Just a reminder that we support configurations where there are both big-endian and little-endinan binaries in the *same* executable file: for example, Mac OS X 32-bit-only universal binaries (.so, .dylib, .exe) contain both i386 and ppc code in the same file.  So any -endian differences must remain isolated to C code or be dynamically tested at execution time in Python code. And ./configure time tests can be problematic.  (The proposed patch doesn't have that problem but it might be an issue later in tests.)