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)
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()
80
79
"""Sort the graph and return as a list.
82
81
After calling this the sorter is empty and you must create a new one.
84
83
return list(self.iter_topo_order())
96
95
def iter_topo_order(self):
97
96
"""Yield the nodes of the graph in a topological order.
99
98
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
102
# 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
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
128
110
if not parents_to_visit:
129
111
# 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)
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()
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))
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)
162
162
def merge_sort(graph, branch_tip, mainline_revisions=None, generate_revno=False):
163
163
"""Topological sort a graph which groups merges.
165
165
:param graph: sequence of pairs of node->parents_list.
166
: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
167
167
reachable from branch_tip are not included in the
169
169
:param mainline_revisions: If not None this forces a mainline to be
230
229
The result is a list sorted so that all parents come before
231
230
their children. Each element of the list is a tuple containing:
232
231
(sequence_number, node_name, merge_depth, end_of_merge)
233
* 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
235
234
* node_name: The node name: opaque text to the merge routine.
236
235
* merge_depth: How many levels of merging deep this node has been
238
237
* revno_sequence: When requested this field provides a sequence of
239
238
revision numbers for all revisions. The format is:
240
(REVNO, BRANCHNUM, BRANCHREVNO). BRANCHNUM is the number of the
239
REVNO[[.BRANCHREVNO.REVNO] ...]. BRANCHREVNO is the number of the
241
240
branch that the revno is on. From left to right the REVNO numbers
242
241
are the sequence numbers within that branch of the revision.
243
242
For instance, the graph {A:[], B:['A'], C:['A', 'B']} will get
283
282
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
283
D is not the end of a cluster from rule 1, but is from rule 2: E
285
284
is not its left most ancestor
286
285
E is the end of a cluster due to rule 1
287
286
F might be but we need more data.
289
288
we show connecting lines to a parent when:
290
289
- 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
290
That is, the merge was not done to the mainline before this cluster
292
291
was merged to the mainline.
293
292
This can be detected thus:
294
* 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
297
296
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
298
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
300
C 1 [H] # note that this is a long line to show back to the
302
301
ancestor - see the end of merge rules.
366
359
self._original_graph = dict(self._graph.items())
367
360
# we need to know the revision numbers of revisions to determine
368
361
# 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
362
# this is a graph from node to [revno_tuple, sequence_number]
363
# where sequence is the number of branches made from the node,
371
364
# and revno_tuple is the tuple that was assigned to the node.
372
365
# 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 = {}
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
379
372
# this is a stack storing the depth first search into the graph.
380
373
self._node_name_stack = []
381
374
# at each level of recursion we need the merge depth this node is at:
382
375
self._node_merge_depth_stack = []
383
376
# 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
377
# stack stores the parents we have not yet checked for the node at the
385
378
# matching depth in _node_name_stack
386
379
self._pending_parents_stack = []
387
380
# When we first look at a node we assign it a seqence number from its
388
381
# leftmost parent.
389
self._first_child_stack = []
382
self._assigned_sequence_stack = []
390
383
# this is a set of the nodes who have been completely analysed for fast
391
384
# membership checking
392
385
self._completed_node_names = set()
402
# the scheduling order is: F, E, D, C, B, A
395
# the scheduling order is: F, E, D, C, B, A
403
396
# that is - 'left subtree, right subtree, node'
404
397
# which would mean that when we schedule A we can emit the entire tree.
405
398
self._scheduled_nodes = []
406
# 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
407
400
# unmerged subtree. After this subtree is scheduled, all other subtrees
408
401
# have their merge depth increased by one from this nodes merge depth.
409
402
# it contains tuples - name, merge_depth
410
403
self._left_subtree_pushed_stack = []
412
405
# 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,)):
406
if branch_tip is not None:
416
407
parents = self._graph.pop(branch_tip)
417
408
self._push_node(branch_tip, 0, parents)
419
410
def sorted(self):
420
411
"""Sort the graph and return as a list.
422
413
After calling this the sorter is empty and you must create a new one.
424
415
return list(self.iter_topo_order())
426
417
def iter_topo_order(self):
427
418
"""Yield the nodes of the graph in a topological order.
429
420
After finishing iteration the sorter is empty and you cannot continue
460
452
node_merge_depth_stack_append(merge_depth)
461
453
left_subtree_pushed_stack_append(False)
462
454
pending_parents_stack_append(list(parents))
463
# 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]
466
458
# 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
459
parent_revno = revnos[parents[0]]
460
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)
463
# no parents, use the root sequence
464
sequence = self._root_sequence
465
self._root_sequence +=1
466
assigned_sequence_stack_append(sequence)
481
468
def pop_node(node_name_stack_pop=node_name_stack.pop,
482
469
node_merge_depth_stack_pop=node_merge_depth_stack.pop,
483
first_child_stack_pop=self._first_child_stack.pop,
470
assigned_sequence_stack_pop=self._assigned_sequence_stack.pop,
484
471
left_subtree_pushed_stack_pop=left_subtree_pushed_stack.pop,
485
472
pending_parents_stack_pop=pending_parents_stack.pop,
486
473
original_graph=self._original_graph,
487
474
revnos=self._revnos,
488
475
completed_node_names_add=self._completed_node_names.add,
489
476
scheduled_nodes_append=scheduled_nodes.append,
490
revno_to_branch_count=self._revno_to_branch_count,
492
478
"""Pop the top node off the stack
497
483
# pop off the local variables
498
484
node_name = node_name_stack_pop()
499
485
merge_depth = node_merge_depth_stack_pop()
500
first_child = first_child_stack_pop()
486
sequence = assigned_sequence_stack_pop()
501
487
# remove this node from the pending lists:
502
488
left_subtree_pushed_stack_pop()
503
489
pending_parents_stack_pop()
505
491
parents = original_graph[node_name]
508
493
# 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:
494
parent_revno = revnos[parents[0]]
516
496
# 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)
497
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,)
499
# increment the sequence number within the branch
500
revno = parent_revno[0][:-1] + (parent_revno[0][-1] + 1,)
528
502
# no parents, use the root sequence
529
root_count = revno_to_branch_count.get(0, -1)
532
revno = (0, root_count, 1)
504
# make a parallel import revision number
505
revno = (0, sequence, 1)
535
revno_to_branch_count[0] = root_count
537
509
# store the revno for this node for future reference
538
510
revnos[node_name][0] = revno
555
527
if not left_subtree_pushed_stack[-1]:
556
528
# recurse depth first into the primary parent
557
529
next_node_name = pending_parents_stack[-1].pop(0)
558
is_left_subtree = True
559
left_subtree_pushed_stack[-1] = True
561
531
# place any merges in right-to-left order for scheduling
562
532
# which gives us left-to-right order after we reverse
563
# the scheduled queue. XXX: This has the effect of
533
# the scheduled queue. XXX: This has the effect of
564
534
# allocating common-new revisions to the right-most
565
# subtree rather than the left most, which will
535
# subtree rather than the left most, which will
566
536
# display nicely (you get smaller trees at the top
567
537
# of the combined merge).
568
538
next_node_name = pending_parents_stack[-1].pop()
569
is_left_subtree = False
570
539
if next_node_name in completed_node_names:
571
540
# this parent was completed by a child on the
572
541
# call stack. skip it.
641
607
self._node_merge_depth_stack.append(merge_depth)
642
608
self._left_subtree_pushed_stack.append(False)
643
609
self._pending_parents_stack.append(list(parents))
644
# 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]
647
613
# 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
614
parent_revno = self._revnos[parents[0]]
615
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)
618
# no parents, use the root sequence
619
sequence = self._root_sequence
620
self._root_sequence +=1
621
self._assigned_sequence_stack.append(sequence)
662
623
def _pop_node(self):
663
"""Pop the top node off the stack
624
"""Pop the top node off the stack
665
626
The node is appended to the sorted output.
668
629
# pop off the local variables
669
630
node_name = self._node_name_stack.pop()
670
631
merge_depth = self._node_merge_depth_stack.pop()
671
first_child = self._first_child_stack.pop()
632
sequence = self._assigned_sequence_stack.pop()
672
633
# remove this node from the pending lists:
673
634
self._left_subtree_pushed_stack.pop()
674
635
self._pending_parents_stack.pop()
676
637
parents = self._original_graph[node_name]
679
639
# 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:
640
parent_revno = self._revnos[parents[0]]
687
642
# 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)
643
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,)
645
# increment the sequence number within the branch
646
revno = parent_revno[0][:-1] + (parent_revno[0][-1] + 1,)
699
648
# 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)
650
# make a parallel import revision number
651
revno = (0, sequence, 1)
707
self._revno_to_branch_count[0] = root_count
709
655
# store the revno for this node for future reference
710
656
self._revnos[node_name][0] = revno