For example, PyUnicode_FromFormat() does not specify the return value but it does not seem to be only one. I only have a basic knowledge of Python C/API, so I am not sure whether it is meaningful.

if you are referring to http://docs.python.org/dev/py3k/c-api/unicode.html#PyUnicode_FromFormat the return type is a PyObject.

See also http://docs.python.org/dev/py3k/c-api/structures.html#PyObject

I meant whether it returns a new reference or not. For instance, documentation for PyUnicode_FromObject() and PyUnicode_AsWideChar() states that a new reference is returned, but this is most specified for most functions in Unicode Object (at least).

I meant whether it returns a new reference or not. For instance, documentation for PyUnicode_FromObject() and PyUnicode_AsWideChar() states that a new reference is returned, but this is not specified for most functions in Unicode Object (at least).

Arnaud Fontaine wrote:
> 
> Arnaud Fontaine <arnau@debian.org> added the comment:
> 
> I meant whether it returns a new reference or not. For instance, documentation for PyUnicode_FromObject() and PyUnicode_AsWideChar() states that a new reference is returned, but this is not specified for most functions in Unicode Object (at least).

Patches are welcome !

FWIW: Unless otherwise noted, all Unicode APIs that return a PyObject*
give you a new reference.

I'd like to add that a patch would need to modify Doc/data/refcounts.dat, not the .rst files.

Hi, I read through unicodeobject.c and added the (IMO) proper reference counts to the missing functions.  I attached a first patch which adds this to Doc/data/refcounts.dat.

The patch also fixes 2 minor glitches in Doc/c-api/unicode.rst, PyUnicode_DecodeMBCSStateful stated int instead of Py_ssize_t for it's arguments and PyUnicode_FromString had it's return value wrongly formated.

Sorry, we missed this issue.

refcounts.dat was updated in other issues.