Do not expect a copysign function to be defined in <cmath> with clang-cl

include/core/SkFloatingPoint.h

Index: include/core/SkFloatingPoint.h
diff --git a/include/core/SkFloatingPoint.h b/include/core/SkFloatingPoint.h
index 5489bac215c8de31b00f4d9c48b59e8ddea7fce5..93d479cf8709e91138196854574d2f852ec02604 100644
--- a/include/core/SkFloatingPoint.h
+++ b/include/core/SkFloatingPoint.h
@@ -31,7 +31,7 @@ static inline float sk_float_pow(float base, float exp) {
 
 static inline float sk_float_copysign(float x, float y) {
 // c++11 contains a 'float copysign(float, float)' function in <cmath>.
-#if __cplusplus >= 201103L || (defined(_MSC_VER) && _MSC_VER >= 1800)
+#if (!defined(_MSC_VER) && __cplusplus >= 201103L) || (defined(_MSC_VER) && _MSC_VER >= 1800)

[bungeman-skia, 2014/09/12 20:54:52]

So this won't break Chromium for now because Chromium does set
'-fmsc-version=1800'. On the other hand, this change doesn't make it very clear
that clang-cl causes '__cplusplus >= 201103L' to be true even when the runtime
library it's going to link against doesn't actually support that.


I think what we might want is a

#define SK_BUILD_CLANG_CL (defined(_MSC_VER) && defined(__clang__))

and change this line to

#if (!SK_BUILD_CLANG_CL && __cplusplus >= 201103L) || (defined(_MSC_VER) &&
_MSC_VER >= 1800)

to state clearly that we don't trust the clang-cl __cplusplus.

     return copysign(x, y);
 
 // Posix has demanded 'float copysignf(float, float)' (from C99) since Issue 6.