@@ -10457,10 +10457,10 @@ _PyUnicode_JoinArray(PyObject *separator, PyObject *const *items, Py_ssize_t seq
1045710457 PyObject * item ;
1045810458 Py_ssize_t sz , i , res_offset ;
1045910459 Py_UCS4 maxchar ;
10460- Py_UCS4 item_maxchar ;
10460+ /* kinds is the bitwise OR of the PyUnicode_KIND() values (1, 2, 4). */
10461+ unsigned int kinds = 0 , all_ascii = 1 ;
1046110462 int use_memcpy ;
1046210463 unsigned char * res_data = NULL , * sep_data = NULL ;
10463- PyObject * last_obj ;
1046410464 int kind = 0 ;
1046510465
1046610466 /* If empty sequence, return u"". */
@@ -10469,14 +10469,12 @@ _PyUnicode_JoinArray(PyObject *separator, PyObject *const *items, Py_ssize_t seq
1046910469 }
1047010470
1047110471 /* If singleton sequence with an exact Unicode, return that. */
10472- last_obj = NULL ;
1047310472 if (seqlen == 1 ) {
1047410473 if (PyUnicode_CheckExact (items [0 ])) {
1047510474 res = items [0 ];
1047610475 return Py_NewRef (res );
1047710476 }
1047810477 seplen = 0 ;
10479- maxchar = 0 ;
1048010478 }
1048110479 else {
1048210480 /* Set up sep and seplen */
@@ -10486,7 +10484,6 @@ _PyUnicode_JoinArray(PyObject *separator, PyObject *const *items, Py_ssize_t seq
1048610484 if (!sep )
1048710485 goto onError ;
1048810486 seplen = 1 ;
10489- maxchar = 32 ;
1049010487 }
1049110488 else {
1049210489 if (!PyUnicode_Check (separator )) {
@@ -10498,12 +10495,12 @@ _PyUnicode_JoinArray(PyObject *separator, PyObject *const *items, Py_ssize_t seq
1049810495 }
1049910496 sep = separator ;
1050010497 seplen = PyUnicode_GET_LENGTH (separator );
10501- maxchar = PyUnicode_MAX_CHAR_VALUE (separator );
1050210498 /* inc refcount to keep this code path symmetric with the
1050310499 above case of a blank separator */
1050410500 Py_INCREF (sep );
1050510501 }
10506- last_obj = sep ;
10502+ kinds = PyUnicode_KIND (sep );
10503+ all_ascii = PyUnicode_IS_ASCII (sep );
1050710504 }
1050810505
1050910506 /* There are at least two things to join, or else we have a subclass
@@ -10527,23 +10524,34 @@ _PyUnicode_JoinArray(PyObject *separator, PyObject *const *items, Py_ssize_t seq
1052710524 i , Py_TYPE (item )-> tp_name );
1052810525 goto onError ;
1052910526 }
10530- add_sz = PyUnicode_GET_LENGTH (item );
10531- item_maxchar = PyUnicode_MAX_CHAR_VALUE (item );
10532- maxchar = Py_MAX (maxchar , item_maxchar );
10533- if (i != 0 ) {
10534- add_sz += seplen ;
10535- }
10527+ /* Count one separator per item; the extra one is removed below. */
10528+ add_sz = (size_t )PyUnicode_GET_LENGTH (item ) + (size_t )seplen ;
10529+ kinds |= PyUnicode_KIND (item );
10530+ all_ascii &= PyUnicode_IS_ASCII (item );
1053610531 if (add_sz > (size_t )(PY_SSIZE_T_MAX - sz )) {
1053710532 PyErr_SetString (PyExc_OverflowError ,
1053810533 "join() result is too long for a Python string" );
1053910534 goto onError ;
1054010535 }
1054110536 sz += add_sz ;
10542- if (use_memcpy && last_obj != NULL ) {
10543- if (PyUnicode_KIND (last_obj ) != PyUnicode_KIND (item ))
10544- use_memcpy = 0 ;
10545- }
10546- last_obj = item ;
10537+ }
10538+ sz -= seplen ;
10539+ /* memcpy() can be used if the separator and all items have the same
10540+ kind: the result then has that kind too. */
10541+ if (kinds & (kinds - 1 )) {
10542+ use_memcpy = 0 ;
10543+ }
10544+ if (all_ascii ) {
10545+ maxchar = 0x7f ;
10546+ }
10547+ else if (kinds & PyUnicode_4BYTE_KIND ) {
10548+ maxchar = MAX_UNICODE ;
10549+ }
10550+ else if (kinds & PyUnicode_2BYTE_KIND ) {
10551+ maxchar = 0xffff ;
10552+ }
10553+ else {
10554+ maxchar = 0xff ;
1054710555 }
1054810556
1054910557 res = PyUnicode_New (sz , maxchar );
0 commit comments