-cosf.c isn't supposed
> to be compiled for MSVC?
>
> Please help me resolve my confusion.
>
> Same for coshf(), cosl() etc. AFAICS.
Note cosl is treated differently as per:
http://git.sv.gnu.org/gitweb/?p=gnulib.git;a=commitdiff;h=d5b42fa0
thanks,
Pádraig.
