• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* Drop in replacement for heapq.py
2 
3 C implementation derived directly from heapq.py in Py2.3
4 which was written by Kevin O'Connor, augmented by Tim Peters,
5 annotated by François Pinard, and converted to C by Raymond Hettinger.
6 
7 */
8 
9 #include "Python.h"
10 
11 /* Older implementations of heapq used Py_LE for comparisons.  Now, it uses
12    Py_LT so it will match min(), sorted(), and bisect().  Unfortunately, some
13    client code (Twisted for example) relied on Py_LE, so this little function
14    restores compatibility by trying both.
15 */
16 static int
cmp_lt(PyObject * x,PyObject * y)17 cmp_lt(PyObject *x, PyObject *y)
18 {
19     int cmp;
20     static PyObject *lt = NULL;
21 
22     if (lt == NULL) {
23         lt = PyString_FromString("__lt__");
24         if (lt == NULL)
25             return -1;
26     }
27     if (PyObject_HasAttr(x, lt))
28         return PyObject_RichCompareBool(x, y, Py_LT);
29     cmp = PyObject_RichCompareBool(y, x, Py_LE);
30     if (cmp != -1)
31         cmp = 1 - cmp;
32     return cmp;
33 }
34 
35 static int
_siftdown(PyListObject * heap,Py_ssize_t startpos,Py_ssize_t pos)36 _siftdown(PyListObject *heap, Py_ssize_t startpos, Py_ssize_t pos)
37 {
38     PyObject *newitem, *parent;
39     Py_ssize_t parentpos, size;
40     int cmp;
41 
42     assert(PyList_Check(heap));
43     size = PyList_GET_SIZE(heap);
44     if (pos >= size) {
45         PyErr_SetString(PyExc_IndexError, "index out of range");
46         return -1;
47     }
48 
49     /* Follow the path to the root, moving parents down until finding
50        a place newitem fits. */
51     newitem = PyList_GET_ITEM(heap, pos);
52     while (pos > startpos) {
53         parentpos = (pos - 1) >> 1;
54         parent = PyList_GET_ITEM(heap, parentpos);
55         cmp = cmp_lt(newitem, parent);
56         if (cmp == -1)
57             return -1;
58         if (size != PyList_GET_SIZE(heap)) {
59             PyErr_SetString(PyExc_RuntimeError,
60                             "list changed size during iteration");
61             return -1;
62         }
63         if (cmp == 0)
64             break;
65         parent = PyList_GET_ITEM(heap, parentpos);
66         newitem = PyList_GET_ITEM(heap, pos);
67         PyList_SET_ITEM(heap, parentpos, newitem);
68         PyList_SET_ITEM(heap, pos, parent);
69         pos = parentpos;
70     }
71     return 0;
72 }
73 
74 static int
_siftup(PyListObject * heap,Py_ssize_t pos)75 _siftup(PyListObject *heap, Py_ssize_t pos)
76 {
77     Py_ssize_t startpos, endpos, childpos, rightpos, limit;
78     PyObject *tmp1, *tmp2;
79     int cmp;
80 
81     assert(PyList_Check(heap));
82     endpos = PyList_GET_SIZE(heap);
83     startpos = pos;
84     if (pos >= endpos) {
85         PyErr_SetString(PyExc_IndexError, "index out of range");
86         return -1;
87     }
88 
89     /* Bubble up the smaller child until hitting a leaf. */
90     limit = endpos / 2;          /* smallest pos that has no child */
91     while (pos < limit) {
92         /* Set childpos to index of smaller child.   */
93         childpos = 2*pos + 1;    /* leftmost child position  */
94         rightpos = childpos + 1;
95         if (rightpos < endpos) {
96             cmp = cmp_lt(
97                 PyList_GET_ITEM(heap, childpos),
98                 PyList_GET_ITEM(heap, rightpos));
99             if (cmp == -1)
100                 return -1;
101             if (cmp == 0)
102                 childpos = rightpos;
103             if (endpos != PyList_GET_SIZE(heap)) {
104                 PyErr_SetString(PyExc_RuntimeError,
105                                 "list changed size during iteration");
106                 return -1;
107             }
108         }
109         /* Move the smaller child up. */
110         tmp1 = PyList_GET_ITEM(heap, childpos);
111         tmp2 = PyList_GET_ITEM(heap, pos);
112         PyList_SET_ITEM(heap, childpos, tmp2);
113         PyList_SET_ITEM(heap, pos, tmp1);
114         pos = childpos;
115     }
116     /* Bubble it up to its final resting place (by sifting its parents down). */
117     return _siftdown(heap, startpos, pos);
118 }
119 
120 static PyObject *
heappush(PyObject * self,PyObject * args)121 heappush(PyObject *self, PyObject *args)
122 {
123     PyObject *heap, *item;
124 
125     if (!PyArg_UnpackTuple(args, "heappush", 2, 2, &heap, &item))
126         return NULL;
127 
128     if (!PyList_Check(heap)) {
129         PyErr_SetString(PyExc_TypeError, "heap argument must be a list");
130         return NULL;
131     }
132 
133     if (PyList_Append(heap, item) == -1)
134         return NULL;
135 
136     if (_siftdown((PyListObject *)heap, 0, PyList_GET_SIZE(heap)-1) == -1)
137         return NULL;
138     Py_INCREF(Py_None);
139     return Py_None;
140 }
141 
142 PyDoc_STRVAR(heappush_doc,
143 "heappush(heap, item) -> None. Push item onto heap, maintaining the heap invariant.");
144 
145 static PyObject *
heappop(PyObject * self,PyObject * heap)146 heappop(PyObject *self, PyObject *heap)
147 {
148     PyObject *lastelt, *returnitem;
149     Py_ssize_t n;
150 
151     if (!PyList_Check(heap)) {
152         PyErr_SetString(PyExc_TypeError, "heap argument must be a list");
153         return NULL;
154     }
155 
156     /* # raises appropriate IndexError if heap is empty */
157     n = PyList_GET_SIZE(heap);
158     if (n == 0) {
159         PyErr_SetString(PyExc_IndexError, "index out of range");
160         return NULL;
161     }
162 
163     lastelt = PyList_GET_ITEM(heap, n-1) ;
164     Py_INCREF(lastelt);
165     PyList_SetSlice(heap, n-1, n, NULL);
166     n--;
167 
168     if (!n)
169         return lastelt;
170     returnitem = PyList_GET_ITEM(heap, 0);
171     PyList_SET_ITEM(heap, 0, lastelt);
172     if (_siftup((PyListObject *)heap, 0) == -1) {
173         Py_DECREF(returnitem);
174         return NULL;
175     }
176     return returnitem;
177 }
178 
179 PyDoc_STRVAR(heappop_doc,
180 "Pop the smallest item off the heap, maintaining the heap invariant.");
181 
182 static PyObject *
heapreplace(PyObject * self,PyObject * args)183 heapreplace(PyObject *self, PyObject *args)
184 {
185     PyObject *heap, *item, *returnitem;
186 
187     if (!PyArg_UnpackTuple(args, "heapreplace", 2, 2, &heap, &item))
188         return NULL;
189 
190     if (!PyList_Check(heap)) {
191         PyErr_SetString(PyExc_TypeError, "heap argument must be a list");
192         return NULL;
193     }
194 
195     if (PyList_GET_SIZE(heap) < 1) {
196         PyErr_SetString(PyExc_IndexError, "index out of range");
197         return NULL;
198     }
199 
200     returnitem = PyList_GET_ITEM(heap, 0);
201     Py_INCREF(item);
202     PyList_SET_ITEM(heap, 0, item);
203     if (_siftup((PyListObject *)heap, 0) == -1) {
204         Py_DECREF(returnitem);
205         return NULL;
206     }
207     return returnitem;
208 }
209 
210 PyDoc_STRVAR(heapreplace_doc,
211 "heapreplace(heap, item) -> value. Pop and return the current smallest value, and add the new item.\n\
212 \n\
213 This is more efficient than heappop() followed by heappush(), and can be\n\
214 more appropriate when using a fixed-size heap.  Note that the value\n\
215 returned may be larger than item!  That constrains reasonable uses of\n\
216 this routine unless written as part of a conditional replacement:\n\n\
217     if item > heap[0]:\n\
218         item = heapreplace(heap, item)\n");
219 
220 static PyObject *
heappushpop(PyObject * self,PyObject * args)221 heappushpop(PyObject *self, PyObject *args)
222 {
223     PyObject *heap, *item, *returnitem;
224     int cmp;
225 
226     if (!PyArg_UnpackTuple(args, "heappushpop", 2, 2, &heap, &item))
227         return NULL;
228 
229     if (!PyList_Check(heap)) {
230         PyErr_SetString(PyExc_TypeError, "heap argument must be a list");
231         return NULL;
232     }
233 
234     if (PyList_GET_SIZE(heap) < 1) {
235         Py_INCREF(item);
236         return item;
237     }
238 
239     cmp = cmp_lt(PyList_GET_ITEM(heap, 0), item);
240     if (cmp == -1)
241         return NULL;
242     if (cmp == 0) {
243         Py_INCREF(item);
244         return item;
245     }
246 
247     if (PyList_GET_SIZE(heap) == 0) {
248         PyErr_SetString(PyExc_IndexError, "index out of range");
249         return NULL;
250     }
251 
252     returnitem = PyList_GET_ITEM(heap, 0);
253     Py_INCREF(item);
254     PyList_SET_ITEM(heap, 0, item);
255     if (_siftup((PyListObject *)heap, 0) == -1) {
256         Py_DECREF(returnitem);
257         return NULL;
258     }
259     return returnitem;
260 }
261 
262 PyDoc_STRVAR(heappushpop_doc,
263 "heappushpop(heap, item) -> value. Push item on the heap, then pop and return the smallest item\n\
264 from the heap. The combined action runs more efficiently than\n\
265 heappush() followed by a separate call to heappop().");
266 
267 static PyObject *
heapify(PyObject * self,PyObject * heap)268 heapify(PyObject *self, PyObject *heap)
269 {
270     Py_ssize_t i, n;
271 
272     if (!PyList_Check(heap)) {
273         PyErr_SetString(PyExc_TypeError, "heap argument must be a list");
274         return NULL;
275     }
276 
277     n = PyList_GET_SIZE(heap);
278     /* Transform bottom-up.  The largest index there's any point to
279        looking at is the largest with a child index in-range, so must
280        have 2*i + 1 < n, or i < (n-1)/2.  If n is even = 2*j, this is
281        (2*j-1)/2 = j-1/2 so j-1 is the largest, which is n//2 - 1.  If
282        n is odd = 2*j+1, this is (2*j+1-1)/2 = j so j-1 is the largest,
283        and that's again n//2-1.
284     */
285     for (i=n/2-1 ; i>=0 ; i--)
286         if(_siftup((PyListObject *)heap, i) == -1)
287             return NULL;
288     Py_INCREF(Py_None);
289     return Py_None;
290 }
291 
292 PyDoc_STRVAR(heapify_doc,
293 "Transform list into a heap, in-place, in O(len(heap)) time.");
294 
295 static PyObject *
nlargest(PyObject * self,PyObject * args)296 nlargest(PyObject *self, PyObject *args)
297 {
298     PyObject *heap=NULL, *elem, *iterable, *sol, *it, *oldelem;
299     Py_ssize_t i, n;
300     int cmp;
301 
302     if (!PyArg_ParseTuple(args, "nO:nlargest", &n, &iterable))
303         return NULL;
304 
305     it = PyObject_GetIter(iterable);
306     if (it == NULL)
307         return NULL;
308 
309     heap = PyList_New(0);
310     if (heap == NULL)
311         goto fail;
312 
313     for (i=0 ; i<n ; i++ ){
314         elem = PyIter_Next(it);
315         if (elem == NULL) {
316             if (PyErr_Occurred())
317                 goto fail;
318             else
319                 goto sortit;
320         }
321         if (PyList_Append(heap, elem) == -1) {
322             Py_DECREF(elem);
323             goto fail;
324         }
325         Py_DECREF(elem);
326     }
327     if (PyList_GET_SIZE(heap) == 0)
328         goto sortit;
329 
330     for (i=n/2-1 ; i>=0 ; i--)
331         if(_siftup((PyListObject *)heap, i) == -1)
332             goto fail;
333 
334     sol = PyList_GET_ITEM(heap, 0);
335     while (1) {
336         elem = PyIter_Next(it);
337         if (elem == NULL) {
338             if (PyErr_Occurred())
339                 goto fail;
340             else
341                 goto sortit;
342         }
343         cmp = cmp_lt(sol, elem);
344         if (cmp == -1) {
345             Py_DECREF(elem);
346             goto fail;
347         }
348         if (cmp == 0) {
349             Py_DECREF(elem);
350             continue;
351         }
352         oldelem = PyList_GET_ITEM(heap, 0);
353         PyList_SET_ITEM(heap, 0, elem);
354         Py_DECREF(oldelem);
355         if (_siftup((PyListObject *)heap, 0) == -1)
356             goto fail;
357         sol = PyList_GET_ITEM(heap, 0);
358     }
359 sortit:
360     if (PyList_Sort(heap) == -1)
361         goto fail;
362     if (PyList_Reverse(heap) == -1)
363         goto fail;
364     Py_DECREF(it);
365     return heap;
366 
367 fail:
368     Py_DECREF(it);
369     Py_XDECREF(heap);
370     return NULL;
371 }
372 
373 PyDoc_STRVAR(nlargest_doc,
374 "Find the n largest elements in a dataset.\n\
375 \n\
376 Equivalent to:  sorted(iterable, reverse=True)[:n]\n");
377 
378 static int
_siftdownmax(PyListObject * heap,Py_ssize_t startpos,Py_ssize_t pos)379 _siftdownmax(PyListObject *heap, Py_ssize_t startpos, Py_ssize_t pos)
380 {
381     PyObject *newitem, *parent;
382     int cmp;
383     Py_ssize_t parentpos;
384 
385     assert(PyList_Check(heap));
386     if (pos >= PyList_GET_SIZE(heap)) {
387         PyErr_SetString(PyExc_IndexError, "index out of range");
388         return -1;
389     }
390 
391     newitem = PyList_GET_ITEM(heap, pos);
392     Py_INCREF(newitem);
393     /* Follow the path to the root, moving parents down until finding
394        a place newitem fits. */
395     while (pos > startpos){
396         parentpos = (pos - 1) >> 1;
397         parent = PyList_GET_ITEM(heap, parentpos);
398         cmp = cmp_lt(parent, newitem);
399         if (cmp == -1) {
400             Py_DECREF(newitem);
401             return -1;
402         }
403         if (cmp == 0)
404             break;
405         Py_INCREF(parent);
406         Py_DECREF(PyList_GET_ITEM(heap, pos));
407         PyList_SET_ITEM(heap, pos, parent);
408         pos = parentpos;
409     }
410     Py_DECREF(PyList_GET_ITEM(heap, pos));
411     PyList_SET_ITEM(heap, pos, newitem);
412     return 0;
413 }
414 
415 static int
_siftupmax(PyListObject * heap,Py_ssize_t pos)416 _siftupmax(PyListObject *heap, Py_ssize_t pos)
417 {
418     Py_ssize_t startpos, endpos, childpos, rightpos, limit;
419     int cmp;
420     PyObject *newitem, *tmp;
421 
422     assert(PyList_Check(heap));
423     endpos = PyList_GET_SIZE(heap);
424     startpos = pos;
425     if (pos >= endpos) {
426         PyErr_SetString(PyExc_IndexError, "index out of range");
427         return -1;
428     }
429     newitem = PyList_GET_ITEM(heap, pos);
430     Py_INCREF(newitem);
431 
432     /* Bubble up the smaller child until hitting a leaf. */
433     limit = endpos / 2;          /* smallest pos that has no child */
434     while (pos < limit) {
435         /* Set childpos to index of smaller child.   */
436         childpos = 2*pos + 1;    /* leftmost child position  */
437         rightpos = childpos + 1;
438         if (rightpos < endpos) {
439             cmp = cmp_lt(
440                 PyList_GET_ITEM(heap, rightpos),
441                 PyList_GET_ITEM(heap, childpos));
442             if (cmp == -1) {
443                 Py_DECREF(newitem);
444                 return -1;
445             }
446             if (cmp == 0)
447                 childpos = rightpos;
448         }
449         /* Move the smaller child up. */
450         tmp = PyList_GET_ITEM(heap, childpos);
451         Py_INCREF(tmp);
452         Py_DECREF(PyList_GET_ITEM(heap, pos));
453         PyList_SET_ITEM(heap, pos, tmp);
454         pos = childpos;
455     }
456 
457     /* The leaf at pos is empty now.  Put newitem there, and bubble
458        it up to its final resting place (by sifting its parents down). */
459     Py_DECREF(PyList_GET_ITEM(heap, pos));
460     PyList_SET_ITEM(heap, pos, newitem);
461     return _siftdownmax(heap, startpos, pos);
462 }
463 
464 static PyObject *
nsmallest(PyObject * self,PyObject * args)465 nsmallest(PyObject *self, PyObject *args)
466 {
467     PyObject *heap=NULL, *elem, *iterable, *los, *it, *oldelem;
468     Py_ssize_t i, n;
469     int cmp;
470 
471     if (!PyArg_ParseTuple(args, "nO:nsmallest", &n, &iterable))
472         return NULL;
473 
474     it = PyObject_GetIter(iterable);
475     if (it == NULL)
476         return NULL;
477 
478     heap = PyList_New(0);
479     if (heap == NULL)
480         goto fail;
481 
482     for (i=0 ; i<n ; i++ ){
483         elem = PyIter_Next(it);
484         if (elem == NULL) {
485             if (PyErr_Occurred())
486                 goto fail;
487             else
488                 goto sortit;
489         }
490         if (PyList_Append(heap, elem) == -1) {
491             Py_DECREF(elem);
492             goto fail;
493         }
494         Py_DECREF(elem);
495     }
496     n = PyList_GET_SIZE(heap);
497     if (n == 0)
498         goto sortit;
499 
500     for (i=n/2-1 ; i>=0 ; i--)
501         if(_siftupmax((PyListObject *)heap, i) == -1)
502             goto fail;
503 
504     los = PyList_GET_ITEM(heap, 0);
505     while (1) {
506         elem = PyIter_Next(it);
507         if (elem == NULL) {
508             if (PyErr_Occurred())
509                 goto fail;
510             else
511                 goto sortit;
512         }
513         cmp = cmp_lt(elem, los);
514         if (cmp == -1) {
515             Py_DECREF(elem);
516             goto fail;
517         }
518         if (cmp == 0) {
519             Py_DECREF(elem);
520             continue;
521         }
522 
523         oldelem = PyList_GET_ITEM(heap, 0);
524         PyList_SET_ITEM(heap, 0, elem);
525         Py_DECREF(oldelem);
526         if (_siftupmax((PyListObject *)heap, 0) == -1)
527             goto fail;
528         los = PyList_GET_ITEM(heap, 0);
529     }
530 
531 sortit:
532     if (PyList_Sort(heap) == -1)
533         goto fail;
534     Py_DECREF(it);
535     return heap;
536 
537 fail:
538     Py_DECREF(it);
539     Py_XDECREF(heap);
540     return NULL;
541 }
542 
543 PyDoc_STRVAR(nsmallest_doc,
544 "Find the n smallest elements in a dataset.\n\
545 \n\
546 Equivalent to:  sorted(iterable)[:n]\n");
547 
548 static PyMethodDef heapq_methods[] = {
549     {"heappush",        (PyCFunction)heappush,
550         METH_VARARGS,           heappush_doc},
551     {"heappushpop",     (PyCFunction)heappushpop,
552         METH_VARARGS,           heappushpop_doc},
553     {"heappop",         (PyCFunction)heappop,
554         METH_O,                 heappop_doc},
555     {"heapreplace",     (PyCFunction)heapreplace,
556         METH_VARARGS,           heapreplace_doc},
557     {"heapify",         (PyCFunction)heapify,
558         METH_O,                 heapify_doc},
559     {"nlargest",        (PyCFunction)nlargest,
560         METH_VARARGS,           nlargest_doc},
561     {"nsmallest",       (PyCFunction)nsmallest,
562         METH_VARARGS,           nsmallest_doc},
563     {NULL,              NULL}           /* sentinel */
564 };
565 
566 PyDoc_STRVAR(module_doc,
567 "Heap queue algorithm (a.k.a. priority queue).\n\
568 \n\
569 Heaps are arrays for which a[k] <= a[2*k+1] and a[k] <= a[2*k+2] for\n\
570 all k, counting elements from 0.  For the sake of comparison,\n\
571 non-existing elements are considered to be infinite.  The interesting\n\
572 property of a heap is that a[0] is always its smallest element.\n\
573 \n\
574 Usage:\n\
575 \n\
576 heap = []            # creates an empty heap\n\
577 heappush(heap, item) # pushes a new item on the heap\n\
578 item = heappop(heap) # pops the smallest item from the heap\n\
579 item = heap[0]       # smallest item on the heap without popping it\n\
580 heapify(x)           # transforms list into a heap, in-place, in linear time\n\
581 item = heapreplace(heap, item) # pops and returns smallest item, and adds\n\
582                                # new item; the heap size is unchanged\n\
583 \n\
584 Our API differs from textbook heap algorithms as follows:\n\
585 \n\
586 - We use 0-based indexing.  This makes the relationship between the\n\
587   index for a node and the indexes for its children slightly less\n\
588   obvious, but is more suitable since Python uses 0-based indexing.\n\
589 \n\
590 - Our heappop() method returns the smallest item, not the largest.\n\
591 \n\
592 These two make it possible to view the heap as a regular Python list\n\
593 without surprises: heap[0] is the smallest item, and heap.sort()\n\
594 maintains the heap invariant!\n");
595 
596 
597 PyDoc_STRVAR(__about__,
598 "Heap queues\n\
599 \n\
600 [explanation by Fran�ois Pinard]\n\
601 \n\
602 Heaps are arrays for which a[k] <= a[2*k+1] and a[k] <= a[2*k+2] for\n\
603 all k, counting elements from 0.  For the sake of comparison,\n\
604 non-existing elements are considered to be infinite.  The interesting\n\
605 property of a heap is that a[0] is always its smallest element.\n"
606 "\n\
607 The strange invariant above is meant to be an efficient memory\n\
608 representation for a tournament.  The numbers below are `k', not a[k]:\n\
609 \n\
610                                    0\n\
611 \n\
612                   1                                 2\n\
613 \n\
614           3               4                5               6\n\
615 \n\
616       7       8       9       10      11      12      13      14\n\
617 \n\
618     15 16   17 18   19 20   21 22   23 24   25 26   27 28   29 30\n\
619 \n\
620 \n\
621 In the tree above, each cell `k' is topping `2*k+1' and `2*k+2'.  In\n\
622 a usual binary tournament we see in sports, each cell is the winner\n\
623 over the two cells it tops, and we can trace the winner down the tree\n\
624 to see all opponents s/he had.  However, in many computer applications\n\
625 of such tournaments, we do not need to trace the history of a winner.\n\
626 To be more memory efficient, when a winner is promoted, we try to\n\
627 replace it by something else at a lower level, and the rule becomes\n\
628 that a cell and the two cells it tops contain three different items,\n\
629 but the top cell \"wins\" over the two topped cells.\n"
630 "\n\
631 If this heap invariant is protected at all time, index 0 is clearly\n\
632 the overall winner.  The simplest algorithmic way to remove it and\n\
633 find the \"next\" winner is to move some loser (let's say cell 30 in the\n\
634 diagram above) into the 0 position, and then percolate this new 0 down\n\
635 the tree, exchanging values, until the invariant is re-established.\n\
636 This is clearly logarithmic on the total number of items in the tree.\n\
637 By iterating over all items, you get an O(n ln n) sort.\n"
638 "\n\
639 A nice feature of this sort is that you can efficiently insert new\n\
640 items while the sort is going on, provided that the inserted items are\n\
641 not \"better\" than the last 0'th element you extracted.  This is\n\
642 especially useful in simulation contexts, where the tree holds all\n\
643 incoming events, and the \"win\" condition means the smallest scheduled\n\
644 time.  When an event schedule other events for execution, they are\n\
645 scheduled into the future, so they can easily go into the heap.  So, a\n\
646 heap is a good structure for implementing schedulers (this is what I\n\
647 used for my MIDI sequencer :-).\n"
648 "\n\
649 Various structures for implementing schedulers have been extensively\n\
650 studied, and heaps are good for this, as they are reasonably speedy,\n\
651 the speed is almost constant, and the worst case is not much different\n\
652 than the average case.  However, there are other representations which\n\
653 are more efficient overall, yet the worst cases might be terrible.\n"
654 "\n\
655 Heaps are also very useful in big disk sorts.  You most probably all\n\
656 know that a big sort implies producing \"runs\" (which are pre-sorted\n\
657 sequences, which size is usually related to the amount of CPU memory),\n\
658 followed by a merging passes for these runs, which merging is often\n\
659 very cleverly organised[1].  It is very important that the initial\n\
660 sort produces the longest runs possible.  Tournaments are a good way\n\
661 to that.  If, using all the memory available to hold a tournament, you\n\
662 replace and percolate items that happen to fit the current run, you'll\n\
663 produce runs which are twice the size of the memory for random input,\n\
664 and much better for input fuzzily ordered.\n"
665 "\n\
666 Moreover, if you output the 0'th item on disk and get an input which\n\
667 may not fit in the current tournament (because the value \"wins\" over\n\
668 the last output value), it cannot fit in the heap, so the size of the\n\
669 heap decreases.  The freed memory could be cleverly reused immediately\n\
670 for progressively building a second heap, which grows at exactly the\n\
671 same rate the first heap is melting.  When the first heap completely\n\
672 vanishes, you switch heaps and start a new run.  Clever and quite\n\
673 effective!\n\
674 \n\
675 In a word, heaps are useful memory structures to know.  I use them in\n\
676 a few applications, and I think it is good to keep a `heap' module\n\
677 around. :-)\n"
678 "\n\
679 --------------------\n\
680 [1] The disk balancing algorithms which are current, nowadays, are\n\
681 more annoying than clever, and this is a consequence of the seeking\n\
682 capabilities of the disks.  On devices which cannot seek, like big\n\
683 tape drives, the story was quite different, and one had to be very\n\
684 clever to ensure (far in advance) that each tape movement will be the\n\
685 most effective possible (that is, will best participate at\n\
686 \"progressing\" the merge).  Some tapes were even able to read\n\
687 backwards, and this was also used to avoid the rewinding time.\n\
688 Believe me, real good tape sorts were quite spectacular to watch!\n\
689 From all times, sorting has always been a Great Art! :-)\n");
690 
691 PyMODINIT_FUNC
init_heapq(void)692 init_heapq(void)
693 {
694     PyObject *m;
695 
696     m = Py_InitModule3("_heapq", heapq_methods, module_doc);
697     if (m == NULL)
698         return;
699     PyModule_AddObject(m, "__about__", PyString_FromString(__about__));
700 }
701 
702