[libc] Make math-macros.h C++-friendly

[frobtech]

The isfinite, isnan, and isinf "functions" are specified by C99..C23 to be macros that act as type-generic functions. Defining them as their _builtin* counterparts works fine for this. However, in C++ the identifiers need to be usable in different contexts, such as being declared inside a C++ namespace. So define inline constexpr template functions for them under #ifdef __cplusplus.

[llvmbot]

@llvm/pr-subscribers-libc

Author: Roland McGrath (frobtech)

Changes

The isfinite, isnan, and isinf "functions" are specified by C99..C23 to be macros that act as type-generic functions. Defining them as their _builtin* counterparts works fine for this. However, in C++ the identifiers need to be usable in different contexts, such as being declared inside a C++ namespace. So define inline constexpr template functions for them under #ifdef __cplusplus.

---

Full diff: https://github.com/llvm/llvm-project/pull/86206.diff

1 Files Affected:

• (modified) libc/include/llvm-libc-macros/math-macros.h (+28-4)

diff --git a/libc/include/llvm-libc-macros/math-macros.h b/libc/include/llvm-libc-macros/math-macros.h
index db8a4ea65bd69a..2605535b927d76 100644
--- a/libc/include/llvm-libc-macros/math-macros.h
+++ b/libc/include/llvm-libc-macros/math-macros.h
@@ -30,10 +30,6 @@
 #define FP_LLOGB0 (-LONG_MAX - 1)
 #define FP_LLOGBNAN LONG_MAX
 
-#define isfinite(x) __builtin_isfinite(x)
-#define isinf(x) __builtin_isinf(x)
-#define isnan(x) __builtin_isnan(x)
-
 #ifdef __FAST_MATH__
 #define math_errhandling 0
 #elif defined(__NO_MATH_ERRNO__)
@@ -44,4 +40,32 @@
 #define math_errhandling (MATH_ERRNO | MATH_ERREXCEPT)
 #endif
 
+// These must be type-generic functions.  The C standard specifies them as
+// being macros rather than functions, in fact.  However, in C++ it's important
+// that there be function declarations that don't interfere with other uses of
+// the identifier, even in places with parentheses where a function-like macro
+// will be expanded (such as a function declaration in a C++ namespace).
+
+#ifdef __cplusplus
+
+template <typename T> inline constexpr bool isfinite(T x) {
+  return __builtin_isfinite(x);
+}
+
+template <typename T> inline constexpr bool isinf(T x) {
+  return __builtin_isinf(x);
+}
+
+template <typename T> inline constexpr bool isnan(T x) {
+  return __builtin_isnan(x);
+}
+
+#else
+
+#define isfinite(x) __builtin_isfinite(x)
+#define isinf(x) __builtin_isinf(x)
+#define isnan(x) __builtin_isnan(x)
+
+#endif
+
 #endif // LLVM_LIBC_MACROS_MATH_MACROS_H

[philnik777]

This patch is actually really problematic for interoperability with C++ standard libraries and the implementation is non-conforming in C++, because it doesn't satisfy http://eel.is/c++draft/c.math#cmath.syn-3.