30
34
graph -- sequence of pairs of node->parents_list.
32
The result is a list of node names, such that all parents come before
36
The result is a list of node names, such that all parents come before their
35
39
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.
37
return TopoSorter(graph).sorted()
49
kg = _mod_graph.KnownGraph(dict(graph))
40
53
class TopoSorter(object):
42
55
def __init__(self, graph):
43
56
"""Topological sorting of a graph.
45
58
:param graph: sequence of pairs of node_name->parent_names_list.
46
59
i.e. [('C', ['B']), ('B', ['A']), ('A', [])]
47
60
For this input the output from the sort or
48
61
iter_topo_order routines will be:
51
64
node identifiers can be any hashable object, and are typically strings.
53
66
If you have a graph like [('a', ['b']), ('a', ['c'])] this will only use
56
69
The graph is sorted lazily: until you iterate or sort the input is
57
70
not processed other than to create an internal representation.
59
iteration or sorting may raise GraphCycleError if a cycle is present
72
iteration or sorting may raise GraphCycleError if a cycle is present
62
# a dict of the graph.
75
# store a dict of the graph.
63
76
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
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
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()
103
117
# now pick a random node in the source graph, and transfer it to the
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
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
111
127
if not parents_to_visit:
112
128
# append the revision to the topo sorted list
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()
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)
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)
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))
165
161
def merge_sort(graph, branch_tip, mainline_revisions=None, generate_revno=False):
166
162
"""Topological sort a graph which groups merges.
168
164
:param graph: sequence of pairs of node->parents_list.
169
:param branch_tip: the tip of the branch to graph. Revisions not
165
:param branch_tip: the tip of the branch to graph. Revisions not
170
166
reachable from branch_tip are not included in the
172
168
:param mainline_revisions: If not None this forces a mainline to be
232
228
The result is a list sorted so that all parents come before
233
229
their children. Each element of the list is a tuple containing:
234
230
(sequence_number, node_name, merge_depth, end_of_merge)
235
* sequence_number: The sequence of this row in the output. Useful for
231
* sequence_number: The sequence of this row in the output. Useful for
237
233
* node_name: The node name: opaque text to the merge routine.
238
234
* merge_depth: How many levels of merging deep this node has been
240
236
* revno_sequence: When requested this field provides a sequence of
241
237
revision numbers for all revisions. The format is:
242
REVNO[[.BRANCHREVNO.REVNO] ...]. BRANCHREVNO is the number of the
238
(REVNO, BRANCHNUM, BRANCHREVNO). BRANCHNUM is the number of the
243
239
branch that the revno is on. From left to right the REVNO numbers
244
240
are the sequence numbers within that branch of the revision.
245
241
For instance, the graph {A:[], B:['A'], C:['A', 'B']} will get
285
281
C is the end of a cluster due to rule 1.
286
D is not the end of a cluster from rule 1, but is from rule 2: E
282
D is not the end of a cluster from rule 1, but is from rule 2: E
287
283
is not its left most ancestor
288
284
E is the end of a cluster due to rule 1
289
285
F might be but we need more data.
291
287
we show connecting lines to a parent when:
292
288
- The parent is the start of a merge within this cluster.
293
That is, the merge was not done to the mainline before this cluster
289
That is, the merge was not done to the mainline before this cluster
294
290
was merged to the mainline.
295
291
This can be detected thus:
296
* The parent has a higher merge depth and is the next revision in
292
* The parent has a higher merge depth and is the next revision in
299
295
The next revision in the list constraint is needed for this case:
301
B 1 [C, F] # we do not want to show a line to F which is depth 2
297
B 1 [C, F] # we do not want to show a line to F which is depth 2
303
C 1 [H] # note that this is a long line to show back to the
299
C 1 [H] # note that this is a long line to show back to the
304
300
ancestor - see the end of merge rules.
362
364
self._original_graph = dict(self._graph.items())
363
365
# we need to know the revision numbers of revisions to determine
364
366
# the revision numbers of their descendants
365
# this is a graph from node to [revno_tuple, sequence_number]
366
# where sequence is the number of branches made from the node,
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
367
369
# and revno_tuple is the tuple that was assigned to the node.
368
370
# we dont know revnos to start with, so we start it seeded with
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
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 = {}
375
377
# this is a stack storing the depth first search into the graph.
376
378
self._node_name_stack = []
377
379
# at each level of recursion we need the merge depth this node is at:
378
380
self._node_merge_depth_stack = []
379
381
# at each level of 'recursion' we have to check each parent. This
380
# stack stores the parents we have not yet checked for the node at the
382
# stack stores the parents we have not yet checked for the node at the
381
383
# matching depth in _node_name_stack
382
384
self._pending_parents_stack = []
383
385
# When we first look at a node we assign it a seqence number from its
384
386
# leftmost parent.
385
self._assigned_sequence_stack = []
387
self._first_child_stack = []
386
388
# this is a set of the nodes who have been completely analysed for fast
387
389
# membership checking
388
390
self._completed_node_names = set()
398
# the scheduling order is: F, E, D, C, B, A
400
# the scheduling order is: F, E, D, C, B, A
399
401
# that is - 'left subtree, right subtree, node'
400
402
# which would mean that when we schedule A we can emit the entire tree.
401
403
self._scheduled_nodes = []
402
# This records for each node when we have processed its left most
404
# This records for each node when we have processed its left most
403
405
# unmerged subtree. After this subtree is scheduled, all other subtrees
404
406
# have their merge depth increased by one from this nodes merge depth.
405
407
# it contains tuples - name, merge_depth
406
408
self._left_subtree_pushed_stack = []
408
410
# seed the search with the tip of the branch
409
if branch_tip is not None:
411
if (branch_tip is not None and
412
branch_tip != _mod_revision.NULL_REVISION and
413
branch_tip != (_mod_revision.NULL_REVISION,)):
410
414
parents = self._graph.pop(branch_tip)
411
415
self._push_node(branch_tip, 0, parents)
413
417
def sorted(self):
414
418
"""Sort the graph and return as a list.
416
420
After calling this the sorter is empty and you must create a new one.
418
422
return list(self.iter_topo_order())
420
424
def iter_topo_order(self):
421
425
"""Yield the nodes of the graph in a topological order.
423
427
After finishing iteration the sorter is empty and you cannot continue
455
458
node_merge_depth_stack_append(merge_depth)
456
459
left_subtree_pushed_stack_append(False)
457
460
pending_parents_stack_append(list(parents))
458
# as we push it, assign it a sequence number against its parent:
459
parents = original_graph[node_name]
461
# as we push it, check if it is the first child
461
463
# node has parents, assign from the left most parent.
462
parent_revno = revnos[parents[0]]
463
sequence = parent_revno[1]
464
parent_info = revnos[parents[0]]
465
first_child = parent_info[1]
466
parent_info[1] = False
466
# no parents, use the root sequence
467
sequence = self._root_sequence
468
self._root_sequence +=1
469
assigned_sequence_stack_append(sequence)
468
# We don't use the same algorithm here, but we need to keep the
471
first_child_stack_append(first_child)
471
473
def pop_node(node_name_stack_pop=node_name_stack.pop,
472
474
node_merge_depth_stack_pop=node_merge_depth_stack.pop,
473
assigned_sequence_stack_pop=self._assigned_sequence_stack.pop,
475
first_child_stack_pop=self._first_child_stack.pop,
474
476
left_subtree_pushed_stack_pop=left_subtree_pushed_stack.pop,
475
477
pending_parents_stack_pop=pending_parents_stack.pop,
476
478
original_graph=self._original_graph,
477
479
revnos=self._revnos,
478
480
completed_node_names_add=self._completed_node_names.add,
479
481
scheduled_nodes_append=scheduled_nodes.append,
482
revno_to_branch_count=self._revno_to_branch_count,
481
484
"""Pop the top node off the stack
494
497
parents = original_graph[node_name]
496
499
# node has parents, assign from the left most parent.
497
parent_revno = revnos[parents[0]]
500
parent_revno = revnos[parents[0]][0]
499
502
# not the first child, make a new branch
500
revno = parent_revno[0] + (sequence, 1)
503
base_revno = parent_revno[0]
504
branch_count = revno_to_branch_count.get(base_revno, 0)
506
revno_to_branch_count[base_revno] = branch_count
507
revno = (parent_revno[0], branch_count, 1)
508
# revno = (parent_revno[0], branch_count, parent_revno[-1]+1)
502
# increment the sequence number within the branch
503
revno = parent_revno[0][:-1] + (parent_revno[0][-1] + 1,)
510
# as the first child, we just increase the final revision
512
revno = parent_revno[:-1] + (parent_revno[-1] + 1,)
505
514
# no parents, use the root sequence
507
# make a parallel import revision number
508
revno = (0, sequence, 1)
515
root_count = revno_to_branch_count.get(0, -1)
518
revno = (0, root_count, 1)
521
revno_to_branch_count[0] = root_count
512
523
# store the revno for this node for future reference
513
524
revnos[node_name][0] = revno
530
541
if not left_subtree_pushed_stack[-1]:
531
542
# recurse depth first into the primary parent
532
543
next_node_name = pending_parents_stack[-1].pop(0)
544
is_left_subtree = True
545
left_subtree_pushed_stack[-1] = True
534
547
# place any merges in right-to-left order for scheduling
535
548
# which gives us left-to-right order after we reverse
536
# the scheduled queue. XXX: This has the effect of
549
# the scheduled queue. XXX: This has the effect of
537
550
# allocating common-new revisions to the right-most
538
# subtree rather than the left most, which will
551
# subtree rather than the left most, which will
539
552
# display nicely (you get smaller trees at the top
540
553
# of the combined merge).
541
554
next_node_name = pending_parents_stack[-1].pop()
555
is_left_subtree = False
542
556
if next_node_name in completed_node_names:
543
557
# this parent was completed by a child on the
544
558
# call stack. skip it.
610
627
self._node_merge_depth_stack.append(merge_depth)
611
628
self._left_subtree_pushed_stack.append(False)
612
629
self._pending_parents_stack.append(list(parents))
613
# as we push it, assign it a sequence number against its parent:
630
# as we push it, figure out if this is the first child
614
631
parents = self._original_graph[node_name]
616
633
# node has parents, assign from the left most parent.
617
parent_revno = self._revnos[parents[0]]
618
sequence = parent_revno[1]
634
parent_info = self._revnos[parents[0]]
635
first_child = parent_info[1]
636
parent_info[1] = False
621
# no parents, use the root sequence
622
sequence = self._root_sequence
623
self._root_sequence +=1
624
self._assigned_sequence_stack.append(sequence)
638
# We don't use the same algorithm here, but we need to keep the
641
self._first_child_stack.append(first_child)
626
643
def _pop_node(self):
627
"""Pop the top node off the stack
644
"""Pop the top node off the stack
629
646
The node is appended to the sorted output.
640
657
parents = self._original_graph[node_name]
642
659
# node has parents, assign from the left most parent.
643
parent_revno = self._revnos[parents[0]]
660
parent_revno = self._revnos[parents[0]][0]
645
662
# not the first child, make a new branch
646
revno = parent_revno[0] + (sequence, 1)
663
base_revno = parent_revno[0]
664
branch_count = self._revno_to_branch_count.get(base_revno, 0)
666
self._revno_to_branch_count[base_revno] = branch_count
667
revno = (parent_revno[0], branch_count, 1)
668
# revno = (parent_revno[0], branch_count, parent_revno[-1]+1)
648
# increment the sequence number within the branch
649
revno = parent_revno[0][:-1] + (parent_revno[0][-1] + 1,)
670
# as the first child, we just increase the final revision
672
revno = parent_revno[:-1] + (parent_revno[-1] + 1,)
651
674
# no parents, use the root sequence
653
# make a parallel import revision number
654
revno = (0, sequence, 1)
675
root_count = self._revno_to_branch_count.get(0, 0)
676
root_count = self._revno_to_branch_count.get(0, -1)
679
revno = (0, root_count, 1)
682
self._revno_to_branch_count[0] = root_count
658
684
# store the revno for this node for future reference
659
685
self._revnos[node_name][0] = revno
added
removed
bzrlib/tsort.py
