diff options
author | Marc-André Lemburg <mal@egenix.com> | 2000-07-05 11:24:13 (GMT) |
---|---|---|
committer | Marc-André Lemburg <mal@egenix.com> | 2000-07-05 11:24:13 (GMT) |
commit | 5f0e29eb45c0ee5913e33b7f94b7bdc5f5b17b79 (patch) | |
tree | 4ccd7c1faff28d905d8b213d5882611760589d30 | |
parent | adf63d6c246ca0356b633e995402b8e4a32f8c0a (diff) | |
download | cpython-5f0e29eb45c0ee5913e33b7f94b7bdc5f5b17b79.zip cpython-5f0e29eb45c0ee5913e33b7f94b7bdc5f5b17b79.tar.gz cpython-5f0e29eb45c0ee5913e33b7f94b7bdc5f5b17b79.tar.bz2 |
Modified the encode interfaces of all builtin codecs to accept
any object which can be converted to Unicode by means of
PyUnicode_FromObject().
-rw-r--r-- | Modules/_codecsmodule.c | 155 |
1 files changed, 105 insertions, 50 deletions
diff --git a/Modules/_codecsmodule.c b/Modules/_codecsmodule.c index 4f368f8..a7be915 100644 --- a/Modules/_codecsmodule.c +++ b/Modules/_codecsmodule.c @@ -20,8 +20,14 @@ <encoding>_decode(char_buffer_obj[,errors='strict']) -> (Unicode object, bytes consumed) + <encoding>_encode() interfaces also accept non-Unicode object as + input. The objects are then converted to Unicode using + PyUnicode_FromObject() prior to applying the conversion. + These <encoding>s are available: utf_8, unicode_escape, - raw_unicode_escape, unicode_internal, latin_1, ascii (7-bit) + raw_unicode_escape, unicode_internal, latin_1, ascii (7-bit), + mbcs (on win32). + Written by Marc-Andre Lemburg (mal@lemburg.com). @@ -344,17 +350,22 @@ static PyObject * utf_8_encode(PyObject *self, PyObject *args) { - PyObject *str; + PyObject *str, *v; const char *errors = NULL; - if (!PyArg_ParseTuple(args, "U|z:utf_8_encode", + if (!PyArg_ParseTuple(args, "O|z:utf_8_encode", &str, &errors)) return NULL; - return codec_tuple(PyUnicode_EncodeUTF8(PyUnicode_AS_UNICODE(str), - PyUnicode_GET_SIZE(str), - errors), - PyUnicode_GET_SIZE(str)); + str = PyUnicode_FromObject(str); + if (str == NULL) + return NULL; + v = codec_tuple(PyUnicode_EncodeUTF8(PyUnicode_AS_UNICODE(str), + PyUnicode_GET_SIZE(str), + errors), + PyUnicode_GET_SIZE(str)); + Py_DECREF(str); + return v; } /* This version provides access to the byteorder parameter of the @@ -368,147 +379,186 @@ static PyObject * utf_16_encode(PyObject *self, PyObject *args) { - PyObject *str; + PyObject *str, *v; const char *errors = NULL; int byteorder = 0; - if (!PyArg_ParseTuple(args, "U|zi:utf_16_encode", + if (!PyArg_ParseTuple(args, "O|zi:utf_16_encode", &str, &errors, &byteorder)) return NULL; - return codec_tuple(PyUnicode_EncodeUTF16(PyUnicode_AS_UNICODE(str), - PyUnicode_GET_SIZE(str), - errors, - byteorder), - PyUnicode_GET_SIZE(str)); + str = PyUnicode_FromObject(str); + if (str == NULL) + return NULL; + v = codec_tuple(PyUnicode_EncodeUTF16(PyUnicode_AS_UNICODE(str), + PyUnicode_GET_SIZE(str), + errors, + byteorder), + PyUnicode_GET_SIZE(str)); + Py_DECREF(str); + return v; } static PyObject * utf_16_le_encode(PyObject *self, PyObject *args) { - PyObject *str; + PyObject *str, *v; const char *errors = NULL; - if (!PyArg_ParseTuple(args, "U|zi:utf_16_le_encode", + if (!PyArg_ParseTuple(args, "O|zi:utf_16_le_encode", &str, &errors)) return NULL; - return codec_tuple(PyUnicode_EncodeUTF16(PyUnicode_AS_UNICODE(str), + str = PyUnicode_FromObject(str); + if (str == NULL) + return NULL; + v = codec_tuple(PyUnicode_EncodeUTF16(PyUnicode_AS_UNICODE(str), PyUnicode_GET_SIZE(str), errors, -1), PyUnicode_GET_SIZE(str)); + Py_DECREF(str); + return v; } static PyObject * utf_16_be_encode(PyObject *self, PyObject *args) { - PyObject *str; + PyObject *str, *v; const char *errors = NULL; - if (!PyArg_ParseTuple(args, "U|zi:utf_16_be_encode", + if (!PyArg_ParseTuple(args, "O|zi:utf_16_be_encode", &str, &errors)) return NULL; - return codec_tuple(PyUnicode_EncodeUTF16(PyUnicode_AS_UNICODE(str), - PyUnicode_GET_SIZE(str), - errors, - +1), - PyUnicode_GET_SIZE(str)); + str = PyUnicode_FromObject(str); + if (str == NULL) + return NULL; + v = codec_tuple(PyUnicode_EncodeUTF16(PyUnicode_AS_UNICODE(str), + PyUnicode_GET_SIZE(str), + errors, + +1), + PyUnicode_GET_SIZE(str)); + Py_DECREF(str); + return v; } static PyObject * unicode_escape_encode(PyObject *self, PyObject *args) { - PyObject *str; + PyObject *str, *v; const char *errors = NULL; - if (!PyArg_ParseTuple(args, "U|z:unicode_escape_encode", + if (!PyArg_ParseTuple(args, "O|z:unicode_escape_encode", &str, &errors)) return NULL; - return codec_tuple(PyUnicode_EncodeUnicodeEscape( - PyUnicode_AS_UNICODE(str), - PyUnicode_GET_SIZE(str)), - PyUnicode_GET_SIZE(str)); + str = PyUnicode_FromObject(str); + if (str == NULL) + return NULL; + v = codec_tuple(PyUnicode_EncodeUnicodeEscape(PyUnicode_AS_UNICODE(str), + PyUnicode_GET_SIZE(str)), + PyUnicode_GET_SIZE(str)); + Py_DECREF(str); + return v; } static PyObject * raw_unicode_escape_encode(PyObject *self, PyObject *args) { - PyObject *str; + PyObject *str, *v; const char *errors = NULL; - if (!PyArg_ParseTuple(args, "U|z:raw_unicode_escape_encode", + if (!PyArg_ParseTuple(args, "O|z:raw_unicode_escape_encode", &str, &errors)) return NULL; - return codec_tuple(PyUnicode_EncodeRawUnicodeEscape( + str = PyUnicode_FromObject(str); + if (str == NULL) + return NULL; + v = codec_tuple(PyUnicode_EncodeRawUnicodeEscape( PyUnicode_AS_UNICODE(str), PyUnicode_GET_SIZE(str)), - PyUnicode_GET_SIZE(str)); + PyUnicode_GET_SIZE(str)); + Py_DECREF(str); + return v; } static PyObject * latin_1_encode(PyObject *self, PyObject *args) { - PyObject *str; + PyObject *str, *v; const char *errors = NULL; - if (!PyArg_ParseTuple(args, "U|z:latin_1_encode", + if (!PyArg_ParseTuple(args, "O|z:latin_1_encode", &str, &errors)) return NULL; - return codec_tuple(PyUnicode_EncodeLatin1( + str = PyUnicode_FromObject(str); + if (str == NULL) + return NULL; + v = codec_tuple(PyUnicode_EncodeLatin1( PyUnicode_AS_UNICODE(str), PyUnicode_GET_SIZE(str), errors), - PyUnicode_GET_SIZE(str)); + PyUnicode_GET_SIZE(str)); + Py_DECREF(str); + return v; } static PyObject * ascii_encode(PyObject *self, PyObject *args) { - PyObject *str; + PyObject *str, *v; const char *errors = NULL; - if (!PyArg_ParseTuple(args, "U|z:ascii_encode", + if (!PyArg_ParseTuple(args, "O|z:ascii_encode", &str, &errors)) return NULL; - return codec_tuple(PyUnicode_EncodeASCII( + str = PyUnicode_FromObject(str); + if (str == NULL) + return NULL; + v = codec_tuple(PyUnicode_EncodeASCII( PyUnicode_AS_UNICODE(str), PyUnicode_GET_SIZE(str), errors), - PyUnicode_GET_SIZE(str)); + PyUnicode_GET_SIZE(str)); + Py_DECREF(str); + return v; } static PyObject * charmap_encode(PyObject *self, PyObject *args) { - PyObject *str; + PyObject *str, *v; const char *errors = NULL; PyObject *mapping = NULL; - if (!PyArg_ParseTuple(args, "U|zO:charmap_encode", + if (!PyArg_ParseTuple(args, "O|zO:charmap_encode", &str, &errors, &mapping)) return NULL; if (mapping == Py_None) mapping = NULL; - return codec_tuple(PyUnicode_EncodeCharmap( + str = PyUnicode_FromObject(str); + if (str == NULL) + return NULL; + v = codec_tuple(PyUnicode_EncodeCharmap( PyUnicode_AS_UNICODE(str), PyUnicode_GET_SIZE(str), mapping, errors), - PyUnicode_GET_SIZE(str)); + PyUnicode_GET_SIZE(str)); + Py_DECREF(str); + return v; } #ifdef MS_WIN32 @@ -517,18 +567,23 @@ static PyObject * mbcs_encode(PyObject *self, PyObject *args) { - PyObject *str; + PyObject *str, *v; const char *errors = NULL; - if (!PyArg_ParseTuple(args, "U|z:mbcs_encode", + if (!PyArg_ParseTuple(args, "O|z:mbcs_encode", &str, &errors)) return NULL; - return codec_tuple(PyUnicode_EncodeMBCS( + str = PyUnicode_FromObject(str); + if (str == NULL) + return NULL; + v = codec_tuple(PyUnicode_EncodeMBCS( PyUnicode_AS_UNICODE(str), PyUnicode_GET_SIZE(str), errors), - PyUnicode_GET_SIZE(str)); + PyUnicode_GET_SIZE(str)); + Py_DECREF(str); + return v; } #endif /* MS_WIN32 */ |