Do not rely on Clang's results to verify our GCC version handling

We use __float128 to distinguish GCC versions, which is not a
distinguishing factor for Clang. (It's supported on several Clang
versions on x86, but not on Apple's M1/ARM.) This commit effectively
disables the gcc_version1 test on macOS (where "gcc" really is Clang).

Fixes: #5828

Author: tautschnig

regression/ansi-c/gcc_version1/gcc-4.c

@@ -1,4 +1,5 @@
-// None of these types should be defined when emulating gcc-4:
+// None of these types should be defined when emulating gcc-4, which clang on
+// macOS also emulates:
 
 typedef float _Float32;
 typedef double _Float32x;

regression/ansi-c/gcc_version1/gcc-5.c

@@ -1,4 +1,6 @@
 // These types should *not* be provided when emulating gcc-5:
+
+#ifndef __clang__
 typedef float _Float32;
 typedef double _Float32x;
 typedef double _Float64;
@@ -8,3 +10,4 @@ typedef long double _Float128x;
 
 // But this type should:
 __float128 f128;
+#endif

regression/ansi-c/gcc_version1/gcc-7.c

@@ -1,4 +1,6 @@
 // All these types should be provided when emulating gcc-7:
+
+#ifndef __clang__
 _Float32 f32;
 _Float32x f32x;
 _Float64 f64;
@@ -7,3 +9,4 @@ _Float128 f128;
 _Float128x f128x;
 
 __float128 gcc_f128;
+#endif