1
# Copyright (C) 2005, 2006 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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18
"""Topological sorting routines."""
21
from bzrlib import errors
24
__all__ = ["topo_sort", "TopoSorter", "merge_sort", "MergeSorter"]
28
"""Topological sort a graph.
30
graph -- sequence of pairs of node->parents_list.
32
The result is a list of node names, such that all parents come before
35
node identifiers can be any hashable object, and are typically strings.
37
return TopoSorter(graph).sorted()
40
class TopoSorter(object):
42
def __init__(self, graph):
43
"""Topological sorting of a graph.
45
:param graph: sequence of pairs of node_name->parent_names_list.
46
i.e. [('C', ['B']), ('B', ['A']), ('A', [])]
47
For this input the output from the sort or
48
iter_topo_order routines will be:
51
node identifiers can be any hashable object, and are typically strings.
53
If you have a graph like [('a', ['b']), ('a', ['c'])] this will only use
54
one of the two values for 'a'.
56
The graph is sorted lazily: until you iterate or sort the input is
57
not processed other than to create an internal representation.
59
iteration or sorting may raise GraphCycleError if a cycle is present
62
# a dict of the graph.
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()
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
# now pick a random node in the source graph, and transfer it to the
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
110
if not parents_to_visit:
111
# append the revision to the topo sorted list
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()
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
def merge_sort(graph, branch_tip, mainline_revisions=None, generate_revno=False):
163
"""Topological sort a graph which groups merges.
165
:param graph: sequence of pairs of node->parents_list.
166
:param branch_tip: the tip of the branch to graph. Revisions not
167
reachable from branch_tip are not included in the
169
:param mainline_revisions: If not None this forces a mainline to be
170
used rather than synthesised from the graph.
171
This must be a valid path through some part
172
of the graph. If the mainline does not cover all
173
the revisions, output stops at the start of the
174
old revision listed in the mainline revisions
176
The order for this parameter is oldest-first.
177
:param generate_revno: Optional parameter controlling the generation of
178
revision number sequences in the output. See the output description of
179
the MergeSorter docstring for details.
180
: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
'_assigned_sequence_stack',
193
'_left_subtree_pushed_stack',
196
'_mainline_revisions',
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[[.BRANCHREVNO.REVNO] ...]. BRANCHREVNO 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
if self._graph[revision][0] == parent:
352
# remove it from its prior spot
353
self._graph[revision].remove(parent)
354
# insert it into the start of the mainline
355
self._graph[revision].insert(0, parent)
356
# we need to do a check late in the process to detect end-of-merges
357
# which requires the parents to be accessible: its easier for now
358
# to just keep the original graph around.
359
self._original_graph = dict(self._graph.items())
360
# we need to know the revision numbers of revisions to determine
361
# the revision numbers of their descendants
362
# this is a graph from node to [revno_tuple, sequence_number]
363
# where sequence is the number of branches made from the node,
364
# and revno_tuple is the tuple that was assigned to the node.
365
# we dont know revnos to start with, so we start it seeded with
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
372
# this is a stack storing the depth first search into the graph.
373
self._node_name_stack = []
374
# at each level of recursion we need the merge depth this node is at:
375
self._node_merge_depth_stack = []
376
# at each level of 'recursion' we have to check each parent. This
377
# stack stores the parents we have not yet checked for the node at the
378
# matching depth in _node_name_stack
379
self._pending_parents_stack = []
380
# When we first look at a node we assign it a seqence number from its
382
self._assigned_sequence_stack = []
383
# this is a set of the nodes who have been completely analysed for fast
384
# membership checking
385
self._completed_node_names = set()
386
# this is the scheduling of nodes list.
387
# Nodes are scheduled
388
# from the bottom left of the tree: in the tree
395
# the scheduling order is: F, E, D, C, B, A
396
# that is - 'left subtree, right subtree, node'
397
# which would mean that when we schedule A we can emit the entire tree.
398
self._scheduled_nodes = []
399
# This records for each node when we have processed its left most
400
# unmerged subtree. After this subtree is scheduled, all other subtrees
401
# have their merge depth increased by one from this nodes merge depth.
402
# it contains tuples - name, merge_depth
403
self._left_subtree_pushed_stack = []
405
# seed the search with the tip of the branch
406
if branch_tip is not None:
407
parents = self._graph.pop(branch_tip)
408
self._push_node(branch_tip, 0, parents)
411
"""Sort the graph and return as a list.
413
After calling this the sorter is empty and you must create a new one.
415
return list(self.iter_topo_order())
417
def iter_topo_order(self):
418
"""Yield the nodes of the graph in a topological order.
420
After finishing iteration the sorter is empty and you cannot continue
423
# These are safe to offload to local variables, because they are used
424
# as a stack and modified in place, never assigned to.
425
node_name_stack = self._node_name_stack
426
node_merge_depth_stack = self._node_merge_depth_stack
427
pending_parents_stack = self._pending_parents_stack
428
left_subtree_pushed_stack = self._left_subtree_pushed_stack
429
completed_node_names = self._completed_node_names
430
scheduled_nodes = self._scheduled_nodes
432
graph_pop = self._graph.pop
434
def push_node(node_name, merge_depth, parents,
435
node_name_stack_append=node_name_stack.append,
436
node_merge_depth_stack_append=node_merge_depth_stack.append,
437
left_subtree_pushed_stack_append=left_subtree_pushed_stack.append,
438
pending_parents_stack_append=pending_parents_stack.append,
439
assigned_sequence_stack_append=self._assigned_sequence_stack.append,
440
original_graph=self._original_graph,
443
"""Add node_name to the pending node stack.
445
Names in this stack will get emitted into the output as they are popped
448
This inlines a lot of self._variable.append functions as local
451
node_name_stack_append(node_name)
452
node_merge_depth_stack_append(merge_depth)
453
left_subtree_pushed_stack_append(False)
454
pending_parents_stack_append(list(parents))
455
# as we push it, assign it a sequence number against its parent:
456
parents = original_graph[node_name]
458
# node has parents, assign from the left most parent.
459
parent_revno = revnos[parents[0]]
460
sequence = parent_revno[1]
463
# no parents, use the root sequence
464
sequence = self._root_sequence
465
self._root_sequence +=1
466
assigned_sequence_stack_append(sequence)
468
def pop_node(node_name_stack_pop=node_name_stack.pop,
469
node_merge_depth_stack_pop=node_merge_depth_stack.pop,
470
assigned_sequence_stack_pop=self._assigned_sequence_stack.pop,
471
left_subtree_pushed_stack_pop=left_subtree_pushed_stack.pop,
472
pending_parents_stack_pop=pending_parents_stack.pop,
473
original_graph=self._original_graph,
475
completed_node_names_add=self._completed_node_names.add,
476
scheduled_nodes_append=scheduled_nodes.append,
478
"""Pop the top node off the stack
480
The node is appended to the sorted output.
482
# we are returning from the flattened call frame:
483
# pop off the local variables
484
node_name = node_name_stack_pop()
485
merge_depth = node_merge_depth_stack_pop()
486
sequence = assigned_sequence_stack_pop()
487
# remove this node from the pending lists:
488
left_subtree_pushed_stack_pop()
489
pending_parents_stack_pop()
491
parents = original_graph[node_name]
493
# node has parents, assign from the left most parent.
494
parent_revno = revnos[parents[0]]
496
# not the first child, make a new branch
497
revno = parent_revno[0] + (sequence, 1)
499
# increment the sequence number within the branch
500
revno = parent_revno[0][:-1] + (parent_revno[0][-1] + 1,)
502
# no parents, use the root sequence
504
# make a parallel import revision number
505
revno = (0, sequence, 1)
509
# store the revno for this node for future reference
510
revnos[node_name][0] = revno
511
completed_node_names_add(node_name)
512
scheduled_nodes_append((node_name, merge_depth, revno))
516
while node_name_stack:
517
# loop until this call completes.
518
parents_to_visit = pending_parents_stack[-1]
519
# if all parents are done, the revision is done
520
if not parents_to_visit:
521
# append the revision to the topo sorted scheduled list:
522
# all the nodes parents have been scheduled added, now
523
# we can add it to the output.
526
while pending_parents_stack[-1]:
527
if not left_subtree_pushed_stack[-1]:
528
# recurse depth first into the primary parent
529
next_node_name = pending_parents_stack[-1].pop(0)
531
# place any merges in right-to-left order for scheduling
532
# which gives us left-to-right order after we reverse
533
# the scheduled queue. XXX: This has the effect of
534
# allocating common-new revisions to the right-most
535
# subtree rather than the left most, which will
536
# display nicely (you get smaller trees at the top
537
# of the combined merge).
538
next_node_name = pending_parents_stack[-1].pop()
539
if next_node_name in completed_node_names:
540
# this parent was completed by a child on the
541
# call stack. skip it.
543
# otherwise transfer it from the source graph into the
544
# top of the current depth first search stack.
546
parents = graph_pop(next_node_name)
548
# if the next node is not in the source graph it has
549
# already been popped from it and placed into the
550
# current search stack (but not completed or we would
551
# have hit the continue 4 lines up.
552
# this indicates a cycle.
553
raise errors.GraphCycleError(node_name_stack)
555
if left_subtree_pushed_stack[-1]:
556
# a new child branch from name_stack[-1]
560
left_subtree_pushed_stack[-1] = True
562
node_merge_depth_stack[-1] + next_merge_depth)
567
# and do not continue processing parents until this 'call'
571
# We have scheduled the graph. Now deliver the ordered output:
573
stop_revision = self._stop_revision
574
generate_revno = self._generate_revno
575
original_graph = self._original_graph
577
while scheduled_nodes:
578
node_name, merge_depth, revno = scheduled_nodes.pop()
579
if node_name == stop_revision:
581
if not len(scheduled_nodes):
582
# last revision is the end of a merge
584
elif scheduled_nodes[-1][1] < merge_depth:
585
# the next node is to our left
587
elif (scheduled_nodes[-1][1] == merge_depth and
588
(scheduled_nodes[-1][0] not in
589
original_graph[node_name])):
590
# the next node was part of a multiple-merge.
595
yield (sequence_number, node_name, merge_depth, revno, end_of_merge)
597
yield (sequence_number, node_name, merge_depth, end_of_merge)
600
def _push_node(self, node_name, merge_depth, parents):
601
"""Add node_name to the pending node stack.
603
Names in this stack will get emitted into the output as they are popped
606
self._node_name_stack.append(node_name)
607
self._node_merge_depth_stack.append(merge_depth)
608
self._left_subtree_pushed_stack.append(False)
609
self._pending_parents_stack.append(list(parents))
610
# as we push it, assign it a sequence number against its parent:
611
parents = self._original_graph[node_name]
613
# node has parents, assign from the left most parent.
614
parent_revno = self._revnos[parents[0]]
615
sequence = parent_revno[1]
618
# no parents, use the root sequence
619
sequence = self._root_sequence
620
self._root_sequence +=1
621
self._assigned_sequence_stack.append(sequence)
624
"""Pop the top node off the stack
626
The node is appended to the sorted output.
628
# we are returning from the flattened call frame:
629
# pop off the local variables
630
node_name = self._node_name_stack.pop()
631
merge_depth = self._node_merge_depth_stack.pop()
632
sequence = self._assigned_sequence_stack.pop()
633
# remove this node from the pending lists:
634
self._left_subtree_pushed_stack.pop()
635
self._pending_parents_stack.pop()
637
parents = self._original_graph[node_name]
639
# node has parents, assign from the left most parent.
640
parent_revno = self._revnos[parents[0]]
642
# not the first child, make a new branch
643
revno = parent_revno[0] + (sequence, 1)
645
# increment the sequence number within the branch
646
revno = parent_revno[0][:-1] + (parent_revno[0][-1] + 1,)
648
# no parents, use the root sequence
650
# make a parallel import revision number
651
revno = (0, sequence, 1)
655
# store the revno for this node for future reference
656
self._revnos[node_name][0] = revno
657
self._completed_node_names.add(node_name)
658
self._scheduled_nodes.append((node_name, merge_depth, self._revnos[node_name][0]))
added
removed
bzrlib/tsort.py
