1
# Copyright (C) 2005, 2006 Canonical Ltd
1
# Copyright (C) 2005-2010 Canonical Ltd
3
3
# This program is free software; you can redistribute it and/or modify
4
4
# it under the terms of the GNU General Public License as published by
5
5
# the Free Software Foundation; either version 2 of the License, or
6
6
# (at your option) any later version.
8
8
# This program is distributed in the hope that it will be useful,
9
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
11
# GNU General Public License for more details.
13
13
# You should have received a copy of the GNU General Public License
14
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
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from binascii import hexlify
18
from copy import deepcopy
19
from cStringIO import StringIO
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
22
from unittest import TestSuite
24
from bzrlib import bzrdir, check, delta, gpg, errors, xml5, ui, transactions, osutils
25
from bzrlib.decorators import needs_read_lock, needs_write_lock
26
from bzrlib.errors import InvalidRevisionId
27
from bzrlib.graph import Graph
43
revision as _mod_revision,
50
from bzrlib.bundle import serializer
51
from bzrlib.revisiontree import RevisionTree
52
from bzrlib.store.versioned import VersionedFileStore
53
from bzrlib.testament import Testament
62
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
28
63
from bzrlib.inter import InterObject
29
from bzrlib.inventory import Inventory
30
from bzrlib.knit import KnitVersionedFile, KnitPlainFactory
31
from bzrlib.lockable_files import LockableFiles, TransportLock
32
from bzrlib.lockdir import LockDir
33
from bzrlib.osutils import (safe_unicode, rand_bytes, compact_date,
35
from bzrlib.revision import NULL_REVISION, Revision
36
from bzrlib.store.versioned import VersionedFileStore, WeaveStore
37
from bzrlib.store.text import TextStore
38
from bzrlib.symbol_versioning import (deprecated_method,
41
from bzrlib.trace import mutter, note
42
from bzrlib.tree import RevisionTree, EmptyTree
43
from bzrlib.tsort import topo_sort
44
from bzrlib.testament import Testament
45
from bzrlib.tree import EmptyTree
46
from bzrlib.weave import WeaveFile
49
class Repository(object):
64
from bzrlib.inventory import (
70
from bzrlib.recordcounter import RecordCounter
71
from bzrlib.lock import _RelockDebugMixin, LogicalLockResult
72
from bzrlib.trace import (
73
log_exception_quietly, note, mutter, mutter_callsite, warning)
76
# Old formats display a warning, but only once
77
_deprecation_warning_done = False
80
class IsInWriteGroupError(errors.InternalBzrError):
82
_fmt = "May not refresh_data of repo %(repo)s while in a write group."
84
def __init__(self, repo):
85
errors.InternalBzrError.__init__(self, repo=repo)
88
class CommitBuilder(object):
89
"""Provides an interface to build up a commit.
91
This allows describing a tree to be committed without needing to
92
know the internals of the format of the repository.
95
# all clients should supply tree roots.
96
record_root_entry = True
97
# the default CommitBuilder does not manage trees whose root is versioned.
98
_versioned_root = False
100
def __init__(self, repository, parents, config, timestamp=None,
101
timezone=None, committer=None, revprops=None,
103
"""Initiate a CommitBuilder.
105
:param repository: Repository to commit to.
106
:param parents: Revision ids of the parents of the new revision.
107
:param config: Configuration to use.
108
:param timestamp: Optional timestamp recorded for commit.
109
:param timezone: Optional timezone for timestamp.
110
:param committer: Optional committer to set for commit.
111
:param revprops: Optional dictionary of revision properties.
112
:param revision_id: Optional revision id.
114
self._config = config
116
if committer is None:
117
self._committer = self._config.username()
118
elif not isinstance(committer, unicode):
119
self._committer = committer.decode() # throw if non-ascii
121
self._committer = committer
123
self.new_inventory = Inventory(None)
124
self._new_revision_id = revision_id
125
self.parents = parents
126
self.repository = repository
129
if revprops is not None:
130
self._validate_revprops(revprops)
131
self._revprops.update(revprops)
133
if timestamp is None:
134
timestamp = time.time()
135
# Restrict resolution to 1ms
136
self._timestamp = round(timestamp, 3)
139
self._timezone = osutils.local_time_offset()
141
self._timezone = int(timezone)
143
self._generate_revision_if_needed()
144
self.__heads = graph.HeadsCache(repository.get_graph()).heads
145
self._basis_delta = []
146
# API compatibility, older code that used CommitBuilder did not call
147
# .record_delete(), which means the delta that is computed would not be
148
# valid. Callers that will call record_delete() should call
149
# .will_record_deletes() to indicate that.
150
self._recording_deletes = False
151
# memo'd check for no-op commits.
152
self._any_changes = False
154
def any_changes(self):
155
"""Return True if any entries were changed.
157
This includes merge-only changes. It is the core for the --unchanged
160
:return: True if any changes have occured.
162
return self._any_changes
164
def _validate_unicode_text(self, text, context):
165
"""Verify things like commit messages don't have bogus characters."""
167
raise ValueError('Invalid value for %s: %r' % (context, text))
169
def _validate_revprops(self, revprops):
170
for key, value in revprops.iteritems():
171
# We know that the XML serializers do not round trip '\r'
172
# correctly, so refuse to accept them
173
if not isinstance(value, basestring):
174
raise ValueError('revision property (%s) is not a valid'
175
' (unicode) string: %r' % (key, value))
176
self._validate_unicode_text(value,
177
'revision property (%s)' % (key,))
179
def commit(self, message):
180
"""Make the actual commit.
182
:return: The revision id of the recorded revision.
184
self._validate_unicode_text(message, 'commit message')
185
rev = _mod_revision.Revision(
186
timestamp=self._timestamp,
187
timezone=self._timezone,
188
committer=self._committer,
190
inventory_sha1=self.inv_sha1,
191
revision_id=self._new_revision_id,
192
properties=self._revprops)
193
rev.parent_ids = self.parents
194
self.repository.add_revision(self._new_revision_id, rev,
195
self.new_inventory, self._config)
196
self.repository.commit_write_group()
197
return self._new_revision_id
200
"""Abort the commit that is being built.
202
self.repository.abort_write_group()
204
def revision_tree(self):
205
"""Return the tree that was just committed.
207
After calling commit() this can be called to get a RevisionTree
208
representing the newly committed tree. This is preferred to
209
calling Repository.revision_tree() because that may require
210
deserializing the inventory, while we already have a copy in
213
if self.new_inventory is None:
214
self.new_inventory = self.repository.get_inventory(
215
self._new_revision_id)
216
return RevisionTree(self.repository, self.new_inventory,
217
self._new_revision_id)
219
def finish_inventory(self):
220
"""Tell the builder that the inventory is finished.
222
:return: The inventory id in the repository, which can be used with
223
repository.get_inventory.
225
if self.new_inventory is None:
226
# an inventory delta was accumulated without creating a new
228
basis_id = self.basis_delta_revision
229
# We ignore the 'inventory' returned by add_inventory_by_delta
230
# because self.new_inventory is used to hint to the rest of the
231
# system what code path was taken
232
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
233
basis_id, self._basis_delta, self._new_revision_id,
236
if self.new_inventory.root is None:
237
raise AssertionError('Root entry should be supplied to'
238
' record_entry_contents, as of bzr 0.10.')
239
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
240
self.new_inventory.revision_id = self._new_revision_id
241
self.inv_sha1 = self.repository.add_inventory(
242
self._new_revision_id,
246
return self._new_revision_id
248
def _gen_revision_id(self):
249
"""Return new revision-id."""
250
return generate_ids.gen_revision_id(self._committer, self._timestamp)
252
def _generate_revision_if_needed(self):
253
"""Create a revision id if None was supplied.
255
If the repository can not support user-specified revision ids
256
they should override this function and raise CannotSetRevisionId
257
if _new_revision_id is not None.
259
:raises: CannotSetRevisionId
261
if self._new_revision_id is None:
262
self._new_revision_id = self._gen_revision_id()
263
self.random_revid = True
265
self.random_revid = False
267
def _heads(self, file_id, revision_ids):
268
"""Calculate the graph heads for revision_ids in the graph of file_id.
270
This can use either a per-file graph or a global revision graph as we
271
have an identity relationship between the two graphs.
273
return self.__heads(revision_ids)
275
def _check_root(self, ie, parent_invs, tree):
276
"""Helper for record_entry_contents.
278
:param ie: An entry being added.
279
:param parent_invs: The inventories of the parent revisions of the
281
:param tree: The tree that is being committed.
283
# In this revision format, root entries have no knit or weave When
284
# serializing out to disk and back in root.revision is always
286
ie.revision = self._new_revision_id
288
def _require_root_change(self, tree):
289
"""Enforce an appropriate root object change.
291
This is called once when record_iter_changes is called, if and only if
292
the root was not in the delta calculated by record_iter_changes.
294
:param tree: The tree which is being committed.
296
if len(self.parents) == 0:
297
raise errors.RootMissing()
298
entry = entry_factory['directory'](tree.path2id(''), '',
300
entry.revision = self._new_revision_id
301
self._basis_delta.append(('', '', entry.file_id, entry))
303
def _get_delta(self, ie, basis_inv, path):
304
"""Get a delta against the basis inventory for ie."""
305
if ie.file_id not in basis_inv:
307
result = (None, path, ie.file_id, ie)
308
self._basis_delta.append(result)
310
elif ie != basis_inv[ie.file_id]:
312
# TODO: avoid tis id2path call.
313
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
314
self._basis_delta.append(result)
320
def get_basis_delta(self):
321
"""Return the complete inventory delta versus the basis inventory.
323
This has been built up with the calls to record_delete and
324
record_entry_contents. The client must have already called
325
will_record_deletes() to indicate that they will be generating a
328
:return: An inventory delta, suitable for use with apply_delta, or
329
Repository.add_inventory_by_delta, etc.
331
if not self._recording_deletes:
332
raise AssertionError("recording deletes not activated.")
333
return self._basis_delta
335
def record_delete(self, path, file_id):
336
"""Record that a delete occured against a basis tree.
338
This is an optional API - when used it adds items to the basis_delta
339
being accumulated by the commit builder. It cannot be called unless the
340
method will_record_deletes() has been called to inform the builder that
341
a delta is being supplied.
343
:param path: The path of the thing deleted.
344
:param file_id: The file id that was deleted.
346
if not self._recording_deletes:
347
raise AssertionError("recording deletes not activated.")
348
delta = (path, None, file_id, None)
349
self._basis_delta.append(delta)
350
self._any_changes = True
353
def will_record_deletes(self):
354
"""Tell the commit builder that deletes are being notified.
356
This enables the accumulation of an inventory delta; for the resulting
357
commit to be valid, deletes against the basis MUST be recorded via
358
builder.record_delete().
360
self._recording_deletes = True
362
basis_id = self.parents[0]
364
basis_id = _mod_revision.NULL_REVISION
365
self.basis_delta_revision = basis_id
367
def record_entry_contents(self, ie, parent_invs, path, tree,
369
"""Record the content of ie from tree into the commit if needed.
371
Side effect: sets ie.revision when unchanged
373
:param ie: An inventory entry present in the commit.
374
:param parent_invs: The inventories of the parent revisions of the
376
:param path: The path the entry is at in the tree.
377
:param tree: The tree which contains this entry and should be used to
379
:param content_summary: Summary data from the tree about the paths
380
content - stat, length, exec, sha/link target. This is only
381
accessed when the entry has a revision of None - that is when it is
382
a candidate to commit.
383
:return: A tuple (change_delta, version_recorded, fs_hash).
384
change_delta is an inventory_delta change for this entry against
385
the basis tree of the commit, or None if no change occured against
387
version_recorded is True if a new version of the entry has been
388
recorded. For instance, committing a merge where a file was only
389
changed on the other side will return (delta, False).
390
fs_hash is either None, or the hash details for the path (currently
391
a tuple of the contents sha1 and the statvalue returned by
392
tree.get_file_with_stat()).
394
if self.new_inventory.root is None:
395
if ie.parent_id is not None:
396
raise errors.RootMissing()
397
self._check_root(ie, parent_invs, tree)
398
if ie.revision is None:
399
kind = content_summary[0]
401
# ie is carried over from a prior commit
403
# XXX: repository specific check for nested tree support goes here - if
404
# the repo doesn't want nested trees we skip it ?
405
if (kind == 'tree-reference' and
406
not self.repository._format.supports_tree_reference):
407
# mismatch between commit builder logic and repository:
408
# this needs the entry creation pushed down into the builder.
409
raise NotImplementedError('Missing repository subtree support.')
410
self.new_inventory.add(ie)
412
# TODO: slow, take it out of the inner loop.
414
basis_inv = parent_invs[0]
416
basis_inv = Inventory(root_id=None)
418
# ie.revision is always None if the InventoryEntry is considered
419
# for committing. We may record the previous parents revision if the
420
# content is actually unchanged against a sole head.
421
if ie.revision is not None:
422
if not self._versioned_root and path == '':
423
# repositories that do not version the root set the root's
424
# revision to the new commit even when no change occurs (more
425
# specifically, they do not record a revision on the root; and
426
# the rev id is assigned to the root during deserialisation -
427
# this masks when a change may have occurred against the basis.
428
# To match this we always issue a delta, because the revision
429
# of the root will always be changing.
430
if ie.file_id in basis_inv:
431
delta = (basis_inv.id2path(ie.file_id), path,
435
delta = (None, path, ie.file_id, ie)
436
self._basis_delta.append(delta)
437
return delta, False, None
439
# we don't need to commit this, because the caller already
440
# determined that an existing revision of this file is
441
# appropriate. If it's not being considered for committing then
442
# it and all its parents to the root must be unaltered so
443
# no-change against the basis.
444
if ie.revision == self._new_revision_id:
445
raise AssertionError("Impossible situation, a skipped "
446
"inventory entry (%r) claims to be modified in this "
447
"commit (%r).", (ie, self._new_revision_id))
448
return None, False, None
449
# XXX: Friction: parent_candidates should return a list not a dict
450
# so that we don't have to walk the inventories again.
451
parent_candiate_entries = ie.parent_candidates(parent_invs)
452
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
454
for inv in parent_invs:
455
if ie.file_id in inv:
456
old_rev = inv[ie.file_id].revision
457
if old_rev in head_set:
458
heads.append(inv[ie.file_id].revision)
459
head_set.remove(inv[ie.file_id].revision)
462
# now we check to see if we need to write a new record to the
464
# We write a new entry unless there is one head to the ancestors, and
465
# the kind-derived content is unchanged.
467
# Cheapest check first: no ancestors, or more the one head in the
468
# ancestors, we write a new node.
472
# There is a single head, look it up for comparison
473
parent_entry = parent_candiate_entries[heads[0]]
474
# if the non-content specific data has changed, we'll be writing a
476
if (parent_entry.parent_id != ie.parent_id or
477
parent_entry.name != ie.name):
479
# now we need to do content specific checks:
481
# if the kind changed the content obviously has
482
if kind != parent_entry.kind:
484
# Stat cache fingerprint feedback for the caller - None as we usually
485
# don't generate one.
488
if content_summary[2] is None:
489
raise ValueError("Files must not have executable = None")
491
# We can't trust a check of the file length because of content
493
if (# if the exec bit has changed we have to store:
494
parent_entry.executable != content_summary[2]):
496
elif parent_entry.text_sha1 == content_summary[3]:
497
# all meta and content is unchanged (using a hash cache
498
# hit to check the sha)
499
ie.revision = parent_entry.revision
500
ie.text_size = parent_entry.text_size
501
ie.text_sha1 = parent_entry.text_sha1
502
ie.executable = parent_entry.executable
503
return self._get_delta(ie, basis_inv, path), False, None
505
# Either there is only a hash change(no hash cache entry,
506
# or same size content change), or there is no change on
508
# Provide the parent's hash to the store layer, so that the
509
# content is unchanged we will not store a new node.
510
nostore_sha = parent_entry.text_sha1
512
# We want to record a new node regardless of the presence or
513
# absence of a content change in the file.
515
ie.executable = content_summary[2]
516
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
518
text = file_obj.read()
522
ie.text_sha1, ie.text_size = self._add_text_to_weave(
523
ie.file_id, text, heads, nostore_sha)
524
# Let the caller know we generated a stat fingerprint.
525
fingerprint = (ie.text_sha1, stat_value)
526
except errors.ExistingContent:
527
# Turns out that the file content was unchanged, and we were
528
# only going to store a new node if it was changed. Carry over
530
ie.revision = parent_entry.revision
531
ie.text_size = parent_entry.text_size
532
ie.text_sha1 = parent_entry.text_sha1
533
ie.executable = parent_entry.executable
534
return self._get_delta(ie, basis_inv, path), False, None
535
elif kind == 'directory':
537
# all data is meta here, nothing specific to directory, so
539
ie.revision = parent_entry.revision
540
return self._get_delta(ie, basis_inv, path), False, None
541
self._add_text_to_weave(ie.file_id, '', heads, None)
542
elif kind == 'symlink':
543
current_link_target = content_summary[3]
545
# symlink target is not generic metadata, check if it has
547
if current_link_target != parent_entry.symlink_target:
550
# unchanged, carry over.
551
ie.revision = parent_entry.revision
552
ie.symlink_target = parent_entry.symlink_target
553
return self._get_delta(ie, basis_inv, path), False, None
554
ie.symlink_target = current_link_target
555
self._add_text_to_weave(ie.file_id, '', heads, None)
556
elif kind == 'tree-reference':
558
if content_summary[3] != parent_entry.reference_revision:
561
# unchanged, carry over.
562
ie.reference_revision = parent_entry.reference_revision
563
ie.revision = parent_entry.revision
564
return self._get_delta(ie, basis_inv, path), False, None
565
ie.reference_revision = content_summary[3]
566
if ie.reference_revision is None:
567
raise AssertionError("invalid content_summary for nested tree: %r"
568
% (content_summary,))
569
self._add_text_to_weave(ie.file_id, '', heads, None)
571
raise NotImplementedError('unknown kind')
572
ie.revision = self._new_revision_id
573
self._any_changes = True
574
return self._get_delta(ie, basis_inv, path), True, fingerprint
576
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
577
_entry_factory=entry_factory):
578
"""Record a new tree via iter_changes.
580
:param tree: The tree to obtain text contents from for changed objects.
581
:param basis_revision_id: The revision id of the tree the iter_changes
582
has been generated against. Currently assumed to be the same
583
as self.parents[0] - if it is not, errors may occur.
584
:param iter_changes: An iter_changes iterator with the changes to apply
585
to basis_revision_id. The iterator must not include any items with
586
a current kind of None - missing items must be either filtered out
587
or errored-on beefore record_iter_changes sees the item.
588
:param _entry_factory: Private method to bind entry_factory locally for
590
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
593
# Create an inventory delta based on deltas between all the parents and
594
# deltas between all the parent inventories. We use inventory delta's
595
# between the inventory objects because iter_changes masks
596
# last-changed-field only changes.
598
# file_id -> change map, change is fileid, paths, changed, versioneds,
599
# parents, names, kinds, executables
601
# {file_id -> revision_id -> inventory entry, for entries in parent
602
# trees that are not parents[0]
606
revtrees = list(self.repository.revision_trees(self.parents))
607
except errors.NoSuchRevision:
608
# one or more ghosts, slow path.
610
for revision_id in self.parents:
612
revtrees.append(self.repository.revision_tree(revision_id))
613
except errors.NoSuchRevision:
615
basis_revision_id = _mod_revision.NULL_REVISION
617
revtrees.append(self.repository.revision_tree(
618
_mod_revision.NULL_REVISION))
619
# The basis inventory from a repository
621
basis_inv = revtrees[0].inventory
623
basis_inv = self.repository.revision_tree(
624
_mod_revision.NULL_REVISION).inventory
625
if len(self.parents) > 0:
626
if basis_revision_id != self.parents[0] and not ghost_basis:
628
"arbitrary basis parents not yet supported with merges")
629
for revtree in revtrees[1:]:
630
for change in revtree.inventory._make_delta(basis_inv):
631
if change[1] is None:
632
# Not present in this parent.
634
if change[2] not in merged_ids:
635
if change[0] is not None:
636
basis_entry = basis_inv[change[2]]
637
merged_ids[change[2]] = [
639
basis_entry.revision,
642
parent_entries[change[2]] = {
644
basis_entry.revision:basis_entry,
646
change[3].revision:change[3],
649
merged_ids[change[2]] = [change[3].revision]
650
parent_entries[change[2]] = {change[3].revision:change[3]}
652
merged_ids[change[2]].append(change[3].revision)
653
parent_entries[change[2]][change[3].revision] = change[3]
656
# Setup the changes from the tree:
657
# changes maps file_id -> (change, [parent revision_ids])
659
for change in iter_changes:
660
# This probably looks up in basis_inv way to much.
661
if change[1][0] is not None:
662
head_candidate = [basis_inv[change[0]].revision]
665
changes[change[0]] = change, merged_ids.get(change[0],
667
unchanged_merged = set(merged_ids) - set(changes)
668
# Extend the changes dict with synthetic changes to record merges of
670
for file_id in unchanged_merged:
671
# Record a merged version of these items that did not change vs the
672
# basis. This can be either identical parallel changes, or a revert
673
# of a specific file after a merge. The recorded content will be
674
# that of the current tree (which is the same as the basis), but
675
# the per-file graph will reflect a merge.
676
# NB:XXX: We are reconstructing path information we had, this
677
# should be preserved instead.
678
# inv delta change: (file_id, (path_in_source, path_in_target),
679
# changed_content, versioned, parent, name, kind,
682
basis_entry = basis_inv[file_id]
683
except errors.NoSuchId:
684
# a change from basis->some_parents but file_id isn't in basis
685
# so was new in the merge, which means it must have changed
686
# from basis -> current, and as it hasn't the add was reverted
687
# by the user. So we discard this change.
691
(basis_inv.id2path(file_id), tree.id2path(file_id)),
693
(basis_entry.parent_id, basis_entry.parent_id),
694
(basis_entry.name, basis_entry.name),
695
(basis_entry.kind, basis_entry.kind),
696
(basis_entry.executable, basis_entry.executable))
697
changes[file_id] = (change, merged_ids[file_id])
698
# changes contains tuples with the change and a set of inventory
699
# candidates for the file.
701
# old_path, new_path, file_id, new_inventory_entry
702
seen_root = False # Is the root in the basis delta?
703
inv_delta = self._basis_delta
704
modified_rev = self._new_revision_id
705
for change, head_candidates in changes.values():
706
if change[3][1]: # versioned in target.
707
# Several things may be happening here:
708
# We may have a fork in the per-file graph
709
# - record a change with the content from tree
710
# We may have a change against < all trees
711
# - carry over the tree that hasn't changed
712
# We may have a change against all trees
713
# - record the change with the content from tree
716
entry = _entry_factory[kind](file_id, change[5][1],
718
head_set = self._heads(change[0], set(head_candidates))
721
for head_candidate in head_candidates:
722
if head_candidate in head_set:
723
heads.append(head_candidate)
724
head_set.remove(head_candidate)
727
# Could be a carry-over situation:
728
parent_entry_revs = parent_entries.get(file_id, None)
729
if parent_entry_revs:
730
parent_entry = parent_entry_revs.get(heads[0], None)
733
if parent_entry is None:
734
# The parent iter_changes was called against is the one
735
# that is the per-file head, so any change is relevant
736
# iter_changes is valid.
737
carry_over_possible = False
739
# could be a carry over situation
740
# A change against the basis may just indicate a merge,
741
# we need to check the content against the source of the
742
# merge to determine if it was changed after the merge
744
if (parent_entry.kind != entry.kind or
745
parent_entry.parent_id != entry.parent_id or
746
parent_entry.name != entry.name):
747
# Metadata common to all entries has changed
748
# against per-file parent
749
carry_over_possible = False
751
carry_over_possible = True
752
# per-type checks for changes against the parent_entry
755
# Cannot be a carry-over situation
756
carry_over_possible = False
757
# Populate the entry in the delta
759
# XXX: There is still a small race here: If someone reverts the content of a file
760
# after iter_changes examines and decides it has changed,
761
# we will unconditionally record a new version even if some
762
# other process reverts it while commit is running (with
763
# the revert happening after iter_changes did its
766
entry.executable = True
768
entry.executable = False
769
if (carry_over_possible and
770
parent_entry.executable == entry.executable):
771
# Check the file length, content hash after reading
773
nostore_sha = parent_entry.text_sha1
776
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
778
text = file_obj.read()
782
entry.text_sha1, entry.text_size = self._add_text_to_weave(
783
file_id, text, heads, nostore_sha)
784
yield file_id, change[1][1], (entry.text_sha1, stat_value)
785
except errors.ExistingContent:
786
# No content change against a carry_over parent
787
# Perhaps this should also yield a fs hash update?
789
entry.text_size = parent_entry.text_size
790
entry.text_sha1 = parent_entry.text_sha1
791
elif kind == 'symlink':
793
entry.symlink_target = tree.get_symlink_target(file_id)
794
if (carry_over_possible and
795
parent_entry.symlink_target == entry.symlink_target):
798
self._add_text_to_weave(change[0], '', heads, None)
799
elif kind == 'directory':
800
if carry_over_possible:
803
# Nothing to set on the entry.
804
# XXX: split into the Root and nonRoot versions.
805
if change[1][1] != '' or self.repository.supports_rich_root():
806
self._add_text_to_weave(change[0], '', heads, None)
807
elif kind == 'tree-reference':
808
if not self.repository._format.supports_tree_reference:
809
# This isn't quite sane as an error, but we shouldn't
810
# ever see this code path in practice: tree's don't
811
# permit references when the repo doesn't support tree
813
raise errors.UnsupportedOperation(tree.add_reference,
815
reference_revision = tree.get_reference_revision(change[0])
816
entry.reference_revision = reference_revision
817
if (carry_over_possible and
818
parent_entry.reference_revision == reference_revision):
821
self._add_text_to_weave(change[0], '', heads, None)
823
raise AssertionError('unknown kind %r' % kind)
825
entry.revision = modified_rev
827
entry.revision = parent_entry.revision
830
new_path = change[1][1]
831
inv_delta.append((change[1][0], new_path, change[0], entry))
834
self.new_inventory = None
836
# This should perhaps be guarded by a check that the basis we
837
# commit against is the basis for the commit and if not do a delta
839
self._any_changes = True
841
# housekeeping root entry changes do not affect no-change commits.
842
self._require_root_change(tree)
843
self.basis_delta_revision = basis_revision_id
845
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
846
parent_keys = tuple([(file_id, parent) for parent in parents])
847
return self.repository.texts._add_text(
848
(file_id, self._new_revision_id), parent_keys, new_text,
849
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
852
class RootCommitBuilder(CommitBuilder):
853
"""This commitbuilder actually records the root id"""
855
# the root entry gets versioned properly by this builder.
856
_versioned_root = True
858
def _check_root(self, ie, parent_invs, tree):
859
"""Helper for record_entry_contents.
861
:param ie: An entry being added.
862
:param parent_invs: The inventories of the parent revisions of the
864
:param tree: The tree that is being committed.
867
def _require_root_change(self, tree):
868
"""Enforce an appropriate root object change.
870
This is called once when record_iter_changes is called, if and only if
871
the root was not in the delta calculated by record_iter_changes.
873
:param tree: The tree which is being committed.
875
# versioned roots do not change unless the tree found a change.
878
class RepositoryWriteLockResult(LogicalLockResult):
879
"""The result of write locking a repository.
881
:ivar repository_token: The token obtained from the underlying lock, or
883
:ivar unlock: A callable which will unlock the lock.
886
def __init__(self, unlock, repository_token):
887
LogicalLockResult.__init__(self, unlock)
888
self.repository_token = repository_token
891
return "RepositoryWriteLockResult(%s, %s)" % (self.repository_token,
895
######################################################################
899
class Repository(_RelockDebugMixin, controldir.ControlComponent):
50
900
"""Repository holding history for one or more branches.
52
902
The repository holds and retrieves historical information including
53
903
revisions and file history. It's normally accessed only by the Branch,
54
904
which views a particular line of development through that history.
56
The Repository builds on top of Stores and a Transport, which respectively
57
describe the disk data format and the way of accessing the (possibly
906
The Repository builds on top of some byte storage facilies (the revisions,
907
signatures, inventories, texts and chk_bytes attributes) and a Transport,
908
which respectively provide byte storage and a means to access the (possibly
911
The byte storage facilities are addressed via tuples, which we refer to
912
as 'keys' throughout the code base. Revision_keys, inventory_keys and
913
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
914
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
915
byte string made up of a hash identifier and a hash value.
916
We use this interface because it allows low friction with the underlying
917
code that implements disk indices, network encoding and other parts of
920
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
921
the serialised revisions for the repository. This can be used to obtain
922
revision graph information or to access raw serialised revisions.
923
The result of trying to insert data into the repository via this store
924
is undefined: it should be considered read-only except for implementors
926
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
927
the serialised signatures for the repository. This can be used to
928
obtain access to raw serialised signatures. The result of trying to
929
insert data into the repository via this store is undefined: it should
930
be considered read-only except for implementors of repositories.
931
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
932
the serialised inventories for the repository. This can be used to
933
obtain unserialised inventories. The result of trying to insert data
934
into the repository via this store is undefined: it should be
935
considered read-only except for implementors of repositories.
936
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
937
texts of files and directories for the repository. This can be used to
938
obtain file texts or file graphs. Note that Repository.iter_file_bytes
939
is usually a better interface for accessing file texts.
940
The result of trying to insert data into the repository via this store
941
is undefined: it should be considered read-only except for implementors
943
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
944
any data the repository chooses to store or have indexed by its hash.
945
The result of trying to insert data into the repository via this store
946
is undefined: it should be considered read-only except for implementors
948
:ivar _transport: Transport for file access to repository, typically
949
pointing to .bzr/repository.
62
def add_inventory(self, revid, inv, parents):
63
"""Add the inventory inv to the repository as revid.
65
:param parents: The revision ids of the parents that revid
952
# What class to use for a CommitBuilder. Often it's simpler to change this
953
# in a Repository class subclass rather than to override
954
# get_commit_builder.
955
_commit_builder_class = CommitBuilder
956
# The search regex used by xml based repositories to determine what things
957
# where changed in a single commit.
958
_file_ids_altered_regex = lazy_regex.lazy_compile(
959
r'file_id="(?P<file_id>[^"]+)"'
960
r'.* revision="(?P<revision_id>[^"]+)"'
963
def abort_write_group(self, suppress_errors=False):
964
"""Commit the contents accrued within the current write group.
966
:param suppress_errors: if true, abort_write_group will catch and log
967
unexpected errors that happen during the abort, rather than
968
allowing them to propagate. Defaults to False.
970
:seealso: start_write_group.
972
if self._write_group is not self.get_transaction():
973
# has an unlock or relock occured ?
976
'(suppressed) mismatched lock context and write group. %r, %r',
977
self._write_group, self.get_transaction())
979
raise errors.BzrError(
980
'mismatched lock context and write group. %r, %r' %
981
(self._write_group, self.get_transaction()))
983
self._abort_write_group()
984
except Exception, exc:
985
self._write_group = None
986
if not suppress_errors:
988
mutter('abort_write_group failed')
989
log_exception_quietly()
990
note('bzr: ERROR (ignored): %s', exc)
991
self._write_group = None
993
def _abort_write_group(self):
994
"""Template method for per-repository write group cleanup.
996
This is called during abort before the write group is considered to be
997
finished and should cleanup any internal state accrued during the write
998
group. There is no requirement that data handed to the repository be
999
*not* made available - this is not a rollback - but neither should any
1000
attempt be made to ensure that data added is fully commited. Abort is
1001
invoked when an error has occured so futher disk or network operations
1002
may not be possible or may error and if possible should not be
1006
def add_fallback_repository(self, repository):
1007
"""Add a repository to use for looking up data not held locally.
1009
:param repository: A repository.
1011
if not self._format.supports_external_lookups:
1012
raise errors.UnstackableRepositoryFormat(self._format, self.base)
1013
if self.is_locked():
1014
# This repository will call fallback.unlock() when we transition to
1015
# the unlocked state, so we make sure to increment the lock count
1016
repository.lock_read()
1017
self._check_fallback_repository(repository)
1018
self._fallback_repositories.append(repository)
1019
self.texts.add_fallback_versioned_files(repository.texts)
1020
self.inventories.add_fallback_versioned_files(repository.inventories)
1021
self.revisions.add_fallback_versioned_files(repository.revisions)
1022
self.signatures.add_fallback_versioned_files(repository.signatures)
1023
if self.chk_bytes is not None:
1024
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
1026
def _check_fallback_repository(self, repository):
1027
"""Check that this repository can fallback to repository safely.
1029
Raise an error if not.
1031
:param repository: A repository to fallback to.
1033
return InterRepository._assert_same_model(self, repository)
1035
def add_inventory(self, revision_id, inv, parents):
1036
"""Add the inventory inv to the repository as revision_id.
1038
:param parents: The revision ids of the parents that revision_id
66
1039
is known to have and are in the repository already.
68
returns the sha1 of the serialized inventory.
70
assert inv.revision_id is None or inv.revision_id == revid, \
71
"Mismatch between inventory revision" \
72
" id and insertion revid (%r, %r)" % (inv.revision_id, revid)
73
inv_text = xml5.serializer_v5.write_inventory_to_string(inv)
74
inv_sha1 = osutils.sha_string(inv_text)
75
inv_vf = self.control_weaves.get_weave('inventory',
76
self.get_transaction())
77
self._inventory_add_lines(inv_vf, revid, parents, osutils.split_lines(inv_text))
80
def _inventory_add_lines(self, inv_vf, revid, parents, lines):
82
for parent in parents:
84
final_parents.append(parent)
86
inv_vf.add_lines(revid, final_parents, lines)
89
def add_revision(self, rev_id, rev, inv=None, config=None):
90
"""Add rev to the revision store as rev_id.
92
:param rev_id: the revision id to use.
1041
:returns: The validator(which is a sha1 digest, though what is sha'd is
1042
repository format specific) of the serialized inventory.
1044
if not self.is_in_write_group():
1045
raise AssertionError("%r not in write group" % (self,))
1046
_mod_revision.check_not_reserved_id(revision_id)
1047
if not (inv.revision_id is None or inv.revision_id == revision_id):
1048
raise AssertionError(
1049
"Mismatch between inventory revision"
1050
" id and insertion revid (%r, %r)"
1051
% (inv.revision_id, revision_id))
1052
if inv.root is None:
1053
raise errors.RootMissing()
1054
return self._add_inventory_checked(revision_id, inv, parents)
1056
def _add_inventory_checked(self, revision_id, inv, parents):
1057
"""Add inv to the repository after checking the inputs.
1059
This function can be overridden to allow different inventory styles.
1061
:seealso: add_inventory, for the contract.
1063
inv_lines = self._serializer.write_inventory_to_lines(inv)
1064
return self._inventory_add_lines(revision_id, parents,
1065
inv_lines, check_content=False)
1067
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1068
parents, basis_inv=None, propagate_caches=False):
1069
"""Add a new inventory expressed as a delta against another revision.
1071
See the inventory developers documentation for the theory behind
1074
:param basis_revision_id: The inventory id the delta was created
1075
against. (This does not have to be a direct parent.)
1076
:param delta: The inventory delta (see Inventory.apply_delta for
1078
:param new_revision_id: The revision id that the inventory is being
1080
:param parents: The revision ids of the parents that revision_id is
1081
known to have and are in the repository already. These are supplied
1082
for repositories that depend on the inventory graph for revision
1083
graph access, as well as for those that pun ancestry with delta
1085
:param basis_inv: The basis inventory if it is already known,
1087
:param propagate_caches: If True, the caches for this inventory are
1088
copied to and updated for the result if possible.
1090
:returns: (validator, new_inv)
1091
The validator(which is a sha1 digest, though what is sha'd is
1092
repository format specific) of the serialized inventory, and the
1093
resulting inventory.
1095
if not self.is_in_write_group():
1096
raise AssertionError("%r not in write group" % (self,))
1097
_mod_revision.check_not_reserved_id(new_revision_id)
1098
basis_tree = self.revision_tree(basis_revision_id)
1099
basis_tree.lock_read()
1101
# Note that this mutates the inventory of basis_tree, which not all
1102
# inventory implementations may support: A better idiom would be to
1103
# return a new inventory, but as there is no revision tree cache in
1104
# repository this is safe for now - RBC 20081013
1105
if basis_inv is None:
1106
basis_inv = basis_tree.inventory
1107
basis_inv.apply_delta(delta)
1108
basis_inv.revision_id = new_revision_id
1109
return (self.add_inventory(new_revision_id, basis_inv, parents),
1114
def _inventory_add_lines(self, revision_id, parents, lines,
1115
check_content=True):
1116
"""Store lines in inv_vf and return the sha1 of the inventory."""
1117
parents = [(parent,) for parent in parents]
1118
result = self.inventories.add_lines((revision_id,), parents, lines,
1119
check_content=check_content)[0]
1120
self.inventories._access.flush()
1123
def add_revision(self, revision_id, rev, inv=None, config=None):
1124
"""Add rev to the revision store as revision_id.
1126
:param revision_id: the revision id to use.
93
1127
:param rev: The revision object.
94
1128
:param inv: The inventory for the revision. if None, it will be looked
95
1129
up in the inventory storer
154
1195
Returns a set of the present revisions.
157
for id in revision_ids:
158
if self.has_revision(id):
1198
graph = self.get_graph()
1199
parent_map = graph.get_parent_map(revision_ids)
1200
# The old API returned a list, should this actually be a set?
1201
return parent_map.keys()
1203
def _check_inventories(self, checker):
1204
"""Check the inventories found from the revision scan.
1206
This is responsible for verifying the sha1 of inventories and
1207
creating a pending_keys set that covers data referenced by inventories.
1209
bar = ui.ui_factory.nested_progress_bar()
1211
self._do_check_inventories(checker, bar)
1215
def _do_check_inventories(self, checker, bar):
1216
"""Helper for _check_inventories."""
1218
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1219
kinds = ['chk_bytes', 'texts']
1220
count = len(checker.pending_keys)
1221
bar.update("inventories", 0, 2)
1222
current_keys = checker.pending_keys
1223
checker.pending_keys = {}
1224
# Accumulate current checks.
1225
for key in current_keys:
1226
if key[0] != 'inventories' and key[0] not in kinds:
1227
checker._report_items.append('unknown key type %r' % (key,))
1228
keys[key[0]].add(key[1:])
1229
if keys['inventories']:
1230
# NB: output order *should* be roughly sorted - topo or
1231
# inverse topo depending on repository - either way decent
1232
# to just delta against. However, pre-CHK formats didn't
1233
# try to optimise inventory layout on disk. As such the
1234
# pre-CHK code path does not use inventory deltas.
1236
for record in self.inventories.check(keys=keys['inventories']):
1237
if record.storage_kind == 'absent':
1238
checker._report_items.append(
1239
'Missing inventory {%s}' % (record.key,))
1241
last_object = self._check_record('inventories', record,
1242
checker, last_object,
1243
current_keys[('inventories',) + record.key])
1244
del keys['inventories']
1247
bar.update("texts", 1)
1248
while (checker.pending_keys or keys['chk_bytes']
1250
# Something to check.
1251
current_keys = checker.pending_keys
1252
checker.pending_keys = {}
1253
# Accumulate current checks.
1254
for key in current_keys:
1255
if key[0] not in kinds:
1256
checker._report_items.append('unknown key type %r' % (key,))
1257
keys[key[0]].add(key[1:])
1258
# Check the outermost kind only - inventories || chk_bytes || texts
1262
for record in getattr(self, kind).check(keys=keys[kind]):
1263
if record.storage_kind == 'absent':
1264
checker._report_items.append(
1265
'Missing %s {%s}' % (kind, record.key,))
1267
last_object = self._check_record(kind, record,
1268
checker, last_object, current_keys[(kind,) + record.key])
1272
def _check_record(self, kind, record, checker, last_object, item_data):
1273
"""Check a single text from this repository."""
1274
if kind == 'inventories':
1275
rev_id = record.key[0]
1276
inv = self._deserialise_inventory(rev_id,
1277
record.get_bytes_as('fulltext'))
1278
if last_object is not None:
1279
delta = inv._make_delta(last_object)
1280
for old_path, path, file_id, ie in delta:
1283
ie.check(checker, rev_id, inv)
1285
for path, ie in inv.iter_entries():
1286
ie.check(checker, rev_id, inv)
1287
if self._format.fast_deltas:
1289
elif kind == 'chk_bytes':
1290
# No code written to check chk_bytes for this repo format.
1291
checker._report_items.append(
1292
'unsupported key type chk_bytes for %s' % (record.key,))
1293
elif kind == 'texts':
1294
self._check_text(record, checker, item_data)
1296
checker._report_items.append(
1297
'unknown key type %s for %s' % (kind, record.key))
1299
def _check_text(self, record, checker, item_data):
1300
"""Check a single text."""
1301
# Check it is extractable.
1302
# TODO: check length.
1303
if record.storage_kind == 'chunked':
1304
chunks = record.get_bytes_as(record.storage_kind)
1305
sha1 = osutils.sha_strings(chunks)
1306
length = sum(map(len, chunks))
1308
content = record.get_bytes_as('fulltext')
1309
sha1 = osutils.sha_string(content)
1310
length = len(content)
1311
if item_data and sha1 != item_data[1]:
1312
checker._report_items.append(
1313
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1314
(record.key, sha1, item_data[1], item_data[2]))
163
1317
def create(a_bzrdir):
164
1318
"""Construct the current default format repository in a_bzrdir."""
165
1319
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
167
def __init__(self, _format, a_bzrdir, control_files, _revision_store, control_store, text_store):
1321
def __init__(self, _format, a_bzrdir, control_files):
168
1322
"""instantiate a Repository.
170
1324
:param _format: The format of the repository on disk.
171
1325
:param a_bzrdir: The BzrDir of the repository.
173
In the future we will have a single api for all stores for
174
getting file texts, inventories and revisions, then
175
this construct will accept instances of those things.
1327
# In the future we will have a single api for all stores for
1328
# getting file texts, inventories and revisions, then
1329
# this construct will accept instances of those things.
177
1330
super(Repository, self).__init__()
178
1331
self._format = _format
179
1332
# the following are part of the public API for Repository:
180
1333
self.bzrdir = a_bzrdir
181
1334
self.control_files = control_files
182
self._revision_store = _revision_store
183
self.text_store = text_store
184
# backwards compatibility
185
self.weave_store = text_store
186
# not right yet - should be more semantically clear ?
188
self.control_store = control_store
189
self.control_weaves = control_store
190
# TODO: make sure to construct the right store classes, etc, depending
191
# on whether escaping is required.
1335
self._transport = control_files._transport
1336
self.base = self._transport.base
1338
self._reconcile_does_inventory_gc = True
1339
self._reconcile_fixes_text_parents = False
1340
self._reconcile_backsup_inventory = True
1341
self._write_group = None
1342
# Additional places to query for data.
1343
self._fallback_repositories = []
1344
# An InventoryEntry cache, used during deserialization
1345
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1346
# Is it safe to return inventory entries directly from the entry cache,
1347
# rather copying them?
1348
self._safe_to_return_from_cache = False
1351
def user_transport(self):
1352
return self.bzrdir.user_transport
1355
def control_transport(self):
1356
return self._transport
193
1358
def __repr__(self):
194
return '%s(%r)' % (self.__class__.__name__,
195
self.bzrdir.transport.base)
1359
if self._fallback_repositories:
1360
return '%s(%r, fallback_repositories=%r)' % (
1361
self.__class__.__name__,
1363
self._fallback_repositories)
1365
return '%s(%r)' % (self.__class__.__name__,
1368
def _has_same_fallbacks(self, other_repo):
1369
"""Returns true if the repositories have the same fallbacks."""
1370
my_fb = self._fallback_repositories
1371
other_fb = other_repo._fallback_repositories
1372
if len(my_fb) != len(other_fb):
1374
for f, g in zip(my_fb, other_fb):
1375
if not f.has_same_location(g):
1379
def has_same_location(self, other):
1380
"""Returns a boolean indicating if this repository is at the same
1381
location as another repository.
1383
This might return False even when two repository objects are accessing
1384
the same physical repository via different URLs.
1386
if self.__class__ is not other.__class__:
1388
return (self._transport.base == other._transport.base)
1390
def is_in_write_group(self):
1391
"""Return True if there is an open write group.
1393
:seealso: start_write_group.
1395
return self._write_group is not None
197
1397
def is_locked(self):
198
1398
return self.control_files.is_locked()
200
def lock_write(self):
201
self.control_files.lock_write()
1400
def is_write_locked(self):
1401
"""Return True if this object is write locked."""
1402
return self.is_locked() and self.control_files._lock_mode == 'w'
1404
def lock_write(self, token=None):
1405
"""Lock this repository for writing.
1407
This causes caching within the repository obejct to start accumlating
1408
data during reads, and allows a 'write_group' to be obtained. Write
1409
groups must be used for actual data insertion.
1411
A token should be passed in if you know that you have locked the object
1412
some other way, and need to synchronise this object's state with that
1415
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1417
:param token: if this is already locked, then lock_write will fail
1418
unless the token matches the existing lock.
1419
:returns: a token if this instance supports tokens, otherwise None.
1420
:raises TokenLockingNotSupported: when a token is given but this
1421
instance doesn't support using token locks.
1422
:raises MismatchedToken: if the specified token doesn't match the token
1423
of the existing lock.
1424
:seealso: start_write_group.
1425
:return: A RepositoryWriteLockResult.
1427
locked = self.is_locked()
1428
token = self.control_files.lock_write(token=token)
1430
self._warn_if_deprecated()
1431
self._note_lock('w')
1432
for repo in self._fallback_repositories:
1433
# Writes don't affect fallback repos
1435
self._refresh_data()
1436
return RepositoryWriteLockResult(self.unlock, token)
203
1438
def lock_read(self):
1439
"""Lock the repository for read operations.
1441
:return: An object with an unlock method which will release the lock
1444
locked = self.is_locked()
204
1445
self.control_files.lock_read()
1447
self._warn_if_deprecated()
1448
self._note_lock('r')
1449
for repo in self._fallback_repositories:
1451
self._refresh_data()
1452
return LogicalLockResult(self.unlock)
206
1454
def get_physical_lock_status(self):
207
1455
return self.control_files.get_physical_lock_status()
210
def missing_revision_ids(self, other, revision_id=None):
1457
def leave_lock_in_place(self):
1458
"""Tell this repository not to release the physical lock when this
1461
If lock_write doesn't return a token, then this method is not supported.
1463
self.control_files.leave_in_place()
1465
def dont_leave_lock_in_place(self):
1466
"""Tell this repository to release the physical lock when this
1467
object is unlocked, even if it didn't originally acquire it.
1469
If lock_write doesn't return a token, then this method is not supported.
1471
self.control_files.dont_leave_in_place()
1474
def gather_stats(self, revid=None, committers=None):
1475
"""Gather statistics from a revision id.
1477
:param revid: The revision id to gather statistics from, if None, then
1478
no revision specific statistics are gathered.
1479
:param committers: Optional parameter controlling whether to grab
1480
a count of committers from the revision specific statistics.
1481
:return: A dictionary of statistics. Currently this contains:
1482
committers: The number of committers if requested.
1483
firstrev: A tuple with timestamp, timezone for the penultimate left
1484
most ancestor of revid, if revid is not the NULL_REVISION.
1485
latestrev: A tuple with timestamp, timezone for revid, if revid is
1486
not the NULL_REVISION.
1487
revisions: The total revision count in the repository.
1488
size: An estimate disk size of the repository in bytes.
1491
if revid and committers:
1492
result['committers'] = 0
1493
if revid and revid != _mod_revision.NULL_REVISION:
1495
all_committers = set()
1496
revisions = self.get_ancestry(revid)
1497
# pop the leading None
1499
first_revision = None
1501
# ignore the revisions in the middle - just grab first and last
1502
revisions = revisions[0], revisions[-1]
1503
for revision in self.get_revisions(revisions):
1504
if not first_revision:
1505
first_revision = revision
1507
all_committers.add(revision.committer)
1508
last_revision = revision
1510
result['committers'] = len(all_committers)
1511
result['firstrev'] = (first_revision.timestamp,
1512
first_revision.timezone)
1513
result['latestrev'] = (last_revision.timestamp,
1514
last_revision.timezone)
1516
# now gather global repository information
1517
# XXX: This is available for many repos regardless of listability.
1518
if self.user_transport.listable():
1519
# XXX: do we want to __define len__() ?
1520
# Maybe the versionedfiles object should provide a different
1521
# method to get the number of keys.
1522
result['revisions'] = len(self.revisions.keys())
1523
# result['size'] = t
1526
def find_branches(self, using=False):
1527
"""Find branches underneath this repository.
1529
This will include branches inside other branches.
1531
:param using: If True, list only branches using this repository.
1533
if using and not self.is_shared():
1534
return self.bzrdir.list_branches()
1535
class Evaluator(object):
1538
self.first_call = True
1540
def __call__(self, bzrdir):
1541
# On the first call, the parameter is always the bzrdir
1542
# containing the current repo.
1543
if not self.first_call:
1545
repository = bzrdir.open_repository()
1546
except errors.NoRepositoryPresent:
1549
return False, ([], repository)
1550
self.first_call = False
1551
value = (bzrdir.list_branches(), None)
1555
for branches, repository in bzrdir.BzrDir.find_bzrdirs(
1556
self.user_transport, evaluate=Evaluator()):
1557
if branches is not None:
1558
ret.extend(branches)
1559
if not using and repository is not None:
1560
ret.extend(repository.find_branches())
1564
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
211
1565
"""Return the revision ids that other has that this does not.
213
1567
These are returned in topological order.
215
1569
revision_id: only return revision ids included by revision_id.
217
return InterRepository.get(other, self).missing_revision_ids(revision_id)
1571
return InterRepository.get(other, self).search_missing_revision_ids(
1572
revision_id, find_ghosts)
256
1761
:param revprops: Optional dictionary of revision properties.
257
1762
:param revision_id: Optional revision id.
259
return CommitBuilder(self, parents, config, timestamp, timezone,
260
committer, revprops, revision_id)
1764
if self._fallback_repositories:
1765
raise errors.BzrError("Cannot commit from a lightweight checkout "
1766
"to a stacked branch. See "
1767
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1768
result = self._commit_builder_class(self, parents, config,
1769
timestamp, timezone, committer, revprops, revision_id)
1770
self.start_write_group()
1773
@only_raises(errors.LockNotHeld, errors.LockBroken)
262
1774
def unlock(self):
1775
if (self.control_files._lock_count == 1 and
1776
self.control_files._lock_mode == 'w'):
1777
if self._write_group is not None:
1778
self.abort_write_group()
1779
self.control_files.unlock()
1780
raise errors.BzrError(
1781
'Must end write groups before releasing write locks.')
263
1782
self.control_files.unlock()
1783
if self.control_files._lock_count == 0:
1784
self._inventory_entry_cache.clear()
1785
for repo in self._fallback_repositories:
265
1788
@needs_read_lock
266
def clone(self, a_bzrdir, revision_id=None, basis=None):
1789
def clone(self, a_bzrdir, revision_id=None):
267
1790
"""Clone this repository into a_bzrdir using the current format.
269
1792
Currently no check is made that the format of this repository and
270
1793
the bzrdir format are compatible. FIXME RBC 20060201.
1795
:return: The newly created destination repository.
1797
# TODO: deprecate after 0.16; cloning this with all its settings is
1798
# probably not very useful -- mbp 20070423
1799
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1800
self.copy_content_into(dest_repo, revision_id)
1803
def start_write_group(self):
1804
"""Start a write group in the repository.
1806
Write groups are used by repositories which do not have a 1:1 mapping
1807
between file ids and backend store to manage the insertion of data from
1808
both fetch and commit operations.
1810
A write lock is required around the start_write_group/commit_write_group
1811
for the support of lock-requiring repository formats.
1813
One can only insert data into a repository inside a write group.
1817
if not self.is_write_locked():
1818
raise errors.NotWriteLocked(self)
1819
if self._write_group:
1820
raise errors.BzrError('already in a write group')
1821
self._start_write_group()
1822
# so we can detect unlock/relock - the write group is now entered.
1823
self._write_group = self.get_transaction()
1825
def _start_write_group(self):
1826
"""Template method for per-repository write group startup.
1828
This is called before the write group is considered to be
1833
def sprout(self, to_bzrdir, revision_id=None):
1834
"""Create a descendent repository for new development.
1836
Unlike clone, this does not copy the settings of the repository.
1838
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1839
dest_repo.fetch(self, revision_id=revision_id)
1842
def _create_sprouting_repo(self, a_bzrdir, shared):
272
1843
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
273
1844
# use target default format.
274
result = a_bzrdir.create_repository()
275
# FIXME RBC 20060209 split out the repository type to avoid this check ?
276
elif isinstance(a_bzrdir._format,
277
(bzrdir.BzrDirFormat4,
278
bzrdir.BzrDirFormat5,
279
bzrdir.BzrDirFormat6)):
280
result = a_bzrdir.open_repository()
1845
dest_repo = a_bzrdir.create_repository()
282
result = self._format.initialize(a_bzrdir, shared=self.is_shared())
283
self.copy_content_into(result, revision_id, basis)
1847
# Most control formats need the repository to be specifically
1848
# created, but on some old all-in-one formats it's not needed
1850
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1851
except errors.UninitializableFormat:
1852
dest_repo = a_bzrdir.open_repository()
1855
def _get_sink(self):
1856
"""Return a sink for streaming into this repository."""
1857
return StreamSink(self)
1859
def _get_source(self, to_format):
1860
"""Return a source for streaming from this repository."""
1861
return StreamSource(self, to_format)
286
1863
@needs_read_lock
287
1864
def has_revision(self, revision_id):
288
1865
"""True if this repository has a copy of the revision."""
289
return self._revision_store.has_revision_id(revision_id,
290
self.get_transaction())
1866
return revision_id in self.has_revisions((revision_id,))
1869
def has_revisions(self, revision_ids):
1870
"""Probe to find out the presence of multiple revisions.
1872
:param revision_ids: An iterable of revision_ids.
1873
:return: A set of the revision_ids that were present.
1875
parent_map = self.revisions.get_parent_map(
1876
[(rev_id,) for rev_id in revision_ids])
1878
if _mod_revision.NULL_REVISION in revision_ids:
1879
result.add(_mod_revision.NULL_REVISION)
1880
result.update([key[0] for key in parent_map])
1884
def get_revision(self, revision_id):
1885
"""Return the Revision object for a named revision."""
1886
return self.get_revisions([revision_id])[0]
292
1888
@needs_read_lock
293
1889
def get_revision_reconcile(self, revision_id):
294
1890
"""'reconcile' helper routine that allows access to a revision always.
296
1892
This variant of get_revision does not cross check the weave graph
297
1893
against the revision one as get_revision does: but it should only
298
1894
be used by reconcile, or reconcile-alike commands that are correcting
299
1895
or testing the revision graph.
301
if not revision_id or not isinstance(revision_id, basestring):
302
raise InvalidRevisionId(revision_id=revision_id, branch=self)
303
return self._revision_store.get_revisions([revision_id],
304
self.get_transaction())[0]
1897
return self._get_revisions([revision_id])[0]
305
1899
@needs_read_lock
306
1900
def get_revisions(self, revision_ids):
307
return self._revision_store.get_revisions(revision_ids,
308
self.get_transaction())
311
def get_revision_xml(self, revision_id):
312
rev = self.get_revision(revision_id)
314
# the current serializer..
315
self._revision_store._serializer.write_revision(rev, rev_tmp)
317
return rev_tmp.getvalue()
320
def get_revision(self, revision_id):
321
"""Return the Revision object for a named revision"""
322
r = self.get_revision_reconcile(revision_id)
323
# weave corruption can lead to absent revision markers that should be
325
# the following test is reasonably cheap (it needs a single weave read)
326
# and the weave is cached in read transactions. In write transactions
327
# it is not cached but typically we only read a small number of
328
# revisions. For knits when they are introduced we will probably want
329
# to ensure that caching write transactions are in use.
330
inv = self.get_inventory_weave()
331
self._check_revision_parents(r, inv)
335
def get_deltas_for_revisions(self, revisions):
1901
"""Get many revisions at once.
1903
Repositories that need to check data on every revision read should
1904
subclass this method.
1906
return self._get_revisions(revision_ids)
1909
def _get_revisions(self, revision_ids):
1910
"""Core work logic to get many revisions without sanity checks."""
1912
for revid, rev in self._iter_revisions(revision_ids):
1914
raise errors.NoSuchRevision(self, revid)
1916
return [revs[revid] for revid in revision_ids]
1918
def _iter_revisions(self, revision_ids):
1919
"""Iterate over revision objects.
1921
:param revision_ids: An iterable of revisions to examine. None may be
1922
passed to request all revisions known to the repository. Note that
1923
not all repositories can find unreferenced revisions; for those
1924
repositories only referenced ones will be returned.
1925
:return: An iterator of (revid, revision) tuples. Absent revisions (
1926
those asked for but not available) are returned as (revid, None).
1928
if revision_ids is None:
1929
revision_ids = self.all_revision_ids()
1931
for rev_id in revision_ids:
1932
if not rev_id or not isinstance(rev_id, basestring):
1933
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1934
keys = [(key,) for key in revision_ids]
1935
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1936
for record in stream:
1937
revid = record.key[0]
1938
if record.storage_kind == 'absent':
1941
text = record.get_bytes_as('fulltext')
1942
rev = self._serializer.read_revision_from_string(text)
1945
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
336
1946
"""Produce a generator of revision deltas.
338
1948
Note that the input is a sequence of REVISIONS, not revision_ids.
339
1949
Trees will be held in memory until the generator exits.
340
1950
Each delta is relative to the revision's lefthand predecessor.
1952
:param specific_fileids: if not None, the result is filtered
1953
so that only those file-ids, their parents and their
1954
children are included.
1956
# Get the revision-ids of interest
342
1957
required_trees = set()
343
1958
for revision in revisions:
344
1959
required_trees.add(revision.revision_id)
345
1960
required_trees.update(revision.parent_ids[:1])
346
trees = dict((t.get_revision_id(), t) for
347
t in self.revision_trees(required_trees))
1962
# Get the matching filtered trees. Note that it's more
1963
# efficient to pass filtered trees to changes_from() rather
1964
# than doing the filtering afterwards. changes_from() could
1965
# arguably do the filtering itself but it's path-based, not
1966
# file-id based, so filtering before or afterwards is
1968
if specific_fileids is None:
1969
trees = dict((t.get_revision_id(), t) for
1970
t in self.revision_trees(required_trees))
1972
trees = dict((t.get_revision_id(), t) for
1973
t in self._filtered_revision_trees(required_trees,
1976
# Calculate the deltas
348
1977
for revision in revisions:
349
1978
if not revision.parent_ids:
350
old_tree = EmptyTree()
1979
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
352
1981
old_tree = trees[revision.parent_ids[0]]
353
yield delta.compare_trees(old_tree, trees[revision.revision_id])
1982
yield trees[revision.revision_id].changes_from(old_tree)
355
1984
@needs_read_lock
356
def get_revision_delta(self, revision_id):
1985
def get_revision_delta(self, revision_id, specific_fileids=None):
357
1986
"""Return the delta for one revision.
359
1988
The delta is relative to the left-hand predecessor of the
1991
:param specific_fileids: if not None, the result is filtered
1992
so that only those file-ids, their parents and their
1993
children are included.
362
1995
r = self.get_revision(revision_id)
363
return list(self.get_deltas_for_revisions([r]))[0]
365
def _check_revision_parents(self, revision, inventory):
366
"""Private to Repository and Fetch.
368
This checks the parentage of revision in an inventory weave for
369
consistency and is only applicable to inventory-weave-for-ancestry
370
using repository formats & fetchers.
372
weave_parents = inventory.get_parents(revision.revision_id)
373
weave_names = inventory.versions()
374
for parent_id in revision.parent_ids:
375
if parent_id in weave_names:
376
# this parent must not be a ghost.
377
if not parent_id in weave_parents:
379
raise errors.CorruptRepository(self)
1996
return list(self.get_deltas_for_revisions([r],
1997
specific_fileids=specific_fileids))[0]
381
1999
@needs_write_lock
382
2000
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
383
2001
signature = gpg_strategy.sign(plaintext)
384
self._revision_store.add_revision_signature_text(revision_id,
386
self.get_transaction())
388
def fileids_altered_by_revision_ids(self, revision_ids):
389
"""Find the file ids and versions affected by revisions.
391
:param revisions: an iterable containing revision ids.
392
:return: a dictionary mapping altered file-ids to an iterable of
393
revision_ids. Each altered file-ids has the exact revision_ids that
394
altered it listed explicitly.
396
assert isinstance(self._format, (RepositoryFormat5,
399
RepositoryFormatKnit1)), \
400
("fileids_altered_by_revision_ids only supported for branches "
401
"which store inventory as unnested xml, not on %r" % self)
402
selected_revision_ids = set(revision_ids)
403
w = self.get_inventory_weave()
2002
self.add_signature_text(revision_id, signature)
2005
def add_signature_text(self, revision_id, signature):
2006
self.signatures.add_lines((revision_id,), (),
2007
osutils.split_lines(signature))
2009
def find_text_key_references(self):
2010
"""Find the text key references within the repository.
2012
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2013
to whether they were referred to by the inventory of the
2014
revision_id that they contain. The inventory texts from all present
2015
revision ids are assessed to generate this report.
2017
revision_keys = self.revisions.keys()
2018
w = self.inventories
2019
pb = ui.ui_factory.nested_progress_bar()
2021
return self._find_text_key_references_from_xml_inventory_lines(
2022
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2026
def _find_text_key_references_from_xml_inventory_lines(self,
2028
"""Core routine for extracting references to texts from inventories.
2030
This performs the translation of xml lines to revision ids.
2032
:param line_iterator: An iterator of lines, origin_version_id
2033
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2034
to whether they were referred to by the inventory of the
2035
revision_id that they contain. Note that if that revision_id was
2036
not part of the line_iterator's output then False will be given -
2037
even though it may actually refer to that key.
2039
if not self._serializer.support_altered_by_hack:
2040
raise AssertionError(
2041
"_find_text_key_references_from_xml_inventory_lines only "
2042
"supported for branches which store inventory as unnested xml"
2043
", not on %r" % self)
406
2046
# this code needs to read every new line in every inventory for the
407
2047
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
408
# not present in one of those inventories is unnecessary but not
2048
# not present in one of those inventories is unnecessary but not
409
2049
# harmful because we are filtering by the revision id marker in the
410
# inventory lines : we only select file ids altered in one of those
2050
# inventory lines : we only select file ids altered in one of those
411
2051
# revisions. We don't need to see all lines in the inventory because
412
2052
# only those added in an inventory in rev X can contain a revision=X
414
for line in w.iter_lines_added_or_present_in_versions(selected_revision_ids):
415
start = line.find('file_id="')+9
416
if start < 9: continue
417
end = line.find('"', start)
419
file_id = _unescape_xml(line[start:end])
421
start = line.find('revision="')+10
422
if start < 10: continue
423
end = line.find('"', start)
425
revision_id = _unescape_xml(line[start:end])
426
if revision_id in selected_revision_ids:
427
result.setdefault(file_id, set()).add(revision_id)
431
def get_inventory_weave(self):
432
return self.control_weaves.get_weave('inventory',
433
self.get_transaction())
2054
unescape_revid_cache = {}
2055
unescape_fileid_cache = {}
2057
# jam 20061218 In a big fetch, this handles hundreds of thousands
2058
# of lines, so it has had a lot of inlining and optimizing done.
2059
# Sorry that it is a little bit messy.
2060
# Move several functions to be local variables, since this is a long
2062
search = self._file_ids_altered_regex.search
2063
unescape = _unescape_xml
2064
setdefault = result.setdefault
2065
for line, line_key in line_iterator:
2066
match = search(line)
2069
# One call to match.group() returning multiple items is quite a
2070
# bit faster than 2 calls to match.group() each returning 1
2071
file_id, revision_id = match.group('file_id', 'revision_id')
2073
# Inlining the cache lookups helps a lot when you make 170,000
2074
# lines and 350k ids, versus 8.4 unique ids.
2075
# Using a cache helps in 2 ways:
2076
# 1) Avoids unnecessary decoding calls
2077
# 2) Re-uses cached strings, which helps in future set and
2079
# (2) is enough that removing encoding entirely along with
2080
# the cache (so we are using plain strings) results in no
2081
# performance improvement.
2083
revision_id = unescape_revid_cache[revision_id]
2085
unescaped = unescape(revision_id)
2086
unescape_revid_cache[revision_id] = unescaped
2087
revision_id = unescaped
2089
# Note that unconditionally unescaping means that we deserialise
2090
# every fileid, which for general 'pull' is not great, but we don't
2091
# really want to have some many fulltexts that this matters anyway.
2094
file_id = unescape_fileid_cache[file_id]
2096
unescaped = unescape(file_id)
2097
unescape_fileid_cache[file_id] = unescaped
2100
key = (file_id, revision_id)
2101
setdefault(key, False)
2102
if revision_id == line_key[-1]:
2106
def _inventory_xml_lines_for_keys(self, keys):
2107
"""Get a line iterator of the sort needed for findind references.
2109
Not relevant for non-xml inventory repositories.
2111
Ghosts in revision_keys are ignored.
2113
:param revision_keys: The revision keys for the inventories to inspect.
2114
:return: An iterator over (inventory line, revid) for the fulltexts of
2115
all of the xml inventories specified by revision_keys.
2117
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2118
for record in stream:
2119
if record.storage_kind != 'absent':
2120
chunks = record.get_bytes_as('chunked')
2121
revid = record.key[-1]
2122
lines = osutils.chunks_to_lines(chunks)
2126
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2128
"""Helper routine for fileids_altered_by_revision_ids.
2130
This performs the translation of xml lines to revision ids.
2132
:param line_iterator: An iterator of lines, origin_version_id
2133
:param revision_keys: The revision ids to filter for. This should be a
2134
set or other type which supports efficient __contains__ lookups, as
2135
the revision key from each parsed line will be looked up in the
2136
revision_keys filter.
2137
:return: a dictionary mapping altered file-ids to an iterable of
2138
revision_ids. Each altered file-ids has the exact revision_ids that
2139
altered it listed explicitly.
2141
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2142
line_iterator).iterkeys())
2143
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2144
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2145
self._inventory_xml_lines_for_keys(parent_keys)))
2146
new_keys = seen - parent_seen
2148
setdefault = result.setdefault
2149
for key in new_keys:
2150
setdefault(key[0], set()).add(key[-1])
2153
def _find_parent_ids_of_revisions(self, revision_ids):
2154
"""Find all parent ids that are mentioned in the revision graph.
2156
:return: set of revisions that are parents of revision_ids which are
2157
not part of revision_ids themselves
2159
parent_map = self.get_parent_map(revision_ids)
2161
map(parent_ids.update, parent_map.itervalues())
2162
parent_ids.difference_update(revision_ids)
2163
parent_ids.discard(_mod_revision.NULL_REVISION)
2166
def _find_parent_keys_of_revisions(self, revision_keys):
2167
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2169
:param revision_keys: An iterable of revision_keys.
2170
:return: The parents of all revision_keys that are not already in
2173
parent_map = self.revisions.get_parent_map(revision_keys)
2175
map(parent_keys.update, parent_map.itervalues())
2176
parent_keys.difference_update(revision_keys)
2177
parent_keys.discard(_mod_revision.NULL_REVISION)
2180
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2181
"""Find the file ids and versions affected by revisions.
2183
:param revisions: an iterable containing revision ids.
2184
:param _inv_weave: The inventory weave from this repository or None.
2185
If None, the inventory weave will be opened automatically.
2186
:return: a dictionary mapping altered file-ids to an iterable of
2187
revision_ids. Each altered file-ids has the exact revision_ids that
2188
altered it listed explicitly.
2190
selected_keys = set((revid,) for revid in revision_ids)
2191
w = _inv_weave or self.inventories
2192
return self._find_file_ids_from_xml_inventory_lines(
2193
w.iter_lines_added_or_present_in_keys(
2194
selected_keys, pb=None),
2197
def iter_files_bytes(self, desired_files):
2198
"""Iterate through file versions.
2200
Files will not necessarily be returned in the order they occur in
2201
desired_files. No specific order is guaranteed.
2203
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2204
value supplied by the caller as part of desired_files. It should
2205
uniquely identify the file version in the caller's context. (Examples:
2206
an index number or a TreeTransform trans_id.)
2208
bytes_iterator is an iterable of bytestrings for the file. The
2209
kind of iterable and length of the bytestrings are unspecified, but for
2210
this implementation, it is a list of bytes produced by
2211
VersionedFile.get_record_stream().
2213
:param desired_files: a list of (file_id, revision_id, identifier)
2217
for file_id, revision_id, callable_data in desired_files:
2218
text_keys[(file_id, revision_id)] = callable_data
2219
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2220
if record.storage_kind == 'absent':
2221
raise errors.RevisionNotPresent(record.key, self)
2222
yield text_keys[record.key], record.get_bytes_as('chunked')
2224
def _generate_text_key_index(self, text_key_references=None,
2226
"""Generate a new text key index for the repository.
2228
This is an expensive function that will take considerable time to run.
2230
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2231
list of parents, also text keys. When a given key has no parents,
2232
the parents list will be [NULL_REVISION].
2234
# All revisions, to find inventory parents.
2235
if ancestors is None:
2236
graph = self.get_graph()
2237
ancestors = graph.get_parent_map(self.all_revision_ids())
2238
if text_key_references is None:
2239
text_key_references = self.find_text_key_references()
2240
pb = ui.ui_factory.nested_progress_bar()
2242
return self._do_generate_text_key_index(ancestors,
2243
text_key_references, pb)
2247
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2248
"""Helper for _generate_text_key_index to avoid deep nesting."""
2249
revision_order = tsort.topo_sort(ancestors)
2250
invalid_keys = set()
2252
for revision_id in revision_order:
2253
revision_keys[revision_id] = set()
2254
text_count = len(text_key_references)
2255
# a cache of the text keys to allow reuse; costs a dict of all the
2256
# keys, but saves a 2-tuple for every child of a given key.
2258
for text_key, valid in text_key_references.iteritems():
2260
invalid_keys.add(text_key)
2262
revision_keys[text_key[1]].add(text_key)
2263
text_key_cache[text_key] = text_key
2264
del text_key_references
2266
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2267
NULL_REVISION = _mod_revision.NULL_REVISION
2268
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2269
# too small for large or very branchy trees. However, for 55K path
2270
# trees, it would be easy to use too much memory trivially. Ideally we
2271
# could gauge this by looking at available real memory etc, but this is
2272
# always a tricky proposition.
2273
inventory_cache = lru_cache.LRUCache(10)
2274
batch_size = 10 # should be ~150MB on a 55K path tree
2275
batch_count = len(revision_order) / batch_size + 1
2277
pb.update("Calculating text parents", processed_texts, text_count)
2278
for offset in xrange(batch_count):
2279
to_query = revision_order[offset * batch_size:(offset + 1) *
2283
for revision_id in to_query:
2284
parent_ids = ancestors[revision_id]
2285
for text_key in revision_keys[revision_id]:
2286
pb.update("Calculating text parents", processed_texts)
2287
processed_texts += 1
2288
candidate_parents = []
2289
for parent_id in parent_ids:
2290
parent_text_key = (text_key[0], parent_id)
2292
check_parent = parent_text_key not in \
2293
revision_keys[parent_id]
2295
# the parent parent_id is a ghost:
2296
check_parent = False
2297
# truncate the derived graph against this ghost.
2298
parent_text_key = None
2300
# look at the parent commit details inventories to
2301
# determine possible candidates in the per file graph.
2304
inv = inventory_cache[parent_id]
2306
inv = self.revision_tree(parent_id).inventory
2307
inventory_cache[parent_id] = inv
2309
parent_entry = inv[text_key[0]]
2310
except (KeyError, errors.NoSuchId):
2312
if parent_entry is not None:
2314
text_key[0], parent_entry.revision)
2316
parent_text_key = None
2317
if parent_text_key is not None:
2318
candidate_parents.append(
2319
text_key_cache[parent_text_key])
2320
parent_heads = text_graph.heads(candidate_parents)
2321
new_parents = list(parent_heads)
2322
new_parents.sort(key=lambda x:candidate_parents.index(x))
2323
if new_parents == []:
2324
new_parents = [NULL_REVISION]
2325
text_index[text_key] = new_parents
2327
for text_key in invalid_keys:
2328
text_index[text_key] = [NULL_REVISION]
2331
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2332
"""Get an iterable listing the keys of all the data introduced by a set
2335
The keys will be ordered so that the corresponding items can be safely
2336
fetched and inserted in that order.
2338
:returns: An iterable producing tuples of (knit-kind, file-id,
2339
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2340
'revisions'. file-id is None unless knit-kind is 'file'.
2342
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2345
for result in self._find_non_file_keys_to_fetch(revision_ids):
2348
def _find_file_keys_to_fetch(self, revision_ids, pb):
2349
# XXX: it's a bit weird to control the inventory weave caching in this
2350
# generator. Ideally the caching would be done in fetch.py I think. Or
2351
# maybe this generator should explicitly have the contract that it
2352
# should not be iterated until the previously yielded item has been
2354
inv_w = self.inventories
2356
# file ids that changed
2357
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2359
num_file_ids = len(file_ids)
2360
for file_id, altered_versions in file_ids.iteritems():
2362
pb.update("Fetch texts", count, num_file_ids)
2364
yield ("file", file_id, altered_versions)
2366
def _find_non_file_keys_to_fetch(self, revision_ids):
2368
yield ("inventory", None, revision_ids)
2371
# XXX: Note ATM no callers actually pay attention to this return
2372
# instead they just use the list of revision ids and ignore
2373
# missing sigs. Consider removing this work entirely
2374
revisions_with_signatures = set(self.signatures.get_parent_map(
2375
[(r,) for r in revision_ids]))
2376
revisions_with_signatures = set(
2377
[r for (r,) in revisions_with_signatures])
2378
revisions_with_signatures.intersection_update(revision_ids)
2379
yield ("signatures", None, revisions_with_signatures)
2382
yield ("revisions", None, revision_ids)
435
2384
@needs_read_lock
436
2385
def get_inventory(self, revision_id):
437
"""Get Inventory object by hash."""
438
return self.deserialise_inventory(
439
revision_id, self.get_inventory_xml(revision_id))
441
def deserialise_inventory(self, revision_id, xml):
442
"""Transform the xml into an inventory object.
2386
"""Get Inventory object by revision id."""
2387
return self.iter_inventories([revision_id]).next()
2389
def iter_inventories(self, revision_ids, ordering=None):
2390
"""Get many inventories by revision_ids.
2392
This will buffer some or all of the texts used in constructing the
2393
inventories in memory, but will only parse a single inventory at a
2396
:param revision_ids: The expected revision ids of the inventories.
2397
:param ordering: optional ordering, e.g. 'topological'. If not
2398
specified, the order of revision_ids will be preserved (by
2399
buffering if necessary).
2400
:return: An iterator of inventories.
2402
if ((None in revision_ids)
2403
or (_mod_revision.NULL_REVISION in revision_ids)):
2404
raise ValueError('cannot get null revision inventory')
2405
return self._iter_inventories(revision_ids, ordering)
2407
def _iter_inventories(self, revision_ids, ordering):
2408
"""single-document based inventory iteration."""
2409
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2410
for text, revision_id in inv_xmls:
2411
yield self._deserialise_inventory(revision_id, text)
2413
def _iter_inventory_xmls(self, revision_ids, ordering):
2414
if ordering is None:
2415
order_as_requested = True
2416
ordering = 'unordered'
2418
order_as_requested = False
2419
keys = [(revision_id,) for revision_id in revision_ids]
2422
if order_as_requested:
2423
key_iter = iter(keys)
2424
next_key = key_iter.next()
2425
stream = self.inventories.get_record_stream(keys, ordering, True)
2427
for record in stream:
2428
if record.storage_kind != 'absent':
2429
chunks = record.get_bytes_as('chunked')
2430
if order_as_requested:
2431
text_chunks[record.key] = chunks
2433
yield ''.join(chunks), record.key[-1]
2435
raise errors.NoSuchRevision(self, record.key)
2436
if order_as_requested:
2437
# Yield as many results as we can while preserving order.
2438
while next_key in text_chunks:
2439
chunks = text_chunks.pop(next_key)
2440
yield ''.join(chunks), next_key[-1]
2442
next_key = key_iter.next()
2443
except StopIteration:
2444
# We still want to fully consume the get_record_stream,
2445
# just in case it is not actually finished at this point
2449
def _deserialise_inventory(self, revision_id, xml):
2450
"""Transform the xml into an inventory object.
444
2452
:param revision_id: The expected revision id of the inventory.
445
2453
:param xml: A serialised inventory.
447
return xml5.serializer_v5.read_inventory_from_string(xml)
2455
result = self._serializer.read_inventory_from_string(xml, revision_id,
2456
entry_cache=self._inventory_entry_cache,
2457
return_from_cache=self._safe_to_return_from_cache)
2458
if result.revision_id != revision_id:
2459
raise AssertionError('revision id mismatch %s != %s' % (
2460
result.revision_id, revision_id))
2463
def get_serializer_format(self):
2464
return self._serializer.format_num
449
2466
@needs_read_lock
450
def get_inventory_xml(self, revision_id):
451
"""Get inventory XML as a file object."""
2467
def _get_inventory_xml(self, revision_id):
2468
"""Get serialized inventory as a string."""
2469
texts = self._iter_inventory_xmls([revision_id], 'unordered')
453
assert isinstance(revision_id, basestring), type(revision_id)
454
iw = self.get_inventory_weave()
455
return iw.get_text(revision_id)
2471
text, revision_id = texts.next()
2472
except StopIteration:
457
2473
raise errors.HistoryMissing(self, 'inventory', revision_id)
460
def get_inventory_sha1(self, revision_id):
461
"""Return the sha1 hash of the inventory entry
463
return self.get_revision(revision_id).inventory_sha1
466
def get_revision_graph(self, revision_id=None):
467
"""Return a dictionary containing the revision graph.
469
:param revision_id: The revision_id to get a graph from. If None, then
470
the entire revision graph is returned. This is a deprecated mode of
471
operation and will be removed in the future.
472
:return: a dictionary of revision_id->revision_parents_list.
474
# special case NULL_REVISION
475
if revision_id == NULL_REVISION:
477
weave = self.get_inventory_weave()
478
all_revisions = self._eliminate_revisions_not_present(weave.versions())
479
entire_graph = dict([(node, weave.get_parents(node)) for
480
node in all_revisions])
481
if revision_id is None:
483
elif revision_id not in entire_graph:
484
raise errors.NoSuchRevision(self, revision_id)
486
# add what can be reached from revision_id
488
pending = set([revision_id])
489
while len(pending) > 0:
491
result[node] = entire_graph[node]
492
for revision_id in result[node]:
493
if revision_id not in result:
494
pending.add(revision_id)
498
def get_revision_graph_with_ghosts(self, revision_ids=None):
499
"""Return a graph of the revisions with ghosts marked as applicable.
501
:param revision_ids: an iterable of revisions to graph or None for all.
502
:return: a Graph object with the graph reachable from revision_ids.
506
pending = set(self.all_revision_ids())
509
pending = set(revision_ids)
510
# special case NULL_REVISION
511
if NULL_REVISION in pending:
512
pending.remove(NULL_REVISION)
513
required = set(pending)
516
revision_id = pending.pop()
518
rev = self.get_revision(revision_id)
519
except errors.NoSuchRevision:
520
if revision_id in required:
523
result.add_ghost(revision_id)
525
for parent_id in rev.parent_ids:
526
# is this queued or done ?
527
if (parent_id not in pending and
528
parent_id not in done):
530
pending.add(parent_id)
531
result.add_node(revision_id, rev.parent_ids)
532
done.add(revision_id)
536
def get_revision_inventory(self, revision_id):
537
"""Return inventory of a past revision."""
538
# TODO: Unify this with get_inventory()
539
# bzr 0.0.6 and later imposes the constraint that the inventory_id
540
# must be the same as its revision, so this is trivial.
541
if revision_id is None:
542
# This does not make sense: if there is no revision,
543
# then it is the current tree inventory surely ?!
544
# and thus get_root_id() is something that looks at the last
545
# commit on the branch, and the get_root_id is an inventory check.
546
raise NotImplementedError
547
# return Inventory(self.get_root_id())
549
return self.get_inventory(revision_id)
2476
def get_rev_id_for_revno(self, revno, known_pair):
2477
"""Return the revision id of a revno, given a later (revno, revid)
2478
pair in the same history.
2480
:return: if found (True, revid). If the available history ran out
2481
before reaching the revno, then this returns
2482
(False, (closest_revno, closest_revid)).
2484
known_revno, known_revid = known_pair
2485
partial_history = [known_revid]
2486
distance_from_known = known_revno - revno
2487
if distance_from_known < 0:
2489
'requested revno (%d) is later than given known revno (%d)'
2490
% (revno, known_revno))
2493
self, partial_history, stop_index=distance_from_known)
2494
except errors.RevisionNotPresent, err:
2495
if err.revision_id == known_revid:
2496
# The start revision (known_revid) wasn't found.
2498
# This is a stacked repository with no fallbacks, or a there's a
2499
# left-hand ghost. Either way, even though the revision named in
2500
# the error isn't in this repo, we know it's the next step in this
2501
# left-hand history.
2502
partial_history.append(err.revision_id)
2503
if len(partial_history) <= distance_from_known:
2504
# Didn't find enough history to get a revid for the revno.
2505
earliest_revno = known_revno - len(partial_history) + 1
2506
return (False, (earliest_revno, partial_history[-1]))
2507
if len(partial_history) - 1 > distance_from_known:
2508
raise AssertionError('_iter_for_revno returned too much history')
2509
return (True, partial_history[-1])
2511
def iter_reverse_revision_history(self, revision_id):
2512
"""Iterate backwards through revision ids in the lefthand history
2514
:param revision_id: The revision id to start with. All its lefthand
2515
ancestors will be traversed.
2517
graph = self.get_graph()
2518
stop_revisions = (None, _mod_revision.NULL_REVISION)
2519
return graph.iter_lefthand_ancestry(revision_id, stop_revisions)
552
2521
def is_shared(self):
553
2522
"""Return True if this repository is flagged as a shared repository."""
554
2523
raise NotImplementedError(self.is_shared)
814
self.control_files._transport.delete('no-working-trees')
2975
self._transport.delete('no-working-trees')
815
2976
except errors.NoSuchFile:
818
self.control_files.put_utf8('no-working-trees', '')
2979
self._transport.put_bytes('no-working-trees', '',
2980
mode=self.bzrdir._get_file_mode())
820
2982
def make_working_trees(self):
821
2983
"""Returns the policy for making working trees on new branches."""
822
return not self.control_files._transport.has('no-working-trees')
825
class KnitRepository(MetaDirRepository):
826
"""Knit format repository."""
828
def _inventory_add_lines(self, inv_vf, revid, parents, lines):
829
inv_vf.add_lines_with_ghosts(revid, parents, lines)
832
def _all_revision_ids(self):
833
"""See Repository.all_revision_ids()."""
834
# Knits get the revision graph from the index of the revision knit, so
835
# it's always possible even if they're on an unlistable transport.
836
return self._revision_store.all_revision_ids(self.get_transaction())
838
def fileid_involved_between_revs(self, from_revid, to_revid):
839
"""Find file_id(s) which are involved in the changes between revisions.
841
This determines the set of revisions which are involved, and then
842
finds all file ids affected by those revisions.
844
vf = self._get_revision_vf()
845
from_set = set(vf.get_ancestry(from_revid))
846
to_set = set(vf.get_ancestry(to_revid))
847
changed = to_set.difference(from_set)
848
return self._fileid_involved_by_set(changed)
850
def fileid_involved(self, last_revid=None):
851
"""Find all file_ids modified in the ancestry of last_revid.
853
:param last_revid: If None, last_revision() will be used.
856
changed = set(self.all_revision_ids())
858
changed = set(self.get_ancestry(last_revid))
861
return self._fileid_involved_by_set(changed)
864
def get_ancestry(self, revision_id):
865
"""Return a list of revision-ids integrated by a revision.
867
This is topologically sorted.
869
if revision_id is None:
871
vf = self._get_revision_vf()
873
return [None] + vf.get_ancestry(revision_id)
874
except errors.RevisionNotPresent:
875
raise errors.NoSuchRevision(self, revision_id)
878
def get_revision(self, revision_id):
879
"""Return the Revision object for a named revision"""
880
return self.get_revision_reconcile(revision_id)
883
def get_revision_graph(self, revision_id=None):
884
"""Return a dictionary containing the revision graph.
886
:param revision_id: The revision_id to get a graph from. If None, then
887
the entire revision graph is returned. This is a deprecated mode of
888
operation and will be removed in the future.
889
:return: a dictionary of revision_id->revision_parents_list.
891
# special case NULL_REVISION
892
if revision_id == NULL_REVISION:
894
weave = self._get_revision_vf()
895
entire_graph = weave.get_graph()
896
if revision_id is None:
897
return weave.get_graph()
898
elif revision_id not in weave:
899
raise errors.NoSuchRevision(self, revision_id)
901
# add what can be reached from revision_id
903
pending = set([revision_id])
904
while len(pending) > 0:
906
result[node] = weave.get_parents(node)
907
for revision_id in result[node]:
908
if revision_id not in result:
909
pending.add(revision_id)
913
def get_revision_graph_with_ghosts(self, revision_ids=None):
914
"""Return a graph of the revisions with ghosts marked as applicable.
916
:param revision_ids: an iterable of revisions to graph or None for all.
917
:return: a Graph object with the graph reachable from revision_ids.
920
vf = self._get_revision_vf()
921
versions = set(vf.versions())
923
pending = set(self.all_revision_ids())
926
pending = set(revision_ids)
927
# special case NULL_REVISION
928
if NULL_REVISION in pending:
929
pending.remove(NULL_REVISION)
930
required = set(pending)
933
revision_id = pending.pop()
934
if not revision_id in versions:
935
if revision_id in required:
936
raise errors.NoSuchRevision(self, revision_id)
938
result.add_ghost(revision_id)
939
# mark it as done so we don't try for it again.
940
done.add(revision_id)
942
parent_ids = vf.get_parents_with_ghosts(revision_id)
943
for parent_id in parent_ids:
944
# is this queued or done ?
945
if (parent_id not in pending and
946
parent_id not in done):
948
pending.add(parent_id)
949
result.add_node(revision_id, parent_ids)
950
done.add(revision_id)
953
def _get_revision_vf(self):
954
""":return: a versioned file containing the revisions."""
955
vf = self._revision_store.get_revision_file(self.get_transaction())
959
def reconcile(self, other=None, thorough=False):
960
"""Reconcile this repository."""
961
from bzrlib.reconcile import KnitReconciler
962
reconciler = KnitReconciler(self, thorough=thorough)
963
reconciler.reconcile()
966
def revision_parents(self, revision_id):
967
return self._get_revision_vf().get_parents(revision_id)
2984
return not self._transport.has('no-working-trees')
2987
class MetaDirVersionedFileRepository(MetaDirRepository):
2988
"""Repositories in a meta-dir, that work via versioned file objects."""
2990
def __init__(self, _format, a_bzrdir, control_files):
2991
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2995
network_format_registry = registry.FormatRegistry()
2996
"""Registry of formats indexed by their network name.
2998
The network name for a repository format is an identifier that can be used when
2999
referring to formats with smart server operations. See
3000
RepositoryFormat.network_name() for more detail.
3004
format_registry = registry.FormatRegistry(network_format_registry)
3005
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
3007
This can contain either format instances themselves, or classes/factories that
3008
can be called to obtain one.
3012
#####################################################################
3013
# Repository Formats
970
3015
class RepositoryFormat(object):
971
3016
"""A repository format.
973
Formats provide three things:
3018
Formats provide four things:
974
3019
* An initialization routine to construct repository data on disk.
975
* a format string which is used when the BzrDir supports versioned
3020
* a optional format string which is used when the BzrDir supports
977
3022
* an open routine which returns a Repository instance.
979
Formats are placed in an dict by their format string for reference
980
during opening. These should be subclasses of RepositoryFormat
3023
* A network name for referring to the format in smart server RPC
3026
There is one and only one Format subclass for each on-disk format. But
3027
there can be one Repository subclass that is used for several different
3028
formats. The _format attribute on a Repository instance can be used to
3029
determine the disk format.
3031
Formats are placed in a registry by their format string for reference
3032
during opening. These should be subclasses of RepositoryFormat for
983
3035
Once a format is deprecated, just deprecate the initialize and open
984
methods on the format class. Do not deprecate the object, as the
985
object will be created every system load.
3036
methods on the format class. Do not deprecate the object, as the
3037
object may be created even when a repository instance hasn't been
987
3040
Common instance attributes:
988
3041
_matchingbzrdir - the bzrdir format that the repository format was
989
3042
originally written to work with. This can be used if manually
990
3043
constructing a bzrdir and repository, or more commonly for test suite
994
_default_format = None
995
"""The default format used for new repositories."""
998
"""The known formats."""
3047
# Set to True or False in derived classes. True indicates that the format
3048
# supports ghosts gracefully.
3049
supports_ghosts = None
3050
# Can this repository be given external locations to lookup additional
3051
# data. Set to True or False in derived classes.
3052
supports_external_lookups = None
3053
# Does this format support CHK bytestring lookups. Set to True or False in
3055
supports_chks = None
3056
# Should commit add an inventory, or an inventory delta to the repository.
3057
_commit_inv_deltas = True
3058
# What order should fetch operations request streams in?
3059
# The default is unordered as that is the cheapest for an origin to
3061
_fetch_order = 'unordered'
3062
# Does this repository format use deltas that can be fetched as-deltas ?
3063
# (E.g. knits, where the knit deltas can be transplanted intact.
3064
# We default to False, which will ensure that enough data to get
3065
# a full text out of any fetch stream will be grabbed.
3066
_fetch_uses_deltas = False
3067
# Should fetch trigger a reconcile after the fetch? Only needed for
3068
# some repository formats that can suffer internal inconsistencies.
3069
_fetch_reconcile = False
3070
# Does this format have < O(tree_size) delta generation. Used to hint what
3071
# code path for commit, amongst other things.
3073
# Does doing a pack operation compress data? Useful for the pack UI command
3074
# (so if there is one pack, the operation can still proceed because it may
3075
# help), and for fetching when data won't have come from the same
3077
pack_compresses = False
3078
# Does the repository inventory storage understand references to trees?
3079
supports_tree_reference = None
3080
# Is the format experimental ?
3081
experimental = False
3084
return "%s()" % self.__class__.__name__
3086
def __eq__(self, other):
3087
# format objects are generally stateless
3088
return isinstance(other, self.__class__)
3090
def __ne__(self, other):
3091
return not self == other
1001
3094
def find_format(klass, a_bzrdir):
1002
"""Return the format for the repository object in a_bzrdir."""
3095
"""Return the format for the repository object in a_bzrdir.
3097
This is used by bzr native formats that have a "format" file in
3098
the repository. Other methods may be used by different types of
1004
3102
transport = a_bzrdir.get_repository_transport(None)
1005
format_string = transport.get("format").read()
1006
return klass._formats[format_string]
3103
format_string = transport.get_bytes("format")
3104
return format_registry.get(format_string)
1007
3105
except errors.NoSuchFile:
1008
3106
raise errors.NoRepositoryPresent(a_bzrdir)
1009
3107
except KeyError:
1010
raise errors.UnknownFormatError(format=format_string)
1012
def _get_control_store(self, repo_transport, control_files):
1013
"""Return the control store for this repository."""
1014
raise NotImplementedError(self._get_control_store)
3108
raise errors.UnknownFormatError(format=format_string,
3112
def register_format(klass, format):
3113
format_registry.register(format.get_format_string(), format)
3116
def unregister_format(klass, format):
3117
format_registry.remove(format.get_format_string())
1017
3120
def get_default_format(klass):
1018
3121
"""Return the current default format."""
1019
return klass._default_format
3122
from bzrlib import bzrdir
3123
return bzrdir.format_registry.make_bzrdir('default').repository_format
1021
3125
def get_format_string(self):
1022
3126
"""Return the ASCII format string that identifies this format.
1024
Note that in pre format ?? repositories the format string is
3128
Note that in pre format ?? repositories the format string is
1025
3129
not permitted nor written to disk.
1027
3131
raise NotImplementedError(self.get_format_string)
1079
3162
:param a_bzrdir: The bzrdir to put the new repository in it.
1080
3163
:param shared: The repository should be initialized as a sharable one.
3164
:returns: The new repository object.
1082
3166
This may raise UninitializableFormat if shared repository are not
1083
3167
compatible the a_bzrdir.
3169
raise NotImplementedError(self.initialize)
1086
3171
def is_supported(self):
1087
3172
"""Is this format supported?
1089
3174
Supported formats must be initializable and openable.
1090
Unsupported formats may not support initialization or committing or
3175
Unsupported formats may not support initialization or committing or
1091
3176
some other features depending on the reason for not being supported.
3180
def network_name(self):
3181
"""A simple byte string uniquely identifying this format for RPC calls.
3183
MetaDir repository formats use their disk format string to identify the
3184
repository over the wire. All in one formats such as bzr < 0.8, and
3185
foreign formats like svn/git and hg should use some marker which is
3186
unique and immutable.
3188
raise NotImplementedError(self.network_name)
3190
def check_conversion_target(self, target_format):
3191
if self.rich_root_data and not target_format.rich_root_data:
3192
raise errors.BadConversionTarget(
3193
'Does not support rich root data.', target_format,
3195
if (self.supports_tree_reference and
3196
not getattr(target_format, 'supports_tree_reference', False)):
3197
raise errors.BadConversionTarget(
3198
'Does not support nested trees', target_format,
1095
3201
def open(self, a_bzrdir, _found=False):
1096
3202
"""Return an instance of this format for the bzrdir a_bzrdir.
1098
3204
_found is a private parameter, do not use it.
1100
3206
raise NotImplementedError(self.open)
1103
def register_format(klass, format):
1104
klass._formats[format.get_format_string()] = format
1107
def set_default_format(klass, format):
1108
klass._default_format = format
1111
def unregister_format(klass, format):
1112
assert klass._formats[format.get_format_string()] is format
1113
del klass._formats[format.get_format_string()]
1116
class PreSplitOutRepositoryFormat(RepositoryFormat):
1117
"""Base class for the pre split out repository formats."""
1119
def initialize(self, a_bzrdir, shared=False, _internal=False):
1120
"""Create a weave repository.
1122
TODO: when creating split out bzr branch formats, move this to a common
1123
base for Format5, Format6. or something like that.
1125
from bzrlib.weavefile import write_weave_v5
1126
from bzrlib.weave import Weave
1129
raise errors.IncompatibleFormat(self, a_bzrdir._format)
1132
# always initialized when the bzrdir is.
1133
return self.open(a_bzrdir, _found=True)
1135
# Create an empty weave
1137
write_weave_v5(Weave(), sio)
1138
empty_weave = sio.getvalue()
1140
mutter('creating repository in %s.', a_bzrdir.transport.base)
1141
dirs = ['revision-store', 'weaves']
1142
files = [('inventory.weave', StringIO(empty_weave)),
1145
# FIXME: RBC 20060125 don't peek under the covers
1146
# NB: no need to escape relative paths that are url safe.
1147
control_files = LockableFiles(a_bzrdir.transport, 'branch-lock',
1149
control_files.create_lock()
1150
control_files.lock_write()
1151
control_files._transport.mkdir_multi(dirs,
1152
mode=control_files._dir_mode)
1154
for file, content in files:
1155
control_files.put(file, content)
1157
control_files.unlock()
1158
return self.open(a_bzrdir, _found=True)
1160
def _get_control_store(self, repo_transport, control_files):
1161
"""Return the control store for this repository."""
1162
return self._get_versioned_file_store('',
1167
def _get_text_store(self, transport, control_files):
1168
"""Get a store for file texts for this format."""
1169
raise NotImplementedError(self._get_text_store)
1171
def open(self, a_bzrdir, _found=False):
1172
"""See RepositoryFormat.open()."""
1174
# we are being called directly and must probe.
1175
raise NotImplementedError
1177
repo_transport = a_bzrdir.get_repository_transport(None)
1178
control_files = a_bzrdir._control_files
1179
text_store = self._get_text_store(repo_transport, control_files)
1180
control_store = self._get_control_store(repo_transport, control_files)
1181
_revision_store = self._get_revision_store(repo_transport, control_files)
1182
return AllInOneRepository(_format=self,
1184
_revision_store=_revision_store,
1185
control_store=control_store,
1186
text_store=text_store)
1189
class RepositoryFormat4(PreSplitOutRepositoryFormat):
1190
"""Bzr repository format 4.
1192
This repository format has:
1194
- TextStores for texts, inventories,revisions.
1196
This format is deprecated: it indexes texts using a text id which is
1197
removed in format 5; initialization and write support for this format
1202
super(RepositoryFormat4, self).__init__()
1203
self._matchingbzrdir = bzrdir.BzrDirFormat4()
1205
def get_format_description(self):
1206
"""See RepositoryFormat.get_format_description()."""
1207
return "Repository format 4"
1209
def initialize(self, url, shared=False, _internal=False):
1210
"""Format 4 branches cannot be created."""
1211
raise errors.UninitializableFormat(self)
1213
def is_supported(self):
1214
"""Format 4 is not supported.
1216
It is not supported because the model changed from 4 to 5 and the
1217
conversion logic is expensive - so doing it on the fly was not
1222
def _get_control_store(self, repo_transport, control_files):
1223
"""Format 4 repositories have no formal control store at this point.
1225
This will cause any control-file-needing apis to fail - this is desired.
1229
def _get_revision_store(self, repo_transport, control_files):
1230
"""See RepositoryFormat._get_revision_store()."""
1231
from bzrlib.xml4 import serializer_v4
1232
return self._get_text_rev_store(repo_transport,
1235
serializer=serializer_v4)
1237
def _get_text_store(self, transport, control_files):
1238
"""See RepositoryFormat._get_text_store()."""
1241
class RepositoryFormat5(PreSplitOutRepositoryFormat):
1242
"""Bzr control format 5.
1244
This repository format has:
1245
- weaves for file texts and inventory
1247
- TextStores for revisions and signatures.
1251
super(RepositoryFormat5, self).__init__()
1252
self._matchingbzrdir = bzrdir.BzrDirFormat5()
1254
def get_format_description(self):
1255
"""See RepositoryFormat.get_format_description()."""
1256
return "Weave repository format 5"
1258
def _get_revision_store(self, repo_transport, control_files):
1259
"""See RepositoryFormat._get_revision_store()."""
1260
"""Return the revision store object for this a_bzrdir."""
1261
return self._get_text_rev_store(repo_transport,
1266
def _get_text_store(self, transport, control_files):
1267
"""See RepositoryFormat._get_text_store()."""
1268
return self._get_versioned_file_store('weaves', transport, control_files, prefixed=False)
1271
class RepositoryFormat6(PreSplitOutRepositoryFormat):
1272
"""Bzr control format 6.
1274
This repository format has:
1275
- weaves for file texts and inventory
1276
- hash subdirectory based stores.
1277
- TextStores for revisions and signatures.
1281
super(RepositoryFormat6, self).__init__()
1282
self._matchingbzrdir = bzrdir.BzrDirFormat6()
1284
def get_format_description(self):
1285
"""See RepositoryFormat.get_format_description()."""
1286
return "Weave repository format 6"
1288
def _get_revision_store(self, repo_transport, control_files):
1289
"""See RepositoryFormat._get_revision_store()."""
1290
return self._get_text_rev_store(repo_transport,
1296
def _get_text_store(self, transport, control_files):
1297
"""See RepositoryFormat._get_text_store()."""
1298
return self._get_versioned_file_store('weaves', transport, control_files)
3208
def _run_post_repo_init_hooks(self, repository, a_bzrdir, shared):
3209
from bzrlib.bzrdir import BzrDir, RepoInitHookParams
3210
hooks = BzrDir.hooks['post_repo_init']
3213
params = RepoInitHookParams(repository, self, a_bzrdir, shared)
1301
3218
class MetaDirRepositoryFormat(RepositoryFormat):
1302
3219
"""Common base class for the new repositories using the metadir layout."""
3221
rich_root_data = False
3222
supports_tree_reference = False
3223
supports_external_lookups = False
3226
def _matchingbzrdir(self):
3227
matching = bzrdir.BzrDirMetaFormat1()
3228
matching.repository_format = self
1304
3231
def __init__(self):
1305
3232
super(MetaDirRepositoryFormat, self).__init__()
1306
self._matchingbzrdir = bzrdir.BzrDirMetaFormat1()
1308
3234
def _create_control_files(self, a_bzrdir):
1309
3235
"""Create the required files and the initial control_files object."""
1310
3236
# FIXME: RBC 20060125 don't peek under the covers
1311
3237
# NB: no need to escape relative paths that are url safe.
1312
3238
repository_transport = a_bzrdir.get_repository_transport(self)
1313
control_files = LockableFiles(repository_transport, 'lock', LockDir)
3239
control_files = lockable_files.LockableFiles(repository_transport,
3240
'lock', lockdir.LockDir)
1314
3241
control_files.create_lock()
1315
3242
return control_files
1318
3245
"""Upload the initial blank content."""
1319
3246
control_files = self._create_control_files(a_bzrdir)
1320
3247
control_files.lock_write()
3248
transport = control_files._transport
3250
utf8_files += [('shared-storage', '')]
1322
control_files._transport.mkdir_multi(dirs,
1323
mode=control_files._dir_mode)
1324
for file, content in files:
1325
control_files.put(file, content)
1326
for file, content in utf8_files:
1327
control_files.put_utf8(file, content)
1329
control_files.put_utf8('shared-storage', '')
3252
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3253
for (filename, content_stream) in files:
3254
transport.put_file(filename, content_stream,
3255
mode=a_bzrdir._get_file_mode())
3256
for (filename, content_bytes) in utf8_files:
3257
transport.put_bytes_non_atomic(filename, content_bytes,
3258
mode=a_bzrdir._get_file_mode())
1331
3260
control_files.unlock()
1334
class RepositoryFormat7(MetaDirRepositoryFormat):
1335
"""Bzr repository 7.
1337
This repository format has:
1338
- weaves for file texts and inventory
1339
- hash subdirectory based stores.
1340
- TextStores for revisions and signatures.
1341
- a format marker of its own
1342
- an optional 'shared-storage' flag
1343
- an optional 'no-working-trees' flag
1346
def _get_control_store(self, repo_transport, control_files):
1347
"""Return the control store for this repository."""
1348
return self._get_versioned_file_store('',
1353
def get_format_string(self):
1354
"""See RepositoryFormat.get_format_string()."""
1355
return "Bazaar-NG Repository format 7"
1357
def get_format_description(self):
1358
"""See RepositoryFormat.get_format_description()."""
1359
return "Weave repository format 7"
1361
def _get_revision_store(self, repo_transport, control_files):
1362
"""See RepositoryFormat._get_revision_store()."""
1363
return self._get_text_rev_store(repo_transport,
1370
def _get_text_store(self, transport, control_files):
1371
"""See RepositoryFormat._get_text_store()."""
1372
return self._get_versioned_file_store('weaves',
1376
def initialize(self, a_bzrdir, shared=False):
1377
"""Create a weave repository.
1379
:param shared: If true the repository will be initialized as a shared
1382
from bzrlib.weavefile import write_weave_v5
1383
from bzrlib.weave import Weave
1385
# Create an empty weave
1387
write_weave_v5(Weave(), sio)
1388
empty_weave = sio.getvalue()
1390
mutter('creating repository in %s.', a_bzrdir.transport.base)
1391
dirs = ['revision-store', 'weaves']
1392
files = [('inventory.weave', StringIO(empty_weave)),
1394
utf8_files = [('format', self.get_format_string())]
1396
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1397
return self.open(a_bzrdir=a_bzrdir, _found=True)
1399
def open(self, a_bzrdir, _found=False, _override_transport=None):
1400
"""See RepositoryFormat.open().
1402
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1403
repository at a slightly different url
1404
than normal. I.e. during 'upgrade'.
1407
format = RepositoryFormat.find_format(a_bzrdir)
1408
assert format.__class__ == self.__class__
1409
if _override_transport is not None:
1410
repo_transport = _override_transport
1412
repo_transport = a_bzrdir.get_repository_transport(None)
1413
control_files = LockableFiles(repo_transport, 'lock', LockDir)
1414
text_store = self._get_text_store(repo_transport, control_files)
1415
control_store = self._get_control_store(repo_transport, control_files)
1416
_revision_store = self._get_revision_store(repo_transport, control_files)
1417
return MetaDirRepository(_format=self,
1419
control_files=control_files,
1420
_revision_store=_revision_store,
1421
control_store=control_store,
1422
text_store=text_store)
1425
class RepositoryFormatKnit1(MetaDirRepositoryFormat):
1426
"""Bzr repository knit format 1.
1428
This repository format has:
1429
- knits for file texts and inventory
1430
- hash subdirectory based stores.
1431
- knits for revisions and signatures
1432
- TextStores for revisions and signatures.
1433
- a format marker of its own
1434
- an optional 'shared-storage' flag
1435
- an optional 'no-working-trees' flag
1438
This format was introduced in bzr 0.8.
1441
def _get_control_store(self, repo_transport, control_files):
1442
"""Return the control store for this repository."""
1443
return VersionedFileStore(
1446
file_mode=control_files._file_mode,
1447
versionedfile_class=KnitVersionedFile,
1448
versionedfile_kwargs={'factory':KnitPlainFactory()},
1451
def get_format_string(self):
1452
"""See RepositoryFormat.get_format_string()."""
1453
return "Bazaar-NG Knit Repository Format 1"
1455
def get_format_description(self):
1456
"""See RepositoryFormat.get_format_description()."""
1457
return "Knit repository format 1"
1459
def _get_revision_store(self, repo_transport, control_files):
1460
"""See RepositoryFormat._get_revision_store()."""
1461
from bzrlib.store.revision.knit import KnitRevisionStore
1462
versioned_file_store = VersionedFileStore(
1464
file_mode=control_files._file_mode,
1467
versionedfile_class=KnitVersionedFile,
1468
versionedfile_kwargs={'delta':False, 'factory':KnitPlainFactory()},
1471
return KnitRevisionStore(versioned_file_store)
1473
def _get_text_store(self, transport, control_files):
1474
"""See RepositoryFormat._get_text_store()."""
1475
return self._get_versioned_file_store('knits',
1478
versionedfile_class=KnitVersionedFile,
1481
def initialize(self, a_bzrdir, shared=False):
1482
"""Create a knit format 1 repository.
1484
:param a_bzrdir: bzrdir to contain the new repository; must already
1486
:param shared: If true the repository will be initialized as a shared
1489
mutter('creating repository in %s.', a_bzrdir.transport.base)
1490
dirs = ['revision-store', 'knits']
1492
utf8_files = [('format', self.get_format_string())]
1494
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1495
repo_transport = a_bzrdir.get_repository_transport(None)
1496
control_files = LockableFiles(repo_transport, 'lock', LockDir)
1497
control_store = self._get_control_store(repo_transport, control_files)
1498
transaction = transactions.WriteTransaction()
1499
# trigger a write of the inventory store.
1500
control_store.get_weave_or_empty('inventory', transaction)
1501
_revision_store = self._get_revision_store(repo_transport, control_files)
1502
_revision_store.has_revision_id('A', transaction)
1503
_revision_store.get_signature_file(transaction)
1504
return self.open(a_bzrdir=a_bzrdir, _found=True)
1506
def open(self, a_bzrdir, _found=False, _override_transport=None):
1507
"""See RepositoryFormat.open().
1509
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1510
repository at a slightly different url
1511
than normal. I.e. during 'upgrade'.
1514
format = RepositoryFormat.find_format(a_bzrdir)
1515
assert format.__class__ == self.__class__
1516
if _override_transport is not None:
1517
repo_transport = _override_transport
1519
repo_transport = a_bzrdir.get_repository_transport(None)
1520
control_files = LockableFiles(repo_transport, 'lock', LockDir)
1521
text_store = self._get_text_store(repo_transport, control_files)
1522
control_store = self._get_control_store(repo_transport, control_files)
1523
_revision_store = self._get_revision_store(repo_transport, control_files)
1524
return KnitRepository(_format=self,
1526
control_files=control_files,
1527
_revision_store=_revision_store,
1528
control_store=control_store,
1529
text_store=text_store)
1532
# formats which have no format string are not discoverable
1533
# and not independently creatable, so are not registered.
1534
RepositoryFormat.register_format(RepositoryFormat7())
1535
_default_format = RepositoryFormatKnit1()
1536
RepositoryFormat.register_format(_default_format)
1537
RepositoryFormat.set_default_format(_default_format)
1538
_legacy_formats = [RepositoryFormat4(),
1539
RepositoryFormat5(),
1540
RepositoryFormat6()]
3262
def network_name(self):
3263
"""Metadir formats have matching disk and network format strings."""
3264
return self.get_format_string()
3267
# Pre-0.8 formats that don't have a disk format string (because they are
3268
# versioned by the matching control directory). We use the control directories
3269
# disk format string as a key for the network_name because they meet the
3270
# constraints (simple string, unique, immutable).
3271
network_format_registry.register_lazy(
3272
"Bazaar-NG branch, format 5\n",
3273
'bzrlib.repofmt.weaverepo',
3274
'RepositoryFormat5',
3276
network_format_registry.register_lazy(
3277
"Bazaar-NG branch, format 6\n",
3278
'bzrlib.repofmt.weaverepo',
3279
'RepositoryFormat6',
3282
# formats which have no format string are not discoverable or independently
3283
# creatable on disk, so are not registered in format_registry. They're
3284
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3285
# needed, it's constructed directly by the BzrDir. Non-native formats where
3286
# the repository is not separately opened are similar.
3288
format_registry.register_lazy(
3289
'Bazaar-NG Repository format 7',
3290
'bzrlib.repofmt.weaverepo',
3294
format_registry.register_lazy(
3295
'Bazaar-NG Knit Repository Format 1',
3296
'bzrlib.repofmt.knitrepo',
3297
'RepositoryFormatKnit1',
3300
format_registry.register_lazy(
3301
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3302
'bzrlib.repofmt.knitrepo',
3303
'RepositoryFormatKnit3',
3306
format_registry.register_lazy(
3307
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3308
'bzrlib.repofmt.knitrepo',
3309
'RepositoryFormatKnit4',
3312
# Pack-based formats. There is one format for pre-subtrees, and one for
3313
# post-subtrees to allow ease of testing.
3314
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3315
format_registry.register_lazy(
3316
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3317
'bzrlib.repofmt.pack_repo',
3318
'RepositoryFormatKnitPack1',
3320
format_registry.register_lazy(
3321
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3322
'bzrlib.repofmt.pack_repo',
3323
'RepositoryFormatKnitPack3',
3325
format_registry.register_lazy(
3326
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3327
'bzrlib.repofmt.pack_repo',
3328
'RepositoryFormatKnitPack4',
3330
format_registry.register_lazy(
3331
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3332
'bzrlib.repofmt.pack_repo',
3333
'RepositoryFormatKnitPack5',
3335
format_registry.register_lazy(
3336
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3337
'bzrlib.repofmt.pack_repo',
3338
'RepositoryFormatKnitPack5RichRoot',
3340
format_registry.register_lazy(
3341
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3342
'bzrlib.repofmt.pack_repo',
3343
'RepositoryFormatKnitPack5RichRootBroken',
3345
format_registry.register_lazy(
3346
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3347
'bzrlib.repofmt.pack_repo',
3348
'RepositoryFormatKnitPack6',
3350
format_registry.register_lazy(
3351
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3352
'bzrlib.repofmt.pack_repo',
3353
'RepositoryFormatKnitPack6RichRoot',
3356
# Development formats.
3357
# Obsolete but kept pending a CHK based subtree format.
3358
format_registry.register_lazy(
3359
("Bazaar development format 2 with subtree support "
3360
"(needs bzr.dev from before 1.8)\n"),
3361
'bzrlib.repofmt.pack_repo',
3362
'RepositoryFormatPackDevelopment2Subtree',
3365
# 1.14->1.16 go below here
3366
format_registry.register_lazy(
3367
'Bazaar development format - group compression and chk inventory'
3368
' (needs bzr.dev from 1.14)\n',
3369
'bzrlib.repofmt.groupcompress_repo',
3370
'RepositoryFormatCHK1',
3373
format_registry.register_lazy(
3374
'Bazaar development format - chk repository with bencode revision '
3375
'serialization (needs bzr.dev from 1.16)\n',
3376
'bzrlib.repofmt.groupcompress_repo',
3377
'RepositoryFormatCHK2',
3379
format_registry.register_lazy(
3380
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3381
'bzrlib.repofmt.groupcompress_repo',
3382
'RepositoryFormat2a',
3384
format_registry.register_lazy(
3385
'Bazaar development format 8\n',
3386
'bzrlib.repofmt.groupcompress_repo',
3387
'RepositoryFormat2aSubtree',
1543
3391
class InterRepository(InterObject):
1544
3392
"""This class represents operations taking place between two repositories.
1546
3394
Its instances have methods like copy_content and fetch, and contain
1547
references to the source and target repositories these operations can be
3395
references to the source and target repositories these operations can be
1548
3396
carried out on.
1550
3398
Often we will provide convenience methods on 'repository' which carry out
1552
3400
InterRepository.get(other).method_name(parameters).
3403
_walk_to_common_revisions_batch_size = 50
1556
3405
"""The available optimised InterRepository types."""
1558
3407
@needs_write_lock
1559
def copy_content(self, revision_id=None, basis=None):
3408
def copy_content(self, revision_id=None):
1560
3409
"""Make a complete copy of the content in self into destination.
1562
This is a destructive operation! Do not use it on existing
3411
This is a destructive operation! Do not use it on existing
1565
3414
:param revision_id: Only copy the content needed to construct
1566
3415
revision_id and its parents.
1567
:param basis: Copy the needed data preferentially from basis.
1570
3418
self.target.set_make_working_trees(self.source.make_working_trees())
1571
3419
except NotImplementedError:
1573
# grab the basis available data
1574
if basis is not None:
1575
self.target.fetch(basis, revision_id=revision_id)
1576
# but don't bother fetching if we have the needed data now.
1577
if (revision_id not in (None, NULL_REVISION) and
1578
self.target.has_revision(revision_id)):
1580
3421
self.target.fetch(self.source, revision_id=revision_id)
1582
def _double_lock(self, lock_source, lock_target):
1583
"""Take out too locks, rolling back the first if the second throws."""
1588
# we want to ensure that we don't leave source locked by mistake.
1589
# and any error on target should not confuse source.
1590
self.source.unlock()
1593
3423
@needs_write_lock
1594
def fetch(self, revision_id=None, pb=None):
3424
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
1595
3426
"""Fetch the content required to construct revision_id.
1597
The content is copied from source to target.
3428
The content is copied from self.source to self.target.
1599
3430
:param revision_id: if None all content is copied, if NULL_REVISION no
1600
3431
content is copied.
1601
:param pb: optional progress bar to use for progress reports. If not
1602
provided a default one will be created.
1604
Returns the copied revision count and the failed revisions in a tuple:
1607
from bzrlib.fetch import GenericRepoFetcher
1608
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1609
self.source, self.source._format, self.target, self.target._format)
1610
f = GenericRepoFetcher(to_repository=self.target,
3435
ui.ui_factory.warn_experimental_format_fetch(self)
3436
from bzrlib.fetch import RepoFetcher
3437
# See <https://launchpad.net/bugs/456077> asking for a warning here
3438
if self.source._format.network_name() != self.target._format.network_name():
3439
ui.ui_factory.show_user_warning('cross_format_fetch',
3440
from_format=self.source._format,
3441
to_format=self.target._format)
3442
f = RepoFetcher(to_repository=self.target,
1611
3443
from_repository=self.source,
1612
3444
last_revision=revision_id,
1614
return f.count_copied, f.failed_revisions
1616
def lock_read(self):
1617
"""Take out a logical read lock.
1619
This will lock the source branch and the target branch. The source gets
1620
a read lock and the target a read lock.
1622
self._double_lock(self.source.lock_read, self.target.lock_read)
1624
def lock_write(self):
1625
"""Take out a logical write lock.
1627
This will lock the source branch and the target branch. The source gets
1628
a read lock and the target a write lock.
1630
self._double_lock(self.source.lock_read, self.target.lock_write)
3445
fetch_spec=fetch_spec,
3446
find_ghosts=find_ghosts)
3448
def _walk_to_common_revisions(self, revision_ids):
3449
"""Walk out from revision_ids in source to revisions target has.
3451
:param revision_ids: The start point for the search.
3452
:return: A set of revision ids.
3454
target_graph = self.target.get_graph()
3455
revision_ids = frozenset(revision_ids)
3456
missing_revs = set()
3457
source_graph = self.source.get_graph()
3458
# ensure we don't pay silly lookup costs.
3459
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3460
null_set = frozenset([_mod_revision.NULL_REVISION])
3461
searcher_exhausted = False
3465
# Iterate the searcher until we have enough next_revs
3466
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3468
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3469
next_revs.update(next_revs_part)
3470
ghosts.update(ghosts_part)
3471
except StopIteration:
3472
searcher_exhausted = True
3474
# If there are ghosts in the source graph, and the caller asked for
3475
# them, make sure that they are present in the target.
3476
# We don't care about other ghosts as we can't fetch them and
3477
# haven't been asked to.
3478
ghosts_to_check = set(revision_ids.intersection(ghosts))
3479
revs_to_get = set(next_revs).union(ghosts_to_check)
3481
have_revs = set(target_graph.get_parent_map(revs_to_get))
3482
# we always have NULL_REVISION present.
3483
have_revs = have_revs.union(null_set)
3484
# Check if the target is missing any ghosts we need.
3485
ghosts_to_check.difference_update(have_revs)
3487
# One of the caller's revision_ids is a ghost in both the
3488
# source and the target.
3489
raise errors.NoSuchRevision(
3490
self.source, ghosts_to_check.pop())
3491
missing_revs.update(next_revs - have_revs)
3492
# Because we may have walked past the original stop point, make
3493
# sure everything is stopped
3494
stop_revs = searcher.find_seen_ancestors(have_revs)
3495
searcher.stop_searching_any(stop_revs)
3496
if searcher_exhausted:
3498
return searcher.get_result()
1632
3500
@needs_read_lock
1633
def missing_revision_ids(self, revision_id=None):
3501
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1634
3502
"""Return the revision ids that source has that target does not.
1636
These are returned in topological order.
1638
3504
:param revision_id: only return revision ids included by this
3506
:param find_ghosts: If True find missing revisions in deep history
3507
rather than just finding the surface difference.
3508
:return: A bzrlib.graph.SearchResult.
3510
# stop searching at found target revisions.
3511
if not find_ghosts and revision_id is not None:
3512
return self._walk_to_common_revisions([revision_id])
1641
3513
# generic, possibly worst case, slow code path.
1642
3514
target_ids = set(self.target.all_revision_ids())
1643
3515
if revision_id is not None:
1644
3516
source_ids = self.source.get_ancestry(revision_id)
1645
assert source_ids[0] == None
3517
if source_ids[0] is not None:
3518
raise AssertionError()
1646
3519
source_ids.pop(0)
1648
3521
source_ids = self.source.all_revision_ids()
1649
3522
result_set = set(source_ids).difference(target_ids)
1650
# this may look like a no-op: its not. It preserves the ordering
1651
# other_ids had while only returning the members from other_ids
1652
# that we've decided we need.
1653
return [rev_id for rev_id in source_ids if rev_id in result_set]
1656
"""Release the locks on source and target."""
3523
return self.source.revision_ids_to_search_result(result_set)
3526
def _same_model(source, target):
3527
"""True if source and target have the same data representation.
3529
Note: this is always called on the base class; overriding it in a
3530
subclass will have no effect.
1658
self.target.unlock()
1660
self.source.unlock()
1663
class InterWeaveRepo(InterRepository):
1664
"""Optimised code paths between Weave based repositories."""
1666
_matching_repo_format = RepositoryFormat7()
1667
"""Repository format for testing with."""
3533
InterRepository._assert_same_model(source, target)
3535
except errors.IncompatibleRepositories, e:
3539
def _assert_same_model(source, target):
3540
"""Raise an exception if two repositories do not use the same model.
3542
if source.supports_rich_root() != target.supports_rich_root():
3543
raise errors.IncompatibleRepositories(source, target,
3544
"different rich-root support")
3545
if source._serializer != target._serializer:
3546
raise errors.IncompatibleRepositories(source, target,
3547
"different serializers")
3550
class InterSameDataRepository(InterRepository):
3551
"""Code for converting between repositories that represent the same data.
3553
Data format and model must match for this to work.
3557
def _get_repo_format_to_test(self):
3558
"""Repository format for testing with.
3560
InterSameData can pull from subtree to subtree and from non-subtree to
3561
non-subtree, so we test this with the richest repository format.
3563
from bzrlib.repofmt import knitrepo
3564
return knitrepo.RepositoryFormatKnit3()
3567
def is_compatible(source, target):
3568
return InterRepository._same_model(source, target)
3571
class InterWeaveRepo(InterSameDataRepository):
3572
"""Optimised code paths between Weave based repositories.
3574
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3575
implemented lazy inter-object optimisation.
3579
def _get_repo_format_to_test(self):
3580
from bzrlib.repofmt import weaverepo
3581
return weaverepo.RepositoryFormat7()
1670
3584
def is_compatible(source, target):
1671
3585
"""Be compatible with known Weave formats.
1673
3587
We don't test for the stores being of specific types because that
1674
could lead to confusing results, and there is no need to be
3588
could lead to confusing results, and there is no need to be
1675
3589
overly general.
3591
from bzrlib.repofmt.weaverepo import (
1678
3597
return (isinstance(source._format, (RepositoryFormat5,
1679
3598
RepositoryFormat6,
1763
3663
# we do not have a revision as that would be pointless.
1764
3664
target_ids = set(self.target._all_possible_ids())
1765
3665
possibly_present_revisions = target_ids.intersection(source_ids_set)
1766
actually_present_revisions = set(self.target._eliminate_revisions_not_present(possibly_present_revisions))
3666
actually_present_revisions = set(
3667
self.target._eliminate_revisions_not_present(possibly_present_revisions))
1767
3668
required_revisions = source_ids_set.difference(actually_present_revisions)
1768
required_topo_revisions = [rev_id for rev_id in source_ids if rev_id in required_revisions]
1769
3669
if revision_id is not None:
1770
3670
# we used get_ancestry to determine source_ids then we are assured all
1771
3671
# revisions referenced are present as they are installed in topological order.
1772
3672
# and the tip revision was validated by get_ancestry.
1773
return required_topo_revisions
3673
result_set = required_revisions
1775
# if we just grabbed the possibly available ids, then
3675
# if we just grabbed the possibly available ids, then
1776
3676
# we only have an estimate of whats available and need to validate
1777
3677
# that against the revision records.
1778
return self.source._eliminate_revisions_not_present(required_topo_revisions)
1781
class InterKnitRepo(InterRepository):
3679
self.source._eliminate_revisions_not_present(required_revisions))
3680
return self.source.revision_ids_to_search_result(result_set)
3683
class InterKnitRepo(InterSameDataRepository):
1782
3684
"""Optimised code paths between Knit based repositories."""
1784
_matching_repo_format = RepositoryFormatKnit1()
1785
"""Repository format for testing with."""
3687
def _get_repo_format_to_test(self):
3688
from bzrlib.repofmt import knitrepo
3689
return knitrepo.RepositoryFormatKnit1()
1788
3692
def is_compatible(source, target):
1789
3693
"""Be compatible with known Knit formats.
1791
3695
We don't test for the stores being of specific types because that
1792
could lead to confusing results, and there is no need to be
3696
could lead to confusing results, and there is no need to be
1793
3697
overly general.
3699
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
1796
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1797
isinstance(target._format, (RepositoryFormatKnit1)))
3701
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3702
isinstance(target._format, RepositoryFormatKnit))
1798
3703
except AttributeError:
1802
def fetch(self, revision_id=None, pb=None):
1803
"""See InterRepository.fetch()."""
1804
from bzrlib.fetch import KnitRepoFetcher
1805
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1806
self.source, self.source._format, self.target, self.target._format)
1807
f = KnitRepoFetcher(to_repository=self.target,
1808
from_repository=self.source,
1809
last_revision=revision_id,
1811
return f.count_copied, f.failed_revisions
3705
return are_knits and InterRepository._same_model(source, target)
1813
3707
@needs_read_lock
1814
def missing_revision_ids(self, revision_id=None):
3708
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1815
3709
"""See InterRepository.missing_revision_ids()."""
1816
3710
if revision_id is not None:
1817
3711
source_ids = self.source.get_ancestry(revision_id)
1818
assert source_ids[0] == None
3712
if source_ids[0] is not None:
3713
raise AssertionError()
1819
3714
source_ids.pop(0)
1821
source_ids = self.source._all_possible_ids()
3716
source_ids = self.source.all_revision_ids()
1822
3717
source_ids_set = set(source_ids)
1823
3718
# source_ids is the worst possible case we may need to pull.
1824
3719
# now we want to filter source_ids against what we actually
1825
3720
# have in target, but don't try to check for existence where we know
1826
3721
# we do not have a revision as that would be pointless.
1827
target_ids = set(self.target._all_possible_ids())
3722
target_ids = set(self.target.all_revision_ids())
1828
3723
possibly_present_revisions = target_ids.intersection(source_ids_set)
1829
actually_present_revisions = set(self.target._eliminate_revisions_not_present(possibly_present_revisions))
3724
actually_present_revisions = set(
3725
self.target._eliminate_revisions_not_present(possibly_present_revisions))
1830
3726
required_revisions = source_ids_set.difference(actually_present_revisions)
1831
required_topo_revisions = [rev_id for rev_id in source_ids if rev_id in required_revisions]
1832
3727
if revision_id is not None:
1833
3728
# we used get_ancestry to determine source_ids then we are assured all
1834
3729
# revisions referenced are present as they are installed in topological order.
1835
3730
# and the tip revision was validated by get_ancestry.
1836
return required_topo_revisions
3731
result_set = required_revisions
1838
# if we just grabbed the possibly available ids, then
3733
# if we just grabbed the possibly available ids, then
1839
3734
# we only have an estimate of whats available and need to validate
1840
3735
# that against the revision records.
1841
return self.source._eliminate_revisions_not_present(required_topo_revisions)
3737
self.source._eliminate_revisions_not_present(required_revisions))
3738
return self.source.revision_ids_to_search_result(result_set)
3741
class InterDifferingSerializer(InterRepository):
3744
def _get_repo_format_to_test(self):
3748
def is_compatible(source, target):
3749
"""Be compatible with Knit2 source and Knit3 target"""
3750
# This is redundant with format.check_conversion_target(), however that
3751
# raises an exception, and we just want to say "False" as in we won't
3752
# support converting between these formats.
3753
if 'IDS_never' in debug.debug_flags:
3755
if source.supports_rich_root() and not target.supports_rich_root():
3757
if (source._format.supports_tree_reference
3758
and not target._format.supports_tree_reference):
3760
if target._fallback_repositories and target._format.supports_chks:
3761
# IDS doesn't know how to copy CHKs for the parent inventories it
3762
# adds to stacked repos.
3764
if 'IDS_always' in debug.debug_flags:
3766
# Only use this code path for local source and target. IDS does far
3767
# too much IO (both bandwidth and roundtrips) over a network.
3768
if not source.bzrdir.transport.base.startswith('file:///'):
3770
if not target.bzrdir.transport.base.startswith('file:///'):
3774
def _get_trees(self, revision_ids, cache):
3776
for rev_id in revision_ids:
3778
possible_trees.append((rev_id, cache[rev_id]))
3780
# Not cached, but inventory might be present anyway.
3782
tree = self.source.revision_tree(rev_id)
3783
except errors.NoSuchRevision:
3784
# Nope, parent is ghost.
3787
cache[rev_id] = tree
3788
possible_trees.append((rev_id, tree))
3789
return possible_trees
3791
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3792
"""Get the best delta and base for this revision.
3794
:return: (basis_id, delta)
3797
# Generate deltas against each tree, to find the shortest.
3798
texts_possibly_new_in_tree = set()
3799
for basis_id, basis_tree in possible_trees:
3800
delta = tree.inventory._make_delta(basis_tree.inventory)
3801
for old_path, new_path, file_id, new_entry in delta:
3802
if new_path is None:
3803
# This file_id isn't present in the new rev, so we don't
3807
# Rich roots are handled elsewhere...
3809
kind = new_entry.kind
3810
if kind != 'directory' and kind != 'file':
3811
# No text record associated with this inventory entry.
3813
# This is a directory or file that has changed somehow.
3814
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3815
deltas.append((len(delta), basis_id, delta))
3817
return deltas[0][1:]
3819
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3820
"""Find all parent revisions that are absent, but for which the
3821
inventory is present, and copy those inventories.
3823
This is necessary to preserve correctness when the source is stacked
3824
without fallbacks configured. (Note that in cases like upgrade the
3825
source may be not have _fallback_repositories even though it is
3829
for parents in parent_map.values():
3830
parent_revs.update(parents)
3831
present_parents = self.source.get_parent_map(parent_revs)
3832
absent_parents = set(parent_revs).difference(present_parents)
3833
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3834
(rev_id,) for rev_id in absent_parents)
3835
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3836
for parent_tree in self.source.revision_trees(parent_inv_ids):
3837
current_revision_id = parent_tree.get_revision_id()
3838
parents_parents_keys = parent_invs_keys_for_stacking[
3839
(current_revision_id,)]
3840
parents_parents = [key[-1] for key in parents_parents_keys]
3841
basis_id = _mod_revision.NULL_REVISION
3842
basis_tree = self.source.revision_tree(basis_id)
3843
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3844
self.target.add_inventory_by_delta(
3845
basis_id, delta, current_revision_id, parents_parents)
3846
cache[current_revision_id] = parent_tree
3848
def _fetch_batch(self, revision_ids, basis_id, cache):
3849
"""Fetch across a few revisions.
3851
:param revision_ids: The revisions to copy
3852
:param basis_id: The revision_id of a tree that must be in cache, used
3853
as a basis for delta when no other base is available
3854
:param cache: A cache of RevisionTrees that we can use.
3855
:return: The revision_id of the last converted tree. The RevisionTree
3856
for it will be in cache
3858
# Walk though all revisions; get inventory deltas, copy referenced
3859
# texts that delta references, insert the delta, revision and
3861
root_keys_to_create = set()
3864
pending_revisions = []
3865
parent_map = self.source.get_parent_map(revision_ids)
3866
self._fetch_parent_invs_for_stacking(parent_map, cache)
3867
self.source._safe_to_return_from_cache = True
3868
for tree in self.source.revision_trees(revision_ids):
3869
# Find a inventory delta for this revision.
3870
# Find text entries that need to be copied, too.
3871
current_revision_id = tree.get_revision_id()
3872
parent_ids = parent_map.get(current_revision_id, ())
3873
parent_trees = self._get_trees(parent_ids, cache)
3874
possible_trees = list(parent_trees)
3875
if len(possible_trees) == 0:
3876
# There either aren't any parents, or the parents are ghosts,
3877
# so just use the last converted tree.
3878
possible_trees.append((basis_id, cache[basis_id]))
3879
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3881
revision = self.source.get_revision(current_revision_id)
3882
pending_deltas.append((basis_id, delta,
3883
current_revision_id, revision.parent_ids))
3884
if self._converting_to_rich_root:
3885
self._revision_id_to_root_id[current_revision_id] = \
3887
# Determine which texts are in present in this revision but not in
3888
# any of the available parents.
3889
texts_possibly_new_in_tree = set()
3890
for old_path, new_path, file_id, entry in delta:
3891
if new_path is None:
3892
# This file_id isn't present in the new rev
3896
if not self.target.supports_rich_root():
3897
# The target doesn't support rich root, so we don't
3900
if self._converting_to_rich_root:
3901
# This can't be copied normally, we have to insert
3903
root_keys_to_create.add((file_id, entry.revision))
3906
texts_possibly_new_in_tree.add((file_id, entry.revision))
3907
for basis_id, basis_tree in possible_trees:
3908
basis_inv = basis_tree.inventory
3909
for file_key in list(texts_possibly_new_in_tree):
3910
file_id, file_revision = file_key
3912
entry = basis_inv[file_id]
3913
except errors.NoSuchId:
3915
if entry.revision == file_revision:
3916
texts_possibly_new_in_tree.remove(file_key)
3917
text_keys.update(texts_possibly_new_in_tree)
3918
pending_revisions.append(revision)
3919
cache[current_revision_id] = tree
3920
basis_id = current_revision_id
3921
self.source._safe_to_return_from_cache = False
3923
from_texts = self.source.texts
3924
to_texts = self.target.texts
3925
if root_keys_to_create:
3926
root_stream = _mod_fetch._new_root_data_stream(
3927
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3929
to_texts.insert_record_stream(root_stream)
3930
to_texts.insert_record_stream(from_texts.get_record_stream(
3931
text_keys, self.target._format._fetch_order,
3932
not self.target._format._fetch_uses_deltas))
3933
# insert inventory deltas
3934
for delta in pending_deltas:
3935
self.target.add_inventory_by_delta(*delta)
3936
if self.target._fallback_repositories:
3937
# Make sure this stacked repository has all the parent inventories
3938
# for the new revisions that we are about to insert. We do this
3939
# before adding the revisions so that no revision is added until
3940
# all the inventories it may depend on are added.
3941
# Note that this is overzealous, as we may have fetched these in an
3944
revision_ids = set()
3945
for revision in pending_revisions:
3946
revision_ids.add(revision.revision_id)
3947
parent_ids.update(revision.parent_ids)
3948
parent_ids.difference_update(revision_ids)
3949
parent_ids.discard(_mod_revision.NULL_REVISION)
3950
parent_map = self.source.get_parent_map(parent_ids)
3951
# we iterate over parent_map and not parent_ids because we don't
3952
# want to try copying any revision which is a ghost
3953
for parent_tree in self.source.revision_trees(parent_map):
3954
current_revision_id = parent_tree.get_revision_id()
3955
parents_parents = parent_map[current_revision_id]
3956
possible_trees = self._get_trees(parents_parents, cache)
3957
if len(possible_trees) == 0:
3958
# There either aren't any parents, or the parents are
3959
# ghosts, so just use the last converted tree.
3960
possible_trees.append((basis_id, cache[basis_id]))
3961
basis_id, delta = self._get_delta_for_revision(parent_tree,
3962
parents_parents, possible_trees)
3963
self.target.add_inventory_by_delta(
3964
basis_id, delta, current_revision_id, parents_parents)
3965
# insert signatures and revisions
3966
for revision in pending_revisions:
3968
signature = self.source.get_signature_text(
3969
revision.revision_id)
3970
self.target.add_signature_text(revision.revision_id,
3972
except errors.NoSuchRevision:
3974
self.target.add_revision(revision.revision_id, revision)
3977
def _fetch_all_revisions(self, revision_ids, pb):
3978
"""Fetch everything for the list of revisions.
3980
:param revision_ids: The list of revisions to fetch. Must be in
3982
:param pb: A ProgressTask
3985
basis_id, basis_tree = self._get_basis(revision_ids[0])
3987
cache = lru_cache.LRUCache(100)
3988
cache[basis_id] = basis_tree
3989
del basis_tree # We don't want to hang on to it here
3993
for offset in range(0, len(revision_ids), batch_size):
3994
self.target.start_write_group()
3996
pb.update('Transferring revisions', offset,
3998
batch = revision_ids[offset:offset+batch_size]
3999
basis_id = self._fetch_batch(batch, basis_id, cache)
4001
self.source._safe_to_return_from_cache = False
4002
self.target.abort_write_group()
4005
hint = self.target.commit_write_group()
4008
if hints and self.target._format.pack_compresses:
4009
self.target.pack(hint=hints)
4010
pb.update('Transferring revisions', len(revision_ids),
4014
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
4016
"""See InterRepository.fetch()."""
4017
if fetch_spec is not None:
4018
raise AssertionError("Not implemented yet...")
4019
ui.ui_factory.warn_experimental_format_fetch(self)
4020
if (not self.source.supports_rich_root()
4021
and self.target.supports_rich_root()):
4022
self._converting_to_rich_root = True
4023
self._revision_id_to_root_id = {}
4025
self._converting_to_rich_root = False
4026
# See <https://launchpad.net/bugs/456077> asking for a warning here
4027
if self.source._format.network_name() != self.target._format.network_name():
4028
ui.ui_factory.show_user_warning('cross_format_fetch',
4029
from_format=self.source._format,
4030
to_format=self.target._format)
4031
revision_ids = self.target.search_missing_revision_ids(self.source,
4032
revision_id, find_ghosts=find_ghosts).get_keys()
4033
if not revision_ids:
4035
revision_ids = tsort.topo_sort(
4036
self.source.get_graph().get_parent_map(revision_ids))
4037
if not revision_ids:
4039
# Walk though all revisions; get inventory deltas, copy referenced
4040
# texts that delta references, insert the delta, revision and
4043
my_pb = ui.ui_factory.nested_progress_bar()
4046
symbol_versioning.warn(
4047
symbol_versioning.deprecated_in((1, 14, 0))
4048
% "pb parameter to fetch()")
4051
self._fetch_all_revisions(revision_ids, pb)
4053
if my_pb is not None:
4055
return len(revision_ids), 0
4057
def _get_basis(self, first_revision_id):
4058
"""Get a revision and tree which exists in the target.
4060
This assumes that first_revision_id is selected for transmission
4061
because all other ancestors are already present. If we can't find an
4062
ancestor we fall back to NULL_REVISION since we know that is safe.
4064
:return: (basis_id, basis_tree)
4066
first_rev = self.source.get_revision(first_revision_id)
4068
basis_id = first_rev.parent_ids[0]
4069
# only valid as a basis if the target has it
4070
self.target.get_revision(basis_id)
4071
# Try to get a basis tree - if it's a ghost it will hit the
4072
# NoSuchRevision case.
4073
basis_tree = self.source.revision_tree(basis_id)
4074
except (IndexError, errors.NoSuchRevision):
4075
basis_id = _mod_revision.NULL_REVISION
4076
basis_tree = self.source.revision_tree(basis_id)
4077
return basis_id, basis_tree
4080
InterRepository.register_optimiser(InterDifferingSerializer)
4081
InterRepository.register_optimiser(InterSameDataRepository)
1843
4082
InterRepository.register_optimiser(InterWeaveRepo)
1844
4083
InterRepository.register_optimiser(InterKnitRepo)
1847
class RepositoryTestProviderAdapter(object):
1848
"""A tool to generate a suite testing multiple repository formats at once.
1850
This is done by copying the test once for each transport and injecting
1851
the transport_server, transport_readonly_server, and bzrdir_format and
1852
repository_format classes into each copy. Each copy is also given a new id()
1853
to make it easy to identify.
1856
def __init__(self, transport_server, transport_readonly_server, formats):
1857
self._transport_server = transport_server
1858
self._transport_readonly_server = transport_readonly_server
1859
self._formats = formats
1861
def adapt(self, test):
1862
result = TestSuite()
1863
for repository_format, bzrdir_format in self._formats:
1864
new_test = deepcopy(test)
1865
new_test.transport_server = self._transport_server
1866
new_test.transport_readonly_server = self._transport_readonly_server
1867
new_test.bzrdir_format = bzrdir_format
1868
new_test.repository_format = repository_format
1869
def make_new_test_id():
1870
new_id = "%s(%s)" % (new_test.id(), repository_format.__class__.__name__)
1871
return lambda: new_id
1872
new_test.id = make_new_test_id()
1873
result.addTest(new_test)
1877
class InterRepositoryTestProviderAdapter(object):
1878
"""A tool to generate a suite testing multiple inter repository formats.
1880
This is done by copying the test once for each interrepo provider and injecting
1881
the transport_server, transport_readonly_server, repository_format and
1882
repository_to_format classes into each copy.
1883
Each copy is also given a new id() to make it easy to identify.
1886
def __init__(self, transport_server, transport_readonly_server, formats):
1887
self._transport_server = transport_server
1888
self._transport_readonly_server = transport_readonly_server
1889
self._formats = formats
1891
def adapt(self, test):
1892
result = TestSuite()
1893
for interrepo_class, repository_format, repository_format_to in self._formats:
1894
new_test = deepcopy(test)
1895
new_test.transport_server = self._transport_server
1896
new_test.transport_readonly_server = self._transport_readonly_server
1897
new_test.interrepo_class = interrepo_class
1898
new_test.repository_format = repository_format
1899
new_test.repository_format_to = repository_format_to
1900
def make_new_test_id():
1901
new_id = "%s(%s)" % (new_test.id(), interrepo_class.__name__)
1902
return lambda: new_id
1903
new_test.id = make_new_test_id()
1904
result.addTest(new_test)
1908
def default_test_list():
1909
"""Generate the default list of interrepo permutations to test."""
1911
# test the default InterRepository between format 6 and the current
1913
# XXX: robertc 20060220 reinstate this when there are two supported
1914
# formats which do not have an optimal code path between them.
1915
result.append((InterRepository,
1916
RepositoryFormat6(),
1917
RepositoryFormatKnit1()))
1918
for optimiser in InterRepository._optimisers:
1919
result.append((optimiser,
1920
optimiser._matching_repo_format,
1921
optimiser._matching_repo_format
1923
# if there are specific combinations we want to use, we can add them
1928
4086
class CopyConverter(object):
1929
4087
"""A repository conversion tool which just performs a copy of the content.
1931
4089
This is slow but quite reliable.
2172
4157
"""Unescape predefined XML entities in a string of data."""
2173
4158
global _unescape_re
2174
4159
if _unescape_re is None:
2175
_unescape_re = re.compile('\&([^;]*);')
4160
_unescape_re = re.compile('\&([^;]*);')
2176
4161
return _unescape_re.sub(_unescaper, data)
4164
class _VersionedFileChecker(object):
4166
def __init__(self, repository, text_key_references=None, ancestors=None):
4167
self.repository = repository
4168
self.text_index = self.repository._generate_text_key_index(
4169
text_key_references=text_key_references, ancestors=ancestors)
4171
def calculate_file_version_parents(self, text_key):
4172
"""Calculate the correct parents for a file version according to
4175
parent_keys = self.text_index[text_key]
4176
if parent_keys == [_mod_revision.NULL_REVISION]:
4178
return tuple(parent_keys)
4180
def check_file_version_parents(self, texts, progress_bar=None):
4181
"""Check the parents stored in a versioned file are correct.
4183
It also detects file versions that are not referenced by their
4184
corresponding revision's inventory.
4186
:returns: A tuple of (wrong_parents, dangling_file_versions).
4187
wrong_parents is a dict mapping {revision_id: (stored_parents,
4188
correct_parents)} for each revision_id where the stored parents
4189
are not correct. dangling_file_versions is a set of (file_id,
4190
revision_id) tuples for versions that are present in this versioned
4191
file, but not used by the corresponding inventory.
4193
local_progress = None
4194
if progress_bar is None:
4195
local_progress = ui.ui_factory.nested_progress_bar()
4196
progress_bar = local_progress
4198
return self._check_file_version_parents(texts, progress_bar)
4201
local_progress.finished()
4203
def _check_file_version_parents(self, texts, progress_bar):
4204
"""See check_file_version_parents."""
4206
self.file_ids = set([file_id for file_id, _ in
4207
self.text_index.iterkeys()])
4208
# text keys is now grouped by file_id
4209
n_versions = len(self.text_index)
4210
progress_bar.update('loading text store', 0, n_versions)
4211
parent_map = self.repository.texts.get_parent_map(self.text_index)
4212
# On unlistable transports this could well be empty/error...
4213
text_keys = self.repository.texts.keys()
4214
unused_keys = frozenset(text_keys) - set(self.text_index)
4215
for num, key in enumerate(self.text_index.iterkeys()):
4216
progress_bar.update('checking text graph', num, n_versions)
4217
correct_parents = self.calculate_file_version_parents(key)
4219
knit_parents = parent_map[key]
4220
except errors.RevisionNotPresent:
4223
if correct_parents != knit_parents:
4224
wrong_parents[key] = (knit_parents, correct_parents)
4225
return wrong_parents, unused_keys
4228
def _old_get_graph(repository, revision_id):
4229
"""DO NOT USE. That is all. I'm serious."""
4230
graph = repository.get_graph()
4231
revision_graph = dict(((key, value) for key, value in
4232
graph.iter_ancestry([revision_id]) if value is not None))
4233
return _strip_NULL_ghosts(revision_graph)
4236
def _strip_NULL_ghosts(revision_graph):
4237
"""Also don't use this. more compatibility code for unmigrated clients."""
4238
# Filter ghosts, and null:
4239
if _mod_revision.NULL_REVISION in revision_graph:
4240
del revision_graph[_mod_revision.NULL_REVISION]
4241
for key, parents in revision_graph.items():
4242
revision_graph[key] = tuple(parent for parent in parents if parent
4244
return revision_graph
4247
class StreamSink(object):
4248
"""An object that can insert a stream into a repository.
4250
This interface handles the complexity of reserialising inventories and
4251
revisions from different formats, and allows unidirectional insertion into
4252
stacked repositories without looking for the missing basis parents
4256
def __init__(self, target_repo):
4257
self.target_repo = target_repo
4259
def insert_stream(self, stream, src_format, resume_tokens):
4260
"""Insert a stream's content into the target repository.
4262
:param src_format: a bzr repository format.
4264
:return: a list of resume tokens and an iterable of keys additional
4265
items required before the insertion can be completed.
4267
self.target_repo.lock_write()
4270
self.target_repo.resume_write_group(resume_tokens)
4273
self.target_repo.start_write_group()
4276
# locked_insert_stream performs a commit|suspend.
4277
return self._locked_insert_stream(stream, src_format,
4280
self.target_repo.abort_write_group(suppress_errors=True)
4283
self.target_repo.unlock()
4285
def _locked_insert_stream(self, stream, src_format, is_resume):
4286
to_serializer = self.target_repo._format._serializer
4287
src_serializer = src_format._serializer
4289
if to_serializer == src_serializer:
4290
# If serializers match and the target is a pack repository, set the
4291
# write cache size on the new pack. This avoids poor performance
4292
# on transports where append is unbuffered (such as
4293
# RemoteTransport). This is safe to do because nothing should read
4294
# back from the target repository while a stream with matching
4295
# serialization is being inserted.
4296
# The exception is that a delta record from the source that should
4297
# be a fulltext may need to be expanded by the target (see
4298
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4299
# explicitly flush any buffered writes first in that rare case.
4301
new_pack = self.target_repo._pack_collection._new_pack
4302
except AttributeError:
4303
# Not a pack repository
4306
new_pack.set_write_cache_size(1024*1024)
4307
for substream_type, substream in stream:
4308
if 'stream' in debug.debug_flags:
4309
mutter('inserting substream: %s', substream_type)
4310
if substream_type == 'texts':
4311
self.target_repo.texts.insert_record_stream(substream)
4312
elif substream_type == 'inventories':
4313
if src_serializer == to_serializer:
4314
self.target_repo.inventories.insert_record_stream(
4317
self._extract_and_insert_inventories(
4318
substream, src_serializer)
4319
elif substream_type == 'inventory-deltas':
4320
self._extract_and_insert_inventory_deltas(
4321
substream, src_serializer)
4322
elif substream_type == 'chk_bytes':
4323
# XXX: This doesn't support conversions, as it assumes the
4324
# conversion was done in the fetch code.
4325
self.target_repo.chk_bytes.insert_record_stream(substream)
4326
elif substream_type == 'revisions':
4327
# This may fallback to extract-and-insert more often than
4328
# required if the serializers are different only in terms of
4330
if src_serializer == to_serializer:
4331
self.target_repo.revisions.insert_record_stream(substream)
4333
self._extract_and_insert_revisions(substream,
4335
elif substream_type == 'signatures':
4336
self.target_repo.signatures.insert_record_stream(substream)
4338
raise AssertionError('kaboom! %s' % (substream_type,))
4339
# Done inserting data, and the missing_keys calculations will try to
4340
# read back from the inserted data, so flush the writes to the new pack
4341
# (if this is pack format).
4342
if new_pack is not None:
4343
new_pack._write_data('', flush=True)
4344
# Find all the new revisions (including ones from resume_tokens)
4345
missing_keys = self.target_repo.get_missing_parent_inventories(
4346
check_for_missing_texts=is_resume)
4348
for prefix, versioned_file in (
4349
('texts', self.target_repo.texts),
4350
('inventories', self.target_repo.inventories),
4351
('revisions', self.target_repo.revisions),
4352
('signatures', self.target_repo.signatures),
4353
('chk_bytes', self.target_repo.chk_bytes),
4355
if versioned_file is None:
4357
# TODO: key is often going to be a StaticTuple object
4358
# I don't believe we can define a method by which
4359
# (prefix,) + StaticTuple will work, though we could
4360
# define a StaticTuple.sq_concat that would allow you to
4361
# pass in either a tuple or a StaticTuple as the second
4362
# object, so instead we could have:
4363
# StaticTuple(prefix) + key here...
4364
missing_keys.update((prefix,) + key for key in
4365
versioned_file.get_missing_compression_parent_keys())
4366
except NotImplementedError:
4367
# cannot even attempt suspending, and missing would have failed
4368
# during stream insertion.
4369
missing_keys = set()
4372
# suspend the write group and tell the caller what we is
4373
# missing. We know we can suspend or else we would not have
4374
# entered this code path. (All repositories that can handle
4375
# missing keys can handle suspending a write group).
4376
write_group_tokens = self.target_repo.suspend_write_group()
4377
return write_group_tokens, missing_keys
4378
hint = self.target_repo.commit_write_group()
4379
if (to_serializer != src_serializer and
4380
self.target_repo._format.pack_compresses):
4381
self.target_repo.pack(hint=hint)
4384
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4385
target_rich_root = self.target_repo._format.rich_root_data
4386
target_tree_refs = self.target_repo._format.supports_tree_reference
4387
for record in substream:
4388
# Insert the delta directly
4389
inventory_delta_bytes = record.get_bytes_as('fulltext')
4390
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4392
parse_result = deserialiser.parse_text_bytes(
4393
inventory_delta_bytes)
4394
except inventory_delta.IncompatibleInventoryDelta, err:
4395
trace.mutter("Incompatible delta: %s", err.msg)
4396
raise errors.IncompatibleRevision(self.target_repo._format)
4397
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4398
revision_id = new_id
4399
parents = [key[0] for key in record.parents]
4400
self.target_repo.add_inventory_by_delta(
4401
basis_id, inv_delta, revision_id, parents)
4403
def _extract_and_insert_inventories(self, substream, serializer,
4405
"""Generate a new inventory versionedfile in target, converting data.
4407
The inventory is retrieved from the source, (deserializing it), and
4408
stored in the target (reserializing it in a different format).
4410
target_rich_root = self.target_repo._format.rich_root_data
4411
target_tree_refs = self.target_repo._format.supports_tree_reference
4412
for record in substream:
4413
# It's not a delta, so it must be a fulltext in the source
4414
# serializer's format.
4415
bytes = record.get_bytes_as('fulltext')
4416
revision_id = record.key[0]
4417
inv = serializer.read_inventory_from_string(bytes, revision_id)
4418
parents = [key[0] for key in record.parents]
4419
self.target_repo.add_inventory(revision_id, inv, parents)
4420
# No need to keep holding this full inv in memory when the rest of
4421
# the substream is likely to be all deltas.
4424
def _extract_and_insert_revisions(self, substream, serializer):
4425
for record in substream:
4426
bytes = record.get_bytes_as('fulltext')
4427
revision_id = record.key[0]
4428
rev = serializer.read_revision_from_string(bytes)
4429
if rev.revision_id != revision_id:
4430
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4431
self.target_repo.add_revision(revision_id, rev)
4434
if self.target_repo._format._fetch_reconcile:
4435
self.target_repo.reconcile()
4438
class StreamSource(object):
4439
"""A source of a stream for fetching between repositories."""
4441
def __init__(self, from_repository, to_format):
4442
"""Create a StreamSource streaming from from_repository."""
4443
self.from_repository = from_repository
4444
self.to_format = to_format
4445
self._record_counter = RecordCounter()
4447
def delta_on_metadata(self):
4448
"""Return True if delta's are permitted on metadata streams.
4450
That is on revisions and signatures.
4452
src_serializer = self.from_repository._format._serializer
4453
target_serializer = self.to_format._serializer
4454
return (self.to_format._fetch_uses_deltas and
4455
src_serializer == target_serializer)
4457
def _fetch_revision_texts(self, revs):
4458
# fetch signatures first and then the revision texts
4459
# may need to be a InterRevisionStore call here.
4460
from_sf = self.from_repository.signatures
4461
# A missing signature is just skipped.
4462
keys = [(rev_id,) for rev_id in revs]
4463
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4465
self.to_format._fetch_order,
4466
not self.to_format._fetch_uses_deltas))
4467
# If a revision has a delta, this is actually expanded inside the
4468
# insert_record_stream code now, which is an alternate fix for
4470
from_rf = self.from_repository.revisions
4471
revisions = from_rf.get_record_stream(
4473
self.to_format._fetch_order,
4474
not self.delta_on_metadata())
4475
return [('signatures', signatures), ('revisions', revisions)]
4477
def _generate_root_texts(self, revs):
4478
"""This will be called by get_stream between fetching weave texts and
4479
fetching the inventory weave.
4481
if self._rich_root_upgrade():
4482
return _mod_fetch.Inter1and2Helper(
4483
self.from_repository).generate_root_texts(revs)
4487
def get_stream(self, search):
4489
revs = search.get_keys()
4490
graph = self.from_repository.get_graph()
4491
revs = tsort.topo_sort(graph.get_parent_map(revs))
4492
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4494
for knit_kind, file_id, revisions in data_to_fetch:
4495
if knit_kind != phase:
4497
# Make a new progress bar for this phase
4498
if knit_kind == "file":
4499
# Accumulate file texts
4500
text_keys.extend([(file_id, revision) for revision in
4502
elif knit_kind == "inventory":
4503
# Now copy the file texts.
4504
from_texts = self.from_repository.texts
4505
yield ('texts', from_texts.get_record_stream(
4506
text_keys, self.to_format._fetch_order,
4507
not self.to_format._fetch_uses_deltas))
4508
# Cause an error if a text occurs after we have done the
4511
# Before we process the inventory we generate the root
4512
# texts (if necessary) so that the inventories references
4514
for _ in self._generate_root_texts(revs):
4516
# we fetch only the referenced inventories because we do not
4517
# know for unselected inventories whether all their required
4518
# texts are present in the other repository - it could be
4520
for info in self._get_inventory_stream(revs):
4522
elif knit_kind == "signatures":
4523
# Nothing to do here; this will be taken care of when
4524
# _fetch_revision_texts happens.
4526
elif knit_kind == "revisions":
4527
for record in self._fetch_revision_texts(revs):
4530
raise AssertionError("Unknown knit kind %r" % knit_kind)
4532
def get_stream_for_missing_keys(self, missing_keys):
4533
# missing keys can only occur when we are byte copying and not
4534
# translating (because translation means we don't send
4535
# unreconstructable deltas ever).
4537
keys['texts'] = set()
4538
keys['revisions'] = set()
4539
keys['inventories'] = set()
4540
keys['chk_bytes'] = set()
4541
keys['signatures'] = set()
4542
for key in missing_keys:
4543
keys[key[0]].add(key[1:])
4544
if len(keys['revisions']):
4545
# If we allowed copying revisions at this point, we could end up
4546
# copying a revision without copying its required texts: a
4547
# violation of the requirements for repository integrity.
4548
raise AssertionError(
4549
'cannot copy revisions to fill in missing deltas %s' % (
4550
keys['revisions'],))
4551
for substream_kind, keys in keys.iteritems():
4552
vf = getattr(self.from_repository, substream_kind)
4553
if vf is None and keys:
4554
raise AssertionError(
4555
"cannot fill in keys for a versioned file we don't"
4556
" have: %s needs %s" % (substream_kind, keys))
4558
# No need to stream something we don't have
4560
if substream_kind == 'inventories':
4561
# Some missing keys are genuinely ghosts, filter those out.
4562
present = self.from_repository.inventories.get_parent_map(keys)
4563
revs = [key[0] for key in present]
4564
# Get the inventory stream more-or-less as we do for the
4565
# original stream; there's no reason to assume that records
4566
# direct from the source will be suitable for the sink. (Think
4567
# e.g. 2a -> 1.9-rich-root).
4568
for info in self._get_inventory_stream(revs, missing=True):
4572
# Ask for full texts always so that we don't need more round trips
4573
# after this stream.
4574
# Some of the missing keys are genuinely ghosts, so filter absent
4575
# records. The Sink is responsible for doing another check to
4576
# ensure that ghosts don't introduce missing data for future
4578
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4579
self.to_format._fetch_order, True))
4580
yield substream_kind, stream
4582
def inventory_fetch_order(self):
4583
if self._rich_root_upgrade():
4584
return 'topological'
4586
return self.to_format._fetch_order
4588
def _rich_root_upgrade(self):
4589
return (not self.from_repository._format.rich_root_data and
4590
self.to_format.rich_root_data)
4592
def _get_inventory_stream(self, revision_ids, missing=False):
4593
from_format = self.from_repository._format
4594
if (from_format.supports_chks and self.to_format.supports_chks and
4595
from_format.network_name() == self.to_format.network_name()):
4596
raise AssertionError(
4597
"this case should be handled by GroupCHKStreamSource")
4598
elif 'forceinvdeltas' in debug.debug_flags:
4599
return self._get_convertable_inventory_stream(revision_ids,
4600
delta_versus_null=missing)
4601
elif from_format.network_name() == self.to_format.network_name():
4603
return self._get_simple_inventory_stream(revision_ids,
4605
elif (not from_format.supports_chks and not self.to_format.supports_chks
4606
and from_format._serializer == self.to_format._serializer):
4607
# Essentially the same format.
4608
return self._get_simple_inventory_stream(revision_ids,
4611
# Any time we switch serializations, we want to use an
4612
# inventory-delta based approach.
4613
return self._get_convertable_inventory_stream(revision_ids,
4614
delta_versus_null=missing)
4616
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4617
# NB: This currently reopens the inventory weave in source;
4618
# using a single stream interface instead would avoid this.
4619
from_weave = self.from_repository.inventories
4621
delta_closure = True
4623
delta_closure = not self.delta_on_metadata()
4624
yield ('inventories', from_weave.get_record_stream(
4625
[(rev_id,) for rev_id in revision_ids],
4626
self.inventory_fetch_order(), delta_closure))
4628
def _get_convertable_inventory_stream(self, revision_ids,
4629
delta_versus_null=False):
4630
# The two formats are sufficiently different that there is no fast
4631
# path, so we need to send just inventorydeltas, which any
4632
# sufficiently modern client can insert into any repository.
4633
# The StreamSink code expects to be able to
4634
# convert on the target, so we need to put bytes-on-the-wire that can
4635
# be converted. That means inventory deltas (if the remote is <1.19,
4636
# RemoteStreamSink will fallback to VFS to insert the deltas).
4637
yield ('inventory-deltas',
4638
self._stream_invs_as_deltas(revision_ids,
4639
delta_versus_null=delta_versus_null))
4641
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4642
"""Return a stream of inventory-deltas for the given rev ids.
4644
:param revision_ids: The list of inventories to transmit
4645
:param delta_versus_null: Don't try to find a minimal delta for this
4646
entry, instead compute the delta versus the NULL_REVISION. This
4647
effectively streams a complete inventory. Used for stuff like
4648
filling in missing parents, etc.
4650
from_repo = self.from_repository
4651
revision_keys = [(rev_id,) for rev_id in revision_ids]
4652
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4653
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4655
inventories = self.from_repository.iter_inventories(
4656
revision_ids, 'topological')
4657
format = from_repo._format
4658
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4659
inventory_cache = lru_cache.LRUCache(50)
4660
null_inventory = from_repo.revision_tree(
4661
_mod_revision.NULL_REVISION).inventory
4662
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4663
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4664
# repo back into a non-rich-root repo ought to be allowed)
4665
serializer = inventory_delta.InventoryDeltaSerializer(
4666
versioned_root=format.rich_root_data,
4667
tree_references=format.supports_tree_reference)
4668
for inv in inventories:
4669
key = (inv.revision_id,)
4670
parent_keys = parent_map.get(key, ())
4672
if not delta_versus_null and parent_keys:
4673
# The caller did not ask for complete inventories and we have
4674
# some parents that we can delta against. Make a delta against
4675
# each parent so that we can find the smallest.
4676
parent_ids = [parent_key[0] for parent_key in parent_keys]
4677
for parent_id in parent_ids:
4678
if parent_id not in invs_sent_so_far:
4679
# We don't know that the remote side has this basis, so
4682
if parent_id == _mod_revision.NULL_REVISION:
4683
parent_inv = null_inventory
4685
parent_inv = inventory_cache.get(parent_id, None)
4686
if parent_inv is None:
4687
parent_inv = from_repo.get_inventory(parent_id)
4688
candidate_delta = inv._make_delta(parent_inv)
4689
if (delta is None or
4690
len(delta) > len(candidate_delta)):
4691
delta = candidate_delta
4692
basis_id = parent_id
4694
# Either none of the parents ended up being suitable, or we
4695
# were asked to delta against NULL
4696
basis_id = _mod_revision.NULL_REVISION
4697
delta = inv._make_delta(null_inventory)
4698
invs_sent_so_far.add(inv.revision_id)
4699
inventory_cache[inv.revision_id] = inv
4700
delta_serialized = ''.join(
4701
serializer.delta_to_lines(basis_id, key[-1], delta))
4702
yield versionedfile.FulltextContentFactory(
4703
key, parent_keys, None, delta_serialized)
4706
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4707
stop_revision=None):
4708
"""Extend the partial history to include a given index
4710
If a stop_index is supplied, stop when that index has been reached.
4711
If a stop_revision is supplied, stop when that revision is
4712
encountered. Otherwise, stop when the beginning of history is
4715
:param stop_index: The index which should be present. When it is
4716
present, history extension will stop.
4717
:param stop_revision: The revision id which should be present. When
4718
it is encountered, history extension will stop.
4720
start_revision = partial_history_cache[-1]
4721
iterator = repo.iter_reverse_revision_history(start_revision)
4723
#skip the last revision in the list
4726
if (stop_index is not None and
4727
len(partial_history_cache) > stop_index):
4729
if partial_history_cache[-1] == stop_revision:
4731
revision_id = iterator.next()
4732
partial_history_cache.append(revision_id)
4733
except StopIteration: