1
# Copyright (C) 2005, 2006, 2008 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
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"]
32
"""Topological sort a graph.
34
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.
181
Node identifiers can be any hashable object, and are typically strings.
183
return MergeSorter(graph, branch_tip, mainline_revisions,
184
generate_revno).sorted()
187
class MergeSorter(object):
189
__slots__ = ['_node_name_stack',
190
'_node_merge_depth_stack',
191
'_pending_parents_stack',
192
'_first_child_stack',
193
'_left_subtree_pushed_stack',
196
'_mainline_revisions',
200
'_revno_to_branch_count',
201
'_completed_node_names',
205
def __init__(self, graph, branch_tip, mainline_revisions=None,
206
generate_revno=False):
207
"""Merge-aware topological sorting of a graph.
209
:param graph: sequence of pairs of node_name->parent_names_list.
210
i.e. [('C', ['B']), ('B', ['A']), ('A', [])]
211
For this input the output from the sort or
212
iter_topo_order routines will be:
214
:param branch_tip: the tip of the branch to graph. Revisions not
215
reachable from branch_tip are not included in the
217
:param mainline_revisions: If not None this forces a mainline to be
218
used rather than synthesised from the graph.
219
This must be a valid path through some part
220
of the graph. If the mainline does not cover all
221
the revisions, output stops at the start of the
222
old revision listed in the mainline revisions
224
The order for this parameter is oldest-first.
225
:param generate_revno: Optional parameter controlling the generation of
226
revision number sequences in the output. See the output description
229
The result is a list sorted so that all parents come before
230
their children. Each element of the list is a tuple containing:
231
(sequence_number, node_name, merge_depth, end_of_merge)
232
* sequence_number: The sequence of this row in the output. Useful for
234
* node_name: The node name: opaque text to the merge routine.
235
* merge_depth: How many levels of merging deep this node has been
237
* revno_sequence: When requested this field provides a sequence of
238
revision numbers for all revisions. The format is:
239
(REVNO, BRANCHNUM, BRANCHREVNO). BRANCHNUM is the number of the
240
branch that the revno is on. From left to right the REVNO numbers
241
are the sequence numbers within that branch of the revision.
242
For instance, the graph {A:[], B:['A'], C:['A', 'B']} will get
243
the following revno_sequences assigned: A:(1,), B:(1,1,1), C:(2,).
244
This should be read as 'A is the first commit in the trunk',
245
'B is the first commit on the first branch made from A', 'C is the
246
second commit in the trunk'.
247
* end_of_merge: When True the next node is part of a different merge.
250
node identifiers can be any hashable object, and are typically strings.
252
If you have a graph like [('a', ['b']), ('a', ['c'])] this will only use
253
one of the two values for 'a'.
255
The graph is sorted lazily: until you iterate or sort the input is
256
not processed other than to create an internal representation.
258
iteration or sorting may raise GraphCycleError if a cycle is present
261
Background information on the design:
262
-------------------------------------
263
definition: the end of any cluster or 'merge' occurs when:
264
1 - the next revision has a lower merge depth than we do.
271
C, D are the ends of clusters, E might be but we need more data.
272
2 - or the next revision at our merge depth is not our left most
274
This is required to handle multiple-merges in one commit.
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
284
is not its left most ancestor
285
E is the end of a cluster due to rule 1
286
F might be but we need more data.
288
we show connecting lines to a parent when:
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
291
was merged to the mainline.
292
This can be detected thus:
293
* The parent has a higher merge depth and is the next revision in
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
300
C 1 [H] # note that this is a long line to show back to the
301
ancestor - see the end of merge rules.
307
- Part of this merges 'branch':
308
The parent has the same merge depth and is our left most parent and we
309
are not the end of the cluster.
310
A 0 [C, B] lines: [B, C]
311
B 1 [E, C] lines: [C]
313
D 0 [F, E] lines: [E, F]
316
- The end of this merge/cluster:
317
we can ONLY have multiple parents at the end of a cluster if this
318
branch was previously merged into the 'mainline'.
319
- if we have one and only one parent, show it
320
Note that this may be to a greater merge depth - for instance if
321
this branch continued from a deeply nested branch to add something
323
- if we have more than one parent - show the second oldest (older ==
324
further down the list) parent with
325
an equal or lower merge depth
326
XXXX revisit when awake. ddaa asks about the relevance of each one
327
- maybe more than one parent is relevant
329
self._generate_revno = generate_revno
330
# a dict of the graph.
331
self._graph = dict(graph)
332
# if there is an explicit mainline, alter the graph to match. This is
333
# easier than checking at every merge whether we are on the mainline and
334
# if so which path to take.
335
if mainline_revisions is None:
336
self._mainline_revisions = []
337
self._stop_revision = None
339
self._mainline_revisions = list(mainline_revisions)
340
self._stop_revision = self._mainline_revisions[0]
341
# skip the first revision, its what we reach and its parents are
342
# therefore irrelevant
343
for index, revision in enumerate(self._mainline_revisions[1:]):
344
# NB: index 0 means self._mainline_revisions[1]
345
# if the mainline matches the graph, nothing to do.
346
parent = self._mainline_revisions[index]
348
# end of mainline_revisions history
350
graph_parent_ids = self._graph[revision]
351
if not graph_parent_ids:
352
# We ran into a ghost, skip over it, this is a workaround for
353
# bug #243536, the _graph has had ghosts stripped, but the
354
# mainline_revisions have not
356
if graph_parent_ids[0] == parent:
358
# remove it from its prior spot
359
self._graph[revision].remove(parent)
360
# insert it into the start of the mainline
361
self._graph[revision].insert(0, parent)
362
# we need to do a check late in the process to detect end-of-merges
363
# which requires the parents to be accessible: its easier for now
364
# to just keep the original graph around.
365
self._original_graph = dict(self._graph.items())
366
# we need to know the revision numbers of revisions to determine
367
# 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
370
# and revno_tuple is the tuple that was assigned to the node.
371
# 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 = {}
378
# this is a stack storing the depth first search into the graph.
379
self._node_name_stack = []
380
# at each level of recursion we need the merge depth this node is at:
381
self._node_merge_depth_stack = []
382
# 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
384
# matching depth in _node_name_stack
385
self._pending_parents_stack = []
386
# When we first look at a node we assign it a seqence number from its
388
self._first_child_stack = []
389
# this is a set of the nodes who have been completely analysed for fast
390
# membership checking
391
self._completed_node_names = set()
392
# this is the scheduling of nodes list.
393
# Nodes are scheduled
394
# from the bottom left of the tree: in the tree
401
# the scheduling order is: F, E, D, C, B, A
402
# that is - 'left subtree, right subtree, node'
403
# which would mean that when we schedule A we can emit the entire tree.
404
self._scheduled_nodes = []
405
# This records for each node when we have processed its left most
406
# unmerged subtree. After this subtree is scheduled, all other subtrees
407
# have their merge depth increased by one from this nodes merge depth.
408
# it contains tuples - name, merge_depth
409
self._left_subtree_pushed_stack = []
411
# 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,)):
415
parents = self._graph.pop(branch_tip)
416
self._push_node(branch_tip, 0, parents)
419
"""Sort the graph and return as a list.
421
After calling this the sorter is empty and you must create a new one.
423
return list(self.iter_topo_order())
425
def iter_topo_order(self):
426
"""Yield the nodes of the graph in a topological order.
428
After finishing iteration the sorter is empty and you cannot continue
431
# These are safe to offload to local variables, because they are used
432
# as a stack and modified in place, never assigned to.
433
node_name_stack = self._node_name_stack
434
node_merge_depth_stack = self._node_merge_depth_stack
435
pending_parents_stack = self._pending_parents_stack
436
left_subtree_pushed_stack = self._left_subtree_pushed_stack
437
completed_node_names = self._completed_node_names
438
scheduled_nodes = self._scheduled_nodes
440
graph_pop = self._graph.pop
442
def push_node(node_name, merge_depth, parents,
443
node_name_stack_append=node_name_stack.append,
444
node_merge_depth_stack_append=node_merge_depth_stack.append,
445
left_subtree_pushed_stack_append=left_subtree_pushed_stack.append,
446
pending_parents_stack_append=pending_parents_stack.append,
447
first_child_stack_append=self._first_child_stack.append,
450
"""Add node_name to the pending node stack.
452
Names in this stack will get emitted into the output as they are popped
455
This inlines a lot of self._variable.append functions as local
458
node_name_stack_append(node_name)
459
node_merge_depth_stack_append(merge_depth)
460
left_subtree_pushed_stack_append(False)
461
pending_parents_stack_append(list(parents))
462
# as we push it, check if it is the first child
465
# 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
475
# We don't use the same algorithm here, but we need to keep the
478
first_child_stack_append(first_child)
480
def pop_node(node_name_stack_pop=node_name_stack.pop,
481
node_merge_depth_stack_pop=node_merge_depth_stack.pop,
482
first_child_stack_pop=self._first_child_stack.pop,
483
left_subtree_pushed_stack_pop=left_subtree_pushed_stack.pop,
484
pending_parents_stack_pop=pending_parents_stack.pop,
485
original_graph=self._original_graph,
487
completed_node_names_add=self._completed_node_names.add,
488
scheduled_nodes_append=scheduled_nodes.append,
489
revno_to_branch_count=self._revno_to_branch_count,
491
"""Pop the top node off the stack
493
The node is appended to the sorted output.
495
# we are returning from the flattened call frame:
496
# pop off the local variables
497
node_name = node_name_stack_pop()
498
merge_depth = node_merge_depth_stack_pop()
499
first_child = first_child_stack_pop()
500
# remove this node from the pending lists:
501
left_subtree_pushed_stack_pop()
502
pending_parents_stack_pop()
504
parents = original_graph[node_name]
507
# 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:
515
# 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)
523
# as the first child, we just increase the final revision
525
revno = parent_revno[:-1] + (parent_revno[-1] + 1,)
527
# no parents, use the root sequence
528
root_count = revno_to_branch_count.get(0, -1)
531
revno = (0, root_count, 1)
534
revno_to_branch_count[0] = root_count
536
# store the revno for this node for future reference
537
revnos[node_name][0] = revno
538
completed_node_names_add(node_name)
539
scheduled_nodes_append((node_name, merge_depth, revno))
543
while node_name_stack:
544
# loop until this call completes.
545
parents_to_visit = pending_parents_stack[-1]
546
# if all parents are done, the revision is done
547
if not parents_to_visit:
548
# append the revision to the topo sorted scheduled list:
549
# all the nodes parents have been scheduled added, now
550
# we can add it to the output.
553
while pending_parents_stack[-1]:
554
if not left_subtree_pushed_stack[-1]:
555
# recurse depth first into the primary parent
556
next_node_name = pending_parents_stack[-1].pop(0)
557
is_left_subtree = True
558
left_subtree_pushed_stack[-1] = True
560
# place any merges in right-to-left order for scheduling
561
# which gives us left-to-right order after we reverse
562
# the scheduled queue. XXX: This has the effect of
563
# allocating common-new revisions to the right-most
564
# subtree rather than the left most, which will
565
# display nicely (you get smaller trees at the top
566
# of the combined merge).
567
next_node_name = pending_parents_stack[-1].pop()
568
is_left_subtree = False
569
if next_node_name in completed_node_names:
570
# this parent was completed by a child on the
571
# call stack. skip it.
573
# otherwise transfer it from the source graph into the
574
# top of the current depth first search stack.
576
parents = graph_pop(next_node_name)
578
# if the next node is not in the source graph it has
579
# already been popped from it and placed into the
580
# current search stack (but not completed or we would
581
# have hit the continue 4 lines up.
582
# this indicates a cycle.
583
if next_node_name in self._original_graph:
584
raise errors.GraphCycleError(node_name_stack)
586
# This is just a ghost parent, ignore it
590
# a new child branch from name_stack[-1]
595
node_merge_depth_stack[-1] + next_merge_depth)
600
# and do not continue processing parents until this 'call'
604
# We have scheduled the graph. Now deliver the ordered output:
606
stop_revision = self._stop_revision
607
generate_revno = self._generate_revno
608
original_graph = self._original_graph
610
while scheduled_nodes:
611
node_name, merge_depth, revno = scheduled_nodes.pop()
612
if node_name == stop_revision:
614
if not len(scheduled_nodes):
615
# last revision is the end of a merge
617
elif scheduled_nodes[-1][1] < merge_depth:
618
# the next node is to our left
620
elif (scheduled_nodes[-1][1] == merge_depth and
621
(scheduled_nodes[-1][0] not in
622
original_graph[node_name])):
623
# the next node was part of a multiple-merge.
628
yield (sequence_number, node_name, merge_depth, revno, end_of_merge)
630
yield (sequence_number, node_name, merge_depth, end_of_merge)
633
def _push_node(self, node_name, merge_depth, parents):
634
"""Add node_name to the pending node stack.
636
Names in this stack will get emitted into the output as they are popped
639
self._node_name_stack.append(node_name)
640
self._node_merge_depth_stack.append(merge_depth)
641
self._left_subtree_pushed_stack.append(False)
642
self._pending_parents_stack.append(list(parents))
643
# as we push it, figure out if this is the first child
646
# 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
656
# We don't use the same algorithm here, but we need to keep the
659
self._first_child_stack.append(first_child)
662
"""Pop the top node off the stack
664
The node is appended to the sorted output.
666
# we are returning from the flattened call frame:
667
# pop off the local variables
668
node_name = self._node_name_stack.pop()
669
merge_depth = self._node_merge_depth_stack.pop()
670
first_child = self._first_child_stack.pop()
671
# remove this node from the pending lists:
672
self._left_subtree_pushed_stack.pop()
673
self._pending_parents_stack.pop()
675
parents = self._original_graph[node_name]
678
# 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:
686
# 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)
694
# as the first child, we just increase the final revision
696
revno = parent_revno[:-1] + (parent_revno[-1] + 1,)
698
# 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)
706
self._revno_to_branch_count[0] = root_count
708
# store the revno for this node for future reference
709
self._revnos[node_name][0] = revno
710
self._completed_node_names.add(node_name)
711
self._scheduled_nodes.append((node_name, merge_depth, self._revnos[node_name][0]))
added
removed
bzrlib/tsort.py
