1
# Copyright (C) 2009 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
"""Definition of a class that is similar to Set with some small changes."""
19
cdef extern from "python-compat.h":
22
cdef extern from "Python.h":
23
ctypedef unsigned long size_t
24
ctypedef long (*hashfunc)(PyObject*)
25
ctypedef PyObject *(*richcmpfunc)(PyObject *, PyObject *, int)
26
ctypedef int (*visitproc)(PyObject *, void *)
27
ctypedef int (*traverseproc)(PyObject *, visitproc, void *)
30
PyObject *Py_NotImplemented
31
void Py_INCREF(PyObject *)
32
void Py_DECREF(PyObject *)
33
ctypedef struct PyTypeObject:
35
richcmpfunc tp_richcompare
36
traverseproc tp_traverse
38
PyTypeObject *Py_TYPE(PyObject *)
39
int PyObject_IsTrue(PyObject *)
40
# Note: *Don't* use hash(), Pyrex 0.9.8.5 thinks it returns an 'int', and
41
# thus silently truncates to 32-bits on 64-bit machines.
42
long PyObject_Hash(PyObject *) except -1
44
void *PyMem_Malloc(size_t nbytes)
45
void PyMem_Free(void *)
46
void memset(void *, int, size_t)
49
# Dummy is an object used to mark nodes that have been deleted. Since
50
# collisions require us to move a node to an alternative location, if we just
51
# set an entry to NULL on delete, we won't find any relocated nodes.
52
# We have to use _dummy_obj because we need to keep a refcount to it, but we
53
# also use _dummy as a pointer, because it avoids having to put <PyObject*> all
55
cdef object _dummy_obj
58
_dummy = <PyObject *>_dummy_obj
61
cdef int _is_equal(PyObject *this, long this_hash, PyObject *other) except -1:
67
other_hash = PyObject_Hash(other)
68
if other_hash != this_hash:
71
# This implements a subset of the PyObject_RichCompareBool functionality.
73
# 1) Doesn't try to do anything with old-style classes
74
# 2) Assumes that both objects have a tp_richcompare implementation, and
75
# that if that is not enough to compare equal, then they are not
76
# equal. (It doesn't try to cast them both to some intermediate form
77
# that would compare equal.)
78
res = Py_TYPE(this).tp_richcompare(this, other, Py_EQ)
79
if res == NULL: # Exception
81
if PyObject_IsTrue(res):
84
if res == Py_NotImplemented:
86
res = Py_TYPE(other).tp_richcompare(other, this, Py_EQ)
89
if PyObject_IsTrue(res):
96
cdef public api class SimpleSet [object SimpleSetObject, type SimpleSet_Type]:
97
"""This class can be used to track canonical forms for objects.
99
It is similar in function to the interned dictionary that is used by
102
1) It assumes that hash(obj) is cheap, so does not need to inline a copy
104
2) It only stores one reference to the object, rather than 2 (key vs
107
As such, it uses 1/3rd the amount of memory to store a pointer to the
110
# Attributes are defined in the .pxd file
111
DEF DEFAULT_SIZE=1024
114
cdef Py_ssize_t size, n_bytes
117
self._mask = size - 1
120
n_bytes = sizeof(PyObject*) * size;
121
self._table = <PyObject **>PyMem_Malloc(n_bytes)
122
if self._table == NULL:
124
memset(self._table, 0, n_bytes)
126
def __dealloc__(self):
127
if self._table != NULL:
128
PyMem_Free(self._table)
143
def _memory_size(self):
144
"""Return the number of bytes of memory consumed by this class."""
145
return sizeof(self) + (sizeof(PyObject*)*(self._mask + 1))
150
def _test_lookup(self, key):
153
slot = _lookup(self, key)
156
elif slot[0] == _dummy:
159
res = <object>slot[0]
160
return <int>(slot - self._table), res
162
def __contains__(self, key):
163
"""Is key present in this SimpleSet."""
166
slot = _lookup(self, key)
167
if slot[0] == NULL or slot[0] == _dummy:
171
cdef PyObject *_get(self, object key) except? NULL:
172
"""Return the object (or nothing) define at the given location."""
175
slot = _lookup(self, key)
176
if slot[0] == NULL or slot[0] == _dummy:
180
def __getitem__(self, key):
181
"""Return a stored item that is equivalent to key."""
182
cdef PyObject *py_val
184
py_val = self._get(key)
186
raise KeyError("Key %s is not present" % key)
187
val = <object>(py_val)
190
cdef int _insert_clean(self, PyObject *key) except -1:
191
"""Insert a key into self.table.
193
This is only meant to be used during times like '_resize',
194
as it makes a lot of assuptions about keys not already being present,
195
and there being no dummy entries.
197
cdef size_t i, n_lookup
199
cdef PyObject **table, **slot
205
the_hash = PyObject_Hash(key)
207
for n_lookup from 0 <= n_lookup <= <size_t>mask: # Don't loop forever
208
slot = &table[i & mask]
211
self._fill = self._fill + 1
212
self._used = self._used + 1
215
raise RuntimeError('ran out of slots.')
217
def _py_resize(self, min_used):
218
"""Do not use this directly, it is only exposed for testing."""
219
return self._resize(min_used)
221
cdef Py_ssize_t _resize(self, Py_ssize_t min_used) except -1:
222
"""Resize the internal table.
224
The final table will be big enough to hold at least min_used entries.
225
We will copy the data from the existing table over, leaving out dummy
228
:return: The new size of the internal table
230
cdef Py_ssize_t new_size, n_bytes, remaining
231
cdef PyObject **new_table, **old_table, **slot
233
new_size = DEFAULT_SIZE
234
while new_size <= min_used and new_size > 0:
235
new_size = new_size << 1
236
# We rolled over our signed size field
239
# Even if min_used == self._mask + 1, and we aren't changing the actual
240
# size, we will still run the algorithm so that dummy entries are
243
# if new_size < self._used:
244
# raise RuntimeError('cannot shrink SimpleSet to something'
245
# ' smaller than the number of used slots.')
246
n_bytes = sizeof(PyObject*) * new_size;
247
new_table = <PyObject **>PyMem_Malloc(n_bytes)
248
if new_table == NULL:
251
old_table = self._table
252
self._table = new_table
253
memset(self._table, 0, n_bytes)
254
self._mask = new_size - 1
256
remaining = self._fill
259
# Moving everything to the other table is refcount neutral, so we don't
263
if slot[0] == NULL: # unused slot
265
elif slot[0] == _dummy: # dummy slot
266
remaining = remaining - 1
268
remaining = remaining - 1
269
self._insert_clean(slot[0])
271
PyMem_Free(old_table)
275
"""Similar to set.add(), start tracking this key.
277
There is one small difference, which is that we return the object that
278
is stored at the given location. (which is closer to the
279
dict.setdefault() functionality.)
281
return self._add(key)
283
cdef object _add(self, key):
284
cdef PyObject **slot, *py_key
287
py_key = <PyObject *>key
288
if (Py_TYPE(py_key).tp_richcompare == NULL
289
or Py_TYPE(py_key).tp_hash == NULL):
290
raise TypeError('Types added to SimpleSet must implement'
291
' both tp_richcompare and tp_hash')
293
# We need at least one empty slot
294
assert self._used < self._mask
295
slot = _lookup(self, key)
296
if (slot[0] == NULL):
298
self._fill = self._fill + 1
299
self._used = self._used + 1
302
elif (slot[0] == _dummy):
304
self._used = self._used + 1
307
# No else: clause. If _lookup returns a pointer to
308
# a live object, then we already have a value at this location.
309
retval = <object>(slot[0])
310
# PySet and PyDict use a 2-3rds full algorithm, we'll follow suit
311
if added and (self._fill * 3) >= ((self._mask + 1) * 2):
312
# However, we always work for a load factor of 2:1
313
self._resize(self._used * 2)
314
# Even if we resized and ended up moving retval into a different slot,
315
# it is still the value that is held at the slot equivalent to 'key',
316
# so we can still return it
319
def discard(self, key):
320
"""Remove key from the set, whether it exists or not.
322
:return: False if the item did not exist, True if it did
324
if self._discard(key):
328
cdef int _discard(self, key) except -1:
329
cdef PyObject **slot, *py_key
331
slot = _lookup(self, key)
332
if slot[0] == NULL or slot[0] == _dummy:
334
self._used = self._used - 1
337
# PySet uses the heuristic: If more than 1/5 are dummies, then resize
339
# if ((so->_fill - so->_used) * 5 < so->mask)
340
# However, we are planning on using this as an interning structure, in
341
# which we will be putting a lot of objects. And we expect that large
342
# groups of them are going to have the same lifetime.
343
# Dummy entries hurt a little bit because they cause the lookup to keep
344
# searching, but resizing is also rather expensive
345
# For now, we'll just use their algorithm, but we may want to revisit
347
if ((self._fill - self._used) * 5 > self._mask):
348
self._resize(self._used * 2)
352
return _SimpleSet_iterator(self)
355
cdef class _SimpleSet_iterator:
356
"""Iterator over the SimpleSet structure."""
360
cdef Py_ssize_t _used # track if things have been mutated while iterating
361
cdef Py_ssize_t len # number of entries left
363
def __init__(self, obj):
366
self._used = self.set._used
367
self.len = self.set._used
373
cdef Py_ssize_t mask, i
378
if self.set._used != self._used:
379
# Force this exception to continue to be raised
381
raise RuntimeError("Set size changed during iteration")
382
if not SimpleSet_Next(self.set, &self.pos, &key):
386
the_key = <object>key # INCREF
387
self.len = self.len - 1
390
def __length_hint__(self):
391
if self.set is not None and self._used == self.set._used:
397
cdef api SimpleSet SimpleSet_New():
398
"""Create a new SimpleSet object."""
402
cdef SimpleSet _check_self(object self):
403
"""Check that the parameter is not None.
405
Pyrex/Cython will do type checking, but only to ensure that an object is
406
either the right type or None. You can say "object foo not None" for pure
407
python functions, but not for C functions.
408
So this is just a helper for all the apis that need to do the check.
410
cdef SimpleSet true_self
412
raise TypeError('self must not be None')
417
cdef PyObject **_lookup(SimpleSet self, object key) except NULL:
418
"""Find the slot where 'key' would fit.
420
This is the same as a dicts 'lookup' function.
422
:param key: An object we are looking up
423
:param hash: The hash for key
424
:return: The location in self.table where key should be put.
425
location == NULL is an exception, but (*location) == NULL just
426
indicates the slot is empty and can be used.
428
# This uses Quadratic Probing:
429
# http://en.wikipedia.org/wiki/Quadratic_probing
431
# This leads to probe locations at:
434
# h2 = h0 + 3 = h1 + 1 + 1
435
# h3 = h0 + 6 = h2 + 1 + 2
436
# h4 = h0 + 10 = h2 + 1 + 3
437
# Note that all of these are '& mask', but that is computed *after* the
439
# This differs from the algorithm used by Set and Dict. Which, effectively,
440
# use double-hashing, and a step size that starts large, but dwindles to
441
# stepping one-by-one.
442
# This gives more 'locality' in that if you have a collision at offset X,
443
# the first fallback is X+1, which is fast to check. However, that means
444
# that an object w/ hash X+1 will also check there, and then X+2 next.
445
# However, for objects with differing hashes, their chains are different.
446
# The former checks X, X+1, X+3, ... the latter checks X+1, X+2, X+4, ...
447
# So different hashes diverge quickly.
448
# A bigger problem is that we *only* ever use the lowest bits of the hash
449
# So all integers (x + SIZE*N) will resolve into the same bucket, and all
450
# use the same collision resolution. We may want to try to find a way to
451
# incorporate the upper bits of the hash with quadratic probing. (For
452
# example, X, X+1, X+3+some_upper_bits, X+6+more_upper_bits, etc.)
453
cdef size_t i, n_lookup
456
cdef PyObject **table, **slot, *cur, **free_slot, *py_key
458
py_key = <PyObject *>key
459
# Note: avoid using hash(obj) because of a bug w/ pyrex 0.9.8.5 and 64-bit
460
# (it treats hash() as returning an 'int' rather than a 'long')
461
key_hash = PyObject_Hash(py_key)
466
for n_lookup from 0 <= n_lookup <= <size_t>mask: # Don't loop forever
467
slot = &table[i & mask]
471
if free_slot != NULL:
472
# Did we find an earlier _dummy entry?
477
# Found an exact pointer to the key
480
if free_slot == NULL:
482
elif _is_equal(py_key, key_hash, cur):
483
# Both py_key and cur belong in this slot, return it
486
raise AssertionError('should never get here')
489
cdef api PyObject **_SimpleSet_Lookup(object self, object key) except NULL:
490
"""Find the slot where 'key' would fit.
492
This is the same as a dicts 'lookup' function. This is a private
493
api because mutating what you get without maintaing the other invariants
496
:param key: An object we are looking up
497
:param hash: The hash for key
498
:return: The location in self._table where key should be put
499
should never be NULL, but may reference a NULL (PyObject*)
501
return _lookup(_check_self(self), key)
504
cdef api object SimpleSet_Add(object self, object key):
505
"""Add a key to the SimpleSet (set).
507
:param self: The SimpleSet to add the key to.
508
:param key: The key to be added. If the key is already present,
509
self will not be modified
510
:return: The current key stored at the location defined by 'key'.
511
This may be the same object, or it may be an equivalent object.
512
(consider dict.setdefault(key, key))
514
return _check_self(self)._add(key)
517
cdef api int SimpleSet_Contains(object self, object key) except -1:
518
"""Is key present in self?"""
519
return (key in _check_self(self))
522
cdef api int SimpleSet_Discard(object self, object key) except -1:
523
"""Remove the object referenced at location 'key'.
525
:param self: The SimpleSet being modified
526
:param key: The key we are checking on
527
:return: 1 if there was an object present, 0 if there was not, and -1 on
530
return _check_self(self)._discard(key)
533
cdef api PyObject *SimpleSet_Get(SimpleSet self, object key) except? NULL:
534
"""Get a pointer to the object present at location 'key'.
536
This returns an object which is equal to key which was previously added to
537
self. This returns a borrowed reference, as it may also return NULL if no
538
value is present at that location.
540
:param key: The value we are looking for
541
:return: The object present at that location
543
return _check_self(self)._get(key)
546
cdef api Py_ssize_t SimpleSet_Size(object self) except -1:
547
"""Get the number of active entries in 'self'"""
548
return _check_self(self)._used
551
cdef api int SimpleSet_Next(object self, Py_ssize_t *pos, PyObject **key):
552
"""Walk over items in a SimpleSet.
554
:param pos: should be initialized to 0 by the caller, and will be updated
556
:param key: Will return a borrowed reference to key
557
:return: 0 if nothing left, 1 if we are returning a new value
559
cdef Py_ssize_t i, mask
560
cdef SimpleSet true_self
561
cdef PyObject **table
562
true_self = _check_self(self)
566
mask = true_self._mask
567
table= true_self._table
568
while (i <= mask and (table[i] == NULL or table[i] == _dummy)):
578
cdef int SimpleSet_traverse(SimpleSet self, visitproc visit, void *arg):
579
"""This is an implementation of 'tp_traverse' that hits the whole table.
581
Cython/Pyrex don't seem to let you define a tp_traverse, and they only
582
define one for you if you have an 'object' attribute. Since they don't
583
support C arrays of objects, we access the PyObject * directly.
586
cdef PyObject *next_key
590
while SimpleSet_Next(self, &pos, &next_key):
591
ret = visit(next_key, arg)
596
# It is a little bit ugly to do this, but it works, and means that Meliae can
597
# dump the total memory consumed by all child objects.
598
(<PyTypeObject *>SimpleSet).tp_traverse = <traverseproc>SimpleSet_traverse
added
removed
bzrlib/_simple_set_pyx.pyx
