To decode byte string from the locale encoding (LC_CTYPE), PyUnicode_DecodeFSDefault() can be used, but this function uses a constant encoding set at startup (the locale encoding at startup). The right method is currently to call _Py_char2wchar() and then PyUnicode_FromWideChar(). _Py_char2wchar() is a low level function, it doesn't raise nice Python exception, but just return NULL on error and write a message to stderr using fprintf() (!).

Attached patch adds PyUnicode_DecodeLocale() and PyUnicode_DecodeLocaleAndSize() to offer a high level API to decode data from the *current* locale encoding. These functions fail with an OSError  or MemoryError if decoding fails (instead of a generic ValueError), and don't write to stderr anymore. They are a surrogateescape argument to choose to escape undecodable bytes or to fail with an error.

The patch only uses the function in _localemodule.c, but other functions may have to be fixed to use the new function. The tzname_encoding.patch of issue #5905 should maybe use it for example.