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
23
from bzrlib.deprecated_graph import (node_distances, select_farthest)
25
# DIAGRAM of terminology
35
# In this diagram, relative to G and H:
36
# A, B, C, D, E are common ancestors.
37
# C, D and E are border ancestors, because each has a non-common descendant.
38
# D and E are least common ancestors because none of their descendants are
40
# C is not a least common ancestor because its descendant, E, is a common
43
# The find_unique_lca algorithm will pick A in two steps:
44
# 1. find_lca('G', 'H') => ['D', 'E']
45
# 2. Since len(['D', 'E']) > 1, find_lca('D', 'E') => ['A']
48
class DictParentsProvider(object):
49
"""A parents provider for Graph objects."""
51
def __init__(self, ancestry):
52
self.ancestry = ancestry
55
return 'DictParentsProvider(%r)' % self.ancestry
57
def get_parent_map(self, keys):
58
"""See _StackedParentsProvider.get_parent_map"""
59
ancestry = self.ancestry
60
return dict((k, ancestry[k]) for k in keys if k in ancestry)
63
class _StackedParentsProvider(object):
65
def __init__(self, parent_providers):
66
self._parent_providers = parent_providers
69
return "_StackedParentsProvider(%r)" % self._parent_providers
71
def get_parent_map(self, keys):
72
"""Get a mapping of keys => parents
74
A dictionary is returned with an entry for each key present in this
75
source. If this source doesn't have information about a key, it should
78
[NULL_REVISION] is used as the parent of the first user-committed
79
revision. Its parent list is empty.
81
:param keys: An iterable returning keys to check (eg revision_ids)
82
:return: A dictionary mapping each key to its parents
86
for parents_provider in self._parent_providers:
87
new_found = parents_provider.get_parent_map(remaining)
88
found.update(new_found)
89
remaining.difference_update(new_found)
95
class CachingParentsProvider(object):
96
"""A parents provider which will cache the revision => parents in a dict.
98
This is useful for providers that have an expensive lookup.
101
def __init__(self, parent_provider):
102
self._real_provider = parent_provider
103
# Theoretically we could use an LRUCache here
107
return "%s(%r)" % (self.__class__.__name__, self._real_provider)
109
def get_parent_map(self, keys):
110
"""See _StackedParentsProvider.get_parent_map"""
112
# If the _real_provider doesn't have a key, we cache a value of None,
113
# which we then later use to realize we cannot provide a value for that
120
if value is not None:
121
parent_map[key] = value
126
new_parents = self._real_provider.get_parent_map(needed)
127
cache.update(new_parents)
128
parent_map.update(new_parents)
129
needed.difference_update(new_parents)
130
cache.update(dict.fromkeys(needed, None))
135
"""Provide incremental access to revision graphs.
137
This is the generic implementation; it is intended to be subclassed to
138
specialize it for other repository types.
141
def __init__(self, parents_provider):
142
"""Construct a Graph that uses several graphs as its input
144
This should not normally be invoked directly, because there may be
145
specialized implementations for particular repository types. See
146
Repository.get_graph().
148
:param parents_provider: An object providing a get_parent_map call
149
conforming to the behavior of
150
StackedParentsProvider.get_parent_map.
152
if getattr(parents_provider, 'get_parents', None) is not None:
153
self.get_parents = parents_provider.get_parents
154
if getattr(parents_provider, 'get_parent_map', None) is not None:
155
self.get_parent_map = parents_provider.get_parent_map
156
self._parents_provider = parents_provider
159
return 'Graph(%r)' % self._parents_provider
161
def find_lca(self, *revisions):
162
"""Determine the lowest common ancestors of the provided revisions
164
A lowest common ancestor is a common ancestor none of whose
165
descendants are common ancestors. In graphs, unlike trees, there may
166
be multiple lowest common ancestors.
168
This algorithm has two phases. Phase 1 identifies border ancestors,
169
and phase 2 filters border ancestors to determine lowest common
172
In phase 1, border ancestors are identified, using a breadth-first
173
search starting at the bottom of the graph. Searches are stopped
174
whenever a node or one of its descendants is determined to be common
176
In phase 2, the border ancestors are filtered to find the least
177
common ancestors. This is done by searching the ancestries of each
180
Phase 2 is perfomed on the principle that a border ancestor that is
181
not an ancestor of any other border ancestor is a least common
184
Searches are stopped when they find a node that is determined to be a
185
common ancestor of all border ancestors, because this shows that it
186
cannot be a descendant of any border ancestor.
188
The scaling of this operation should be proportional to
189
1. The number of uncommon ancestors
190
2. The number of border ancestors
191
3. The length of the shortest path between a border ancestor and an
192
ancestor of all border ancestors.
194
border_common, common, sides = self._find_border_ancestors(revisions)
195
# We may have common ancestors that can be reached from each other.
196
# - ask for the heads of them to filter it down to only ones that
197
# cannot be reached from each other - phase 2.
198
return self.heads(border_common)
200
def find_difference(self, left_revision, right_revision):
201
"""Determine the graph difference between two revisions"""
202
border, common, (left, right) = self._find_border_ancestors(
203
[left_revision, right_revision])
204
return (left.difference(right).difference(common),
205
right.difference(left).difference(common))
207
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
208
def get_parents(self, revisions):
209
"""Find revision ids of the parents of a list of revisions
211
A list is returned of the same length as the input. Each entry
212
is a list of parent ids for the corresponding input revision.
214
[NULL_REVISION] is used as the parent of the first user-committed
215
revision. Its parent list is empty.
217
If the revision is not present (i.e. a ghost), None is used in place
218
of the list of parents.
220
Deprecated in bzr 1.2 - please see get_parent_map.
222
parents = self.get_parent_map(revisions)
223
return [parent.get(r, None) for r in revisions]
225
def get_parent_map(self, revisions):
226
"""Get a map of key:parent_list for revisions.
228
This implementation delegates to get_parents, for old parent_providers
229
that do not supply get_parent_map.
232
for rev, parents in self.get_parents(revisions):
233
if parents is not None:
234
result[rev] = parents
237
def _make_breadth_first_searcher(self, revisions):
238
return _BreadthFirstSearcher(revisions, self)
240
def _find_border_ancestors(self, revisions):
241
"""Find common ancestors with at least one uncommon descendant.
243
Border ancestors are identified using a breadth-first
244
search starting at the bottom of the graph. Searches are stopped
245
whenever a node or one of its descendants is determined to be common.
247
This will scale with the number of uncommon ancestors.
249
As well as the border ancestors, a set of seen common ancestors and a
250
list of sets of seen ancestors for each input revision is returned.
251
This allows calculation of graph difference from the results of this
254
if None in revisions:
255
raise errors.InvalidRevisionId(None, self)
256
common_searcher = self._make_breadth_first_searcher([])
257
common_ancestors = set()
258
searchers = [self._make_breadth_first_searcher([r])
260
active_searchers = searchers[:]
261
border_ancestors = set()
262
def update_common(searcher, revisions):
263
w_seen_ancestors = searcher.find_seen_ancestors(
265
stopped = searcher.stop_searching_any(w_seen_ancestors)
266
common_ancestors.update(w_seen_ancestors)
267
common_searcher.start_searching(stopped)
270
if len(active_searchers) == 0:
271
return border_ancestors, common_ancestors, [s.seen for s in
274
new_common = common_searcher.next()
275
common_ancestors.update(new_common)
276
except StopIteration:
279
for searcher in active_searchers:
280
for revision in new_common.intersection(searcher.seen):
281
update_common(searcher, revision)
284
new_active_searchers = []
285
for searcher in active_searchers:
287
newly_seen.update(searcher.next())
288
except StopIteration:
291
new_active_searchers.append(searcher)
292
active_searchers = new_active_searchers
293
for revision in newly_seen:
294
if revision in common_ancestors:
295
for searcher in searchers:
296
update_common(searcher, revision)
298
for searcher in searchers:
299
if revision not in searcher.seen:
302
border_ancestors.add(revision)
303
for searcher in searchers:
304
update_common(searcher, revision)
306
def heads(self, keys):
307
"""Return the heads from amongst keys.
309
This is done by searching the ancestries of each key. Any key that is
310
reachable from another key is not returned; all the others are.
312
This operation scales with the relative depth between any two keys. If
313
any two keys are completely disconnected all ancestry of both sides
316
:param keys: An iterable of keys.
317
:return: A set of the heads. Note that as a set there is no ordering
318
information. Callers will need to filter their input to create
319
order if they need it.
321
candidate_heads = set(keys)
322
if revision.NULL_REVISION in candidate_heads:
323
# NULL_REVISION is only a head if it is the only entry
324
candidate_heads.remove(revision.NULL_REVISION)
325
if not candidate_heads:
326
return set([revision.NULL_REVISION])
327
if len(candidate_heads) < 2:
328
return candidate_heads
329
searchers = dict((c, self._make_breadth_first_searcher([c]))
330
for c in candidate_heads)
331
active_searchers = dict(searchers)
332
# skip over the actual candidate for each searcher
333
for searcher in active_searchers.itervalues():
335
# The common walker finds nodes that are common to two or more of the
336
# input keys, so that we don't access all history when a currently
337
# uncommon search point actually meets up with something behind a
338
# common search point. Common search points do not keep searches
339
# active; they just allow us to make searches inactive without
340
# accessing all history.
341
common_walker = self._make_breadth_first_searcher([])
342
while len(active_searchers) > 0:
347
except StopIteration:
348
# No common points being searched at this time.
350
for candidate in active_searchers.keys():
352
searcher = active_searchers[candidate]
354
# rare case: we deleted candidate in a previous iteration
355
# through this for loop, because it was determined to be
356
# a descendant of another candidate.
359
ancestors.update(searcher.next())
360
except StopIteration:
361
del active_searchers[candidate]
363
# process found nodes
365
for ancestor in ancestors:
366
if ancestor in candidate_heads:
367
candidate_heads.remove(ancestor)
368
del searchers[ancestor]
369
if ancestor in active_searchers:
370
del active_searchers[ancestor]
371
# it may meet up with a known common node
372
if ancestor in common_walker.seen:
373
# some searcher has encountered our known common nodes:
375
ancestor_set = set([ancestor])
376
for searcher in searchers.itervalues():
377
searcher.stop_searching_any(ancestor_set)
379
# or it may have been just reached by all the searchers:
380
for searcher in searchers.itervalues():
381
if ancestor not in searcher.seen:
384
# The final active searcher has just reached this node,
385
# making it be known as a descendant of all candidates,
386
# so we can stop searching it, and any seen ancestors
387
new_common.add(ancestor)
388
for searcher in searchers.itervalues():
390
searcher.find_seen_ancestors(ancestor)
391
searcher.stop_searching_any(seen_ancestors)
392
common_walker.start_searching(new_common)
393
return candidate_heads
395
def find_unique_lca(self, left_revision, right_revision,
397
"""Find a unique LCA.
399
Find lowest common ancestors. If there is no unique common
400
ancestor, find the lowest common ancestors of those ancestors.
402
Iteration stops when a unique lowest common ancestor is found.
403
The graph origin is necessarily a unique lowest common ancestor.
405
Note that None is not an acceptable substitute for NULL_REVISION.
406
in the input for this method.
408
:param count_steps: If True, the return value will be a tuple of
409
(unique_lca, steps) where steps is the number of times that
410
find_lca was run. If False, only unique_lca is returned.
412
revisions = [left_revision, right_revision]
416
lca = self.find_lca(*revisions)
424
raise errors.NoCommonAncestor(left_revision, right_revision)
427
def iter_ancestry(self, revision_ids):
428
"""Iterate the ancestry of this revision.
430
:param revision_ids: Nodes to start the search
431
:return: Yield tuples mapping a revision_id to its parents for the
432
ancestry of revision_id.
433
Ghosts will be returned with None as their parents, and nodes
434
with no parents will have NULL_REVISION as their only parent. (As
435
defined by get_parent_map.)
436
There will also be a node for (NULL_REVISION, ())
438
pending = set(revision_ids)
441
processed.update(pending)
442
next_map = self.get_parent_map(pending)
444
for item in next_map.iteritems():
446
next_pending.update(p for p in item[1] if p not in processed)
447
ghosts = pending.difference(next_map)
450
pending = next_pending
452
def iter_topo_order(self, revisions):
453
"""Iterate through the input revisions in topological order.
455
This sorting only ensures that parents come before their children.
456
An ancestor may sort after a descendant if the relationship is not
457
visible in the supplied list of revisions.
459
sorter = tsort.TopoSorter(self.get_parent_map(revisions))
460
return sorter.iter_topo_order()
462
def is_ancestor(self, candidate_ancestor, candidate_descendant):
463
"""Determine whether a revision is an ancestor of another.
465
We answer this using heads() as heads() has the logic to perform the
466
smallest number of parent lookups to determine the ancestral
467
relationship between N revisions.
469
return set([candidate_descendant]) == self.heads(
470
[candidate_ancestor, candidate_descendant])
473
class HeadsCache(object):
474
"""A cache of results for graph heads calls."""
476
def __init__(self, graph):
480
def heads(self, keys):
481
"""Return the heads of keys.
483
This matches the API of Graph.heads(), specifically the return value is
484
a set which can be mutated, and ordering of the input is not preserved
487
:see also: Graph.heads.
488
:param keys: The keys to calculate heads for.
489
:return: A set containing the heads, which may be mutated without
490
affecting future lookups.
492
keys = frozenset(keys)
494
return set(self._heads[keys])
496
heads = self.graph.heads(keys)
497
self._heads[keys] = heads
501
class _BreadthFirstSearcher(object):
502
"""Parallel search breadth-first the ancestry of revisions.
504
This class implements the iterator protocol, but additionally
505
1. provides a set of seen ancestors, and
506
2. allows some ancestries to be unsearched, via stop_searching_any
509
def __init__(self, revisions, parents_provider):
511
self._next_query = set(revisions)
513
self._started_keys = set(self._next_query)
514
self._stopped_keys = set()
515
self._parents_provider = parents_provider
516
self._returning = 'next_with_ghosts'
517
self._current_present = set()
518
self._current_ghosts = set()
519
self._current_parents = {}
526
search = '%s=%r' % (prefix, list(self._next_query))
527
return ('_BreadthFirstSearcher(iterations=%d, %s,'
528
' seen=%r)' % (self._iterations, search, list(self.seen)))
530
def get_result(self):
531
"""Get a SearchResult for the current state of this searcher.
533
:return: A SearchResult for this search so far. The SearchResult is
534
static - the search can be advanced and the search result will not
535
be invalidated or altered.
537
if self._returning == 'next':
538
# We have to know the current nodes children to be able to list the
539
# exclude keys for them. However, while we could have a second
540
# look-ahead result buffer and shuffle things around, this method
541
# is typically only called once per search - when memoising the
542
# results of the search.
543
found, ghosts, next, parents = self._do_query(self._next_query)
544
# pretend we didn't query: perhaps we should tweak _do_query to be
545
# entirely stateless?
546
self.seen.difference_update(next)
547
next_query = next.union(ghosts)
549
next_query = self._next_query
550
excludes = self._stopped_keys.union(next_query)
551
included_keys = self.seen.difference(excludes)
552
return SearchResult(self._started_keys, excludes, len(included_keys),
556
"""Return the next ancestors of this revision.
558
Ancestors are returned in the order they are seen in a breadth-first
559
traversal. No ancestor will be returned more than once. Ancestors are
560
returned before their parentage is queried, so ghosts and missing
561
revisions (including the start revisions) are included in the result.
562
This can save a round trip in LCA style calculation by allowing
563
convergence to be detected without reading the data for the revision
564
the convergence occurs on.
566
:return: A set of revision_ids.
568
if self._returning != 'next':
569
# switch to returning the query, not the results.
570
self._returning = 'next'
571
self._iterations += 1
574
if len(self._next_query) == 0:
575
raise StopIteration()
576
# We have seen what we're querying at this point as we are returning
577
# the query, not the results.
578
self.seen.update(self._next_query)
579
return self._next_query
581
def next_with_ghosts(self):
582
"""Return the next found ancestors, with ghosts split out.
584
Ancestors are returned in the order they are seen in a breadth-first
585
traversal. No ancestor will be returned more than once. Ancestors are
586
returned only after asking for their parents, which allows us to detect
587
which revisions are ghosts and which are not.
589
:return: A tuple with (present ancestors, ghost ancestors) sets.
591
if self._returning != 'next_with_ghosts':
592
# switch to returning the results, not the current query.
593
self._returning = 'next_with_ghosts'
595
if len(self._next_query) == 0:
596
raise StopIteration()
598
return self._current_present, self._current_ghosts
601
"""Advance the search.
603
Updates self.seen, self._next_query, self._current_present,
604
self._current_ghosts, self._current_parents and self._iterations.
606
self._iterations += 1
607
found, ghosts, next, parents = self._do_query(self._next_query)
608
self._current_present = found
609
self._current_ghosts = ghosts
610
self._next_query = next
611
self._current_parents = parents
612
# ghosts are implicit stop points, otherwise the search cannot be
613
# repeated when ghosts are filled.
614
self._stopped_keys.update(ghosts)
616
def _do_query(self, revisions):
617
"""Query for revisions.
619
Adds revisions to the seen set.
621
:param revisions: Revisions to query.
622
:return: A tuple: (set(found_revisions), set(ghost_revisions),
623
set(parents_of_found_revisions), dict(found_revisions:parents)).
625
found_parents = set()
626
parents_of_found = set()
627
# revisions may contain nodes that point to other nodes in revisions:
628
# we want to filter them out.
629
self.seen.update(revisions)
630
parent_map = self._parents_provider.get_parent_map(revisions)
631
for rev_id, parents in parent_map.iteritems():
632
found_parents.add(rev_id)
633
parents_of_found.update(p for p in parents if p not in self.seen)
634
ghost_parents = revisions - found_parents
635
return found_parents, ghost_parents, parents_of_found, parent_map
640
def find_seen_ancestors(self, revision):
641
"""Find ancestors of this revision that have already been seen."""
642
searcher = _BreadthFirstSearcher([revision], self._parents_provider)
643
seen_ancestors = set()
644
for ancestors in searcher:
645
for ancestor in ancestors:
646
if ancestor not in self.seen:
647
searcher.stop_searching_any([ancestor])
649
seen_ancestors.add(ancestor)
650
return seen_ancestors
652
def stop_searching_any(self, revisions):
654
Remove any of the specified revisions from the search list.
656
None of the specified revisions are required to be present in the
657
search list. In this case, the call is a no-op.
659
revisions = frozenset(revisions)
660
if self._returning == 'next':
661
stopped = self._next_query.intersection(revisions)
662
self._next_query = self._next_query.difference(revisions)
664
stopped_present = self._current_present.intersection(revisions)
665
stopped = stopped_present.union(
666
self._current_ghosts.intersection(revisions))
667
self._current_present.difference_update(stopped)
668
self._current_ghosts.difference_update(stopped)
669
# stopping 'x' should stop returning parents of 'x', but
670
# not if 'y' always references those same parents
671
stop_rev_references = {}
672
for rev in stopped_present:
673
for parent_id in self._current_parents[rev]:
674
if parent_id not in stop_rev_references:
675
stop_rev_references[parent_id] = 0
676
stop_rev_references[parent_id] += 1
677
# if only the stopped revisions reference it, the ref count will be
679
for parents in self._current_parents.itervalues():
680
for parent_id in parents:
682
stop_rev_references[parent_id] -= 1
686
for rev_id, refs in stop_rev_references.iteritems():
688
stop_parents.add(rev_id)
689
self._next_query.difference_update(stop_parents)
690
self._stopped_keys.update(stopped)
693
def start_searching(self, revisions):
694
"""Add revisions to the search.
696
The parents of revisions will be returned from the next call to next()
697
or next_with_ghosts(). If next_with_ghosts was the most recently used
698
next* call then the return value is the result of looking up the
699
ghost/not ghost status of revisions. (A tuple (present, ghosted)).
701
revisions = frozenset(revisions)
702
self._started_keys.update(revisions)
703
new_revisions = revisions.difference(self.seen)
704
revs, ghosts, query, parents = self._do_query(revisions)
705
self._stopped_keys.update(ghosts)
706
if self._returning == 'next':
707
self._next_query.update(new_revisions)
709
# perform a query on revisions
710
self._current_present.update(revs)
711
self._current_ghosts.update(ghosts)
712
self._next_query.update(query)
713
self._current_parents.update(parents)
717
class SearchResult(object):
718
"""The result of a breadth first search.
720
A SearchResult provides the ability to reconstruct the search or access a
721
set of the keys the search found.
724
def __init__(self, start_keys, exclude_keys, key_count, keys):
725
"""Create a SearchResult.
727
:param start_keys: The keys the search started at.
728
:param exclude_keys: The keys the search excludes.
729
:param key_count: The total number of keys (from start to but not
731
:param keys: The keys the search found. Note that in future we may get
732
a SearchResult from a smart server, in which case the keys list is
733
not necessarily immediately available.
735
self._recipe = (start_keys, exclude_keys, key_count)
736
self._keys = frozenset(keys)
738
def get_recipe(self):
739
"""Return a recipe that can be used to replay this search.
741
The recipe allows reconstruction of the same results at a later date
742
without knowing all the found keys. The essential elements are a list
743
of keys to start and and to stop at. In order to give reproducible
744
results when ghosts are encountered by a search they are automatically
745
added to the exclude list (or else ghost filling may alter the
748
:return: A tuple (start_keys_set, exclude_keys_set, revision_count). To
749
recreate the results of this search, create a breadth first
750
searcher on the same graph starting at start_keys. Then call next()
751
(or next_with_ghosts()) repeatedly, and on every result, call
752
stop_searching_any on any keys from the exclude_keys set. The
753
revision_count value acts as a trivial cross-check - the found
754
revisions of the new search should have as many elements as
755
revision_count. If it does not, then additional revisions have been
756
ghosted since the search was executed the first time and the second
762
"""Return the keys found in this search.
764
:return: A set of keys.
added
removed
bzrlib/graph.py
