tokenize.detect_encoding() and open it in read only mode.

_Py_char2wchar(), instead of L"" VAR hack, to escape undecodable bytes using
the surrogateescape error handler.

Explain in the documentation that conversion errors should never happen.

 * Add error_pos optional argument to _Py_wchar2char()
 * PyUnicode_EncodeFS() raises a UnicodeEncodeError or MemoryError if
   _Py_wchar2char() fails

All recent modifications to distutils should now be covered in NEWS.

...unless these are commands you only run once a year :)

 * Specify the default encoding: write 'utf-8' instead of
   sys.getdefaultencoding(), because the default encoding is now constant
 * Specify the default errors value

encoding, instead of the locale encoding. Patch written by Alexander
Belopolsky.

doesn't preserve the zero sign (if TANH_PRESERVES_ZERO_SIGN define is 0).

Gay's dtoa.c (but not for current versions of Python's dtoa.c).
Thanks Rick Regan for finding and reporting this.