34
30
graph -- sequence of pairs of node->parents_list.
36
The result is a list of node names, such that all parents come before their
32
The result is a list of node names, such that all parents come before
39
35
node identifiers can be any hashable object, and are typically strings.
41
This function has the same purpose as the TopoSorter class, but uses a
42
different algorithm to sort the graph. That means that while both return a
43
list with parents before their child nodes, the exact ordering can be
46
topo_sort is faster when the whole list is needed, while when iterating
47
over a part of the list, TopoSorter.iter_topo_order should be used.
49
kg = _mod_graph.KnownGraph(dict(graph))
37
return TopoSorter(graph).sorted()
53
40
class TopoSorter(object):
55
42
def __init__(self, graph):
56
43
"""Topological sorting of a graph.
58
45
:param graph: sequence of pairs of node_name->parent_names_list.
59
46
i.e. [('C', ['B']), ('B', ['A']), ('A', [])]
60
47
For this input the output from the sort or
61
48
iter_topo_order routines will be:
64
51
node identifiers can be any hashable object, and are typically strings.
66
53
If you have a graph like [('a', ['b']), ('a', ['c'])] this will only use
69
56
The graph is sorted lazily: until you iterate or sort the input is
70
57
not processed other than to create an internal representation.
72
iteration or sorting may raise GraphCycleError if a cycle is present
59
iteration or sorting may raise GraphCycleError if a cycle is present
75
# store a dict of the graph.
62
# a dict of the graph.
76
63
self._graph = dict(graph)
64
self._visitable = set(self._graph)
66
# self._original_graph = dict(graph)
68
# this is a stack storing the depth first search into the graph.
69
self._node_name_stack = []
70
# at each level of 'recursion' we have to check each parent. This
71
# stack stores the parents we have not yet checked for the node at the
72
# matching depth in _node_name_stack
73
self._pending_parents_stack = []
74
# this is a set of the completed nodes for fast checking whether a
75
# parent in a node we are processing on the stack has already been
76
# emitted and thus can be skipped.
77
self._completed_node_names = set()
79
80
"""Sort the graph and return as a list.
81
82
After calling this the sorter is empty and you must create a new one.
83
84
return list(self.iter_topo_order())
95
96
def iter_topo_order(self):
96
97
"""Yield the nodes of the graph in a topological order.
98
99
After finishing iteration the sorter is empty and you cannot continue
102
visitable = set(graph)
104
# this is a stack storing the depth first search into the graph.
105
pending_node_stack = []
106
# at each level of 'recursion' we have to check each parent. This
107
# stack stores the parents we have not yet checked for the node at the
108
# matching depth in pending_node_stack
109
pending_parents_stack = []
111
# this is a set of the completed nodes for fast checking whether a
112
# parent in a node we are processing on the stack has already been
113
# emitted and thus can be skipped.
114
completed_node_names = set()
117
103
# now pick a random node in the source graph, and transfer it to the
118
# top of the depth first search stack of pending nodes.
119
node_name, parents = graph.popitem()
120
pending_node_stack.append(node_name)
121
pending_parents_stack.append(list(parents))
123
# loop until pending_node_stack is empty
124
while pending_node_stack:
125
parents_to_visit = pending_parents_stack[-1]
126
# if there are no parents left, the revision is done
104
# top of the depth first search stack.
105
node_name, parents = self._graph.popitem()
106
self._push_node(node_name, parents)
107
while self._node_name_stack:
108
# loop until this call completes.
109
parents_to_visit = self._pending_parents_stack[-1]
110
# if all parents are done, the revision is done
127
111
if not parents_to_visit:
128
112
# append the revision to the topo sorted list
129
# all the nodes parents have been added to the output,
130
# now we can add it to the output.
131
popped_node = pending_node_stack.pop()
132
pending_parents_stack.pop()
133
completed_node_names.add(popped_node)
113
# all the nodes parents have been added to the output, now
114
# we can add it to the output.
115
yield self._pop_node()
136
# recurse depth first into a single parent
137
next_node_name = parents_to_visit.pop()
139
if next_node_name in completed_node_names:
140
# parent was already completed by a child, skip it.
142
if next_node_name not in visitable:
143
# parent is not a node in the original graph, skip it.
146
# transfer it along with its parents from the source graph
147
# into the top of the current depth first search stack.
149
parents = graph.pop(next_node_name)
151
# if the next node is not in the source graph it has
152
# already been popped from it and placed into the
153
# current search stack (but not completed or we would
154
# have hit the continue 6 lines up). this indicates a
156
raise errors.GraphCycleError(pending_node_stack)
157
pending_node_stack.append(next_node_name)
158
pending_parents_stack.append(list(parents))
117
while self._pending_parents_stack[-1]:
118
# recurse depth first into a single parent
119
next_node_name = self._pending_parents_stack[-1].pop()
120
if next_node_name in self._completed_node_names:
121
# this parent was completed by a child on the
122
# call stack. skip it.
124
if next_node_name not in self._visitable:
126
# otherwise transfer it from the source graph into the
127
# top of the current depth first search stack.
129
parents = self._graph.pop(next_node_name)
131
# if the next node is not in the source graph it has
132
# already been popped from it and placed into the
133
# current search stack (but not completed or we would
134
# have hit the continue 4 lines up.
135
# this indicates a cycle.
136
raise errors.GraphCycleError(self._node_name_stack)
137
self._push_node(next_node_name, parents)
138
# and do not continue processing parents until this 'call'
142
def _push_node(self, node_name, parents):
143
"""Add node_name to the pending node stack.
145
Names in this stack will get emitted into the output as they are popped
148
self._node_name_stack.append(node_name)
149
self._pending_parents_stack.append(list(parents))
152
"""Pop the top node off the stack
154
The node is appended to the sorted output.
156
# we are returning from the flattened call frame:
157
# pop off the local variables
158
node_name = self._node_name_stack.pop()
159
self._pending_parents_stack.pop()
161
self._completed_node_names.add(node_name)
161
165
def merge_sort(graph, branch_tip, mainline_revisions=None, generate_revno=False):
162
166
"""Topological sort a graph which groups merges.
164
168
:param graph: sequence of pairs of node->parents_list.
165
:param branch_tip: the tip of the branch to graph. Revisions not
169
:param branch_tip: the tip of the branch to graph. Revisions not
166
170
reachable from branch_tip are not included in the
168
172
:param mainline_revisions: If not None this forces a mainline to be
228
232
The result is a list sorted so that all parents come before
229
233
their children. Each element of the list is a tuple containing:
230
234
(sequence_number, node_name, merge_depth, end_of_merge)
231
* sequence_number: The sequence of this row in the output. Useful for
235
* sequence_number: The sequence of this row in the output. Useful for
233
237
* node_name: The node name: opaque text to the merge routine.
234
238
* merge_depth: How many levels of merging deep this node has been
236
240
* revno_sequence: When requested this field provides a sequence of
237
241
revision numbers for all revisions. The format is:
238
(REVNO, BRANCHNUM, BRANCHREVNO). BRANCHNUM is the number of the
242
REVNO[[.BRANCHREVNO.REVNO] ...]. BRANCHREVNO is the number of the
239
243
branch that the revno is on. From left to right the REVNO numbers
240
244
are the sequence numbers within that branch of the revision.
241
245
For instance, the graph {A:[], B:['A'], C:['A', 'B']} will get
281
285
C is the end of a cluster due to rule 1.
282
D is not the end of a cluster from rule 1, but is from rule 2: E
286
D is not the end of a cluster from rule 1, but is from rule 2: E
283
287
is not its left most ancestor
284
288
E is the end of a cluster due to rule 1
285
289
F might be but we need more data.
287
291
we show connecting lines to a parent when:
288
292
- The parent is the start of a merge within this cluster.
289
That is, the merge was not done to the mainline before this cluster
293
That is, the merge was not done to the mainline before this cluster
290
294
was merged to the mainline.
291
295
This can be detected thus:
292
* The parent has a higher merge depth and is the next revision in
296
* The parent has a higher merge depth and is the next revision in
295
299
The next revision in the list constraint is needed for this case:
297
B 1 [C, F] # we do not want to show a line to F which is depth 2
301
B 1 [C, F] # we do not want to show a line to F which is depth 2
299
C 1 [H] # note that this is a long line to show back to the
303
C 1 [H] # note that this is a long line to show back to the
300
304
ancestor - see the end of merge rules.
364
362
self._original_graph = dict(self._graph.items())
365
363
# we need to know the revision numbers of revisions to determine
366
364
# the revision numbers of their descendants
367
# this is a graph from node to [revno_tuple, first_child]
368
# where first_child is True if no other children have seen this node
365
# this is a graph from node to [revno_tuple, sequence_number]
366
# where sequence is the number of branches made from the node,
369
367
# and revno_tuple is the tuple that was assigned to the node.
370
368
# we dont know revnos to start with, so we start it seeded with
372
self._revnos = dict((revision, [None, True])
373
for revision in self._graph)
374
# Each mainline revision counts how many child branches have spawned from it.
375
self._revno_to_branch_count = {}
370
self._revnos = dict((revision, [None, 0]) for revision in self._graph)
371
# the global implicit root node has revno 0, but we need to know
372
# the sequence number for it too:
373
self._root_sequence = 0
377
375
# this is a stack storing the depth first search into the graph.
378
376
self._node_name_stack = []
379
377
# at each level of recursion we need the merge depth this node is at:
380
378
self._node_merge_depth_stack = []
381
379
# at each level of 'recursion' we have to check each parent. This
382
# stack stores the parents we have not yet checked for the node at the
380
# stack stores the parents we have not yet checked for the node at the
383
381
# matching depth in _node_name_stack
384
382
self._pending_parents_stack = []
385
383
# When we first look at a node we assign it a seqence number from its
386
384
# leftmost parent.
387
self._first_child_stack = []
385
self._assigned_sequence_stack = []
388
386
# this is a set of the nodes who have been completely analysed for fast
389
387
# membership checking
390
388
self._completed_node_names = set()
400
# the scheduling order is: F, E, D, C, B, A
398
# the scheduling order is: F, E, D, C, B, A
401
399
# that is - 'left subtree, right subtree, node'
402
400
# which would mean that when we schedule A we can emit the entire tree.
403
401
self._scheduled_nodes = []
404
# This records for each node when we have processed its left most
402
# This records for each node when we have processed its left most
405
403
# unmerged subtree. After this subtree is scheduled, all other subtrees
406
404
# have their merge depth increased by one from this nodes merge depth.
407
405
# it contains tuples - name, merge_depth
408
406
self._left_subtree_pushed_stack = []
410
408
# seed the search with the tip of the branch
411
if (branch_tip is not None and
412
branch_tip != _mod_revision.NULL_REVISION and
413
branch_tip != (_mod_revision.NULL_REVISION,)):
409
if branch_tip is not None:
414
410
parents = self._graph.pop(branch_tip)
415
411
self._push_node(branch_tip, 0, parents)
417
413
def sorted(self):
418
414
"""Sort the graph and return as a list.
420
416
After calling this the sorter is empty and you must create a new one.
422
418
return list(self.iter_topo_order())
424
420
def iter_topo_order(self):
425
421
"""Yield the nodes of the graph in a topological order.
427
423
After finishing iteration the sorter is empty and you cannot continue
458
455
node_merge_depth_stack_append(merge_depth)
459
456
left_subtree_pushed_stack_append(False)
460
457
pending_parents_stack_append(list(parents))
461
# as we push it, check if it is the first child
458
# as we push it, assign it a sequence number against its parent:
459
parents = original_graph[node_name]
464
461
# node has parents, assign from the left most parent.
466
parent_info = revnos[parents[0]]
468
# Left-hand parent is a ghost, consider it not to exist
470
if parent_info is not None:
471
first_child = parent_info[1]
472
parent_info[1] = False
462
parent_revno = revnos[parents[0]]
463
sequence = parent_revno[1]
474
# We don't use the same algorithm here, but we need to keep the
477
first_child_stack_append(first_child)
466
# no parents, use the root sequence
467
sequence = self._root_sequence
468
self._root_sequence +=1
469
assigned_sequence_stack_append(sequence)
479
471
def pop_node(node_name_stack_pop=node_name_stack.pop,
480
472
node_merge_depth_stack_pop=node_merge_depth_stack.pop,
481
first_child_stack_pop=self._first_child_stack.pop,
473
assigned_sequence_stack_pop=self._assigned_sequence_stack.pop,
482
474
left_subtree_pushed_stack_pop=left_subtree_pushed_stack.pop,
483
475
pending_parents_stack_pop=pending_parents_stack.pop,
484
476
original_graph=self._original_graph,
485
477
revnos=self._revnos,
486
478
completed_node_names_add=self._completed_node_names.add,
487
479
scheduled_nodes_append=scheduled_nodes.append,
488
revno_to_branch_count=self._revno_to_branch_count,
490
481
"""Pop the top node off the stack
495
486
# pop off the local variables
496
487
node_name = node_name_stack_pop()
497
488
merge_depth = node_merge_depth_stack_pop()
498
first_child = first_child_stack_pop()
489
sequence = assigned_sequence_stack_pop()
499
490
# remove this node from the pending lists:
500
491
left_subtree_pushed_stack_pop()
501
492
pending_parents_stack_pop()
503
494
parents = original_graph[node_name]
506
496
# node has parents, assign from the left most parent.
508
parent_revno = revnos[parents[0]][0]
510
# left-hand parent is a ghost, treat it as not existing
512
if parent_revno is not None:
497
parent_revno = revnos[parents[0]]
514
499
# not the first child, make a new branch
515
base_revno = parent_revno[0]
516
branch_count = revno_to_branch_count.get(base_revno, 0)
518
revno_to_branch_count[base_revno] = branch_count
519
revno = (parent_revno[0], branch_count, 1)
520
# revno = (parent_revno[0], branch_count, parent_revno[-1]+1)
500
revno = parent_revno[0] + (sequence, 1)
522
# as the first child, we just increase the final revision
524
revno = parent_revno[:-1] + (parent_revno[-1] + 1,)
502
# increment the sequence number within the branch
503
revno = parent_revno[0][:-1] + (parent_revno[0][-1] + 1,)
526
505
# no parents, use the root sequence
527
root_count = revno_to_branch_count.get(0, -1)
530
revno = (0, root_count, 1)
507
# make a parallel import revision number
508
revno = (0, sequence, 1)
533
revno_to_branch_count[0] = root_count
535
512
# store the revno for this node for future reference
536
513
revnos[node_name][0] = revno
553
530
if not left_subtree_pushed_stack[-1]:
554
531
# recurse depth first into the primary parent
555
532
next_node_name = pending_parents_stack[-1].pop(0)
556
is_left_subtree = True
557
left_subtree_pushed_stack[-1] = True
559
534
# place any merges in right-to-left order for scheduling
560
535
# which gives us left-to-right order after we reverse
561
# the scheduled queue. XXX: This has the effect of
536
# the scheduled queue. XXX: This has the effect of
562
537
# allocating common-new revisions to the right-most
563
# subtree rather than the left most, which will
538
# subtree rather than the left most, which will
564
539
# display nicely (you get smaller trees at the top
565
540
# of the combined merge).
566
541
next_node_name = pending_parents_stack[-1].pop()
567
is_left_subtree = False
568
542
if next_node_name in completed_node_names:
569
543
# this parent was completed by a child on the
570
544
# call stack. skip it.
639
610
self._node_merge_depth_stack.append(merge_depth)
640
611
self._left_subtree_pushed_stack.append(False)
641
612
self._pending_parents_stack.append(list(parents))
642
# as we push it, figure out if this is the first child
613
# as we push it, assign it a sequence number against its parent:
614
parents = self._original_graph[node_name]
645
616
# node has parents, assign from the left most parent.
647
parent_info = self._revnos[parents[0]]
649
# Left-hand parent is a ghost, consider it not to exist
651
if parent_info is not None:
652
first_child = parent_info[1]
653
parent_info[1] = False
617
parent_revno = self._revnos[parents[0]]
618
sequence = parent_revno[1]
655
# We don't use the same algorithm here, but we need to keep the
658
self._first_child_stack.append(first_child)
621
# no parents, use the root sequence
622
sequence = self._root_sequence
623
self._root_sequence +=1
624
self._assigned_sequence_stack.append(sequence)
660
626
def _pop_node(self):
661
"""Pop the top node off the stack
627
"""Pop the top node off the stack
663
629
The node is appended to the sorted output.
666
632
# pop off the local variables
667
633
node_name = self._node_name_stack.pop()
668
634
merge_depth = self._node_merge_depth_stack.pop()
669
first_child = self._first_child_stack.pop()
635
sequence = self._assigned_sequence_stack.pop()
670
636
# remove this node from the pending lists:
671
637
self._left_subtree_pushed_stack.pop()
672
638
self._pending_parents_stack.pop()
674
640
parents = self._original_graph[node_name]
677
642
# node has parents, assign from the left most parent.
679
parent_revno = self._revnos[parents[0]][0]
681
# left-hand parent is a ghost, treat it as not existing
683
if parent_revno is not None:
643
parent_revno = self._revnos[parents[0]]
685
645
# not the first child, make a new branch
686
base_revno = parent_revno[0]
687
branch_count = self._revno_to_branch_count.get(base_revno, 0)
689
self._revno_to_branch_count[base_revno] = branch_count
690
revno = (parent_revno[0], branch_count, 1)
691
# revno = (parent_revno[0], branch_count, parent_revno[-1]+1)
646
revno = parent_revno[0] + (sequence, 1)
693
# as the first child, we just increase the final revision
695
revno = parent_revno[:-1] + (parent_revno[-1] + 1,)
648
# increment the sequence number within the branch
649
revno = parent_revno[0][:-1] + (parent_revno[0][-1] + 1,)
697
651
# no parents, use the root sequence
698
root_count = self._revno_to_branch_count.get(0, 0)
699
root_count = self._revno_to_branch_count.get(0, -1)
702
revno = (0, root_count, 1)
653
# make a parallel import revision number
654
revno = (0, sequence, 1)
705
self._revno_to_branch_count[0] = root_count
707
658
# store the revno for this node for future reference
708
659
self._revnos[node_name][0] = revno
added
removed
bzrlib/tsort.py
