2.12.1: compilation error with `is*()` symbols

[0-wiz-0]

With gcc 12.5.0 (the default) on NetBSD 11.99.2, I see:

c++ -O2 -g -fstack-clash-protection -I/usr/include -I/usr/pkg/include/python3.13 -I/usr/pkg/include -g -I/usr/include -I/usr/pkg/include/python3.13 -I/usr/pkg/include -fPIC -DNPY_TARGET_VERSION=NPY_1_23_API_VERSION -I/us
r/pkg/lib/python3.13/site-packages/numpy/_core/include -I/usr/pkg/include/python3.13 -c numexpr/interpreter.cpp -o build/temp.netbsd-11.99.2-amd64-cpython-313/numexpr/interpreter.o
In file included from /scratch/math/py-numexpr/work/.buildlink/include/python3.13/Python.h:23,
                 from numexpr/module.hpp:13,
                 from numexpr/interpreter.cpp:10:
numexpr/numexpr_config.hpp: In function 'bool isfinitef_(float)':
numexpr/numexpr_config.hpp:54:49: error: expected unqualified-id before '(' token
   54 | inline bool isfinitef_(float x) { return !!std::isfinite(x); }
      |                                                 ^~~~~~~~
numexpr/numexpr_config.hpp: In function 'bool isnanf_(float)':
numexpr/numexpr_config.hpp:55:49: error: expected unqualified-id before '(' token
   55 | inline bool isnanf_(float x)    { return !!std::isnan(x); }
      |                                                 ^~~~~
numexpr/numexpr_config.hpp: In function 'bool isfinited(double)':
numexpr/numexpr_config.hpp:56:49: error: expected unqualified-id before '(' token
   56 | inline bool isfinited(double x) { return !!std::isfinite(x); }
      |                                                 ^~~~~~~~
numexpr/numexpr_config.hpp: In function 'bool isnand(double)':
numexpr/numexpr_config.hpp:57:49: error: expected unqualified-id before '(' token
   57 | inline bool isnand(double x)    { return !!std::isnan(x); }
      |                                                 ^~~~~
numexpr/numexpr_config.hpp: In function 'bool isinff_(float)':
numexpr/numexpr_config.hpp:58:46: error: expected unqualified-id before '(' token
   58 | inline bool isinff_(float x) { return !!std::isinf(x); }
      |                                              ^~~~~
numexpr/numexpr_config.hpp: In function 'bool isinfd(double)':
numexpr/numexpr_config.hpp:59:49: error: expected unqualified-id before '(' token
   59 | inline bool isinfd(double x)    { return !!std::isinf(x); }
      |                                                 ^~~~~

When I add an include of cmath (since this is the header that defines the symbols), I get:

c++ -O2 -g -fstack-clash-protection -I/usr/include -I/usr/pkg/include/python3.13 -I/usr/pkg/include -g -I/usr/include -I/usr/pkg/include/python3.13 -I/usr/pkg/include -fPIC -DNPY_TARGET_VERSION=NPY_1_23_API_VERSION -I/us
r/pkg/lib/python3.13/site-packages/numpy/_core/include -I/usr/pkg/include/python3.13 -c numexpr/interpreter.cpp -o build/temp.netbsd-11.99.2-amd64-cpython-313/numexpr/interpreter.o
In file included from numexpr/module.hpp:17,
                 from numexpr/interpreter.cpp:10:
numexpr/numexpr_config.hpp: In function 'bool isfinitef_(float)':
numexpr/numexpr_config.hpp:55:46: error: '::isfinite' has not been declared; did you mean 'std::isfinite'?
   55 | inline bool isfinitef_(float x) { return !!::isfinite(x); }
      |                                              ^~~~~~~~
      |                                              std::isfinite
In file included from numexpr/numexpr_config.hpp:43:
/usr/include/g++/cmath:578:5: note: 'std::isfinite' declared here
  578 |     isfinite(_Tp __x)
      |     ^~~~~~~~
numexpr/numexpr_config.hpp: In function 'bool isnanf_(float)':
numexpr/numexpr_config.hpp:56:46: error: '::isnan' has not been declared; did you mean 'std::isnan'?
   56 | inline bool isnanf_(float x)    { return !!::isnan(x); }
      |                                              ^~~~~
      |                                              std::isnan
/usr/include/g++/cmath:632:5: note: 'std::isnan' declared here
  632 |     isnan(_Tp __x)
      |     ^~~~~
numexpr/numexpr_config.hpp: In function 'bool isfinited(double)':
numexpr/numexpr_config.hpp:57:46: error: '::isfinite' has not been declared; did you mean 'std::isfinite'?
   57 | inline bool isfinited(double x) { return !!::isfinite(x); }
      |                                              ^~~~~~~~
      |                                              std::isfinite
/usr/include/g++/cmath:578:5: note: 'std::isfinite' declared here
  578 |     isfinite(_Tp __x)
      |     ^~~~~~~~
numexpr/numexpr_config.hpp: In function 'bool isnand(double)':
numexpr/numexpr_config.hpp:58:46: error: '::isnan' has not been declared; did you mean 'std::isnan'?
   58 | inline bool isnand(double x)    { return !!::isnan(x); }
      |                                              ^~~~~
      |                                              std::isnan
/usr/include/g++/cmath:632:5: note: 'std::isnan' declared here
  632 |     isnan(_Tp __x)
      |     ^~~~~
numexpr/numexpr_config.hpp: In function 'bool isinff_(float)':
numexpr/numexpr_config.hpp:59:43: error: '::isinf' has not been declared; did you mean 'std::isinf'?
   59 | inline bool isinff_(float x) { return !!::isinf(x); }
      |                                           ^~~~~
      |                                           std::isinf
/usr/include/g++/cmath:605:5: note: 'std::isinf' declared here
  605 |     isinf(_Tp __x)
      |     ^~~~~
numexpr/numexpr_config.hpp: In function 'bool isinfd(double)':
numexpr/numexpr_config.hpp:60:46: error: '::isinf' has not been declared; did you mean 'std::isinf'?
   60 | inline bool isinfd(double x)    { return !!::isinf(x); }
      |                                              ^~~~~
      |                                              std::isinf
/usr/include/g++/cmath:605:5: note: 'std::isinf' declared here
  605 |     isinf(_Tp __x)
      |     ^~~~~

so I came up with this patch, which fixes the problem for me:

--- numexpr/numexpr_config.hpp.orig     2025-09-14 13:57:31.046041558 +0000
+++ numexpr/numexpr_config.hpp
@@ -40,6 +40,7 @@
 #include "mkl_vml.h"
 #include "mkl_service.h"
 #endif
+#include <cmath>

 #ifdef _WIN32
   #ifndef __MINGW32__
@@ -51,12 +52,12 @@
 msvc_function_stubs contains windows alternatives
 /* Due to casting problems (normally return ints not bools, easiest to define
 non-overloaded wrappers for these functions) */
-inline bool isfinitef_(float x) { return !!::isfinite(x); }
-inline bool isnanf_(float x)    { return !!::isnan(x); }
-inline bool isfinited(double x) { return !!::isfinite(x); }
-inline bool isnand(double x)    { return !!::isnan(x); }
-inline bool isinff_(float x) { return !!::isinf(x); }
-inline bool isinfd(double x)    { return !!::isinf(x); }
+inline bool isfinitef_(float x) { return !!std::isfinite(x); }
+inline bool isnanf_(float x)    { return !!std::isnan(x); }
+inline bool isfinited(double x) { return !!std::isfinite(x); }
+inline bool isnand(double x)    { return !!std::isnan(x); }
+inline bool isinff_(float x) { return !!std::isinf(x); }
+inline bool isinfd(double x)    { return !!std::isinf(x); }
 #endif

 #endif // NUMEXPR_CONFIG_HPP

but I have no wait to test this on Windows.

[FrancescAlted]

Ok. I suggest to send a PR so that we can run your patch in the Windows part of CI.

[0-wiz-0]

@FrancescAlted Sure, it's in #529.