cmath vs. math.h vs isfinite

[Thomas Klausner <wiz%netbsd.org@localhost>]

Hi!

cython 3.3.0 uncovered a problem in our headers, at least compared to
other systems.

Using this test program test.cc:

#include <cmath>
#include <math.h>
#include <cstdio>
#include <cstdlib>
int main() {
  if (isfinite(0.0)) {
    printf("finite\n");
  }
  exit(0);
}

# make test
c++ -O2    -o test test.cc
test.cc: In function 'int main()':
test.cc:6:7: error: 'isfinite' was not declared in this scope; did you mean 'std::isfinite'?
    6 |   if (isfinite(0.0)) {
      |       ^~~~~~~~
      |       std::isfinite
In file included from test.cc:1:
/usr/include/g++/cmath:1142:5: note: 'std::isfinite' declared here
 1142 |     isfinite(_Tp)
      |     ^~~~~~~~
*** Error code 1


but when I remove <cmath> and only include <math.h> it's fine:

# make test
c++ -O2    -o test test.cc
#

I tried the same program on godbolt.org and do not see the same issue
- as long as math.h is included, a plain isfinite is fine; only if
<cmath> is included alone, it doesn't work.

Is there some standard lawyering that makes NetBSD ok, or is this
something that should be fixed in our headers?
 Thomas