cpython/Objects/listobject.c

2924 lines
69 KiB
C
Raw Normal View History

1990-10-14 09:07:46 -03:00
/* List object implementation */
1997-05-02 00:12:38 -03:00
#include "Python.h"
1994-08-29 09:45:32 -03:00
#ifdef STDC_HEADERS
#include <stddef.h>
#else
#include <sys/types.h> /* For size_t */
#endif
1990-10-14 09:07:46 -03:00
/* Ensure ob_item has room for at least newsize elements, and set
* ob_size to newsize. If newsize > ob_size on entry, the content
* of the new slots at exit is undefined heap trash; it's the caller's
* responsiblity to overwrite them with sane values.
* The number of allocated elements may grow, shrink, or stay the same.
* Failure is impossible if newsize <= self.allocated on entry, although
* that partly relies on an assumption that the system realloc() never
* fails when passed a number of bytes <= the number of bytes last
* allocated (the C standard doesn't guarantee this, but it's hard to
* imagine a realloc implementation where it wouldn't be true).
* Note that self->ob_item may change, and even if newsize is less
* than ob_size on entry.
*/
static int
2006-02-15 13:27:45 -04:00
list_resize(PyListObject *self, Py_ssize_t newsize)
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
{
PyObject **items;
size_t new_allocated;
2006-02-15 13:27:45 -04:00
Py_ssize_t allocated = self->allocated;
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
/* Bypass realloc() when a previous overallocation is large enough
to accommodate the newsize. If the newsize falls lower than half
the allocated size, then proceed with the realloc() to shrink the list.
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
*/
if (allocated >= newsize && newsize >= (allocated >> 1)) {
assert(self->ob_item != NULL || newsize == 0);
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
self->ob_size = newsize;
return 0;
}
/* This over-allocates proportional to the list size, making room
* for additional growth. The over-allocation is mild, but is
* enough to give linear-time amortized behavior over a long
* sequence of appends() in the presence of a poorly-performing
* system realloc().
* The growth pattern is: 0, 4, 8, 16, 25, 35, 46, 58, 72, 88, ...
*/
new_allocated = (newsize >> 3) + (newsize < 9 ? 3 : 6) + newsize;
if (newsize == 0)
new_allocated = 0;
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
items = self->ob_item;
if (new_allocated <= ((~(size_t)0) / sizeof(PyObject *)))
PyMem_RESIZE(items, PyObject *, new_allocated);
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
else
items = NULL;
if (items == NULL) {
PyErr_NoMemory();
return -1;
}
self->ob_item = items;
self->ob_size = newsize;
self->allocated = new_allocated;
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
return 0;
}
/* Empty list reuse scheme to save calls to malloc and free */
#define MAXFREELISTS 80
static PyListObject *free_lists[MAXFREELISTS];
static int num_free_lists = 0;
2004-10-07 00:58:07 -03:00
void
PyList_Fini(void)
{
PyListObject *op;
while (num_free_lists) {
num_free_lists--;
op = free_lists[num_free_lists];
assert(PyList_CheckExact(op));
PyObject_GC_Del(op);
}
}
1997-05-02 00:12:38 -03:00
PyObject *
2006-02-15 13:27:45 -04:00
PyList_New(Py_ssize_t size)
1990-10-14 09:07:46 -03:00
{
1997-05-02 00:12:38 -03:00
PyListObject *op;
1994-08-29 09:45:32 -03:00
size_t nbytes;
1990-10-14 09:07:46 -03:00
if (size < 0) {
1997-05-02 00:12:38 -03:00
PyErr_BadInternalCall();
1990-10-14 09:07:46 -03:00
return NULL;
}
nbytes = size * sizeof(PyObject *);
/* Check for overflow */
if (nbytes / sizeof(PyObject *) != (size_t)size)
1997-05-02 00:12:38 -03:00
return PyErr_NoMemory();
if (num_free_lists) {
num_free_lists--;
op = free_lists[num_free_lists];
_Py_NewReference((PyObject *)op);
} else {
op = PyObject_GC_New(PyListObject, &PyList_Type);
if (op == NULL)
return NULL;
1990-10-14 09:07:46 -03:00
}
if (size <= 0)
1990-10-14 09:07:46 -03:00
op->ob_item = NULL;
else {
op->ob_item = (PyObject **) PyMem_MALLOC(nbytes);
if (op->ob_item == NULL)
1997-05-02 00:12:38 -03:00
return PyErr_NoMemory();
memset(op->ob_item, 0, nbytes);
1990-10-14 09:07:46 -03:00
}
2001-08-29 20:54:21 -03:00
op->ob_size = size;
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
op->allocated = size;
2001-08-29 20:54:21 -03:00
_PyObject_GC_TRACK(op);
1997-05-02 00:12:38 -03:00
return (PyObject *) op;
1990-10-14 09:07:46 -03:00
}
2006-02-15 13:27:45 -04:00
Py_ssize_t
2000-07-09 12:16:51 -03:00
PyList_Size(PyObject *op)
1990-10-14 09:07:46 -03:00
{
1997-05-02 00:12:38 -03:00
if (!PyList_Check(op)) {
PyErr_BadInternalCall();
1990-10-14 09:07:46 -03:00
return -1;
}
else
1997-05-02 00:12:38 -03:00
return ((PyListObject *)op) -> ob_size;
1990-10-14 09:07:46 -03:00
}
static PyObject *indexerr = NULL;
1997-05-02 00:12:38 -03:00
PyObject *
2006-02-15 13:27:45 -04:00
PyList_GetItem(PyObject *op, Py_ssize_t i)
1990-10-14 09:07:46 -03:00
{
1997-05-02 00:12:38 -03:00
if (!PyList_Check(op)) {
PyErr_BadInternalCall();
1990-10-14 09:07:46 -03:00
return NULL;
}
1997-05-02 00:12:38 -03:00
if (i < 0 || i >= ((PyListObject *)op) -> ob_size) {
if (indexerr == NULL)
1997-05-02 00:12:38 -03:00
indexerr = PyString_FromString(
"list index out of range");
PyErr_SetObject(PyExc_IndexError, indexerr);
1990-10-14 09:07:46 -03:00
return NULL;
}
1997-05-02 00:12:38 -03:00
return ((PyListObject *)op) -> ob_item[i];
1990-10-14 09:07:46 -03:00
}
int
2006-02-15 13:27:45 -04:00
PyList_SetItem(register PyObject *op, register Py_ssize_t i,
2000-07-09 12:16:51 -03:00
register PyObject *newitem)
1990-10-14 09:07:46 -03:00
{
1997-05-02 00:12:38 -03:00
register PyObject *olditem;
register PyObject **p;
if (!PyList_Check(op)) {
Py_XDECREF(newitem);
PyErr_BadInternalCall();
1990-10-21 19:15:08 -03:00
return -1;
1990-10-14 09:07:46 -03:00
}
1997-05-02 00:12:38 -03:00
if (i < 0 || i >= ((PyListObject *)op) -> ob_size) {
Py_XDECREF(newitem);
PyErr_SetString(PyExc_IndexError,
"list assignment index out of range");
1990-10-21 19:15:08 -03:00
return -1;
1990-10-14 09:07:46 -03:00
}
1997-05-02 00:12:38 -03:00
p = ((PyListObject *)op) -> ob_item + i;
1995-03-09 08:12:50 -04:00
olditem = *p;
*p = newitem;
1997-05-02 00:12:38 -03:00
Py_XDECREF(olditem);
1990-10-14 09:07:46 -03:00
return 0;
}
static int
2006-02-15 13:27:45 -04:00
ins1(PyListObject *self, Py_ssize_t where, PyObject *v)
1990-10-14 09:07:46 -03:00
{
2006-02-15 13:27:45 -04:00
Py_ssize_t i, n = self->ob_size;
1997-05-02 00:12:38 -03:00
PyObject **items;
1990-10-21 19:15:08 -03:00
if (v == NULL) {
1997-05-02 00:12:38 -03:00
PyErr_BadInternalCall();
1990-10-21 19:15:08 -03:00
return -1;
}
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
if (n == INT_MAX) {
PyErr_SetString(PyExc_OverflowError,
"cannot add more objects to list");
return -1;
}
2004-07-28 23:29:26 -03:00
if (list_resize(self, n+1) == -1)
1990-10-21 19:15:08 -03:00
return -1;
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
if (where < 0) {
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
where += n;
if (where < 0)
where = 0;
}
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
if (where > n)
where = n;
items = self->ob_item;
for (i = n; --i >= where; )
1990-10-14 09:07:46 -03:00
items[i+1] = items[i];
1997-05-02 00:12:38 -03:00
Py_INCREF(v);
1990-10-14 09:07:46 -03:00
items[where] = v;
return 0;
}
int
2006-02-15 13:27:45 -04:00
PyList_Insert(PyObject *op, Py_ssize_t where, PyObject *newitem)
1990-10-14 09:07:46 -03:00
{
1997-05-02 00:12:38 -03:00
if (!PyList_Check(op)) {
PyErr_BadInternalCall();
1990-10-21 19:15:08 -03:00
return -1;
}
1997-05-02 00:12:38 -03:00
return ins1((PyListObject *)op, where, newitem);
1990-10-14 09:07:46 -03:00
}
static int
app1(PyListObject *self, PyObject *v)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t n = PyList_GET_SIZE(self);
assert (v != NULL);
if (n == INT_MAX) {
PyErr_SetString(PyExc_OverflowError,
"cannot add more objects to list");
return -1;
}
if (list_resize(self, n+1) == -1)
return -1;
Py_INCREF(v);
PyList_SET_ITEM(self, n, v);
return 0;
}
1990-10-14 09:07:46 -03:00
int
2000-07-09 12:16:51 -03:00
PyList_Append(PyObject *op, PyObject *newitem)
1990-10-14 09:07:46 -03:00
{
if (PyList_Check(op) && (newitem != NULL))
return app1((PyListObject *)op, newitem);
PyErr_BadInternalCall();
return -1;
1990-10-14 09:07:46 -03:00
}
/* Methods */
static void
2000-07-09 12:16:51 -03:00
list_dealloc(PyListObject *op)
1990-10-14 09:07:46 -03:00
{
2006-02-15 13:27:45 -04:00
Py_ssize_t i;
PyObject_GC_UnTrack(op);
Py_TRASHCAN_SAFE_BEGIN(op)
if (op->ob_item != NULL) {
/* Do it backwards, for Christian Tismer.
There's a simple test case where somehow this reduces
thrashing when a *very* large list is created and
immediately deleted. */
i = op->ob_size;
while (--i >= 0) {
1997-05-02 00:12:38 -03:00
Py_XDECREF(op->ob_item[i]);
}
PyMem_FREE(op->ob_item);
}
if (num_free_lists < MAXFREELISTS && PyList_CheckExact(op))
free_lists[num_free_lists++] = op;
2004-07-28 23:29:26 -03:00
else
op->ob_type->tp_free((PyObject *)op);
Py_TRASHCAN_SAFE_END(op)
1990-10-14 09:07:46 -03:00
}
1991-06-07 13:10:43 -03:00
static int
2000-07-09 12:16:51 -03:00
list_print(PyListObject *op, FILE *fp, int flags)
1990-10-14 09:07:46 -03:00
{
2006-02-15 13:27:45 -04:00
int rc;
Py_ssize_t i;
2006-02-15 13:27:45 -04:00
rc = Py_ReprEnter((PyObject*)op);
if (rc != 0) {
if (rc < 0)
return rc;
fprintf(fp, "[...]");
return 0;
}
1990-10-14 09:07:46 -03:00
fprintf(fp, "[");
1991-06-07 13:10:43 -03:00
for (i = 0; i < op->ob_size; i++) {
if (i > 0)
1990-10-14 09:07:46 -03:00
fprintf(fp, ", ");
if (PyObject_Print(op->ob_item[i], fp, 0) != 0) {
Py_ReprLeave((PyObject *)op);
1991-06-07 13:10:43 -03:00
return -1;
}
1990-10-14 09:07:46 -03:00
}
fprintf(fp, "]");
Py_ReprLeave((PyObject *)op);
1991-06-07 13:10:43 -03:00
return 0;
1990-10-14 09:07:46 -03:00
}
1997-05-02 00:12:38 -03:00
static PyObject *
2000-07-09 12:16:51 -03:00
list_repr(PyListObject *v)
1990-10-14 09:07:46 -03:00
{
2006-02-15 13:27:45 -04:00
Py_ssize_t i;
PyObject *s, *temp;
PyObject *pieces = NULL, *result = NULL;
i = Py_ReprEnter((PyObject*)v);
if (i != 0) {
return i > 0 ? PyString_FromString("[...]") : NULL;
}
if (v->ob_size == 0) {
result = PyString_FromString("[]");
goto Done;
1990-10-14 09:07:46 -03:00
}
pieces = PyList_New(0);
if (pieces == NULL)
goto Done;
/* Do repr() on each element. Note that this may mutate the list,
so must refetch the list size on each iteration. */
for (i = 0; i < v->ob_size; ++i) {
int status;
s = PyObject_Repr(v->ob_item[i]);
if (s == NULL)
goto Done;
status = PyList_Append(pieces, s);
Py_DECREF(s); /* append created a new ref */
if (status < 0)
goto Done;
}
/* Add "[]" decorations to the first and last items. */
assert(PyList_GET_SIZE(pieces) > 0);
s = PyString_FromString("[");
if (s == NULL)
goto Done;
temp = PyList_GET_ITEM(pieces, 0);
PyString_ConcatAndDel(&s, temp);
PyList_SET_ITEM(pieces, 0, s);
if (s == NULL)
goto Done;
s = PyString_FromString("]");
if (s == NULL)
goto Done;
temp = PyList_GET_ITEM(pieces, PyList_GET_SIZE(pieces) - 1);
PyString_ConcatAndDel(&temp, s);
PyList_SET_ITEM(pieces, PyList_GET_SIZE(pieces) - 1, temp);
if (temp == NULL)
goto Done;
/* Paste them all together with ", " between. */
s = PyString_FromString(", ");
if (s == NULL)
goto Done;
result = _PyString_Join(s, pieces);
2002-07-18 23:35:45 -03:00
Py_DECREF(s);
Done:
Py_XDECREF(pieces);
Py_ReprLeave((PyObject *)v);
return result;
1990-10-14 09:07:46 -03:00
}
2006-02-15 13:27:45 -04:00
static Py_ssize_t
2000-07-09 12:16:51 -03:00
list_length(PyListObject *a)
1990-10-14 09:07:46 -03:00
{
return a->ob_size;
}
static int
2000-07-09 12:16:51 -03:00
list_contains(PyListObject *a, PyObject *el)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t i;
int cmp;
for (i = 0, cmp = 0 ; cmp == 0 && i < a->ob_size; ++i)
cmp = PyObject_RichCompareBool(el, PyList_GET_ITEM(a, i),
Py_EQ);
return cmp;
}
1997-05-02 00:12:38 -03:00
static PyObject *
2006-02-15 13:27:45 -04:00
list_item(PyListObject *a, Py_ssize_t i)
1990-10-14 09:07:46 -03:00
{
if (i < 0 || i >= a->ob_size) {
if (indexerr == NULL)
1997-05-02 00:12:38 -03:00
indexerr = PyString_FromString(
"list index out of range");
PyErr_SetObject(PyExc_IndexError, indexerr);
1990-10-14 09:07:46 -03:00
return NULL;
}
1997-05-02 00:12:38 -03:00
Py_INCREF(a->ob_item[i]);
1990-10-14 09:07:46 -03:00
return a->ob_item[i];
}
1997-05-02 00:12:38 -03:00
static PyObject *
2006-02-15 13:27:45 -04:00
list_slice(PyListObject *a, Py_ssize_t ilow, Py_ssize_t ihigh)
1990-10-14 09:07:46 -03:00
{
1997-05-02 00:12:38 -03:00
PyListObject *np;
PyObject **src, **dest;
2006-02-15 13:27:45 -04:00
Py_ssize_t i, len;
1990-10-14 09:07:46 -03:00
if (ilow < 0)
ilow = 0;
else if (ilow > a->ob_size)
ilow = a->ob_size;
if (ihigh < ilow)
ihigh = ilow;
else if (ihigh > a->ob_size)
ihigh = a->ob_size;
len = ihigh - ilow;
np = (PyListObject *) PyList_New(len);
1990-10-14 09:07:46 -03:00
if (np == NULL)
return NULL;
src = a->ob_item + ilow;
dest = np->ob_item;
for (i = 0; i < len; i++) {
PyObject *v = src[i];
1997-05-02 00:12:38 -03:00
Py_INCREF(v);
dest[i] = v;
1990-10-14 09:07:46 -03:00
}
1997-05-02 00:12:38 -03:00
return (PyObject *)np;
1990-10-14 09:07:46 -03:00
}
1997-05-02 00:12:38 -03:00
PyObject *
2006-02-15 13:27:45 -04:00
PyList_GetSlice(PyObject *a, Py_ssize_t ilow, Py_ssize_t ihigh)
{
1997-05-02 00:12:38 -03:00
if (!PyList_Check(a)) {
PyErr_BadInternalCall();
return NULL;
}
1997-05-02 00:12:38 -03:00
return list_slice((PyListObject *)a, ilow, ihigh);
}
1997-05-02 00:12:38 -03:00
static PyObject *
2000-07-09 12:16:51 -03:00
list_concat(PyListObject *a, PyObject *bb)
1990-10-14 09:07:46 -03:00
{
2006-02-15 13:27:45 -04:00
Py_ssize_t size;
Py_ssize_t i;
PyObject **src, **dest;
1997-05-02 00:12:38 -03:00
PyListObject *np;
if (!PyList_Check(bb)) {
PyErr_Format(PyExc_TypeError,
"can only concatenate list (not \"%.200s\") to list",
bb->ob_type->tp_name);
1990-10-14 09:07:46 -03:00
return NULL;
}
1997-05-02 00:12:38 -03:00
#define b ((PyListObject *)bb)
1990-10-14 09:07:46 -03:00
size = a->ob_size + b->ob_size;
if (size < 0)
return PyErr_NoMemory();
1997-05-02 00:12:38 -03:00
np = (PyListObject *) PyList_New(size);
1990-10-14 09:07:46 -03:00
if (np == NULL) {
1991-06-07 13:10:43 -03:00
return NULL;
1990-10-14 09:07:46 -03:00
}
src = a->ob_item;
dest = np->ob_item;
1990-10-14 09:07:46 -03:00
for (i = 0; i < a->ob_size; i++) {
PyObject *v = src[i];
1997-05-02 00:12:38 -03:00
Py_INCREF(v);
dest[i] = v;
1990-10-14 09:07:46 -03:00
}
src = b->ob_item;
dest = np->ob_item + a->ob_size;
1990-10-14 09:07:46 -03:00
for (i = 0; i < b->ob_size; i++) {
PyObject *v = src[i];
1997-05-02 00:12:38 -03:00
Py_INCREF(v);
dest[i] = v;
1990-10-14 09:07:46 -03:00
}
1997-05-02 00:12:38 -03:00
return (PyObject *)np;
1990-10-14 09:07:46 -03:00
#undef b
}
1997-05-02 00:12:38 -03:00
static PyObject *
2006-02-15 13:27:45 -04:00
list_repeat(PyListObject *a, Py_ssize_t n)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t i, j;
Py_ssize_t size;
1997-05-02 00:12:38 -03:00
PyListObject *np;
PyObject **p, **items;
PyObject *elem;
if (n < 0)
n = 0;
size = a->ob_size * n;
if (size == 0)
return PyList_New(0);
if (n && size/n != a->ob_size)
return PyErr_NoMemory();
1997-05-02 00:12:38 -03:00
np = (PyListObject *) PyList_New(size);
if (np == NULL)
return NULL;
items = np->ob_item;
if (a->ob_size == 1) {
elem = a->ob_item[0];
for (i = 0; i < n; i++) {
items[i] = elem;
Py_INCREF(elem);
}
return (PyObject *) np;
}
p = np->ob_item;
items = a->ob_item;
for (i = 0; i < n; i++) {
for (j = 0; j < a->ob_size; j++) {
*p = items[j];
1997-05-02 00:12:38 -03:00
Py_INCREF(*p);
p++;
}
}
1997-05-02 00:12:38 -03:00
return (PyObject *) np;
}
static int
list_clear(PyListObject *a)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t i;
PyObject **item = a->ob_item;
if (item != NULL) {
/* Because XDECREF can recursively invoke operations on
this list, we make it empty first. */
i = a->ob_size;
a->ob_size = 0;
a->ob_item = NULL;
a->allocated = 0;
while (--i >= 0) {
Py_XDECREF(item[i]);
}
PyMem_FREE(item);
}
/* Never fails; the return value can be ignored.
Note that there is no guarantee that the list is actually empty
at this point, because XDECREF may have populated it again! */
return 0;
}
/* a[ilow:ihigh] = v if v != NULL.
* del a[ilow:ihigh] if v == NULL.
*
* Special speed gimmick: when v is NULL and ihigh - ilow <= 8, it's
* guaranteed the call cannot fail.
*/
1990-10-14 09:07:46 -03:00
static int
2006-02-15 13:27:45 -04:00
list_ass_slice(PyListObject *a, Py_ssize_t ilow, Py_ssize_t ihigh, PyObject *v)
1990-10-14 09:07:46 -03:00
{
1995-01-17 06:21:11 -04:00
/* Because [X]DECREF can recursively invoke list operations on
this list, we must postpone all [X]DECREF activity until
after the list is back in its canonical shape. Therefore
we must allocate an additional array, 'recycle', into which
we temporarily copy the items that are deleted from the
list. :-( */
PyObject *recycle_on_stack[8];
PyObject **recycle = recycle_on_stack; /* will allocate more if needed */
1997-05-02 00:12:38 -03:00
PyObject **item;
PyObject **vitem = NULL;
PyObject *v_as_SF = NULL; /* PySequence_Fast(v) */
2006-02-15 13:27:45 -04:00
Py_ssize_t n; /* # of elements in replacement list */
Py_ssize_t norig; /* # of elements in list getting replaced */
Py_ssize_t d; /* Change in size */
Py_ssize_t k;
size_t s;
int result = -1; /* guilty until proved innocent */
1997-05-02 00:12:38 -03:00
#define b ((PyListObject *)v)
1990-10-14 09:07:46 -03:00
if (v == NULL)
n = 0;
else {
if (a == b) {
/* Special case "a[i:j] = a" -- copy b first */
v = list_slice(b, 0, b->ob_size);
if (v == NULL)
return result;
result = list_ass_slice(a, ilow, ihigh, v);
1997-05-02 00:12:38 -03:00
Py_DECREF(v);
return result;
}
v_as_SF = PySequence_Fast(v, "can only assign an iterable");
if(v_as_SF == NULL)
goto Error;
n = PySequence_Fast_GET_SIZE(v_as_SF);
vitem = PySequence_Fast_ITEMS(v_as_SF);
}
1990-10-14 09:07:46 -03:00
if (ilow < 0)
ilow = 0;
else if (ilow > a->ob_size)
ilow = a->ob_size;
1990-10-14 09:07:46 -03:00
if (ihigh < ilow)
ihigh = ilow;
else if (ihigh > a->ob_size)
ihigh = a->ob_size;
norig = ihigh - ilow;
assert(norig >= 0);
d = n - norig;
if (a->ob_size + d == 0) {
Py_XDECREF(v_as_SF);
return list_clear(a);
}
item = a->ob_item;
/* recycle the items that we are about to remove */
s = norig * sizeof(PyObject *);
if (s > sizeof(recycle_on_stack)) {
recycle = (PyObject **)PyMem_MALLOC(s);
if (recycle == NULL) {
PyErr_NoMemory();
goto Error;
}
}
memcpy(recycle, &item[ilow], s);
if (d < 0) { /* Delete -d items */
memmove(&item[ihigh+d], &item[ihigh],
(a->ob_size - ihigh)*sizeof(PyObject *));
list_resize(a, a->ob_size + d);
item = a->ob_item;
1990-10-14 09:07:46 -03:00
}
else if (d > 0) { /* Insert d items */
k = a->ob_size;
if (list_resize(a, k+d) < 0)
goto Error;
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
item = a->ob_item;
memmove(&item[ihigh+d], &item[ihigh],
(k - ihigh)*sizeof(PyObject *));
1990-10-14 09:07:46 -03:00
}
for (k = 0; k < n; k++, ilow++) {
PyObject *w = vitem[k];
1997-05-02 00:12:38 -03:00
Py_XINCREF(w);
1990-10-14 09:07:46 -03:00
item[ilow] = w;
}
for (k = norig - 1; k >= 0; --k)
Py_XDECREF(recycle[k]);
result = 0;
Error:
if (recycle != recycle_on_stack)
PyMem_FREE(recycle);
Py_XDECREF(v_as_SF);
return result;
1990-10-14 09:07:46 -03:00
#undef b
}
int
2006-02-15 13:27:45 -04:00
PyList_SetSlice(PyObject *a, Py_ssize_t ilow, Py_ssize_t ihigh, PyObject *v)
{
1997-05-02 00:12:38 -03:00
if (!PyList_Check(a)) {
PyErr_BadInternalCall();
return -1;
}
1997-05-02 00:12:38 -03:00
return list_ass_slice((PyListObject *)a, ilow, ihigh, v);
}
static PyObject *
2006-02-15 13:27:45 -04:00
list_inplace_repeat(PyListObject *self, Py_ssize_t n)
{
PyObject **items;
2006-02-15 13:27:45 -04:00
Py_ssize_t size, i, j, p;
size = PyList_GET_SIZE(self);
if (size == 0) {
Py_INCREF(self);
return (PyObject *)self;
}
if (n < 1) {
(void)list_clear(self);
Py_INCREF(self);
return (PyObject *)self;
}
2004-07-28 23:29:26 -03:00
if (list_resize(self, size*n) == -1)
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
return NULL;
p = size;
items = self->ob_item;
for (i = 1; i < n; i++) { /* Start counting at 1, not 0 */
for (j = 0; j < size; j++) {
PyObject *o = items[j];
Py_INCREF(o);
items[p++] = o;
}
}
Py_INCREF(self);
return (PyObject *)self;
}
static int
2006-02-15 13:27:45 -04:00
list_ass_item(PyListObject *a, Py_ssize_t i, PyObject *v)
{
1997-05-02 00:12:38 -03:00
PyObject *old_value;
if (i < 0 || i >= a->ob_size) {
1997-05-02 00:12:38 -03:00
PyErr_SetString(PyExc_IndexError,
"list assignment index out of range");
return -1;
}
if (v == NULL)
return list_ass_slice(a, i, i+1, v);
1997-05-02 00:12:38 -03:00
Py_INCREF(v);
old_value = a->ob_item[i];
a->ob_item[i] = v;
2002-07-18 23:35:45 -03:00
Py_DECREF(old_value);
return 0;
}
1997-05-02 00:12:38 -03:00
static PyObject *
2000-07-09 12:16:51 -03:00
listinsert(PyListObject *self, PyObject *args)
1990-10-14 09:07:46 -03:00
{
2006-02-15 13:27:45 -04:00
Py_ssize_t i;
1997-05-02 00:12:38 -03:00
PyObject *v;
2006-02-15 13:27:45 -04:00
if (!PyArg_ParseTuple(args, "nO:insert", &i, &v))
1990-10-14 09:07:46 -03:00
return NULL;
if (ins1(self, i, v) == 0)
Py_RETURN_NONE;
return NULL;
1990-10-14 09:07:46 -03:00
}
1997-05-02 00:12:38 -03:00
static PyObject *
listappend(PyListObject *self, PyObject *v)
1990-10-14 09:07:46 -03:00
{
if (app1(self, v) == 0)
Py_RETURN_NONE;
return NULL;
1990-10-14 09:07:46 -03:00
}
static PyObject *
listextend(PyListObject *self, PyObject *b)
{
PyObject *it; /* iter(v) */
2006-02-15 13:27:45 -04:00
Py_ssize_t m; /* size of self */
Py_ssize_t n; /* guess for size of b */
Py_ssize_t mn; /* m + n */
Py_ssize_t i;
PyObject *(*iternext)(PyObject *);
/* Special cases:
2004-07-28 23:29:26 -03:00
1) lists and tuples which can use PySequence_Fast ops
2) extending self to self requires making a copy first
*/
if (PyList_CheckExact(b) || PyTuple_CheckExact(b) || (PyObject *)self == b) {
PyObject **src, **dest;
b = PySequence_Fast(b, "argument must be iterable");
if (!b)
return NULL;
n = PySequence_Fast_GET_SIZE(b);
if (n == 0) {
/* short circuit when b is empty */
Py_DECREF(b);
Py_RETURN_NONE;
}
m = self->ob_size;
if (list_resize(self, m + n) == -1) {
Py_DECREF(b);
return NULL;
}
/* note that we may still have self == b here for the
* situation a.extend(a), but the following code works
* in that case too. Just make sure to resize self
* before calling PySequence_Fast_ITEMS.
*/
/* populate the end of self with b's items */
src = PySequence_Fast_ITEMS(b);
dest = self->ob_item + m;
for (i = 0; i < n; i++) {
PyObject *o = src[i];
Py_INCREF(o);
dest[i] = o;
}
Py_DECREF(b);
Py_RETURN_NONE;
}
it = PyObject_GetIter(b);
if (it == NULL)
return NULL;
iternext = *it->ob_type->tp_iternext;
/* Guess a result list size. */
n = _PyObject_LengthHint(b);
if (n < 0) {
if (!PyErr_ExceptionMatches(PyExc_TypeError) &&
!PyErr_ExceptionMatches(PyExc_AttributeError)) {
Py_DECREF(it);
return NULL;
}
PyErr_Clear();
n = 8; /* arbitrary */
}
m = self->ob_size;
mn = m + n;
if (mn >= m) {
/* Make room. */
if (list_resize(self, mn) == -1)
goto error;
/* Make the list sane again. */
self->ob_size = m;
}
/* Else m + n overflowed; on the chance that n lied, and there really
* is enough room, ignore it. If n was telling the truth, we'll
* eventually run out of memory during the loop.
*/
/* Run iterator to exhaustion. */
for (;;) {
PyObject *item = iternext(it);
if (item == NULL) {
if (PyErr_Occurred()) {
if (PyErr_ExceptionMatches(PyExc_StopIteration))
PyErr_Clear();
else
goto error;
}
break;
}
if (self->ob_size < self->allocated) {
/* steals ref */
PyList_SET_ITEM(self, self->ob_size, item);
++self->ob_size;
}
else {
int status = app1(self, item);
Py_DECREF(item); /* append creates a new ref */
if (status < 0)
goto error;
}
}
/* Cut back result list if initial guess was too large. */
if (self->ob_size < self->allocated)
list_resize(self, self->ob_size); /* shrinking can't fail */
Py_DECREF(it);
Py_RETURN_NONE;
error:
Py_DECREF(it);
return NULL;
}
PyObject *
_PyList_Extend(PyListObject *self, PyObject *b)
{
return listextend(self, b);
}
static PyObject *
list_inplace_concat(PyListObject *self, PyObject *other)
{
PyObject *result;
result = listextend(self, other);
if (result == NULL)
return result;
Py_DECREF(result);
Py_INCREF(self);
return (PyObject *)self;
}
static PyObject *
2000-07-09 12:16:51 -03:00
listpop(PyListObject *self, PyObject *args)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t i = -1;
PyObject *v, *arg = NULL;
int status;
if (!PyArg_UnpackTuple(args, "pop", 0, 1, &arg))
return NULL;
if (arg != NULL) {
if (PyInt_Check(arg))
2006-02-15 13:27:45 -04:00
i = PyInt_AS_LONG((PyIntObject*) arg);
else if (!PyArg_ParseTuple(args, "|n:pop", &i))
return NULL;
}
if (self->ob_size == 0) {
/* Special-case most common failure cause */
PyErr_SetString(PyExc_IndexError, "pop from empty list");
return NULL;
}
if (i < 0)
i += self->ob_size;
if (i < 0 || i >= self->ob_size) {
PyErr_SetString(PyExc_IndexError, "pop index out of range");
return NULL;
}
v = self->ob_item[i];
if (i == self->ob_size - 1) {
status = list_resize(self, self->ob_size - 1);
assert(status >= 0);
return v; /* and v now owns the reference the list had */
}
Py_INCREF(v);
status = list_ass_slice(self, i, i+1, (PyObject *)NULL);
assert(status >= 0);
/* Use status, so that in a release build compilers don't
* complain about the unused name.
*/
(void) status;
return v;
}
/* Reverse a slice of a list in place, from lo up to (exclusive) hi. */
static void
reverse_slice(PyObject **lo, PyObject **hi)
{
assert(lo && hi);
--hi;
while (lo < hi) {
PyObject *t = *lo;
*lo = *hi;
*hi = t;
++lo;
--hi;
}
}
/* Lots of code for an adaptive, stable, natural mergesort. There are many
* pieces to this algorithm; read listsort.txt for overviews and details.
*/
/* Comparison function. Takes care of calling a user-supplied
* comparison function (any callable Python object), which must not be
* NULL (use the ISLT macro if you don't know, or call PyObject_RichCompareBool
* with Py_LT if you know it's NULL).
* Returns -1 on error, 1 if x < y, 0 if x >= y.
*/
static int
islt(PyObject *x, PyObject *y, PyObject *compare)
{
PyObject *res;
PyObject *args;
2006-02-15 13:27:45 -04:00
Py_ssize_t i;
assert(compare != NULL);
/* Call the user's comparison function and translate the 3-way
* result into true or false (or error).
*/
args = PyTuple_New(2);
if (args == NULL)
return -1;
Py_INCREF(x);
Py_INCREF(y);
PyTuple_SET_ITEM(args, 0, x);
PyTuple_SET_ITEM(args, 1, y);
res = PyObject_Call(compare, args, NULL);
1997-05-02 00:12:38 -03:00
Py_DECREF(args);
if (res == NULL)
return -1;
1997-05-02 00:12:38 -03:00
if (!PyInt_Check(res)) {
Py_DECREF(res);
PyErr_SetString(PyExc_TypeError,
"comparison function must return int");
return -1;
}
1997-05-02 00:12:38 -03:00
i = PyInt_AsLong(res);
Py_DECREF(res);
return i < 0;
}
/* If COMPARE is NULL, calls PyObject_RichCompareBool with Py_LT, else calls
* islt. This avoids a layer of function call in the usual case, and
* sorting does many comparisons.
* Returns -1 on error, 1 if x < y, 0 if x >= y.
*/
#define ISLT(X, Y, COMPARE) ((COMPARE) == NULL ? \
PyObject_RichCompareBool(X, Y, Py_LT) : \
islt(X, Y, COMPARE))
/* Compare X to Y via "<". Goto "fail" if the comparison raises an
error. Else "k" is set to true iff X<Y, and an "if (k)" block is
started. It makes more sense in context <wink>. X and Y are PyObject*s.
*/
#define IFLT(X, Y) if ((k = ISLT(X, Y, compare)) < 0) goto fail; \
if (k)
/* binarysort is the best method for sorting small arrays: it does
few compares, but can do data movement quadratic in the number of
elements.
[lo, hi) is a contiguous slice of a list, and is sorted via
binary insertion. This sort is stable.
On entry, must have lo <= start <= hi, and that [lo, start) is already
sorted (pass start == lo if you don't know!).
If islt() complains return -1, else 0.
Even in case of error, the output slice will be some permutation of
the input (nothing is lost or duplicated).
*/
static int
2000-07-09 12:16:51 -03:00
binarysort(PyObject **lo, PyObject **hi, PyObject **start, PyObject *compare)
/* compare -- comparison function object, or NULL for default */
{
2006-02-15 13:27:45 -04:00
register Py_ssize_t k;
register PyObject **l, **p, **r;
register PyObject *pivot;
assert(lo <= start && start <= hi);
/* assert [lo, start) is sorted */
if (lo == start)
++start;
for (; start < hi; ++start) {
/* set l to where *start belongs */
l = lo;
r = start;
pivot = *r;
/* Invariants:
* pivot >= all in [lo, l).
* pivot < all in [r, start).
* The second is vacuously true at the start.
*/
assert(l < r);
do {
p = l + ((r - l) >> 1);
IFLT(pivot, *p)
r = p;
else
l = p+1;
} while (l < r);
assert(l == r);
/* The invariants still hold, so pivot >= all in [lo, l) and
pivot < all in [l, start), so pivot belongs at l. Note
that if there are elements equal to pivot, l points to the
first slot after them -- that's why this sort is stable.
Slide over to make room.
Caution: using memmove is much slower under MSVC 5;
we're not usually moving many slots. */
for (p = start; p > l; --p)
*p = *(p-1);
*l = pivot;
}
return 0;
fail:
return -1;
}
/*
Return the length of the run beginning at lo, in the slice [lo, hi). lo < hi
is required on entry. "A run" is the longest ascending sequence, with
lo[0] <= lo[1] <= lo[2] <= ...
or the longest descending sequence, with
lo[0] > lo[1] > lo[2] > ...
Boolean *descending is set to 0 in the former case, or to 1 in the latter.
For its intended use in a stable mergesort, the strictness of the defn of
"descending" is needed so that the caller can safely reverse a descending
sequence without violating stability (strict > ensures there are no equal
elements to get out of order).
Returns -1 in case of error.
*/
2006-02-15 13:27:45 -04:00
static Py_ssize_t
count_run(PyObject **lo, PyObject **hi, PyObject *compare, int *descending)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t k;
Py_ssize_t n;
assert(lo < hi);
*descending = 0;
++lo;
if (lo == hi)
return 1;
n = 2;
IFLT(*lo, *(lo-1)) {
*descending = 1;
for (lo = lo+1; lo < hi; ++lo, ++n) {
IFLT(*lo, *(lo-1))
;
else
break;
}
}
else {
for (lo = lo+1; lo < hi; ++lo, ++n) {
IFLT(*lo, *(lo-1))
break;
}
}
return n;
fail:
return -1;
}
/*
Locate the proper position of key in a sorted vector; if the vector contains
an element equal to key, return the position immediately to the left of
the leftmost equal element. [gallop_right() does the same except returns
the position to the right of the rightmost equal element (if any).]
"a" is a sorted vector with n elements, starting at a[0]. n must be > 0.
1998-05-26 12:06:32 -03:00
"hint" is an index at which to begin the search, 0 <= hint < n. The closer
hint is to the final result, the faster this runs.
The return value is the int k in 0..n such that
a[k-1] < key <= a[k]
pretending that *(a-1) is minus infinity and a[n] is plus infinity. IOW,
key belongs at index k; or, IOW, the first k elements of a should precede
key, and the last n-k should follow key.
Returns -1 on error. See listsort.txt for info on the method.
*/
2006-02-15 13:27:45 -04:00
static Py_ssize_t
gallop_left(PyObject *key, PyObject **a, Py_ssize_t n, Py_ssize_t hint, PyObject *compare)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t ofs;
Py_ssize_t lastofs;
Py_ssize_t k;
assert(key && a && n > 0 && hint >= 0 && hint < n);
a += hint;
lastofs = 0;
ofs = 1;
IFLT(*a, key) {
/* a[hint] < key -- gallop right, until
* a[hint + lastofs] < key <= a[hint + ofs]
*/
2006-02-15 13:27:45 -04:00
const Py_ssize_t maxofs = n - hint; /* &a[n-1] is highest */
while (ofs < maxofs) {
IFLT(a[ofs], key) {
lastofs = ofs;
ofs = (ofs << 1) + 1;
if (ofs <= 0) /* int overflow */
ofs = maxofs;
1998-05-26 12:06:32 -03:00
}
else /* key <= a[hint + ofs] */
break;
1998-05-26 12:06:32 -03:00
}
if (ofs > maxofs)
ofs = maxofs;
/* Translate back to offsets relative to &a[0]. */
lastofs += hint;
ofs += hint;
}
else {
/* key <= a[hint] -- gallop left, until
* a[hint - ofs] < key <= a[hint - lastofs]
*/
2006-02-15 13:27:45 -04:00
const Py_ssize_t maxofs = hint + 1; /* &a[0] is lowest */
while (ofs < maxofs) {
IFLT(*(a-ofs), key)
break;
/* key <= a[hint - ofs] */
lastofs = ofs;
ofs = (ofs << 1) + 1;
if (ofs <= 0) /* int overflow */
ofs = maxofs;
}
if (ofs > maxofs)
ofs = maxofs;
/* Translate back to positive offsets relative to &a[0]. */
k = lastofs;
lastofs = hint - ofs;
ofs = hint - k;
}
a -= hint;
assert(-1 <= lastofs && lastofs < ofs && ofs <= n);
/* Now a[lastofs] < key <= a[ofs], so key belongs somewhere to the
* right of lastofs but no farther right than ofs. Do a binary
* search, with invariant a[lastofs-1] < key <= a[ofs].
*/
++lastofs;
while (lastofs < ofs) {
2006-02-15 13:27:45 -04:00
Py_ssize_t m = lastofs + ((ofs - lastofs) >> 1);
IFLT(a[m], key)
lastofs = m+1; /* a[m] < key */
else
ofs = m; /* key <= a[m] */
}
assert(lastofs == ofs); /* so a[ofs-1] < key <= a[ofs] */
return ofs;
fail:
return -1;
}
/*
Exactly like gallop_left(), except that if key already exists in a[0:n],
finds the position immediately to the right of the rightmost equal value.
The return value is the int k in 0..n such that
a[k-1] <= key < a[k]
or -1 if error.
The code duplication is massive, but this is enough different given that
we're sticking to "<" comparisons that it's much harder to follow if
written as one routine with yet another "left or right?" flag.
*/
2006-02-15 13:27:45 -04:00
static Py_ssize_t
gallop_right(PyObject *key, PyObject **a, Py_ssize_t n, Py_ssize_t hint, PyObject *compare)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t ofs;
Py_ssize_t lastofs;
Py_ssize_t k;
assert(key && a && n > 0 && hint >= 0 && hint < n);
a += hint;
lastofs = 0;
ofs = 1;
IFLT(key, *a) {
/* key < a[hint] -- gallop left, until
* a[hint - ofs] <= key < a[hint - lastofs]
*/
2006-02-15 13:27:45 -04:00
const Py_ssize_t maxofs = hint + 1; /* &a[0] is lowest */
while (ofs < maxofs) {
IFLT(key, *(a-ofs)) {
lastofs = ofs;
ofs = (ofs << 1) + 1;
if (ofs <= 0) /* int overflow */
ofs = maxofs;
}
else /* a[hint - ofs] <= key */
break;
}
if (ofs > maxofs)
ofs = maxofs;
/* Translate back to positive offsets relative to &a[0]. */
k = lastofs;
lastofs = hint - ofs;
ofs = hint - k;
}
else {
/* a[hint] <= key -- gallop right, until
* a[hint + lastofs] <= key < a[hint + ofs]
*/
2006-02-15 13:27:45 -04:00
const Py_ssize_t maxofs = n - hint; /* &a[n-1] is highest */
while (ofs < maxofs) {
IFLT(key, a[ofs])
break;
/* a[hint + ofs] <= key */
lastofs = ofs;
ofs = (ofs << 1) + 1;
if (ofs <= 0) /* int overflow */
ofs = maxofs;
}
if (ofs > maxofs)
ofs = maxofs;
/* Translate back to offsets relative to &a[0]. */
lastofs += hint;
ofs += hint;
}
a -= hint;
assert(-1 <= lastofs && lastofs < ofs && ofs <= n);
/* Now a[lastofs] <= key < a[ofs], so key belongs somewhere to the
* right of lastofs but no farther right than ofs. Do a binary
* search, with invariant a[lastofs-1] <= key < a[ofs].
*/
++lastofs;
while (lastofs < ofs) {
2006-02-15 13:27:45 -04:00
Py_ssize_t m = lastofs + ((ofs - lastofs) >> 1);
IFLT(key, a[m])
ofs = m; /* key < a[m] */
else
lastofs = m+1; /* a[m] <= key */
}
assert(lastofs == ofs); /* so a[ofs-1] <= key < a[ofs] */
return ofs;
fail:
return -1;
}
/* The maximum number of entries in a MergeState's pending-runs stack.
* This is enough to sort arrays of size up to about
* 32 * phi ** MAX_MERGE_PENDING
* where phi ~= 1.618. 85 is ridiculouslylarge enough, good for an array
* with 2**64 elements.
*/
#define MAX_MERGE_PENDING 85
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
/* When we get into galloping mode, we stay there until both runs win less
* often than MIN_GALLOP consecutive times. See listsort.txt for more info.
*/
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
#define MIN_GALLOP 7
/* Avoid malloc for small temp arrays. */
#define MERGESTATE_TEMP_SIZE 256
/* One MergeState exists on the stack per invocation of mergesort. It's just
* a convenient way to pass state around among the helper functions.
*/
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
struct s_slice {
PyObject **base;
2006-02-15 13:27:45 -04:00
Py_ssize_t len;
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
};
typedef struct s_MergeState {
/* The user-supplied comparison function. or NULL if none given. */
PyObject *compare;
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
/* This controls when we get *into* galloping mode. It's initialized
* to MIN_GALLOP. merge_lo and merge_hi tend to nudge it higher for
* random data, and lower for highly structured data.
*/
2006-02-15 13:27:45 -04:00
Py_ssize_t min_gallop;
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
/* 'a' is temp storage to help with merges. It contains room for
* alloced entries.
*/
PyObject **a; /* may point to temparray below */
2006-02-15 13:27:45 -04:00
Py_ssize_t alloced;
/* A stack of n pending runs yet to be merged. Run #i starts at
* address base[i] and extends for len[i] elements. It's always
* true (so long as the indices are in bounds) that
*
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
* pending[i].base + pending[i].len == pending[i+1].base
*
* so we could cut the storage for this, but it's a minor amount,
* and keeping all the info explicit simplifies the code.
*/
int n;
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
struct s_slice pending[MAX_MERGE_PENDING];
/* 'a' points to this when possible, rather than muck with malloc. */
PyObject *temparray[MERGESTATE_TEMP_SIZE];
} MergeState;
/* Conceptually a MergeState's constructor. */
static void
merge_init(MergeState *ms, PyObject *compare)
{
assert(ms != NULL);
ms->compare = compare;
ms->a = ms->temparray;
ms->alloced = MERGESTATE_TEMP_SIZE;
ms->n = 0;
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
ms->min_gallop = MIN_GALLOP;
}
/* Free all the temp memory owned by the MergeState. This must be called
* when you're done with a MergeState, and may be called before then if
* you want to free the temp memory early.
*/
static void
merge_freemem(MergeState *ms)
{
assert(ms != NULL);
if (ms->a != ms->temparray)
PyMem_Free(ms->a);
ms->a = ms->temparray;
ms->alloced = MERGESTATE_TEMP_SIZE;
}
/* Ensure enough temp memory for 'need' array slots is available.
* Returns 0 on success and -1 if the memory can't be gotten.
*/
static int
2006-02-15 13:27:45 -04:00
merge_getmem(MergeState *ms, Py_ssize_t need)
{
assert(ms != NULL);
if (need <= ms->alloced)
return 0;
/* Don't realloc! That can cost cycles to copy the old data, but
* we don't care what's in the block.
*/
merge_freemem(ms);
ms->a = (PyObject **)PyMem_Malloc(need * sizeof(PyObject*));
if (ms->a) {
ms->alloced = need;
return 0;
}
PyErr_NoMemory();
merge_freemem(ms); /* reset to sane state */
return -1;
}
#define MERGE_GETMEM(MS, NEED) ((NEED) <= (MS)->alloced ? 0 : \
merge_getmem(MS, NEED))
/* Merge the na elements starting at pa with the nb elements starting at pb
* in a stable way, in-place. na and nb must be > 0, and pa + na == pb.
* Must also have that *pb < *pa, that pa[na-1] belongs at the end of the
* merge, and should have na <= nb. See listsort.txt for more info.
* Return 0 if successful, -1 if error.
*/
2006-02-15 13:27:45 -04:00
static Py_ssize_t
merge_lo(MergeState *ms, PyObject **pa, Py_ssize_t na,
PyObject **pb, Py_ssize_t nb)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t k;
PyObject *compare;
PyObject **dest;
int result = -1; /* guilty until proved innocent */
2006-02-15 13:27:45 -04:00
Py_ssize_t min_gallop = ms->min_gallop;
assert(ms && pa && pb && na > 0 && nb > 0 && pa + na == pb);
if (MERGE_GETMEM(ms, na) < 0)
return -1;
memcpy(ms->a, pa, na * sizeof(PyObject*));
dest = pa;
pa = ms->a;
*dest++ = *pb++;
--nb;
if (nb == 0)
goto Succeed;
if (na == 1)
goto CopyB;
compare = ms->compare;
for (;;) {
2006-02-15 13:27:45 -04:00
Py_ssize_t acount = 0; /* # of times A won in a row */
Py_ssize_t bcount = 0; /* # of times B won in a row */
/* Do the straightforward thing until (if ever) one run
* appears to win consistently.
*/
for (;;) {
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
assert(na > 1 && nb > 0);
k = ISLT(*pb, *pa, compare);
if (k) {
if (k < 0)
goto Fail;
*dest++ = *pb++;
++bcount;
acount = 0;
--nb;
if (nb == 0)
goto Succeed;
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
if (bcount >= min_gallop)
break;
}
else {
*dest++ = *pa++;
++acount;
bcount = 0;
--na;
if (na == 1)
goto CopyB;
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
if (acount >= min_gallop)
break;
}
}
/* One run is winning so consistently that galloping may
* be a huge win. So try that, and continue galloping until
* (if ever) neither run appears to be winning consistently
* anymore.
*/
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
++min_gallop;
do {
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
assert(na > 1 && nb > 0);
min_gallop -= min_gallop > 1;
ms->min_gallop = min_gallop;
k = gallop_right(*pb, pa, na, 0, compare);
acount = k;
if (k) {
if (k < 0)
goto Fail;
memcpy(dest, pa, k * sizeof(PyObject *));
dest += k;
pa += k;
na -= k;
if (na == 1)
goto CopyB;
/* na==0 is impossible now if the comparison
* function is consistent, but we can't assume
* that it is.
*/
if (na == 0)
goto Succeed;
}
*dest++ = *pb++;
--nb;
if (nb == 0)
goto Succeed;
k = gallop_left(*pa, pb, nb, 0, compare);
bcount = k;
if (k) {
if (k < 0)
goto Fail;
memmove(dest, pb, k * sizeof(PyObject *));
dest += k;
pb += k;
nb -= k;
if (nb == 0)
goto Succeed;
}
*dest++ = *pa++;
--na;
if (na == 1)
goto CopyB;
} while (acount >= MIN_GALLOP || bcount >= MIN_GALLOP);
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
++min_gallop; /* penalize it for leaving galloping mode */
ms->min_gallop = min_gallop;
}
Succeed:
result = 0;
Fail:
if (na)
memcpy(dest, pa, na * sizeof(PyObject*));
return result;
CopyB:
assert(na == 1 && nb > 0);
/* The last element of pa belongs at the end of the merge. */
memmove(dest, pb, nb * sizeof(PyObject *));
dest[nb] = *pa;
return 0;
}
/* Merge the na elements starting at pa with the nb elements starting at pb
* in a stable way, in-place. na and nb must be > 0, and pa + na == pb.
* Must also have that *pb < *pa, that pa[na-1] belongs at the end of the
* merge, and should have na >= nb. See listsort.txt for more info.
* Return 0 if successful, -1 if error.
*/
2006-02-15 13:27:45 -04:00
static Py_ssize_t
merge_hi(MergeState *ms, PyObject **pa, Py_ssize_t na, PyObject **pb, Py_ssize_t nb)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t k;
PyObject *compare;
PyObject **dest;
int result = -1; /* guilty until proved innocent */
PyObject **basea;
PyObject **baseb;
2006-02-15 13:27:45 -04:00
Py_ssize_t min_gallop = ms->min_gallop;
assert(ms && pa && pb && na > 0 && nb > 0 && pa + na == pb);
if (MERGE_GETMEM(ms, nb) < 0)
return -1;
dest = pb + nb - 1;
memcpy(ms->a, pb, nb * sizeof(PyObject*));
basea = pa;
baseb = ms->a;
pb = ms->a + nb - 1;
pa += na - 1;
*dest-- = *pa--;
--na;
if (na == 0)
goto Succeed;
if (nb == 1)
goto CopyA;
compare = ms->compare;
for (;;) {
2006-02-15 13:27:45 -04:00
Py_ssize_t acount = 0; /* # of times A won in a row */
Py_ssize_t bcount = 0; /* # of times B won in a row */
/* Do the straightforward thing until (if ever) one run
* appears to win consistently.
*/
for (;;) {
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
assert(na > 0 && nb > 1);
k = ISLT(*pb, *pa, compare);
if (k) {
if (k < 0)
goto Fail;
*dest-- = *pa--;
++acount;
bcount = 0;
--na;
if (na == 0)
goto Succeed;
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
if (acount >= min_gallop)
break;
}
else {
*dest-- = *pb--;
++bcount;
acount = 0;
--nb;
if (nb == 1)
goto CopyA;
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
if (bcount >= min_gallop)
break;
}
}
/* One run is winning so consistently that galloping may
* be a huge win. So try that, and continue galloping until
* (if ever) neither run appears to be winning consistently
* anymore.
*/
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
++min_gallop;
do {
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
assert(na > 0 && nb > 1);
min_gallop -= min_gallop > 1;
ms->min_gallop = min_gallop;
k = gallop_right(*pb, basea, na, na-1, compare);
if (k < 0)
goto Fail;
k = na - k;
acount = k;
if (k) {
dest -= k;
pa -= k;
memmove(dest+1, pa+1, k * sizeof(PyObject *));
na -= k;
if (na == 0)
goto Succeed;
}
*dest-- = *pb--;
--nb;
if (nb == 1)
goto CopyA;
k = gallop_left(*pa, baseb, nb, nb-1, compare);
if (k < 0)
goto Fail;
k = nb - k;
bcount = k;
if (k) {
dest -= k;
pb -= k;
memcpy(dest+1, pb+1, k * sizeof(PyObject *));
nb -= k;
if (nb == 1)
goto CopyA;
/* nb==0 is impossible now if the comparison
* function is consistent, but we can't assume
* that it is.
*/
if (nb == 0)
goto Succeed;
}
*dest-- = *pa--;
--na;
if (na == 0)
goto Succeed;
} while (acount >= MIN_GALLOP || bcount >= MIN_GALLOP);
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
++min_gallop; /* penalize it for leaving galloping mode */
ms->min_gallop = min_gallop;
}
Succeed:
result = 0;
Fail:
if (nb)
memcpy(dest-(nb-1), baseb, nb * sizeof(PyObject*));
return result;
CopyA:
assert(nb == 1 && na > 0);
/* The first element of pb belongs at the front of the merge. */
dest -= na;
pa -= na;
memmove(dest+1, pa+1, na * sizeof(PyObject *));
*dest = *pb;
return 0;
}
1998-05-13 18:20:49 -03:00
/* Merge the two runs at stack indices i and i+1.
* Returns 0 on success, -1 on error.
*/
2006-02-15 13:27:45 -04:00
static Py_ssize_t
merge_at(MergeState *ms, Py_ssize_t i)
{
PyObject **pa, **pb;
2006-02-15 13:27:45 -04:00
Py_ssize_t na, nb;
Py_ssize_t k;
PyObject *compare;
assert(ms != NULL);
assert(ms->n >= 2);
assert(i >= 0);
assert(i == ms->n - 2 || i == ms->n - 3);
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
pa = ms->pending[i].base;
na = ms->pending[i].len;
pb = ms->pending[i+1].base;
nb = ms->pending[i+1].len;
assert(na > 0 && nb > 0);
assert(pa + na == pb);
/* Record the length of the combined runs; if i is the 3rd-last
* run now, also slide over the last run (which isn't involved
* in this merge). The current run i+1 goes away in any case.
*/
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
ms->pending[i].len = na + nb;
if (i == ms->n - 3)
ms->pending[i+1] = ms->pending[i+2];
--ms->n;
1998-05-13 18:20:49 -03:00
/* Where does b start in a? Elements in a before that can be
* ignored (already in place).
*/
compare = ms->compare;
k = gallop_right(*pb, pa, na, 0, compare);
if (k < 0)
return -1;
pa += k;
na -= k;
if (na == 0)
return 0;
1998-05-26 12:06:32 -03:00
/* Where does a end in b? Elements in b after that can be
* ignored (already in place).
*/
nb = gallop_left(pa[na-1], pb, nb, nb-1, compare);
if (nb <= 0)
return nb;
/* Merge what remains of the runs, using a temp array with
* min(na, nb) elements.
*/
if (na <= nb)
return merge_lo(ms, pa, na, pb, nb);
else
return merge_hi(ms, pa, na, pb, nb);
}
/* Examine the stack of runs waiting to be merged, merging adjacent runs
* until the stack invariants are re-established:
*
* 1. len[-3] > len[-2] + len[-1]
* 2. len[-2] > len[-1]
*
* See listsort.txt for more info.
*
* Returns 0 on success, -1 on error.
*/
static int
merge_collapse(MergeState *ms)
{
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
struct s_slice *p = ms->pending;
assert(ms);
while (ms->n > 1) {
2006-02-15 13:27:45 -04:00
Py_ssize_t n = ms->n - 2;
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
if (n > 0 && p[n-1].len <= p[n].len + p[n+1].len) {
if (p[n-1].len < p[n+1].len)
--n;
if (merge_at(ms, n) < 0)
return -1;
}
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
else if (p[n].len <= p[n+1].len) {
if (merge_at(ms, n) < 0)
return -1;
}
else
break;
}
return 0;
}
/* Regardless of invariants, merge all runs on the stack until only one
* remains. This is used at the end of the mergesort.
*
* Returns 0 on success, -1 on error.
*/
static int
merge_force_collapse(MergeState *ms)
{
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
struct s_slice *p = ms->pending;
assert(ms);
while (ms->n > 1) {
2006-02-15 13:27:45 -04:00
Py_ssize_t n = ms->n - 2;
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
if (n > 0 && p[n-1].len < p[n+1].len)
--n;
if (merge_at(ms, n) < 0)
return -1;
}
return 0;
}
/* Compute a good value for the minimum run length; natural runs shorter
* than this are boosted artificially via binary insertion.
*
* If n < 64, return n (it's too small to bother with fancy stuff).
* Else if n is an exact power of 2, return 32.
* Else return an int k, 32 <= k <= 64, such that n/k is close to, but
* strictly less than, an exact power of 2.
*
* See listsort.txt for more info.
*/
2006-02-15 13:27:45 -04:00
static Py_ssize_t
merge_compute_minrun(Py_ssize_t n)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t r = 0; /* becomes 1 if any 1 bits are shifted off */
assert(n >= 0);
while (n >= 64) {
r |= n & 1;
n >>= 1;
}
return n + r;
}
/* Special wrapper to support stable sorting using the decorate-sort-undecorate
2004-09-10 09:59:54 -03:00
pattern. Holds a key which is used for comparisons and the original record
2004-07-28 23:29:26 -03:00
which is returned during the undecorate phase. By exposing only the key
during comparisons, the underlying sort stability characteristics are left
unchanged. Also, if a custom comparison function is used, it will only see
the key instead of a full record. */
typedef struct {
PyObject_HEAD
PyObject *key;
PyObject *value;
} sortwrapperobject;
static PyTypeObject sortwrapper_type;
static PyObject *
sortwrapper_richcompare(sortwrapperobject *a, sortwrapperobject *b, int op)
{
if (!PyObject_TypeCheck(b, &sortwrapper_type)) {
2004-07-28 23:29:26 -03:00
PyErr_SetString(PyExc_TypeError,
"expected a sortwrapperobject");
return NULL;
}
return PyObject_RichCompare(a->key, b->key, op);
}
static void
sortwrapper_dealloc(sortwrapperobject *so)
{
Py_XDECREF(so->key);
Py_XDECREF(so->value);
PyObject_Del(so);
}
PyDoc_STRVAR(sortwrapper_doc, "Object wrapper with a custom sort key.");
static PyTypeObject sortwrapper_type = {
PyObject_HEAD_INIT(&PyType_Type)
0, /* ob_size */
"sortwrapper", /* tp_name */
sizeof(sortwrapperobject), /* tp_basicsize */
0, /* tp_itemsize */
/* methods */
(destructor)sortwrapper_dealloc, /* tp_dealloc */
0, /* tp_print */
0, /* tp_getattr */
0, /* tp_setattr */
0, /* tp_compare */
0, /* tp_repr */
0, /* tp_as_number */
0, /* tp_as_sequence */
0, /* tp_as_mapping */
0, /* tp_hash */
0, /* tp_call */
0, /* tp_str */
PyObject_GenericGetAttr, /* tp_getattro */
0, /* tp_setattro */
0, /* tp_as_buffer */
2004-07-28 23:29:26 -03:00
Py_TPFLAGS_DEFAULT |
Py_TPFLAGS_HAVE_RICHCOMPARE, /* tp_flags */
sortwrapper_doc, /* tp_doc */
0, /* tp_traverse */
0, /* tp_clear */
(richcmpfunc)sortwrapper_richcompare, /* tp_richcompare */
};
/* Returns a new reference to a sortwrapper.
Consumes the references to the two underlying objects. */
static PyObject *
build_sortwrapper(PyObject *key, PyObject *value)
{
sortwrapperobject *so;
2004-07-28 23:29:26 -03:00
so = PyObject_New(sortwrapperobject, &sortwrapper_type);
if (so == NULL)
return NULL;
so->key = key;
so->value = value;
return (PyObject *)so;
}
/* Returns a new reference to the value underlying the wrapper. */
static PyObject *
sortwrapper_getvalue(PyObject *so)
{
PyObject *value;
if (!PyObject_TypeCheck(so, &sortwrapper_type)) {
2004-07-28 23:29:26 -03:00
PyErr_SetString(PyExc_TypeError,
"expected a sortwrapperobject");
return NULL;
}
value = ((sortwrapperobject *)so)->value;
Py_INCREF(value);
return value;
}
/* Wrapper for user specified cmp functions in combination with a
specified key function. Makes sure the cmp function is presented
with the actual key instead of the sortwrapper */
typedef struct {
PyObject_HEAD
PyObject *func;
} cmpwrapperobject;
static void
cmpwrapper_dealloc(cmpwrapperobject *co)
{
Py_XDECREF(co->func);
PyObject_Del(co);
}
static PyObject *
cmpwrapper_call(cmpwrapperobject *co, PyObject *args, PyObject *kwds)
{
PyObject *x, *y, *xx, *yy;
if (!PyArg_UnpackTuple(args, "", 2, 2, &x, &y))
return NULL;
if (!PyObject_TypeCheck(x, &sortwrapper_type) ||
2003-10-16 14:16:30 -03:00
!PyObject_TypeCheck(y, &sortwrapper_type)) {
2004-07-28 23:29:26 -03:00
PyErr_SetString(PyExc_TypeError,
"expected a sortwrapperobject");
return NULL;
}
xx = ((sortwrapperobject *)x)->key;
yy = ((sortwrapperobject *)y)->key;
return PyObject_CallFunctionObjArgs(co->func, xx, yy, NULL);
}
PyDoc_STRVAR(cmpwrapper_doc, "cmp() wrapper for sort with custom keys.");
static PyTypeObject cmpwrapper_type = {
PyObject_HEAD_INIT(&PyType_Type)
0, /* ob_size */
"cmpwrapper", /* tp_name */
sizeof(cmpwrapperobject), /* tp_basicsize */
0, /* tp_itemsize */
/* methods */
(destructor)cmpwrapper_dealloc, /* tp_dealloc */
0, /* tp_print */
0, /* tp_getattr */
0, /* tp_setattr */
0, /* tp_compare */
0, /* tp_repr */
0, /* tp_as_number */
0, /* tp_as_sequence */
0, /* tp_as_mapping */
0, /* tp_hash */
(ternaryfunc)cmpwrapper_call, /* tp_call */
0, /* tp_str */
PyObject_GenericGetAttr, /* tp_getattro */
0, /* tp_setattro */
0, /* tp_as_buffer */
Py_TPFLAGS_DEFAULT, /* tp_flags */
cmpwrapper_doc, /* tp_doc */
};
static PyObject *
build_cmpwrapper(PyObject *cmpfunc)
{
cmpwrapperobject *co;
2004-07-28 23:29:26 -03:00
co = PyObject_New(cmpwrapperobject, &cmpwrapper_type);
if (co == NULL)
return NULL;
Py_INCREF(cmpfunc);
co->func = cmpfunc;
return (PyObject *)co;
}
/* An adaptive, stable, natural mergesort. See listsort.txt.
* Returns Py_None on success, NULL on error. Even in case of error, the
* list will be some permutation of its input state (nothing is lost or
* duplicated).
*/
1997-05-02 00:12:38 -03:00
static PyObject *
listsort(PyListObject *self, PyObject *args, PyObject *kwds)
{
MergeState ms;
PyObject **lo, **hi;
2006-02-15 13:27:45 -04:00
Py_ssize_t nremaining;
Py_ssize_t minrun;
Py_ssize_t saved_ob_size, saved_allocated;
PyObject **saved_ob_item;
PyObject **final_ob_item;
PyObject *compare = NULL;
PyObject *result = NULL; /* guilty until proved innocent */
int reverse = 0;
PyObject *keyfunc = NULL;
2006-02-15 13:27:45 -04:00
Py_ssize_t i;
PyObject *key, *value, *kvpair;
static char *kwlist[] = {"cmp", "key", "reverse", 0};
assert(self != NULL);
assert (PyList_Check(self));
if (args != NULL) {
if (!PyArg_ParseTupleAndKeywords(args, kwds, "|OOi:sort",
kwlist, &compare, &keyfunc, &reverse))
return NULL;
}
if (compare == Py_None)
compare = NULL;
if (keyfunc == Py_None)
keyfunc = NULL;
if (compare != NULL && keyfunc != NULL) {
compare = build_cmpwrapper(compare);
if (compare == NULL)
return NULL;
} else
Py_XINCREF(compare);
/* The list is temporarily made empty, so that mutations performed
* by comparison functions can't affect the slice of memory we're
* sorting (allowing mutations during sorting is a core-dump
* factory, since ob_item may change).
*/
saved_ob_size = self->ob_size;
saved_ob_item = self->ob_item;
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
saved_allocated = self->allocated;
self->ob_size = 0;
self->ob_item = NULL;
self->allocated = -1; /* any operation will reset it to >= 0 */
if (keyfunc != NULL) {
for (i=0 ; i < saved_ob_size ; i++) {
value = saved_ob_item[i];
2004-07-28 23:29:26 -03:00
key = PyObject_CallFunctionObjArgs(keyfunc, value,
NULL);
if (key == NULL) {
for (i=i-1 ; i>=0 ; i--) {
kvpair = saved_ob_item[i];
value = sortwrapper_getvalue(kvpair);
saved_ob_item[i] = value;
Py_DECREF(kvpair);
}
goto dsu_fail;
}
kvpair = build_sortwrapper(key, value);
if (kvpair == NULL)
goto dsu_fail;
saved_ob_item[i] = kvpair;
}
}
/* Reverse sort stability achieved by initially reversing the list,
applying a stable forward sort, then reversing the final result. */
if (reverse && saved_ob_size > 1)
reverse_slice(saved_ob_item, saved_ob_item + saved_ob_size);
merge_init(&ms, compare);
nremaining = saved_ob_size;
if (nremaining < 2)
goto succeed;
/* March over the array once, left to right, finding natural runs,
* and extending short natural runs to minrun elements.
*/
lo = saved_ob_item;
hi = lo + nremaining;
minrun = merge_compute_minrun(nremaining);
do {
int descending;
2006-02-15 13:27:45 -04:00
Py_ssize_t n;
/* Identify next run. */
n = count_run(lo, hi, compare, &descending);
if (n < 0)
goto fail;
if (descending)
reverse_slice(lo, lo + n);
/* If short, extend to min(minrun, nremaining). */
if (n < minrun) {
2006-02-15 13:27:45 -04:00
const Py_ssize_t force = nremaining <= minrun ?
nremaining : minrun;
if (binarysort(lo, lo + force, lo + n, compare) < 0)
goto fail;
n = force;
}
/* Push run onto pending-runs stack, and maybe merge. */
assert(ms.n < MAX_MERGE_PENDING);
1. Combined the base and length arrays into a single array of structs. This is friendlier for caches. 2. Cut MIN_GALLOP to 7, but added a per-sort min_gallop vrbl that adapts the "get into galloping mode" threshold higher when galloping isn't paying, and lower when it is. There's no known case where this hurts. It's (of course) neutral for /sort, \sort and =sort. It also happens to be neutral for !sort. It cuts a tiny # of compares in 3sort and +sort. For *sort, it reduces the # of compares to better than what this used to do when MIN_GALLOP was hardcoded to 10 (it did about 0.1% more *sort compares before, but given how close we are to the limit, this is "a lot"!). %sort used to do about 1.5% more compares, and ~sort about 3.6% more. Here are exact counts: i *sort 3sort +sort %sort ~sort !sort 15 449235 33019 33016 51328 188720 65534 before 448885 33016 33007 50426 182083 65534 after 0.08% 0.01% 0.03% 1.79% 3.65% 0.00% %ch from after 16 963714 65824 65809 103409 377634 131070 962991 65821 65808 101667 364341 131070 0.08% 0.00% 0.00% 1.71% 3.65% 0.00% 17 2059092 131413 131362 209130 755476 262142 2057533 131410 131361 206193 728871 262142 0.08% 0.00% 0.00% 1.42% 3.65% 0.00% 18 4380687 262440 262460 421998 1511174 524286 4377402 262437 262459 416347 1457945 524286 0.08% 0.00% 0.00% 1.36% 3.65% 0.00% 19 9285709 524581 524634 848590 3022584 1048574 9278734 524580 524633 837947 2916107 1048574 0.08% 0.00% 0.00% 1.27% 3.65% 0.00% 20 19621118 1048960 1048942 1715806 6045418 2097150 19606028 1048958 1048941 1694896 5832445 2097150 0.08% 0.00% 0.00% 1.23% 3.65% 0.00% 3. Added some key asserts I overlooked before. 4. Updated the doc file.
2002-08-10 02:21:15 -03:00
ms.pending[ms.n].base = lo;
ms.pending[ms.n].len = n;
++ms.n;
if (merge_collapse(&ms) < 0)
goto fail;
/* Advance to find next run. */
lo += n;
nremaining -= n;
} while (nremaining);
assert(lo == hi);
if (merge_force_collapse(&ms) < 0)
goto fail;
assert(ms.n == 1);
assert(ms.pending[0].base == saved_ob_item);
assert(ms.pending[0].len == saved_ob_size);
succeed:
result = Py_None;
fail:
if (keyfunc != NULL) {
for (i=0 ; i < saved_ob_size ; i++) {
kvpair = saved_ob_item[i];
value = sortwrapper_getvalue(kvpair);
saved_ob_item[i] = value;
Py_DECREF(kvpair);
}
}
if (self->allocated != -1 && result != NULL) {
/* The user mucked with the list during the sort,
* and we don't already have another error to report.
*/
PyErr_SetString(PyExc_ValueError, "list modified during sort");
result = NULL;
}
if (reverse && saved_ob_size > 1)
reverse_slice(saved_ob_item, saved_ob_item + saved_ob_size);
merge_freemem(&ms);
dsu_fail:
final_ob_item = self->ob_item;
i = self->ob_size;
self->ob_size = saved_ob_size;
self->ob_item = saved_ob_item;
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
self->allocated = saved_allocated;
if (final_ob_item != NULL) {
/* we cannot use list_clear() for this because it does not
guarantee that the list is really empty when it returns */
while (--i >= 0) {
Py_XDECREF(final_ob_item[i]);
}
PyMem_FREE(final_ob_item);
}
Py_XDECREF(compare);
Py_XINCREF(result);
return result;
}
#undef IFLT
#undef ISLT
int
2000-07-09 12:16:51 -03:00
PyList_Sort(PyObject *v)
{
if (v == NULL || !PyList_Check(v)) {
PyErr_BadInternalCall();
return -1;
}
v = listsort((PyListObject *)v, (PyObject *)NULL, (PyObject *)NULL);
if (v == NULL)
return -1;
Py_DECREF(v);
return 0;
}
static PyObject *
listreverse(PyListObject *self)
{
if (self->ob_size > 1)
reverse_slice(self->ob_item, self->ob_item + self->ob_size);
Py_RETURN_NONE;
}
int
2000-07-09 12:16:51 -03:00
PyList_Reverse(PyObject *v)
{
PyListObject *self = (PyListObject *)v;
1997-05-02 00:12:38 -03:00
if (v == NULL || !PyList_Check(v)) {
PyErr_BadInternalCall();
return -1;
}
if (self->ob_size > 1)
reverse_slice(self->ob_item, self->ob_item + self->ob_size);
return 0;
}
1997-05-02 00:12:38 -03:00
PyObject *
2000-07-09 12:16:51 -03:00
PyList_AsTuple(PyObject *v)
1994-08-29 09:45:32 -03:00
{
1997-05-02 00:12:38 -03:00
PyObject *w;
PyObject **p;
2006-02-15 13:27:45 -04:00
Py_ssize_t n;
1997-05-02 00:12:38 -03:00
if (v == NULL || !PyList_Check(v)) {
PyErr_BadInternalCall();
1994-08-29 09:45:32 -03:00
return NULL;
}
1997-05-02 00:12:38 -03:00
n = ((PyListObject *)v)->ob_size;
w = PyTuple_New(n);
1994-08-29 09:45:32 -03:00
if (w == NULL)
return NULL;
1997-05-02 00:12:38 -03:00
p = ((PyTupleObject *)w)->ob_item;
memcpy((void *)p,
(void *)((PyListObject *)v)->ob_item,
1997-05-02 00:12:38 -03:00
n*sizeof(PyObject *));
1994-08-29 09:45:32 -03:00
while (--n >= 0) {
1997-05-02 00:12:38 -03:00
Py_INCREF(*p);
1994-08-29 09:45:32 -03:00
p++;
}
return w;
}
1997-05-02 00:12:38 -03:00
static PyObject *
listindex(PyListObject *self, PyObject *args)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t i, start=0, stop=self->ob_size;
PyObject *v;
if (!PyArg_ParseTuple(args, "O|O&O&:index", &v,
_PyEval_SliceIndex, &start,
_PyEval_SliceIndex, &stop))
return NULL;
if (start < 0) {
start += self->ob_size;
if (start < 0)
start = 0;
}
if (stop < 0) {
stop += self->ob_size;
if (stop < 0)
stop = 0;
}
for (i = start; i < stop && i < self->ob_size; i++) {
int cmp = PyObject_RichCompareBool(self->ob_item[i], v, Py_EQ);
if (cmp > 0)
2006-02-15 13:27:45 -04:00
return PyInt_FromSsize_t(i);
else if (cmp < 0)
return NULL;
}
1997-05-02 00:12:38 -03:00
PyErr_SetString(PyExc_ValueError, "list.index(x): x not in list");
return NULL;
}
1997-05-02 00:12:38 -03:00
static PyObject *
listcount(PyListObject *self, PyObject *v)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t count = 0;
Py_ssize_t i;
for (i = 0; i < self->ob_size; i++) {
int cmp = PyObject_RichCompareBool(self->ob_item[i], v, Py_EQ);
if (cmp > 0)
count++;
else if (cmp < 0)
return NULL;
}
2006-02-15 13:27:45 -04:00
return PyInt_FromSsize_t(count);
}
1997-05-02 00:12:38 -03:00
static PyObject *
listremove(PyListObject *self, PyObject *v)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t i;
for (i = 0; i < self->ob_size; i++) {
int cmp = PyObject_RichCompareBool(self->ob_item[i], v, Py_EQ);
if (cmp > 0) {
1997-05-02 00:12:38 -03:00
if (list_ass_slice(self, i, i+1,
(PyObject *)NULL) == 0)
Py_RETURN_NONE;
return NULL;
}
else if (cmp < 0)
return NULL;
}
1997-05-02 00:12:38 -03:00
PyErr_SetString(PyExc_ValueError, "list.remove(x): x not in list");
return NULL;
}
static int
list_traverse(PyListObject *o, visitproc visit, void *arg)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t i;
PyObject *x;
for (i = o->ob_size; --i >= 0; ) {
x = o->ob_item[i];
if (x != NULL) {
2006-02-15 13:27:45 -04:00
int err = visit(x, arg);
if (err)
return err;
}
}
return 0;
}
static PyObject *
list_richcompare(PyObject *v, PyObject *w, int op)
{
PyListObject *vl, *wl;
2006-02-15 13:27:45 -04:00
Py_ssize_t i;
if (!PyList_Check(v) || !PyList_Check(w)) {
Py_INCREF(Py_NotImplemented);
return Py_NotImplemented;
}
vl = (PyListObject *)v;
wl = (PyListObject *)w;
if (vl->ob_size != wl->ob_size && (op == Py_EQ || op == Py_NE)) {
/* Shortcut: if the lengths differ, the lists differ */
PyObject *res;
if (op == Py_EQ)
res = Py_False;
else
res = Py_True;
Py_INCREF(res);
return res;
}
/* Search for the first index where items are different */
for (i = 0; i < vl->ob_size && i < wl->ob_size; i++) {
int k = PyObject_RichCompareBool(vl->ob_item[i],
wl->ob_item[i], Py_EQ);
if (k < 0)
return NULL;
if (!k)
break;
}
if (i >= vl->ob_size || i >= wl->ob_size) {
/* No more items to compare -- compare sizes */
2006-02-15 13:27:45 -04:00
Py_ssize_t vs = vl->ob_size;
Py_ssize_t ws = wl->ob_size;
int cmp;
PyObject *res;
switch (op) {
case Py_LT: cmp = vs < ws; break;
case Py_LE: cmp = vs <= ws; break;
case Py_EQ: cmp = vs == ws; break;
case Py_NE: cmp = vs != ws; break;
case Py_GT: cmp = vs > ws; break;
case Py_GE: cmp = vs >= ws; break;
default: return NULL; /* cannot happen */
}
if (cmp)
res = Py_True;
else
res = Py_False;
Py_INCREF(res);
return res;
}
/* We have an item that differs -- shortcuts for EQ/NE */
if (op == Py_EQ) {
Py_INCREF(Py_False);
return Py_False;
}
if (op == Py_NE) {
Py_INCREF(Py_True);
return Py_True;
}
/* Compare the final item again using the proper operator */
return PyObject_RichCompare(vl->ob_item[i], wl->ob_item[i], op);
}
2001-08-02 01:15:00 -03:00
static int
list_init(PyListObject *self, PyObject *args, PyObject *kw)
{
PyObject *arg = NULL;
static char *kwlist[] = {"sequence", 0};
2001-08-02 01:15:00 -03:00
if (!PyArg_ParseTupleAndKeywords(args, kw, "|O:list", kwlist, &arg))
return -1;
/* Verify list invariants established by PyType_GenericAlloc() */
assert(0 <= self->ob_size);
assert(self->ob_size <= self->allocated || self->allocated == -1);
assert(self->ob_item != NULL ||
self->allocated == 0 || self->allocated == -1);
/* Empty previous contents */
if (self->ob_item != NULL) {
(void)list_clear(self);
}
if (arg != NULL) {
PyObject *rv = listextend(self, arg);
if (rv == NULL)
return -1;
Py_DECREF(rv);
}
2001-08-02 01:15:00 -03:00
return 0;
}
static long
list_nohash(PyObject *self)
{
PyErr_SetString(PyExc_TypeError, "list objects are unhashable");
return -1;
}
static PyObject *list_iter(PyObject *seq);
static PyObject *list_reversed(PyListObject* seq, PyObject* unused);
PyDoc_STRVAR(getitem_doc,
"x.__getitem__(y) <==> x[y]");
PyDoc_STRVAR(reversed_doc,
"L.__reversed__() -- return a reverse iterator over the list");
2002-06-13 17:33:02 -03:00
PyDoc_STRVAR(append_doc,
"L.append(object) -- append object to end");
PyDoc_STRVAR(extend_doc,
"L.extend(iterable) -- extend list by appending elements from the iterable");
2002-06-13 17:33:02 -03:00
PyDoc_STRVAR(insert_doc,
"L.insert(index, object) -- insert object before index");
PyDoc_STRVAR(pop_doc,
"L.pop([index]) -> item -- remove and return item at index (default last)");
PyDoc_STRVAR(remove_doc,
"L.remove(value) -- remove first occurrence of value");
PyDoc_STRVAR(index_doc,
"L.index(value, [start, [stop]]) -> integer -- return first index of value");
2002-06-13 17:33:02 -03:00
PyDoc_STRVAR(count_doc,
"L.count(value) -> integer -- return number of occurrences of value");
PyDoc_STRVAR(reverse_doc,
"L.reverse() -- reverse *IN PLACE*");
PyDoc_STRVAR(sort_doc,
"L.sort(cmp=None, key=None, reverse=False) -- stable sort *IN PLACE*;\n\
cmp(x, y) -> -1, 0, 1");
static PyObject *list_subscript(PyListObject*, PyObject*);
1997-05-02 00:12:38 -03:00
static PyMethodDef list_methods[] = {
{"__getitem__", (PyCFunction)list_subscript, METH_O|METH_COEXIST, getitem_doc},
{"__reversed__",(PyCFunction)list_reversed, METH_NOARGS, reversed_doc},
{"append", (PyCFunction)listappend, METH_O, append_doc},
{"insert", (PyCFunction)listinsert, METH_VARARGS, insert_doc},
{"extend", (PyCFunction)listextend, METH_O, extend_doc},
{"pop", (PyCFunction)listpop, METH_VARARGS, pop_doc},
{"remove", (PyCFunction)listremove, METH_O, remove_doc},
{"index", (PyCFunction)listindex, METH_VARARGS, index_doc},
{"count", (PyCFunction)listcount, METH_O, count_doc},
{"reverse", (PyCFunction)listreverse, METH_NOARGS, reverse_doc},
{"sort", (PyCFunction)listsort, METH_VARARGS | METH_KEYWORDS, sort_doc},
{NULL, NULL} /* sentinel */
1990-10-14 09:07:46 -03:00
};
1997-05-02 00:12:38 -03:00
static PySequenceMethods list_as_sequence = {
2006-02-15 13:27:45 -04:00
(lenfunc)list_length, /* sq_length */
(binaryfunc)list_concat, /* sq_concat */
2006-02-15 13:27:45 -04:00
(ssizeargfunc)list_repeat, /* sq_repeat */
(ssizeargfunc)list_item, /* sq_item */
(ssizessizeargfunc)list_slice, /* sq_slice */
(ssizeobjargproc)list_ass_item, /* sq_ass_item */
(ssizessizeobjargproc)list_ass_slice, /* sq_ass_slice */
(objobjproc)list_contains, /* sq_contains */
(binaryfunc)list_inplace_concat, /* sq_inplace_concat */
2006-02-15 13:27:45 -04:00
(ssizeargfunc)list_inplace_repeat, /* sq_inplace_repeat */
1990-10-14 09:07:46 -03:00
};
PyDoc_STRVAR(list_doc,
2001-08-02 01:15:00 -03:00
"list() -> new list\n"
"list(sequence) -> new list initialized from sequence's items");
2001-08-02 01:15:00 -03:00
#define HASINDEX(o) PyType_HasFeature((o)->ob_type, Py_TPFLAGS_HAVE_INDEX)
static PyObject *
list_subscript(PyListObject* self, PyObject* item)
{
PyNumberMethods *nb = item->ob_type->tp_as_number;
if (nb != NULL && HASINDEX(item) && nb->nb_index != NULL) {
Py_ssize_t i = nb->nb_index(item);
if (i == -1 && PyErr_Occurred())
return NULL;
if (i < 0)
i += PyList_GET_SIZE(self);
return list_item(self, i);
}
else if (PySlice_Check(item)) {
2006-02-15 13:27:45 -04:00
Py_ssize_t start, stop, step, slicelength, cur, i;
PyObject* result;
PyObject* it;
PyObject **src, **dest;
if (PySlice_GetIndicesEx((PySliceObject*)item, self->ob_size,
&start, &stop, &step, &slicelength) < 0) {
return NULL;
}
if (slicelength <= 0) {
return PyList_New(0);
}
else {
result = PyList_New(slicelength);
if (!result) return NULL;
src = self->ob_item;
dest = ((PyListObject *)result)->ob_item;
2002-07-18 23:35:45 -03:00
for (cur = start, i = 0; i < slicelength;
cur += step, i++) {
it = src[cur];
Py_INCREF(it);
dest[i] = it;
}
2002-07-18 23:35:45 -03:00
return result;
}
}
else {
PyErr_SetString(PyExc_TypeError,
"list indices must be integers");
return NULL;
}
}
2002-07-18 23:35:45 -03:00
static int
list_ass_subscript(PyListObject* self, PyObject* item, PyObject* value)
{
PyNumberMethods *nb = item->ob_type->tp_as_number;
if (nb != NULL && HASINDEX(item) && nb->nb_index != NULL) {
Py_ssize_t i = nb->nb_index(item);
if (i == -1 && PyErr_Occurred())
return -1;
if (i < 0)
i += PyList_GET_SIZE(self);
return list_ass_item(self, i, value);
}
else if (PySlice_Check(item)) {
2006-02-15 13:27:45 -04:00
Py_ssize_t start, stop, step, slicelength;
if (PySlice_GetIndicesEx((PySliceObject*)item, self->ob_size,
&start, &stop, &step, &slicelength) < 0) {
return -1;
}
/* treat L[slice(a,b)] = v _exactly_ like L[a:b] = v */
if (step == 1 && ((PySliceObject*)item)->step == Py_None)
return list_ass_slice(self, start, stop, value);
if (value == NULL) {
/* delete slice */
* Optimized list appends and pops by making fewer calls the underlying system realloc(). This is achieved by tracking the overallocation size in a new field and using that information to skip calls to realloc() whenever possible. * Simplified and tightened the amount of overallocation. For larger lists, this overallocates by 1/8th (compared to the previous scheme which ranged between 1/4th to 1/32nd over-allocation). For smaller lists (n<6), the maximum overallocation is one byte (formerly it could be upto eight bytes). This saves memory in applications with large numbers of small lists. * Eliminated the NRESIZE macro in favor of a new, static list_resize function that encapsulates the resizing logic. Coverting this back to macro would give a small (under 1%) speed-up. This was too small to warrant the loss of readability, maintainability, and de-coupling. * Some functions using NRESIZE had grown unnecessarily complex in their efforts to bend to the macro's calling pattern. With the new list_resize function in place, those other functions could be simplified. That is being saved for a separate patch. * The ob_item==NULL check could be eliminated from the new list_resize function. This would entail finding each piece of code that sets ob_item to NULL and adding a new line to invalidate the overallocation tracking field. Rather than impose a new requirement on other pieces of list code, it was preferred to leave the NULL check in place and retain the benefits of decoupling, maintainability and information hiding (only PyList_New() and list_sort() need to know about the new field). This approach also reduces the odds of breaking an extension module. (Collaborative effort by Raymond Hettinger, Hye-Shik Chang, Tim Peters, and Armin Rigo.)
2004-02-13 07:36:39 -04:00
PyObject **garbage;
2006-02-15 13:27:45 -04:00
Py_ssize_t cur, i;
2002-07-18 23:35:45 -03:00
if (slicelength <= 0)
return 0;
if (step < 0) {
stop = start + 1;
start = stop + step*(slicelength - 1) - 1;
step = -step;
}
garbage = (PyObject**)
PyMem_MALLOC(slicelength*sizeof(PyObject*));
2002-07-18 23:35:45 -03:00
/* drawing pictures might help
understand these for loops */
2002-06-11 09:22:28 -03:00
for (cur = start, i = 0;
cur < stop;
cur += step, i++) {
2006-02-15 13:27:45 -04:00
Py_ssize_t lim = step;
garbage[i] = PyList_GET_ITEM(self, cur);
if (cur + step >= self->ob_size) {
lim = self->ob_size - cur - 1;
}
2004-07-28 23:29:26 -03:00
memmove(self->ob_item + cur - i,
self->ob_item + cur + 1,
lim * sizeof(PyObject *));
}
for (cur = start + slicelength*step + 1;
cur < self->ob_size; cur++) {
PyList_SET_ITEM(self, cur - slicelength,
PyList_GET_ITEM(self, cur));
}
self->ob_size -= slicelength;
list_resize(self, self->ob_size);
for (i = 0; i < slicelength; i++) {
Py_DECREF(garbage[i]);
}
PyMem_FREE(garbage);
return 0;
}
else {
/* assign slice */
PyObject **garbage, *ins, *seq, **seqitems, **selfitems;
2006-02-15 13:27:45 -04:00
Py_ssize_t cur, i;
/* protect against a[::-1] = a */
if (self == (PyListObject*)value) {
seq = list_slice((PyListObject*)value, 0,
PyList_GET_SIZE(value));
}
else {
2004-07-28 23:29:26 -03:00
seq = PySequence_Fast(value,
"must assign iterable to extended slice");
if (!seq)
return -1;
}
if (PySequence_Fast_GET_SIZE(seq) != slicelength) {
PyErr_Format(PyExc_ValueError,
"attempt to assign sequence of size %zd to extended slice of size %zd",
2006-02-16 02:59:22 -04:00
PySequence_Fast_GET_SIZE(seq),
slicelength);
Py_DECREF(seq);
return -1;
}
if (!slicelength) {
Py_DECREF(seq);
return 0;
}
garbage = (PyObject**)
PyMem_MALLOC(slicelength*sizeof(PyObject*));
2002-07-18 23:35:45 -03:00
selfitems = self->ob_item;
seqitems = PySequence_Fast_ITEMS(seq);
2002-07-18 23:35:45 -03:00
for (cur = start, i = 0; i < slicelength;
cur += step, i++) {
garbage[i] = selfitems[cur];
ins = seqitems[i];
Py_INCREF(ins);
selfitems[cur] = ins;
}
for (i = 0; i < slicelength; i++) {
Py_DECREF(garbage[i]);
}
2002-07-18 23:35:45 -03:00
PyMem_FREE(garbage);
Py_DECREF(seq);
2002-07-18 23:35:45 -03:00
return 0;
}
2002-07-18 23:35:45 -03:00
}
else {
2002-07-18 23:35:45 -03:00
PyErr_SetString(PyExc_TypeError,
"list indices must be integers");
return -1;
}
}
static PyMappingMethods list_as_mapping = {
2006-02-15 13:27:45 -04:00
(lenfunc)list_length,
(binaryfunc)list_subscript,
(objobjargproc)list_ass_subscript
};
1997-05-02 00:12:38 -03:00
PyTypeObject PyList_Type = {
PyObject_HEAD_INIT(&PyType_Type)
1990-10-14 09:07:46 -03:00
0,
"list",
2001-08-29 20:54:21 -03:00
sizeof(PyListObject),
1990-10-14 09:07:46 -03:00
0,
(destructor)list_dealloc, /* tp_dealloc */
(printfunc)list_print, /* tp_print */
2001-08-02 01:15:00 -03:00
0, /* tp_getattr */
0, /* tp_setattr */
0, /* tp_compare */
(reprfunc)list_repr, /* tp_repr */
0, /* tp_as_number */
&list_as_sequence, /* tp_as_sequence */
&list_as_mapping, /* tp_as_mapping */
list_nohash, /* tp_hash */
0, /* tp_call */
0, /* tp_str */
2001-08-02 01:15:00 -03:00
PyObject_GenericGetAttr, /* tp_getattro */
0, /* tp_setattro */
0, /* tp_as_buffer */
2001-08-29 20:54:21 -03:00
Py_TPFLAGS_DEFAULT | Py_TPFLAGS_HAVE_GC |
2001-08-02 01:15:00 -03:00
Py_TPFLAGS_BASETYPE, /* tp_flags */
list_doc, /* tp_doc */
(traverseproc)list_traverse, /* tp_traverse */
(inquiry)list_clear, /* tp_clear */
list_richcompare, /* tp_richcompare */
2001-08-02 01:15:00 -03:00
0, /* tp_weaklistoffset */
list_iter, /* tp_iter */
2001-08-02 01:15:00 -03:00
0, /* tp_iternext */
list_methods, /* tp_methods */
0, /* tp_members */
0, /* tp_getset */
0, /* tp_base */
0, /* tp_dict */
0, /* tp_descr_get */
0, /* tp_descr_set */
0, /* tp_dictoffset */
(initproc)list_init, /* tp_init */
PyType_GenericAlloc, /* tp_alloc */
PyType_GenericNew, /* tp_new */
2002-07-16 17:10:23 -03:00
PyObject_GC_Del, /* tp_free */
1990-10-14 09:07:46 -03:00
};
/*********************** List Iterator **************************/
typedef struct {
2002-07-16 17:10:23 -03:00
PyObject_HEAD
long it_index;
PyListObject *it_seq; /* Set to NULL when iterator is exhausted */
} listiterobject;
PyTypeObject PyListIter_Type;
2002-07-16 12:56:52 -03:00
static PyObject *
list_iter(PyObject *seq)
{
2002-07-16 17:10:23 -03:00
listiterobject *it;
if (!PyList_Check(seq)) {
PyErr_BadInternalCall();
return NULL;
}
it = PyObject_GC_New(listiterobject, &PyListIter_Type);
if (it == NULL)
return NULL;
it->it_index = 0;
Py_INCREF(seq);
it->it_seq = (PyListObject *)seq;
_PyObject_GC_TRACK(it);
return (PyObject *)it;
}
static void
listiter_dealloc(listiterobject *it)
{
2002-07-16 17:10:23 -03:00
_PyObject_GC_UNTRACK(it);
Py_XDECREF(it->it_seq);
2002-07-16 17:10:23 -03:00
PyObject_GC_Del(it);
}
static int
listiter_traverse(listiterobject *it, visitproc visit, void *arg)
{
if (it->it_seq == NULL)
return 0;
2002-07-16 17:10:23 -03:00
return visit((PyObject *)it->it_seq, arg);
}
static PyObject *
listiter_next(listiterobject *it)
{
2002-07-16 17:10:23 -03:00
PyListObject *seq;
PyObject *item;
assert(it != NULL);
2002-07-16 17:10:23 -03:00
seq = it->it_seq;
if (seq == NULL)
return NULL;
2002-07-16 17:10:23 -03:00
assert(PyList_Check(seq));
2002-07-16 17:10:23 -03:00
if (it->it_index < PyList_GET_SIZE(seq)) {
item = PyList_GET_ITEM(seq, it->it_index);
++it->it_index;
2002-07-16 17:10:23 -03:00
Py_INCREF(item);
return item;
}
Py_DECREF(seq);
it->it_seq = NULL;
2002-07-16 17:10:23 -03:00
return NULL;
}
static PyObject *
listiter_len(listiterobject *it)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t len;
if (it->it_seq) {
len = PyList_GET_SIZE(it->it_seq) - it->it_index;
if (len >= 0)
2006-02-15 13:27:45 -04:00
return PyInt_FromSsize_t(len);
}
return PyInt_FromLong(0);
}
PyDoc_STRVAR(length_hint_doc, "Private method returning an estimate of len(list(it)).");
static PyMethodDef listiter_methods[] = {
{"__length_hint__", (PyCFunction)listiter_len, METH_NOARGS, length_hint_doc},
{NULL, NULL} /* sentinel */
};
PyTypeObject PyListIter_Type = {
2002-07-16 17:10:23 -03:00
PyObject_HEAD_INIT(&PyType_Type)
0, /* ob_size */
"listiterator", /* tp_name */
sizeof(listiterobject), /* tp_basicsize */
0, /* tp_itemsize */
/* methods */
(destructor)listiter_dealloc, /* tp_dealloc */
0, /* tp_print */
0, /* tp_getattr */
0, /* tp_setattr */
0, /* tp_compare */
0, /* tp_repr */
0, /* tp_as_number */
0, /* tp_as_sequence */
2002-07-16 17:10:23 -03:00
0, /* tp_as_mapping */
0, /* tp_hash */
0, /* tp_call */
0, /* tp_str */
PyObject_GenericGetAttr, /* tp_getattro */
0, /* tp_setattro */
0, /* tp_as_buffer */
Py_TPFLAGS_DEFAULT | Py_TPFLAGS_HAVE_GC,/* tp_flags */
0, /* tp_doc */
(traverseproc)listiter_traverse, /* tp_traverse */
0, /* tp_clear */
0, /* tp_richcompare */
0, /* tp_weaklistoffset */
PyObject_SelfIter, /* tp_iter */
2002-07-16 17:10:23 -03:00
(iternextfunc)listiter_next, /* tp_iternext */
listiter_methods, /* tp_methods */
2002-07-16 17:10:23 -03:00
0, /* tp_members */
};
/*********************** List Reverse Iterator **************************/
typedef struct {
PyObject_HEAD
Py_ssize_t it_index;
PyListObject *it_seq; /* Set to NULL when iterator is exhausted */
} listreviterobject;
PyTypeObject PyListRevIter_Type;
static PyObject *
list_reversed(PyListObject *seq, PyObject *unused)
{
listreviterobject *it;
it = PyObject_GC_New(listreviterobject, &PyListRevIter_Type);
if (it == NULL)
return NULL;
assert(PyList_Check(seq));
it->it_index = PyList_GET_SIZE(seq) - 1;
Py_INCREF(seq);
it->it_seq = seq;
PyObject_GC_Track(it);
return (PyObject *)it;
}
static void
listreviter_dealloc(listreviterobject *it)
{
PyObject_GC_UnTrack(it);
Py_XDECREF(it->it_seq);
PyObject_GC_Del(it);
}
static int
listreviter_traverse(listreviterobject *it, visitproc visit, void *arg)
{
if (it->it_seq == NULL)
return 0;
return visit((PyObject *)it->it_seq, arg);
}
static PyObject *
listreviter_next(listreviterobject *it)
{
PyObject *item;
Py_ssize_t index = it->it_index;
PyListObject *seq = it->it_seq;
if (index>=0 && index < PyList_GET_SIZE(seq)) {
item = PyList_GET_ITEM(seq, index);
it->it_index--;
Py_INCREF(item);
return item;
}
it->it_index = -1;
if (seq != NULL) {
it->it_seq = NULL;
Py_DECREF(seq);
}
return NULL;
}
2006-02-15 13:27:45 -04:00
static Py_ssize_t
listreviter_len(listreviterobject *it)
{
2006-02-15 13:27:45 -04:00
Py_ssize_t len = it->it_index + 1;
if (it->it_seq == NULL || PyList_GET_SIZE(it->it_seq) < len)
return 0;
return len;
}
static PySequenceMethods listreviter_as_sequence = {
2006-02-15 13:27:45 -04:00
(lenfunc)listreviter_len, /* sq_length */
0, /* sq_concat */
};
PyTypeObject PyListRevIter_Type = {
PyObject_HEAD_INIT(&PyType_Type)
0, /* ob_size */
"listreverseiterator", /* tp_name */
sizeof(listreviterobject), /* tp_basicsize */
0, /* tp_itemsize */
/* methods */
(destructor)listreviter_dealloc, /* tp_dealloc */
0, /* tp_print */
0, /* tp_getattr */
0, /* tp_setattr */
0, /* tp_compare */
0, /* tp_repr */
0, /* tp_as_number */
&listreviter_as_sequence, /* tp_as_sequence */
0, /* tp_as_mapping */
0, /* tp_hash */
0, /* tp_call */
0, /* tp_str */
PyObject_GenericGetAttr, /* tp_getattro */
0, /* tp_setattro */
0, /* tp_as_buffer */
Py_TPFLAGS_DEFAULT | Py_TPFLAGS_HAVE_GC,/* tp_flags */
0, /* tp_doc */
(traverseproc)listreviter_traverse, /* tp_traverse */
0, /* tp_clear */
0, /* tp_richcompare */
0, /* tp_weaklistoffset */
PyObject_SelfIter, /* tp_iter */
(iternextfunc)listreviter_next, /* tp_iternext */
0,
};