Proposed PEP 349: Generalized String Coercion

• Goal: ease transition from str to unicode
• Unicode-safe: can take unicode, doesn't try to coerce to str
• Str-stable: if passed str, doesn't coerce to unicode
• Using str () makes non-Unicode-safe, unicode () not str-stable
• "%s" % obj now coerces obj which has only __unicode__ to '<foo instance>' (note that "%s" % unistr works)
• Returning unicode from __str__ causes str () to complain UnicodeEncodeError if it can't encode.
• Want: promotion to unicode. PEP 349 defines text(), which will call __str__, check that isinstance (res, basestring), and return without encoding.

Guido still prefers a bytes type (and immutable frozenbytes).

PEP 349 solves a problem I don't have often (probably because of my lack of writing such libraries), but doesn't seem to solve the problem I do have, of reading from external data sources: foo_string.decode ('ascii') is type unicode and doesn't need to be, but foo_string.decode ('iso-8859-1') might need to be or might not, depending on what it contains.