Guido wrote: 
> Why copy the default encoding before mangling it?  With a little
> extra care you will only have to copy it once.  Also, consider not
> mangling at all, but assuming the encoding comes in a canonical form
> -- several other functions assume that, e.g. PyUnicode_Decode() and
> PyUnicode_AsEncodedString().

It is impossible guarantee that Py_FileSystemDefaultEncoding is
normalized, since its value can be set using nl_langinfo(CODESET)
during the bootstrapping process. PyUnicode_Decode() and other
decoding/encoding functions use the codec module, which is not available
during the early bootstrapping process, to normalize the encoding name.