13
13
# You should have received a copy of the GNU General Public License
14
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18
"""Topological sorting routines."""
24
revision as _mod_revision,
28
__all__ = ["topo_sort", "TopoSorter", "merge_sort", "MergeSorter"]
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19
from bzrlib.errors import GraphCycleError
31
21
def topo_sort(graph):
32
22
"""Topological sort a graph.
34
24
graph -- sequence of pairs of node->parents_list.
36
The result is a list of node names, such that all parents come before their
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.
49
kg = _mod_graph.KnownGraph(dict(graph))
53
class TopoSorter(object):
55
def __init__(self, graph):
56
"""Topological sorting of a graph.
58
:param graph: sequence of pairs of node_name->parent_names_list.
59
i.e. [('C', ['B']), ('B', ['A']), ('A', [])]
60
For this input the output from the sort or
61
iter_topo_order routines will be:
64
node identifiers can be any hashable object, and are typically strings.
66
If you have a graph like [('a', ['b']), ('a', ['c'])] this will only use
67
one of the two values for 'a'.
69
The graph is sorted lazily: until you iterate or sort the input is
70
not processed other than to create an internal representation.
72
iteration or sorting may raise GraphCycleError if a cycle is present
75
# store a dict of the graph.
76
self._graph = dict(graph)
79
"""Sort the graph and return as a list.
81
After calling this the sorter is empty and you must create a new one.
83
return list(self.iter_topo_order())
85
### Useful if fiddling with this code.
87
### sorted_names = list(self.iter_topo_order())
88
### for index in range(len(sorted_names)):
89
### rev = sorted_names[index]
90
### for left_index in range(index):
91
### if rev in self.original_graph[sorted_names[left_index]]:
92
### print "revision in parent list of earlier revision"
93
### import pdb;pdb.set_trace()
95
def iter_topo_order(self):
96
"""Yield the nodes of the graph in a topological order.
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
# 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
127
if not parents_to_visit:
128
# 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)
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))
161
def merge_sort(graph, branch_tip, mainline_revisions=None, generate_revno=False):
162
"""Topological sort a graph which groups merges.
164
:param graph: sequence of pairs of node->parents_list.
165
:param branch_tip: the tip of the branch to graph. Revisions not
166
reachable from branch_tip are not included in the
168
:param mainline_revisions: If not None this forces a mainline to be
169
used rather than synthesised from the graph.
170
This must be a valid path through some part
171
of the graph. If the mainline does not cover all
172
the revisions, output stops at the start of the
173
old revision listed in the mainline revisions
175
The order for this parameter is oldest-first.
176
:param generate_revno: Optional parameter controlling the generation of
177
revision number sequences in the output. See the output description of
178
the MergeSorter docstring for details.
179
:result: See the MergeSorter docstring for details.
180
node identifiers can be any hashable object, and are typically strings.
182
return MergeSorter(graph, branch_tip, mainline_revisions,
183
generate_revno).sorted()
186
class MergeSorter(object):
188
__slots__ = ['_node_name_stack',
189
'_node_merge_depth_stack',
190
'_pending_parents_stack',
191
'_first_child_stack',
192
'_left_subtree_pushed_stack',
195
'_mainline_revisions',
199
'_revno_to_branch_count',
200
'_completed_node_names',
204
def __init__(self, graph, branch_tip, mainline_revisions=None,
205
generate_revno=False):
206
"""Merge-aware topological sorting of a graph.
208
:param graph: sequence of pairs of node_name->parent_names_list.
209
i.e. [('C', ['B']), ('B', ['A']), ('A', [])]
210
For this input the output from the sort or
211
iter_topo_order routines will be:
213
:param branch_tip: the tip of the branch to graph. Revisions not
214
reachable from branch_tip are not included in the
216
:param mainline_revisions: If not None this forces a mainline to be
217
used rather than synthesised from the graph.
218
This must be a valid path through some part
219
of the graph. If the mainline does not cover all
220
the revisions, output stops at the start of the
221
old revision listed in the mainline revisions
223
The order for this parameter is oldest-first.
224
:param generate_revno: Optional parameter controlling the generation of
225
revision number sequences in the output. See the output description
228
The result is a list sorted so that all parents come before
229
their children. Each element of the list is a tuple containing:
230
(sequence_number, node_name, merge_depth, end_of_merge)
231
* sequence_number: The sequence of this row in the output. Useful for
233
* node_name: The node name: opaque text to the merge routine.
234
* merge_depth: How many levels of merging deep this node has been
236
* revno_sequence: When requested this field provides a sequence of
237
revision numbers for all revisions. The format is:
238
(REVNO, BRANCHNUM, BRANCHREVNO). BRANCHNUM is the number of the
239
branch that the revno is on. From left to right the REVNO numbers
240
are the sequence numbers within that branch of the revision.
241
For instance, the graph {A:[], B:['A'], C:['A', 'B']} will get
242
the following revno_sequences assigned: A:(1,), B:(1,1,1), C:(2,).
243
This should be read as 'A is the first commit in the trunk',
244
'B is the first commit on the first branch made from A', 'C is the
245
second commit in the trunk'.
246
* end_of_merge: When True the next node is part of a different merge.
249
node identifiers can be any hashable object, and are typically strings.
251
If you have a graph like [('a', ['b']), ('a', ['c'])] this will only use
252
one of the two values for 'a'.
254
The graph is sorted lazily: until you iterate or sort the input is
255
not processed other than to create an internal representation.
257
iteration or sorting may raise GraphCycleError if a cycle is present
260
Background information on the design:
261
-------------------------------------
262
definition: the end of any cluster or 'merge' occurs when:
263
1 - the next revision has a lower merge depth than we do.
270
C, D are the ends of clusters, E might be but we need more data.
271
2 - or the next revision at our merge depth is not our left most
273
This is required to handle multiple-merges in one commit.
281
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
283
is not its left most ancestor
284
E is the end of a cluster due to rule 1
285
F might be but we need more data.
287
we show connecting lines to a parent when:
288
- 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
290
was merged to the mainline.
291
This can be detected thus:
292
* The parent has a higher merge depth and is the next revision in
295
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
299
C 1 [H] # note that this is a long line to show back to the
300
ancestor - see the end of merge rules.
306
- Part of this merges 'branch':
307
The parent has the same merge depth and is our left most parent and we
308
are not the end of the cluster.
309
A 0 [C, B] lines: [B, C]
310
B 1 [E, C] lines: [C]
312
D 0 [F, E] lines: [E, F]
315
- The end of this merge/cluster:
316
we can ONLY have multiple parents at the end of a cluster if this
317
branch was previously merged into the 'mainline'.
318
- if we have one and only one parent, show it
319
Note that this may be to a greater merge depth - for instance if
320
this branch continued from a deeply nested branch to add something
322
- if we have more than one parent - show the second oldest (older ==
323
further down the list) parent with
324
an equal or lower merge depth
325
XXXX revisit when awake. ddaa asks about the relevance of each one
326
- maybe more than one parent is relevant
328
self._generate_revno = generate_revno
329
# a dict of the graph.
330
self._graph = dict(graph)
331
# if there is an explicit mainline, alter the graph to match. This is
332
# easier than checking at every merge whether we are on the mainline and
333
# if so which path to take.
334
if mainline_revisions is None:
335
self._mainline_revisions = []
336
self._stop_revision = None
338
self._mainline_revisions = list(mainline_revisions)
339
self._stop_revision = self._mainline_revisions[0]
340
# skip the first revision, its what we reach and its parents are
341
# therefore irrelevant
342
for index, revision in enumerate(self._mainline_revisions[1:]):
343
# NB: index 0 means self._mainline_revisions[1]
344
# if the mainline matches the graph, nothing to do.
345
parent = self._mainline_revisions[index]
347
# end of mainline_revisions history
349
graph_parent_ids = self._graph[revision]
350
if not graph_parent_ids:
351
# We ran into a ghost, skip over it, this is a workaround for
352
# bug #243536, the _graph has had ghosts stripped, but the
353
# mainline_revisions have not
355
if graph_parent_ids[0] == parent:
357
# remove it from its prior spot
358
self._graph[revision].remove(parent)
359
# insert it into the start of the mainline
360
self._graph[revision].insert(0, parent)
361
# we need to do a check late in the process to detect end-of-merges
362
# which requires the parents to be accessible: its easier for now
363
# to just keep the original graph around.
364
self._original_graph = dict(self._graph.items())
365
# we need to know the revision numbers of revisions to determine
366
# 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
369
# and revno_tuple is the tuple that was assigned to the node.
370
# 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 = {}
377
# this is a stack storing the depth first search into the graph.
378
self._node_name_stack = []
379
# at each level of recursion we need the merge depth this node is at:
380
self._node_merge_depth_stack = []
381
# 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
383
# matching depth in _node_name_stack
384
self._pending_parents_stack = []
385
# When we first look at a node we assign it a seqence number from its
387
self._first_child_stack = []
388
# this is a set of the nodes who have been completely analysed for fast
389
# membership checking
390
self._completed_node_names = set()
391
# this is the scheduling of nodes list.
392
# Nodes are scheduled
393
# from the bottom left of the tree: in the tree
400
# the scheduling order is: F, E, D, C, B, A
401
# that is - 'left subtree, right subtree, node'
402
# which would mean that when we schedule A we can emit the entire tree.
403
self._scheduled_nodes = []
404
# This records for each node when we have processed its left most
405
# unmerged subtree. After this subtree is scheduled, all other subtrees
406
# have their merge depth increased by one from this nodes merge depth.
407
# it contains tuples - name, merge_depth
408
self._left_subtree_pushed_stack = []
410
# 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,)):
414
parents = self._graph.pop(branch_tip)
415
self._push_node(branch_tip, 0, parents)
418
"""Sort the graph and return as a list.
420
After calling this the sorter is empty and you must create a new one.
422
return list(self.iter_topo_order())
424
def iter_topo_order(self):
425
"""Yield the nodes of the graph in a topological order.
427
After finishing iteration the sorter is empty and you cannot continue
430
# These are safe to offload to local variables, because they are used
431
# as a stack and modified in place, never assigned to.
432
node_name_stack = self._node_name_stack
433
node_merge_depth_stack = self._node_merge_depth_stack
434
pending_parents_stack = self._pending_parents_stack
435
left_subtree_pushed_stack = self._left_subtree_pushed_stack
436
completed_node_names = self._completed_node_names
437
scheduled_nodes = self._scheduled_nodes
439
graph_pop = self._graph.pop
441
def push_node(node_name, merge_depth, parents,
442
node_name_stack_append=node_name_stack.append,
443
node_merge_depth_stack_append=node_merge_depth_stack.append,
444
left_subtree_pushed_stack_append=left_subtree_pushed_stack.append,
445
pending_parents_stack_append=pending_parents_stack.append,
446
first_child_stack_append=self._first_child_stack.append,
449
"""Add node_name to the pending node stack.
451
Names in this stack will get emitted into the output as they are popped
454
This inlines a lot of self._variable.append functions as local
457
node_name_stack_append(node_name)
458
node_merge_depth_stack_append(merge_depth)
459
left_subtree_pushed_stack_append(False)
460
pending_parents_stack_append(list(parents))
461
# as we push it, check if it is the first child
463
# node has parents, assign from the left most parent.
464
parent_info = revnos[parents[0]]
465
first_child = parent_info[1]
466
parent_info[1] = False
468
# We don't use the same algorithm here, but we need to keep the
471
first_child_stack_append(first_child)
473
def pop_node(node_name_stack_pop=node_name_stack.pop,
474
node_merge_depth_stack_pop=node_merge_depth_stack.pop,
475
first_child_stack_pop=self._first_child_stack.pop,
476
left_subtree_pushed_stack_pop=left_subtree_pushed_stack.pop,
477
pending_parents_stack_pop=pending_parents_stack.pop,
478
original_graph=self._original_graph,
480
completed_node_names_add=self._completed_node_names.add,
481
scheduled_nodes_append=scheduled_nodes.append,
482
revno_to_branch_count=self._revno_to_branch_count,
484
"""Pop the top node off the stack
486
The node is appended to the sorted output.
488
# we are returning from the flattened call frame:
489
# pop off the local variables
490
node_name = node_name_stack_pop()
491
merge_depth = node_merge_depth_stack_pop()
492
first_child = first_child_stack_pop()
493
# remove this node from the pending lists:
494
left_subtree_pushed_stack_pop()
495
pending_parents_stack_pop()
497
parents = original_graph[node_name]
499
# node has parents, assign from the left most parent.
500
parent_revno = revnos[parents[0]][0]
502
# not the first child, make a new branch
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)
510
# as the first child, we just increase the final revision
512
revno = parent_revno[:-1] + (parent_revno[-1] + 1,)
514
# no parents, use the root sequence
515
root_count = revno_to_branch_count.get(0, -1)
518
revno = (0, root_count, 1)
521
revno_to_branch_count[0] = root_count
523
# store the revno for this node for future reference
524
revnos[node_name][0] = revno
525
completed_node_names_add(node_name)
526
scheduled_nodes_append((node_name, merge_depth, revno))
530
while node_name_stack:
531
# loop until this call completes.
532
parents_to_visit = pending_parents_stack[-1]
533
# if all parents are done, the revision is done
534
if not parents_to_visit:
535
# append the revision to the topo sorted scheduled list:
536
# all the nodes parents have been scheduled added, now
537
# we can add it to the output.
540
while pending_parents_stack[-1]:
541
if not left_subtree_pushed_stack[-1]:
542
# recurse depth first into the primary parent
543
next_node_name = pending_parents_stack[-1].pop(0)
544
is_left_subtree = True
545
left_subtree_pushed_stack[-1] = True
547
# place any merges in right-to-left order for scheduling
548
# which gives us left-to-right order after we reverse
549
# the scheduled queue. XXX: This has the effect of
550
# allocating common-new revisions to the right-most
551
# subtree rather than the left most, which will
552
# display nicely (you get smaller trees at the top
553
# of the combined merge).
554
next_node_name = pending_parents_stack[-1].pop()
555
is_left_subtree = False
556
if next_node_name in completed_node_names:
557
# this parent was completed by a child on the
558
# call stack. skip it.
560
# otherwise transfer it from the source graph into the
561
# top of the current depth first search stack.
563
parents = graph_pop(next_node_name)
565
# if the next node is not in the source graph it has
566
# already been popped from it and placed into the
567
# current search stack (but not completed or we would
568
# have hit the continue 4 lines up.
569
# this indicates a cycle.
570
if next_node_name in self._original_graph:
571
raise errors.GraphCycleError(node_name_stack)
573
# This is just a ghost parent, ignore it
577
# a new child branch from name_stack[-1]
582
node_merge_depth_stack[-1] + next_merge_depth)
587
# and do not continue processing parents until this 'call'
591
# We have scheduled the graph. Now deliver the ordered output:
593
stop_revision = self._stop_revision
594
generate_revno = self._generate_revno
595
original_graph = self._original_graph
597
while scheduled_nodes:
598
node_name, merge_depth, revno = scheduled_nodes.pop()
599
if node_name == stop_revision:
601
if not len(scheduled_nodes):
602
# last revision is the end of a merge
604
elif scheduled_nodes[-1][1] < merge_depth:
605
# the next node is to our left
607
elif (scheduled_nodes[-1][1] == merge_depth and
608
(scheduled_nodes[-1][0] not in
609
original_graph[node_name])):
610
# the next node was part of a multiple-merge.
615
yield (sequence_number, node_name, merge_depth, revno, end_of_merge)
617
yield (sequence_number, node_name, merge_depth, end_of_merge)
620
def _push_node(self, node_name, merge_depth, parents):
621
"""Add node_name to the pending node stack.
623
Names in this stack will get emitted into the output as they are popped
626
self._node_name_stack.append(node_name)
627
self._node_merge_depth_stack.append(merge_depth)
628
self._left_subtree_pushed_stack.append(False)
629
self._pending_parents_stack.append(list(parents))
630
# as we push it, figure out if this is the first child
631
parents = self._original_graph[node_name]
633
# node has parents, assign from the left most parent.
634
parent_info = self._revnos[parents[0]]
635
first_child = parent_info[1]
636
parent_info[1] = False
638
# We don't use the same algorithm here, but we need to keep the
641
self._first_child_stack.append(first_child)
644
"""Pop the top node off the stack
646
The node is appended to the sorted output.
648
# we are returning from the flattened call frame:
649
# pop off the local variables
650
node_name = self._node_name_stack.pop()
651
merge_depth = self._node_merge_depth_stack.pop()
652
first_child = self._first_child_stack.pop()
653
# remove this node from the pending lists:
654
self._left_subtree_pushed_stack.pop()
655
self._pending_parents_stack.pop()
657
parents = self._original_graph[node_name]
659
# node has parents, assign from the left most parent.
660
parent_revno = self._revnos[parents[0]][0]
662
# not the first child, make a new branch
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)
670
# as the first child, we just increase the final revision
672
revno = parent_revno[:-1] + (parent_revno[-1] + 1,)
674
# no parents, use the root sequence
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
684
# store the revno for this node for future reference
685
self._revnos[node_name][0] = revno
686
self._completed_node_names.add(node_name)
687
self._scheduled_nodes.append((node_name, merge_depth, self._revnos[node_name][0]))
26
The result is a list of node names, such that all parents come before
29
Nodes at the same depth are returned in sorted order.
31
node identifiers can be any hashable object, and are typically strings.
33
parents = {} # node -> list of parents
34
children = {} # node -> list of children
35
for node, node_parents in graph:
36
assert node not in parents, \
37
('node %r repeated in graph' % node)
38
parents[node] = set(node_parents)
39
if node not in children:
40
children[node] = set()
41
for parent in node_parents:
42
if parent in children:
43
children[parent].add(node)
45
children[parent] = set([node])
48
# find nodes with no parents, and take them now
49
no_parents = [n for n in parents if len(parents[n]) == 0]
52
raise GraphCycleError(parents)
55
for child in children[n]:
56
assert n in parents[child]
57
parents[child].remove(n)
added
removed
bzrlib/tsort.py
