26
32
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)
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):
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.
178
The result is a list of node names, such that all parents come before
181
node identifiers can be any hashable object, and are typically strings.
183
return MergeSorter(graph, branch_tip, mainline_revisions).sorted()
186
class MergeSorter(object):
188
def __init__(self, graph, branch_tip, mainline_revisions=None):
189
"""Merge-aware topological sorting of a graph.
191
:param graph: sequence of pairs of node_name->parent_names_list.
192
i.e. [('C', ['B']), ('B', ['A']), ('A', [])]
193
For this input the output from the sort or
194
iter_topo_order routines will be:
196
:param branch_tip: the tip of the branch to graph. Revisions not
197
reachable from branch_tip are not included in the
199
:param mainline_revisions: If not None this forces a mainline to be
200
used rather than synthesised from the graph.
201
This must be a valid path through some part
202
of the graph. If the mainline does not cover all
203
the revisions, output stops at the start of the
204
old revision listed in the mainline revisions
206
The order for this parameter is oldest-first.
209
node identifiers can be any hashable object, and are typically strings.
211
If you have a graph like [('a', ['b']), ('a', ['c'])] this will only use
212
one of the two values for 'a'.
214
The graph is sorted lazily: until you iterate or sort the input is
215
not processed other than to create an internal representation.
217
iteration or sorting may raise GraphCycleError if a cycle is present
220
Background information on the design:
221
-------------------------------------
222
definition: the end of any cluster or 'merge' occurs when:
223
1 - the next revision has a lower merge depth than we do.
230
C, D are the ends of clusters, E might be but we need more data.
231
2 - or the next revision at our merge depth is not our left most
233
This is required to handle multiple-merges in one commit.
241
C is the end of a cluster due to rule 1.
242
D is not the end of a cluster from rule 1, but is from rule 2: E
243
is not its left most ancestor
244
E is the end of a cluster due to rule 1
245
F might be but we need more data.
247
we show connecting lines to a parent when:
248
- The parent is the start of a merge within this cluster.
249
That is, the merge was not done to the mainline before this cluster
250
was merged to the mainline.
251
This can be detected thus:
252
* The parent has a higher merge depth and is the next revision in
255
The next revision in the list constraint is needed for this case:
257
B 1 [C, F] # we do not want to show a line to F which is depth 2
259
C 1 [H] # note that this is a long line to show back to the
260
ancestor - see the end of merge rules.
266
- Part of this merges 'branch':
267
The parent has the same merge depth and is our left most parent and we
268
are not the end of the cluster.
269
A 0 [C, B] lines: [B, C]
270
B 1 [E, C] lines: [C]
272
D 0 [F, E] lines: [E, F]
275
- The end of this merge/cluster:
276
we can ONLY have multiple parents at the end of a cluster if this
277
branch was previously merged into the 'mainline'.
278
- if we have one and only one parent, show it
279
Note that this may be to a greater merge depth - for instance if
280
this branch continued from a deeply nested branch to add something
282
- if we have more than one parent - show the second oldest (older ==
283
further down the list) parent with
284
an equal or lower merge depth
285
XXXX revisit when awake. ddaa asks about the relevance of each one
286
- maybe more than one parent is relevant
288
# a dict of the graph.
289
self._graph = dict(graph)
290
# if there is an explicit mainline, alter the graph to match. This is
291
# easier than checking at every merge whether we are on the mainline and
292
# if so which path to take.
293
if mainline_revisions is None:
294
self._mainline_revisions = []
295
self._stop_revision = None
297
self._mainline_revisions = list(mainline_revisions)
298
self._stop_revision = self._mainline_revisions[0]
299
# skip the first revision, its what we reach and its parents are
300
# therefore irrelevant
301
for index, revision in enumerate(self._mainline_revisions[1:]):
302
# NB: index 0 means self._mainline_revisions[1]
303
# if the mainline matches the graph, nothing to do.
304
parent = self._mainline_revisions[index]
306
# end of mainline_revisions history
308
if self._graph[revision][0] == parent:
310
# remove it from its prior spot
311
self._graph[revision].remove(parent)
312
# insert it into the start of the mainline
313
self._graph[revision].insert(0, parent)
314
# we need to do a check late in the process to detect end-of-merges
315
# which requires the parents to be accessible: its easier for now
316
# to just keep the original graph around.
317
self._original_graph = dict(self._graph.items())
319
# this is a stack storing the depth first search into the graph.
320
self._node_name_stack = []
321
# at each level of recursion we need the merge depth this node is at:
322
self._node_merge_depth_stack = []
323
# at each level of 'recursion' we have to check each parent. This
324
# stack stores the parents we have not yet checked for the node at the
325
# matching depth in _node_name_stack
326
self._pending_parents_stack = []
327
# this is a set of the nodes who have been completely analysed for fast
328
# membership checking
329
self._completed_node_names = set()
330
# this is the scheduling of nodes list.
331
# Nodes are scheduled
332
# from the bottom left of the tree: in the tree
339
# the scheduling order is: F, E, D, C, B, A
340
# that is - 'left subtree, right subtree, node'
341
# which would mean that when we schedule A we can emit the entire tree.
342
self._scheduled_nodes = []
343
# This records for each node when we have processed its left most
344
# unmerged subtree. After this subtree is scheduled, all other subtrees
345
# have their merge depth increased by one from this nodes merge depth.
346
self._left_subtree_done_stack = []
348
# seed the search with the tip of the branch
349
if branch_tip is not None:
350
parents = self._graph.pop(branch_tip)
351
self._push_node(branch_tip, 0, parents)
354
"""Sort the graph and return as a list.
356
After calling this the sorter is empty and you must create a new one.
358
return list(self.iter_topo_order())
360
def iter_topo_order(self):
361
"""Yield the nodes of the graph in a topological order.
363
After finishing iteration the sorter is empty and you cannot continue
366
while self._node_name_stack:
367
# loop until this call completes.
368
parents_to_visit = self._pending_parents_stack[-1]
369
# if all parents are done, the revision is done
370
if not parents_to_visit:
371
# append the revision to the topo sorted scheduled list:
372
# all the nodes parents have been scheduled added, now
373
# we can add it to the output.
376
while self._pending_parents_stack[-1]:
377
if not self._left_subtree_done_stack[-1]:
378
# recurse depth first into the primary parent
379
next_node_name = self._pending_parents_stack[-1].pop(0)
381
# place any merges in right-to-left order for scheduling
382
# which gives us left-to-right order after we reverse
383
# the scheduled queue. XXX: This has the effect of
384
# allocating common-new revisions to the right-most
385
# subtree rather than the left most, which will
386
# display nicely (you get smaller trees at the top
387
# of the combined merge).
388
next_node_name = self._pending_parents_stack[-1].pop()
389
if next_node_name in self._completed_node_names:
390
# this parent was completed by a child on the
391
# call stack. skip it.
393
# otherwise transfer it from the source graph into the
394
# top of the current depth first search stack.
396
parents = self._graph.pop(next_node_name)
398
# if the next node is not in the source graph it has
399
# already been popped from it and placed into the
400
# current search stack (but not completed or we would
401
# have hit the continue 4 lines up.
402
# this indicates a cycle.
403
raise errors.GraphCycleError(self._node_name_stack)
405
if self._left_subtree_done_stack[-1]:
409
self._left_subtree_done_stack[-1] = True
411
self._node_merge_depth_stack[-1] + next_merge_depth)
416
# and do not continue processing parents until this 'call'
419
# We have scheduled the graph. Now deliver the ordered output:
421
while self._scheduled_nodes:
422
node_name, merge_depth = self._scheduled_nodes.pop()
423
if node_name == self._stop_revision:
425
if not len(self._scheduled_nodes):
427
elif self._scheduled_nodes[-1][1] < merge_depth:
428
# the next node is to our left
430
elif (self._scheduled_nodes[-1][1] == merge_depth and
431
(self._scheduled_nodes[-1][0] not in
432
self._original_graph[node_name])):
433
# the next node was part of a multiple-merge.
437
yield (sequence_number, node_name, merge_depth, end_of_merge)
440
def _push_node(self, node_name, merge_depth, parents):
441
"""Add node_name to the pending node stack.
443
Names in this stack will get emitted into the output as they are popped
446
self._node_name_stack.append(node_name)
447
self._node_merge_depth_stack.append(merge_depth)
448
self._left_subtree_done_stack.append(False)
449
self._pending_parents_stack.append(list(parents))
452
"""Pop the top node off the stack
454
The node is appended to the sorted output.
456
# we are returning from the flattened call frame:
457
# pop off the local variables
458
node_name = self._node_name_stack.pop()
459
merge_depth = self._node_merge_depth_stack.pop()
460
self._left_subtree_done_stack.pop()
461
self._pending_parents_stack.pop()
463
self._completed_node_names.add(node_name)
464
self._scheduled_nodes.append((node_name, merge_depth))
added
removed
bzrlib/tsort.py
