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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
from cStringIO import StringIO
17
19
from bzrlib.lazy_import import lazy_import
18
20
lazy_import(globals(), """
23
24
from bzrlib import (
41
39
revision as _mod_revision,
49
44
from bzrlib.bundle import serializer
50
45
from bzrlib.revisiontree import RevisionTree
51
46
from bzrlib.store.versioned import VersionedFileStore
47
from bzrlib.store.text import TextStore
52
48
from bzrlib.testament import Testament
59
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
51
from bzrlib.decorators import needs_read_lock, needs_write_lock
60
52
from bzrlib.inter import InterObject
61
from bzrlib.inventory import (
67
from bzrlib.lock import _RelockDebugMixin, LogicalLockResult
68
from bzrlib.trace import (
69
log_exception_quietly, note, mutter, mutter_callsite, warning)
53
from bzrlib.inventory import Inventory, InventoryDirectory, ROOT_ID
54
from bzrlib.symbol_versioning import (
57
from bzrlib.trace import mutter, mutter_callsite, note, warning
72
60
# Old formats display a warning, but only once
73
61
_deprecation_warning_done = False
76
class IsInWriteGroupError(errors.InternalBzrError):
78
_fmt = "May not refresh_data of repo %(repo)s while in a write group."
80
def __init__(self, repo):
81
errors.InternalBzrError.__init__(self, repo=repo)
84
class CommitBuilder(object):
85
"""Provides an interface to build up a commit.
87
This allows describing a tree to be committed without needing to
88
know the internals of the format of the repository.
91
# all clients should supply tree roots.
92
record_root_entry = True
93
# the default CommitBuilder does not manage trees whose root is versioned.
94
_versioned_root = False
96
def __init__(self, repository, parents, config, timestamp=None,
97
timezone=None, committer=None, revprops=None,
99
"""Initiate a CommitBuilder.
101
:param repository: Repository to commit to.
102
:param parents: Revision ids of the parents of the new revision.
103
:param config: Configuration to use.
104
:param timestamp: Optional timestamp recorded for commit.
105
:param timezone: Optional timezone for timestamp.
106
:param committer: Optional committer to set for commit.
107
:param revprops: Optional dictionary of revision properties.
108
:param revision_id: Optional revision id.
110
self._config = config
112
if committer is None:
113
self._committer = self._config.username()
115
self._committer = committer
117
self.new_inventory = Inventory(None)
118
self._new_revision_id = revision_id
119
self.parents = parents
120
self.repository = repository
123
if revprops is not None:
124
self._validate_revprops(revprops)
125
self._revprops.update(revprops)
127
if timestamp is None:
128
timestamp = time.time()
129
# Restrict resolution to 1ms
130
self._timestamp = round(timestamp, 3)
133
self._timezone = osutils.local_time_offset()
135
self._timezone = int(timezone)
137
self._generate_revision_if_needed()
138
self.__heads = graph.HeadsCache(repository.get_graph()).heads
139
self._basis_delta = []
140
# API compatibility, older code that used CommitBuilder did not call
141
# .record_delete(), which means the delta that is computed would not be
142
# valid. Callers that will call record_delete() should call
143
# .will_record_deletes() to indicate that.
144
self._recording_deletes = False
145
# memo'd check for no-op commits.
146
self._any_changes = False
148
def any_changes(self):
149
"""Return True if any entries were changed.
151
This includes merge-only changes. It is the core for the --unchanged
154
:return: True if any changes have occured.
156
return self._any_changes
158
def _validate_unicode_text(self, text, context):
159
"""Verify things like commit messages don't have bogus characters."""
161
raise ValueError('Invalid value for %s: %r' % (context, text))
163
def _validate_revprops(self, revprops):
164
for key, value in revprops.iteritems():
165
# We know that the XML serializers do not round trip '\r'
166
# correctly, so refuse to accept them
167
if not isinstance(value, basestring):
168
raise ValueError('revision property (%s) is not a valid'
169
' (unicode) string: %r' % (key, value))
170
self._validate_unicode_text(value,
171
'revision property (%s)' % (key,))
173
def commit(self, message):
174
"""Make the actual commit.
176
:return: The revision id of the recorded revision.
178
self._validate_unicode_text(message, 'commit message')
179
rev = _mod_revision.Revision(
180
timestamp=self._timestamp,
181
timezone=self._timezone,
182
committer=self._committer,
184
inventory_sha1=self.inv_sha1,
185
revision_id=self._new_revision_id,
186
properties=self._revprops)
187
rev.parent_ids = self.parents
188
self.repository.add_revision(self._new_revision_id, rev,
189
self.new_inventory, self._config)
190
self.repository.commit_write_group()
191
return self._new_revision_id
194
"""Abort the commit that is being built.
196
self.repository.abort_write_group()
198
def revision_tree(self):
199
"""Return the tree that was just committed.
201
After calling commit() this can be called to get a RevisionTree
202
representing the newly committed tree. This is preferred to
203
calling Repository.revision_tree() because that may require
204
deserializing the inventory, while we already have a copy in
207
if self.new_inventory is None:
208
self.new_inventory = self.repository.get_inventory(
209
self._new_revision_id)
210
return RevisionTree(self.repository, self.new_inventory,
211
self._new_revision_id)
213
def finish_inventory(self):
214
"""Tell the builder that the inventory is finished.
216
:return: The inventory id in the repository, which can be used with
217
repository.get_inventory.
219
if self.new_inventory is None:
220
# an inventory delta was accumulated without creating a new
222
basis_id = self.basis_delta_revision
223
# We ignore the 'inventory' returned by add_inventory_by_delta
224
# because self.new_inventory is used to hint to the rest of the
225
# system what code path was taken
226
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
227
basis_id, self._basis_delta, self._new_revision_id,
230
if self.new_inventory.root is None:
231
raise AssertionError('Root entry should be supplied to'
232
' record_entry_contents, as of bzr 0.10.')
233
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
234
self.new_inventory.revision_id = self._new_revision_id
235
self.inv_sha1 = self.repository.add_inventory(
236
self._new_revision_id,
240
return self._new_revision_id
242
def _gen_revision_id(self):
243
"""Return new revision-id."""
244
return generate_ids.gen_revision_id(self._config.username(),
247
def _generate_revision_if_needed(self):
248
"""Create a revision id if None was supplied.
250
If the repository can not support user-specified revision ids
251
they should override this function and raise CannotSetRevisionId
252
if _new_revision_id is not None.
254
:raises: CannotSetRevisionId
256
if self._new_revision_id is None:
257
self._new_revision_id = self._gen_revision_id()
258
self.random_revid = True
260
self.random_revid = False
262
def _heads(self, file_id, revision_ids):
263
"""Calculate the graph heads for revision_ids in the graph of file_id.
265
This can use either a per-file graph or a global revision graph as we
266
have an identity relationship between the two graphs.
268
return self.__heads(revision_ids)
270
def _check_root(self, ie, parent_invs, tree):
271
"""Helper for record_entry_contents.
273
:param ie: An entry being added.
274
:param parent_invs: The inventories of the parent revisions of the
276
:param tree: The tree that is being committed.
278
# In this revision format, root entries have no knit or weave When
279
# serializing out to disk and back in root.revision is always
281
ie.revision = self._new_revision_id
283
def _require_root_change(self, tree):
284
"""Enforce an appropriate root object change.
286
This is called once when record_iter_changes is called, if and only if
287
the root was not in the delta calculated by record_iter_changes.
289
:param tree: The tree which is being committed.
291
# NB: if there are no parents then this method is not called, so no
292
# need to guard on parents having length.
293
entry = entry_factory['directory'](tree.path2id(''), '',
295
entry.revision = self._new_revision_id
296
self._basis_delta.append(('', '', entry.file_id, entry))
298
def _get_delta(self, ie, basis_inv, path):
299
"""Get a delta against the basis inventory for ie."""
300
if ie.file_id not in basis_inv:
302
result = (None, path, ie.file_id, ie)
303
self._basis_delta.append(result)
305
elif ie != basis_inv[ie.file_id]:
307
# TODO: avoid tis id2path call.
308
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
309
self._basis_delta.append(result)
315
def get_basis_delta(self):
316
"""Return the complete inventory delta versus the basis inventory.
318
This has been built up with the calls to record_delete and
319
record_entry_contents. The client must have already called
320
will_record_deletes() to indicate that they will be generating a
323
:return: An inventory delta, suitable for use with apply_delta, or
324
Repository.add_inventory_by_delta, etc.
326
if not self._recording_deletes:
327
raise AssertionError("recording deletes not activated.")
328
return self._basis_delta
330
def record_delete(self, path, file_id):
331
"""Record that a delete occured against a basis tree.
333
This is an optional API - when used it adds items to the basis_delta
334
being accumulated by the commit builder. It cannot be called unless the
335
method will_record_deletes() has been called to inform the builder that
336
a delta is being supplied.
338
:param path: The path of the thing deleted.
339
:param file_id: The file id that was deleted.
341
if not self._recording_deletes:
342
raise AssertionError("recording deletes not activated.")
343
delta = (path, None, file_id, None)
344
self._basis_delta.append(delta)
345
self._any_changes = True
348
def will_record_deletes(self):
349
"""Tell the commit builder that deletes are being notified.
351
This enables the accumulation of an inventory delta; for the resulting
352
commit to be valid, deletes against the basis MUST be recorded via
353
builder.record_delete().
355
self._recording_deletes = True
357
basis_id = self.parents[0]
359
basis_id = _mod_revision.NULL_REVISION
360
self.basis_delta_revision = basis_id
362
def record_entry_contents(self, ie, parent_invs, path, tree,
364
"""Record the content of ie from tree into the commit if needed.
366
Side effect: sets ie.revision when unchanged
368
:param ie: An inventory entry present in the commit.
369
:param parent_invs: The inventories of the parent revisions of the
371
:param path: The path the entry is at in the tree.
372
:param tree: The tree which contains this entry and should be used to
374
:param content_summary: Summary data from the tree about the paths
375
content - stat, length, exec, sha/link target. This is only
376
accessed when the entry has a revision of None - that is when it is
377
a candidate to commit.
378
:return: A tuple (change_delta, version_recorded, fs_hash).
379
change_delta is an inventory_delta change for this entry against
380
the basis tree of the commit, or None if no change occured against
382
version_recorded is True if a new version of the entry has been
383
recorded. For instance, committing a merge where a file was only
384
changed on the other side will return (delta, False).
385
fs_hash is either None, or the hash details for the path (currently
386
a tuple of the contents sha1 and the statvalue returned by
387
tree.get_file_with_stat()).
389
if self.new_inventory.root is None:
390
if ie.parent_id is not None:
391
raise errors.RootMissing()
392
self._check_root(ie, parent_invs, tree)
393
if ie.revision is None:
394
kind = content_summary[0]
396
# ie is carried over from a prior commit
398
# XXX: repository specific check for nested tree support goes here - if
399
# the repo doesn't want nested trees we skip it ?
400
if (kind == 'tree-reference' and
401
not self.repository._format.supports_tree_reference):
402
# mismatch between commit builder logic and repository:
403
# this needs the entry creation pushed down into the builder.
404
raise NotImplementedError('Missing repository subtree support.')
405
self.new_inventory.add(ie)
407
# TODO: slow, take it out of the inner loop.
409
basis_inv = parent_invs[0]
411
basis_inv = Inventory(root_id=None)
413
# ie.revision is always None if the InventoryEntry is considered
414
# for committing. We may record the previous parents revision if the
415
# content is actually unchanged against a sole head.
416
if ie.revision is not None:
417
if not self._versioned_root and path == '':
418
# repositories that do not version the root set the root's
419
# revision to the new commit even when no change occurs (more
420
# specifically, they do not record a revision on the root; and
421
# the rev id is assigned to the root during deserialisation -
422
# this masks when a change may have occurred against the basis.
423
# To match this we always issue a delta, because the revision
424
# of the root will always be changing.
425
if ie.file_id in basis_inv:
426
delta = (basis_inv.id2path(ie.file_id), path,
430
delta = (None, path, ie.file_id, ie)
431
self._basis_delta.append(delta)
432
return delta, False, None
434
# we don't need to commit this, because the caller already
435
# determined that an existing revision of this file is
436
# appropriate. If its not being considered for committing then
437
# it and all its parents to the root must be unaltered so
438
# no-change against the basis.
439
if ie.revision == self._new_revision_id:
440
raise AssertionError("Impossible situation, a skipped "
441
"inventory entry (%r) claims to be modified in this "
442
"commit (%r).", (ie, self._new_revision_id))
443
return None, False, None
444
# XXX: Friction: parent_candidates should return a list not a dict
445
# so that we don't have to walk the inventories again.
446
parent_candiate_entries = ie.parent_candidates(parent_invs)
447
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
449
for inv in parent_invs:
450
if ie.file_id in inv:
451
old_rev = inv[ie.file_id].revision
452
if old_rev in head_set:
453
heads.append(inv[ie.file_id].revision)
454
head_set.remove(inv[ie.file_id].revision)
457
# now we check to see if we need to write a new record to the
459
# We write a new entry unless there is one head to the ancestors, and
460
# the kind-derived content is unchanged.
462
# Cheapest check first: no ancestors, or more the one head in the
463
# ancestors, we write a new node.
467
# There is a single head, look it up for comparison
468
parent_entry = parent_candiate_entries[heads[0]]
469
# if the non-content specific data has changed, we'll be writing a
471
if (parent_entry.parent_id != ie.parent_id or
472
parent_entry.name != ie.name):
474
# now we need to do content specific checks:
476
# if the kind changed the content obviously has
477
if kind != parent_entry.kind:
479
# Stat cache fingerprint feedback for the caller - None as we usually
480
# don't generate one.
483
if content_summary[2] is None:
484
raise ValueError("Files must not have executable = None")
486
# We can't trust a check of the file length because of content
488
if (# if the exec bit has changed we have to store:
489
parent_entry.executable != content_summary[2]):
491
elif parent_entry.text_sha1 == content_summary[3]:
492
# all meta and content is unchanged (using a hash cache
493
# hit to check the sha)
494
ie.revision = parent_entry.revision
495
ie.text_size = parent_entry.text_size
496
ie.text_sha1 = parent_entry.text_sha1
497
ie.executable = parent_entry.executable
498
return self._get_delta(ie, basis_inv, path), False, None
500
# Either there is only a hash change(no hash cache entry,
501
# or same size content change), or there is no change on
503
# Provide the parent's hash to the store layer, so that the
504
# content is unchanged we will not store a new node.
505
nostore_sha = parent_entry.text_sha1
507
# We want to record a new node regardless of the presence or
508
# absence of a content change in the file.
510
ie.executable = content_summary[2]
511
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
513
text = file_obj.read()
517
ie.text_sha1, ie.text_size = self._add_text_to_weave(
518
ie.file_id, text, heads, nostore_sha)
519
# Let the caller know we generated a stat fingerprint.
520
fingerprint = (ie.text_sha1, stat_value)
521
except errors.ExistingContent:
522
# Turns out that the file content was unchanged, and we were
523
# only going to store a new node if it was changed. Carry over
525
ie.revision = parent_entry.revision
526
ie.text_size = parent_entry.text_size
527
ie.text_sha1 = parent_entry.text_sha1
528
ie.executable = parent_entry.executable
529
return self._get_delta(ie, basis_inv, path), False, None
530
elif kind == 'directory':
532
# all data is meta here, nothing specific to directory, so
534
ie.revision = parent_entry.revision
535
return self._get_delta(ie, basis_inv, path), False, None
536
self._add_text_to_weave(ie.file_id, '', heads, None)
537
elif kind == 'symlink':
538
current_link_target = content_summary[3]
540
# symlink target is not generic metadata, check if it has
542
if current_link_target != parent_entry.symlink_target:
545
# unchanged, carry over.
546
ie.revision = parent_entry.revision
547
ie.symlink_target = parent_entry.symlink_target
548
return self._get_delta(ie, basis_inv, path), False, None
549
ie.symlink_target = current_link_target
550
self._add_text_to_weave(ie.file_id, '', heads, None)
551
elif kind == 'tree-reference':
553
if content_summary[3] != parent_entry.reference_revision:
556
# unchanged, carry over.
557
ie.reference_revision = parent_entry.reference_revision
558
ie.revision = parent_entry.revision
559
return self._get_delta(ie, basis_inv, path), False, None
560
ie.reference_revision = content_summary[3]
561
if ie.reference_revision is None:
562
raise AssertionError("invalid content_summary for nested tree: %r"
563
% (content_summary,))
564
self._add_text_to_weave(ie.file_id, '', heads, None)
566
raise NotImplementedError('unknown kind')
567
ie.revision = self._new_revision_id
568
self._any_changes = True
569
return self._get_delta(ie, basis_inv, path), True, fingerprint
571
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
572
_entry_factory=entry_factory):
573
"""Record a new tree via iter_changes.
575
:param tree: The tree to obtain text contents from for changed objects.
576
:param basis_revision_id: The revision id of the tree the iter_changes
577
has been generated against. Currently assumed to be the same
578
as self.parents[0] - if it is not, errors may occur.
579
:param iter_changes: An iter_changes iterator with the changes to apply
580
to basis_revision_id. The iterator must not include any items with
581
a current kind of None - missing items must be either filtered out
582
or errored-on beefore record_iter_changes sees the item.
583
:param _entry_factory: Private method to bind entry_factory locally for
585
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
588
# Create an inventory delta based on deltas between all the parents and
589
# deltas between all the parent inventories. We use inventory delta's
590
# between the inventory objects because iter_changes masks
591
# last-changed-field only changes.
593
# file_id -> change map, change is fileid, paths, changed, versioneds,
594
# parents, names, kinds, executables
596
# {file_id -> revision_id -> inventory entry, for entries in parent
597
# trees that are not parents[0]
601
revtrees = list(self.repository.revision_trees(self.parents))
602
except errors.NoSuchRevision:
603
# one or more ghosts, slow path.
605
for revision_id in self.parents:
607
revtrees.append(self.repository.revision_tree(revision_id))
608
except errors.NoSuchRevision:
610
basis_revision_id = _mod_revision.NULL_REVISION
612
revtrees.append(self.repository.revision_tree(
613
_mod_revision.NULL_REVISION))
614
# The basis inventory from a repository
616
basis_inv = revtrees[0].inventory
618
basis_inv = self.repository.revision_tree(
619
_mod_revision.NULL_REVISION).inventory
620
if len(self.parents) > 0:
621
if basis_revision_id != self.parents[0] and not ghost_basis:
623
"arbitrary basis parents not yet supported with merges")
624
for revtree in revtrees[1:]:
625
for change in revtree.inventory._make_delta(basis_inv):
626
if change[1] is None:
627
# Not present in this parent.
629
if change[2] not in merged_ids:
630
if change[0] is not None:
631
basis_entry = basis_inv[change[2]]
632
merged_ids[change[2]] = [
634
basis_entry.revision,
637
parent_entries[change[2]] = {
639
basis_entry.revision:basis_entry,
641
change[3].revision:change[3],
644
merged_ids[change[2]] = [change[3].revision]
645
parent_entries[change[2]] = {change[3].revision:change[3]}
647
merged_ids[change[2]].append(change[3].revision)
648
parent_entries[change[2]][change[3].revision] = change[3]
651
# Setup the changes from the tree:
652
# changes maps file_id -> (change, [parent revision_ids])
654
for change in iter_changes:
655
# This probably looks up in basis_inv way to much.
656
if change[1][0] is not None:
657
head_candidate = [basis_inv[change[0]].revision]
660
changes[change[0]] = change, merged_ids.get(change[0],
662
unchanged_merged = set(merged_ids) - set(changes)
663
# Extend the changes dict with synthetic changes to record merges of
665
for file_id in unchanged_merged:
666
# Record a merged version of these items that did not change vs the
667
# basis. This can be either identical parallel changes, or a revert
668
# of a specific file after a merge. The recorded content will be
669
# that of the current tree (which is the same as the basis), but
670
# the per-file graph will reflect a merge.
671
# NB:XXX: We are reconstructing path information we had, this
672
# should be preserved instead.
673
# inv delta change: (file_id, (path_in_source, path_in_target),
674
# changed_content, versioned, parent, name, kind,
677
basis_entry = basis_inv[file_id]
678
except errors.NoSuchId:
679
# a change from basis->some_parents but file_id isn't in basis
680
# so was new in the merge, which means it must have changed
681
# from basis -> current, and as it hasn't the add was reverted
682
# by the user. So we discard this change.
686
(basis_inv.id2path(file_id), tree.id2path(file_id)),
688
(basis_entry.parent_id, basis_entry.parent_id),
689
(basis_entry.name, basis_entry.name),
690
(basis_entry.kind, basis_entry.kind),
691
(basis_entry.executable, basis_entry.executable))
692
changes[file_id] = (change, merged_ids[file_id])
693
# changes contains tuples with the change and a set of inventory
694
# candidates for the file.
696
# old_path, new_path, file_id, new_inventory_entry
697
seen_root = False # Is the root in the basis delta?
698
inv_delta = self._basis_delta
699
modified_rev = self._new_revision_id
700
for change, head_candidates in changes.values():
701
if change[3][1]: # versioned in target.
702
# Several things may be happening here:
703
# We may have a fork in the per-file graph
704
# - record a change with the content from tree
705
# We may have a change against < all trees
706
# - carry over the tree that hasn't changed
707
# We may have a change against all trees
708
# - record the change with the content from tree
711
entry = _entry_factory[kind](file_id, change[5][1],
713
head_set = self._heads(change[0], set(head_candidates))
716
for head_candidate in head_candidates:
717
if head_candidate in head_set:
718
heads.append(head_candidate)
719
head_set.remove(head_candidate)
722
# Could be a carry-over situation:
723
parent_entry_revs = parent_entries.get(file_id, None)
724
if parent_entry_revs:
725
parent_entry = parent_entry_revs.get(heads[0], None)
728
if parent_entry is None:
729
# The parent iter_changes was called against is the one
730
# that is the per-file head, so any change is relevant
731
# iter_changes is valid.
732
carry_over_possible = False
734
# could be a carry over situation
735
# A change against the basis may just indicate a merge,
736
# we need to check the content against the source of the
737
# merge to determine if it was changed after the merge
739
if (parent_entry.kind != entry.kind or
740
parent_entry.parent_id != entry.parent_id or
741
parent_entry.name != entry.name):
742
# Metadata common to all entries has changed
743
# against per-file parent
744
carry_over_possible = False
746
carry_over_possible = True
747
# per-type checks for changes against the parent_entry
750
# Cannot be a carry-over situation
751
carry_over_possible = False
752
# Populate the entry in the delta
754
# XXX: There is still a small race here: If someone reverts the content of a file
755
# after iter_changes examines and decides it has changed,
756
# we will unconditionally record a new version even if some
757
# other process reverts it while commit is running (with
758
# the revert happening after iter_changes did it's
761
entry.executable = True
763
entry.executable = False
764
if (carry_over_possible and
765
parent_entry.executable == entry.executable):
766
# Check the file length, content hash after reading
768
nostore_sha = parent_entry.text_sha1
771
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
773
text = file_obj.read()
777
entry.text_sha1, entry.text_size = self._add_text_to_weave(
778
file_id, text, heads, nostore_sha)
779
yield file_id, change[1][1], (entry.text_sha1, stat_value)
780
except errors.ExistingContent:
781
# No content change against a carry_over parent
782
# Perhaps this should also yield a fs hash update?
784
entry.text_size = parent_entry.text_size
785
entry.text_sha1 = parent_entry.text_sha1
786
elif kind == 'symlink':
788
entry.symlink_target = tree.get_symlink_target(file_id)
789
if (carry_over_possible and
790
parent_entry.symlink_target == entry.symlink_target):
793
self._add_text_to_weave(change[0], '', heads, None)
794
elif kind == 'directory':
795
if carry_over_possible:
798
# Nothing to set on the entry.
799
# XXX: split into the Root and nonRoot versions.
800
if change[1][1] != '' or self.repository.supports_rich_root():
801
self._add_text_to_weave(change[0], '', heads, None)
802
elif kind == 'tree-reference':
803
if not self.repository._format.supports_tree_reference:
804
# This isn't quite sane as an error, but we shouldn't
805
# ever see this code path in practice: tree's don't
806
# permit references when the repo doesn't support tree
808
raise errors.UnsupportedOperation(tree.add_reference,
810
reference_revision = tree.get_reference_revision(change[0])
811
entry.reference_revision = reference_revision
812
if (carry_over_possible and
813
parent_entry.reference_revision == reference_revision):
816
self._add_text_to_weave(change[0], '', heads, None)
818
raise AssertionError('unknown kind %r' % kind)
820
entry.revision = modified_rev
822
entry.revision = parent_entry.revision
825
new_path = change[1][1]
826
inv_delta.append((change[1][0], new_path, change[0], entry))
829
self.new_inventory = None
831
# This should perhaps be guarded by a check that the basis we
832
# commit against is the basis for the commit and if not do a delta
834
self._any_changes = True
836
# housekeeping root entry changes do not affect no-change commits.
837
self._require_root_change(tree)
838
self.basis_delta_revision = basis_revision_id
840
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
841
parent_keys = tuple([(file_id, parent) for parent in parents])
842
return self.repository.texts._add_text(
843
(file_id, self._new_revision_id), parent_keys, new_text,
844
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
847
class RootCommitBuilder(CommitBuilder):
848
"""This commitbuilder actually records the root id"""
850
# the root entry gets versioned properly by this builder.
851
_versioned_root = True
853
def _check_root(self, ie, parent_invs, tree):
854
"""Helper for record_entry_contents.
856
:param ie: An entry being added.
857
:param parent_invs: The inventories of the parent revisions of the
859
:param tree: The tree that is being committed.
862
def _require_root_change(self, tree):
863
"""Enforce an appropriate root object change.
865
This is called once when record_iter_changes is called, if and only if
866
the root was not in the delta calculated by record_iter_changes.
868
:param tree: The tree which is being committed.
870
# versioned roots do not change unless the tree found a change.
873
class RepositoryWriteLockResult(LogicalLockResult):
874
"""The result of write locking a repository.
876
:ivar repository_token: The token obtained from the underlying lock, or
878
:ivar unlock: A callable which will unlock the lock.
881
def __init__(self, unlock, repository_token):
882
LogicalLockResult.__init__(self, unlock)
883
self.repository_token = repository_token
886
return "RepositoryWriteLockResult(%s, %s)" % (self.repository_token,
890
64
######################################################################
894
class Repository(_RelockDebugMixin, bzrdir.ControlComponent):
67
class Repository(object):
895
68
"""Repository holding history for one or more branches.
897
70
The repository holds and retrieves historical information including
898
71
revisions and file history. It's normally accessed only by the Branch,
899
72
which views a particular line of development through that history.
901
The Repository builds on top of some byte storage facilies (the revisions,
902
signatures, inventories, texts and chk_bytes attributes) and a Transport,
903
which respectively provide byte storage and a means to access the (possibly
74
The Repository builds on top of Stores and a Transport, which respectively
75
describe the disk data format and the way of accessing the (possibly
906
The byte storage facilities are addressed via tuples, which we refer to
907
as 'keys' throughout the code base. Revision_keys, inventory_keys and
908
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
909
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
910
byte string made up of a hash identifier and a hash value.
911
We use this interface because it allows low friction with the underlying
912
code that implements disk indices, network encoding and other parts of
915
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
916
the serialised revisions for the repository. This can be used to obtain
917
revision graph information or to access raw serialised revisions.
918
The result of trying to insert data into the repository via this store
919
is undefined: it should be considered read-only except for implementors
921
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
922
the serialised signatures for the repository. This can be used to
923
obtain access to raw serialised signatures. The result of trying to
924
insert data into the repository via this store is undefined: it should
925
be considered read-only except for implementors of repositories.
926
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
927
the serialised inventories for the repository. This can be used to
928
obtain unserialised inventories. The result of trying to insert data
929
into the repository via this store is undefined: it should be
930
considered read-only except for implementors of repositories.
931
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
932
texts of files and directories for the repository. This can be used to
933
obtain file texts or file graphs. Note that Repository.iter_file_bytes
934
is usually a better interface for accessing file texts.
935
The result of trying to insert data into the repository via this store
936
is undefined: it should be considered read-only except for implementors
938
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
939
any data the repository chooses to store or have indexed by its hash.
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 _transport: Transport for file access to repository, typically
944
pointing to .bzr/repository.
947
# What class to use for a CommitBuilder. Often its simpler to change this
948
# in a Repository class subclass rather than to override
949
# get_commit_builder.
950
_commit_builder_class = CommitBuilder
951
# The search regex used by xml based repositories to determine what things
952
# where changed in a single commit.
953
79
_file_ids_altered_regex = lazy_regex.lazy_compile(
954
80
r'file_id="(?P<file_id>[^"]+)"'
955
r'.* revision="(?P<revision_id>[^"]+)"'
81
r'.*revision="(?P<revision_id>[^"]+)"'
958
def abort_write_group(self, suppress_errors=False):
84
def abort_write_group(self):
959
85
"""Commit the contents accrued within the current write group.
961
:param suppress_errors: if true, abort_write_group will catch and log
962
unexpected errors that happen during the abort, rather than
963
allowing them to propagate. Defaults to False.
965
87
:seealso: start_write_group.
967
89
if self._write_group is not self.get_transaction():
968
90
# has an unlock or relock occured ?
971
'(suppressed) mismatched lock context and write group. %r, %r',
972
self._write_group, self.get_transaction())
974
raise errors.BzrError(
975
'mismatched lock context and write group. %r, %r' %
976
(self._write_group, self.get_transaction()))
978
self._abort_write_group()
979
except Exception, exc:
980
self._write_group = None
981
if not suppress_errors:
983
mutter('abort_write_group failed')
984
log_exception_quietly()
985
note('bzr: ERROR (ignored): %s', exc)
91
raise errors.BzrError('mismatched lock context and write group.')
92
self._abort_write_group()
986
93
self._write_group = None
988
95
def _abort_write_group(self):
989
96
"""Template method for per-repository write group cleanup.
991
This is called during abort before the write group is considered to be
98
This is called during abort before the write group is considered to be
992
99
finished and should cleanup any internal state accrued during the write
993
100
group. There is no requirement that data handed to the repository be
994
101
*not* made available - this is not a rollback - but neither should any
1190
218
Returns a set of the present revisions.
1193
graph = self.get_graph()
1194
parent_map = graph.get_parent_map(revision_ids)
1195
# The old API returned a list, should this actually be a set?
1196
return parent_map.keys()
1198
def _check_inventories(self, checker):
1199
"""Check the inventories found from the revision scan.
1201
This is responsible for verifying the sha1 of inventories and
1202
creating a pending_keys set that covers data referenced by inventories.
1204
bar = ui.ui_factory.nested_progress_bar()
1206
self._do_check_inventories(checker, bar)
1210
def _do_check_inventories(self, checker, bar):
1211
"""Helper for _check_inventories."""
1213
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1214
kinds = ['chk_bytes', 'texts']
1215
count = len(checker.pending_keys)
1216
bar.update("inventories", 0, 2)
1217
current_keys = checker.pending_keys
1218
checker.pending_keys = {}
1219
# Accumulate current checks.
1220
for key in current_keys:
1221
if key[0] != 'inventories' and key[0] not in kinds:
1222
checker._report_items.append('unknown key type %r' % (key,))
1223
keys[key[0]].add(key[1:])
1224
if keys['inventories']:
1225
# NB: output order *should* be roughly sorted - topo or
1226
# inverse topo depending on repository - either way decent
1227
# to just delta against. However, pre-CHK formats didn't
1228
# try to optimise inventory layout on disk. As such the
1229
# pre-CHK code path does not use inventory deltas.
1231
for record in self.inventories.check(keys=keys['inventories']):
1232
if record.storage_kind == 'absent':
1233
checker._report_items.append(
1234
'Missing inventory {%s}' % (record.key,))
1236
last_object = self._check_record('inventories', record,
1237
checker, last_object,
1238
current_keys[('inventories',) + record.key])
1239
del keys['inventories']
1242
bar.update("texts", 1)
1243
while (checker.pending_keys or keys['chk_bytes']
1245
# Something to check.
1246
current_keys = checker.pending_keys
1247
checker.pending_keys = {}
1248
# Accumulate current checks.
1249
for key in current_keys:
1250
if key[0] not in kinds:
1251
checker._report_items.append('unknown key type %r' % (key,))
1252
keys[key[0]].add(key[1:])
1253
# Check the outermost kind only - inventories || chk_bytes || texts
1257
for record in getattr(self, kind).check(keys=keys[kind]):
1258
if record.storage_kind == 'absent':
1259
checker._report_items.append(
1260
'Missing %s {%s}' % (kind, record.key,))
1262
last_object = self._check_record(kind, record,
1263
checker, last_object, current_keys[(kind,) + record.key])
1267
def _check_record(self, kind, record, checker, last_object, item_data):
1268
"""Check a single text from this repository."""
1269
if kind == 'inventories':
1270
rev_id = record.key[0]
1271
inv = self._deserialise_inventory(rev_id,
1272
record.get_bytes_as('fulltext'))
1273
if last_object is not None:
1274
delta = inv._make_delta(last_object)
1275
for old_path, path, file_id, ie in delta:
1278
ie.check(checker, rev_id, inv)
1280
for path, ie in inv.iter_entries():
1281
ie.check(checker, rev_id, inv)
1282
if self._format.fast_deltas:
1284
elif kind == 'chk_bytes':
1285
# No code written to check chk_bytes for this repo format.
1286
checker._report_items.append(
1287
'unsupported key type chk_bytes for %s' % (record.key,))
1288
elif kind == 'texts':
1289
self._check_text(record, checker, item_data)
1291
checker._report_items.append(
1292
'unknown key type %s for %s' % (kind, record.key))
1294
def _check_text(self, record, checker, item_data):
1295
"""Check a single text."""
1296
# Check it is extractable.
1297
# TODO: check length.
1298
if record.storage_kind == 'chunked':
1299
chunks = record.get_bytes_as(record.storage_kind)
1300
sha1 = osutils.sha_strings(chunks)
1301
length = sum(map(len, chunks))
1303
content = record.get_bytes_as('fulltext')
1304
sha1 = osutils.sha_string(content)
1305
length = len(content)
1306
if item_data and sha1 != item_data[1]:
1307
checker._report_items.append(
1308
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1309
(record.key, sha1, item_data[1], item_data[2]))
221
for id in revision_ids:
222
if self.has_revision(id):
1312
227
def create(a_bzrdir):
1313
228
"""Construct the current default format repository in a_bzrdir."""
1314
229
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1316
def __init__(self, _format, a_bzrdir, control_files):
231
def __init__(self, _format, a_bzrdir, control_files, _revision_store, control_store, text_store):
1317
232
"""instantiate a Repository.
1319
234
:param _format: The format of the repository on disk.
1320
235
:param a_bzrdir: The BzrDir of the repository.
237
In the future we will have a single api for all stores for
238
getting file texts, inventories and revisions, then
239
this construct will accept instances of those things.
1322
# In the future we will have a single api for all stores for
1323
# getting file texts, inventories and revisions, then
1324
# this construct will accept instances of those things.
1325
241
super(Repository, self).__init__()
1326
242
self._format = _format
1327
243
# the following are part of the public API for Repository:
1328
244
self.bzrdir = a_bzrdir
1329
245
self.control_files = control_files
1330
self._transport = control_files._transport
1331
self.base = self._transport.base
246
self._revision_store = _revision_store
247
# backwards compatibility
248
self.weave_store = text_store
1333
250
self._reconcile_does_inventory_gc = True
1334
self._reconcile_fixes_text_parents = False
1335
self._reconcile_backsup_inventory = True
251
# not right yet - should be more semantically clear ?
253
self.control_store = control_store
254
self.control_weaves = control_store
255
# TODO: make sure to construct the right store classes, etc, depending
256
# on whether escaping is required.
257
self._warn_if_deprecated()
1336
258
self._write_group = None
1337
# Additional places to query for data.
1338
self._fallback_repositories = []
1339
# An InventoryEntry cache, used during deserialization
1340
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1341
# Is it safe to return inventory entries directly from the entry cache,
1342
# rather copying them?
1343
self._safe_to_return_from_cache = False
1346
def user_transport(self):
1347
return self.bzrdir.user_transport
1350
def control_transport(self):
1351
return self._transport
1353
260
def __repr__(self):
1354
if self._fallback_repositories:
1355
return '%s(%r, fallback_repositories=%r)' % (
1356
self.__class__.__name__,
1358
self._fallback_repositories)
1360
return '%s(%r)' % (self.__class__.__name__,
1363
def _has_same_fallbacks(self, other_repo):
1364
"""Returns true if the repositories have the same fallbacks."""
1365
my_fb = self._fallback_repositories
1366
other_fb = other_repo._fallback_repositories
1367
if len(my_fb) != len(other_fb):
1369
for f, g in zip(my_fb, other_fb):
1370
if not f.has_same_location(g):
261
return '%s(%r)' % (self.__class__.__name__,
262
self.bzrdir.transport.base)
1374
264
def has_same_location(self, other):
1375
265
"""Returns a boolean indicating if this repository is at the same
1579
406
def copy_content_into(self, destination, revision_id=None):
1580
407
"""Make a complete copy of the content in self into destination.
1582
This is a destructive operation! Do not use it on existing
409
This is a destructive operation! Do not use it on existing
412
revision_id = osutils.safe_revision_id(revision_id)
1585
413
return InterRepository.get(self, destination).copy_content(revision_id)
1587
415
def commit_write_group(self):
1588
416
"""Commit the contents accrued within the current write group.
1590
418
:seealso: start_write_group.
1592
:return: it may return an opaque hint that can be passed to 'pack'.
1594
420
if self._write_group is not self.get_transaction():
1595
421
# has an unlock or relock occured ?
1596
raise errors.BzrError('mismatched lock context %r and '
1598
(self.get_transaction(), self._write_group))
1599
result = self._commit_write_group()
422
raise errors.BzrError('mismatched lock context and write group.')
423
self._commit_write_group()
1600
424
self._write_group = None
1603
426
def _commit_write_group(self):
1604
427
"""Template method for per-repository write group cleanup.
1606
This is called before the write group is considered to be
429
This is called before the write group is considered to be
1607
430
finished and should ensure that all data handed to the repository
1608
for writing during the write group is safely committed (to the
431
for writing during the write group is safely committed (to the
1609
432
extent possible considering file system caching etc).
1612
def suspend_write_group(self):
1613
raise errors.UnsuspendableWriteGroup(self)
1615
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1616
"""Return the keys of missing inventory parents for revisions added in
1619
A revision is not complete if the inventory delta for that revision
1620
cannot be calculated. Therefore if the parent inventories of a
1621
revision are not present, the revision is incomplete, and e.g. cannot
1622
be streamed by a smart server. This method finds missing inventory
1623
parents for revisions added in this write group.
1625
if not self._format.supports_external_lookups:
1626
# This is only an issue for stacked repositories
1628
if not self.is_in_write_group():
1629
raise AssertionError('not in a write group')
1631
# XXX: We assume that every added revision already has its
1632
# corresponding inventory, so we only check for parent inventories that
1633
# might be missing, rather than all inventories.
1634
parents = set(self.revisions._index.get_missing_parents())
1635
parents.discard(_mod_revision.NULL_REVISION)
1636
unstacked_inventories = self.inventories._index
1637
present_inventories = unstacked_inventories.get_parent_map(
1638
key[-1:] for key in parents)
1639
parents.difference_update(present_inventories)
1640
if len(parents) == 0:
1641
# No missing parent inventories.
1643
if not check_for_missing_texts:
1644
return set(('inventories', rev_id) for (rev_id,) in parents)
1645
# Ok, now we have a list of missing inventories. But these only matter
1646
# if the inventories that reference them are missing some texts they
1647
# appear to introduce.
1648
# XXX: Texts referenced by all added inventories need to be present,
1649
# but at the moment we're only checking for texts referenced by
1650
# inventories at the graph's edge.
1651
key_deps = self.revisions._index._key_dependencies
1652
key_deps.satisfy_refs_for_keys(present_inventories)
1653
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1654
file_ids = self.fileids_altered_by_revision_ids(referrers)
1655
missing_texts = set()
1656
for file_id, version_ids in file_ids.iteritems():
1657
missing_texts.update(
1658
(file_id, version_id) for version_id in version_ids)
1659
present_texts = self.texts.get_parent_map(missing_texts)
1660
missing_texts.difference_update(present_texts)
1661
if not missing_texts:
1662
# No texts are missing, so all revisions and their deltas are
1665
# Alternatively the text versions could be returned as the missing
1666
# keys, but this is likely to be less data.
1667
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1670
def refresh_data(self):
1671
"""Re-read any data needed to synchronise with disk.
1673
This method is intended to be called after another repository instance
1674
(such as one used by a smart server) has inserted data into the
1675
repository. On all repositories this will work outside of write groups.
1676
Some repository formats (pack and newer for bzrlib native formats)
1677
support refresh_data inside write groups. If called inside a write
1678
group on a repository that does not support refreshing in a write group
1679
IsInWriteGroupError will be raised.
1681
self._refresh_data()
1683
def resume_write_group(self, tokens):
1684
if not self.is_write_locked():
1685
raise errors.NotWriteLocked(self)
1686
if self._write_group:
1687
raise errors.BzrError('already in a write group')
1688
self._resume_write_group(tokens)
1689
# so we can detect unlock/relock - the write group is now entered.
1690
self._write_group = self.get_transaction()
1692
def _resume_write_group(self, tokens):
1693
raise errors.UnsuspendableWriteGroup(self)
1695
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
435
def fetch(self, source, revision_id=None, pb=None):
1697
436
"""Fetch the content required to construct revision_id from source.
1699
If revision_id is None and fetch_spec is None, then all content is
1702
fetch() may not be used when the repository is in a write group -
1703
either finish the current write group before using fetch, or use
1704
fetch before starting the write group.
1706
:param find_ghosts: Find and copy revisions in the source that are
1707
ghosts in the target (and not reachable directly by walking out to
1708
the first-present revision in target from revision_id).
1709
:param revision_id: If specified, all the content needed for this
1710
revision ID will be copied to the target. Fetch will determine for
1711
itself which content needs to be copied.
1712
:param fetch_spec: If specified, a SearchResult or
1713
PendingAncestryResult that describes which revisions to copy. This
1714
allows copying multiple heads at once. Mutually exclusive with
438
If revision_id is None all content is copied.
1717
if fetch_spec is not None and revision_id is not None:
1718
raise AssertionError(
1719
"fetch_spec and revision_id are mutually exclusive.")
1720
if self.is_in_write_group():
1721
raise errors.InternalBzrError(
1722
"May not fetch while in a write group.")
1723
# fast path same-url fetch operations
1724
# TODO: lift out to somewhere common with RemoteRepository
1725
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1726
if (self.has_same_location(source)
1727
and fetch_spec is None
1728
and self._has_same_fallbacks(source)):
1729
# check that last_revision is in 'from' and then return a
1731
if (revision_id is not None and
1732
not _mod_revision.is_null(revision_id)):
1733
self.get_revision(revision_id)
1735
# if there is no specific appropriate InterRepository, this will get
1736
# the InterRepository base class, which raises an
1737
# IncompatibleRepositories when asked to fetch.
440
revision_id = osutils.safe_revision_id(revision_id)
1738
441
inter = InterRepository.get(source, self)
1739
return inter.fetch(revision_id=revision_id, pb=pb,
1740
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
443
return inter.fetch(revision_id=revision_id, pb=pb)
444
except NotImplementedError:
445
raise errors.IncompatibleRepositories(source, self)
1742
447
def create_bundle(self, target, base, fileobj, format=None):
1743
448
return serializer.write_bundle(self, target, base, fileobj, format)
1745
def get_commit_builder(self, branch, parents, config, timestamp=None,
1746
timezone=None, committer=None, revprops=None,
450
def get_commit_builder(self, branch, parents, config, timestamp=None,
451
timezone=None, committer=None, revprops=None,
1747
452
revision_id=None):
1748
453
"""Obtain a CommitBuilder for this repository.
1750
455
:param branch: Branch to commit to.
1751
456
:param parents: Revision ids of the parents of the new revision.
1752
457
:param config: Configuration to use.
1847
542
dest_repo = a_bzrdir.open_repository()
1848
543
return dest_repo
1850
def _get_sink(self):
1851
"""Return a sink for streaming into this repository."""
1852
return StreamSink(self)
1854
def _get_source(self, to_format):
1855
"""Return a source for streaming from this repository."""
1856
return StreamSource(self, to_format)
1858
545
@needs_read_lock
1859
546
def has_revision(self, revision_id):
1860
547
"""True if this repository has a copy of the revision."""
1861
return revision_id in self.has_revisions((revision_id,))
1864
def has_revisions(self, revision_ids):
1865
"""Probe to find out the presence of multiple revisions.
1867
:param revision_ids: An iterable of revision_ids.
1868
:return: A set of the revision_ids that were present.
1870
parent_map = self.revisions.get_parent_map(
1871
[(rev_id,) for rev_id in revision_ids])
1873
if _mod_revision.NULL_REVISION in revision_ids:
1874
result.add(_mod_revision.NULL_REVISION)
1875
result.update([key[0] for key in parent_map])
1879
def get_revision(self, revision_id):
1880
"""Return the Revision object for a named revision."""
1881
return self.get_revisions([revision_id])[0]
548
if 'evil' in debug.debug_flags:
549
mutter_callsite(2, "has_revision is a LBYL symptom.")
550
revision_id = osutils.safe_revision_id(revision_id)
551
return self._revision_store.has_revision_id(revision_id,
552
self.get_transaction())
1883
554
@needs_read_lock
1884
555
def get_revision_reconcile(self, revision_id):
1885
556
"""'reconcile' helper routine that allows access to a revision always.
1887
558
This variant of get_revision does not cross check the weave graph
1888
559
against the revision one as get_revision does: but it should only
1889
560
be used by reconcile, or reconcile-alike commands that are correcting
1890
561
or testing the revision graph.
1892
return self._get_revisions([revision_id])[0]
563
if not revision_id or not isinstance(revision_id, basestring):
564
raise errors.InvalidRevisionId(revision_id=revision_id,
566
return self.get_revisions([revision_id])[0]
1894
568
@needs_read_lock
1895
569
def get_revisions(self, revision_ids):
1896
"""Get many revisions at once.
1898
Repositories that need to check data on every revision read should
1899
subclass this method.
1901
return self._get_revisions(revision_ids)
1904
def _get_revisions(self, revision_ids):
1905
"""Core work logic to get many revisions without sanity checks."""
1907
for revid, rev in self._iter_revisions(revision_ids):
1909
raise errors.NoSuchRevision(self, revid)
1911
return [revs[revid] for revid in revision_ids]
1913
def _iter_revisions(self, revision_ids):
1914
"""Iterate over revision objects.
1916
:param revision_ids: An iterable of revisions to examine. None may be
1917
passed to request all revisions known to the repository. Note that
1918
not all repositories can find unreferenced revisions; for those
1919
repositories only referenced ones will be returned.
1920
:return: An iterator of (revid, revision) tuples. Absent revisions (
1921
those asked for but not available) are returned as (revid, None).
1923
if revision_ids is None:
1924
revision_ids = self.all_revision_ids()
1926
for rev_id in revision_ids:
1927
if not rev_id or not isinstance(rev_id, basestring):
1928
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1929
keys = [(key,) for key in revision_ids]
1930
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1931
for record in stream:
1932
revid = record.key[0]
1933
if record.storage_kind == 'absent':
1936
text = record.get_bytes_as('fulltext')
1937
rev = self._serializer.read_revision_from_string(text)
1940
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
570
revision_ids = [osutils.safe_revision_id(r) for r in revision_ids]
571
revs = self._revision_store.get_revisions(revision_ids,
572
self.get_transaction())
574
assert not isinstance(rev.revision_id, unicode)
575
for parent_id in rev.parent_ids:
576
assert not isinstance(parent_id, unicode)
580
def get_revision_xml(self, revision_id):
581
# TODO: jam 20070210 This shouldn't be necessary since get_revision
582
# would have already do it.
583
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
584
revision_id = osutils.safe_revision_id(revision_id)
585
rev = self.get_revision(revision_id)
587
# the current serializer..
588
self._revision_store._serializer.write_revision(rev, rev_tmp)
590
return rev_tmp.getvalue()
593
def get_revision(self, revision_id):
594
"""Return the Revision object for a named revision"""
595
# TODO: jam 20070210 get_revision_reconcile should do this for us
596
revision_id = osutils.safe_revision_id(revision_id)
597
r = self.get_revision_reconcile(revision_id)
598
# weave corruption can lead to absent revision markers that should be
600
# the following test is reasonably cheap (it needs a single weave read)
601
# and the weave is cached in read transactions. In write transactions
602
# it is not cached but typically we only read a small number of
603
# revisions. For knits when they are introduced we will probably want
604
# to ensure that caching write transactions are in use.
605
inv = self.get_inventory_weave()
606
self._check_revision_parents(r, inv)
610
def get_deltas_for_revisions(self, revisions):
1941
611
"""Produce a generator of revision deltas.
1943
613
Note that the input is a sequence of REVISIONS, not revision_ids.
1944
614
Trees will be held in memory until the generator exits.
1945
615
Each delta is relative to the revision's lefthand predecessor.
1947
:param specific_fileids: if not None, the result is filtered
1948
so that only those file-ids, their parents and their
1949
children are included.
1951
# Get the revision-ids of interest
1952
617
required_trees = set()
1953
618
for revision in revisions:
1954
619
required_trees.add(revision.revision_id)
1955
620
required_trees.update(revision.parent_ids[:1])
1957
# Get the matching filtered trees. Note that it's more
1958
# efficient to pass filtered trees to changes_from() rather
1959
# than doing the filtering afterwards. changes_from() could
1960
# arguably do the filtering itself but it's path-based, not
1961
# file-id based, so filtering before or afterwards is
1963
if specific_fileids is None:
1964
trees = dict((t.get_revision_id(), t) for
1965
t in self.revision_trees(required_trees))
1967
trees = dict((t.get_revision_id(), t) for
1968
t in self._filtered_revision_trees(required_trees,
1971
# Calculate the deltas
621
trees = dict((t.get_revision_id(), t) for
622
t in self.revision_trees(required_trees))
1972
623
for revision in revisions:
1973
624
if not revision.parent_ids:
1974
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
625
old_tree = self.revision_tree(None)
1976
627
old_tree = trees[revision.parent_ids[0]]
1977
628
yield trees[revision.revision_id].changes_from(old_tree)
1979
630
@needs_read_lock
1980
def get_revision_delta(self, revision_id, specific_fileids=None):
631
def get_revision_delta(self, revision_id):
1981
632
"""Return the delta for one revision.
1983
634
The delta is relative to the left-hand predecessor of the
1986
:param specific_fileids: if not None, the result is filtered
1987
so that only those file-ids, their parents and their
1988
children are included.
1990
637
r = self.get_revision(revision_id)
1991
return list(self.get_deltas_for_revisions([r],
1992
specific_fileids=specific_fileids))[0]
638
return list(self.get_deltas_for_revisions([r]))[0]
640
def _check_revision_parents(self, revision, inventory):
641
"""Private to Repository and Fetch.
643
This checks the parentage of revision in an inventory weave for
644
consistency and is only applicable to inventory-weave-for-ancestry
645
using repository formats & fetchers.
647
weave_parents = inventory.get_parents(revision.revision_id)
648
weave_names = inventory.versions()
649
for parent_id in revision.parent_ids:
650
if parent_id in weave_names:
651
# this parent must not be a ghost.
652
if not parent_id in weave_parents:
654
raise errors.CorruptRepository(self)
1994
656
@needs_write_lock
1995
657
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
658
revision_id = osutils.safe_revision_id(revision_id)
1996
659
signature = gpg_strategy.sign(plaintext)
1997
self.add_signature_text(revision_id, signature)
2000
def add_signature_text(self, revision_id, signature):
2001
self.signatures.add_lines((revision_id,), (),
2002
osutils.split_lines(signature))
2004
def find_text_key_references(self):
2005
"""Find the text key references within the repository.
2007
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2008
to whether they were referred to by the inventory of the
2009
revision_id that they contain. The inventory texts from all present
2010
revision ids are assessed to generate this report.
2012
revision_keys = self.revisions.keys()
2013
w = self.inventories
2014
pb = ui.ui_factory.nested_progress_bar()
2016
return self._find_text_key_references_from_xml_inventory_lines(
2017
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2021
def _find_text_key_references_from_xml_inventory_lines(self,
2023
"""Core routine for extracting references to texts from inventories.
2025
This performs the translation of xml lines to revision ids.
2027
:param line_iterator: An iterator of lines, origin_version_id
2028
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2029
to whether they were referred to by the inventory of the
2030
revision_id that they contain. Note that if that revision_id was
2031
not part of the line_iterator's output then False will be given -
2032
even though it may actually refer to that key.
2034
if not self._serializer.support_altered_by_hack:
2035
raise AssertionError(
2036
"_find_text_key_references_from_xml_inventory_lines only "
2037
"supported for branches which store inventory as unnested xml"
2038
", not on %r" % self)
660
self._revision_store.add_revision_signature_text(revision_id,
662
self.get_transaction())
664
def fileids_altered_by_revision_ids(self, revision_ids):
665
"""Find the file ids and versions affected by revisions.
667
:param revisions: an iterable containing revision ids.
668
:return: a dictionary mapping altered file-ids to an iterable of
669
revision_ids. Each altered file-ids has the exact revision_ids that
670
altered it listed explicitly.
672
assert self._serializer.support_altered_by_hack, \
673
("fileids_altered_by_revision_ids only supported for branches "
674
"which store inventory as unnested xml, not on %r" % self)
675
selected_revision_ids = set(osutils.safe_revision_id(r)
676
for r in revision_ids)
677
w = self.get_inventory_weave()
2041
680
# this code needs to read every new line in every inventory for the
2042
681
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2043
# not present in one of those inventories is unnecessary but not
682
# not present in one of those inventories is unnecessary but not
2044
683
# harmful because we are filtering by the revision id marker in the
2045
# inventory lines : we only select file ids altered in one of those
684
# inventory lines : we only select file ids altered in one of those
2046
685
# revisions. We don't need to see all lines in the inventory because
2047
686
# only those added in an inventory in rev X can contain a revision=X
2334
773
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2335
774
'revisions'. file-id is None unless knit-kind is 'file'.
2337
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2340
for result in self._find_non_file_keys_to_fetch(revision_ids):
2343
def _find_file_keys_to_fetch(self, revision_ids, pb):
2344
776
# XXX: it's a bit weird to control the inventory weave caching in this
2345
777
# generator. Ideally the caching would be done in fetch.py I think. Or
2346
778
# maybe this generator should explicitly have the contract that it
2347
779
# should not be iterated until the previously yielded item has been
2349
inv_w = self.inventories
781
inv_w = self.get_inventory_weave()
2351
784
# file ids that changed
2352
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
785
file_ids = self.fileids_altered_by_revision_ids(revision_ids)
2354
787
num_file_ids = len(file_ids)
2355
788
for file_id, altered_versions in file_ids.iteritems():
2357
pb.update("Fetch texts", count, num_file_ids)
789
if _files_pb is not None:
790
_files_pb.update("fetch texts", count, num_file_ids)
2359
792
yield ("file", file_id, altered_versions)
793
# We're done with the files_pb. Note that it finished by the caller,
794
# just as it was created by the caller.
2361
def _find_non_file_keys_to_fetch(self, revision_ids):
2363
798
yield ("inventory", None, revision_ids)
2366
# XXX: Note ATM no callers actually pay attention to this return
2367
# instead they just use the list of revision ids and ignore
2368
# missing sigs. Consider removing this work entirely
2369
revisions_with_signatures = set(self.signatures.get_parent_map(
2370
[(r,) for r in revision_ids]))
2371
revisions_with_signatures = set(
2372
[r for (r,) in revisions_with_signatures])
2373
revisions_with_signatures.intersection_update(revision_ids)
802
revisions_with_signatures = set()
803
for rev_id in revision_ids:
805
self.get_signature_text(rev_id)
806
except errors.NoSuchRevision:
810
revisions_with_signatures.add(rev_id)
2374
811
yield ("signatures", None, revisions_with_signatures)
2377
814
yield ("revisions", None, revision_ids)
2379
816
@needs_read_lock
817
def get_inventory_weave(self):
818
return self.control_weaves.get_weave('inventory',
819
self.get_transaction())
2380
822
def get_inventory(self, revision_id):
2381
"""Get Inventory object by revision id."""
2382
return self.iter_inventories([revision_id]).next()
2384
def iter_inventories(self, revision_ids, ordering=None):
2385
"""Get many inventories by revision_ids.
2387
This will buffer some or all of the texts used in constructing the
2388
inventories in memory, but will only parse a single inventory at a
2391
:param revision_ids: The expected revision ids of the inventories.
2392
:param ordering: optional ordering, e.g. 'topological'. If not
2393
specified, the order of revision_ids will be preserved (by
2394
buffering if necessary).
2395
:return: An iterator of inventories.
2397
if ((None in revision_ids)
2398
or (_mod_revision.NULL_REVISION in revision_ids)):
2399
raise ValueError('cannot get null revision inventory')
2400
return self._iter_inventories(revision_ids, ordering)
2402
def _iter_inventories(self, revision_ids, ordering):
2403
"""single-document based inventory iteration."""
2404
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2405
for text, revision_id in inv_xmls:
2406
yield self._deserialise_inventory(revision_id, text)
2408
def _iter_inventory_xmls(self, revision_ids, ordering):
2409
if ordering is None:
2410
order_as_requested = True
2411
ordering = 'unordered'
2413
order_as_requested = False
2414
keys = [(revision_id,) for revision_id in revision_ids]
2417
if order_as_requested:
2418
key_iter = iter(keys)
2419
next_key = key_iter.next()
2420
stream = self.inventories.get_record_stream(keys, ordering, True)
2422
for record in stream:
2423
if record.storage_kind != 'absent':
2424
chunks = record.get_bytes_as('chunked')
2425
if order_as_requested:
2426
text_chunks[record.key] = chunks
2428
yield ''.join(chunks), record.key[-1]
2430
raise errors.NoSuchRevision(self, record.key)
2431
if order_as_requested:
2432
# Yield as many results as we can while preserving order.
2433
while next_key in text_chunks:
2434
chunks = text_chunks.pop(next_key)
2435
yield ''.join(chunks), next_key[-1]
2437
next_key = key_iter.next()
2438
except StopIteration:
2439
# We still want to fully consume the get_record_stream,
2440
# just in case it is not actually finished at this point
2444
def _deserialise_inventory(self, revision_id, xml):
2445
"""Transform the xml into an inventory object.
823
"""Get Inventory object by hash."""
824
# TODO: jam 20070210 Technically we don't need to sanitize, since all
825
# called functions must sanitize.
826
revision_id = osutils.safe_revision_id(revision_id)
827
return self.deserialise_inventory(
828
revision_id, self.get_inventory_xml(revision_id))
830
def deserialise_inventory(self, revision_id, xml):
831
"""Transform the xml into an inventory object.
2447
833
:param revision_id: The expected revision id of the inventory.
2448
834
:param xml: A serialised inventory.
2450
result = self._serializer.read_inventory_from_string(xml, revision_id,
2451
entry_cache=self._inventory_entry_cache,
2452
return_from_cache=self._safe_to_return_from_cache)
2453
if result.revision_id != revision_id:
2454
raise AssertionError('revision id mismatch %s != %s' % (
2455
result.revision_id, revision_id))
836
revision_id = osutils.safe_revision_id(revision_id)
837
result = self._serializer.read_inventory_from_string(xml)
838
result.root.revision = revision_id
841
def serialise_inventory(self, inv):
842
return self._serializer.write_inventory_to_string(inv)
2458
844
def get_serializer_format(self):
2459
845
return self._serializer.format_num
2461
847
@needs_read_lock
2462
def _get_inventory_xml(self, revision_id):
2463
"""Get serialized inventory as a string."""
2464
texts = self._iter_inventory_xmls([revision_id], 'unordered')
848
def get_inventory_xml(self, revision_id):
849
"""Get inventory XML as a file object."""
850
revision_id = osutils.safe_revision_id(revision_id)
2466
text, revision_id = texts.next()
2467
except StopIteration:
852
assert isinstance(revision_id, str), type(revision_id)
853
iw = self.get_inventory_weave()
854
return iw.get_text(revision_id)
2468
856
raise errors.HistoryMissing(self, 'inventory', revision_id)
2471
def get_rev_id_for_revno(self, revno, known_pair):
2472
"""Return the revision id of a revno, given a later (revno, revid)
2473
pair in the same history.
2475
:return: if found (True, revid). If the available history ran out
2476
before reaching the revno, then this returns
2477
(False, (closest_revno, closest_revid)).
2479
known_revno, known_revid = known_pair
2480
partial_history = [known_revid]
2481
distance_from_known = known_revno - revno
2482
if distance_from_known < 0:
2484
'requested revno (%d) is later than given known revno (%d)'
2485
% (revno, known_revno))
2488
self, partial_history, stop_index=distance_from_known)
2489
except errors.RevisionNotPresent, err:
2490
if err.revision_id == known_revid:
2491
# The start revision (known_revid) wasn't found.
2493
# This is a stacked repository with no fallbacks, or a there's a
2494
# left-hand ghost. Either way, even though the revision named in
2495
# the error isn't in this repo, we know it's the next step in this
2496
# left-hand history.
2497
partial_history.append(err.revision_id)
2498
if len(partial_history) <= distance_from_known:
2499
# Didn't find enough history to get a revid for the revno.
2500
earliest_revno = known_revno - len(partial_history) + 1
2501
return (False, (earliest_revno, partial_history[-1]))
2502
if len(partial_history) - 1 > distance_from_known:
2503
raise AssertionError('_iter_for_revno returned too much history')
2504
return (True, partial_history[-1])
859
def get_inventory_sha1(self, revision_id):
860
"""Return the sha1 hash of the inventory entry
862
# TODO: jam 20070210 Shouldn't this be deprecated / removed?
863
revision_id = osutils.safe_revision_id(revision_id)
864
return self.get_revision(revision_id).inventory_sha1
867
def get_revision_graph(self, revision_id=None):
868
"""Return a dictionary containing the revision graph.
870
:param revision_id: The revision_id to get a graph from. If None, then
871
the entire revision graph is returned. This is a deprecated mode of
872
operation and will be removed in the future.
873
:return: a dictionary of revision_id->revision_parents_list.
875
if 'evil' in debug.debug_flags:
877
"get_revision_graph scales with size of history.")
878
# special case NULL_REVISION
879
if revision_id == _mod_revision.NULL_REVISION:
881
revision_id = osutils.safe_revision_id(revision_id)
882
a_weave = self.get_inventory_weave()
883
all_revisions = self._eliminate_revisions_not_present(
885
entire_graph = dict([(node, tuple(a_weave.get_parents(node))) for
886
node in all_revisions])
887
if revision_id is None:
889
elif revision_id not in entire_graph:
890
raise errors.NoSuchRevision(self, revision_id)
892
# add what can be reached from revision_id
894
pending = set([revision_id])
895
while len(pending) > 0:
897
result[node] = entire_graph[node]
898
for revision_id in result[node]:
899
if revision_id not in result:
900
pending.add(revision_id)
904
def get_revision_graph_with_ghosts(self, revision_ids=None):
905
"""Return a graph of the revisions with ghosts marked as applicable.
907
:param revision_ids: an iterable of revisions to graph or None for all.
908
:return: a Graph object with the graph reachable from revision_ids.
910
if 'evil' in debug.debug_flags:
912
"get_revision_graph_with_ghosts scales with size of history.")
913
result = deprecated_graph.Graph()
915
pending = set(self.all_revision_ids())
918
pending = set(osutils.safe_revision_id(r) for r in revision_ids)
919
# special case NULL_REVISION
920
if _mod_revision.NULL_REVISION in pending:
921
pending.remove(_mod_revision.NULL_REVISION)
922
required = set(pending)
925
revision_id = pending.pop()
927
rev = self.get_revision(revision_id)
928
except errors.NoSuchRevision:
929
if revision_id in required:
932
result.add_ghost(revision_id)
934
for parent_id in rev.parent_ids:
935
# is this queued or done ?
936
if (parent_id not in pending and
937
parent_id not in done):
939
pending.add(parent_id)
940
result.add_node(revision_id, rev.parent_ids)
941
done.add(revision_id)
944
def _get_history_vf(self):
945
"""Get a versionedfile whose history graph reflects all revisions.
947
For weave repositories, this is the inventory weave.
949
return self.get_inventory_weave()
2506
951
def iter_reverse_revision_history(self, revision_id):
2507
952
"""Iterate backwards through revision ids in the lefthand history
2543
1003
for repositories to maintain loaded indices across multiple locks
2544
1004
by checking inside their implementation of this method to see
2545
1005
whether their indices are still valid. This depends of course on
2546
the disk format being validatable in this manner. This method is
2547
also called by the refresh_data() public interface to cause a refresh
2548
to occur while in a write lock so that data inserted by a smart server
2549
push operation is visible on the client's instance of the physical
1006
the disk format being validatable in this manner.
2553
1009
@needs_read_lock
2554
1010
def revision_tree(self, revision_id):
2555
1011
"""Return Tree for a revision on this branch.
2557
`revision_id` may be NULL_REVISION for the empty tree revision.
1013
`revision_id` may be None for the empty tree revision.
2559
revision_id = _mod_revision.ensure_null(revision_id)
2560
1015
# TODO: refactor this to use an existing revision object
2561
1016
# so we don't need to read it in twice.
2562
if revision_id == _mod_revision.NULL_REVISION:
2563
return RevisionTree(self, Inventory(root_id=None),
1017
if revision_id is None or revision_id == _mod_revision.NULL_REVISION:
1018
return RevisionTree(self, Inventory(root_id=None),
2564
1019
_mod_revision.NULL_REVISION)
2566
inv = self.get_inventory(revision_id)
1021
revision_id = osutils.safe_revision_id(revision_id)
1022
inv = self.get_revision_inventory(revision_id)
2567
1023
return RevisionTree(self, inv, revision_id)
2569
1026
def revision_trees(self, revision_ids):
2570
"""Return Trees for revisions in this repository.
2572
:param revision_ids: a sequence of revision-ids;
2573
a revision-id may not be None or 'null:'
2575
inventories = self.iter_inventories(revision_ids)
2576
for inv in inventories:
2577
yield RevisionTree(self, inv, inv.revision_id)
2579
def _filtered_revision_trees(self, revision_ids, file_ids):
2580
"""Return Tree for a revision on this branch with only some files.
2582
:param revision_ids: a sequence of revision-ids;
2583
a revision-id may not be None or 'null:'
2584
:param file_ids: if not None, the result is filtered
2585
so that only those file-ids, their parents and their
2586
children are included.
2588
inventories = self.iter_inventories(revision_ids)
2589
for inv in inventories:
2590
# Should we introduce a FilteredRevisionTree class rather
2591
# than pre-filter the inventory here?
2592
filtered_inv = inv.filter(file_ids)
2593
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
1027
"""Return Tree for a revision on this branch.
1029
`revision_id` may not be None or 'null:'"""
1030
assert None not in revision_ids
1031
assert _mod_revision.NULL_REVISION not in revision_ids
1032
texts = self.get_inventory_weave().get_texts(revision_ids)
1033
for text, revision_id in zip(texts, revision_ids):
1034
inv = self.deserialise_inventory(revision_id, text)
1035
yield RevisionTree(self, inv, revision_id)
2595
1037
@needs_read_lock
2596
1038
def get_ancestry(self, revision_id, topo_sorted=True):
2597
1039
"""Return a list of revision-ids integrated by a revision.
2599
The first element of the list is always None, indicating the origin
2600
revision. This might change when we have history horizons, or
1041
The first element of the list is always None, indicating the origin
1042
revision. This might change when we have history horizons, or
2601
1043
perhaps we should have a new API.
2603
1045
This is topologically sorted.
2605
1047
if _mod_revision.is_null(revision_id):
1049
revision_id = osutils.safe_revision_id(revision_id)
2607
1050
if not self.has_revision(revision_id):
2608
1051
raise errors.NoSuchRevision(self, revision_id)
2609
graph = self.get_graph()
2611
search = graph._make_breadth_first_searcher([revision_id])
2614
found, ghosts = search.next_with_ghosts()
2615
except StopIteration:
2618
if _mod_revision.NULL_REVISION in keys:
2619
keys.remove(_mod_revision.NULL_REVISION)
2621
parent_map = graph.get_parent_map(keys)
2622
keys = tsort.topo_sort(parent_map)
2623
return [None] + list(keys)
1052
w = self.get_inventory_weave()
1053
candidates = w.get_ancestry(revision_id, topo_sorted)
1054
return [None] + candidates # self._eliminate_revisions_not_present(candidates)
2625
def pack(self, hint=None, clean_obsolete_packs=False):
2626
1057
"""Compress the data within the repository.
2628
1059
This operation only makes sense for some repository types. For other
2629
1060
types it should be a no-op that just returns.
2631
1062
This stub method does not require a lock, but subclasses should use
2632
@needs_write_lock as this is a long running call its reasonable to
1063
@needs_write_lock as this is a long running call its reasonable to
2633
1064
implicitly lock for the user.
2635
:param hint: If not supplied, the whole repository is packed.
2636
If supplied, the repository may use the hint parameter as a
2637
hint for the parts of the repository to pack. A hint can be
2638
obtained from the result of commit_write_group(). Out of
2639
date hints are simply ignored, because concurrent operations
2640
can obsolete them rapidly.
2642
:param clean_obsolete_packs: Clean obsolete packs immediately after
1068
def print_file(self, file, revision_id):
1069
"""Print `file` to stdout.
1071
FIXME RBC 20060125 as John Meinel points out this is a bad api
1072
- it writes to stdout, it assumes that that is valid etc. Fix
1073
by creating a new more flexible convenience function.
1075
revision_id = osutils.safe_revision_id(revision_id)
1076
tree = self.revision_tree(revision_id)
1077
# use inventory as it was in that revision
1078
file_id = tree.inventory.path2id(file)
1080
# TODO: jam 20060427 Write a test for this code path
1081
# it had a bug in it, and was raising the wrong
1083
raise errors.BzrError("%r is not present in revision %s" % (file, revision_id))
1084
tree.print_file(file_id)
2646
1086
def get_transaction(self):
2647
1087
return self.control_files.get_transaction()
2649
def get_parent_map(self, revision_ids):
2650
"""See graph.StackedParentsProvider.get_parent_map"""
2651
# revisions index works in keys; this just works in revisions
2652
# therefore wrap and unwrap
1089
def revision_parents(self, revision_id):
1090
revision_id = osutils.safe_revision_id(revision_id)
1091
return self.get_inventory_weave().parent_names(revision_id)
1093
def get_parents(self, revision_ids):
1094
"""See StackedParentsProvider.get_parents"""
2655
1096
for revision_id in revision_ids:
2656
1097
if revision_id == _mod_revision.NULL_REVISION:
2657
result[revision_id] = ()
2658
elif revision_id is None:
2659
raise ValueError('get_parent_map(None) is not valid')
2661
query_keys.append((revision_id ,))
2662
for ((revision_id,), parent_keys) in \
2663
self.revisions.get_parent_map(query_keys).iteritems():
2665
result[revision_id] = tuple([parent_revid
2666
for (parent_revid,) in parent_keys])
2668
result[revision_id] = (_mod_revision.NULL_REVISION,)
1101
parents = self.get_revision(revision_id).parent_ids
1102
except errors.NoSuchRevision:
1105
if len(parents) == 0:
1106
parents = [_mod_revision.NULL_REVISION]
1107
parents_list.append(parents)
2671
1110
def _make_parents_provider(self):
2675
def get_known_graph_ancestry(self, revision_ids):
2676
"""Return the known graph for a set of revision ids and their ancestors.
2678
st = static_tuple.StaticTuple
2679
revision_keys = [st(r_id).intern() for r_id in revision_ids]
2680
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
2681
return graph.GraphThunkIdsToKeys(known_graph)
2683
1113
def get_graph(self, other_repository=None):
2684
1114
"""Return the graph walker for this repository format"""
2685
1115
parents_provider = self._make_parents_provider()
2686
1116
if (other_repository is not None and
2687
not self.has_same_location(other_repository)):
2688
parents_provider = graph.StackedParentsProvider(
1117
other_repository.bzrdir.transport.base !=
1118
self.bzrdir.transport.base):
1119
parents_provider = graph._StackedParentsProvider(
2689
1120
[parents_provider, other_repository._make_parents_provider()])
2690
1121
return graph.Graph(parents_provider)
2692
def _get_versioned_file_checker(self, text_key_references=None,
2694
"""Return an object suitable for checking versioned files.
2696
:param text_key_references: if non-None, an already built
2697
dictionary mapping text keys ((fileid, revision_id) tuples)
2698
to whether they were referred to by the inventory of the
2699
revision_id that they contain. If None, this will be
2701
:param ancestors: Optional result from
2702
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2705
return _VersionedFileChecker(self,
2706
text_key_references=text_key_references, ancestors=ancestors)
2708
def revision_ids_to_search_result(self, result_set):
2709
"""Convert a set of revision ids to a graph SearchResult."""
2710
result_parents = set()
2711
for parents in self.get_graph().get_parent_map(
2712
result_set).itervalues():
2713
result_parents.update(parents)
2714
included_keys = result_set.intersection(result_parents)
2715
start_keys = result_set.difference(included_keys)
2716
exclude_keys = result_parents.difference(result_set)
2717
result = graph.SearchResult(start_keys, exclude_keys,
2718
len(result_set), result_set)
2721
1123
@needs_write_lock
2722
1124
def set_make_working_trees(self, new_value):
2723
1125
"""Set the policy flag for making working trees when creating branches.
3402
1589
InterRepository.get(other).method_name(parameters).
3405
_walk_to_common_revisions_batch_size = 50
3406
1592
_optimisers = []
3407
1593
"""The available optimised InterRepository types."""
3410
1595
def copy_content(self, revision_id=None):
3411
"""Make a complete copy of the content in self into destination.
3413
This is a destructive operation! Do not use it on existing
3416
:param revision_id: Only copy the content needed to construct
3417
revision_id and its parents.
3420
self.target.set_make_working_trees(self.source.make_working_trees())
3421
except NotImplementedError:
3423
self.target.fetch(self.source, revision_id=revision_id)
3426
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
1596
raise NotImplementedError(self.copy_content)
1598
def fetch(self, revision_id=None, pb=None):
3428
1599
"""Fetch the content required to construct revision_id.
3430
1601
The content is copied from self.source to self.target.
3432
1603
:param revision_id: if None all content is copied, if NULL_REVISION no
3433
1604
content is copied.
3437
ui.ui_factory.warn_experimental_format_fetch(self)
3438
from bzrlib.fetch import RepoFetcher
3439
# See <https://launchpad.net/bugs/456077> asking for a warning here
3440
if self.source._format.network_name() != self.target._format.network_name():
3441
ui.ui_factory.show_user_warning('cross_format_fetch',
3442
from_format=self.source._format,
3443
to_format=self.target._format)
3444
f = RepoFetcher(to_repository=self.target,
3445
from_repository=self.source,
3446
last_revision=revision_id,
3447
fetch_spec=fetch_spec,
3448
find_ghosts=find_ghosts)
3450
def _walk_to_common_revisions(self, revision_ids):
3451
"""Walk out from revision_ids in source to revisions target has.
3453
:param revision_ids: The start point for the search.
3454
:return: A set of revision ids.
3456
target_graph = self.target.get_graph()
3457
revision_ids = frozenset(revision_ids)
3458
missing_revs = set()
3459
source_graph = self.source.get_graph()
3460
# ensure we don't pay silly lookup costs.
3461
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3462
null_set = frozenset([_mod_revision.NULL_REVISION])
3463
searcher_exhausted = False
3467
# Iterate the searcher until we have enough next_revs
3468
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3470
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3471
next_revs.update(next_revs_part)
3472
ghosts.update(ghosts_part)
3473
except StopIteration:
3474
searcher_exhausted = True
3476
# If there are ghosts in the source graph, and the caller asked for
3477
# them, make sure that they are present in the target.
3478
# We don't care about other ghosts as we can't fetch them and
3479
# haven't been asked to.
3480
ghosts_to_check = set(revision_ids.intersection(ghosts))
3481
revs_to_get = set(next_revs).union(ghosts_to_check)
3483
have_revs = set(target_graph.get_parent_map(revs_to_get))
3484
# we always have NULL_REVISION present.
3485
have_revs = have_revs.union(null_set)
3486
# Check if the target is missing any ghosts we need.
3487
ghosts_to_check.difference_update(have_revs)
3489
# One of the caller's revision_ids is a ghost in both the
3490
# source and the target.
3491
raise errors.NoSuchRevision(
3492
self.source, ghosts_to_check.pop())
3493
missing_revs.update(next_revs - have_revs)
3494
# Because we may have walked past the original stop point, make
3495
# sure everything is stopped
3496
stop_revs = searcher.find_seen_ancestors(have_revs)
3497
searcher.stop_searching_any(stop_revs)
3498
if searcher_exhausted:
3500
return searcher.get_result()
1605
:param pb: optional progress bar to use for progress reports. If not
1606
provided a default one will be created.
1608
Returns the copied revision count and the failed revisions in a tuple:
1611
raise NotImplementedError(self.fetch)
3502
1613
@needs_read_lock
3503
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1614
def missing_revision_ids(self, revision_id=None):
3504
1615
"""Return the revision ids that source has that target does not.
1617
These are returned in topological order.
3506
1619
:param revision_id: only return revision ids included by this
3508
:param find_ghosts: If True find missing revisions in deep history
3509
rather than just finding the surface difference.
3510
:return: A bzrlib.graph.SearchResult.
3512
# stop searching at found target revisions.
3513
if not find_ghosts and revision_id is not None:
3514
return self._walk_to_common_revisions([revision_id])
3515
1622
# generic, possibly worst case, slow code path.
3516
1623
target_ids = set(self.target.all_revision_ids())
3517
1624
if revision_id is not None:
1625
# TODO: jam 20070210 InterRepository is internal enough that it
1626
# should assume revision_ids are already utf-8
1627
revision_id = osutils.safe_revision_id(revision_id)
3518
1628
source_ids = self.source.get_ancestry(revision_id)
3519
if source_ids[0] is not None:
3520
raise AssertionError()
1629
assert source_ids[0] is None
3521
1630
source_ids.pop(0)
3523
1632
source_ids = self.source.all_revision_ids()
3524
1633
result_set = set(source_ids).difference(target_ids)
3525
return self.source.revision_ids_to_search_result(result_set)
3528
def _same_model(source, target):
3529
"""True if source and target have the same data representation.
3531
Note: this is always called on the base class; overriding it in a
3532
subclass will have no effect.
3535
InterRepository._assert_same_model(source, target)
3537
except errors.IncompatibleRepositories, e:
3541
def _assert_same_model(source, target):
3542
"""Raise an exception if two repositories do not use the same model.
3544
if source.supports_rich_root() != target.supports_rich_root():
3545
raise errors.IncompatibleRepositories(source, target,
3546
"different rich-root support")
3547
if source._serializer != target._serializer:
3548
raise errors.IncompatibleRepositories(source, target,
3549
"different serializers")
1634
# this may look like a no-op: its not. It preserves the ordering
1635
# other_ids had while only returning the members from other_ids
1636
# that we've decided we need.
1637
return [rev_id for rev_id in source_ids if rev_id in result_set]
3552
1640
class InterSameDataRepository(InterRepository):
3553
1641
"""Code for converting between repositories that represent the same data.
3555
1643
Data format and model must match for this to work.
3559
1647
def _get_repo_format_to_test(self):
3560
"""Repository format for testing with.
3562
InterSameData can pull from subtree to subtree and from non-subtree to
3563
non-subtree, so we test this with the richest repository format.
3565
from bzrlib.repofmt import knitrepo
3566
return knitrepo.RepositoryFormatKnit3()
1648
"""Repository format for testing with."""
1649
return RepositoryFormat.get_default_format()
3569
1652
def is_compatible(source, target):
3570
return InterRepository._same_model(source, target)
1653
if source.supports_rich_root() != target.supports_rich_root():
1655
if source._serializer != target._serializer:
1660
def copy_content(self, revision_id=None):
1661
"""Make a complete copy of the content in self into destination.
1663
This copies both the repository's revision data, and configuration information
1664
such as the make_working_trees setting.
1666
This is a destructive operation! Do not use it on existing
1669
:param revision_id: Only copy the content needed to construct
1670
revision_id and its parents.
1673
self.target.set_make_working_trees(self.source.make_working_trees())
1674
except NotImplementedError:
1676
# TODO: jam 20070210 This is fairly internal, so we should probably
1677
# just assert that revision_id is not unicode.
1678
revision_id = osutils.safe_revision_id(revision_id)
1679
# but don't bother fetching if we have the needed data now.
1680
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1681
self.target.has_revision(revision_id)):
1683
self.target.fetch(self.source, revision_id=revision_id)
1686
def fetch(self, revision_id=None, pb=None):
1687
"""See InterRepository.fetch()."""
1688
from bzrlib.fetch import GenericRepoFetcher
1689
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1690
self.source, self.source._format, self.target,
1691
self.target._format)
1692
# TODO: jam 20070210 This should be an assert, not a translate
1693
revision_id = osutils.safe_revision_id(revision_id)
1694
f = GenericRepoFetcher(to_repository=self.target,
1695
from_repository=self.source,
1696
last_revision=revision_id,
1698
return f.count_copied, f.failed_revisions
3573
1701
class InterWeaveRepo(InterSameDataRepository):
3574
"""Optimised code paths between Weave based repositories.
3576
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3577
implemented lazy inter-object optimisation.
1702
"""Optimised code paths between Weave based repositories."""
3581
1705
def _get_repo_format_to_test(self):
3694
1829
def is_compatible(source, target):
3695
1830
"""Be compatible with known Knit formats.
3697
1832
We don't test for the stores being of specific types because that
3698
could lead to confusing results, and there is no need to be
1833
could lead to confusing results, and there is no need to be
3699
1834
overly general.
3701
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
1836
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1
3703
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3704
isinstance(target._format, RepositoryFormatKnit))
1838
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1839
isinstance(target._format, (RepositoryFormatKnit1)))
3705
1840
except AttributeError:
3707
return are_knits and InterRepository._same_model(source, target)
1844
def fetch(self, revision_id=None, pb=None):
1845
"""See InterRepository.fetch()."""
1846
from bzrlib.fetch import KnitRepoFetcher
1847
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1848
self.source, self.source._format, self.target, self.target._format)
1849
# TODO: jam 20070210 This should be an assert, not a translate
1850
revision_id = osutils.safe_revision_id(revision_id)
1851
f = KnitRepoFetcher(to_repository=self.target,
1852
from_repository=self.source,
1853
last_revision=revision_id,
1855
return f.count_copied, f.failed_revisions
3709
1857
@needs_read_lock
3710
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1858
def missing_revision_ids(self, revision_id=None):
3711
1859
"""See InterRepository.missing_revision_ids()."""
3712
1860
if revision_id is not None:
3713
1861
source_ids = self.source.get_ancestry(revision_id)
3714
if source_ids[0] is not None:
3715
raise AssertionError()
1862
assert source_ids[0] is None
3716
1863
source_ids.pop(0)
3718
source_ids = self.source.all_revision_ids()
1865
source_ids = self.source._all_possible_ids()
3719
1866
source_ids_set = set(source_ids)
3720
1867
# source_ids is the worst possible case we may need to pull.
3721
1868
# now we want to filter source_ids against what we actually
3722
1869
# have in target, but don't try to check for existence where we know
3723
1870
# we do not have a revision as that would be pointless.
3724
target_ids = set(self.target.all_revision_ids())
1871
target_ids = set(self.target._all_possible_ids())
3725
1872
possibly_present_revisions = target_ids.intersection(source_ids_set)
3726
actually_present_revisions = set(
3727
self.target._eliminate_revisions_not_present(possibly_present_revisions))
1873
actually_present_revisions = set(self.target._eliminate_revisions_not_present(possibly_present_revisions))
3728
1874
required_revisions = source_ids_set.difference(actually_present_revisions)
1875
required_topo_revisions = [rev_id for rev_id in source_ids if rev_id in required_revisions]
3729
1876
if revision_id is not None:
3730
1877
# we used get_ancestry to determine source_ids then we are assured all
3731
1878
# revisions referenced are present as they are installed in topological order.
3732
1879
# and the tip revision was validated by get_ancestry.
3733
result_set = required_revisions
1880
return required_topo_revisions
3735
# if we just grabbed the possibly available ids, then
1882
# if we just grabbed the possibly available ids, then
3736
1883
# we only have an estimate of whats available and need to validate
3737
1884
# that against the revision records.
3739
self.source._eliminate_revisions_not_present(required_revisions))
3740
return self.source.revision_ids_to_search_result(result_set)
3743
class InterDifferingSerializer(InterRepository):
3746
def _get_repo_format_to_test(self):
3750
def is_compatible(source, target):
3751
"""Be compatible with Knit2 source and Knit3 target"""
3752
# This is redundant with format.check_conversion_target(), however that
3753
# raises an exception, and we just want to say "False" as in we won't
3754
# support converting between these formats.
3755
if 'IDS_never' in debug.debug_flags:
3757
if source.supports_rich_root() and not target.supports_rich_root():
3759
if (source._format.supports_tree_reference
3760
and not target._format.supports_tree_reference):
3762
if target._fallback_repositories and target._format.supports_chks:
3763
# IDS doesn't know how to copy CHKs for the parent inventories it
3764
# adds to stacked repos.
3766
if 'IDS_always' in debug.debug_flags:
3768
# Only use this code path for local source and target. IDS does far
3769
# too much IO (both bandwidth and roundtrips) over a network.
3770
if not source.bzrdir.transport.base.startswith('file:///'):
3772
if not target.bzrdir.transport.base.startswith('file:///'):
3776
def _get_trees(self, revision_ids, cache):
3778
for rev_id in revision_ids:
3780
possible_trees.append((rev_id, cache[rev_id]))
3782
# Not cached, but inventory might be present anyway.
3784
tree = self.source.revision_tree(rev_id)
3785
except errors.NoSuchRevision:
3786
# Nope, parent is ghost.
3789
cache[rev_id] = tree
3790
possible_trees.append((rev_id, tree))
3791
return possible_trees
3793
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3794
"""Get the best delta and base for this revision.
3796
:return: (basis_id, delta)
3799
# Generate deltas against each tree, to find the shortest.
3800
texts_possibly_new_in_tree = set()
3801
for basis_id, basis_tree in possible_trees:
3802
delta = tree.inventory._make_delta(basis_tree.inventory)
3803
for old_path, new_path, file_id, new_entry in delta:
3804
if new_path is None:
3805
# This file_id isn't present in the new rev, so we don't
3809
# Rich roots are handled elsewhere...
3811
kind = new_entry.kind
3812
if kind != 'directory' and kind != 'file':
3813
# No text record associated with this inventory entry.
3815
# This is a directory or file that has changed somehow.
3816
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3817
deltas.append((len(delta), basis_id, delta))
3819
return deltas[0][1:]
3821
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3822
"""Find all parent revisions that are absent, but for which the
3823
inventory is present, and copy those inventories.
3825
This is necessary to preserve correctness when the source is stacked
3826
without fallbacks configured. (Note that in cases like upgrade the
3827
source may be not have _fallback_repositories even though it is
3831
for parents in parent_map.values():
3832
parent_revs.update(parents)
3833
present_parents = self.source.get_parent_map(parent_revs)
3834
absent_parents = set(parent_revs).difference(present_parents)
3835
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3836
(rev_id,) for rev_id in absent_parents)
3837
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3838
for parent_tree in self.source.revision_trees(parent_inv_ids):
3839
current_revision_id = parent_tree.get_revision_id()
3840
parents_parents_keys = parent_invs_keys_for_stacking[
3841
(current_revision_id,)]
3842
parents_parents = [key[-1] for key in parents_parents_keys]
3843
basis_id = _mod_revision.NULL_REVISION
3844
basis_tree = self.source.revision_tree(basis_id)
3845
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3846
self.target.add_inventory_by_delta(
3847
basis_id, delta, current_revision_id, parents_parents)
3848
cache[current_revision_id] = parent_tree
3850
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3851
"""Fetch across a few revisions.
3853
:param revision_ids: The revisions to copy
3854
:param basis_id: The revision_id of a tree that must be in cache, used
3855
as a basis for delta when no other base is available
3856
:param cache: A cache of RevisionTrees that we can use.
3857
:param a_graph: A Graph object to determine the heads() of the
3858
rich-root data stream.
3859
:return: The revision_id of the last converted tree. The RevisionTree
3860
for it will be in cache
3862
# Walk though all revisions; get inventory deltas, copy referenced
3863
# texts that delta references, insert the delta, revision and
3865
root_keys_to_create = set()
3868
pending_revisions = []
3869
parent_map = self.source.get_parent_map(revision_ids)
3870
self._fetch_parent_invs_for_stacking(parent_map, cache)
3871
self.source._safe_to_return_from_cache = True
3872
for tree in self.source.revision_trees(revision_ids):
3873
# Find a inventory delta for this revision.
3874
# Find text entries that need to be copied, too.
3875
current_revision_id = tree.get_revision_id()
3876
parent_ids = parent_map.get(current_revision_id, ())
3877
parent_trees = self._get_trees(parent_ids, cache)
3878
possible_trees = list(parent_trees)
3879
if len(possible_trees) == 0:
3880
# There either aren't any parents, or the parents are ghosts,
3881
# so just use the last converted tree.
3882
possible_trees.append((basis_id, cache[basis_id]))
3883
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3885
revision = self.source.get_revision(current_revision_id)
3886
pending_deltas.append((basis_id, delta,
3887
current_revision_id, revision.parent_ids))
3888
if self._converting_to_rich_root:
3889
self._revision_id_to_root_id[current_revision_id] = \
3891
# Determine which texts are in present in this revision but not in
3892
# any of the available parents.
3893
texts_possibly_new_in_tree = set()
3894
for old_path, new_path, file_id, entry in delta:
3895
if new_path is None:
3896
# This file_id isn't present in the new rev
3900
if not self.target.supports_rich_root():
3901
# The target doesn't support rich root, so we don't
3904
if self._converting_to_rich_root:
3905
# This can't be copied normally, we have to insert
3907
root_keys_to_create.add((file_id, entry.revision))
3910
texts_possibly_new_in_tree.add((file_id, entry.revision))
3911
for basis_id, basis_tree in possible_trees:
3912
basis_inv = basis_tree.inventory
3913
for file_key in list(texts_possibly_new_in_tree):
3914
file_id, file_revision = file_key
3916
entry = basis_inv[file_id]
3917
except errors.NoSuchId:
3919
if entry.revision == file_revision:
3920
texts_possibly_new_in_tree.remove(file_key)
3921
text_keys.update(texts_possibly_new_in_tree)
3922
pending_revisions.append(revision)
3923
cache[current_revision_id] = tree
3924
basis_id = current_revision_id
3925
self.source._safe_to_return_from_cache = False
3927
from_texts = self.source.texts
3928
to_texts = self.target.texts
3929
if root_keys_to_create:
3930
root_stream = _mod_fetch._new_root_data_stream(
3931
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3932
self.source, graph=a_graph)
3933
to_texts.insert_record_stream(root_stream)
3934
to_texts.insert_record_stream(from_texts.get_record_stream(
3935
text_keys, self.target._format._fetch_order,
3936
not self.target._format._fetch_uses_deltas))
3937
# insert inventory deltas
3938
for delta in pending_deltas:
3939
self.target.add_inventory_by_delta(*delta)
3940
if self.target._fallback_repositories:
3941
# Make sure this stacked repository has all the parent inventories
3942
# for the new revisions that we are about to insert. We do this
3943
# before adding the revisions so that no revision is added until
3944
# all the inventories it may depend on are added.
3945
# Note that this is overzealous, as we may have fetched these in an
3948
revision_ids = set()
3949
for revision in pending_revisions:
3950
revision_ids.add(revision.revision_id)
3951
parent_ids.update(revision.parent_ids)
3952
parent_ids.difference_update(revision_ids)
3953
parent_ids.discard(_mod_revision.NULL_REVISION)
3954
parent_map = self.source.get_parent_map(parent_ids)
3955
# we iterate over parent_map and not parent_ids because we don't
3956
# want to try copying any revision which is a ghost
3957
for parent_tree in self.source.revision_trees(parent_map):
3958
current_revision_id = parent_tree.get_revision_id()
3959
parents_parents = parent_map[current_revision_id]
3960
possible_trees = self._get_trees(parents_parents, cache)
3961
if len(possible_trees) == 0:
3962
# There either aren't any parents, or the parents are
3963
# ghosts, so just use the last converted tree.
3964
possible_trees.append((basis_id, cache[basis_id]))
3965
basis_id, delta = self._get_delta_for_revision(parent_tree,
3966
parents_parents, possible_trees)
3967
self.target.add_inventory_by_delta(
3968
basis_id, delta, current_revision_id, parents_parents)
3969
# insert signatures and revisions
3970
for revision in pending_revisions:
3972
signature = self.source.get_signature_text(
3973
revision.revision_id)
3974
self.target.add_signature_text(revision.revision_id,
3976
except errors.NoSuchRevision:
3978
self.target.add_revision(revision.revision_id, revision)
3981
def _fetch_all_revisions(self, revision_ids, pb):
3982
"""Fetch everything for the list of revisions.
3984
:param revision_ids: The list of revisions to fetch. Must be in
3986
:param pb: A ProgressTask
3989
basis_id, basis_tree = self._get_basis(revision_ids[0])
3991
cache = lru_cache.LRUCache(100)
3992
cache[basis_id] = basis_tree
3993
del basis_tree # We don't want to hang on to it here
3995
if self._converting_to_rich_root and len(revision_ids) > 100:
3996
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
4001
for offset in range(0, len(revision_ids), batch_size):
4002
self.target.start_write_group()
4004
pb.update('Transferring revisions', offset,
4006
batch = revision_ids[offset:offset+batch_size]
4007
basis_id = self._fetch_batch(batch, basis_id, cache,
4010
self.source._safe_to_return_from_cache = False
4011
self.target.abort_write_group()
4014
hint = self.target.commit_write_group()
4017
if hints and self.target._format.pack_compresses:
4018
self.target.pack(hint=hints)
4019
pb.update('Transferring revisions', len(revision_ids),
4023
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
4025
"""See InterRepository.fetch()."""
4026
if fetch_spec is not None:
4027
raise AssertionError("Not implemented yet...")
4028
ui.ui_factory.warn_experimental_format_fetch(self)
4029
if (not self.source.supports_rich_root()
4030
and self.target.supports_rich_root()):
4031
self._converting_to_rich_root = True
4032
self._revision_id_to_root_id = {}
4034
self._converting_to_rich_root = False
4035
# See <https://launchpad.net/bugs/456077> asking for a warning here
4036
if self.source._format.network_name() != self.target._format.network_name():
4037
ui.ui_factory.show_user_warning('cross_format_fetch',
4038
from_format=self.source._format,
4039
to_format=self.target._format)
4040
revision_ids = self.target.search_missing_revision_ids(self.source,
4041
revision_id, find_ghosts=find_ghosts).get_keys()
4042
if not revision_ids:
4044
revision_ids = tsort.topo_sort(
4045
self.source.get_graph().get_parent_map(revision_ids))
4046
if not revision_ids:
4048
# Walk though all revisions; get inventory deltas, copy referenced
4049
# texts that delta references, insert the delta, revision and
4052
my_pb = ui.ui_factory.nested_progress_bar()
4055
symbol_versioning.warn(
4056
symbol_versioning.deprecated_in((1, 14, 0))
4057
% "pb parameter to fetch()")
4060
self._fetch_all_revisions(revision_ids, pb)
4062
if my_pb is not None:
4064
return len(revision_ids), 0
4066
def _get_basis(self, first_revision_id):
4067
"""Get a revision and tree which exists in the target.
4069
This assumes that first_revision_id is selected for transmission
4070
because all other ancestors are already present. If we can't find an
4071
ancestor we fall back to NULL_REVISION since we know that is safe.
4073
:return: (basis_id, basis_tree)
4075
first_rev = self.source.get_revision(first_revision_id)
4077
basis_id = first_rev.parent_ids[0]
4078
# only valid as a basis if the target has it
4079
self.target.get_revision(basis_id)
4080
# Try to get a basis tree - if its a ghost it will hit the
4081
# NoSuchRevision case.
4082
basis_tree = self.source.revision_tree(basis_id)
4083
except (IndexError, errors.NoSuchRevision):
4084
basis_id = _mod_revision.NULL_REVISION
4085
basis_tree = self.source.revision_tree(basis_id)
4086
return basis_id, basis_tree
4089
InterRepository.register_optimiser(InterDifferingSerializer)
1885
return self.source._eliminate_revisions_not_present(required_topo_revisions)
1888
class InterModel1and2(InterRepository):
1891
def _get_repo_format_to_test(self):
1895
def is_compatible(source, target):
1896
if not source.supports_rich_root() and target.supports_rich_root():
1902
def fetch(self, revision_id=None, pb=None):
1903
"""See InterRepository.fetch()."""
1904
from bzrlib.fetch import Model1toKnit2Fetcher
1905
# TODO: jam 20070210 This should be an assert, not a translate
1906
revision_id = osutils.safe_revision_id(revision_id)
1907
f = Model1toKnit2Fetcher(to_repository=self.target,
1908
from_repository=self.source,
1909
last_revision=revision_id,
1911
return f.count_copied, f.failed_revisions
1914
def copy_content(self, revision_id=None):
1915
"""Make a complete copy of the content in self into destination.
1917
This is a destructive operation! Do not use it on existing
1920
:param revision_id: Only copy the content needed to construct
1921
revision_id and its parents.
1924
self.target.set_make_working_trees(self.source.make_working_trees())
1925
except NotImplementedError:
1927
# TODO: jam 20070210 Internal, assert, don't translate
1928
revision_id = osutils.safe_revision_id(revision_id)
1929
# but don't bother fetching if we have the needed data now.
1930
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1931
self.target.has_revision(revision_id)):
1933
self.target.fetch(self.source, revision_id=revision_id)
1936
class InterKnit1and2(InterKnitRepo):
1939
def _get_repo_format_to_test(self):
1943
def is_compatible(source, target):
1944
"""Be compatible with Knit1 source and Knit3 target"""
1945
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit3
1947
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1, \
1948
RepositoryFormatKnit3
1949
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1950
isinstance(target._format, (RepositoryFormatKnit3)))
1951
except AttributeError:
1955
def fetch(self, revision_id=None, pb=None):
1956
"""See InterRepository.fetch()."""
1957
from bzrlib.fetch import Knit1to2Fetcher
1958
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1959
self.source, self.source._format, self.target,
1960
self.target._format)
1961
# TODO: jam 20070210 This should be an assert, not a translate
1962
revision_id = osutils.safe_revision_id(revision_id)
1963
f = Knit1to2Fetcher(to_repository=self.target,
1964
from_repository=self.source,
1965
last_revision=revision_id,
1967
return f.count_copied, f.failed_revisions
1970
class InterRemoteRepository(InterRepository):
1971
"""Code for converting between RemoteRepository objects.
1973
This just gets an non-remote repository from the RemoteRepository, and calls
1974
InterRepository.get again.
1977
def __init__(self, source, target):
1978
if isinstance(source, remote.RemoteRepository):
1979
source._ensure_real()
1980
real_source = source._real_repository
1982
real_source = source
1983
if isinstance(target, remote.RemoteRepository):
1984
target._ensure_real()
1985
real_target = target._real_repository
1987
real_target = target
1988
self.real_inter = InterRepository.get(real_source, real_target)
1991
def is_compatible(source, target):
1992
if isinstance(source, remote.RemoteRepository):
1994
if isinstance(target, remote.RemoteRepository):
1998
def copy_content(self, revision_id=None):
1999
self.real_inter.copy_content(revision_id=revision_id)
2001
def fetch(self, revision_id=None, pb=None):
2002
self.real_inter.fetch(revision_id=revision_id, pb=pb)
2005
def _get_repo_format_to_test(self):
4090
2009
InterRepository.register_optimiser(InterSameDataRepository)
4091
2010
InterRepository.register_optimiser(InterWeaveRepo)
4092
2011
InterRepository.register_optimiser(InterKnitRepo)
2012
InterRepository.register_optimiser(InterModel1and2)
2013
InterRepository.register_optimiser(InterKnit1and2)
2014
InterRepository.register_optimiser(InterRemoteRepository)
4095
2017
class CopyConverter(object):
4096
2018
"""A repository conversion tool which just performs a copy of the content.
4098
2020
This is slow but quite reliable.
4104
2026
:param target_format: The format the resulting repository should be.
4106
2028
self.target_format = target_format
4108
2030
def convert(self, repo, pb):
4109
2031
"""Perform the conversion of to_convert, giving feedback via pb.
4111
2033
:param to_convert: The disk object to convert.
4112
2034
:param pb: a progress bar to use for progress information.
4114
pb = ui.ui_factory.nested_progress_bar()
4117
2039
# this is only useful with metadir layouts - separated repo content.
4118
2040
# trigger an assertion if not such
4119
2041
repo._format.get_format_string()
4120
2042
self.repo_dir = repo.bzrdir
4121
pb.update('Moving repository to repository.backup')
2043
self.step('Moving repository to repository.backup')
4122
2044
self.repo_dir.transport.move('repository', 'repository.backup')
4123
2045
backup_transport = self.repo_dir.transport.clone('repository.backup')
4124
2046
repo._format.check_conversion_target(self.target_format)
4125
2047
self.source_repo = repo._format.open(self.repo_dir,
4127
2049
_override_transport=backup_transport)
4128
pb.update('Creating new repository')
2050
self.step('Creating new repository')
4129
2051
converted = self.target_format.initialize(self.repo_dir,
4130
2052
self.source_repo.is_shared())
4131
2053
converted.lock_write()
4133
pb.update('Copying content')
2055
self.step('Copying content into repository.')
4134
2056
self.source_repo.copy_content_into(converted)
4136
2058
converted.unlock()
4137
pb.update('Deleting old repository content')
2059
self.step('Deleting old repository content.')
4138
2060
self.repo_dir.transport.delete_tree('repository.backup')
4139
ui.ui_factory.note('repository converted')
2061
self.pb.note('repository converted')
2063
def step(self, message):
2064
"""Update the pb by a step."""
2066
self.pb.update(message, self.count, self.total)
2069
class CommitBuilder(object):
2070
"""Provides an interface to build up a commit.
2072
This allows describing a tree to be committed without needing to
2073
know the internals of the format of the repository.
2076
record_root_entry = False
2077
def __init__(self, repository, parents, config, timestamp=None,
2078
timezone=None, committer=None, revprops=None,
2080
"""Initiate a CommitBuilder.
2082
:param repository: Repository to commit to.
2083
:param parents: Revision ids of the parents of the new revision.
2084
:param config: Configuration to use.
2085
:param timestamp: Optional timestamp recorded for commit.
2086
:param timezone: Optional timezone for timestamp.
2087
:param committer: Optional committer to set for commit.
2088
:param revprops: Optional dictionary of revision properties.
2089
:param revision_id: Optional revision id.
2091
self._config = config
2093
if committer is None:
2094
self._committer = self._config.username()
2096
assert isinstance(committer, basestring), type(committer)
2097
self._committer = committer
2099
self.new_inventory = Inventory(None)
2100
self._new_revision_id = osutils.safe_revision_id(revision_id)
2101
self.parents = parents
2102
self.repository = repository
2105
if revprops is not None:
2106
self._revprops.update(revprops)
2108
if timestamp is None:
2109
timestamp = time.time()
2110
# Restrict resolution to 1ms
2111
self._timestamp = round(timestamp, 3)
2113
if timezone is None:
2114
self._timezone = osutils.local_time_offset()
2116
self._timezone = int(timezone)
2118
self._generate_revision_if_needed()
2120
def commit(self, message):
2121
"""Make the actual commit.
2123
:return: The revision id of the recorded revision.
2125
rev = _mod_revision.Revision(
2126
timestamp=self._timestamp,
2127
timezone=self._timezone,
2128
committer=self._committer,
2130
inventory_sha1=self.inv_sha1,
2131
revision_id=self._new_revision_id,
2132
properties=self._revprops)
2133
rev.parent_ids = self.parents
2134
self.repository.add_revision(self._new_revision_id, rev,
2135
self.new_inventory, self._config)
2136
self.repository.commit_write_group()
2137
return self._new_revision_id
2140
"""Abort the commit that is being built.
2142
self.repository.abort_write_group()
2144
def revision_tree(self):
2145
"""Return the tree that was just committed.
2147
After calling commit() this can be called to get a RevisionTree
2148
representing the newly committed tree. This is preferred to
2149
calling Repository.revision_tree() because that may require
2150
deserializing the inventory, while we already have a copy in
2153
return RevisionTree(self.repository, self.new_inventory,
2154
self._new_revision_id)
2156
def finish_inventory(self):
2157
"""Tell the builder that the inventory is finished."""
2158
if self.new_inventory.root is None:
2159
symbol_versioning.warn('Root entry should be supplied to'
2160
' record_entry_contents, as of bzr 0.10.',
2161
DeprecationWarning, stacklevel=2)
2162
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
2163
self.new_inventory.revision_id = self._new_revision_id
2164
self.inv_sha1 = self.repository.add_inventory(
2165
self._new_revision_id,
2170
def _gen_revision_id(self):
2171
"""Return new revision-id."""
2172
return generate_ids.gen_revision_id(self._config.username(),
2175
def _generate_revision_if_needed(self):
2176
"""Create a revision id if None was supplied.
2178
If the repository can not support user-specified revision ids
2179
they should override this function and raise CannotSetRevisionId
2180
if _new_revision_id is not None.
2182
:raises: CannotSetRevisionId
2184
if self._new_revision_id is None:
2185
self._new_revision_id = self._gen_revision_id()
2187
def record_entry_contents(self, ie, parent_invs, path, tree):
2188
"""Record the content of ie from tree into the commit if needed.
2190
Side effect: sets ie.revision when unchanged
2192
:param ie: An inventory entry present in the commit.
2193
:param parent_invs: The inventories of the parent revisions of the
2195
:param path: The path the entry is at in the tree.
2196
:param tree: The tree which contains this entry and should be used to
2199
if self.new_inventory.root is None and ie.parent_id is not None:
2200
symbol_versioning.warn('Root entry should be supplied to'
2201
' record_entry_contents, as of bzr 0.10.',
2202
DeprecationWarning, stacklevel=2)
2203
self.record_entry_contents(tree.inventory.root.copy(), parent_invs,
2205
self.new_inventory.add(ie)
2207
# ie.revision is always None if the InventoryEntry is considered
2208
# for committing. ie.snapshot will record the correct revision
2209
# which may be the sole parent if it is untouched.
2210
if ie.revision is not None:
2213
# In this revision format, root entries have no knit or weave
2214
if ie is self.new_inventory.root:
2215
# When serializing out to disk and back in
2216
# root.revision is always _new_revision_id
2217
ie.revision = self._new_revision_id
2219
previous_entries = ie.find_previous_heads(
2221
self.repository.weave_store,
2222
self.repository.get_transaction())
2223
# we are creating a new revision for ie in the history store
2225
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
2227
def modified_directory(self, file_id, file_parents):
2228
"""Record the presence of a symbolic link.
2230
:param file_id: The file_id of the link to record.
2231
:param file_parents: The per-file parent revision ids.
2233
self._add_text_to_weave(file_id, [], file_parents.keys())
2235
def modified_reference(self, file_id, file_parents):
2236
"""Record the modification of a reference.
2238
:param file_id: The file_id of the link to record.
2239
:param file_parents: The per-file parent revision ids.
2241
self._add_text_to_weave(file_id, [], file_parents.keys())
2243
def modified_file_text(self, file_id, file_parents,
2244
get_content_byte_lines, text_sha1=None,
2246
"""Record the text of file file_id
2248
:param file_id: The file_id of the file to record the text of.
2249
:param file_parents: The per-file parent revision ids.
2250
:param get_content_byte_lines: A callable which will return the byte
2252
:param text_sha1: Optional SHA1 of the file contents.
2253
:param text_size: Optional size of the file contents.
2255
# mutter('storing text of file {%s} in revision {%s} into %r',
2256
# file_id, self._new_revision_id, self.repository.weave_store)
2257
# special case to avoid diffing on renames or
2259
if (len(file_parents) == 1
2260
and text_sha1 == file_parents.values()[0].text_sha1
2261
and text_size == file_parents.values()[0].text_size):
2262
previous_ie = file_parents.values()[0]
2263
versionedfile = self.repository.weave_store.get_weave(file_id,
2264
self.repository.get_transaction())
2265
versionedfile.clone_text(self._new_revision_id,
2266
previous_ie.revision, file_parents.keys())
2267
return text_sha1, text_size
2269
new_lines = get_content_byte_lines()
2270
# TODO: Rather than invoking sha_strings here, _add_text_to_weave
2271
# should return the SHA1 and size
2272
self._add_text_to_weave(file_id, new_lines, file_parents.keys())
2273
return osutils.sha_strings(new_lines), \
2274
sum(map(len, new_lines))
2276
def modified_link(self, file_id, file_parents, link_target):
2277
"""Record the presence of a symbolic link.
2279
:param file_id: The file_id of the link to record.
2280
:param file_parents: The per-file parent revision ids.
2281
:param link_target: Target location of this link.
2283
self._add_text_to_weave(file_id, [], file_parents.keys())
2285
def _add_text_to_weave(self, file_id, new_lines, parents):
2286
versionedfile = self.repository.weave_store.get_weave_or_empty(
2287
file_id, self.repository.get_transaction())
2288
versionedfile.add_lines(self._new_revision_id, parents, new_lines)
2289
versionedfile.clear_cache()
2292
class _CommitBuilder(CommitBuilder):
2293
"""Temporary class so old CommitBuilders are detected properly
2295
Note: CommitBuilder works whether or not root entry is recorded.
2298
record_root_entry = True
2301
class RootCommitBuilder(CommitBuilder):
2302
"""This commitbuilder actually records the root id"""
2304
record_root_entry = True
2306
def record_entry_contents(self, ie, parent_invs, path, tree):
2307
"""Record the content of ie from tree into the commit if needed.
2309
Side effect: sets ie.revision when unchanged
2311
:param ie: An inventory entry present in the commit.
2312
:param parent_invs: The inventories of the parent revisions of the
2314
:param path: The path the entry is at in the tree.
2315
:param tree: The tree which contains this entry and should be used to
2318
assert self.new_inventory.root is not None or ie.parent_id is None
2319
self.new_inventory.add(ie)
2321
# ie.revision is always None if the InventoryEntry is considered
2322
# for committing. ie.snapshot will record the correct revision
2323
# which may be the sole parent if it is untouched.
2324
if ie.revision is not None:
2327
previous_entries = ie.find_previous_heads(
2329
self.repository.weave_store,
2330
self.repository.get_transaction())
2331
# we are creating a new revision for ie in the history store
2333
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
4143
2336
_unescape_map = {
4168
2361
if _unescape_re is None:
4169
2362
_unescape_re = re.compile('\&([^;]*);')
4170
2363
return _unescape_re.sub(_unescaper, data)
4173
class _VersionedFileChecker(object):
4175
def __init__(self, repository, text_key_references=None, ancestors=None):
4176
self.repository = repository
4177
self.text_index = self.repository._generate_text_key_index(
4178
text_key_references=text_key_references, ancestors=ancestors)
4180
def calculate_file_version_parents(self, text_key):
4181
"""Calculate the correct parents for a file version according to
4184
parent_keys = self.text_index[text_key]
4185
if parent_keys == [_mod_revision.NULL_REVISION]:
4187
return tuple(parent_keys)
4189
def check_file_version_parents(self, texts, progress_bar=None):
4190
"""Check the parents stored in a versioned file are correct.
4192
It also detects file versions that are not referenced by their
4193
corresponding revision's inventory.
4195
:returns: A tuple of (wrong_parents, dangling_file_versions).
4196
wrong_parents is a dict mapping {revision_id: (stored_parents,
4197
correct_parents)} for each revision_id where the stored parents
4198
are not correct. dangling_file_versions is a set of (file_id,
4199
revision_id) tuples for versions that are present in this versioned
4200
file, but not used by the corresponding inventory.
4202
local_progress = None
4203
if progress_bar is None:
4204
local_progress = ui.ui_factory.nested_progress_bar()
4205
progress_bar = local_progress
4207
return self._check_file_version_parents(texts, progress_bar)
4210
local_progress.finished()
4212
def _check_file_version_parents(self, texts, progress_bar):
4213
"""See check_file_version_parents."""
4215
self.file_ids = set([file_id for file_id, _ in
4216
self.text_index.iterkeys()])
4217
# text keys is now grouped by file_id
4218
n_versions = len(self.text_index)
4219
progress_bar.update('loading text store', 0, n_versions)
4220
parent_map = self.repository.texts.get_parent_map(self.text_index)
4221
# On unlistable transports this could well be empty/error...
4222
text_keys = self.repository.texts.keys()
4223
unused_keys = frozenset(text_keys) - set(self.text_index)
4224
for num, key in enumerate(self.text_index.iterkeys()):
4225
progress_bar.update('checking text graph', num, n_versions)
4226
correct_parents = self.calculate_file_version_parents(key)
4228
knit_parents = parent_map[key]
4229
except errors.RevisionNotPresent:
4232
if correct_parents != knit_parents:
4233
wrong_parents[key] = (knit_parents, correct_parents)
4234
return wrong_parents, unused_keys
4237
def _old_get_graph(repository, revision_id):
4238
"""DO NOT USE. That is all. I'm serious."""
4239
graph = repository.get_graph()
4240
revision_graph = dict(((key, value) for key, value in
4241
graph.iter_ancestry([revision_id]) if value is not None))
4242
return _strip_NULL_ghosts(revision_graph)
4245
def _strip_NULL_ghosts(revision_graph):
4246
"""Also don't use this. more compatibility code for unmigrated clients."""
4247
# Filter ghosts, and null:
4248
if _mod_revision.NULL_REVISION in revision_graph:
4249
del revision_graph[_mod_revision.NULL_REVISION]
4250
for key, parents in revision_graph.items():
4251
revision_graph[key] = tuple(parent for parent in parents if parent
4253
return revision_graph
4256
class StreamSink(object):
4257
"""An object that can insert a stream into a repository.
4259
This interface handles the complexity of reserialising inventories and
4260
revisions from different formats, and allows unidirectional insertion into
4261
stacked repositories without looking for the missing basis parents
4265
def __init__(self, target_repo):
4266
self.target_repo = target_repo
4268
def insert_stream(self, stream, src_format, resume_tokens):
4269
"""Insert a stream's content into the target repository.
4271
:param src_format: a bzr repository format.
4273
:return: a list of resume tokens and an iterable of keys additional
4274
items required before the insertion can be completed.
4276
self.target_repo.lock_write()
4279
self.target_repo.resume_write_group(resume_tokens)
4282
self.target_repo.start_write_group()
4285
# locked_insert_stream performs a commit|suspend.
4286
return self._locked_insert_stream(stream, src_format, is_resume)
4288
self.target_repo.abort_write_group(suppress_errors=True)
4291
self.target_repo.unlock()
4293
def _locked_insert_stream(self, stream, src_format, is_resume):
4294
to_serializer = self.target_repo._format._serializer
4295
src_serializer = src_format._serializer
4297
if to_serializer == src_serializer:
4298
# If serializers match and the target is a pack repository, set the
4299
# write cache size on the new pack. This avoids poor performance
4300
# on transports where append is unbuffered (such as
4301
# RemoteTransport). This is safe to do because nothing should read
4302
# back from the target repository while a stream with matching
4303
# serialization is being inserted.
4304
# The exception is that a delta record from the source that should
4305
# be a fulltext may need to be expanded by the target (see
4306
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4307
# explicitly flush any buffered writes first in that rare case.
4309
new_pack = self.target_repo._pack_collection._new_pack
4310
except AttributeError:
4311
# Not a pack repository
4314
new_pack.set_write_cache_size(1024*1024)
4315
for substream_type, substream in stream:
4316
if 'stream' in debug.debug_flags:
4317
mutter('inserting substream: %s', substream_type)
4318
if substream_type == 'texts':
4319
self.target_repo.texts.insert_record_stream(substream)
4320
elif substream_type == 'inventories':
4321
if src_serializer == to_serializer:
4322
self.target_repo.inventories.insert_record_stream(
4325
self._extract_and_insert_inventories(
4326
substream, src_serializer)
4327
elif substream_type == 'inventory-deltas':
4328
self._extract_and_insert_inventory_deltas(
4329
substream, src_serializer)
4330
elif substream_type == 'chk_bytes':
4331
# XXX: This doesn't support conversions, as it assumes the
4332
# conversion was done in the fetch code.
4333
self.target_repo.chk_bytes.insert_record_stream(substream)
4334
elif substream_type == 'revisions':
4335
# This may fallback to extract-and-insert more often than
4336
# required if the serializers are different only in terms of
4338
if src_serializer == to_serializer:
4339
self.target_repo.revisions.insert_record_stream(
4342
self._extract_and_insert_revisions(substream,
4344
elif substream_type == 'signatures':
4345
self.target_repo.signatures.insert_record_stream(substream)
4347
raise AssertionError('kaboom! %s' % (substream_type,))
4348
# Done inserting data, and the missing_keys calculations will try to
4349
# read back from the inserted data, so flush the writes to the new pack
4350
# (if this is pack format).
4351
if new_pack is not None:
4352
new_pack._write_data('', flush=True)
4353
# Find all the new revisions (including ones from resume_tokens)
4354
missing_keys = self.target_repo.get_missing_parent_inventories(
4355
check_for_missing_texts=is_resume)
4357
for prefix, versioned_file in (
4358
('texts', self.target_repo.texts),
4359
('inventories', self.target_repo.inventories),
4360
('revisions', self.target_repo.revisions),
4361
('signatures', self.target_repo.signatures),
4362
('chk_bytes', self.target_repo.chk_bytes),
4364
if versioned_file is None:
4366
# TODO: key is often going to be a StaticTuple object
4367
# I don't believe we can define a method by which
4368
# (prefix,) + StaticTuple will work, though we could
4369
# define a StaticTuple.sq_concat that would allow you to
4370
# pass in either a tuple or a StaticTuple as the second
4371
# object, so instead we could have:
4372
# StaticTuple(prefix) + key here...
4373
missing_keys.update((prefix,) + key for key in
4374
versioned_file.get_missing_compression_parent_keys())
4375
except NotImplementedError:
4376
# cannot even attempt suspending, and missing would have failed
4377
# during stream insertion.
4378
missing_keys = set()
4381
# suspend the write group and tell the caller what we is
4382
# missing. We know we can suspend or else we would not have
4383
# entered this code path. (All repositories that can handle
4384
# missing keys can handle suspending a write group).
4385
write_group_tokens = self.target_repo.suspend_write_group()
4386
return write_group_tokens, missing_keys
4387
hint = self.target_repo.commit_write_group()
4388
if (to_serializer != src_serializer and
4389
self.target_repo._format.pack_compresses):
4390
self.target_repo.pack(hint=hint)
4393
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4394
target_rich_root = self.target_repo._format.rich_root_data
4395
target_tree_refs = self.target_repo._format.supports_tree_reference
4396
for record in substream:
4397
# Insert the delta directly
4398
inventory_delta_bytes = record.get_bytes_as('fulltext')
4399
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4401
parse_result = deserialiser.parse_text_bytes(
4402
inventory_delta_bytes)
4403
except inventory_delta.IncompatibleInventoryDelta, err:
4404
trace.mutter("Incompatible delta: %s", err.msg)
4405
raise errors.IncompatibleRevision(self.target_repo._format)
4406
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4407
revision_id = new_id
4408
parents = [key[0] for key in record.parents]
4409
self.target_repo.add_inventory_by_delta(
4410
basis_id, inv_delta, revision_id, parents)
4412
def _extract_and_insert_inventories(self, substream, serializer,
4414
"""Generate a new inventory versionedfile in target, converting data.
4416
The inventory is retrieved from the source, (deserializing it), and
4417
stored in the target (reserializing it in a different format).
4419
target_rich_root = self.target_repo._format.rich_root_data
4420
target_tree_refs = self.target_repo._format.supports_tree_reference
4421
for record in substream:
4422
# It's not a delta, so it must be a fulltext in the source
4423
# serializer's format.
4424
bytes = record.get_bytes_as('fulltext')
4425
revision_id = record.key[0]
4426
inv = serializer.read_inventory_from_string(bytes, revision_id)
4427
parents = [key[0] for key in record.parents]
4428
self.target_repo.add_inventory(revision_id, inv, parents)
4429
# No need to keep holding this full inv in memory when the rest of
4430
# the substream is likely to be all deltas.
4433
def _extract_and_insert_revisions(self, substream, serializer):
4434
for record in substream:
4435
bytes = record.get_bytes_as('fulltext')
4436
revision_id = record.key[0]
4437
rev = serializer.read_revision_from_string(bytes)
4438
if rev.revision_id != revision_id:
4439
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4440
self.target_repo.add_revision(revision_id, rev)
4443
if self.target_repo._format._fetch_reconcile:
4444
self.target_repo.reconcile()
4447
class StreamSource(object):
4448
"""A source of a stream for fetching between repositories."""
4450
def __init__(self, from_repository, to_format):
4451
"""Create a StreamSource streaming from from_repository."""
4452
self.from_repository = from_repository
4453
self.to_format = to_format
4455
def delta_on_metadata(self):
4456
"""Return True if delta's are permitted on metadata streams.
4458
That is on revisions and signatures.
4460
src_serializer = self.from_repository._format._serializer
4461
target_serializer = self.to_format._serializer
4462
return (self.to_format._fetch_uses_deltas and
4463
src_serializer == target_serializer)
4465
def _fetch_revision_texts(self, revs):
4466
# fetch signatures first and then the revision texts
4467
# may need to be a InterRevisionStore call here.
4468
from_sf = self.from_repository.signatures
4469
# A missing signature is just skipped.
4470
keys = [(rev_id,) for rev_id in revs]
4471
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4473
self.to_format._fetch_order,
4474
not self.to_format._fetch_uses_deltas))
4475
# If a revision has a delta, this is actually expanded inside the
4476
# insert_record_stream code now, which is an alternate fix for
4478
from_rf = self.from_repository.revisions
4479
revisions = from_rf.get_record_stream(
4481
self.to_format._fetch_order,
4482
not self.delta_on_metadata())
4483
return [('signatures', signatures), ('revisions', revisions)]
4485
def _generate_root_texts(self, revs):
4486
"""This will be called by get_stream between fetching weave texts and
4487
fetching the inventory weave.
4489
if self._rich_root_upgrade():
4490
return _mod_fetch.Inter1and2Helper(
4491
self.from_repository).generate_root_texts(revs)
4495
def get_stream(self, search):
4497
revs = search.get_keys()
4498
graph = self.from_repository.get_graph()
4499
revs = tsort.topo_sort(graph.get_parent_map(revs))
4500
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4502
for knit_kind, file_id, revisions in data_to_fetch:
4503
if knit_kind != phase:
4505
# Make a new progress bar for this phase
4506
if knit_kind == "file":
4507
# Accumulate file texts
4508
text_keys.extend([(file_id, revision) for revision in
4510
elif knit_kind == "inventory":
4511
# Now copy the file texts.
4512
from_texts = self.from_repository.texts
4513
yield ('texts', from_texts.get_record_stream(
4514
text_keys, self.to_format._fetch_order,
4515
not self.to_format._fetch_uses_deltas))
4516
# Cause an error if a text occurs after we have done the
4519
# Before we process the inventory we generate the root
4520
# texts (if necessary) so that the inventories references
4522
for _ in self._generate_root_texts(revs):
4524
# we fetch only the referenced inventories because we do not
4525
# know for unselected inventories whether all their required
4526
# texts are present in the other repository - it could be
4528
for info in self._get_inventory_stream(revs):
4530
elif knit_kind == "signatures":
4531
# Nothing to do here; this will be taken care of when
4532
# _fetch_revision_texts happens.
4534
elif knit_kind == "revisions":
4535
for record in self._fetch_revision_texts(revs):
4538
raise AssertionError("Unknown knit kind %r" % knit_kind)
4540
def get_stream_for_missing_keys(self, missing_keys):
4541
# missing keys can only occur when we are byte copying and not
4542
# translating (because translation means we don't send
4543
# unreconstructable deltas ever).
4545
keys['texts'] = set()
4546
keys['revisions'] = set()
4547
keys['inventories'] = set()
4548
keys['chk_bytes'] = set()
4549
keys['signatures'] = set()
4550
for key in missing_keys:
4551
keys[key[0]].add(key[1:])
4552
if len(keys['revisions']):
4553
# If we allowed copying revisions at this point, we could end up
4554
# copying a revision without copying its required texts: a
4555
# violation of the requirements for repository integrity.
4556
raise AssertionError(
4557
'cannot copy revisions to fill in missing deltas %s' % (
4558
keys['revisions'],))
4559
for substream_kind, keys in keys.iteritems():
4560
vf = getattr(self.from_repository, substream_kind)
4561
if vf is None and keys:
4562
raise AssertionError(
4563
"cannot fill in keys for a versioned file we don't"
4564
" have: %s needs %s" % (substream_kind, keys))
4566
# No need to stream something we don't have
4568
if substream_kind == 'inventories':
4569
# Some missing keys are genuinely ghosts, filter those out.
4570
present = self.from_repository.inventories.get_parent_map(keys)
4571
revs = [key[0] for key in present]
4572
# Get the inventory stream more-or-less as we do for the
4573
# original stream; there's no reason to assume that records
4574
# direct from the source will be suitable for the sink. (Think
4575
# e.g. 2a -> 1.9-rich-root).
4576
for info in self._get_inventory_stream(revs, missing=True):
4580
# Ask for full texts always so that we don't need more round trips
4581
# after this stream.
4582
# Some of the missing keys are genuinely ghosts, so filter absent
4583
# records. The Sink is responsible for doing another check to
4584
# ensure that ghosts don't introduce missing data for future
4586
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4587
self.to_format._fetch_order, True))
4588
yield substream_kind, stream
4590
def inventory_fetch_order(self):
4591
if self._rich_root_upgrade():
4592
return 'topological'
4594
return self.to_format._fetch_order
4596
def _rich_root_upgrade(self):
4597
return (not self.from_repository._format.rich_root_data and
4598
self.to_format.rich_root_data)
4600
def _get_inventory_stream(self, revision_ids, missing=False):
4601
from_format = self.from_repository._format
4602
if (from_format.supports_chks and self.to_format.supports_chks and
4603
from_format.network_name() == self.to_format.network_name()):
4604
raise AssertionError(
4605
"this case should be handled by GroupCHKStreamSource")
4606
elif 'forceinvdeltas' in debug.debug_flags:
4607
return self._get_convertable_inventory_stream(revision_ids,
4608
delta_versus_null=missing)
4609
elif from_format.network_name() == self.to_format.network_name():
4611
return self._get_simple_inventory_stream(revision_ids,
4613
elif (not from_format.supports_chks and not self.to_format.supports_chks
4614
and from_format._serializer == self.to_format._serializer):
4615
# Essentially the same format.
4616
return self._get_simple_inventory_stream(revision_ids,
4619
# Any time we switch serializations, we want to use an
4620
# inventory-delta based approach.
4621
return self._get_convertable_inventory_stream(revision_ids,
4622
delta_versus_null=missing)
4624
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4625
# NB: This currently reopens the inventory weave in source;
4626
# using a single stream interface instead would avoid this.
4627
from_weave = self.from_repository.inventories
4629
delta_closure = True
4631
delta_closure = not self.delta_on_metadata()
4632
yield ('inventories', from_weave.get_record_stream(
4633
[(rev_id,) for rev_id in revision_ids],
4634
self.inventory_fetch_order(), delta_closure))
4636
def _get_convertable_inventory_stream(self, revision_ids,
4637
delta_versus_null=False):
4638
# The two formats are sufficiently different that there is no fast
4639
# path, so we need to send just inventorydeltas, which any
4640
# sufficiently modern client can insert into any repository.
4641
# The StreamSink code expects to be able to
4642
# convert on the target, so we need to put bytes-on-the-wire that can
4643
# be converted. That means inventory deltas (if the remote is <1.19,
4644
# RemoteStreamSink will fallback to VFS to insert the deltas).
4645
yield ('inventory-deltas',
4646
self._stream_invs_as_deltas(revision_ids,
4647
delta_versus_null=delta_versus_null))
4649
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4650
"""Return a stream of inventory-deltas for the given rev ids.
4652
:param revision_ids: The list of inventories to transmit
4653
:param delta_versus_null: Don't try to find a minimal delta for this
4654
entry, instead compute the delta versus the NULL_REVISION. This
4655
effectively streams a complete inventory. Used for stuff like
4656
filling in missing parents, etc.
4658
from_repo = self.from_repository
4659
revision_keys = [(rev_id,) for rev_id in revision_ids]
4660
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4661
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4663
inventories = self.from_repository.iter_inventories(
4664
revision_ids, 'topological')
4665
format = from_repo._format
4666
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4667
inventory_cache = lru_cache.LRUCache(50)
4668
null_inventory = from_repo.revision_tree(
4669
_mod_revision.NULL_REVISION).inventory
4670
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4671
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4672
# repo back into a non-rich-root repo ought to be allowed)
4673
serializer = inventory_delta.InventoryDeltaSerializer(
4674
versioned_root=format.rich_root_data,
4675
tree_references=format.supports_tree_reference)
4676
for inv in inventories:
4677
key = (inv.revision_id,)
4678
parent_keys = parent_map.get(key, ())
4680
if not delta_versus_null and parent_keys:
4681
# The caller did not ask for complete inventories and we have
4682
# some parents that we can delta against. Make a delta against
4683
# each parent so that we can find the smallest.
4684
parent_ids = [parent_key[0] for parent_key in parent_keys]
4685
for parent_id in parent_ids:
4686
if parent_id not in invs_sent_so_far:
4687
# We don't know that the remote side has this basis, so
4690
if parent_id == _mod_revision.NULL_REVISION:
4691
parent_inv = null_inventory
4693
parent_inv = inventory_cache.get(parent_id, None)
4694
if parent_inv is None:
4695
parent_inv = from_repo.get_inventory(parent_id)
4696
candidate_delta = inv._make_delta(parent_inv)
4697
if (delta is None or
4698
len(delta) > len(candidate_delta)):
4699
delta = candidate_delta
4700
basis_id = parent_id
4702
# Either none of the parents ended up being suitable, or we
4703
# were asked to delta against NULL
4704
basis_id = _mod_revision.NULL_REVISION
4705
delta = inv._make_delta(null_inventory)
4706
invs_sent_so_far.add(inv.revision_id)
4707
inventory_cache[inv.revision_id] = inv
4708
delta_serialized = ''.join(
4709
serializer.delta_to_lines(basis_id, key[-1], delta))
4710
yield versionedfile.FulltextContentFactory(
4711
key, parent_keys, None, delta_serialized)
4714
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4715
stop_revision=None):
4716
"""Extend the partial history to include a given index
4718
If a stop_index is supplied, stop when that index has been reached.
4719
If a stop_revision is supplied, stop when that revision is
4720
encountered. Otherwise, stop when the beginning of history is
4723
:param stop_index: The index which should be present. When it is
4724
present, history extension will stop.
4725
:param stop_revision: The revision id which should be present. When
4726
it is encountered, history extension will stop.
4728
start_revision = partial_history_cache[-1]
4729
iterator = repo.iter_reverse_revision_history(start_revision)
4731
#skip the last revision in the list
4734
if (stop_index is not None and
4735
len(partial_history_cache) > stop_index):
4737
if partial_history_cache[-1] == stop_revision:
4739
revision_id = iterator.next()
4740
partial_history_cache.append(revision_id)
4741
except StopIteration:
added
removed
bzrlib/repository.py
