diff options
author | Ezio Melotti <ezio.melotti@gmail.com> | 2012-02-26 07:39:55 (GMT) |
---|---|---|
committer | Ezio Melotti <ezio.melotti@gmail.com> | 2012-02-26 07:39:55 (GMT) |
commit | cda6b6d60d96e6f755da92deb5e4066839095791 (patch) | |
tree | 0835098963d975f54d46707bae270e08df2fc4fa /Objects/bytesobject.c | |
parent | 408026c7e8c019cf04372a4267c832241e18c62c (diff) | |
download | cpython-cda6b6d60d96e6f755da92deb5e4066839095791.zip cpython-cda6b6d60d96e6f755da92deb5e4066839095791.tar.gz cpython-cda6b6d60d96e6f755da92deb5e4066839095791.tar.bz2 |
#14081: The sep and maxsplit parameter to str.split, bytes.split, and bytearray.split may now be passed as keyword arguments.
Diffstat (limited to 'Objects/bytesobject.c')
-rw-r--r-- | Objects/bytesobject.c | 20 |
1 files changed, 12 insertions, 8 deletions
diff --git a/Objects/bytesobject.c b/Objects/bytesobject.c index a98cdcf..1d2fed7 100644 --- a/Objects/bytesobject.c +++ b/Objects/bytesobject.c @@ -972,7 +972,7 @@ static const char *stripformat[] = {"|O:lstrip", "|O:rstrip", "|O:strip"}; #define STRIPNAME(i) (stripformat[i]+3) PyDoc_STRVAR(split__doc__, -"B.split([sep[, maxsplit]]) -> list of bytes\n\ +"B.split(sep=None, maxsplit=-1) -> list of bytes\n\ \n\ Return a list of the sections in B, using sep as the delimiter.\n\ If sep is not specified or is None, B is split on ASCII whitespace\n\ @@ -980,15 +980,17 @@ characters (space, tab, return, newline, formfeed, vertical tab).\n\ If maxsplit is given, at most maxsplit splits are done."); static PyObject * -bytes_split(PyBytesObject *self, PyObject *args) +bytes_split(PyBytesObject *self, PyObject *args, PyObject *kwds) { + static char *kwlist[] = {"sep", "maxsplit", 0}; Py_ssize_t len = PyBytes_GET_SIZE(self), n; Py_ssize_t maxsplit = -1; const char *s = PyBytes_AS_STRING(self), *sub; Py_buffer vsub; PyObject *list, *subobj = Py_None; - if (!PyArg_ParseTuple(args, "|On:split", &subobj, &maxsplit)) + if (!PyArg_ParseTupleAndKeywords(args, kwds, "|On:split", + kwlist, &subobj, &maxsplit)) return NULL; if (maxsplit < 0) maxsplit = PY_SSIZE_T_MAX; @@ -1060,7 +1062,7 @@ bytes_rpartition(PyBytesObject *self, PyObject *sep_obj) } PyDoc_STRVAR(rsplit__doc__, -"B.rsplit([sep[, maxsplit]]) -> list of bytes\n\ +"B.rsplit(sep=None, maxsplit=-1) -> list of bytes\n\ \n\ Return a list of the sections in B, using sep as the delimiter,\n\ starting at the end of B and working to the front.\n\ @@ -1070,15 +1072,17 @@ If maxsplit is given, at most maxsplit splits are done."); static PyObject * -bytes_rsplit(PyBytesObject *self, PyObject *args) +bytes_rsplit(PyBytesObject *self, PyObject *args, PyObject *kwds) { + static char *kwlist[] = {"sep", "maxsplit", 0}; Py_ssize_t len = PyBytes_GET_SIZE(self), n; Py_ssize_t maxsplit = -1; const char *s = PyBytes_AS_STRING(self), *sub; Py_buffer vsub; PyObject *list, *subobj = Py_None; - if (!PyArg_ParseTuple(args, "|On:rsplit", &subobj, &maxsplit)) + if (!PyArg_ParseTupleAndKeywords(args, kwds, "|On:rsplit", + kwlist, &subobj, &maxsplit)) return NULL; if (maxsplit < 0) maxsplit = PY_SSIZE_T_MAX; @@ -2470,9 +2474,9 @@ bytes_methods[] = { {"rjust", (PyCFunction)stringlib_rjust, METH_VARARGS, rjust__doc__}, {"rpartition", (PyCFunction)bytes_rpartition, METH_O, rpartition__doc__}, - {"rsplit", (PyCFunction)bytes_rsplit, METH_VARARGS, rsplit__doc__}, + {"rsplit", (PyCFunction)bytes_rsplit, METH_VARARGS | METH_KEYWORDS, rsplit__doc__}, {"rstrip", (PyCFunction)bytes_rstrip, METH_VARARGS, rstrip__doc__}, - {"split", (PyCFunction)bytes_split, METH_VARARGS, split__doc__}, + {"split", (PyCFunction)bytes_split, METH_VARARGS | METH_KEYWORDS, split__doc__}, {"splitlines", (PyCFunction)bytes_splitlines, METH_VARARGS | METH_KEYWORDS, splitlines__doc__}, {"startswith", (PyCFunction)bytes_startswith, METH_VARARGS, |