gh-70450: Fix MSVC /fp:strict builds

Replace floating-point expressions that MSVC rejects in static initializers under strict semantics. Initialize cmath NaN table entries once because NAN is not a constant expression in MSVC C mode.

Author: yhay81

Misc/NEWS.d/next/Build/2026-07-29-12-10-00.gh-issue-70450.fpstrict.rst

@@ -0,0 +1 @@
+Fix the Windows build when using the MSVC ``/fp:strict`` option.

Modules/cmathmodule.c

@@ -8,6 +8,7 @@
 
 #include "Python.h"
 #include "pycore_complexobject.h" // _Py_c_neg()
+#include "pycore_lock.h"          // _PyOnceFlag_CallOnce()
 #include "pycore_pymath.h"        // _PY_SHORT_FLOAT_REPR
 /* we need DBL_MAX, DBL_MIN, DBL_EPSILON, DBL_MANT_DIG and FLT_RADIX from
    float.h.  We assume that FLT_RADIX is either 2 or 16. */
@@ -147,11 +148,11 @@ special_type(double d)
     }
 
 #define P Py_MATH_PI
-#define P14 0.25*Py_MATH_PI
-#define P12 0.5*Py_MATH_PI
-#define P34 0.75*Py_MATH_PI
+#define P14 0.78539816339744830962
+#define P12 1.57079632679489661923
+#define P34 2.35619449019234492885
 #define INF INFINITY
-#define N Py_NAN
+#define N 9.5426319407711027e33 /* finite placeholder for NaN */
 #define U -9.5426319407711027e33 /* unlikely value, used as placeholder */
 
 /* First, the C functions that do the real work.  Each of the c_*
@@ -1230,6 +1231,41 @@ PyDoc_STRVAR(module_doc,
 "This module provides access to mathematical functions for complex\n"
 "numbers.");
 
+static _PyOnceFlag special_values_init_once = {0};
+
+static int
+init_special_values(void *Py_UNUSED(arg))
+{
+    Py_complex *tables[] = {
+        acos_special_values[0],
+        acosh_special_values[0],
+        asinh_special_values[0],
+        atanh_special_values[0],
+        cosh_special_values[0],
+        exp_special_values[0],
+        log_special_values[0],
+        sinh_special_values[0],
+        sqrt_special_values[0],
+        tanh_special_values[0],
+        rect_special_values[0],
+    };
+
+    /* MSVC with /fp:strict does not treat NAN as a constant expression.
+       Replace its finite placeholders once all static initialization is
+       complete. */
+    for (size_t i = 0; i < Py_ARRAY_LENGTH(tables); i++) {
+        for (size_t j = 0; j < 7 * 7; j++) {
+            if (tables[i][j].real == N) {
+                tables[i][j].real = Py_NAN;
+            }
+            if (tables[i][j].imag == N) {
+                tables[i][j].imag = Py_NAN;
+            }
+        }
+    }
+    return 0;
+}
+
 static PyMethodDef cmath_methods[] = {
     CMATH_ACOS_METHODDEF
     CMATH_ACOSH_METHODDEF
@@ -1260,6 +1296,10 @@ static PyMethodDef cmath_methods[] = {
 static int
 cmath_exec(PyObject *mod)
 {
+    /* init cannot fail */
+    (void)_PyOnceFlag_CallOnce(
+        &special_values_init_once, init_special_values, NULL);
+
     if (PyModule_Add(mod, "pi", PyFloat_FromDouble(Py_MATH_PI)) < 0) {
         return -1;
     }

Modules/mathmodule.c

@@ -2610,8 +2610,8 @@ math_pow_impl(PyObject *module, double x, double y)
 }
 
 
-static const double degToRad = Py_MATH_PI / 180.0;
-static const double radToDeg = 180.0 / Py_MATH_PI;
+static const double degToRad = 0.017453292519943295769236907684886;
+static const double radToDeg = 57.295779513082320876798154814105;
 
 /*[clinic input]
 math.degrees

Python/dtoa.c

@@ -1108,7 +1108,7 @@ tens[] = {
 static const double
 bigtens[] = { 1e16, 1e32, 1e64, 1e128, 1e256 };
 static const double tinytens[] = { 1e-16, 1e-32, 1e-64, 1e-128,
-                                   9007199254740992.*9007199254740992.e-256
+                                   8.112963841460668e-225
                                    /* = 2^106 * 1e-256 */
 };
 /* The factor of 2^53 in tinytens[4] helps us avoid setting the underflow */