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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18
"""Topological sorting routines."""
21
from bzrlib import errors
22
import bzrlib.revision as _mod_revision
25
__all__ = ["topo_sort", "TopoSorter", "merge_sort", "MergeSorter"]
29
"""Topological sort a graph.
31
graph -- sequence of pairs of node->parents_list.
33
The result is a list of node names, such that all parents come before
36
node identifiers can be any hashable object, and are typically strings.
38
return TopoSorter(graph).sorted()
41
class TopoSorter(object):
43
def __init__(self, graph):
44
"""Topological sorting of a graph.
46
:param graph: sequence of pairs of node_name->parent_names_list.
47
i.e. [('C', ['B']), ('B', ['A']), ('A', [])]
48
For this input the output from the sort or
49
iter_topo_order routines will be:
52
node identifiers can be any hashable object, and are typically strings.
54
If you have a graph like [('a', ['b']), ('a', ['c'])] this will only use
55
one of the two values for 'a'.
57
The graph is sorted lazily: until you iterate or sort the input is
58
not processed other than to create an internal representation.
60
iteration or sorting may raise GraphCycleError if a cycle is present
63
# a dict of the graph.
64
self._graph = dict(graph)
65
self._visitable = set(self._graph)
67
# self._original_graph = dict(graph)
69
# this is a stack storing the depth first search into the graph.
70
self._node_name_stack = []
71
# at each level of 'recursion' we have to check each parent. This
72
# stack stores the parents we have not yet checked for the node at the
73
# matching depth in _node_name_stack
74
self._pending_parents_stack = []
75
# this is a set of the completed nodes for fast checking whether a
76
# parent in a node we are processing on the stack has already been
77
# emitted and thus can be skipped.
78
self._completed_node_names = set()
81
"""Sort the graph and return as a list.
83
After calling this the sorter is empty and you must create a new one.
85
return list(self.iter_topo_order())
87
### Useful if fiddling with this code.
89
### sorted_names = list(self.iter_topo_order())
90
### for index in range(len(sorted_names)):
91
### rev = sorted_names[index]
92
### for left_index in range(index):
93
### if rev in self.original_graph[sorted_names[left_index]]:
94
### print "revision in parent list of earlier revision"
95
### import pdb;pdb.set_trace()
97
def iter_topo_order(self):
98
"""Yield the nodes of the graph in a topological order.
100
After finishing iteration the sorter is empty and you cannot continue
104
# now pick a random node in the source graph, and transfer it to the
105
# top of the depth first search stack.
106
node_name, parents = self._graph.popitem()
107
self._push_node(node_name, parents)
108
while self._node_name_stack:
109
# loop until this call completes.
110
parents_to_visit = self._pending_parents_stack[-1]
111
# if all parents are done, the revision is done
112
if not parents_to_visit:
113
# append the revision to the topo sorted list
114
# all the nodes parents have been added to the output, now
115
# we can add it to the output.
116
yield self._pop_node()
118
while self._pending_parents_stack[-1]:
119
# recurse depth first into a single parent
120
next_node_name = self._pending_parents_stack[-1].pop()
121
if next_node_name in self._completed_node_names:
122
# this parent was completed by a child on the
123
# call stack. skip it.
125
if next_node_name not in self._visitable:
127
# otherwise transfer it from the source graph into the
128
# top of the current depth first search stack.
130
parents = self._graph.pop(next_node_name)
132
# if the next node is not in the source graph it has
133
# already been popped from it and placed into the
134
# current search stack (but not completed or we would
135
# have hit the continue 4 lines up.
136
# this indicates a cycle.
137
raise errors.GraphCycleError(self._node_name_stack)
138
self._push_node(next_node_name, parents)
139
# and do not continue processing parents until this 'call'
143
def _push_node(self, node_name, parents):
144
"""Add node_name to the pending node stack.
146
Names in this stack will get emitted into the output as they are popped
149
self._node_name_stack.append(node_name)
150
self._pending_parents_stack.append(list(parents))
153
"""Pop the top node off the stack
155
The node is appended to the sorted output.
157
# we are returning from the flattened call frame:
158
# pop off the local variables
159
node_name = self._node_name_stack.pop()
160
self._pending_parents_stack.pop()
162
self._completed_node_names.add(node_name)
166
def merge_sort(graph, branch_tip, mainline_revisions=None, generate_revno=False):
167
"""Topological sort a graph which groups merges.
169
:param graph: sequence of pairs of node->parents_list.
170
:param branch_tip: the tip of the branch to graph. Revisions not
171
reachable from branch_tip are not included in the
173
:param mainline_revisions: If not None this forces a mainline to be
174
used rather than synthesised from the graph.
175
This must be a valid path through some part
176
of the graph. If the mainline does not cover all
177
the revisions, output stops at the start of the
178
old revision listed in the mainline revisions
180
The order for this parameter is oldest-first.
181
:param generate_revno: Optional parameter controlling the generation of
182
revision number sequences in the output. See the output description of
183
the MergeSorter docstring for details.
184
:result: See the MergeSorter docstring for details.
185
node identifiers can be any hashable object, and are typically strings.
187
return MergeSorter(graph, branch_tip, mainline_revisions,
188
generate_revno).sorted()
191
class MergeSorter(object):
193
__slots__ = ['_node_name_stack',
194
'_node_merge_depth_stack',
195
'_pending_parents_stack',
196
'_first_child_stack',
197
'_left_subtree_pushed_stack',
200
'_mainline_revisions',
204
'_revno_to_branch_count',
205
'_completed_node_names',
209
def __init__(self, graph, branch_tip, mainline_revisions=None,
210
generate_revno=False):
211
"""Merge-aware topological sorting of a graph.
213
:param graph: sequence of pairs of node_name->parent_names_list.
214
i.e. [('C', ['B']), ('B', ['A']), ('A', [])]
215
For this input the output from the sort or
216
iter_topo_order routines will be:
218
:param branch_tip: the tip of the branch to graph. Revisions not
219
reachable from branch_tip are not included in the
221
:param mainline_revisions: If not None this forces a mainline to be
222
used rather than synthesised from the graph.
223
This must be a valid path through some part
224
of the graph. If the mainline does not cover all
225
the revisions, output stops at the start of the
226
old revision listed in the mainline revisions
228
The order for this parameter is oldest-first.
229
:param generate_revno: Optional parameter controlling the generation of
230
revision number sequences in the output. See the output description
233
The result is a list sorted so that all parents come before
234
their children. Each element of the list is a tuple containing:
235
(sequence_number, node_name, merge_depth, end_of_merge)
236
* sequence_number: The sequence of this row in the output. Useful for
238
* node_name: The node name: opaque text to the merge routine.
239
* merge_depth: How many levels of merging deep this node has been
241
* revno_sequence: When requested this field provides a sequence of
242
revision numbers for all revisions. The format is:
243
(REVNO, BRANCHNUM, BRANCHREVNO). BRANCHNUM is the number of the
244
branch that the revno is on. From left to right the REVNO numbers
245
are the sequence numbers within that branch of the revision.
246
For instance, the graph {A:[], B:['A'], C:['A', 'B']} will get
247
the following revno_sequences assigned: A:(1,), B:(1,1,1), C:(2,).
248
This should be read as 'A is the first commit in the trunk',
249
'B is the first commit on the first branch made from A', 'C is the
250
second commit in the trunk'.
251
* end_of_merge: When True the next node is part of a different merge.
254
node identifiers can be any hashable object, and are typically strings.
256
If you have a graph like [('a', ['b']), ('a', ['c'])] this will only use
257
one of the two values for 'a'.
259
The graph is sorted lazily: until you iterate or sort the input is
260
not processed other than to create an internal representation.
262
iteration or sorting may raise GraphCycleError if a cycle is present
265
Background information on the design:
266
-------------------------------------
267
definition: the end of any cluster or 'merge' occurs when:
268
1 - the next revision has a lower merge depth than we do.
275
C, D are the ends of clusters, E might be but we need more data.
276
2 - or the next revision at our merge depth is not our left most
278
This is required to handle multiple-merges in one commit.
286
C is the end of a cluster due to rule 1.
287
D is not the end of a cluster from rule 1, but is from rule 2: E
288
is not its left most ancestor
289
E is the end of a cluster due to rule 1
290
F might be but we need more data.
292
we show connecting lines to a parent when:
293
- The parent is the start of a merge within this cluster.
294
That is, the merge was not done to the mainline before this cluster
295
was merged to the mainline.
296
This can be detected thus:
297
* The parent has a higher merge depth and is the next revision in
300
The next revision in the list constraint is needed for this case:
302
B 1 [C, F] # we do not want to show a line to F which is depth 2
304
C 1 [H] # note that this is a long line to show back to the
305
ancestor - see the end of merge rules.
311
- Part of this merges 'branch':
312
The parent has the same merge depth and is our left most parent and we
313
are not the end of the cluster.
314
A 0 [C, B] lines: [B, C]
315
B 1 [E, C] lines: [C]
317
D 0 [F, E] lines: [E, F]
320
- The end of this merge/cluster:
321
we can ONLY have multiple parents at the end of a cluster if this
322
branch was previously merged into the 'mainline'.
323
- if we have one and only one parent, show it
324
Note that this may be to a greater merge depth - for instance if
325
this branch continued from a deeply nested branch to add something
327
- if we have more than one parent - show the second oldest (older ==
328
further down the list) parent with
329
an equal or lower merge depth
330
XXXX revisit when awake. ddaa asks about the relevance of each one
331
- maybe more than one parent is relevant
333
self._generate_revno = generate_revno
334
# a dict of the graph.
335
self._graph = dict(graph)
336
# if there is an explicit mainline, alter the graph to match. This is
337
# easier than checking at every merge whether we are on the mainline and
338
# if so which path to take.
339
if mainline_revisions is None:
340
self._mainline_revisions = []
341
self._stop_revision = None
343
self._mainline_revisions = list(mainline_revisions)
344
self._stop_revision = self._mainline_revisions[0]
345
# skip the first revision, its what we reach and its parents are
346
# therefore irrelevant
347
for index, revision in enumerate(self._mainline_revisions[1:]):
348
# NB: index 0 means self._mainline_revisions[1]
349
# if the mainline matches the graph, nothing to do.
350
parent = self._mainline_revisions[index]
352
# end of mainline_revisions history
354
if self._graph[revision][0] == parent:
356
# remove it from its prior spot
357
self._graph[revision].remove(parent)
358
# insert it into the start of the mainline
359
self._graph[revision].insert(0, parent)
360
# we need to do a check late in the process to detect end-of-merges
361
# which requires the parents to be accessible: its easier for now
362
# to just keep the original graph around.
363
self._original_graph = dict(self._graph.items())
364
# we need to know the revision numbers of revisions to determine
365
# the revision numbers of their descendants
366
# this is a graph from node to [revno_tuple, first_child]
367
# where first_child is True if no other children have seen this node
368
# and revno_tuple is the tuple that was assigned to the node.
369
# we dont know revnos to start with, so we start it seeded with
371
self._revnos = dict((revision, [None, True])
372
for revision in self._graph)
373
# Each mainline revision counts how many child branches have spawned from it.
374
self._revno_to_branch_count = {}
376
# this is a stack storing the depth first search into the graph.
377
self._node_name_stack = []
378
# at each level of recursion we need the merge depth this node is at:
379
self._node_merge_depth_stack = []
380
# at each level of 'recursion' we have to check each parent. This
381
# stack stores the parents we have not yet checked for the node at the
382
# matching depth in _node_name_stack
383
self._pending_parents_stack = []
384
# When we first look at a node we assign it a seqence number from its
386
self._first_child_stack = []
387
# this is a set of the nodes who have been completely analysed for fast
388
# membership checking
389
self._completed_node_names = set()
390
# this is the scheduling of nodes list.
391
# Nodes are scheduled
392
# from the bottom left of the tree: in the tree
399
# the scheduling order is: F, E, D, C, B, A
400
# that is - 'left subtree, right subtree, node'
401
# which would mean that when we schedule A we can emit the entire tree.
402
self._scheduled_nodes = []
403
# This records for each node when we have processed its left most
404
# unmerged subtree. After this subtree is scheduled, all other subtrees
405
# have their merge depth increased by one from this nodes merge depth.
406
# it contains tuples - name, merge_depth
407
self._left_subtree_pushed_stack = []
409
# seed the search with the tip of the branch
410
if (branch_tip is not None and
411
branch_tip != _mod_revision.NULL_REVISION):
412
parents = self._graph.pop(branch_tip)
413
self._push_node(branch_tip, 0, parents)
416
"""Sort the graph and return as a list.
418
After calling this the sorter is empty and you must create a new one.
420
return list(self.iter_topo_order())
422
def iter_topo_order(self):
423
"""Yield the nodes of the graph in a topological order.
425
After finishing iteration the sorter is empty and you cannot continue
428
# These are safe to offload to local variables, because they are used
429
# as a stack and modified in place, never assigned to.
430
node_name_stack = self._node_name_stack
431
node_merge_depth_stack = self._node_merge_depth_stack
432
pending_parents_stack = self._pending_parents_stack
433
left_subtree_pushed_stack = self._left_subtree_pushed_stack
434
completed_node_names = self._completed_node_names
435
scheduled_nodes = self._scheduled_nodes
437
graph_pop = self._graph.pop
439
def push_node(node_name, merge_depth, parents,
440
node_name_stack_append=node_name_stack.append,
441
node_merge_depth_stack_append=node_merge_depth_stack.append,
442
left_subtree_pushed_stack_append=left_subtree_pushed_stack.append,
443
pending_parents_stack_append=pending_parents_stack.append,
444
first_child_stack_append=self._first_child_stack.append,
447
"""Add node_name to the pending node stack.
449
Names in this stack will get emitted into the output as they are popped
452
This inlines a lot of self._variable.append functions as local
455
node_name_stack_append(node_name)
456
node_merge_depth_stack_append(merge_depth)
457
left_subtree_pushed_stack_append(False)
458
pending_parents_stack_append(list(parents))
459
# as we push it, check if it is the first child
461
# node has parents, assign from the left most parent.
462
parent_info = revnos[parents[0]]
463
first_child = parent_info[1]
464
parent_info[1] = False
466
# We don't use the same algorithm here, but we need to keep the
469
first_child_stack_append(first_child)
471
def pop_node(node_name_stack_pop=node_name_stack.pop,
472
node_merge_depth_stack_pop=node_merge_depth_stack.pop,
473
first_child_stack_pop=self._first_child_stack.pop,
474
left_subtree_pushed_stack_pop=left_subtree_pushed_stack.pop,
475
pending_parents_stack_pop=pending_parents_stack.pop,
476
original_graph=self._original_graph,
478
completed_node_names_add=self._completed_node_names.add,
479
scheduled_nodes_append=scheduled_nodes.append,
480
revno_to_branch_count=self._revno_to_branch_count,
482
"""Pop the top node off the stack
484
The node is appended to the sorted output.
486
# we are returning from the flattened call frame:
487
# pop off the local variables
488
node_name = node_name_stack_pop()
489
merge_depth = node_merge_depth_stack_pop()
490
first_child = first_child_stack_pop()
491
# remove this node from the pending lists:
492
left_subtree_pushed_stack_pop()
493
pending_parents_stack_pop()
495
parents = original_graph[node_name]
497
# node has parents, assign from the left most parent.
498
parent_revno = revnos[parents[0]][0]
500
# not the first child, make a new branch
501
base_revno = parent_revno[0]
502
branch_count = revno_to_branch_count.get(base_revno, 0)
504
revno_to_branch_count[base_revno] = branch_count
505
revno = (parent_revno[0], branch_count, 1)
506
# revno = (parent_revno[0], branch_count, parent_revno[-1]+1)
508
# as the first child, we just increase the final revision
510
revno = parent_revno[:-1] + (parent_revno[-1] + 1,)
512
# no parents, use the root sequence
513
root_count = revno_to_branch_count.get(0, 0)
515
revno = (0, root_count, 1)
519
revno_to_branch_count[0] = root_count
521
# store the revno for this node for future reference
522
revnos[node_name][0] = revno
523
completed_node_names_add(node_name)
524
scheduled_nodes_append((node_name, merge_depth, revno))
528
while node_name_stack:
529
# loop until this call completes.
530
parents_to_visit = pending_parents_stack[-1]
531
# if all parents are done, the revision is done
532
if not parents_to_visit:
533
# append the revision to the topo sorted scheduled list:
534
# all the nodes parents have been scheduled added, now
535
# we can add it to the output.
538
while pending_parents_stack[-1]:
539
if not left_subtree_pushed_stack[-1]:
540
# recurse depth first into the primary parent
541
next_node_name = pending_parents_stack[-1].pop(0)
543
# place any merges in right-to-left order for scheduling
544
# which gives us left-to-right order after we reverse
545
# the scheduled queue. XXX: This has the effect of
546
# allocating common-new revisions to the right-most
547
# subtree rather than the left most, which will
548
# display nicely (you get smaller trees at the top
549
# of the combined merge).
550
next_node_name = pending_parents_stack[-1].pop()
551
if next_node_name in completed_node_names:
552
# this parent was completed by a child on the
553
# call stack. skip it.
555
# otherwise transfer it from the source graph into the
556
# top of the current depth first search stack.
558
parents = graph_pop(next_node_name)
560
# if the next node is not in the source graph it has
561
# already been popped from it and placed into the
562
# current search stack (but not completed or we would
563
# have hit the continue 4 lines up.
564
# this indicates a cycle.
565
raise errors.GraphCycleError(node_name_stack)
567
if left_subtree_pushed_stack[-1]:
568
# a new child branch from name_stack[-1]
572
left_subtree_pushed_stack[-1] = True
574
node_merge_depth_stack[-1] + next_merge_depth)
579
# and do not continue processing parents until this 'call'
583
# We have scheduled the graph. Now deliver the ordered output:
585
stop_revision = self._stop_revision
586
generate_revno = self._generate_revno
587
original_graph = self._original_graph
589
while scheduled_nodes:
590
node_name, merge_depth, revno = scheduled_nodes.pop()
591
if node_name == stop_revision:
593
if not len(scheduled_nodes):
594
# last revision is the end of a merge
596
elif scheduled_nodes[-1][1] < merge_depth:
597
# the next node is to our left
599
elif (scheduled_nodes[-1][1] == merge_depth and
600
(scheduled_nodes[-1][0] not in
601
original_graph[node_name])):
602
# the next node was part of a multiple-merge.
607
yield (sequence_number, node_name, merge_depth, revno, end_of_merge)
609
yield (sequence_number, node_name, merge_depth, end_of_merge)
612
def _push_node(self, node_name, merge_depth, parents):
613
"""Add node_name to the pending node stack.
615
Names in this stack will get emitted into the output as they are popped
618
self._node_name_stack.append(node_name)
619
self._node_merge_depth_stack.append(merge_depth)
620
self._left_subtree_pushed_stack.append(False)
621
self._pending_parents_stack.append(list(parents))
622
# as we push it, figure out if this is the first child
623
parents = self._original_graph[node_name]
625
# node has parents, assign from the left most parent.
626
parent_info = self._revnos[parents[0]]
627
first_child = parent_info[1]
628
parent_info[1] = False
630
# We don't use the same algorithm here, but we need to keep the
633
self._first_child_stack.append(first_child)
636
"""Pop the top node off the stack
638
The node is appended to the sorted output.
640
# we are returning from the flattened call frame:
641
# pop off the local variables
642
node_name = self._node_name_stack.pop()
643
merge_depth = self._node_merge_depth_stack.pop()
644
first_child = self._first_child_stack.pop()
645
# remove this node from the pending lists:
646
self._left_subtree_pushed_stack.pop()
647
self._pending_parents_stack.pop()
649
parents = self._original_graph[node_name]
651
# node has parents, assign from the left most parent.
652
parent_revno = self._revnos[parents[0]][0]
654
# not the first child, make a new branch
655
base_revno = parent_revno[0]
656
branch_count = self._revno_to_branch_count.get(base_revno, 0)
658
self._revno_to_branch_count[base_revno] = branch_count
659
revno = (parent_revno[0], branch_count, 1)
660
# revno = (parent_revno[0], branch_count, parent_revno[-1]+1)
662
# as the first child, we just increase the final revision
664
revno = parent_revno[:-1] + (parent_revno[-1] + 1,)
666
# no parents, use the root sequence
667
root_count = self._revno_to_branch_count.get(0, 0)
669
revno = (0, root_count, 1)
673
self._revno_to_branch_count[0] = root_count
675
# store the revno for this node for future reference
676
self._revnos[node_name][0] = revno
677
self._completed_node_names.add(node_name)
678
self._scheduled_nodes.append((node_name, merge_depth, self._revnos[node_name][0]))
added
removed
bzrlib/tsort.py
