Bruno Haible <[EMAIL PROTECTED]> writes: > OK, so let's try a runtime test instead of a compile-time test.
OK, but for cross-compilation we need to assume that the bug is absent (contrary to our usual practice for run-time tests), because otherwise cross-compilation will generate the wrong code for the hosts that do 'long double' correctly. The idea will be that we don't support cross-compilation to "buggy" hosts like FreeBSD 6.1.
