| 1 |
/* stringlib: bytes joining implementation */ |
| 2 |
|
| 3 |
#if STRINGLIB_IS_UNICODE |
| 4 |
#error join.h only compatible with byte-wise strings |
| 5 |
#endif |
| 6 |
|
| 7 |
Py_LOCAL_INLINE(PyObject *) |
| 8 |
STRINGLIB(bytes_join)(PyObject *sep, PyObject *iterable) |
| 9 |
{ |
| 10 |
const char *sepstr = STRINGLIB_STR(sep); |
| 11 |
Py_ssize_t seplen = STRINGLIB_LEN(sep); |
| 12 |
PyObject *res = NULL; |
| 13 |
char *p; |
| 14 |
Py_ssize_t seqlen = 0; |
| 15 |
Py_ssize_t sz = 0; |
| 16 |
Py_ssize_t i, nbufs; |
| 17 |
PyObject *seq, *item; |
| 18 |
Py_buffer *buffers = NULL; |
| 19 |
#define NB_STATIC_BUFFERS 10 |
| 20 |
Py_buffer static_buffers[NB_STATIC_BUFFERS]; |
| 21 |
#define GIL_THRESHOLD 1048576 |
| 22 |
int drop_gil = 1; |
| 23 |
PyThreadState *save = NULL; |
| 24 |
|
| 25 |
seq = PySequence_Fast(iterable, "can only join an iterable"); |
| 26 |
if (seq == NULL) { |
| 27 |
return NULL; |
| 28 |
} |
| 29 |
|
| 30 |
seqlen = PySequence_Fast_GET_SIZE(seq); |
| 31 |
if (seqlen == 0) { |
| 32 |
Py_DECREF(seq); |
| 33 |
return STRINGLIB_NEW(NULL, 0); |
| 34 |
} |
| 35 |
#if !STRINGLIB_MUTABLE |
| 36 |
if (seqlen == 1) { |
| 37 |
item = PySequence_Fast_GET_ITEM(seq, 0); |
| 38 |
if (STRINGLIB_CHECK_EXACT(item)) { |
| 39 |
Py_INCREF(item); |
| 40 |
Py_DECREF(seq); |
| 41 |
return item; |
| 42 |
} |
| 43 |
} |
| 44 |
#endif |
| 45 |
if (seqlen > NB_STATIC_BUFFERS) { |
| 46 |
buffers = PyMem_NEW(Py_buffer, seqlen); |
| 47 |
if (buffers == NULL) { |
| 48 |
Py_DECREF(seq); |
| 49 |
PyErr_NoMemory(); |
| 50 |
return NULL; |
| 51 |
} |
| 52 |
} |
| 53 |
else { |
| 54 |
buffers = static_buffers; |
| 55 |
} |
| 56 |
|
| 57 |
/* Here is the general case. Do a pre-pass to figure out the total |
| 58 |
* amount of space we'll need (sz), and see whether all arguments are |
| 59 |
* bytes-like. |
| 60 |
*/ |
| 61 |
for (i = 0, nbufs = 0; i < seqlen; i++) { |
| 62 |
Py_ssize_t itemlen; |
| 63 |
item = PySequence_Fast_GET_ITEM(seq, i); |
| 64 |
if (PyBytes_CheckExact(item)) { |
| 65 |
/* Fast path. */ |
| 66 |
buffers[i].obj = Py_NewRef(item); |
| 67 |
buffers[i].buf = PyBytes_AS_STRING(item); |
| 68 |
buffers[i].len = PyBytes_GET_SIZE(item); |
| 69 |
} |
| 70 |
else { |
| 71 |
if (PyObject_GetBuffer(item, &buffers[i], PyBUF_SIMPLE) != 0) { |
| 72 |
PyErr_Format(PyExc_TypeError, |
| 73 |
"sequence item %zd: expected a bytes-like object, " |
| 74 |
"%.80s found", |
| 75 |
i, Py_TYPE(item)->tp_name); |
| 76 |
goto error; |
| 77 |
} |
| 78 |
/* If the backing objects are mutable, then dropping the GIL |
| 79 |
* opens up race conditions where another thread tries to modify |
| 80 |
* the object which we hold a buffer on it. Such code has data |
| 81 |
* races anyway, but this is a conservative approach that avoids |
| 82 |
* changing the behaviour of that data race. |
| 83 |
*/ |
| 84 |
drop_gil = 0; |
| 85 |
} |
| 86 |
nbufs = i + 1; /* for error cleanup */ |
| 87 |
itemlen = buffers[i].len; |
| 88 |
if (itemlen > PY_SSIZE_T_MAX - sz) { |
| 89 |
PyErr_SetString(PyExc_OverflowError, |
| 90 |
"join() result is too long"); |
| 91 |
goto error; |
| 92 |
} |
| 93 |
sz += itemlen; |
| 94 |
if (i != 0) { |
| 95 |
if (seplen > PY_SSIZE_T_MAX - sz) { |
| 96 |
PyErr_SetString(PyExc_OverflowError, |
| 97 |
"join() result is too long"); |
| 98 |
goto error; |
| 99 |
} |
| 100 |
sz += seplen; |
| 101 |
} |
| 102 |
if (seqlen != PySequence_Fast_GET_SIZE(seq)) { |
| 103 |
PyErr_SetString(PyExc_RuntimeError, |
| 104 |
"sequence changed size during iteration"); |
| 105 |
goto error; |
| 106 |
} |
| 107 |
} |
| 108 |
|
| 109 |
/* Allocate result space. */ |
| 110 |
res = STRINGLIB_NEW(NULL, sz); |
| 111 |
if (res == NULL) |
| 112 |
goto error; |
| 113 |
|
| 114 |
/* Catenate everything. */ |
| 115 |
p = STRINGLIB_STR(res); |
| 116 |
if (sz < GIL_THRESHOLD) { |
| 117 |
drop_gil = 0; /* Benefits are likely outweighed by the overheads */ |
| 118 |
} |
| 119 |
if (drop_gil) { |
| 120 |
save = PyEval_SaveThread(); |
| 121 |
} |
| 122 |
if (!seplen) { |
| 123 |
/* fast path */ |
| 124 |
for (i = 0; i < nbufs; i++) { |
| 125 |
Py_ssize_t n = buffers[i].len; |
| 126 |
char *q = buffers[i].buf; |
| 127 |
memcpy(p, q, n); |
| 128 |
p += n; |
| 129 |
} |
| 130 |
} |
| 131 |
else { |
| 132 |
for (i = 0; i < nbufs; i++) { |
| 133 |
Py_ssize_t n; |
| 134 |
char *q; |
| 135 |
if (i) { |
| 136 |
memcpy(p, sepstr, seplen); |
| 137 |
p += seplen; |
| 138 |
} |
| 139 |
n = buffers[i].len; |
| 140 |
q = buffers[i].buf; |
| 141 |
memcpy(p, q, n); |
| 142 |
p += n; |
| 143 |
} |
| 144 |
} |
| 145 |
if (drop_gil) { |
| 146 |
PyEval_RestoreThread(save); |
| 147 |
} |
| 148 |
goto done; |
| 149 |
|
| 150 |
error: |
| 151 |
res = NULL; |
| 152 |
done: |
| 153 |
Py_DECREF(seq); |
| 154 |
for (i = 0; i < nbufs; i++) |
| 155 |
PyBuffer_Release(&buffers[i]); |
| 156 |
if (buffers != static_buffers) |
| 157 |
PyMem_Free(buffers); |
| 158 |
return res; |
| 159 |
} |
| 160 |
|
| 161 |
#undef NB_STATIC_BUFFERS |
| 162 |
#undef GIL_THRESHOLD |