35
30
graph -- sequence of pairs of node->parents_list.
37
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
40
35
node identifiers can be any hashable object, and are typically strings.
42
This function has the same purpose as the TopoSorter class, but uses a
43
different algorithm to sort the graph. That means that while both return a
44
list with parents before their child nodes, the exact ordering can be
47
topo_sort is faster when the whole list is needed, while when iterating
48
over a part of the list, TopoSorter.iter_topo_order should be used.
50
kg = _mod_graph.KnownGraph(dict(graph))
37
return TopoSorter(graph).sorted()
54
40
class TopoSorter(object):
56
42
def __init__(self, graph):
57
43
"""Topological sorting of a graph.
59
45
:param graph: sequence of pairs of node_name->parent_names_list.
60
46
i.e. [('C', ['B']), ('B', ['A']), ('A', [])]
61
47
For this input the output from the sort or
62
48
iter_topo_order routines will be:
65
51
node identifiers can be any hashable object, and are typically strings.
67
53
If you have a graph like [('a', ['b']), ('a', ['c'])] this will only use
70
56
The graph is sorted lazily: until you iterate or sort the input is
71
57
not processed other than to create an internal representation.
73
iteration or sorting may raise GraphCycleError if a cycle is present
59
iteration or sorting may raise GraphCycleError if a cycle is present
76
# store a dict of the graph.
62
# a dict of the graph.
77
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()
80
80
"""Sort the graph and return as a list.
82
82
After calling this the sorter is empty and you must create a new one.
84
84
return list(self.iter_topo_order())
96
96
def iter_topo_order(self):
97
97
"""Yield the nodes of the graph in a topological order.
99
99
After finishing iteration the sorter is empty and you cannot continue
103
visitable = set(graph)
105
# this is a stack storing the depth first search into the graph.
106
pending_node_stack = []
107
# at each level of 'recursion' we have to check each parent. This
108
# stack stores the parents we have not yet checked for the node at the
109
# matching depth in pending_node_stack
110
pending_parents_stack = []
112
# this is a set of the completed nodes for fast checking whether a
113
# parent in a node we are processing on the stack has already been
114
# emitted and thus can be skipped.
115
completed_node_names = set()
118
103
# now pick a random node in the source graph, and transfer it to the
119
# top of the depth first search stack of pending nodes.
120
node_name, parents = graph.popitem()
121
pending_node_stack.append(node_name)
122
pending_parents_stack.append(list(parents))
124
# loop until pending_node_stack is empty
125
while pending_node_stack:
126
parents_to_visit = pending_parents_stack[-1]
127
# 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
128
111
if not parents_to_visit:
129
112
# append the revision to the topo sorted list
130
# all the nodes parents have been added to the output,
131
# now we can add it to the output.
132
popped_node = pending_node_stack.pop()
133
pending_parents_stack.pop()
134
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()
137
# recurse depth first into a single parent
138
next_node_name = parents_to_visit.pop()
140
if next_node_name in completed_node_names:
141
# parent was already completed by a child, skip it.
143
if next_node_name not in visitable:
144
# parent is not a node in the original graph, skip it.
147
# transfer it along with its parents from the source graph
148
# into the top of the current depth first search stack.
150
parents = graph.pop(next_node_name)
152
# if the next node is not in the source graph it has
153
# already been popped from it and placed into the
154
# current search stack (but not completed or we would
155
# have hit the continue 6 lines up). this indicates a
157
raise errors.GraphCycleError(pending_node_stack)
158
pending_node_stack.append(next_node_name)
159
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)
162
165
def merge_sort(graph, branch_tip, mainline_revisions=None, generate_revno=False):
163
166
"""Topological sort a graph which groups merges.
165
168
:param graph: sequence of pairs of node->parents_list.
166
: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
167
170
reachable from branch_tip are not included in the
169
172
:param mainline_revisions: If not None this forces a mainline to be
230
232
The result is a list sorted so that all parents come before
231
233
their children. Each element of the list is a tuple containing:
232
234
(sequence_number, node_name, merge_depth, end_of_merge)
233
* 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
235
237
* node_name: The node name: opaque text to the merge routine.
236
238
* merge_depth: How many levels of merging deep this node has been
238
240
* revno_sequence: When requested this field provides a sequence of
239
241
revision numbers for all revisions. The format is:
240
(REVNO, BRANCHNUM, BRANCHREVNO). BRANCHNUM is the number of the
242
REVNO[[.BRANCHREVNO.REVNO] ...]. BRANCHREVNO is the number of the
241
243
branch that the revno is on. From left to right the REVNO numbers
242
244
are the sequence numbers within that branch of the revision.
243
245
For instance, the graph {A:[], B:['A'], C:['A', 'B']} will get
283
285
C is the end of a cluster due to rule 1.
284
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
285
287
is not its left most ancestor
286
288
E is the end of a cluster due to rule 1
287
289
F might be but we need more data.
289
291
we show connecting lines to a parent when:
290
292
- The parent is the start of a merge within this cluster.
291
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
292
294
was merged to the mainline.
293
295
This can be detected thus:
294
* 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
297
299
The next revision in the list constraint is needed for this case:
299
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
301
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
302
304
ancestor - see the end of merge rules.
366
362
self._original_graph = dict(self._graph.items())
367
363
# we need to know the revision numbers of revisions to determine
368
364
# the revision numbers of their descendants
369
# this is a graph from node to [revno_tuple, first_child]
370
# 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,
371
367
# and revno_tuple is the tuple that was assigned to the node.
372
368
# we dont know revnos to start with, so we start it seeded with
374
self._revnos = dict((revision, [None, True])
375
for revision in self._graph)
376
# Each mainline revision counts how many child branches have spawned from it.
377
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
379
375
# this is a stack storing the depth first search into the graph.
380
376
self._node_name_stack = []
381
377
# at each level of recursion we need the merge depth this node is at:
382
378
self._node_merge_depth_stack = []
383
379
# at each level of 'recursion' we have to check each parent. This
384
# 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
385
381
# matching depth in _node_name_stack
386
382
self._pending_parents_stack = []
387
383
# When we first look at a node we assign it a seqence number from its
388
384
# leftmost parent.
389
self._first_child_stack = []
385
self._assigned_sequence_stack = []
390
386
# this is a set of the nodes who have been completely analysed for fast
391
387
# membership checking
392
388
self._completed_node_names = set()
402
# the scheduling order is: F, E, D, C, B, A
398
# the scheduling order is: F, E, D, C, B, A
403
399
# that is - 'left subtree, right subtree, node'
404
400
# which would mean that when we schedule A we can emit the entire tree.
405
401
self._scheduled_nodes = []
406
# 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
407
403
# unmerged subtree. After this subtree is scheduled, all other subtrees
408
404
# have their merge depth increased by one from this nodes merge depth.
409
405
# it contains tuples - name, merge_depth
410
406
self._left_subtree_pushed_stack = []
412
408
# seed the search with the tip of the branch
413
if (branch_tip is not None and
414
branch_tip != _mod_revision.NULL_REVISION and
415
branch_tip != (_mod_revision.NULL_REVISION,)):
409
if branch_tip is not None:
416
410
parents = self._graph.pop(branch_tip)
417
411
self._push_node(branch_tip, 0, parents)
419
413
def sorted(self):
420
414
"""Sort the graph and return as a list.
422
416
After calling this the sorter is empty and you must create a new one.
424
418
return list(self.iter_topo_order())
426
420
def iter_topo_order(self):
427
421
"""Yield the nodes of the graph in a topological order.
429
423
After finishing iteration the sorter is empty and you cannot continue
460
455
node_merge_depth_stack_append(merge_depth)
461
456
left_subtree_pushed_stack_append(False)
462
457
pending_parents_stack_append(list(parents))
463
# 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]
466
461
# node has parents, assign from the left most parent.
468
parent_info = revnos[parents[0]]
470
# Left-hand parent is a ghost, consider it not to exist
472
if parent_info is not None:
473
first_child = parent_info[1]
474
parent_info[1] = False
462
parent_revno = revnos[parents[0]]
463
sequence = parent_revno[1]
476
# We don't use the same algorithm here, but we need to keep the
479
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)
481
471
def pop_node(node_name_stack_pop=node_name_stack.pop,
482
472
node_merge_depth_stack_pop=node_merge_depth_stack.pop,
483
first_child_stack_pop=self._first_child_stack.pop,
473
assigned_sequence_stack_pop=self._assigned_sequence_stack.pop,
484
474
left_subtree_pushed_stack_pop=left_subtree_pushed_stack.pop,
485
475
pending_parents_stack_pop=pending_parents_stack.pop,
486
476
original_graph=self._original_graph,
487
477
revnos=self._revnos,
488
478
completed_node_names_add=self._completed_node_names.add,
489
479
scheduled_nodes_append=scheduled_nodes.append,
490
revno_to_branch_count=self._revno_to_branch_count,
492
481
"""Pop the top node off the stack
497
486
# pop off the local variables
498
487
node_name = node_name_stack_pop()
499
488
merge_depth = node_merge_depth_stack_pop()
500
first_child = first_child_stack_pop()
489
sequence = assigned_sequence_stack_pop()
501
490
# remove this node from the pending lists:
502
491
left_subtree_pushed_stack_pop()
503
492
pending_parents_stack_pop()
505
494
parents = original_graph[node_name]
508
496
# node has parents, assign from the left most parent.
510
parent_revno = revnos[parents[0]][0]
512
# left-hand parent is a ghost, treat it as not existing
514
if parent_revno is not None:
497
parent_revno = revnos[parents[0]]
516
499
# not the first child, make a new branch
517
base_revno = parent_revno[0]
518
branch_count = revno_to_branch_count.get(base_revno, 0)
520
revno_to_branch_count[base_revno] = branch_count
521
revno = (parent_revno[0], branch_count, 1)
522
# revno = (parent_revno[0], branch_count, parent_revno[-1]+1)
500
revno = parent_revno[0] + (sequence, 1)
524
# as the first child, we just increase the final revision
526
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,)
528
505
# no parents, use the root sequence
529
root_count = revno_to_branch_count.get(0, -1)
532
revno = (0, root_count, 1)
507
# make a parallel import revision number
508
revno = (0, sequence, 1)
535
revno_to_branch_count[0] = root_count
537
512
# store the revno for this node for future reference
538
513
revnos[node_name][0] = revno
555
530
if not left_subtree_pushed_stack[-1]:
556
531
# recurse depth first into the primary parent
557
532
next_node_name = pending_parents_stack[-1].pop(0)
558
is_left_subtree = True
559
left_subtree_pushed_stack[-1] = True
561
534
# place any merges in right-to-left order for scheduling
562
535
# which gives us left-to-right order after we reverse
563
# the scheduled queue. XXX: This has the effect of
536
# the scheduled queue. XXX: This has the effect of
564
537
# allocating common-new revisions to the right-most
565
# subtree rather than the left most, which will
538
# subtree rather than the left most, which will
566
539
# display nicely (you get smaller trees at the top
567
540
# of the combined merge).
568
541
next_node_name = pending_parents_stack[-1].pop()
569
is_left_subtree = False
570
542
if next_node_name in completed_node_names:
571
543
# this parent was completed by a child on the
572
544
# call stack. skip it.
641
610
self._node_merge_depth_stack.append(merge_depth)
642
611
self._left_subtree_pushed_stack.append(False)
643
612
self._pending_parents_stack.append(list(parents))
644
# 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]
647
616
# node has parents, assign from the left most parent.
649
parent_info = self._revnos[parents[0]]
651
# Left-hand parent is a ghost, consider it not to exist
653
if parent_info is not None:
654
first_child = parent_info[1]
655
parent_info[1] = False
617
parent_revno = self._revnos[parents[0]]
618
sequence = parent_revno[1]
657
# We don't use the same algorithm here, but we need to keep the
660
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)
662
626
def _pop_node(self):
663
"""Pop the top node off the stack
627
"""Pop the top node off the stack
665
629
The node is appended to the sorted output.
668
632
# pop off the local variables
669
633
node_name = self._node_name_stack.pop()
670
634
merge_depth = self._node_merge_depth_stack.pop()
671
first_child = self._first_child_stack.pop()
635
sequence = self._assigned_sequence_stack.pop()
672
636
# remove this node from the pending lists:
673
637
self._left_subtree_pushed_stack.pop()
674
638
self._pending_parents_stack.pop()
676
640
parents = self._original_graph[node_name]
679
642
# node has parents, assign from the left most parent.
681
parent_revno = self._revnos[parents[0]][0]
683
# left-hand parent is a ghost, treat it as not existing
685
if parent_revno is not None:
643
parent_revno = self._revnos[parents[0]]
687
645
# not the first child, make a new branch
688
base_revno = parent_revno[0]
689
branch_count = self._revno_to_branch_count.get(base_revno, 0)
691
self._revno_to_branch_count[base_revno] = branch_count
692
revno = (parent_revno[0], branch_count, 1)
693
# revno = (parent_revno[0], branch_count, parent_revno[-1]+1)
646
revno = parent_revno[0] + (sequence, 1)
695
# as the first child, we just increase the final revision
697
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,)
699
651
# no parents, use the root sequence
700
root_count = self._revno_to_branch_count.get(0, 0)
701
root_count = self._revno_to_branch_count.get(0, -1)
704
revno = (0, root_count, 1)
653
# make a parallel import revision number
654
revno = (0, sequence, 1)
707
self._revno_to_branch_count[0] = root_count
709
658
# store the revno for this node for future reference
710
659
self._revnos[node_name][0] = revno
added
removed
bzrlib/tsort.py
