1
# Copyright (C) 2007, 2008, 2009 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25
from bzrlib.symbol_versioning import deprecated_function, deprecated_in
27
STEP_UNIQUE_SEARCHER_EVERY = 5
29
# DIAGRAM of terminology
39
# In this diagram, relative to G and H:
40
# A, B, C, D, E are common ancestors.
41
# C, D and E are border ancestors, because each has a non-common descendant.
42
# D and E are least common ancestors because none of their descendants are
44
# C is not a least common ancestor because its descendant, E, is a common
47
# The find_unique_lca algorithm will pick A in two steps:
48
# 1. find_lca('G', 'H') => ['D', 'E']
49
# 2. Since len(['D', 'E']) > 1, find_lca('D', 'E') => ['A']
52
class DictParentsProvider(object):
53
"""A parents provider for Graph objects."""
55
def __init__(self, ancestry):
56
self.ancestry = ancestry
59
return 'DictParentsProvider(%r)' % self.ancestry
61
def get_parent_map(self, keys):
62
"""See StackedParentsProvider.get_parent_map"""
63
ancestry = self.ancestry
64
return dict((k, ancestry[k]) for k in keys if k in ancestry)
66
@deprecated_function(deprecated_in((1, 16, 0)))
67
def _StackedParentsProvider(*args, **kwargs):
68
return StackedParentsProvider(*args, **kwargs)
70
class StackedParentsProvider(object):
71
"""A parents provider which stacks (or unions) multiple providers.
73
The providers are queries in the order of the provided parent_providers.
76
def __init__(self, parent_providers):
77
self._parent_providers = parent_providers
80
return "%s(%r)" % (self.__class__.__name__, self._parent_providers)
82
def get_parent_map(self, keys):
83
"""Get a mapping of keys => parents
85
A dictionary is returned with an entry for each key present in this
86
source. If this source doesn't have information about a key, it should
89
[NULL_REVISION] is used as the parent of the first user-committed
90
revision. Its parent list is empty.
92
:param keys: An iterable returning keys to check (eg revision_ids)
93
:return: A dictionary mapping each key to its parents
97
for parents_provider in self._parent_providers:
98
new_found = parents_provider.get_parent_map(remaining)
99
found.update(new_found)
100
remaining.difference_update(new_found)
106
class CachingParentsProvider(object):
107
"""A parents provider which will cache the revision => parents as a dict.
109
This is useful for providers which have an expensive look up.
111
Either a ParentsProvider or a get_parent_map-like callback may be
112
supplied. If it provides extra un-asked-for parents, they will be cached,
113
but filtered out of get_parent_map.
115
The cache is enabled by default, but may be disabled and re-enabled.
117
def __init__(self, parent_provider=None, get_parent_map=None):
120
:param parent_provider: The ParentProvider to use. It or
121
get_parent_map must be supplied.
122
:param get_parent_map: The get_parent_map callback to use. It or
123
parent_provider must be supplied.
125
self._real_provider = parent_provider
126
if get_parent_map is None:
127
self._get_parent_map = self._real_provider.get_parent_map
129
self._get_parent_map = get_parent_map
131
self.enable_cache(True)
134
return "%s(%r)" % (self.__class__.__name__, self._real_provider)
136
def enable_cache(self, cache_misses=True):
138
if self._cache is not None:
139
raise AssertionError('Cache enabled when already enabled.')
141
self._cache_misses = cache_misses
142
self.missing_keys = set()
144
def disable_cache(self):
145
"""Disable and clear the cache."""
147
self._cache_misses = None
148
self.missing_keys = set()
150
def get_cached_map(self):
151
"""Return any cached get_parent_map values."""
152
if self._cache is None:
154
return dict(self._cache)
156
def get_parent_map(self, keys):
157
"""See StackedParentsProvider.get_parent_map."""
160
cache = self._get_parent_map(keys)
162
needed_revisions = set(key for key in keys if key not in cache)
163
# Do not ask for negatively cached keys
164
needed_revisions.difference_update(self.missing_keys)
166
parent_map = self._get_parent_map(needed_revisions)
167
cache.update(parent_map)
168
if self._cache_misses:
169
for key in needed_revisions:
170
if key not in parent_map:
171
self.note_missing_key(key)
174
value = cache.get(key)
175
if value is not None:
179
def note_missing_key(self, key):
180
"""Note that key is a missing key."""
181
if self._cache_misses:
182
self.missing_keys.add(key)
186
"""Provide incremental access to revision graphs.
188
This is the generic implementation; it is intended to be subclassed to
189
specialize it for other repository types.
192
def __init__(self, parents_provider):
193
"""Construct a Graph that uses several graphs as its input
195
This should not normally be invoked directly, because there may be
196
specialized implementations for particular repository types. See
197
Repository.get_graph().
199
:param parents_provider: An object providing a get_parent_map call
200
conforming to the behavior of
201
StackedParentsProvider.get_parent_map.
203
if getattr(parents_provider, 'get_parents', None) is not None:
204
self.get_parents = parents_provider.get_parents
205
if getattr(parents_provider, 'get_parent_map', None) is not None:
206
self.get_parent_map = parents_provider.get_parent_map
207
self._parents_provider = parents_provider
210
return 'Graph(%r)' % self._parents_provider
212
def find_lca(self, *revisions):
213
"""Determine the lowest common ancestors of the provided revisions
215
A lowest common ancestor is a common ancestor none of whose
216
descendants are common ancestors. In graphs, unlike trees, there may
217
be multiple lowest common ancestors.
219
This algorithm has two phases. Phase 1 identifies border ancestors,
220
and phase 2 filters border ancestors to determine lowest common
223
In phase 1, border ancestors are identified, using a breadth-first
224
search starting at the bottom of the graph. Searches are stopped
225
whenever a node or one of its descendants is determined to be common
227
In phase 2, the border ancestors are filtered to find the least
228
common ancestors. This is done by searching the ancestries of each
231
Phase 2 is perfomed on the principle that a border ancestor that is
232
not an ancestor of any other border ancestor is a least common
235
Searches are stopped when they find a node that is determined to be a
236
common ancestor of all border ancestors, because this shows that it
237
cannot be a descendant of any border ancestor.
239
The scaling of this operation should be proportional to
240
1. The number of uncommon ancestors
241
2. The number of border ancestors
242
3. The length of the shortest path between a border ancestor and an
243
ancestor of all border ancestors.
245
border_common, common, sides = self._find_border_ancestors(revisions)
246
# We may have common ancestors that can be reached from each other.
247
# - ask for the heads of them to filter it down to only ones that
248
# cannot be reached from each other - phase 2.
249
return self.heads(border_common)
251
def find_difference(self, left_revision, right_revision):
252
"""Determine the graph difference between two revisions"""
253
border, common, searchers = self._find_border_ancestors(
254
[left_revision, right_revision])
255
self._search_for_extra_common(common, searchers)
256
left = searchers[0].seen
257
right = searchers[1].seen
258
return (left.difference(right), right.difference(left))
260
def find_distance_to_null(self, target_revision_id, known_revision_ids):
261
"""Find the left-hand distance to the NULL_REVISION.
263
(This can also be considered the revno of a branch at
266
:param target_revision_id: A revision_id which we would like to know
268
:param known_revision_ids: [(revision_id, revno)] A list of known
269
revno, revision_id tuples. We'll use this to seed the search.
271
# Map from revision_ids to a known value for their revno
272
known_revnos = dict(known_revision_ids)
273
cur_tip = target_revision_id
275
NULL_REVISION = revision.NULL_REVISION
276
known_revnos[NULL_REVISION] = 0
278
searching_known_tips = list(known_revnos.keys())
280
unknown_searched = {}
282
while cur_tip not in known_revnos:
283
unknown_searched[cur_tip] = num_steps
285
to_search = set([cur_tip])
286
to_search.update(searching_known_tips)
287
parent_map = self.get_parent_map(to_search)
288
parents = parent_map.get(cur_tip, None)
289
if not parents: # An empty list or None is a ghost
290
raise errors.GhostRevisionsHaveNoRevno(target_revision_id,
294
for revision_id in searching_known_tips:
295
parents = parent_map.get(revision_id, None)
299
next_revno = known_revnos[revision_id] - 1
300
if next in unknown_searched:
301
# We have enough information to return a value right now
302
return next_revno + unknown_searched[next]
303
if next in known_revnos:
305
known_revnos[next] = next_revno
306
next_known_tips.append(next)
307
searching_known_tips = next_known_tips
309
# We reached a known revision, so just add in how many steps it took to
311
return known_revnos[cur_tip] + num_steps
313
def find_lefthand_distances(self, keys):
314
"""Find the distance to null for all the keys in keys.
316
:param keys: keys to lookup.
317
:return: A dict key->distance for all of keys.
319
# Optimisable by concurrent searching, but a random spread should get
320
# some sort of hit rate.
327
(key, self.find_distance_to_null(key, known_revnos)))
328
except errors.GhostRevisionsHaveNoRevno:
331
known_revnos.append((key, -1))
332
return dict(known_revnos)
334
def find_unique_ancestors(self, unique_revision, common_revisions):
335
"""Find the unique ancestors for a revision versus others.
337
This returns the ancestry of unique_revision, excluding all revisions
338
in the ancestry of common_revisions. If unique_revision is in the
339
ancestry, then the empty set will be returned.
341
:param unique_revision: The revision_id whose ancestry we are
343
XXX: Would this API be better if we allowed multiple revisions on
345
:param common_revisions: Revision_ids of ancestries to exclude.
346
:return: A set of revisions in the ancestry of unique_revision
348
if unique_revision in common_revisions:
351
# Algorithm description
352
# 1) Walk backwards from the unique node and all common nodes.
353
# 2) When a node is seen by both sides, stop searching it in the unique
354
# walker, include it in the common walker.
355
# 3) Stop searching when there are no nodes left for the unique walker.
356
# At this point, you have a maximal set of unique nodes. Some of
357
# them may actually be common, and you haven't reached them yet.
358
# 4) Start new searchers for the unique nodes, seeded with the
359
# information you have so far.
360
# 5) Continue searching, stopping the common searches when the search
361
# tip is an ancestor of all unique nodes.
362
# 6) Aggregate together unique searchers when they are searching the
363
# same tips. When all unique searchers are searching the same node,
364
# stop move it to a single 'all_unique_searcher'.
365
# 7) The 'all_unique_searcher' represents the very 'tip' of searching.
366
# Most of the time this produces very little important information.
367
# So don't step it as quickly as the other searchers.
368
# 8) Search is done when all common searchers have completed.
370
unique_searcher, common_searcher = self._find_initial_unique_nodes(
371
[unique_revision], common_revisions)
373
unique_nodes = unique_searcher.seen.difference(common_searcher.seen)
377
(all_unique_searcher,
378
unique_tip_searchers) = self._make_unique_searchers(unique_nodes,
379
unique_searcher, common_searcher)
381
self._refine_unique_nodes(unique_searcher, all_unique_searcher,
382
unique_tip_searchers, common_searcher)
383
true_unique_nodes = unique_nodes.difference(common_searcher.seen)
384
if 'graph' in debug.debug_flags:
385
trace.mutter('Found %d truly unique nodes out of %d',
386
len(true_unique_nodes), len(unique_nodes))
387
return true_unique_nodes
389
def _find_initial_unique_nodes(self, unique_revisions, common_revisions):
390
"""Steps 1-3 of find_unique_ancestors.
392
Find the maximal set of unique nodes. Some of these might actually
393
still be common, but we are sure that there are no other unique nodes.
395
:return: (unique_searcher, common_searcher)
398
unique_searcher = self._make_breadth_first_searcher(unique_revisions)
399
# we know that unique_revisions aren't in common_revisions, so skip
401
unique_searcher.next()
402
common_searcher = self._make_breadth_first_searcher(common_revisions)
404
# As long as we are still finding unique nodes, keep searching
405
while unique_searcher._next_query:
406
next_unique_nodes = set(unique_searcher.step())
407
next_common_nodes = set(common_searcher.step())
409
# Check if either searcher encounters new nodes seen by the other
411
unique_are_common_nodes = next_unique_nodes.intersection(
412
common_searcher.seen)
413
unique_are_common_nodes.update(
414
next_common_nodes.intersection(unique_searcher.seen))
415
if unique_are_common_nodes:
416
ancestors = unique_searcher.find_seen_ancestors(
417
unique_are_common_nodes)
418
# TODO: This is a bit overboard, we only really care about
419
# the ancestors of the tips because the rest we
420
# already know. This is *correct* but causes us to
421
# search too much ancestry.
422
ancestors.update(common_searcher.find_seen_ancestors(ancestors))
423
unique_searcher.stop_searching_any(ancestors)
424
common_searcher.start_searching(ancestors)
426
return unique_searcher, common_searcher
428
def _make_unique_searchers(self, unique_nodes, unique_searcher,
430
"""Create a searcher for all the unique search tips (step 4).
432
As a side effect, the common_searcher will stop searching any nodes
433
that are ancestors of the unique searcher tips.
435
:return: (all_unique_searcher, unique_tip_searchers)
437
unique_tips = self._remove_simple_descendants(unique_nodes,
438
self.get_parent_map(unique_nodes))
440
if len(unique_tips) == 1:
441
unique_tip_searchers = []
442
ancestor_all_unique = unique_searcher.find_seen_ancestors(unique_tips)
444
unique_tip_searchers = []
445
for tip in unique_tips:
446
revs_to_search = unique_searcher.find_seen_ancestors([tip])
447
revs_to_search.update(
448
common_searcher.find_seen_ancestors(revs_to_search))
449
searcher = self._make_breadth_first_searcher(revs_to_search)
450
# We don't care about the starting nodes.
451
searcher._label = tip
453
unique_tip_searchers.append(searcher)
455
ancestor_all_unique = None
456
for searcher in unique_tip_searchers:
457
if ancestor_all_unique is None:
458
ancestor_all_unique = set(searcher.seen)
460
ancestor_all_unique = ancestor_all_unique.intersection(
462
# Collapse all the common nodes into a single searcher
463
all_unique_searcher = self._make_breadth_first_searcher(
465
if ancestor_all_unique:
466
# We've seen these nodes in all the searchers, so we'll just go to
468
all_unique_searcher.step()
470
# Stop any search tips that are already known as ancestors of the
472
stopped_common = common_searcher.stop_searching_any(
473
common_searcher.find_seen_ancestors(ancestor_all_unique))
476
for searcher in unique_tip_searchers:
477
total_stopped += len(searcher.stop_searching_any(
478
searcher.find_seen_ancestors(ancestor_all_unique)))
479
if 'graph' in debug.debug_flags:
480
trace.mutter('For %d unique nodes, created %d + 1 unique searchers'
481
' (%d stopped search tips, %d common ancestors'
482
' (%d stopped common)',
483
len(unique_nodes), len(unique_tip_searchers),
484
total_stopped, len(ancestor_all_unique),
486
return all_unique_searcher, unique_tip_searchers
488
def _step_unique_and_common_searchers(self, common_searcher,
489
unique_tip_searchers,
491
"""Step all the searchers"""
492
newly_seen_common = set(common_searcher.step())
493
newly_seen_unique = set()
494
for searcher in unique_tip_searchers:
495
next = set(searcher.step())
496
next.update(unique_searcher.find_seen_ancestors(next))
497
next.update(common_searcher.find_seen_ancestors(next))
498
for alt_searcher in unique_tip_searchers:
499
if alt_searcher is searcher:
501
next.update(alt_searcher.find_seen_ancestors(next))
502
searcher.start_searching(next)
503
newly_seen_unique.update(next)
504
return newly_seen_common, newly_seen_unique
506
def _find_nodes_common_to_all_unique(self, unique_tip_searchers,
508
newly_seen_unique, step_all_unique):
509
"""Find nodes that are common to all unique_tip_searchers.
511
If it is time, step the all_unique_searcher, and add its nodes to the
514
common_to_all_unique_nodes = newly_seen_unique.copy()
515
for searcher in unique_tip_searchers:
516
common_to_all_unique_nodes.intersection_update(searcher.seen)
517
common_to_all_unique_nodes.intersection_update(
518
all_unique_searcher.seen)
519
# Step all-unique less frequently than the other searchers.
520
# In the common case, we don't need to spider out far here, so
521
# avoid doing extra work.
523
tstart = time.clock()
524
nodes = all_unique_searcher.step()
525
common_to_all_unique_nodes.update(nodes)
526
if 'graph' in debug.debug_flags:
527
tdelta = time.clock() - tstart
528
trace.mutter('all_unique_searcher step() took %.3fs'
529
'for %d nodes (%d total), iteration: %s',
530
tdelta, len(nodes), len(all_unique_searcher.seen),
531
all_unique_searcher._iterations)
532
return common_to_all_unique_nodes
534
def _collapse_unique_searchers(self, unique_tip_searchers,
535
common_to_all_unique_nodes):
536
"""Combine searchers that are searching the same tips.
538
When two searchers are searching the same tips, we can stop one of the
539
searchers. We also know that the maximal set of common ancestors is the
540
intersection of the two original searchers.
542
:return: A list of searchers that are searching unique nodes.
544
# Filter out searchers that don't actually search different
545
# nodes. We already have the ancestry intersection for them
546
unique_search_tips = {}
547
for searcher in unique_tip_searchers:
548
stopped = searcher.stop_searching_any(common_to_all_unique_nodes)
549
will_search_set = frozenset(searcher._next_query)
550
if not will_search_set:
551
if 'graph' in debug.debug_flags:
552
trace.mutter('Unique searcher %s was stopped.'
553
' (%s iterations) %d nodes stopped',
555
searcher._iterations,
557
elif will_search_set not in unique_search_tips:
558
# This searcher is searching a unique set of nodes, let it
559
unique_search_tips[will_search_set] = [searcher]
561
unique_search_tips[will_search_set].append(searcher)
562
# TODO: it might be possible to collapse searchers faster when they
563
# only have *some* search tips in common.
564
next_unique_searchers = []
565
for searchers in unique_search_tips.itervalues():
566
if len(searchers) == 1:
567
# Searching unique tips, go for it
568
next_unique_searchers.append(searchers[0])
570
# These searchers have started searching the same tips, we
571
# don't need them to cover the same ground. The
572
# intersection of their ancestry won't change, so create a
573
# new searcher, combining their histories.
574
next_searcher = searchers[0]
575
for searcher in searchers[1:]:
576
next_searcher.seen.intersection_update(searcher.seen)
577
if 'graph' in debug.debug_flags:
578
trace.mutter('Combining %d searchers into a single'
579
' searcher searching %d nodes with'
582
len(next_searcher._next_query),
583
len(next_searcher.seen))
584
next_unique_searchers.append(next_searcher)
585
return next_unique_searchers
587
def _refine_unique_nodes(self, unique_searcher, all_unique_searcher,
588
unique_tip_searchers, common_searcher):
589
"""Steps 5-8 of find_unique_ancestors.
591
This function returns when common_searcher has stopped searching for
594
# We step the ancestor_all_unique searcher only every
595
# STEP_UNIQUE_SEARCHER_EVERY steps.
596
step_all_unique_counter = 0
597
# While we still have common nodes to search
598
while common_searcher._next_query:
600
newly_seen_unique) = self._step_unique_and_common_searchers(
601
common_searcher, unique_tip_searchers, unique_searcher)
602
# These nodes are common ancestors of all unique nodes
603
common_to_all_unique_nodes = self._find_nodes_common_to_all_unique(
604
unique_tip_searchers, all_unique_searcher, newly_seen_unique,
605
step_all_unique_counter==0)
606
step_all_unique_counter = ((step_all_unique_counter + 1)
607
% STEP_UNIQUE_SEARCHER_EVERY)
609
if newly_seen_common:
610
# If a 'common' node is an ancestor of all unique searchers, we
611
# can stop searching it.
612
common_searcher.stop_searching_any(
613
all_unique_searcher.seen.intersection(newly_seen_common))
614
if common_to_all_unique_nodes:
615
common_to_all_unique_nodes.update(
616
common_searcher.find_seen_ancestors(
617
common_to_all_unique_nodes))
618
# The all_unique searcher can start searching the common nodes
619
# but everyone else can stop.
620
# This is the sort of thing where we would like to not have it
621
# start_searching all of the nodes, but only mark all of them
622
# as seen, and have it search only the actual tips. Otherwise
623
# it is another get_parent_map() traversal for it to figure out
624
# what we already should know.
625
all_unique_searcher.start_searching(common_to_all_unique_nodes)
626
common_searcher.stop_searching_any(common_to_all_unique_nodes)
628
next_unique_searchers = self._collapse_unique_searchers(
629
unique_tip_searchers, common_to_all_unique_nodes)
630
if len(unique_tip_searchers) != len(next_unique_searchers):
631
if 'graph' in debug.debug_flags:
632
trace.mutter('Collapsed %d unique searchers => %d'
634
len(unique_tip_searchers),
635
len(next_unique_searchers),
636
all_unique_searcher._iterations)
637
unique_tip_searchers = next_unique_searchers
639
def get_parent_map(self, revisions):
640
"""Get a map of key:parent_list for revisions.
642
This implementation delegates to get_parents, for old parent_providers
643
that do not supply get_parent_map.
646
for rev, parents in self.get_parents(revisions):
647
if parents is not None:
648
result[rev] = parents
651
def _make_breadth_first_searcher(self, revisions):
652
return _BreadthFirstSearcher(revisions, self)
654
def _find_border_ancestors(self, revisions):
655
"""Find common ancestors with at least one uncommon descendant.
657
Border ancestors are identified using a breadth-first
658
search starting at the bottom of the graph. Searches are stopped
659
whenever a node or one of its descendants is determined to be common.
661
This will scale with the number of uncommon ancestors.
663
As well as the border ancestors, a set of seen common ancestors and a
664
list of sets of seen ancestors for each input revision is returned.
665
This allows calculation of graph difference from the results of this
668
if None in revisions:
669
raise errors.InvalidRevisionId(None, self)
670
common_ancestors = set()
671
searchers = [self._make_breadth_first_searcher([r])
673
active_searchers = searchers[:]
674
border_ancestors = set()
678
for searcher in searchers:
679
new_ancestors = searcher.step()
681
newly_seen.update(new_ancestors)
683
for revision in newly_seen:
684
if revision in common_ancestors:
685
# Not a border ancestor because it was seen as common
687
new_common.add(revision)
689
for searcher in searchers:
690
if revision not in searcher.seen:
693
# This is a border because it is a first common that we see
694
# after walking for a while.
695
border_ancestors.add(revision)
696
new_common.add(revision)
698
for searcher in searchers:
699
new_common.update(searcher.find_seen_ancestors(new_common))
700
for searcher in searchers:
701
searcher.start_searching(new_common)
702
common_ancestors.update(new_common)
704
# Figure out what the searchers will be searching next, and if
705
# there is only 1 set being searched, then we are done searching,
706
# since all searchers would have to be searching the same data,
707
# thus it *must* be in common.
708
unique_search_sets = set()
709
for searcher in searchers:
710
will_search_set = frozenset(searcher._next_query)
711
if will_search_set not in unique_search_sets:
712
# This searcher is searching a unique set of nodes, let it
713
unique_search_sets.add(will_search_set)
715
if len(unique_search_sets) == 1:
716
nodes = unique_search_sets.pop()
717
uncommon_nodes = nodes.difference(common_ancestors)
719
raise AssertionError("Somehow we ended up converging"
720
" without actually marking them as"
723
"\nuncommon_nodes: %s"
724
% (revisions, uncommon_nodes))
726
return border_ancestors, common_ancestors, searchers
728
def heads(self, keys):
729
"""Return the heads from amongst keys.
731
This is done by searching the ancestries of each key. Any key that is
732
reachable from another key is not returned; all the others are.
734
This operation scales with the relative depth between any two keys. If
735
any two keys are completely disconnected all ancestry of both sides
738
:param keys: An iterable of keys.
739
:return: A set of the heads. Note that as a set there is no ordering
740
information. Callers will need to filter their input to create
741
order if they need it.
743
candidate_heads = set(keys)
744
if revision.NULL_REVISION in candidate_heads:
745
# NULL_REVISION is only a head if it is the only entry
746
candidate_heads.remove(revision.NULL_REVISION)
747
if not candidate_heads:
748
return set([revision.NULL_REVISION])
749
if len(candidate_heads) < 2:
750
return candidate_heads
751
searchers = dict((c, self._make_breadth_first_searcher([c]))
752
for c in candidate_heads)
753
active_searchers = dict(searchers)
754
# skip over the actual candidate for each searcher
755
for searcher in active_searchers.itervalues():
757
# The common walker finds nodes that are common to two or more of the
758
# input keys, so that we don't access all history when a currently
759
# uncommon search point actually meets up with something behind a
760
# common search point. Common search points do not keep searches
761
# active; they just allow us to make searches inactive without
762
# accessing all history.
763
common_walker = self._make_breadth_first_searcher([])
764
while len(active_searchers) > 0:
769
except StopIteration:
770
# No common points being searched at this time.
772
for candidate in active_searchers.keys():
774
searcher = active_searchers[candidate]
776
# rare case: we deleted candidate in a previous iteration
777
# through this for loop, because it was determined to be
778
# a descendant of another candidate.
781
ancestors.update(searcher.next())
782
except StopIteration:
783
del active_searchers[candidate]
785
# process found nodes
787
for ancestor in ancestors:
788
if ancestor in candidate_heads:
789
candidate_heads.remove(ancestor)
790
del searchers[ancestor]
791
if ancestor in active_searchers:
792
del active_searchers[ancestor]
793
# it may meet up with a known common node
794
if ancestor in common_walker.seen:
795
# some searcher has encountered our known common nodes:
797
ancestor_set = set([ancestor])
798
for searcher in searchers.itervalues():
799
searcher.stop_searching_any(ancestor_set)
801
# or it may have been just reached by all the searchers:
802
for searcher in searchers.itervalues():
803
if ancestor not in searcher.seen:
806
# The final active searcher has just reached this node,
807
# making it be known as a descendant of all candidates,
808
# so we can stop searching it, and any seen ancestors
809
new_common.add(ancestor)
810
for searcher in searchers.itervalues():
812
searcher.find_seen_ancestors([ancestor])
813
searcher.stop_searching_any(seen_ancestors)
814
common_walker.start_searching(new_common)
815
return candidate_heads
817
def find_merge_order(self, tip_revision_id, lca_revision_ids):
818
"""Find the order that each revision was merged into tip.
820
This basically just walks backwards with a stack, and walks left-first
821
until it finds a node to stop.
823
if len(lca_revision_ids) == 1:
824
return list(lca_revision_ids)
825
looking_for = set(lca_revision_ids)
826
# TODO: Is there a way we could do this "faster" by batching up the
827
# get_parent_map requests?
828
# TODO: Should we also be culling the ancestry search right away? We
829
# could add looking_for to the "stop" list, and walk their
830
# ancestry in batched mode. The flip side is it might mean we walk a
831
# lot of "stop" nodes, rather than only the minimum.
832
# Then again, without it we may trace back into ancestry we could have
834
stack = [tip_revision_id]
837
while stack and looking_for:
840
if next in looking_for:
842
looking_for.remove(next)
843
if len(looking_for) == 1:
844
found.append(looking_for.pop())
847
parent_ids = self.get_parent_map([next]).get(next, None)
848
if not parent_ids: # Ghost, nothing to search here
850
for parent_id in reversed(parent_ids):
851
# TODO: (performance) We see the parent at this point, but we
852
# wait to mark it until later to make sure we get left
853
# parents before right parents. However, instead of
854
# waiting until we have traversed enough parents, we
855
# could instead note that we've found it, and once all
856
# parents are in the stack, just reverse iterate the
858
if parent_id not in stop:
859
# this will need to be searched
860
stack.append(parent_id)
864
def find_unique_lca(self, left_revision, right_revision,
866
"""Find a unique LCA.
868
Find lowest common ancestors. If there is no unique common
869
ancestor, find the lowest common ancestors of those ancestors.
871
Iteration stops when a unique lowest common ancestor is found.
872
The graph origin is necessarily a unique lowest common ancestor.
874
Note that None is not an acceptable substitute for NULL_REVISION.
875
in the input for this method.
877
:param count_steps: If True, the return value will be a tuple of
878
(unique_lca, steps) where steps is the number of times that
879
find_lca was run. If False, only unique_lca is returned.
881
revisions = [left_revision, right_revision]
885
lca = self.find_lca(*revisions)
893
raise errors.NoCommonAncestor(left_revision, right_revision)
896
def iter_ancestry(self, revision_ids):
897
"""Iterate the ancestry of this revision.
899
:param revision_ids: Nodes to start the search
900
:return: Yield tuples mapping a revision_id to its parents for the
901
ancestry of revision_id.
902
Ghosts will be returned with None as their parents, and nodes
903
with no parents will have NULL_REVISION as their only parent. (As
904
defined by get_parent_map.)
905
There will also be a node for (NULL_REVISION, ())
907
pending = set(revision_ids)
910
processed.update(pending)
911
next_map = self.get_parent_map(pending)
913
for item in next_map.iteritems():
915
next_pending.update(p for p in item[1] if p not in processed)
916
ghosts = pending.difference(next_map)
919
pending = next_pending
921
def iter_topo_order(self, revisions):
922
"""Iterate through the input revisions in topological order.
924
This sorting only ensures that parents come before their children.
925
An ancestor may sort after a descendant if the relationship is not
926
visible in the supplied list of revisions.
928
from bzrlib import tsort
929
sorter = tsort.TopoSorter(self.get_parent_map(revisions))
930
return sorter.iter_topo_order()
932
def is_ancestor(self, candidate_ancestor, candidate_descendant):
933
"""Determine whether a revision is an ancestor of another.
935
We answer this using heads() as heads() has the logic to perform the
936
smallest number of parent lookups to determine the ancestral
937
relationship between N revisions.
939
return set([candidate_descendant]) == self.heads(
940
[candidate_ancestor, candidate_descendant])
942
def is_between(self, revid, lower_bound_revid, upper_bound_revid):
943
"""Determine whether a revision is between two others.
945
returns true if and only if:
946
lower_bound_revid <= revid <= upper_bound_revid
948
return ((upper_bound_revid is None or
949
self.is_ancestor(revid, upper_bound_revid)) and
950
(lower_bound_revid is None or
951
self.is_ancestor(lower_bound_revid, revid)))
953
def _search_for_extra_common(self, common, searchers):
954
"""Make sure that unique nodes are genuinely unique.
956
After _find_border_ancestors, all nodes marked "common" are indeed
957
common. Some of the nodes considered unique are not, due to history
958
shortcuts stopping the searches early.
960
We know that we have searched enough when all common search tips are
961
descended from all unique (uncommon) nodes because we know that a node
962
cannot be an ancestor of its own ancestor.
964
:param common: A set of common nodes
965
:param searchers: The searchers returned from _find_border_ancestors
969
# A) The passed in searchers should all be on the same tips, thus
970
# they should be considered the "common" searchers.
971
# B) We find the difference between the searchers, these are the
972
# "unique" nodes for each side.
973
# C) We do a quick culling so that we only start searching from the
974
# more interesting unique nodes. (A unique ancestor is more
975
# interesting than any of its children.)
976
# D) We start searching for ancestors common to all unique nodes.
977
# E) We have the common searchers stop searching any ancestors of
979
# F) When there are no more common search tips, we stop
981
# TODO: We need a way to remove unique_searchers when they overlap with
982
# other unique searchers.
983
if len(searchers) != 2:
984
raise NotImplementedError(
985
"Algorithm not yet implemented for > 2 searchers")
986
common_searchers = searchers
987
left_searcher = searchers[0]
988
right_searcher = searchers[1]
989
unique = left_searcher.seen.symmetric_difference(right_searcher.seen)
990
if not unique: # No unique nodes, nothing to do
992
total_unique = len(unique)
993
unique = self._remove_simple_descendants(unique,
994
self.get_parent_map(unique))
995
simple_unique = len(unique)
997
unique_searchers = []
998
for revision_id in unique:
999
if revision_id in left_searcher.seen:
1000
parent_searcher = left_searcher
1002
parent_searcher = right_searcher
1003
revs_to_search = parent_searcher.find_seen_ancestors([revision_id])
1004
if not revs_to_search: # XXX: This shouldn't be possible
1005
revs_to_search = [revision_id]
1006
searcher = self._make_breadth_first_searcher(revs_to_search)
1007
# We don't care about the starting nodes.
1009
unique_searchers.append(searcher)
1011
# possible todo: aggregate the common searchers into a single common
1012
# searcher, just make sure that we include the nodes into the .seen
1013
# properties of the original searchers
1015
ancestor_all_unique = None
1016
for searcher in unique_searchers:
1017
if ancestor_all_unique is None:
1018
ancestor_all_unique = set(searcher.seen)
1020
ancestor_all_unique = ancestor_all_unique.intersection(
1023
trace.mutter('Started %s unique searchers for %s unique revisions',
1024
simple_unique, total_unique)
1026
while True: # If we have no more nodes we have nothing to do
1027
newly_seen_common = set()
1028
for searcher in common_searchers:
1029
newly_seen_common.update(searcher.step())
1030
newly_seen_unique = set()
1031
for searcher in unique_searchers:
1032
newly_seen_unique.update(searcher.step())
1033
new_common_unique = set()
1034
for revision in newly_seen_unique:
1035
for searcher in unique_searchers:
1036
if revision not in searcher.seen:
1039
# This is a border because it is a first common that we see
1040
# after walking for a while.
1041
new_common_unique.add(revision)
1042
if newly_seen_common:
1043
# These are nodes descended from one of the 'common' searchers.
1044
# Make sure all searchers are on the same page
1045
for searcher in common_searchers:
1046
newly_seen_common.update(
1047
searcher.find_seen_ancestors(newly_seen_common))
1048
# We start searching the whole ancestry. It is a bit wasteful,
1049
# though. We really just want to mark all of these nodes as
1050
# 'seen' and then start just the tips. However, it requires a
1051
# get_parent_map() call to figure out the tips anyway, and all
1052
# redundant requests should be fairly fast.
1053
for searcher in common_searchers:
1054
searcher.start_searching(newly_seen_common)
1056
# If a 'common' node is an ancestor of all unique searchers, we
1057
# can stop searching it.
1058
stop_searching_common = ancestor_all_unique.intersection(
1060
if stop_searching_common:
1061
for searcher in common_searchers:
1062
searcher.stop_searching_any(stop_searching_common)
1063
if new_common_unique:
1064
# We found some ancestors that are common
1065
for searcher in unique_searchers:
1066
new_common_unique.update(
1067
searcher.find_seen_ancestors(new_common_unique))
1068
# Since these are common, we can grab another set of ancestors
1070
for searcher in common_searchers:
1071
new_common_unique.update(
1072
searcher.find_seen_ancestors(new_common_unique))
1074
# We can tell all of the unique searchers to start at these
1075
# nodes, and tell all of the common searchers to *stop*
1076
# searching these nodes
1077
for searcher in unique_searchers:
1078
searcher.start_searching(new_common_unique)
1079
for searcher in common_searchers:
1080
searcher.stop_searching_any(new_common_unique)
1081
ancestor_all_unique.update(new_common_unique)
1083
# Filter out searchers that don't actually search different
1084
# nodes. We already have the ancestry intersection for them
1085
next_unique_searchers = []
1086
unique_search_sets = set()
1087
for searcher in unique_searchers:
1088
will_search_set = frozenset(searcher._next_query)
1089
if will_search_set not in unique_search_sets:
1090
# This searcher is searching a unique set of nodes, let it
1091
unique_search_sets.add(will_search_set)
1092
next_unique_searchers.append(searcher)
1093
unique_searchers = next_unique_searchers
1094
for searcher in common_searchers:
1095
if searcher._next_query:
1098
# All common searcher have stopped searching
1101
def _remove_simple_descendants(self, revisions, parent_map):
1102
"""remove revisions which are children of other ones in the set
1104
This doesn't do any graph searching, it just checks the immediate
1105
parent_map to find if there are any children which can be removed.
1107
:param revisions: A set of revision_ids
1108
:return: A set of revision_ids with the children removed
1110
simple_ancestors = revisions.copy()
1111
# TODO: jam 20071214 we *could* restrict it to searching only the
1112
# parent_map of revisions already present in 'revisions', but
1113
# considering the general use case, I think this is actually
1116
# This is the same as the following loop. I don't know that it is any
1118
## simple_ancestors.difference_update(r for r, p_ids in parent_map.iteritems()
1119
## if p_ids is not None and revisions.intersection(p_ids))
1120
## return simple_ancestors
1122
# Yet Another Way, invert the parent map (which can be cached)
1124
## for revision_id, parent_ids in parent_map.iteritems():
1125
## for p_id in parent_ids:
1126
## descendants.setdefault(p_id, []).append(revision_id)
1127
## for revision in revisions.intersection(descendants):
1128
## simple_ancestors.difference_update(descendants[revision])
1129
## return simple_ancestors
1130
for revision, parent_ids in parent_map.iteritems():
1131
if parent_ids is None:
1133
for parent_id in parent_ids:
1134
if parent_id in revisions:
1135
# This node has a parent present in the set, so we can
1137
simple_ancestors.discard(revision)
1139
return simple_ancestors
1142
class HeadsCache(object):
1143
"""A cache of results for graph heads calls."""
1145
def __init__(self, graph):
1149
def heads(self, keys):
1150
"""Return the heads of keys.
1152
This matches the API of Graph.heads(), specifically the return value is
1153
a set which can be mutated, and ordering of the input is not preserved
1156
:see also: Graph.heads.
1157
:param keys: The keys to calculate heads for.
1158
:return: A set containing the heads, which may be mutated without
1159
affecting future lookups.
1161
keys = frozenset(keys)
1163
return set(self._heads[keys])
1165
heads = self.graph.heads(keys)
1166
self._heads[keys] = heads
1170
class FrozenHeadsCache(object):
1171
"""Cache heads() calls, assuming the caller won't modify them."""
1173
def __init__(self, graph):
1177
def heads(self, keys):
1178
"""Return the heads of keys.
1180
Similar to Graph.heads(). The main difference is that the return value
1181
is a frozen set which cannot be mutated.
1183
:see also: Graph.heads.
1184
:param keys: The keys to calculate heads for.
1185
:return: A frozenset containing the heads.
1187
keys = frozenset(keys)
1189
return self._heads[keys]
1191
heads = frozenset(self.graph.heads(keys))
1192
self._heads[keys] = heads
1195
def cache(self, keys, heads):
1196
"""Store a known value."""
1197
self._heads[frozenset(keys)] = frozenset(heads)
1200
class _BreadthFirstSearcher(object):
1201
"""Parallel search breadth-first the ancestry of revisions.
1203
This class implements the iterator protocol, but additionally
1204
1. provides a set of seen ancestors, and
1205
2. allows some ancestries to be unsearched, via stop_searching_any
1208
def __init__(self, revisions, parents_provider):
1209
self._iterations = 0
1210
self._next_query = set(revisions)
1212
self._started_keys = set(self._next_query)
1213
self._stopped_keys = set()
1214
self._parents_provider = parents_provider
1215
self._returning = 'next_with_ghosts'
1216
self._current_present = set()
1217
self._current_ghosts = set()
1218
self._current_parents = {}
1221
if self._iterations:
1222
prefix = "searching"
1225
search = '%s=%r' % (prefix, list(self._next_query))
1226
return ('_BreadthFirstSearcher(iterations=%d, %s,'
1227
' seen=%r)' % (self._iterations, search, list(self.seen)))
1229
def get_result(self):
1230
"""Get a SearchResult for the current state of this searcher.
1232
:return: A SearchResult for this search so far. The SearchResult is
1233
static - the search can be advanced and the search result will not
1234
be invalidated or altered.
1236
if self._returning == 'next':
1237
# We have to know the current nodes children to be able to list the
1238
# exclude keys for them. However, while we could have a second
1239
# look-ahead result buffer and shuffle things around, this method
1240
# is typically only called once per search - when memoising the
1241
# results of the search.
1242
found, ghosts, next, parents = self._do_query(self._next_query)
1243
# pretend we didn't query: perhaps we should tweak _do_query to be
1244
# entirely stateless?
1245
self.seen.difference_update(next)
1246
next_query = next.union(ghosts)
1248
next_query = self._next_query
1249
excludes = self._stopped_keys.union(next_query)
1250
included_keys = self.seen.difference(excludes)
1251
return SearchResult(self._started_keys, excludes, len(included_keys),
1257
except StopIteration:
1261
"""Return the next ancestors of this revision.
1263
Ancestors are returned in the order they are seen in a breadth-first
1264
traversal. No ancestor will be returned more than once. Ancestors are
1265
returned before their parentage is queried, so ghosts and missing
1266
revisions (including the start revisions) are included in the result.
1267
This can save a round trip in LCA style calculation by allowing
1268
convergence to be detected without reading the data for the revision
1269
the convergence occurs on.
1271
:return: A set of revision_ids.
1273
if self._returning != 'next':
1274
# switch to returning the query, not the results.
1275
self._returning = 'next'
1276
self._iterations += 1
1279
if len(self._next_query) == 0:
1280
raise StopIteration()
1281
# We have seen what we're querying at this point as we are returning
1282
# the query, not the results.
1283
self.seen.update(self._next_query)
1284
return self._next_query
1286
def next_with_ghosts(self):
1287
"""Return the next found ancestors, with ghosts split out.
1289
Ancestors are returned in the order they are seen in a breadth-first
1290
traversal. No ancestor will be returned more than once. Ancestors are
1291
returned only after asking for their parents, which allows us to detect
1292
which revisions are ghosts and which are not.
1294
:return: A tuple with (present ancestors, ghost ancestors) sets.
1296
if self._returning != 'next_with_ghosts':
1297
# switch to returning the results, not the current query.
1298
self._returning = 'next_with_ghosts'
1300
if len(self._next_query) == 0:
1301
raise StopIteration()
1303
return self._current_present, self._current_ghosts
1306
"""Advance the search.
1308
Updates self.seen, self._next_query, self._current_present,
1309
self._current_ghosts, self._current_parents and self._iterations.
1311
self._iterations += 1
1312
found, ghosts, next, parents = self._do_query(self._next_query)
1313
self._current_present = found
1314
self._current_ghosts = ghosts
1315
self._next_query = next
1316
self._current_parents = parents
1317
# ghosts are implicit stop points, otherwise the search cannot be
1318
# repeated when ghosts are filled.
1319
self._stopped_keys.update(ghosts)
1321
def _do_query(self, revisions):
1322
"""Query for revisions.
1324
Adds revisions to the seen set.
1326
:param revisions: Revisions to query.
1327
:return: A tuple: (set(found_revisions), set(ghost_revisions),
1328
set(parents_of_found_revisions), dict(found_revisions:parents)).
1330
found_revisions = set()
1331
parents_of_found = set()
1332
# revisions may contain nodes that point to other nodes in revisions:
1333
# we want to filter them out.
1334
self.seen.update(revisions)
1335
parent_map = self._parents_provider.get_parent_map(revisions)
1336
found_revisions.update(parent_map)
1337
for rev_id, parents in parent_map.iteritems():
1340
new_found_parents = [p for p in parents if p not in self.seen]
1341
if new_found_parents:
1342
# Calling set.update() with an empty generator is actually
1344
parents_of_found.update(new_found_parents)
1345
ghost_revisions = revisions - found_revisions
1346
return found_revisions, ghost_revisions, parents_of_found, parent_map
1351
def find_seen_ancestors(self, revisions):
1352
"""Find ancestors of these revisions that have already been seen.
1354
This function generally makes the assumption that querying for the
1355
parents of a node that has already been queried is reasonably cheap.
1356
(eg, not a round trip to a remote host).
1358
# TODO: Often we might ask one searcher for its seen ancestors, and
1359
# then ask another searcher the same question. This can result in
1360
# searching the same revisions repeatedly if the two searchers
1361
# have a lot of overlap.
1362
all_seen = self.seen
1363
pending = set(revisions).intersection(all_seen)
1364
seen_ancestors = set(pending)
1366
if self._returning == 'next':
1367
# self.seen contains what nodes have been returned, not what nodes
1368
# have been queried. We don't want to probe for nodes that haven't
1369
# been searched yet.
1370
not_searched_yet = self._next_query
1372
not_searched_yet = ()
1373
pending.difference_update(not_searched_yet)
1374
get_parent_map = self._parents_provider.get_parent_map
1376
parent_map = get_parent_map(pending)
1378
# We don't care if it is a ghost, since it can't be seen if it is
1380
for parent_ids in parent_map.itervalues():
1381
all_parents.extend(parent_ids)
1382
next_pending = all_seen.intersection(all_parents).difference(seen_ancestors)
1383
seen_ancestors.update(next_pending)
1384
next_pending.difference_update(not_searched_yet)
1385
pending = next_pending
1387
return seen_ancestors
1389
def stop_searching_any(self, revisions):
1391
Remove any of the specified revisions from the search list.
1393
None of the specified revisions are required to be present in the
1396
It is okay to call stop_searching_any() for revisions which were seen
1397
in previous iterations. It is the callers responsibility to call
1398
find_seen_ancestors() to make sure that current search tips that are
1399
ancestors of those revisions are also stopped. All explicitly stopped
1400
revisions will be excluded from the search result's get_keys(), though.
1402
# TODO: does this help performance?
1405
revisions = frozenset(revisions)
1406
if self._returning == 'next':
1407
stopped = self._next_query.intersection(revisions)
1408
self._next_query = self._next_query.difference(revisions)
1410
stopped_present = self._current_present.intersection(revisions)
1411
stopped = stopped_present.union(
1412
self._current_ghosts.intersection(revisions))
1413
self._current_present.difference_update(stopped)
1414
self._current_ghosts.difference_update(stopped)
1415
# stopping 'x' should stop returning parents of 'x', but
1416
# not if 'y' always references those same parents
1417
stop_rev_references = {}
1418
for rev in stopped_present:
1419
for parent_id in self._current_parents[rev]:
1420
if parent_id not in stop_rev_references:
1421
stop_rev_references[parent_id] = 0
1422
stop_rev_references[parent_id] += 1
1423
# if only the stopped revisions reference it, the ref count will be
1425
for parents in self._current_parents.itervalues():
1426
for parent_id in parents:
1428
stop_rev_references[parent_id] -= 1
1431
stop_parents = set()
1432
for rev_id, refs in stop_rev_references.iteritems():
1434
stop_parents.add(rev_id)
1435
self._next_query.difference_update(stop_parents)
1436
self._stopped_keys.update(stopped)
1437
self._stopped_keys.update(revisions)
1440
def start_searching(self, revisions):
1441
"""Add revisions to the search.
1443
The parents of revisions will be returned from the next call to next()
1444
or next_with_ghosts(). If next_with_ghosts was the most recently used
1445
next* call then the return value is the result of looking up the
1446
ghost/not ghost status of revisions. (A tuple (present, ghosted)).
1448
revisions = frozenset(revisions)
1449
self._started_keys.update(revisions)
1450
new_revisions = revisions.difference(self.seen)
1451
if self._returning == 'next':
1452
self._next_query.update(new_revisions)
1453
self.seen.update(new_revisions)
1455
# perform a query on revisions
1456
revs, ghosts, query, parents = self._do_query(revisions)
1457
self._stopped_keys.update(ghosts)
1458
self._current_present.update(revs)
1459
self._current_ghosts.update(ghosts)
1460
self._next_query.update(query)
1461
self._current_parents.update(parents)
1465
class SearchResult(object):
1466
"""The result of a breadth first search.
1468
A SearchResult provides the ability to reconstruct the search or access a
1469
set of the keys the search found.
1472
def __init__(self, start_keys, exclude_keys, key_count, keys):
1473
"""Create a SearchResult.
1475
:param start_keys: The keys the search started at.
1476
:param exclude_keys: The keys the search excludes.
1477
:param key_count: The total number of keys (from start to but not
1479
:param keys: The keys the search found. Note that in future we may get
1480
a SearchResult from a smart server, in which case the keys list is
1481
not necessarily immediately available.
1483
self._recipe = ('search', start_keys, exclude_keys, key_count)
1484
self._keys = frozenset(keys)
1486
def get_recipe(self):
1487
"""Return a recipe that can be used to replay this search.
1489
The recipe allows reconstruction of the same results at a later date
1490
without knowing all the found keys. The essential elements are a list
1491
of keys to start and to stop at. In order to give reproducible
1492
results when ghosts are encountered by a search they are automatically
1493
added to the exclude list (or else ghost filling may alter the
1496
:return: A tuple ('search', start_keys_set, exclude_keys_set,
1497
revision_count). To recreate the results of this search, create a
1498
breadth first searcher on the same graph starting at start_keys.
1499
Then call next() (or next_with_ghosts()) repeatedly, and on every
1500
result, call stop_searching_any on any keys from the exclude_keys
1501
set. The revision_count value acts as a trivial cross-check - the
1502
found revisions of the new search should have as many elements as
1503
revision_count. If it does not, then additional revisions have been
1504
ghosted since the search was executed the first time and the second
1510
"""Return the keys found in this search.
1512
:return: A set of keys.
1517
"""Return false if the search lists 1 or more revisions."""
1518
return self._recipe[3] == 0
1520
def refine(self, seen, referenced):
1521
"""Create a new search by refining this search.
1523
:param seen: Revisions that have been satisfied.
1524
:param referenced: Revision references observed while satisfying some
1527
start = self._recipe[1]
1528
exclude = self._recipe[2]
1529
count = self._recipe[3]
1530
keys = self.get_keys()
1531
# New heads = referenced + old heads - seen things - exclude
1532
pending_refs = set(referenced)
1533
pending_refs.update(start)
1534
pending_refs.difference_update(seen)
1535
pending_refs.difference_update(exclude)
1536
# New exclude = old exclude + satisfied heads
1537
seen_heads = start.intersection(seen)
1538
exclude.update(seen_heads)
1539
# keys gets seen removed
1541
# length is reduced by len(seen)
1543
return SearchResult(pending_refs, exclude, count, keys)
1546
class PendingAncestryResult(object):
1547
"""A search result that will reconstruct the ancestry for some graph heads.
1549
Unlike SearchResult, this doesn't hold the complete search result in
1550
memory, it just holds a description of how to generate it.
1553
def __init__(self, heads, repo):
1556
:param heads: an iterable of graph heads.
1557
:param repo: a repository to use to generate the ancestry for the given
1560
self.heads = frozenset(heads)
1563
def get_recipe(self):
1564
"""Return a recipe that can be used to replay this search.
1566
The recipe allows reconstruction of the same results at a later date.
1568
:seealso SearchResult.get_recipe:
1570
:return: A tuple ('proxy-search', start_keys_set, set(), -1)
1571
To recreate this result, create a PendingAncestryResult with the
1574
return ('proxy-search', self.heads, set(), -1)
1577
"""See SearchResult.get_keys.
1579
Returns all the keys for the ancestry of the heads, excluding
1582
return self._get_keys(self.repo.get_graph())
1584
def _get_keys(self, graph):
1585
NULL_REVISION = revision.NULL_REVISION
1586
keys = [key for (key, parents) in graph.iter_ancestry(self.heads)
1587
if key != NULL_REVISION and parents is not None]
1591
"""Return false if the search lists 1 or more revisions."""
1592
if revision.NULL_REVISION in self.heads:
1593
return len(self.heads) == 1
1595
return len(self.heads) == 0
1597
def refine(self, seen, referenced):
1598
"""Create a new search by refining this search.
1600
:param seen: Revisions that have been satisfied.
1601
:param referenced: Revision references observed while satisfying some
1604
referenced = self.heads.union(referenced)
1605
return PendingAncestryResult(referenced - seen, self.repo)
1608
def collapse_linear_regions(parent_map):
1609
"""Collapse regions of the graph that are 'linear'.
1615
can be collapsed by removing B and getting::
1619
:param parent_map: A dictionary mapping children to their parents
1620
:return: Another dictionary with 'linear' chains collapsed
1622
# Note: this isn't a strictly minimal collapse. For example:
1630
# Will not have 'D' removed, even though 'E' could fit. Also:
1636
# A and C are both kept because they are edges of the graph. We *could* get
1637
# rid of A if we wanted.
1645
# Will not have any nodes removed, even though you do have an
1646
# 'uninteresting' linear D->B and E->C
1648
for child, parents in parent_map.iteritems():
1649
children.setdefault(child, [])
1651
children.setdefault(p, []).append(child)
1653
orig_children = dict(children)
1655
result = dict(parent_map)
1656
for node in parent_map:
1657
parents = result[node]
1658
if len(parents) == 1:
1659
parent_children = children[parents[0]]
1660
if len(parent_children) != 1:
1661
# This is not the only child
1663
node_children = children[node]
1664
if len(node_children) != 1:
1666
child_parents = result.get(node_children[0], None)
1667
if len(child_parents) != 1:
1668
# This is not its only parent
1670
# The child of this node only points at it, and the parent only has
1671
# this as a child. remove this node, and join the others together
1672
result[node_children[0]] = parents
1673
children[parents[0]] = node_children
1681
_counters = [0,0,0,0,0,0,0]
1683
from bzrlib._known_graph_pyx import KnownGraph
1685
from bzrlib._known_graph_py import KnownGraph
added
removed
bzrlib/graph.py
