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)
65
# self._original_graph = dict(graph)
67
# this is a stack storing the depth first search into the graph.
68
self._node_name_stack = []
69
# at each level of 'recursion' we have to check each parent. This
70
# stack stores the parents we have not yet checked for the node at the
71
# matching depth in _node_name_stack
72
self._pending_parents_stack = []
73
# this is a set of the completed nodes for fast checking whether a
74
# parent in a node we are processing on the stack has already been
75
# emitted and thus can be skipped.
76
self._completed_node_names = set()
79
79
"""Sort the graph and return as a list.
81
81
After calling this the sorter is empty and you must create a new one.
83
83
return list(self.iter_topo_order())
95
95
def iter_topo_order(self):
96
96
"""Yield the nodes of the graph in a topological order.
98
98
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
102
# 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
103
# top of the depth first search stack.
104
node_name, parents = self._graph.popitem()
105
self._push_node(node_name, parents)
106
while self._node_name_stack:
107
# loop until this call completes.
108
parents_to_visit = self._pending_parents_stack[-1]
109
# if all parents are done, the revision is done
127
110
if not parents_to_visit:
128
111
# 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)
112
# all the nodes parents have been added to the output, now
113
# we can add it to the output.
114
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))
116
while self._pending_parents_stack[-1]:
117
# recurse depth first into a single parent
118
next_node_name = self._pending_parents_stack[-1].pop()
119
if next_node_name in self._completed_node_names:
120
# this parent was completed by a child on the
121
# call stack. skip it.
123
# otherwise transfer it from the source graph into the
124
# top of the current depth first search stack.
126
parents = self._graph.pop(next_node_name)
128
# if the next node is not in the source graph it has
129
# already been popped from it and placed into the
130
# current search stack (but not completed or we would
131
# have hit the continue 4 lines up.
132
# this indicates a cycle.
133
raise errors.GraphCycleError(self._node_name_stack)
134
self._push_node(next_node_name, parents)
135
# and do not continue processing parents until this 'call'
139
def _push_node(self, node_name, parents):
140
"""Add node_name to the pending node stack.
142
Names in this stack will get emitted into the output as they are popped
145
self._node_name_stack.append(node_name)
146
self._pending_parents_stack.append(list(parents))
149
"""Pop the top node off the stack
151
The node is appended to the sorted output.
153
# we are returning from the flattened call frame:
154
# pop off the local variables
155
node_name = self._node_name_stack.pop()
156
self._pending_parents_stack.pop()
158
self._completed_node_names.add(node_name)
161
162
def merge_sort(graph, branch_tip, mainline_revisions=None, generate_revno=False):
162
163
"""Topological sort a graph which groups merges.
164
165
:param graph: sequence of pairs of node->parents_list.
165
:param branch_tip: the tip of the branch to graph. Revisions not
166
:param branch_tip: the tip of the branch to graph. Revisions not
166
167
reachable from branch_tip are not included in the
168
169
:param mainline_revisions: If not None this forces a mainline to be
229
229
The result is a list sorted so that all parents come before
230
230
their children. Each element of the list is a tuple containing:
231
231
(sequence_number, node_name, merge_depth, end_of_merge)
232
* sequence_number: The sequence of this row in the output. Useful for
232
* sequence_number: The sequence of this row in the output. Useful for
234
234
* node_name: The node name: opaque text to the merge routine.
235
235
* merge_depth: How many levels of merging deep this node has been
237
237
* revno_sequence: When requested this field provides a sequence of
238
238
revision numbers for all revisions. The format is:
239
(REVNO, BRANCHNUM, BRANCHREVNO). BRANCHNUM is the number of the
239
REVNO[[.BRANCHREVNO.REVNO] ...]. BRANCHREVNO is the number of the
240
240
branch that the revno is on. From left to right the REVNO numbers
241
241
are the sequence numbers within that branch of the revision.
242
242
For instance, the graph {A:[], B:['A'], C:['A', 'B']} will get
282
282
C is the end of a cluster due to rule 1.
283
D is not the end of a cluster from rule 1, but is from rule 2: E
283
D is not the end of a cluster from rule 1, but is from rule 2: E
284
284
is not its left most ancestor
285
285
E is the end of a cluster due to rule 1
286
286
F might be but we need more data.
288
288
we show connecting lines to a parent when:
289
289
- The parent is the start of a merge within this cluster.
290
That is, the merge was not done to the mainline before this cluster
290
That is, the merge was not done to the mainline before this cluster
291
291
was merged to the mainline.
292
292
This can be detected thus:
293
* The parent has a higher merge depth and is the next revision in
293
* The parent has a higher merge depth and is the next revision in
296
296
The next revision in the list constraint is needed for this case:
298
B 1 [C, F] # we do not want to show a line to F which is depth 2
298
B 1 [C, F] # we do not want to show a line to F which is depth 2
300
C 1 [H] # note that this is a long line to show back to the
300
C 1 [H] # note that this is a long line to show back to the
301
301
ancestor - see the end of merge rules.
365
359
self._original_graph = dict(self._graph.items())
366
360
# we need to know the revision numbers of revisions to determine
367
361
# the revision numbers of their descendants
368
# this is a graph from node to [revno_tuple, first_child]
369
# where first_child is True if no other children have seen this node
362
# this is a graph from node to [revno_tuple, sequence_number]
363
# where sequence is the number of branches made from the node,
370
364
# and revno_tuple is the tuple that was assigned to the node.
371
365
# we dont know revnos to start with, so we start it seeded with
373
self._revnos = dict((revision, [None, True])
374
for revision in self._graph)
375
# Each mainline revision counts how many child branches have spawned from it.
376
self._revno_to_branch_count = {}
367
self._revnos = dict((revision, [None, 0]) for revision in self._graph)
368
# the global implicit root node has revno 0, but we need to know
369
# the sequence number for it too:
370
self._root_sequence = 0
378
372
# this is a stack storing the depth first search into the graph.
379
373
self._node_name_stack = []
380
374
# at each level of recursion we need the merge depth this node is at:
381
375
self._node_merge_depth_stack = []
382
376
# at each level of 'recursion' we have to check each parent. This
383
# stack stores the parents we have not yet checked for the node at the
377
# stack stores the parents we have not yet checked for the node at the
384
378
# matching depth in _node_name_stack
385
379
self._pending_parents_stack = []
386
380
# When we first look at a node we assign it a seqence number from its
387
381
# leftmost parent.
388
self._first_child_stack = []
382
self._assigned_sequence_stack = []
389
383
# this is a set of the nodes who have been completely analysed for fast
390
384
# membership checking
391
385
self._completed_node_names = set()
401
# the scheduling order is: F, E, D, C, B, A
395
# the scheduling order is: F, E, D, C, B, A
402
396
# that is - 'left subtree, right subtree, node'
403
397
# which would mean that when we schedule A we can emit the entire tree.
404
398
self._scheduled_nodes = []
405
# This records for each node when we have processed its left most
399
# This records for each node when we have processed its left most
406
400
# unmerged subtree. After this subtree is scheduled, all other subtrees
407
401
# have their merge depth increased by one from this nodes merge depth.
408
402
# it contains tuples - name, merge_depth
409
403
self._left_subtree_pushed_stack = []
411
405
# seed the search with the tip of the branch
412
if (branch_tip is not None and
413
branch_tip != _mod_revision.NULL_REVISION and
414
branch_tip != (_mod_revision.NULL_REVISION,)):
406
if branch_tip is not None:
415
407
parents = self._graph.pop(branch_tip)
416
408
self._push_node(branch_tip, 0, parents)
418
410
def sorted(self):
419
411
"""Sort the graph and return as a list.
421
413
After calling this the sorter is empty and you must create a new one.
423
415
return list(self.iter_topo_order())
425
417
def iter_topo_order(self):
426
418
"""Yield the nodes of the graph in a topological order.
428
420
After finishing iteration the sorter is empty and you cannot continue
459
452
node_merge_depth_stack_append(merge_depth)
460
453
left_subtree_pushed_stack_append(False)
461
454
pending_parents_stack_append(list(parents))
462
# as we push it, check if it is the first child
455
# as we push it, assign it a sequence number against its parent:
456
parents = original_graph[node_name]
465
458
# node has parents, assign from the left most parent.
467
parent_info = revnos[parents[0]]
469
# Left-hand parent is a ghost, consider it not to exist
471
if parent_info is not None:
472
first_child = parent_info[1]
473
parent_info[1] = False
459
parent_revno = revnos[parents[0]]
460
sequence = parent_revno[1]
475
# We don't use the same algorithm here, but we need to keep the
478
first_child_stack_append(first_child)
463
# no parents, use the root sequence
464
sequence = self._root_sequence
465
self._root_sequence +=1
466
assigned_sequence_stack_append(sequence)
480
468
def pop_node(node_name_stack_pop=node_name_stack.pop,
481
469
node_merge_depth_stack_pop=node_merge_depth_stack.pop,
482
first_child_stack_pop=self._first_child_stack.pop,
470
assigned_sequence_stack_pop=self._assigned_sequence_stack.pop,
483
471
left_subtree_pushed_stack_pop=left_subtree_pushed_stack.pop,
484
472
pending_parents_stack_pop=pending_parents_stack.pop,
485
473
original_graph=self._original_graph,
486
474
revnos=self._revnos,
487
475
completed_node_names_add=self._completed_node_names.add,
488
476
scheduled_nodes_append=scheduled_nodes.append,
489
revno_to_branch_count=self._revno_to_branch_count,
491
478
"""Pop the top node off the stack
496
483
# pop off the local variables
497
484
node_name = node_name_stack_pop()
498
485
merge_depth = node_merge_depth_stack_pop()
499
first_child = first_child_stack_pop()
486
sequence = assigned_sequence_stack_pop()
500
487
# remove this node from the pending lists:
501
488
left_subtree_pushed_stack_pop()
502
489
pending_parents_stack_pop()
504
491
parents = original_graph[node_name]
507
493
# node has parents, assign from the left most parent.
509
parent_revno = revnos[parents[0]][0]
511
# left-hand parent is a ghost, treat it as not existing
513
if parent_revno is not None:
494
parent_revno = revnos[parents[0]]
515
496
# not the first child, make a new branch
516
base_revno = parent_revno[0]
517
branch_count = revno_to_branch_count.get(base_revno, 0)
519
revno_to_branch_count[base_revno] = branch_count
520
revno = (parent_revno[0], branch_count, 1)
521
# revno = (parent_revno[0], branch_count, parent_revno[-1]+1)
497
revno = parent_revno[0] + (sequence, 1)
523
# as the first child, we just increase the final revision
525
revno = parent_revno[:-1] + (parent_revno[-1] + 1,)
499
# increment the sequence number within the branch
500
revno = parent_revno[0][:-1] + (parent_revno[0][-1] + 1,)
527
502
# no parents, use the root sequence
528
root_count = revno_to_branch_count.get(0, -1)
531
revno = (0, root_count, 1)
504
# make a parallel import revision number
505
revno = (0, sequence, 1)
534
revno_to_branch_count[0] = root_count
536
509
# store the revno for this node for future reference
537
510
revnos[node_name][0] = revno
554
527
if not left_subtree_pushed_stack[-1]:
555
528
# recurse depth first into the primary parent
556
529
next_node_name = pending_parents_stack[-1].pop(0)
557
is_left_subtree = True
558
left_subtree_pushed_stack[-1] = True
560
531
# place any merges in right-to-left order for scheduling
561
532
# which gives us left-to-right order after we reverse
562
# the scheduled queue. XXX: This has the effect of
533
# the scheduled queue. XXX: This has the effect of
563
534
# allocating common-new revisions to the right-most
564
# subtree rather than the left most, which will
535
# subtree rather than the left most, which will
565
536
# display nicely (you get smaller trees at the top
566
537
# of the combined merge).
567
538
next_node_name = pending_parents_stack[-1].pop()
568
is_left_subtree = False
569
539
if next_node_name in completed_node_names:
570
540
# this parent was completed by a child on the
571
541
# call stack. skip it.
640
607
self._node_merge_depth_stack.append(merge_depth)
641
608
self._left_subtree_pushed_stack.append(False)
642
609
self._pending_parents_stack.append(list(parents))
643
# as we push it, figure out if this is the first child
610
# as we push it, assign it a sequence number against its parent:
611
parents = self._original_graph[node_name]
646
613
# node has parents, assign from the left most parent.
648
parent_info = self._revnos[parents[0]]
650
# Left-hand parent is a ghost, consider it not to exist
652
if parent_info is not None:
653
first_child = parent_info[1]
654
parent_info[1] = False
614
parent_revno = self._revnos[parents[0]]
615
sequence = parent_revno[1]
656
# We don't use the same algorithm here, but we need to keep the
659
self._first_child_stack.append(first_child)
618
# no parents, use the root sequence
619
sequence = self._root_sequence
620
self._root_sequence +=1
621
self._assigned_sequence_stack.append(sequence)
661
623
def _pop_node(self):
662
"""Pop the top node off the stack
624
"""Pop the top node off the stack
664
626
The node is appended to the sorted output.
667
629
# pop off the local variables
668
630
node_name = self._node_name_stack.pop()
669
631
merge_depth = self._node_merge_depth_stack.pop()
670
first_child = self._first_child_stack.pop()
632
sequence = self._assigned_sequence_stack.pop()
671
633
# remove this node from the pending lists:
672
634
self._left_subtree_pushed_stack.pop()
673
635
self._pending_parents_stack.pop()
675
637
parents = self._original_graph[node_name]
678
639
# node has parents, assign from the left most parent.
680
parent_revno = self._revnos[parents[0]][0]
682
# left-hand parent is a ghost, treat it as not existing
684
if parent_revno is not None:
640
parent_revno = self._revnos[parents[0]]
686
642
# not the first child, make a new branch
687
base_revno = parent_revno[0]
688
branch_count = self._revno_to_branch_count.get(base_revno, 0)
690
self._revno_to_branch_count[base_revno] = branch_count
691
revno = (parent_revno[0], branch_count, 1)
692
# revno = (parent_revno[0], branch_count, parent_revno[-1]+1)
643
revno = parent_revno[0] + (sequence, 1)
694
# as the first child, we just increase the final revision
696
revno = parent_revno[:-1] + (parent_revno[-1] + 1,)
645
# increment the sequence number within the branch
646
revno = parent_revno[0][:-1] + (parent_revno[0][-1] + 1,)
698
648
# no parents, use the root sequence
699
root_count = self._revno_to_branch_count.get(0, 0)
700
root_count = self._revno_to_branch_count.get(0, -1)
703
revno = (0, root_count, 1)
650
# make a parallel import revision number
651
revno = (0, sequence, 1)
706
self._revno_to_branch_count[0] = root_count
708
655
# store the revno for this node for future reference
709
656
self._revnos[node_name][0] = revno
added
removed
bzrlib/tsort.py
