Am 09.01.2026 um 11:07 schrieb LIU Hao:
在 2026-1-9 15:48, Markus Muetzel 写道:
+#if !defined(_USE_MATH_DEFINES)
+#define _USE_MATH_DEFINES
+#endif
+
+#include <math.h>
Does it make sense to just `#define _USE_MATH_DEFINES` without the
`#if`? There's such a define in xaudio2.h but I think no other code in
the CRT defines that.
My thinking was that, if at some point in the future `_USE_MATH_DEFINES`
was defined in `AM_CPPFLAGS` or a user manually added that flag (e.g.,
as a parameter to the `configure` script), that would cause a
compilation error that could be avoided by guarding the definition like
that.
But that point is mute in the updated patch (see below).
+float __cdecl acospif(float x)
+{
+ return acosf(x)/M_PI;
+}
Becasuse `M_PI` is a `double`, this intermediate result is calculated as
return (float) ((double) acosf(x) / M_PI);
When looking into this, I realize that the `long double` variants are
probably not correct either. Despite having 21 significant figures,
`(long double)(double) M_PI` might not yield the same value as
`3.14159265358979323846L`. (When suffixed with `L`, this is accurate
as a `long double`.) So maybe local definitions of PI are inevitable..?
Good point.
Instead of defining the value of pi in the appropriate type repeatedly,
I opted for doing that in a header that can be included in all new
compilation units. (That could be simplified if _Generic macros from C11
were allowed.)
Does that look good to you?
Markus
_______________________________________________
Mingw-w64-public mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
