1
# Copyright (C) 2007 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
28
STEP_UNIQUE_SEARCHER_EVERY = 5
30
# DIAGRAM of terminology
40
# In this diagram, relative to G and H:
41
# A, B, C, D, E are common ancestors.
42
# C, D and E are border ancestors, because each has a non-common descendant.
43
# D and E are least common ancestors because none of their descendants are
45
# C is not a least common ancestor because its descendant, E, is a common
48
# The find_unique_lca algorithm will pick A in two steps:
49
# 1. find_lca('G', 'H') => ['D', 'E']
50
# 2. Since len(['D', 'E']) > 1, find_lca('D', 'E') => ['A']
53
class DictParentsProvider(object):
54
"""A parents provider for Graph objects."""
56
def __init__(self, ancestry):
57
self.ancestry = ancestry
60
return 'DictParentsProvider(%r)' % self.ancestry
62
def get_parent_map(self, keys):
63
"""See _StackedParentsProvider.get_parent_map"""
64
ancestry = self.ancestry
65
return dict((k, ancestry[k]) for k in keys if k in ancestry)
68
class _StackedParentsProvider(object):
70
def __init__(self, parent_providers):
71
self._parent_providers = parent_providers
74
return "_StackedParentsProvider(%r)" % self._parent_providers
76
def get_parent_map(self, keys):
77
"""Get a mapping of keys => parents
79
A dictionary is returned with an entry for each key present in this
80
source. If this source doesn't have information about a key, it should
83
[NULL_REVISION] is used as the parent of the first user-committed
84
revision. Its parent list is empty.
86
:param keys: An iterable returning keys to check (eg revision_ids)
87
:return: A dictionary mapping each key to its parents
91
for parents_provider in self._parent_providers:
92
new_found = parents_provider.get_parent_map(remaining)
93
found.update(new_found)
94
remaining.difference_update(new_found)
100
class CachingParentsProvider(object):
101
"""A parents provider which will cache the revision => parents as a dict.
103
This is useful for providers which have an expensive look up.
105
Either a ParentsProvider or a get_parent_map-like callback may be
106
supplied. If it provides extra un-asked-for parents, they will be cached,
107
but filtered out of get_parent_map.
109
The cache is enabled by default, but may be disabled and re-enabled.
111
def __init__(self, parent_provider=None, get_parent_map=None):
114
:param parent_provider: The ParentProvider to use. It or
115
get_parent_map must be supplied.
116
:param get_parent_map: The get_parent_map callback to use. It or
117
parent_provider must be supplied.
119
self._real_provider = parent_provider
120
if get_parent_map is None:
121
self._get_parent_map = self._real_provider.get_parent_map
123
self._get_parent_map = get_parent_map
125
self._cache_misses = True
128
return "%s(%r)" % (self.__class__.__name__, self._real_provider)
130
def enable_cache(self, cache_misses=True):
132
if self._cache is not None:
133
raise AssertionError('Cache enabled when already enabled.')
135
self._cache_misses = cache_misses
137
def disable_cache(self):
138
"""Disable and clear the cache."""
141
def get_cached_map(self):
142
"""Return any cached get_parent_map values."""
143
if self._cache is None:
145
return dict((k, v) for k, v in self._cache.items()
148
def get_parent_map(self, keys):
149
"""See _StackedParentsProvider.get_parent_map."""
150
# Hack to build up the caching logic.
151
ancestry = self._cache
153
# Caching is disabled.
154
missing_revisions = set(keys)
157
missing_revisions = set(key for key in keys if key not in ancestry)
158
if missing_revisions:
159
parent_map = self._get_parent_map(missing_revisions)
160
ancestry.update(parent_map)
161
if self._cache_misses:
162
# None is never a valid parents list, so it can be used to
164
ancestry.update(dict((k, None) for k in missing_revisions
165
if k not in parent_map))
166
present_keys = [k for k in keys if ancestry.get(k) is not None]
167
return dict((k, ancestry[k]) for k in present_keys)
171
"""Provide incremental access to revision graphs.
173
This is the generic implementation; it is intended to be subclassed to
174
specialize it for other repository types.
177
def __init__(self, parents_provider):
178
"""Construct a Graph that uses several graphs as its input
180
This should not normally be invoked directly, because there may be
181
specialized implementations for particular repository types. See
182
Repository.get_graph().
184
:param parents_provider: An object providing a get_parent_map call
185
conforming to the behavior of
186
StackedParentsProvider.get_parent_map.
188
if getattr(parents_provider, 'get_parents', None) is not None:
189
self.get_parents = parents_provider.get_parents
190
if getattr(parents_provider, 'get_parent_map', None) is not None:
191
self.get_parent_map = parents_provider.get_parent_map
192
self._parents_provider = parents_provider
195
return 'Graph(%r)' % self._parents_provider
197
def find_lca(self, *revisions):
198
"""Determine the lowest common ancestors of the provided revisions
200
A lowest common ancestor is a common ancestor none of whose
201
descendants are common ancestors. In graphs, unlike trees, there may
202
be multiple lowest common ancestors.
204
This algorithm has two phases. Phase 1 identifies border ancestors,
205
and phase 2 filters border ancestors to determine lowest common
208
In phase 1, border ancestors are identified, using a breadth-first
209
search starting at the bottom of the graph. Searches are stopped
210
whenever a node or one of its descendants is determined to be common
212
In phase 2, the border ancestors are filtered to find the least
213
common ancestors. This is done by searching the ancestries of each
216
Phase 2 is perfomed on the principle that a border ancestor that is
217
not an ancestor of any other border ancestor is a least common
220
Searches are stopped when they find a node that is determined to be a
221
common ancestor of all border ancestors, because this shows that it
222
cannot be a descendant of any border ancestor.
224
The scaling of this operation should be proportional to
225
1. The number of uncommon ancestors
226
2. The number of border ancestors
227
3. The length of the shortest path between a border ancestor and an
228
ancestor of all border ancestors.
230
border_common, common, sides = self._find_border_ancestors(revisions)
231
# We may have common ancestors that can be reached from each other.
232
# - ask for the heads of them to filter it down to only ones that
233
# cannot be reached from each other - phase 2.
234
return self.heads(border_common)
236
def find_difference(self, left_revision, right_revision):
237
"""Determine the graph difference between two revisions"""
238
border, common, searchers = self._find_border_ancestors(
239
[left_revision, right_revision])
240
self._search_for_extra_common(common, searchers)
241
left = searchers[0].seen
242
right = searchers[1].seen
243
return (left.difference(right), right.difference(left))
245
def find_distance_to_null(self, target_revision_id, known_revision_ids):
246
"""Find the left-hand distance to the NULL_REVISION.
248
(This can also be considered the revno of a branch at
251
:param target_revision_id: A revision_id which we would like to know
253
:param known_revision_ids: [(revision_id, revno)] A list of known
254
revno, revision_id tuples. We'll use this to seed the search.
256
# Map from revision_ids to a known value for their revno
257
known_revnos = dict(known_revision_ids)
258
cur_tip = target_revision_id
260
NULL_REVISION = revision.NULL_REVISION
261
known_revnos[NULL_REVISION] = 0
263
searching_known_tips = list(known_revnos.keys())
265
unknown_searched = {}
267
while cur_tip not in known_revnos:
268
unknown_searched[cur_tip] = num_steps
270
to_search = set([cur_tip])
271
to_search.update(searching_known_tips)
272
parent_map = self.get_parent_map(to_search)
273
parents = parent_map.get(cur_tip, None)
274
if not parents: # An empty list or None is a ghost
275
raise errors.GhostRevisionsHaveNoRevno(target_revision_id,
279
for revision_id in searching_known_tips:
280
parents = parent_map.get(revision_id, None)
284
next_revno = known_revnos[revision_id] - 1
285
if next in unknown_searched:
286
# We have enough information to return a value right now
287
return next_revno + unknown_searched[next]
288
if next in known_revnos:
290
known_revnos[next] = next_revno
291
next_known_tips.append(next)
292
searching_known_tips = next_known_tips
294
# We reached a known revision, so just add in how many steps it took to
296
return known_revnos[cur_tip] + num_steps
298
def find_unique_ancestors(self, unique_revision, common_revisions):
299
"""Find the unique ancestors for a revision versus others.
301
This returns the ancestry of unique_revision, excluding all revisions
302
in the ancestry of common_revisions. If unique_revision is in the
303
ancestry, then the empty set will be returned.
305
:param unique_revision: The revision_id whose ancestry we are
307
XXX: Would this API be better if we allowed multiple revisions on
309
:param common_revisions: Revision_ids of ancestries to exclude.
310
:return: A set of revisions in the ancestry of unique_revision
312
if unique_revision in common_revisions:
315
# Algorithm description
316
# 1) Walk backwards from the unique node and all common nodes.
317
# 2) When a node is seen by both sides, stop searching it in the unique
318
# walker, include it in the common walker.
319
# 3) Stop searching when there are no nodes left for the unique walker.
320
# At this point, you have a maximal set of unique nodes. Some of
321
# them may actually be common, and you haven't reached them yet.
322
# 4) Start new searchers for the unique nodes, seeded with the
323
# information you have so far.
324
# 5) Continue searching, stopping the common searches when the search
325
# tip is an ancestor of all unique nodes.
326
# 6) Aggregate together unique searchers when they are searching the
327
# same tips. When all unique searchers are searching the same node,
328
# stop move it to a single 'all_unique_searcher'.
329
# 7) The 'all_unique_searcher' represents the very 'tip' of searching.
330
# Most of the time this produces very little important information.
331
# So don't step it as quickly as the other searchers.
332
# 8) Search is done when all common searchers have completed.
334
unique_searcher, common_searcher = self._find_initial_unique_nodes(
335
[unique_revision], common_revisions)
337
unique_nodes = unique_searcher.seen.difference(common_searcher.seen)
341
(all_unique_searcher,
342
unique_tip_searchers) = self._make_unique_searchers(unique_nodes,
343
unique_searcher, common_searcher)
345
self._refine_unique_nodes(unique_searcher, all_unique_searcher,
346
unique_tip_searchers, common_searcher)
347
true_unique_nodes = unique_nodes.difference(common_searcher.seen)
348
if 'graph' in debug.debug_flags:
349
trace.mutter('Found %d truly unique nodes out of %d',
350
len(true_unique_nodes), len(unique_nodes))
351
return true_unique_nodes
353
def _find_initial_unique_nodes(self, unique_revisions, common_revisions):
354
"""Steps 1-3 of find_unique_ancestors.
356
Find the maximal set of unique nodes. Some of these might actually
357
still be common, but we are sure that there are no other unique nodes.
359
:return: (unique_searcher, common_searcher)
362
unique_searcher = self._make_breadth_first_searcher(unique_revisions)
363
# we know that unique_revisions aren't in common_revisions, so skip
365
unique_searcher.next()
366
common_searcher = self._make_breadth_first_searcher(common_revisions)
368
# As long as we are still finding unique nodes, keep searching
369
while unique_searcher._next_query:
370
next_unique_nodes = set(unique_searcher.step())
371
next_common_nodes = set(common_searcher.step())
373
# Check if either searcher encounters new nodes seen by the other
375
unique_are_common_nodes = next_unique_nodes.intersection(
376
common_searcher.seen)
377
unique_are_common_nodes.update(
378
next_common_nodes.intersection(unique_searcher.seen))
379
if unique_are_common_nodes:
380
ancestors = unique_searcher.find_seen_ancestors(
381
unique_are_common_nodes)
382
# TODO: This is a bit overboard, we only really care about
383
# the ancestors of the tips because the rest we
384
# already know. This is *correct* but causes us to
385
# search too much ancestry.
386
ancestors.update(common_searcher.find_seen_ancestors(ancestors))
387
unique_searcher.stop_searching_any(ancestors)
388
common_searcher.start_searching(ancestors)
390
return unique_searcher, common_searcher
392
def _make_unique_searchers(self, unique_nodes, unique_searcher,
394
"""Create a searcher for all the unique search tips (step 4).
396
As a side effect, the common_searcher will stop searching any nodes
397
that are ancestors of the unique searcher tips.
399
:return: (all_unique_searcher, unique_tip_searchers)
401
unique_tips = self._remove_simple_descendants(unique_nodes,
402
self.get_parent_map(unique_nodes))
404
if len(unique_tips) == 1:
405
unique_tip_searchers = []
406
ancestor_all_unique = unique_searcher.find_seen_ancestors(unique_tips)
408
unique_tip_searchers = []
409
for tip in unique_tips:
410
revs_to_search = unique_searcher.find_seen_ancestors([tip])
411
revs_to_search.update(
412
common_searcher.find_seen_ancestors(revs_to_search))
413
searcher = self._make_breadth_first_searcher(revs_to_search)
414
# We don't care about the starting nodes.
415
searcher._label = tip
417
unique_tip_searchers.append(searcher)
419
ancestor_all_unique = None
420
for searcher in unique_tip_searchers:
421
if ancestor_all_unique is None:
422
ancestor_all_unique = set(searcher.seen)
424
ancestor_all_unique = ancestor_all_unique.intersection(
426
# Collapse all the common nodes into a single searcher
427
all_unique_searcher = self._make_breadth_first_searcher(
429
if ancestor_all_unique:
430
# We've seen these nodes in all the searchers, so we'll just go to
432
all_unique_searcher.step()
434
# Stop any search tips that are already known as ancestors of the
436
stopped_common = common_searcher.stop_searching_any(
437
common_searcher.find_seen_ancestors(ancestor_all_unique))
440
for searcher in unique_tip_searchers:
441
total_stopped += len(searcher.stop_searching_any(
442
searcher.find_seen_ancestors(ancestor_all_unique)))
443
if 'graph' in debug.debug_flags:
444
trace.mutter('For %d unique nodes, created %d + 1 unique searchers'
445
' (%d stopped search tips, %d common ancestors'
446
' (%d stopped common)',
447
len(unique_nodes), len(unique_tip_searchers),
448
total_stopped, len(ancestor_all_unique),
450
return all_unique_searcher, unique_tip_searchers
452
def _step_unique_and_common_searchers(self, common_searcher,
453
unique_tip_searchers,
455
"""Step all the searchers"""
456
newly_seen_common = set(common_searcher.step())
457
newly_seen_unique = set()
458
for searcher in unique_tip_searchers:
459
next = set(searcher.step())
460
next.update(unique_searcher.find_seen_ancestors(next))
461
next.update(common_searcher.find_seen_ancestors(next))
462
for alt_searcher in unique_tip_searchers:
463
if alt_searcher is searcher:
465
next.update(alt_searcher.find_seen_ancestors(next))
466
searcher.start_searching(next)
467
newly_seen_unique.update(next)
468
return newly_seen_common, newly_seen_unique
470
def _find_nodes_common_to_all_unique(self, unique_tip_searchers,
472
newly_seen_unique, step_all_unique):
473
"""Find nodes that are common to all unique_tip_searchers.
475
If it is time, step the all_unique_searcher, and add its nodes to the
478
common_to_all_unique_nodes = newly_seen_unique.copy()
479
for searcher in unique_tip_searchers:
480
common_to_all_unique_nodes.intersection_update(searcher.seen)
481
common_to_all_unique_nodes.intersection_update(
482
all_unique_searcher.seen)
483
# Step all-unique less frequently than the other searchers.
484
# In the common case, we don't need to spider out far here, so
485
# avoid doing extra work.
487
tstart = time.clock()
488
nodes = all_unique_searcher.step()
489
common_to_all_unique_nodes.update(nodes)
490
if 'graph' in debug.debug_flags:
491
tdelta = time.clock() - tstart
492
trace.mutter('all_unique_searcher step() took %.3fs'
493
'for %d nodes (%d total), iteration: %s',
494
tdelta, len(nodes), len(all_unique_searcher.seen),
495
all_unique_searcher._iterations)
496
return common_to_all_unique_nodes
498
def _collapse_unique_searchers(self, unique_tip_searchers,
499
common_to_all_unique_nodes):
500
"""Combine searchers that are searching the same tips.
502
When two searchers are searching the same tips, we can stop one of the
503
searchers. We also know that the maximal set of common ancestors is the
504
intersection of the two original searchers.
506
:return: A list of searchers that are searching unique nodes.
508
# Filter out searchers that don't actually search different
509
# nodes. We already have the ancestry intersection for them
510
unique_search_tips = {}
511
for searcher in unique_tip_searchers:
512
stopped = searcher.stop_searching_any(common_to_all_unique_nodes)
513
will_search_set = frozenset(searcher._next_query)
514
if not will_search_set:
515
if 'graph' in debug.debug_flags:
516
trace.mutter('Unique searcher %s was stopped.'
517
' (%s iterations) %d nodes stopped',
519
searcher._iterations,
521
elif will_search_set not in unique_search_tips:
522
# This searcher is searching a unique set of nodes, let it
523
unique_search_tips[will_search_set] = [searcher]
525
unique_search_tips[will_search_set].append(searcher)
526
# TODO: it might be possible to collapse searchers faster when they
527
# only have *some* search tips in common.
528
next_unique_searchers = []
529
for searchers in unique_search_tips.itervalues():
530
if len(searchers) == 1:
531
# Searching unique tips, go for it
532
next_unique_searchers.append(searchers[0])
534
# These searchers have started searching the same tips, we
535
# don't need them to cover the same ground. The
536
# intersection of their ancestry won't change, so create a
537
# new searcher, combining their histories.
538
next_searcher = searchers[0]
539
for searcher in searchers[1:]:
540
next_searcher.seen.intersection_update(searcher.seen)
541
if 'graph' in debug.debug_flags:
542
trace.mutter('Combining %d searchers into a single'
543
' searcher searching %d nodes with'
546
len(next_searcher._next_query),
547
len(next_searcher.seen))
548
next_unique_searchers.append(next_searcher)
549
return next_unique_searchers
551
def _refine_unique_nodes(self, unique_searcher, all_unique_searcher,
552
unique_tip_searchers, common_searcher):
553
"""Steps 5-8 of find_unique_ancestors.
555
This function returns when common_searcher has stopped searching for
558
# We step the ancestor_all_unique searcher only every
559
# STEP_UNIQUE_SEARCHER_EVERY steps.
560
step_all_unique_counter = 0
561
# While we still have common nodes to search
562
while common_searcher._next_query:
564
newly_seen_unique) = self._step_unique_and_common_searchers(
565
common_searcher, unique_tip_searchers, unique_searcher)
566
# These nodes are common ancestors of all unique nodes
567
common_to_all_unique_nodes = self._find_nodes_common_to_all_unique(
568
unique_tip_searchers, all_unique_searcher, newly_seen_unique,
569
step_all_unique_counter==0)
570
step_all_unique_counter = ((step_all_unique_counter + 1)
571
% STEP_UNIQUE_SEARCHER_EVERY)
573
if newly_seen_common:
574
# If a 'common' node is an ancestor of all unique searchers, we
575
# can stop searching it.
576
common_searcher.stop_searching_any(
577
all_unique_searcher.seen.intersection(newly_seen_common))
578
if common_to_all_unique_nodes:
579
common_to_all_unique_nodes.update(
580
common_searcher.find_seen_ancestors(
581
common_to_all_unique_nodes))
582
# The all_unique searcher can start searching the common nodes
583
# but everyone else can stop.
584
# This is the sort of thing where we would like to not have it
585
# start_searching all of the nodes, but only mark all of them
586
# as seen, and have it search only the actual tips. Otherwise
587
# it is another get_parent_map() traversal for it to figure out
588
# what we already should know.
589
all_unique_searcher.start_searching(common_to_all_unique_nodes)
590
common_searcher.stop_searching_any(common_to_all_unique_nodes)
592
next_unique_searchers = self._collapse_unique_searchers(
593
unique_tip_searchers, common_to_all_unique_nodes)
594
if len(unique_tip_searchers) != len(next_unique_searchers):
595
if 'graph' in debug.debug_flags:
596
trace.mutter('Collapsed %d unique searchers => %d'
598
len(unique_tip_searchers),
599
len(next_unique_searchers),
600
all_unique_searcher._iterations)
601
unique_tip_searchers = next_unique_searchers
603
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
604
def get_parents(self, revisions):
605
"""Find revision ids of the parents of a list of revisions
607
A list is returned of the same length as the input. Each entry
608
is a list of parent ids for the corresponding input revision.
610
[NULL_REVISION] is used as the parent of the first user-committed
611
revision. Its parent list is empty.
613
If the revision is not present (i.e. a ghost), None is used in place
614
of the list of parents.
616
Deprecated in bzr 1.2 - please see get_parent_map.
618
parents = self.get_parent_map(revisions)
619
return [parents.get(r, None) for r in revisions]
621
def get_parent_map(self, revisions):
622
"""Get a map of key:parent_list for revisions.
624
This implementation delegates to get_parents, for old parent_providers
625
that do not supply get_parent_map.
628
for rev, parents in self.get_parents(revisions):
629
if parents is not None:
630
result[rev] = parents
633
def _make_breadth_first_searcher(self, revisions):
634
return _BreadthFirstSearcher(revisions, self)
636
def _find_border_ancestors(self, revisions):
637
"""Find common ancestors with at least one uncommon descendant.
639
Border ancestors are identified using a breadth-first
640
search starting at the bottom of the graph. Searches are stopped
641
whenever a node or one of its descendants is determined to be common.
643
This will scale with the number of uncommon ancestors.
645
As well as the border ancestors, a set of seen common ancestors and a
646
list of sets of seen ancestors for each input revision is returned.
647
This allows calculation of graph difference from the results of this
650
if None in revisions:
651
raise errors.InvalidRevisionId(None, self)
652
common_ancestors = set()
653
searchers = [self._make_breadth_first_searcher([r])
655
active_searchers = searchers[:]
656
border_ancestors = set()
660
for searcher in searchers:
661
new_ancestors = searcher.step()
663
newly_seen.update(new_ancestors)
665
for revision in newly_seen:
666
if revision in common_ancestors:
667
# Not a border ancestor because it was seen as common
669
new_common.add(revision)
671
for searcher in searchers:
672
if revision not in searcher.seen:
675
# This is a border because it is a first common that we see
676
# after walking for a while.
677
border_ancestors.add(revision)
678
new_common.add(revision)
680
for searcher in searchers:
681
new_common.update(searcher.find_seen_ancestors(new_common))
682
for searcher in searchers:
683
searcher.start_searching(new_common)
684
common_ancestors.update(new_common)
686
# Figure out what the searchers will be searching next, and if
687
# there is only 1 set being searched, then we are done searching,
688
# since all searchers would have to be searching the same data,
689
# thus it *must* be in common.
690
unique_search_sets = set()
691
for searcher in searchers:
692
will_search_set = frozenset(searcher._next_query)
693
if will_search_set not in unique_search_sets:
694
# This searcher is searching a unique set of nodes, let it
695
unique_search_sets.add(will_search_set)
697
if len(unique_search_sets) == 1:
698
nodes = unique_search_sets.pop()
699
uncommon_nodes = nodes.difference(common_ancestors)
701
raise AssertionError("Somehow we ended up converging"
702
" without actually marking them as"
705
"\nuncommon_nodes: %s"
706
% (revisions, uncommon_nodes))
708
return border_ancestors, common_ancestors, searchers
710
def heads(self, keys):
711
"""Return the heads from amongst keys.
713
This is done by searching the ancestries of each key. Any key that is
714
reachable from another key is not returned; all the others are.
716
This operation scales with the relative depth between any two keys. If
717
any two keys are completely disconnected all ancestry of both sides
720
:param keys: An iterable of keys.
721
:return: A set of the heads. Note that as a set there is no ordering
722
information. Callers will need to filter their input to create
723
order if they need it.
725
candidate_heads = set(keys)
726
if revision.NULL_REVISION in candidate_heads:
727
# NULL_REVISION is only a head if it is the only entry
728
candidate_heads.remove(revision.NULL_REVISION)
729
if not candidate_heads:
730
return set([revision.NULL_REVISION])
731
if len(candidate_heads) < 2:
732
return candidate_heads
733
searchers = dict((c, self._make_breadth_first_searcher([c]))
734
for c in candidate_heads)
735
active_searchers = dict(searchers)
736
# skip over the actual candidate for each searcher
737
for searcher in active_searchers.itervalues():
739
# The common walker finds nodes that are common to two or more of the
740
# input keys, so that we don't access all history when a currently
741
# uncommon search point actually meets up with something behind a
742
# common search point. Common search points do not keep searches
743
# active; they just allow us to make searches inactive without
744
# accessing all history.
745
common_walker = self._make_breadth_first_searcher([])
746
while len(active_searchers) > 0:
751
except StopIteration:
752
# No common points being searched at this time.
754
for candidate in active_searchers.keys():
756
searcher = active_searchers[candidate]
758
# rare case: we deleted candidate in a previous iteration
759
# through this for loop, because it was determined to be
760
# a descendant of another candidate.
763
ancestors.update(searcher.next())
764
except StopIteration:
765
del active_searchers[candidate]
767
# process found nodes
769
for ancestor in ancestors:
770
if ancestor in candidate_heads:
771
candidate_heads.remove(ancestor)
772
del searchers[ancestor]
773
if ancestor in active_searchers:
774
del active_searchers[ancestor]
775
# it may meet up with a known common node
776
if ancestor in common_walker.seen:
777
# some searcher has encountered our known common nodes:
779
ancestor_set = set([ancestor])
780
for searcher in searchers.itervalues():
781
searcher.stop_searching_any(ancestor_set)
783
# or it may have been just reached by all the searchers:
784
for searcher in searchers.itervalues():
785
if ancestor not in searcher.seen:
788
# The final active searcher has just reached this node,
789
# making it be known as a descendant of all candidates,
790
# so we can stop searching it, and any seen ancestors
791
new_common.add(ancestor)
792
for searcher in searchers.itervalues():
794
searcher.find_seen_ancestors([ancestor])
795
searcher.stop_searching_any(seen_ancestors)
796
common_walker.start_searching(new_common)
797
return candidate_heads
799
def find_merge_order(self, tip_revision_id, lca_revision_ids):
800
"""Find the order that each revision was merged into tip.
802
This basically just walks backwards with a stack, and walks left-first
803
until it finds a node to stop.
805
if len(lca_revision_ids) == 1:
806
return list(lca_revision_ids)
807
looking_for = set(lca_revision_ids)
808
# TODO: Is there a way we could do this "faster" by batching up the
809
# get_parent_map requests?
810
# TODO: Should we also be culling the ancestry search right away? We
811
# could add looking_for to the "stop" list, and walk their
812
# ancestry in batched mode. The flip side is it might mean we walk a
813
# lot of "stop" nodes, rather than only the minimum.
814
# Then again, without it we may trace back into ancestry we could have
816
stack = [tip_revision_id]
819
while stack and looking_for:
822
if next in looking_for:
824
looking_for.remove(next)
825
if len(looking_for) == 1:
826
found.append(looking_for.pop())
829
parent_ids = self.get_parent_map([next]).get(next, None)
830
if not parent_ids: # Ghost, nothing to search here
832
for parent_id in reversed(parent_ids):
833
# TODO: (performance) We see the parent at this point, but we
834
# wait to mark it until later to make sure we get left
835
# parents before right parents. However, instead of
836
# waiting until we have traversed enough parents, we
837
# could instead note that we've found it, and once all
838
# parents are in the stack, just reverse iterate the
840
if parent_id not in stop:
841
# this will need to be searched
842
stack.append(parent_id)
846
def find_unique_lca(self, left_revision, right_revision,
848
"""Find a unique LCA.
850
Find lowest common ancestors. If there is no unique common
851
ancestor, find the lowest common ancestors of those ancestors.
853
Iteration stops when a unique lowest common ancestor is found.
854
The graph origin is necessarily a unique lowest common ancestor.
856
Note that None is not an acceptable substitute for NULL_REVISION.
857
in the input for this method.
859
:param count_steps: If True, the return value will be a tuple of
860
(unique_lca, steps) where steps is the number of times that
861
find_lca was run. If False, only unique_lca is returned.
863
revisions = [left_revision, right_revision]
867
lca = self.find_lca(*revisions)
875
raise errors.NoCommonAncestor(left_revision, right_revision)
878
def iter_ancestry(self, revision_ids):
879
"""Iterate the ancestry of this revision.
881
:param revision_ids: Nodes to start the search
882
:return: Yield tuples mapping a revision_id to its parents for the
883
ancestry of revision_id.
884
Ghosts will be returned with None as their parents, and nodes
885
with no parents will have NULL_REVISION as their only parent. (As
886
defined by get_parent_map.)
887
There will also be a node for (NULL_REVISION, ())
889
pending = set(revision_ids)
892
processed.update(pending)
893
next_map = self.get_parent_map(pending)
895
for item in next_map.iteritems():
897
next_pending.update(p for p in item[1] if p not in processed)
898
ghosts = pending.difference(next_map)
901
pending = next_pending
903
def iter_topo_order(self, revisions):
904
"""Iterate through the input revisions in topological order.
906
This sorting only ensures that parents come before their children.
907
An ancestor may sort after a descendant if the relationship is not
908
visible in the supplied list of revisions.
910
sorter = tsort.TopoSorter(self.get_parent_map(revisions))
911
return sorter.iter_topo_order()
913
def is_ancestor(self, candidate_ancestor, candidate_descendant):
914
"""Determine whether a revision is an ancestor of another.
916
We answer this using heads() as heads() has the logic to perform the
917
smallest number of parent lookups to determine the ancestral
918
relationship between N revisions.
920
return set([candidate_descendant]) == self.heads(
921
[candidate_ancestor, candidate_descendant])
923
def is_between(self, revid, lower_bound_revid, upper_bound_revid):
924
"""Determine whether a revision is between two others.
926
returns true if and only if:
927
lower_bound_revid <= revid <= upper_bound_revid
929
return ((upper_bound_revid is None or
930
self.is_ancestor(revid, upper_bound_revid)) and
931
(lower_bound_revid is None or
932
self.is_ancestor(lower_bound_revid, revid)))
934
def _search_for_extra_common(self, common, searchers):
935
"""Make sure that unique nodes are genuinely unique.
937
After _find_border_ancestors, all nodes marked "common" are indeed
938
common. Some of the nodes considered unique are not, due to history
939
shortcuts stopping the searches early.
941
We know that we have searched enough when all common search tips are
942
descended from all unique (uncommon) nodes because we know that a node
943
cannot be an ancestor of its own ancestor.
945
:param common: A set of common nodes
946
:param searchers: The searchers returned from _find_border_ancestors
950
# A) The passed in searchers should all be on the same tips, thus
951
# they should be considered the "common" searchers.
952
# B) We find the difference between the searchers, these are the
953
# "unique" nodes for each side.
954
# C) We do a quick culling so that we only start searching from the
955
# more interesting unique nodes. (A unique ancestor is more
956
# interesting than any of its children.)
957
# D) We start searching for ancestors common to all unique nodes.
958
# E) We have the common searchers stop searching any ancestors of
960
# F) When there are no more common search tips, we stop
962
# TODO: We need a way to remove unique_searchers when they overlap with
963
# other unique searchers.
964
if len(searchers) != 2:
965
raise NotImplementedError(
966
"Algorithm not yet implemented for > 2 searchers")
967
common_searchers = searchers
968
left_searcher = searchers[0]
969
right_searcher = searchers[1]
970
unique = left_searcher.seen.symmetric_difference(right_searcher.seen)
971
if not unique: # No unique nodes, nothing to do
973
total_unique = len(unique)
974
unique = self._remove_simple_descendants(unique,
975
self.get_parent_map(unique))
976
simple_unique = len(unique)
978
unique_searchers = []
979
for revision_id in unique:
980
if revision_id in left_searcher.seen:
981
parent_searcher = left_searcher
983
parent_searcher = right_searcher
984
revs_to_search = parent_searcher.find_seen_ancestors([revision_id])
985
if not revs_to_search: # XXX: This shouldn't be possible
986
revs_to_search = [revision_id]
987
searcher = self._make_breadth_first_searcher(revs_to_search)
988
# We don't care about the starting nodes.
990
unique_searchers.append(searcher)
992
# possible todo: aggregate the common searchers into a single common
993
# searcher, just make sure that we include the nodes into the .seen
994
# properties of the original searchers
996
ancestor_all_unique = None
997
for searcher in unique_searchers:
998
if ancestor_all_unique is None:
999
ancestor_all_unique = set(searcher.seen)
1001
ancestor_all_unique = ancestor_all_unique.intersection(
1004
trace.mutter('Started %s unique searchers for %s unique revisions',
1005
simple_unique, total_unique)
1007
while True: # If we have no more nodes we have nothing to do
1008
newly_seen_common = set()
1009
for searcher in common_searchers:
1010
newly_seen_common.update(searcher.step())
1011
newly_seen_unique = set()
1012
for searcher in unique_searchers:
1013
newly_seen_unique.update(searcher.step())
1014
new_common_unique = set()
1015
for revision in newly_seen_unique:
1016
for searcher in unique_searchers:
1017
if revision not in searcher.seen:
1020
# This is a border because it is a first common that we see
1021
# after walking for a while.
1022
new_common_unique.add(revision)
1023
if newly_seen_common:
1024
# These are nodes descended from one of the 'common' searchers.
1025
# Make sure all searchers are on the same page
1026
for searcher in common_searchers:
1027
newly_seen_common.update(
1028
searcher.find_seen_ancestors(newly_seen_common))
1029
# We start searching the whole ancestry. It is a bit wasteful,
1030
# though. We really just want to mark all of these nodes as
1031
# 'seen' and then start just the tips. However, it requires a
1032
# get_parent_map() call to figure out the tips anyway, and all
1033
# redundant requests should be fairly fast.
1034
for searcher in common_searchers:
1035
searcher.start_searching(newly_seen_common)
1037
# If a 'common' node is an ancestor of all unique searchers, we
1038
# can stop searching it.
1039
stop_searching_common = ancestor_all_unique.intersection(
1041
if stop_searching_common:
1042
for searcher in common_searchers:
1043
searcher.stop_searching_any(stop_searching_common)
1044
if new_common_unique:
1045
# We found some ancestors that are common
1046
for searcher in unique_searchers:
1047
new_common_unique.update(
1048
searcher.find_seen_ancestors(new_common_unique))
1049
# Since these are common, we can grab another set of ancestors
1051
for searcher in common_searchers:
1052
new_common_unique.update(
1053
searcher.find_seen_ancestors(new_common_unique))
1055
# We can tell all of the unique searchers to start at these
1056
# nodes, and tell all of the common searchers to *stop*
1057
# searching these nodes
1058
for searcher in unique_searchers:
1059
searcher.start_searching(new_common_unique)
1060
for searcher in common_searchers:
1061
searcher.stop_searching_any(new_common_unique)
1062
ancestor_all_unique.update(new_common_unique)
1064
# Filter out searchers that don't actually search different
1065
# nodes. We already have the ancestry intersection for them
1066
next_unique_searchers = []
1067
unique_search_sets = set()
1068
for searcher in unique_searchers:
1069
will_search_set = frozenset(searcher._next_query)
1070
if will_search_set not in unique_search_sets:
1071
# This searcher is searching a unique set of nodes, let it
1072
unique_search_sets.add(will_search_set)
1073
next_unique_searchers.append(searcher)
1074
unique_searchers = next_unique_searchers
1075
for searcher in common_searchers:
1076
if searcher._next_query:
1079
# All common searcher have stopped searching
1082
def _remove_simple_descendants(self, revisions, parent_map):
1083
"""remove revisions which are children of other ones in the set
1085
This doesn't do any graph searching, it just checks the immediate
1086
parent_map to find if there are any children which can be removed.
1088
:param revisions: A set of revision_ids
1089
:return: A set of revision_ids with the children removed
1091
simple_ancestors = revisions.copy()
1092
# TODO: jam 20071214 we *could* restrict it to searching only the
1093
# parent_map of revisions already present in 'revisions', but
1094
# considering the general use case, I think this is actually
1097
# This is the same as the following loop. I don't know that it is any
1099
## simple_ancestors.difference_update(r for r, p_ids in parent_map.iteritems()
1100
## if p_ids is not None and revisions.intersection(p_ids))
1101
## return simple_ancestors
1103
# Yet Another Way, invert the parent map (which can be cached)
1105
## for revision_id, parent_ids in parent_map.iteritems():
1106
## for p_id in parent_ids:
1107
## descendants.setdefault(p_id, []).append(revision_id)
1108
## for revision in revisions.intersection(descendants):
1109
## simple_ancestors.difference_update(descendants[revision])
1110
## return simple_ancestors
1111
for revision, parent_ids in parent_map.iteritems():
1112
if parent_ids is None:
1114
for parent_id in parent_ids:
1115
if parent_id in revisions:
1116
# This node has a parent present in the set, so we can
1118
simple_ancestors.discard(revision)
1120
return simple_ancestors
1123
class HeadsCache(object):
1124
"""A cache of results for graph heads calls."""
1126
def __init__(self, graph):
1130
def heads(self, keys):
1131
"""Return the heads of keys.
1133
This matches the API of Graph.heads(), specifically the return value is
1134
a set which can be mutated, and ordering of the input is not preserved
1137
:see also: Graph.heads.
1138
:param keys: The keys to calculate heads for.
1139
:return: A set containing the heads, which may be mutated without
1140
affecting future lookups.
1142
keys = frozenset(keys)
1144
return set(self._heads[keys])
1146
heads = self.graph.heads(keys)
1147
self._heads[keys] = heads
1151
class FrozenHeadsCache(object):
1152
"""Cache heads() calls, assuming the caller won't modify them."""
1154
def __init__(self, graph):
1158
def heads(self, keys):
1159
"""Return the heads of keys.
1161
Similar to Graph.heads(). The main difference is that the return value
1162
is a frozen set which cannot be mutated.
1164
:see also: Graph.heads.
1165
:param keys: The keys to calculate heads for.
1166
:return: A frozenset containing the heads.
1168
keys = frozenset(keys)
1170
return self._heads[keys]
1172
heads = frozenset(self.graph.heads(keys))
1173
self._heads[keys] = heads
1176
def cache(self, keys, heads):
1177
"""Store a known value."""
1178
self._heads[frozenset(keys)] = frozenset(heads)
1181
class _BreadthFirstSearcher(object):
1182
"""Parallel search breadth-first the ancestry of revisions.
1184
This class implements the iterator protocol, but additionally
1185
1. provides a set of seen ancestors, and
1186
2. allows some ancestries to be unsearched, via stop_searching_any
1189
def __init__(self, revisions, parents_provider):
1190
self._iterations = 0
1191
self._next_query = set(revisions)
1193
self._started_keys = set(self._next_query)
1194
self._stopped_keys = set()
1195
self._parents_provider = parents_provider
1196
self._returning = 'next_with_ghosts'
1197
self._current_present = set()
1198
self._current_ghosts = set()
1199
self._current_parents = {}
1202
if self._iterations:
1203
prefix = "searching"
1206
search = '%s=%r' % (prefix, list(self._next_query))
1207
return ('_BreadthFirstSearcher(iterations=%d, %s,'
1208
' seen=%r)' % (self._iterations, search, list(self.seen)))
1210
def get_result(self):
1211
"""Get a SearchResult for the current state of this searcher.
1213
:return: A SearchResult for this search so far. The SearchResult is
1214
static - the search can be advanced and the search result will not
1215
be invalidated or altered.
1217
if self._returning == 'next':
1218
# We have to know the current nodes children to be able to list the
1219
# exclude keys for them. However, while we could have a second
1220
# look-ahead result buffer and shuffle things around, this method
1221
# is typically only called once per search - when memoising the
1222
# results of the search.
1223
found, ghosts, next, parents = self._do_query(self._next_query)
1224
# pretend we didn't query: perhaps we should tweak _do_query to be
1225
# entirely stateless?
1226
self.seen.difference_update(next)
1227
next_query = next.union(ghosts)
1229
next_query = self._next_query
1230
excludes = self._stopped_keys.union(next_query)
1231
included_keys = self.seen.difference(excludes)
1232
return SearchResult(self._started_keys, excludes, len(included_keys),
1238
except StopIteration:
1242
"""Return the next ancestors of this revision.
1244
Ancestors are returned in the order they are seen in a breadth-first
1245
traversal. No ancestor will be returned more than once. Ancestors are
1246
returned before their parentage is queried, so ghosts and missing
1247
revisions (including the start revisions) are included in the result.
1248
This can save a round trip in LCA style calculation by allowing
1249
convergence to be detected without reading the data for the revision
1250
the convergence occurs on.
1252
:return: A set of revision_ids.
1254
if self._returning != 'next':
1255
# switch to returning the query, not the results.
1256
self._returning = 'next'
1257
self._iterations += 1
1260
if len(self._next_query) == 0:
1261
raise StopIteration()
1262
# We have seen what we're querying at this point as we are returning
1263
# the query, not the results.
1264
self.seen.update(self._next_query)
1265
return self._next_query
1267
def next_with_ghosts(self):
1268
"""Return the next found ancestors, with ghosts split out.
1270
Ancestors are returned in the order they are seen in a breadth-first
1271
traversal. No ancestor will be returned more than once. Ancestors are
1272
returned only after asking for their parents, which allows us to detect
1273
which revisions are ghosts and which are not.
1275
:return: A tuple with (present ancestors, ghost ancestors) sets.
1277
if self._returning != 'next_with_ghosts':
1278
# switch to returning the results, not the current query.
1279
self._returning = 'next_with_ghosts'
1281
if len(self._next_query) == 0:
1282
raise StopIteration()
1284
return self._current_present, self._current_ghosts
1287
"""Advance the search.
1289
Updates self.seen, self._next_query, self._current_present,
1290
self._current_ghosts, self._current_parents and self._iterations.
1292
self._iterations += 1
1293
found, ghosts, next, parents = self._do_query(self._next_query)
1294
self._current_present = found
1295
self._current_ghosts = ghosts
1296
self._next_query = next
1297
self._current_parents = parents
1298
# ghosts are implicit stop points, otherwise the search cannot be
1299
# repeated when ghosts are filled.
1300
self._stopped_keys.update(ghosts)
1302
def _do_query(self, revisions):
1303
"""Query for revisions.
1305
Adds revisions to the seen set.
1307
:param revisions: Revisions to query.
1308
:return: A tuple: (set(found_revisions), set(ghost_revisions),
1309
set(parents_of_found_revisions), dict(found_revisions:parents)).
1311
found_revisions = set()
1312
parents_of_found = set()
1313
# revisions may contain nodes that point to other nodes in revisions:
1314
# we want to filter them out.
1315
self.seen.update(revisions)
1316
parent_map = self._parents_provider.get_parent_map(revisions)
1317
found_revisions.update(parent_map)
1318
for rev_id, parents in parent_map.iteritems():
1321
new_found_parents = [p for p in parents if p not in self.seen]
1322
if new_found_parents:
1323
# Calling set.update() with an empty generator is actually
1325
parents_of_found.update(new_found_parents)
1326
ghost_revisions = revisions - found_revisions
1327
return found_revisions, ghost_revisions, parents_of_found, parent_map
1332
def find_seen_ancestors(self, revisions):
1333
"""Find ancestors of these revisions that have already been seen.
1335
This function generally makes the assumption that querying for the
1336
parents of a node that has already been queried is reasonably cheap.
1337
(eg, not a round trip to a remote host).
1339
# TODO: Often we might ask one searcher for its seen ancestors, and
1340
# then ask another searcher the same question. This can result in
1341
# searching the same revisions repeatedly if the two searchers
1342
# have a lot of overlap.
1343
all_seen = self.seen
1344
pending = set(revisions).intersection(all_seen)
1345
seen_ancestors = set(pending)
1347
if self._returning == 'next':
1348
# self.seen contains what nodes have been returned, not what nodes
1349
# have been queried. We don't want to probe for nodes that haven't
1350
# been searched yet.
1351
not_searched_yet = self._next_query
1353
not_searched_yet = ()
1354
pending.difference_update(not_searched_yet)
1355
get_parent_map = self._parents_provider.get_parent_map
1357
parent_map = get_parent_map(pending)
1359
# We don't care if it is a ghost, since it can't be seen if it is
1361
for parent_ids in parent_map.itervalues():
1362
all_parents.extend(parent_ids)
1363
next_pending = all_seen.intersection(all_parents).difference(seen_ancestors)
1364
seen_ancestors.update(next_pending)
1365
next_pending.difference_update(not_searched_yet)
1366
pending = next_pending
1368
return seen_ancestors
1370
def stop_searching_any(self, revisions):
1372
Remove any of the specified revisions from the search list.
1374
None of the specified revisions are required to be present in the
1377
It is okay to call stop_searching_any() for revisions which were seen
1378
in previous iterations. It is the callers responsibility to call
1379
find_seen_ancestors() to make sure that current search tips that are
1380
ancestors of those revisions are also stopped. All explicitly stopped
1381
revisions will be excluded from the search result's get_keys(), though.
1383
# TODO: does this help performance?
1386
revisions = frozenset(revisions)
1387
if self._returning == 'next':
1388
stopped = self._next_query.intersection(revisions)
1389
self._next_query = self._next_query.difference(revisions)
1391
stopped_present = self._current_present.intersection(revisions)
1392
stopped = stopped_present.union(
1393
self._current_ghosts.intersection(revisions))
1394
self._current_present.difference_update(stopped)
1395
self._current_ghosts.difference_update(stopped)
1396
# stopping 'x' should stop returning parents of 'x', but
1397
# not if 'y' always references those same parents
1398
stop_rev_references = {}
1399
for rev in stopped_present:
1400
for parent_id in self._current_parents[rev]:
1401
if parent_id not in stop_rev_references:
1402
stop_rev_references[parent_id] = 0
1403
stop_rev_references[parent_id] += 1
1404
# if only the stopped revisions reference it, the ref count will be
1406
for parents in self._current_parents.itervalues():
1407
for parent_id in parents:
1409
stop_rev_references[parent_id] -= 1
1412
stop_parents = set()
1413
for rev_id, refs in stop_rev_references.iteritems():
1415
stop_parents.add(rev_id)
1416
self._next_query.difference_update(stop_parents)
1417
self._stopped_keys.update(stopped)
1418
self._stopped_keys.update(revisions - set([revision.NULL_REVISION]))
1421
def start_searching(self, revisions):
1422
"""Add revisions to the search.
1424
The parents of revisions will be returned from the next call to next()
1425
or next_with_ghosts(). If next_with_ghosts was the most recently used
1426
next* call then the return value is the result of looking up the
1427
ghost/not ghost status of revisions. (A tuple (present, ghosted)).
1429
revisions = frozenset(revisions)
1430
self._started_keys.update(revisions)
1431
new_revisions = revisions.difference(self.seen)
1432
if self._returning == 'next':
1433
self._next_query.update(new_revisions)
1434
self.seen.update(new_revisions)
1436
# perform a query on revisions
1437
revs, ghosts, query, parents = self._do_query(revisions)
1438
self._stopped_keys.update(ghosts)
1439
self._current_present.update(revs)
1440
self._current_ghosts.update(ghosts)
1441
self._next_query.update(query)
1442
self._current_parents.update(parents)
1446
class SearchResult(object):
1447
"""The result of a breadth first search.
1449
A SearchResult provides the ability to reconstruct the search or access a
1450
set of the keys the search found.
1453
def __init__(self, start_keys, exclude_keys, key_count, keys):
1454
"""Create a SearchResult.
1456
:param start_keys: The keys the search started at.
1457
:param exclude_keys: The keys the search excludes.
1458
:param key_count: The total number of keys (from start to but not
1460
:param keys: The keys the search found. Note that in future we may get
1461
a SearchResult from a smart server, in which case the keys list is
1462
not necessarily immediately available.
1464
self._recipe = (start_keys, exclude_keys, key_count)
1465
self._keys = frozenset(keys)
1467
def get_recipe(self):
1468
"""Return a recipe that can be used to replay this search.
1470
The recipe allows reconstruction of the same results at a later date
1471
without knowing all the found keys. The essential elements are a list
1472
of keys to start and and to stop at. In order to give reproducible
1473
results when ghosts are encountered by a search they are automatically
1474
added to the exclude list (or else ghost filling may alter the
1477
:return: A tuple (start_keys_set, exclude_keys_set, revision_count). To
1478
recreate the results of this search, create a breadth first
1479
searcher on the same graph starting at start_keys. Then call next()
1480
(or next_with_ghosts()) repeatedly, and on every result, call
1481
stop_searching_any on any keys from the exclude_keys set. The
1482
revision_count value acts as a trivial cross-check - the found
1483
revisions of the new search should have as many elements as
1484
revision_count. If it does not, then additional revisions have been
1485
ghosted since the search was executed the first time and the second
1491
"""Return the keys found in this search.
1493
:return: A set of keys.
1498
def collapse_linear_regions(parent_map):
1499
"""Collapse regions of the graph that are 'linear'.
1505
can be collapsed by removing B and getting::
1509
:param parent_map: A dictionary mapping children to their parents
1510
:return: Another dictionary with 'linear' chains collapsed
1512
# Note: this isn't a strictly minimal collapse. For example:
1520
# Will not have 'D' removed, even though 'E' could fit. Also:
1526
# A and C are both kept because they are edges of the graph. We *could* get
1527
# rid of A if we wanted.
1535
# Will not have any nodes removed, even though you do have an
1536
# 'uninteresting' linear D->B and E->C
1538
for child, parents in parent_map.iteritems():
1539
children.setdefault(child, [])
1541
children.setdefault(p, []).append(child)
1543
orig_children = dict(children)
1545
result = dict(parent_map)
1546
for node in parent_map:
1547
parents = result[node]
1548
if len(parents) == 1:
1549
parent_children = children[parents[0]]
1550
if len(parent_children) != 1:
1551
# This is not the only child
1553
node_children = children[node]
1554
if len(node_children) != 1:
1556
child_parents = result.get(node_children[0], None)
1557
if len(child_parents) != 1:
1558
# This is not its only parent
1560
# The child of this node only points at it, and the parent only has
1561
# this as a child. remove this node, and join the others together
1562
result[node_children[0]] = parents
1563
children[parents[0]] = node_children
added
removed
bzrlib/graph.py
