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 bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
17
from binascii import hexlify
18
from copy import deepcopy
19
from cStringIO import StringIO
22
from unittest import TestSuite
42
revision as _mod_revision,
49
from bzrlib.bundle import serializer
24
from bzrlib import bzrdir, check, delta, gpg, errors, xml5, ui, transactions, osutils
25
from bzrlib.decorators import needs_read_lock, needs_write_lock
26
from bzrlib.errors import InvalidRevisionId
27
from bzrlib.graph import Graph
28
from bzrlib.inter import InterObject
29
from bzrlib.inventory import Inventory
30
from bzrlib.knit import KnitVersionedFile, KnitPlainFactory
31
from bzrlib.lockable_files import LockableFiles, TransportLock
32
from bzrlib.lockdir import LockDir
33
from bzrlib.osutils import (safe_unicode, rand_bytes, compact_date,
35
from bzrlib.revision import NULL_REVISION, Revision
50
36
from bzrlib.revisiontree import RevisionTree
51
from bzrlib.store.versioned import VersionedFileStore
37
from bzrlib.store.versioned import VersionedFileStore, WeaveStore
38
from bzrlib.store.text import TextStore
39
from bzrlib.symbol_versioning import (deprecated_method,
52
42
from bzrlib.testament import Testament
60
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
61
from bzrlib.inter import InterObject
62
from bzrlib.inventory import (
68
from bzrlib.recordcounter import RecordCounter
69
from bzrlib.lock import _RelockDebugMixin, LogicalLockResult
70
from bzrlib.trace import (
71
log_exception_quietly, note, mutter, mutter_callsite, warning)
74
# Old formats display a warning, but only once
75
_deprecation_warning_done = False
78
class IsInWriteGroupError(errors.InternalBzrError):
80
_fmt = "May not refresh_data of repo %(repo)s while in a write group."
82
def __init__(self, repo):
83
errors.InternalBzrError.__init__(self, repo=repo)
86
class CommitBuilder(object):
87
"""Provides an interface to build up a commit.
89
This allows describing a tree to be committed without needing to
90
know the internals of the format of the repository.
93
# all clients should supply tree roots.
94
record_root_entry = True
95
# the default CommitBuilder does not manage trees whose root is versioned.
96
_versioned_root = False
98
def __init__(self, repository, parents, config, timestamp=None,
99
timezone=None, committer=None, revprops=None,
101
"""Initiate a CommitBuilder.
103
:param repository: Repository to commit to.
104
:param parents: Revision ids of the parents of the new revision.
105
:param config: Configuration to use.
106
:param timestamp: Optional timestamp recorded for commit.
107
:param timezone: Optional timezone for timestamp.
108
:param committer: Optional committer to set for commit.
109
:param revprops: Optional dictionary of revision properties.
110
:param revision_id: Optional revision id.
112
self._config = config
114
if committer is None:
115
self._committer = self._config.username()
117
self._committer = committer
119
self.new_inventory = Inventory(None)
120
self._new_revision_id = revision_id
121
self.parents = parents
122
self.repository = repository
125
if revprops is not None:
126
self._validate_revprops(revprops)
127
self._revprops.update(revprops)
129
if timestamp is None:
130
timestamp = time.time()
131
# Restrict resolution to 1ms
132
self._timestamp = round(timestamp, 3)
135
self._timezone = osutils.local_time_offset()
137
self._timezone = int(timezone)
139
self._generate_revision_if_needed()
140
self.__heads = graph.HeadsCache(repository.get_graph()).heads
141
self._basis_delta = []
142
# API compatibility, older code that used CommitBuilder did not call
143
# .record_delete(), which means the delta that is computed would not be
144
# valid. Callers that will call record_delete() should call
145
# .will_record_deletes() to indicate that.
146
self._recording_deletes = False
147
# memo'd check for no-op commits.
148
self._any_changes = False
150
def any_changes(self):
151
"""Return True if any entries were changed.
153
This includes merge-only changes. It is the core for the --unchanged
156
:return: True if any changes have occured.
158
return self._any_changes
160
def _validate_unicode_text(self, text, context):
161
"""Verify things like commit messages don't have bogus characters."""
163
raise ValueError('Invalid value for %s: %r' % (context, text))
165
def _validate_revprops(self, revprops):
166
for key, value in revprops.iteritems():
167
# We know that the XML serializers do not round trip '\r'
168
# correctly, so refuse to accept them
169
if not isinstance(value, basestring):
170
raise ValueError('revision property (%s) is not a valid'
171
' (unicode) string: %r' % (key, value))
172
self._validate_unicode_text(value,
173
'revision property (%s)' % (key,))
175
def commit(self, message):
176
"""Make the actual commit.
178
:return: The revision id of the recorded revision.
180
self._validate_unicode_text(message, 'commit message')
181
rev = _mod_revision.Revision(
182
timestamp=self._timestamp,
183
timezone=self._timezone,
184
committer=self._committer,
186
inventory_sha1=self.inv_sha1,
187
revision_id=self._new_revision_id,
188
properties=self._revprops)
189
rev.parent_ids = self.parents
190
self.repository.add_revision(self._new_revision_id, rev,
191
self.new_inventory, self._config)
192
self.repository.commit_write_group()
193
return self._new_revision_id
196
"""Abort the commit that is being built.
198
self.repository.abort_write_group()
200
def revision_tree(self):
201
"""Return the tree that was just committed.
203
After calling commit() this can be called to get a RevisionTree
204
representing the newly committed tree. This is preferred to
205
calling Repository.revision_tree() because that may require
206
deserializing the inventory, while we already have a copy in
209
if self.new_inventory is None:
210
self.new_inventory = self.repository.get_inventory(
211
self._new_revision_id)
212
return RevisionTree(self.repository, self.new_inventory,
213
self._new_revision_id)
215
def finish_inventory(self):
216
"""Tell the builder that the inventory is finished.
218
:return: The inventory id in the repository, which can be used with
219
repository.get_inventory.
221
if self.new_inventory is None:
222
# an inventory delta was accumulated without creating a new
224
basis_id = self.basis_delta_revision
225
# We ignore the 'inventory' returned by add_inventory_by_delta
226
# because self.new_inventory is used to hint to the rest of the
227
# system what code path was taken
228
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
229
basis_id, self._basis_delta, self._new_revision_id,
232
if self.new_inventory.root is None:
233
raise AssertionError('Root entry should be supplied to'
234
' record_entry_contents, as of bzr 0.10.')
235
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
236
self.new_inventory.revision_id = self._new_revision_id
237
self.inv_sha1 = self.repository.add_inventory(
238
self._new_revision_id,
242
return self._new_revision_id
244
def _gen_revision_id(self):
245
"""Return new revision-id."""
246
return generate_ids.gen_revision_id(self._committer, self._timestamp)
248
def _generate_revision_if_needed(self):
249
"""Create a revision id if None was supplied.
251
If the repository can not support user-specified revision ids
252
they should override this function and raise CannotSetRevisionId
253
if _new_revision_id is not None.
255
:raises: CannotSetRevisionId
257
if self._new_revision_id is None:
258
self._new_revision_id = self._gen_revision_id()
259
self.random_revid = True
261
self.random_revid = False
263
def _heads(self, file_id, revision_ids):
264
"""Calculate the graph heads for revision_ids in the graph of file_id.
266
This can use either a per-file graph or a global revision graph as we
267
have an identity relationship between the two graphs.
269
return self.__heads(revision_ids)
271
def _check_root(self, ie, parent_invs, tree):
272
"""Helper for record_entry_contents.
274
:param ie: An entry being added.
275
:param parent_invs: The inventories of the parent revisions of the
277
:param tree: The tree that is being committed.
279
# In this revision format, root entries have no knit or weave When
280
# serializing out to disk and back in root.revision is always
282
ie.revision = self._new_revision_id
284
def _require_root_change(self, tree):
285
"""Enforce an appropriate root object change.
287
This is called once when record_iter_changes is called, if and only if
288
the root was not in the delta calculated by record_iter_changes.
290
:param tree: The tree which is being committed.
292
if len(self.parents) == 0:
293
raise errors.RootMissing()
294
entry = entry_factory['directory'](tree.path2id(''), '',
296
entry.revision = self._new_revision_id
297
self._basis_delta.append(('', '', entry.file_id, entry))
299
def _get_delta(self, ie, basis_inv, path):
300
"""Get a delta against the basis inventory for ie."""
301
if ie.file_id not in basis_inv:
303
result = (None, path, ie.file_id, ie)
304
self._basis_delta.append(result)
306
elif ie != basis_inv[ie.file_id]:
308
# TODO: avoid tis id2path call.
309
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
310
self._basis_delta.append(result)
316
def get_basis_delta(self):
317
"""Return the complete inventory delta versus the basis inventory.
319
This has been built up with the calls to record_delete and
320
record_entry_contents. The client must have already called
321
will_record_deletes() to indicate that they will be generating a
324
:return: An inventory delta, suitable for use with apply_delta, or
325
Repository.add_inventory_by_delta, etc.
327
if not self._recording_deletes:
328
raise AssertionError("recording deletes not activated.")
329
return self._basis_delta
331
def record_delete(self, path, file_id):
332
"""Record that a delete occured against a basis tree.
334
This is an optional API - when used it adds items to the basis_delta
335
being accumulated by the commit builder. It cannot be called unless the
336
method will_record_deletes() has been called to inform the builder that
337
a delta is being supplied.
339
:param path: The path of the thing deleted.
340
:param file_id: The file id that was deleted.
342
if not self._recording_deletes:
343
raise AssertionError("recording deletes not activated.")
344
delta = (path, None, file_id, None)
345
self._basis_delta.append(delta)
346
self._any_changes = True
349
def will_record_deletes(self):
350
"""Tell the commit builder that deletes are being notified.
352
This enables the accumulation of an inventory delta; for the resulting
353
commit to be valid, deletes against the basis MUST be recorded via
354
builder.record_delete().
356
self._recording_deletes = True
358
basis_id = self.parents[0]
360
basis_id = _mod_revision.NULL_REVISION
361
self.basis_delta_revision = basis_id
363
def record_entry_contents(self, ie, parent_invs, path, tree,
365
"""Record the content of ie from tree into the commit if needed.
367
Side effect: sets ie.revision when unchanged
369
:param ie: An inventory entry present in the commit.
370
:param parent_invs: The inventories of the parent revisions of the
372
:param path: The path the entry is at in the tree.
373
:param tree: The tree which contains this entry and should be used to
375
:param content_summary: Summary data from the tree about the paths
376
content - stat, length, exec, sha/link target. This is only
377
accessed when the entry has a revision of None - that is when it is
378
a candidate to commit.
379
:return: A tuple (change_delta, version_recorded, fs_hash).
380
change_delta is an inventory_delta change for this entry against
381
the basis tree of the commit, or None if no change occured against
383
version_recorded is True if a new version of the entry has been
384
recorded. For instance, committing a merge where a file was only
385
changed on the other side will return (delta, False).
386
fs_hash is either None, or the hash details for the path (currently
387
a tuple of the contents sha1 and the statvalue returned by
388
tree.get_file_with_stat()).
390
if self.new_inventory.root is None:
391
if ie.parent_id is not None:
392
raise errors.RootMissing()
393
self._check_root(ie, parent_invs, tree)
394
if ie.revision is None:
395
kind = content_summary[0]
397
# ie is carried over from a prior commit
399
# XXX: repository specific check for nested tree support goes here - if
400
# the repo doesn't want nested trees we skip it ?
401
if (kind == 'tree-reference' and
402
not self.repository._format.supports_tree_reference):
403
# mismatch between commit builder logic and repository:
404
# this needs the entry creation pushed down into the builder.
405
raise NotImplementedError('Missing repository subtree support.')
406
self.new_inventory.add(ie)
408
# TODO: slow, take it out of the inner loop.
410
basis_inv = parent_invs[0]
412
basis_inv = Inventory(root_id=None)
414
# ie.revision is always None if the InventoryEntry is considered
415
# for committing. We may record the previous parents revision if the
416
# content is actually unchanged against a sole head.
417
if ie.revision is not None:
418
if not self._versioned_root and path == '':
419
# repositories that do not version the root set the root's
420
# revision to the new commit even when no change occurs (more
421
# specifically, they do not record a revision on the root; and
422
# the rev id is assigned to the root during deserialisation -
423
# this masks when a change may have occurred against the basis.
424
# To match this we always issue a delta, because the revision
425
# of the root will always be changing.
426
if ie.file_id in basis_inv:
427
delta = (basis_inv.id2path(ie.file_id), path,
431
delta = (None, path, ie.file_id, ie)
432
self._basis_delta.append(delta)
433
return delta, False, None
435
# we don't need to commit this, because the caller already
436
# determined that an existing revision of this file is
437
# appropriate. If its not being considered for committing then
438
# it and all its parents to the root must be unaltered so
439
# no-change against the basis.
440
if ie.revision == self._new_revision_id:
441
raise AssertionError("Impossible situation, a skipped "
442
"inventory entry (%r) claims to be modified in this "
443
"commit (%r).", (ie, self._new_revision_id))
444
return None, False, None
445
# XXX: Friction: parent_candidates should return a list not a dict
446
# so that we don't have to walk the inventories again.
447
parent_candiate_entries = ie.parent_candidates(parent_invs)
448
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
450
for inv in parent_invs:
451
if ie.file_id in inv:
452
old_rev = inv[ie.file_id].revision
453
if old_rev in head_set:
454
heads.append(inv[ie.file_id].revision)
455
head_set.remove(inv[ie.file_id].revision)
458
# now we check to see if we need to write a new record to the
460
# We write a new entry unless there is one head to the ancestors, and
461
# the kind-derived content is unchanged.
463
# Cheapest check first: no ancestors, or more the one head in the
464
# ancestors, we write a new node.
468
# There is a single head, look it up for comparison
469
parent_entry = parent_candiate_entries[heads[0]]
470
# if the non-content specific data has changed, we'll be writing a
472
if (parent_entry.parent_id != ie.parent_id or
473
parent_entry.name != ie.name):
475
# now we need to do content specific checks:
477
# if the kind changed the content obviously has
478
if kind != parent_entry.kind:
480
# Stat cache fingerprint feedback for the caller - None as we usually
481
# don't generate one.
484
if content_summary[2] is None:
485
raise ValueError("Files must not have executable = None")
487
# We can't trust a check of the file length because of content
489
if (# if the exec bit has changed we have to store:
490
parent_entry.executable != content_summary[2]):
492
elif parent_entry.text_sha1 == content_summary[3]:
493
# all meta and content is unchanged (using a hash cache
494
# hit to check the sha)
495
ie.revision = parent_entry.revision
496
ie.text_size = parent_entry.text_size
497
ie.text_sha1 = parent_entry.text_sha1
498
ie.executable = parent_entry.executable
499
return self._get_delta(ie, basis_inv, path), False, None
501
# Either there is only a hash change(no hash cache entry,
502
# or same size content change), or there is no change on
504
# Provide the parent's hash to the store layer, so that the
505
# content is unchanged we will not store a new node.
506
nostore_sha = parent_entry.text_sha1
508
# We want to record a new node regardless of the presence or
509
# absence of a content change in the file.
511
ie.executable = content_summary[2]
512
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
514
text = file_obj.read()
518
ie.text_sha1, ie.text_size = self._add_text_to_weave(
519
ie.file_id, text, heads, nostore_sha)
520
# Let the caller know we generated a stat fingerprint.
521
fingerprint = (ie.text_sha1, stat_value)
522
except errors.ExistingContent:
523
# Turns out that the file content was unchanged, and we were
524
# only going to store a new node if it was changed. Carry over
526
ie.revision = parent_entry.revision
527
ie.text_size = parent_entry.text_size
528
ie.text_sha1 = parent_entry.text_sha1
529
ie.executable = parent_entry.executable
530
return self._get_delta(ie, basis_inv, path), False, None
531
elif kind == 'directory':
533
# all data is meta here, nothing specific to directory, so
535
ie.revision = parent_entry.revision
536
return self._get_delta(ie, basis_inv, path), False, None
537
self._add_text_to_weave(ie.file_id, '', heads, None)
538
elif kind == 'symlink':
539
current_link_target = content_summary[3]
541
# symlink target is not generic metadata, check if it has
543
if current_link_target != parent_entry.symlink_target:
546
# unchanged, carry over.
547
ie.revision = parent_entry.revision
548
ie.symlink_target = parent_entry.symlink_target
549
return self._get_delta(ie, basis_inv, path), False, None
550
ie.symlink_target = current_link_target
551
self._add_text_to_weave(ie.file_id, '', heads, None)
552
elif kind == 'tree-reference':
554
if content_summary[3] != parent_entry.reference_revision:
557
# unchanged, carry over.
558
ie.reference_revision = parent_entry.reference_revision
559
ie.revision = parent_entry.revision
560
return self._get_delta(ie, basis_inv, path), False, None
561
ie.reference_revision = content_summary[3]
562
if ie.reference_revision is None:
563
raise AssertionError("invalid content_summary for nested tree: %r"
564
% (content_summary,))
565
self._add_text_to_weave(ie.file_id, '', heads, None)
567
raise NotImplementedError('unknown kind')
568
ie.revision = self._new_revision_id
569
self._any_changes = True
570
return self._get_delta(ie, basis_inv, path), True, fingerprint
572
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
573
_entry_factory=entry_factory):
574
"""Record a new tree via iter_changes.
576
:param tree: The tree to obtain text contents from for changed objects.
577
:param basis_revision_id: The revision id of the tree the iter_changes
578
has been generated against. Currently assumed to be the same
579
as self.parents[0] - if it is not, errors may occur.
580
:param iter_changes: An iter_changes iterator with the changes to apply
581
to basis_revision_id. The iterator must not include any items with
582
a current kind of None - missing items must be either filtered out
583
or errored-on beefore record_iter_changes sees the item.
584
:param _entry_factory: Private method to bind entry_factory locally for
586
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
589
# Create an inventory delta based on deltas between all the parents and
590
# deltas between all the parent inventories. We use inventory delta's
591
# between the inventory objects because iter_changes masks
592
# last-changed-field only changes.
594
# file_id -> change map, change is fileid, paths, changed, versioneds,
595
# parents, names, kinds, executables
597
# {file_id -> revision_id -> inventory entry, for entries in parent
598
# trees that are not parents[0]
602
revtrees = list(self.repository.revision_trees(self.parents))
603
except errors.NoSuchRevision:
604
# one or more ghosts, slow path.
606
for revision_id in self.parents:
608
revtrees.append(self.repository.revision_tree(revision_id))
609
except errors.NoSuchRevision:
611
basis_revision_id = _mod_revision.NULL_REVISION
613
revtrees.append(self.repository.revision_tree(
614
_mod_revision.NULL_REVISION))
615
# The basis inventory from a repository
617
basis_inv = revtrees[0].inventory
619
basis_inv = self.repository.revision_tree(
620
_mod_revision.NULL_REVISION).inventory
621
if len(self.parents) > 0:
622
if basis_revision_id != self.parents[0] and not ghost_basis:
624
"arbitrary basis parents not yet supported with merges")
625
for revtree in revtrees[1:]:
626
for change in revtree.inventory._make_delta(basis_inv):
627
if change[1] is None:
628
# Not present in this parent.
630
if change[2] not in merged_ids:
631
if change[0] is not None:
632
basis_entry = basis_inv[change[2]]
633
merged_ids[change[2]] = [
635
basis_entry.revision,
638
parent_entries[change[2]] = {
640
basis_entry.revision:basis_entry,
642
change[3].revision:change[3],
645
merged_ids[change[2]] = [change[3].revision]
646
parent_entries[change[2]] = {change[3].revision:change[3]}
648
merged_ids[change[2]].append(change[3].revision)
649
parent_entries[change[2]][change[3].revision] = change[3]
652
# Setup the changes from the tree:
653
# changes maps file_id -> (change, [parent revision_ids])
655
for change in iter_changes:
656
# This probably looks up in basis_inv way to much.
657
if change[1][0] is not None:
658
head_candidate = [basis_inv[change[0]].revision]
661
changes[change[0]] = change, merged_ids.get(change[0],
663
unchanged_merged = set(merged_ids) - set(changes)
664
# Extend the changes dict with synthetic changes to record merges of
666
for file_id in unchanged_merged:
667
# Record a merged version of these items that did not change vs the
668
# basis. This can be either identical parallel changes, or a revert
669
# of a specific file after a merge. The recorded content will be
670
# that of the current tree (which is the same as the basis), but
671
# the per-file graph will reflect a merge.
672
# NB:XXX: We are reconstructing path information we had, this
673
# should be preserved instead.
674
# inv delta change: (file_id, (path_in_source, path_in_target),
675
# changed_content, versioned, parent, name, kind,
678
basis_entry = basis_inv[file_id]
679
except errors.NoSuchId:
680
# a change from basis->some_parents but file_id isn't in basis
681
# so was new in the merge, which means it must have changed
682
# from basis -> current, and as it hasn't the add was reverted
683
# by the user. So we discard this change.
687
(basis_inv.id2path(file_id), tree.id2path(file_id)),
689
(basis_entry.parent_id, basis_entry.parent_id),
690
(basis_entry.name, basis_entry.name),
691
(basis_entry.kind, basis_entry.kind),
692
(basis_entry.executable, basis_entry.executable))
693
changes[file_id] = (change, merged_ids[file_id])
694
# changes contains tuples with the change and a set of inventory
695
# candidates for the file.
697
# old_path, new_path, file_id, new_inventory_entry
698
seen_root = False # Is the root in the basis delta?
699
inv_delta = self._basis_delta
700
modified_rev = self._new_revision_id
701
for change, head_candidates in changes.values():
702
if change[3][1]: # versioned in target.
703
# Several things may be happening here:
704
# We may have a fork in the per-file graph
705
# - record a change with the content from tree
706
# We may have a change against < all trees
707
# - carry over the tree that hasn't changed
708
# We may have a change against all trees
709
# - record the change with the content from tree
712
entry = _entry_factory[kind](file_id, change[5][1],
714
head_set = self._heads(change[0], set(head_candidates))
717
for head_candidate in head_candidates:
718
if head_candidate in head_set:
719
heads.append(head_candidate)
720
head_set.remove(head_candidate)
723
# Could be a carry-over situation:
724
parent_entry_revs = parent_entries.get(file_id, None)
725
if parent_entry_revs:
726
parent_entry = parent_entry_revs.get(heads[0], None)
729
if parent_entry is None:
730
# The parent iter_changes was called against is the one
731
# that is the per-file head, so any change is relevant
732
# iter_changes is valid.
733
carry_over_possible = False
735
# could be a carry over situation
736
# A change against the basis may just indicate a merge,
737
# we need to check the content against the source of the
738
# merge to determine if it was changed after the merge
740
if (parent_entry.kind != entry.kind or
741
parent_entry.parent_id != entry.parent_id or
742
parent_entry.name != entry.name):
743
# Metadata common to all entries has changed
744
# against per-file parent
745
carry_over_possible = False
747
carry_over_possible = True
748
# per-type checks for changes against the parent_entry
751
# Cannot be a carry-over situation
752
carry_over_possible = False
753
# Populate the entry in the delta
755
# XXX: There is still a small race here: If someone reverts the content of a file
756
# after iter_changes examines and decides it has changed,
757
# we will unconditionally record a new version even if some
758
# other process reverts it while commit is running (with
759
# the revert happening after iter_changes did it's
762
entry.executable = True
764
entry.executable = False
765
if (carry_over_possible and
766
parent_entry.executable == entry.executable):
767
# Check the file length, content hash after reading
769
nostore_sha = parent_entry.text_sha1
772
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
774
text = file_obj.read()
778
entry.text_sha1, entry.text_size = self._add_text_to_weave(
779
file_id, text, heads, nostore_sha)
780
yield file_id, change[1][1], (entry.text_sha1, stat_value)
781
except errors.ExistingContent:
782
# No content change against a carry_over parent
783
# Perhaps this should also yield a fs hash update?
785
entry.text_size = parent_entry.text_size
786
entry.text_sha1 = parent_entry.text_sha1
787
elif kind == 'symlink':
789
entry.symlink_target = tree.get_symlink_target(file_id)
790
if (carry_over_possible and
791
parent_entry.symlink_target == entry.symlink_target):
794
self._add_text_to_weave(change[0], '', heads, None)
795
elif kind == 'directory':
796
if carry_over_possible:
799
# Nothing to set on the entry.
800
# XXX: split into the Root and nonRoot versions.
801
if change[1][1] != '' or self.repository.supports_rich_root():
802
self._add_text_to_weave(change[0], '', heads, None)
803
elif kind == 'tree-reference':
804
if not self.repository._format.supports_tree_reference:
805
# This isn't quite sane as an error, but we shouldn't
806
# ever see this code path in practice: tree's don't
807
# permit references when the repo doesn't support tree
809
raise errors.UnsupportedOperation(tree.add_reference,
811
reference_revision = tree.get_reference_revision(change[0])
812
entry.reference_revision = reference_revision
813
if (carry_over_possible and
814
parent_entry.reference_revision == reference_revision):
817
self._add_text_to_weave(change[0], '', heads, None)
819
raise AssertionError('unknown kind %r' % kind)
821
entry.revision = modified_rev
823
entry.revision = parent_entry.revision
826
new_path = change[1][1]
827
inv_delta.append((change[1][0], new_path, change[0], entry))
830
self.new_inventory = None
832
# This should perhaps be guarded by a check that the basis we
833
# commit against is the basis for the commit and if not do a delta
835
self._any_changes = True
837
# housekeeping root entry changes do not affect no-change commits.
838
self._require_root_change(tree)
839
self.basis_delta_revision = basis_revision_id
841
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
842
parent_keys = tuple([(file_id, parent) for parent in parents])
843
return self.repository.texts._add_text(
844
(file_id, self._new_revision_id), parent_keys, new_text,
845
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
848
class RootCommitBuilder(CommitBuilder):
849
"""This commitbuilder actually records the root id"""
851
# the root entry gets versioned properly by this builder.
852
_versioned_root = True
854
def _check_root(self, ie, parent_invs, tree):
855
"""Helper for record_entry_contents.
857
:param ie: An entry being added.
858
:param parent_invs: The inventories of the parent revisions of the
860
:param tree: The tree that is being committed.
863
def _require_root_change(self, tree):
864
"""Enforce an appropriate root object change.
866
This is called once when record_iter_changes is called, if and only if
867
the root was not in the delta calculated by record_iter_changes.
869
:param tree: The tree which is being committed.
871
# versioned roots do not change unless the tree found a change.
874
class RepositoryWriteLockResult(LogicalLockResult):
875
"""The result of write locking a repository.
877
:ivar repository_token: The token obtained from the underlying lock, or
879
:ivar unlock: A callable which will unlock the lock.
882
def __init__(self, unlock, repository_token):
883
LogicalLockResult.__init__(self, unlock)
884
self.repository_token = repository_token
887
return "RepositoryWriteLockResult(%s, %s)" % (self.repository_token,
891
######################################################################
895
class Repository(_RelockDebugMixin, controldir.ControlComponent):
43
from bzrlib.trace import mutter, note
44
from bzrlib.tsort import topo_sort
45
from bzrlib.weave import WeaveFile
48
class Repository(object):
896
49
"""Repository holding history for one or more branches.
898
51
The repository holds and retrieves historical information including
899
52
revisions and file history. It's normally accessed only by the Branch,
900
53
which views a particular line of development through that history.
902
The Repository builds on top of some byte storage facilies (the revisions,
903
signatures, inventories, texts and chk_bytes attributes) and a Transport,
904
which respectively provide byte storage and a means to access the (possibly
55
The Repository builds on top of Stores and a Transport, which respectively
56
describe the disk data format and the way of accessing the (possibly
907
The byte storage facilities are addressed via tuples, which we refer to
908
as 'keys' throughout the code base. Revision_keys, inventory_keys and
909
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
910
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
911
byte string made up of a hash identifier and a hash value.
912
We use this interface because it allows low friction with the underlying
913
code that implements disk indices, network encoding and other parts of
916
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
917
the serialised revisions for the repository. This can be used to obtain
918
revision graph information or to access raw serialised revisions.
919
The result of trying to insert data into the repository via this store
920
is undefined: it should be considered read-only except for implementors
922
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
923
the serialised signatures for the repository. This can be used to
924
obtain access to raw serialised signatures. The result of trying to
925
insert data into the repository via this store is undefined: it should
926
be considered read-only except for implementors of repositories.
927
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
928
the serialised inventories for the repository. This can be used to
929
obtain unserialised inventories. The result of trying to insert data
930
into the repository via this store is undefined: it should be
931
considered read-only except for implementors of repositories.
932
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
933
texts of files and directories for the repository. This can be used to
934
obtain file texts or file graphs. Note that Repository.iter_file_bytes
935
is usually a better interface for accessing file texts.
936
The result of trying to insert data into the repository via this store
937
is undefined: it should be considered read-only except for implementors
939
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
940
any data the repository chooses to store or have indexed by its hash.
941
The result of trying to insert data into the repository via this store
942
is undefined: it should be considered read-only except for implementors
944
:ivar _transport: Transport for file access to repository, typically
945
pointing to .bzr/repository.
948
# What class to use for a CommitBuilder. Often its simpler to change this
949
# in a Repository class subclass rather than to override
950
# get_commit_builder.
951
_commit_builder_class = CommitBuilder
952
# The search regex used by xml based repositories to determine what things
953
# where changed in a single commit.
954
_file_ids_altered_regex = lazy_regex.lazy_compile(
955
r'file_id="(?P<file_id>[^"]+)"'
956
r'.* revision="(?P<revision_id>[^"]+)"'
959
def abort_write_group(self, suppress_errors=False):
960
"""Commit the contents accrued within the current write group.
962
:param suppress_errors: if true, abort_write_group will catch and log
963
unexpected errors that happen during the abort, rather than
964
allowing them to propagate. Defaults to False.
966
:seealso: start_write_group.
968
if self._write_group is not self.get_transaction():
969
# has an unlock or relock occured ?
972
'(suppressed) mismatched lock context and write group. %r, %r',
973
self._write_group, self.get_transaction())
975
raise errors.BzrError(
976
'mismatched lock context and write group. %r, %r' %
977
(self._write_group, self.get_transaction()))
979
self._abort_write_group()
980
except Exception, exc:
981
self._write_group = None
982
if not suppress_errors:
984
mutter('abort_write_group failed')
985
log_exception_quietly()
986
note('bzr: ERROR (ignored): %s', exc)
987
self._write_group = None
989
def _abort_write_group(self):
990
"""Template method for per-repository write group cleanup.
992
This is called during abort before the write group is considered to be
993
finished and should cleanup any internal state accrued during the write
994
group. There is no requirement that data handed to the repository be
995
*not* made available - this is not a rollback - but neither should any
996
attempt be made to ensure that data added is fully commited. Abort is
997
invoked when an error has occured so futher disk or network operations
998
may not be possible or may error and if possible should not be
1002
def add_fallback_repository(self, repository):
1003
"""Add a repository to use for looking up data not held locally.
1005
:param repository: A repository.
1007
if not self._format.supports_external_lookups:
1008
raise errors.UnstackableRepositoryFormat(self._format, self.base)
1009
if self.is_locked():
1010
# This repository will call fallback.unlock() when we transition to
1011
# the unlocked state, so we make sure to increment the lock count
1012
repository.lock_read()
1013
self._check_fallback_repository(repository)
1014
self._fallback_repositories.append(repository)
1015
self.texts.add_fallback_versioned_files(repository.texts)
1016
self.inventories.add_fallback_versioned_files(repository.inventories)
1017
self.revisions.add_fallback_versioned_files(repository.revisions)
1018
self.signatures.add_fallback_versioned_files(repository.signatures)
1019
if self.chk_bytes is not None:
1020
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
1022
def _check_fallback_repository(self, repository):
1023
"""Check that this repository can fallback to repository safely.
1025
Raise an error if not.
1027
:param repository: A repository to fallback to.
1029
return InterRepository._assert_same_model(self, repository)
1031
def add_inventory(self, revision_id, inv, parents):
1032
"""Add the inventory inv to the repository as revision_id.
1034
:param parents: The revision ids of the parents that revision_id
61
def add_inventory(self, revid, inv, parents):
62
"""Add the inventory inv to the repository as revid.
64
:param parents: The revision ids of the parents that revid
1035
65
is known to have and are in the repository already.
1037
:returns: The validator(which is a sha1 digest, though what is sha'd is
1038
repository format specific) of the serialized inventory.
1040
if not self.is_in_write_group():
1041
raise AssertionError("%r not in write group" % (self,))
1042
_mod_revision.check_not_reserved_id(revision_id)
1043
if not (inv.revision_id is None or inv.revision_id == revision_id):
1044
raise AssertionError(
1045
"Mismatch between inventory revision"
1046
" id and insertion revid (%r, %r)"
1047
% (inv.revision_id, revision_id))
1048
if inv.root is None:
1049
raise errors.RootMissing()
1050
return self._add_inventory_checked(revision_id, inv, parents)
1052
def _add_inventory_checked(self, revision_id, inv, parents):
1053
"""Add inv to the repository after checking the inputs.
1055
This function can be overridden to allow different inventory styles.
1057
:seealso: add_inventory, for the contract.
1059
inv_lines = self._serializer.write_inventory_to_lines(inv)
1060
return self._inventory_add_lines(revision_id, parents,
1061
inv_lines, check_content=False)
1063
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1064
parents, basis_inv=None, propagate_caches=False):
1065
"""Add a new inventory expressed as a delta against another revision.
1067
See the inventory developers documentation for the theory behind
1070
:param basis_revision_id: The inventory id the delta was created
1071
against. (This does not have to be a direct parent.)
1072
:param delta: The inventory delta (see Inventory.apply_delta for
1074
:param new_revision_id: The revision id that the inventory is being
1076
:param parents: The revision ids of the parents that revision_id is
1077
known to have and are in the repository already. These are supplied
1078
for repositories that depend on the inventory graph for revision
1079
graph access, as well as for those that pun ancestry with delta
1081
:param basis_inv: The basis inventory if it is already known,
1083
:param propagate_caches: If True, the caches for this inventory are
1084
copied to and updated for the result if possible.
1086
:returns: (validator, new_inv)
1087
The validator(which is a sha1 digest, though what is sha'd is
1088
repository format specific) of the serialized inventory, and the
1089
resulting inventory.
1091
if not self.is_in_write_group():
1092
raise AssertionError("%r not in write group" % (self,))
1093
_mod_revision.check_not_reserved_id(new_revision_id)
1094
basis_tree = self.revision_tree(basis_revision_id)
1095
basis_tree.lock_read()
1097
# Note that this mutates the inventory of basis_tree, which not all
1098
# inventory implementations may support: A better idiom would be to
1099
# return a new inventory, but as there is no revision tree cache in
1100
# repository this is safe for now - RBC 20081013
1101
if basis_inv is None:
1102
basis_inv = basis_tree.inventory
1103
basis_inv.apply_delta(delta)
1104
basis_inv.revision_id = new_revision_id
1105
return (self.add_inventory(new_revision_id, basis_inv, parents),
1110
def _inventory_add_lines(self, revision_id, parents, lines,
1111
check_content=True):
1112
"""Store lines in inv_vf and return the sha1 of the inventory."""
1113
parents = [(parent,) for parent in parents]
1114
result = self.inventories.add_lines((revision_id,), parents, lines,
1115
check_content=check_content)[0]
1116
self.inventories._access.flush()
1119
def add_revision(self, revision_id, rev, inv=None, config=None):
1120
"""Add rev to the revision store as revision_id.
1122
:param revision_id: the revision id to use.
67
returns the sha1 of the serialized inventory.
69
assert inv.revision_id is None or inv.revision_id == revid, \
70
"Mismatch between inventory revision" \
71
" id and insertion revid (%r, %r)" % (inv.revision_id, revid)
72
inv_text = xml5.serializer_v5.write_inventory_to_string(inv)
73
inv_sha1 = osutils.sha_string(inv_text)
74
inv_vf = self.control_weaves.get_weave('inventory',
75
self.get_transaction())
76
self._inventory_add_lines(inv_vf, revid, parents, osutils.split_lines(inv_text))
79
def _inventory_add_lines(self, inv_vf, revid, parents, lines):
81
for parent in parents:
83
final_parents.append(parent)
85
inv_vf.add_lines(revid, final_parents, lines)
88
def add_revision(self, rev_id, rev, inv=None, config=None):
89
"""Add rev to the revision store as rev_id.
91
:param rev_id: the revision id to use.
1123
92
:param rev: The revision object.
1124
93
:param inv: The inventory for the revision. if None, it will be looked
1125
94
up in the inventory storer
1191
153
Returns a set of the present revisions.
1194
graph = self.get_graph()
1195
parent_map = graph.get_parent_map(revision_ids)
1196
# The old API returned a list, should this actually be a set?
1197
return parent_map.keys()
1199
def _check_inventories(self, checker):
1200
"""Check the inventories found from the revision scan.
1202
This is responsible for verifying the sha1 of inventories and
1203
creating a pending_keys set that covers data referenced by inventories.
1205
bar = ui.ui_factory.nested_progress_bar()
1207
self._do_check_inventories(checker, bar)
1211
def _do_check_inventories(self, checker, bar):
1212
"""Helper for _check_inventories."""
1214
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1215
kinds = ['chk_bytes', 'texts']
1216
count = len(checker.pending_keys)
1217
bar.update("inventories", 0, 2)
1218
current_keys = checker.pending_keys
1219
checker.pending_keys = {}
1220
# Accumulate current checks.
1221
for key in current_keys:
1222
if key[0] != 'inventories' and key[0] not in kinds:
1223
checker._report_items.append('unknown key type %r' % (key,))
1224
keys[key[0]].add(key[1:])
1225
if keys['inventories']:
1226
# NB: output order *should* be roughly sorted - topo or
1227
# inverse topo depending on repository - either way decent
1228
# to just delta against. However, pre-CHK formats didn't
1229
# try to optimise inventory layout on disk. As such the
1230
# pre-CHK code path does not use inventory deltas.
1232
for record in self.inventories.check(keys=keys['inventories']):
1233
if record.storage_kind == 'absent':
1234
checker._report_items.append(
1235
'Missing inventory {%s}' % (record.key,))
1237
last_object = self._check_record('inventories', record,
1238
checker, last_object,
1239
current_keys[('inventories',) + record.key])
1240
del keys['inventories']
1243
bar.update("texts", 1)
1244
while (checker.pending_keys or keys['chk_bytes']
1246
# Something to check.
1247
current_keys = checker.pending_keys
1248
checker.pending_keys = {}
1249
# Accumulate current checks.
1250
for key in current_keys:
1251
if key[0] not in kinds:
1252
checker._report_items.append('unknown key type %r' % (key,))
1253
keys[key[0]].add(key[1:])
1254
# Check the outermost kind only - inventories || chk_bytes || texts
1258
for record in getattr(self, kind).check(keys=keys[kind]):
1259
if record.storage_kind == 'absent':
1260
checker._report_items.append(
1261
'Missing %s {%s}' % (kind, record.key,))
1263
last_object = self._check_record(kind, record,
1264
checker, last_object, current_keys[(kind,) + record.key])
1268
def _check_record(self, kind, record, checker, last_object, item_data):
1269
"""Check a single text from this repository."""
1270
if kind == 'inventories':
1271
rev_id = record.key[0]
1272
inv = self._deserialise_inventory(rev_id,
1273
record.get_bytes_as('fulltext'))
1274
if last_object is not None:
1275
delta = inv._make_delta(last_object)
1276
for old_path, path, file_id, ie in delta:
1279
ie.check(checker, rev_id, inv)
1281
for path, ie in inv.iter_entries():
1282
ie.check(checker, rev_id, inv)
1283
if self._format.fast_deltas:
1285
elif kind == 'chk_bytes':
1286
# No code written to check chk_bytes for this repo format.
1287
checker._report_items.append(
1288
'unsupported key type chk_bytes for %s' % (record.key,))
1289
elif kind == 'texts':
1290
self._check_text(record, checker, item_data)
1292
checker._report_items.append(
1293
'unknown key type %s for %s' % (kind, record.key))
1295
def _check_text(self, record, checker, item_data):
1296
"""Check a single text."""
1297
# Check it is extractable.
1298
# TODO: check length.
1299
if record.storage_kind == 'chunked':
1300
chunks = record.get_bytes_as(record.storage_kind)
1301
sha1 = osutils.sha_strings(chunks)
1302
length = sum(map(len, chunks))
1304
content = record.get_bytes_as('fulltext')
1305
sha1 = osutils.sha_string(content)
1306
length = len(content)
1307
if item_data and sha1 != item_data[1]:
1308
checker._report_items.append(
1309
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1310
(record.key, sha1, item_data[1], item_data[2]))
156
for id in revision_ids:
157
if self.has_revision(id):
1313
162
def create(a_bzrdir):
1314
163
"""Construct the current default format repository in a_bzrdir."""
1315
164
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1317
def __init__(self, _format, a_bzrdir, control_files):
166
def __init__(self, _format, a_bzrdir, control_files, _revision_store, control_store, text_store):
1318
167
"""instantiate a Repository.
1320
169
:param _format: The format of the repository on disk.
1321
170
:param a_bzrdir: The BzrDir of the repository.
172
In the future we will have a single api for all stores for
173
getting file texts, inventories and revisions, then
174
this construct will accept instances of those things.
1323
# In the future we will have a single api for all stores for
1324
# getting file texts, inventories and revisions, then
1325
# this construct will accept instances of those things.
1326
176
super(Repository, self).__init__()
1327
177
self._format = _format
1328
178
# the following are part of the public API for Repository:
1329
179
self.bzrdir = a_bzrdir
1330
180
self.control_files = control_files
1331
self._transport = control_files._transport
1332
self.base = self._transport.base
1334
self._reconcile_does_inventory_gc = True
1335
self._reconcile_fixes_text_parents = False
1336
self._reconcile_backsup_inventory = True
1337
self._write_group = None
1338
# Additional places to query for data.
1339
self._fallback_repositories = []
1340
# An InventoryEntry cache, used during deserialization
1341
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1342
# Is it safe to return inventory entries directly from the entry cache,
1343
# rather copying them?
1344
self._safe_to_return_from_cache = False
1347
def user_transport(self):
1348
return self.bzrdir.user_transport
1351
def control_transport(self):
1352
return self._transport
181
self._revision_store = _revision_store
182
self.text_store = text_store
183
# backwards compatibility
184
self.weave_store = text_store
185
# not right yet - should be more semantically clear ?
187
self.control_store = control_store
188
self.control_weaves = control_store
189
# TODO: make sure to construct the right store classes, etc, depending
190
# on whether escaping is required.
1354
192
def __repr__(self):
1355
if self._fallback_repositories:
1356
return '%s(%r, fallback_repositories=%r)' % (
1357
self.__class__.__name__,
1359
self._fallback_repositories)
1361
return '%s(%r)' % (self.__class__.__name__,
1364
def _has_same_fallbacks(self, other_repo):
1365
"""Returns true if the repositories have the same fallbacks."""
1366
my_fb = self._fallback_repositories
1367
other_fb = other_repo._fallback_repositories
1368
if len(my_fb) != len(other_fb):
1370
for f, g in zip(my_fb, other_fb):
1371
if not f.has_same_location(g):
1375
def has_same_location(self, other):
1376
"""Returns a boolean indicating if this repository is at the same
1377
location as another repository.
1379
This might return False even when two repository objects are accessing
1380
the same physical repository via different URLs.
1382
if self.__class__ is not other.__class__:
1384
return (self._transport.base == other._transport.base)
1386
def is_in_write_group(self):
1387
"""Return True if there is an open write group.
1389
:seealso: start_write_group.
1391
return self._write_group is not None
193
return '%s(%r)' % (self.__class__.__name__,
194
self.bzrdir.transport.base)
1393
196
def is_locked(self):
1394
197
return self.control_files.is_locked()
1396
def is_write_locked(self):
1397
"""Return True if this object is write locked."""
1398
return self.is_locked() and self.control_files._lock_mode == 'w'
1400
def lock_write(self, token=None):
1401
"""Lock this repository for writing.
1403
This causes caching within the repository obejct to start accumlating
1404
data during reads, and allows a 'write_group' to be obtained. Write
1405
groups must be used for actual data insertion.
1407
A token should be passed in if you know that you have locked the object
1408
some other way, and need to synchronise this object's state with that
1411
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1413
:param token: if this is already locked, then lock_write will fail
1414
unless the token matches the existing lock.
1415
:returns: a token if this instance supports tokens, otherwise None.
1416
:raises TokenLockingNotSupported: when a token is given but this
1417
instance doesn't support using token locks.
1418
:raises MismatchedToken: if the specified token doesn't match the token
1419
of the existing lock.
1420
:seealso: start_write_group.
1421
:return: A RepositoryWriteLockResult.
1423
locked = self.is_locked()
1424
token = self.control_files.lock_write(token=token)
1426
self._warn_if_deprecated()
1427
self._note_lock('w')
1428
for repo in self._fallback_repositories:
1429
# Writes don't affect fallback repos
1431
self._refresh_data()
1432
return RepositoryWriteLockResult(self.unlock, token)
199
def lock_write(self):
200
self.control_files.lock_write()
1434
202
def lock_read(self):
1435
"""Lock the repository for read operations.
1437
:return: An object with an unlock method which will release the lock
1440
locked = self.is_locked()
1441
203
self.control_files.lock_read()
1443
self._warn_if_deprecated()
1444
self._note_lock('r')
1445
for repo in self._fallback_repositories:
1447
self._refresh_data()
1448
return LogicalLockResult(self.unlock)
1450
205
def get_physical_lock_status(self):
1451
206
return self.control_files.get_physical_lock_status()
1453
def leave_lock_in_place(self):
1454
"""Tell this repository not to release the physical lock when this
1457
If lock_write doesn't return a token, then this method is not supported.
1459
self.control_files.leave_in_place()
1461
def dont_leave_lock_in_place(self):
1462
"""Tell this repository to release the physical lock when this
1463
object is unlocked, even if it didn't originally acquire it.
1465
If lock_write doesn't return a token, then this method is not supported.
1467
self.control_files.dont_leave_in_place()
1470
def gather_stats(self, revid=None, committers=None):
1471
"""Gather statistics from a revision id.
1473
:param revid: The revision id to gather statistics from, if None, then
1474
no revision specific statistics are gathered.
1475
:param committers: Optional parameter controlling whether to grab
1476
a count of committers from the revision specific statistics.
1477
:return: A dictionary of statistics. Currently this contains:
1478
committers: The number of committers if requested.
1479
firstrev: A tuple with timestamp, timezone for the penultimate left
1480
most ancestor of revid, if revid is not the NULL_REVISION.
1481
latestrev: A tuple with timestamp, timezone for revid, if revid is
1482
not the NULL_REVISION.
1483
revisions: The total revision count in the repository.
1484
size: An estimate disk size of the repository in bytes.
1487
if revid and committers:
1488
result['committers'] = 0
1489
if revid and revid != _mod_revision.NULL_REVISION:
1491
all_committers = set()
1492
revisions = self.get_ancestry(revid)
1493
# pop the leading None
1495
first_revision = None
1497
# ignore the revisions in the middle - just grab first and last
1498
revisions = revisions[0], revisions[-1]
1499
for revision in self.get_revisions(revisions):
1500
if not first_revision:
1501
first_revision = revision
1503
all_committers.add(revision.committer)
1504
last_revision = revision
1506
result['committers'] = len(all_committers)
1507
result['firstrev'] = (first_revision.timestamp,
1508
first_revision.timezone)
1509
result['latestrev'] = (last_revision.timestamp,
1510
last_revision.timezone)
1512
# now gather global repository information
1513
# XXX: This is available for many repos regardless of listability.
1514
if self.user_transport.listable():
1515
# XXX: do we want to __define len__() ?
1516
# Maybe the versionedfiles object should provide a different
1517
# method to get the number of keys.
1518
result['revisions'] = len(self.revisions.keys())
1519
# result['size'] = t
1522
def find_branches(self, using=False):
1523
"""Find branches underneath this repository.
1525
This will include branches inside other branches.
1527
:param using: If True, list only branches using this repository.
1529
if using and not self.is_shared():
1530
return self.bzrdir.list_branches()
1531
class Evaluator(object):
1534
self.first_call = True
1536
def __call__(self, bzrdir):
1537
# On the first call, the parameter is always the bzrdir
1538
# containing the current repo.
1539
if not self.first_call:
1541
repository = bzrdir.open_repository()
1542
except errors.NoRepositoryPresent:
1545
return False, ([], repository)
1546
self.first_call = False
1547
value = (bzrdir.list_branches(), None)
1551
for branches, repository in bzrdir.BzrDir.find_bzrdirs(
1552
self.user_transport, evaluate=Evaluator()):
1553
if branches is not None:
1554
ret.extend(branches)
1555
if not using and repository is not None:
1556
ret.extend(repository.find_branches())
1560
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
209
def missing_revision_ids(self, other, revision_id=None):
1561
210
"""Return the revision ids that other has that this does not.
1563
212
These are returned in topological order.
1565
214
revision_id: only return revision ids included by revision_id.
1567
return InterRepository.get(other, self).search_missing_revision_ids(
1568
revision_id, find_ghosts)
216
return InterRepository.get(other, self).missing_revision_ids(revision_id)
1757
255
:param revprops: Optional dictionary of revision properties.
1758
256
:param revision_id: Optional revision id.
1760
if self._fallback_repositories:
1761
raise errors.BzrError("Cannot commit from a lightweight checkout "
1762
"to a stacked branch. See "
1763
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1764
result = self._commit_builder_class(self, parents, config,
1765
timestamp, timezone, committer, revprops, revision_id)
1766
self.start_write_group()
258
return CommitBuilder(self, parents, config, timestamp, timezone,
259
committer, revprops, revision_id)
1769
@only_raises(errors.LockNotHeld, errors.LockBroken)
1770
261
def unlock(self):
1771
if (self.control_files._lock_count == 1 and
1772
self.control_files._lock_mode == 'w'):
1773
if self._write_group is not None:
1774
self.abort_write_group()
1775
self.control_files.unlock()
1776
raise errors.BzrError(
1777
'Must end write groups before releasing write locks.')
1778
262
self.control_files.unlock()
1779
if self.control_files._lock_count == 0:
1780
self._inventory_entry_cache.clear()
1781
for repo in self._fallback_repositories:
1784
264
@needs_read_lock
1785
def clone(self, a_bzrdir, revision_id=None):
265
def clone(self, a_bzrdir, revision_id=None, basis=None):
1786
266
"""Clone this repository into a_bzrdir using the current format.
1788
268
Currently no check is made that the format of this repository and
1789
269
the bzrdir format are compatible. FIXME RBC 20060201.
1791
:return: The newly created destination repository.
1793
# TODO: deprecate after 0.16; cloning this with all its settings is
1794
# probably not very useful -- mbp 20070423
1795
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1796
self.copy_content_into(dest_repo, revision_id)
1799
def start_write_group(self):
1800
"""Start a write group in the repository.
1802
Write groups are used by repositories which do not have a 1:1 mapping
1803
between file ids and backend store to manage the insertion of data from
1804
both fetch and commit operations.
1806
A write lock is required around the start_write_group/commit_write_group
1807
for the support of lock-requiring repository formats.
1809
One can only insert data into a repository inside a write group.
1813
if not self.is_write_locked():
1814
raise errors.NotWriteLocked(self)
1815
if self._write_group:
1816
raise errors.BzrError('already in a write group')
1817
self._start_write_group()
1818
# so we can detect unlock/relock - the write group is now entered.
1819
self._write_group = self.get_transaction()
1821
def _start_write_group(self):
1822
"""Template method for per-repository write group startup.
1824
This is called before the write group is considered to be
1829
def sprout(self, to_bzrdir, revision_id=None):
1830
"""Create a descendent repository for new development.
1832
Unlike clone, this does not copy the settings of the repository.
1834
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1835
dest_repo.fetch(self, revision_id=revision_id)
1838
def _create_sprouting_repo(self, a_bzrdir, shared):
1839
271
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1840
272
# use target default format.
1841
dest_repo = a_bzrdir.create_repository()
273
result = a_bzrdir.create_repository()
274
# FIXME RBC 20060209 split out the repository type to avoid this check ?
275
elif isinstance(a_bzrdir._format,
276
(bzrdir.BzrDirFormat4,
277
bzrdir.BzrDirFormat5,
278
bzrdir.BzrDirFormat6)):
279
result = a_bzrdir.open_repository()
1843
# Most control formats need the repository to be specifically
1844
# created, but on some old all-in-one formats it's not needed
1846
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1847
except errors.UninitializableFormat:
1848
dest_repo = a_bzrdir.open_repository()
1851
def _get_sink(self):
1852
"""Return a sink for streaming into this repository."""
1853
return StreamSink(self)
1855
def _get_source(self, to_format):
1856
"""Return a source for streaming from this repository."""
1857
return StreamSource(self, to_format)
281
result = self._format.initialize(a_bzrdir, shared=self.is_shared())
282
self.copy_content_into(result, revision_id, basis)
1859
285
@needs_read_lock
1860
286
def has_revision(self, revision_id):
1861
287
"""True if this repository has a copy of the revision."""
1862
return revision_id in self.has_revisions((revision_id,))
1865
def has_revisions(self, revision_ids):
1866
"""Probe to find out the presence of multiple revisions.
1868
:param revision_ids: An iterable of revision_ids.
1869
:return: A set of the revision_ids that were present.
1871
parent_map = self.revisions.get_parent_map(
1872
[(rev_id,) for rev_id in revision_ids])
1874
if _mod_revision.NULL_REVISION in revision_ids:
1875
result.add(_mod_revision.NULL_REVISION)
1876
result.update([key[0] for key in parent_map])
1880
def get_revision(self, revision_id):
1881
"""Return the Revision object for a named revision."""
1882
return self.get_revisions([revision_id])[0]
288
return self._revision_store.has_revision_id(revision_id,
289
self.get_transaction())
1884
291
@needs_read_lock
1885
292
def get_revision_reconcile(self, revision_id):
1886
293
"""'reconcile' helper routine that allows access to a revision always.
1888
295
This variant of get_revision does not cross check the weave graph
1889
296
against the revision one as get_revision does: but it should only
1890
297
be used by reconcile, or reconcile-alike commands that are correcting
1891
298
or testing the revision graph.
1893
return self._get_revisions([revision_id])[0]
300
if not revision_id or not isinstance(revision_id, basestring):
301
raise InvalidRevisionId(revision_id=revision_id, branch=self)
302
return self._revision_store.get_revisions([revision_id],
303
self.get_transaction())[0]
1895
304
@needs_read_lock
1896
305
def get_revisions(self, revision_ids):
1897
"""Get many revisions at once.
1899
Repositories that need to check data on every revision read should
1900
subclass this method.
1902
return self._get_revisions(revision_ids)
1905
def _get_revisions(self, revision_ids):
1906
"""Core work logic to get many revisions without sanity checks."""
1908
for revid, rev in self._iter_revisions(revision_ids):
1910
raise errors.NoSuchRevision(self, revid)
1912
return [revs[revid] for revid in revision_ids]
1914
def _iter_revisions(self, revision_ids):
1915
"""Iterate over revision objects.
1917
:param revision_ids: An iterable of revisions to examine. None may be
1918
passed to request all revisions known to the repository. Note that
1919
not all repositories can find unreferenced revisions; for those
1920
repositories only referenced ones will be returned.
1921
:return: An iterator of (revid, revision) tuples. Absent revisions (
1922
those asked for but not available) are returned as (revid, None).
1924
if revision_ids is None:
1925
revision_ids = self.all_revision_ids()
1927
for rev_id in revision_ids:
1928
if not rev_id or not isinstance(rev_id, basestring):
1929
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1930
keys = [(key,) for key in revision_ids]
1931
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1932
for record in stream:
1933
revid = record.key[0]
1934
if record.storage_kind == 'absent':
1937
text = record.get_bytes_as('fulltext')
1938
rev = self._serializer.read_revision_from_string(text)
1941
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
306
return self._revision_store.get_revisions(revision_ids,
307
self.get_transaction())
310
def get_revision_xml(self, revision_id):
311
rev = self.get_revision(revision_id)
313
# the current serializer..
314
self._revision_store._serializer.write_revision(rev, rev_tmp)
316
return rev_tmp.getvalue()
319
def get_revision(self, revision_id):
320
"""Return the Revision object for a named revision"""
321
r = self.get_revision_reconcile(revision_id)
322
# weave corruption can lead to absent revision markers that should be
324
# the following test is reasonably cheap (it needs a single weave read)
325
# and the weave is cached in read transactions. In write transactions
326
# it is not cached but typically we only read a small number of
327
# revisions. For knits when they are introduced we will probably want
328
# to ensure that caching write transactions are in use.
329
inv = self.get_inventory_weave()
330
self._check_revision_parents(r, inv)
334
def get_deltas_for_revisions(self, revisions):
1942
335
"""Produce a generator of revision deltas.
1944
337
Note that the input is a sequence of REVISIONS, not revision_ids.
1945
338
Trees will be held in memory until the generator exits.
1946
339
Each delta is relative to the revision's lefthand predecessor.
1948
:param specific_fileids: if not None, the result is filtered
1949
so that only those file-ids, their parents and their
1950
children are included.
1952
# Get the revision-ids of interest
1953
341
required_trees = set()
1954
342
for revision in revisions:
1955
343
required_trees.add(revision.revision_id)
1956
344
required_trees.update(revision.parent_ids[:1])
1958
# Get the matching filtered trees. Note that it's more
1959
# efficient to pass filtered trees to changes_from() rather
1960
# than doing the filtering afterwards. changes_from() could
1961
# arguably do the filtering itself but it's path-based, not
1962
# file-id based, so filtering before or afterwards is
1964
if specific_fileids is None:
1965
trees = dict((t.get_revision_id(), t) for
1966
t in self.revision_trees(required_trees))
1968
trees = dict((t.get_revision_id(), t) for
1969
t in self._filtered_revision_trees(required_trees,
1972
# Calculate the deltas
345
trees = dict((t.get_revision_id(), t) for
346
t in self.revision_trees(required_trees))
1973
347
for revision in revisions:
1974
348
if not revision.parent_ids:
1975
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
349
old_tree = self.revision_tree(None)
1977
351
old_tree = trees[revision.parent_ids[0]]
1978
352
yield trees[revision.revision_id].changes_from(old_tree)
1980
354
@needs_read_lock
1981
def get_revision_delta(self, revision_id, specific_fileids=None):
355
def get_revision_delta(self, revision_id):
1982
356
"""Return the delta for one revision.
1984
358
The delta is relative to the left-hand predecessor of the
1987
:param specific_fileids: if not None, the result is filtered
1988
so that only those file-ids, their parents and their
1989
children are included.
1991
361
r = self.get_revision(revision_id)
1992
return list(self.get_deltas_for_revisions([r],
1993
specific_fileids=specific_fileids))[0]
362
return list(self.get_deltas_for_revisions([r]))[0]
364
def _check_revision_parents(self, revision, inventory):
365
"""Private to Repository and Fetch.
367
This checks the parentage of revision in an inventory weave for
368
consistency and is only applicable to inventory-weave-for-ancestry
369
using repository formats & fetchers.
371
weave_parents = inventory.get_parents(revision.revision_id)
372
weave_names = inventory.versions()
373
for parent_id in revision.parent_ids:
374
if parent_id in weave_names:
375
# this parent must not be a ghost.
376
if not parent_id in weave_parents:
378
raise errors.CorruptRepository(self)
1995
380
@needs_write_lock
1996
381
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1997
382
signature = gpg_strategy.sign(plaintext)
1998
self.add_signature_text(revision_id, signature)
2001
def add_signature_text(self, revision_id, signature):
2002
self.signatures.add_lines((revision_id,), (),
2003
osutils.split_lines(signature))
2005
def find_text_key_references(self):
2006
"""Find the text key references within the repository.
2008
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2009
to whether they were referred to by the inventory of the
2010
revision_id that they contain. The inventory texts from all present
2011
revision ids are assessed to generate this report.
2013
revision_keys = self.revisions.keys()
2014
w = self.inventories
2015
pb = ui.ui_factory.nested_progress_bar()
2017
return self._find_text_key_references_from_xml_inventory_lines(
2018
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2022
def _find_text_key_references_from_xml_inventory_lines(self,
2024
"""Core routine for extracting references to texts from inventories.
2026
This performs the translation of xml lines to revision ids.
2028
:param line_iterator: An iterator of lines, origin_version_id
2029
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2030
to whether they were referred to by the inventory of the
2031
revision_id that they contain. Note that if that revision_id was
2032
not part of the line_iterator's output then False will be given -
2033
even though it may actually refer to that key.
2035
if not self._serializer.support_altered_by_hack:
2036
raise AssertionError(
2037
"_find_text_key_references_from_xml_inventory_lines only "
2038
"supported for branches which store inventory as unnested xml"
2039
", not on %r" % self)
383
self._revision_store.add_revision_signature_text(revision_id,
385
self.get_transaction())
387
def fileids_altered_by_revision_ids(self, revision_ids):
388
"""Find the file ids and versions affected by revisions.
390
:param revisions: an iterable containing revision ids.
391
:return: a dictionary mapping altered file-ids to an iterable of
392
revision_ids. Each altered file-ids has the exact revision_ids that
393
altered it listed explicitly.
395
assert isinstance(self._format, (RepositoryFormat5,
398
RepositoryFormatKnit1)), \
399
("fileids_altered_by_revision_ids only supported for branches "
400
"which store inventory as unnested xml, not on %r" % self)
401
selected_revision_ids = set(revision_ids)
402
w = self.get_inventory_weave()
2042
405
# this code needs to read every new line in every inventory for the
2043
406
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2044
# not present in one of those inventories is unnecessary but not
407
# not present in one of those inventories is unnecessary but not
2045
408
# harmful because we are filtering by the revision id marker in the
2046
# inventory lines : we only select file ids altered in one of those
409
# inventory lines : we only select file ids altered in one of those
2047
410
# revisions. We don't need to see all lines in the inventory because
2048
411
# only those added in an inventory in rev X can contain a revision=X
2050
unescape_revid_cache = {}
2051
unescape_fileid_cache = {}
2053
# jam 20061218 In a big fetch, this handles hundreds of thousands
2054
# of lines, so it has had a lot of inlining and optimizing done.
2055
# Sorry that it is a little bit messy.
2056
# Move several functions to be local variables, since this is a long
2058
search = self._file_ids_altered_regex.search
2059
unescape = _unescape_xml
2060
setdefault = result.setdefault
2061
for line, line_key in line_iterator:
2062
match = search(line)
2065
# One call to match.group() returning multiple items is quite a
2066
# bit faster than 2 calls to match.group() each returning 1
2067
file_id, revision_id = match.group('file_id', 'revision_id')
2069
# Inlining the cache lookups helps a lot when you make 170,000
2070
# lines and 350k ids, versus 8.4 unique ids.
2071
# Using a cache helps in 2 ways:
2072
# 1) Avoids unnecessary decoding calls
2073
# 2) Re-uses cached strings, which helps in future set and
2075
# (2) is enough that removing encoding entirely along with
2076
# the cache (so we are using plain strings) results in no
2077
# performance improvement.
2079
revision_id = unescape_revid_cache[revision_id]
2081
unescaped = unescape(revision_id)
2082
unescape_revid_cache[revision_id] = unescaped
2083
revision_id = unescaped
2085
# Note that unconditionally unescaping means that we deserialise
2086
# every fileid, which for general 'pull' is not great, but we don't
2087
# really want to have some many fulltexts that this matters anyway.
2090
file_id = unescape_fileid_cache[file_id]
2092
unescaped = unescape(file_id)
2093
unescape_fileid_cache[file_id] = unescaped
2096
key = (file_id, revision_id)
2097
setdefault(key, False)
2098
if revision_id == line_key[-1]:
2102
def _inventory_xml_lines_for_keys(self, keys):
2103
"""Get a line iterator of the sort needed for findind references.
2105
Not relevant for non-xml inventory repositories.
2107
Ghosts in revision_keys are ignored.
2109
:param revision_keys: The revision keys for the inventories to inspect.
2110
:return: An iterator over (inventory line, revid) for the fulltexts of
2111
all of the xml inventories specified by revision_keys.
2113
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2114
for record in stream:
2115
if record.storage_kind != 'absent':
2116
chunks = record.get_bytes_as('chunked')
2117
revid = record.key[-1]
2118
lines = osutils.chunks_to_lines(chunks)
2122
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2124
"""Helper routine for fileids_altered_by_revision_ids.
2126
This performs the translation of xml lines to revision ids.
2128
:param line_iterator: An iterator of lines, origin_version_id
2129
:param revision_keys: The revision ids to filter for. This should be a
2130
set or other type which supports efficient __contains__ lookups, as
2131
the revision key from each parsed line will be looked up in the
2132
revision_keys filter.
2133
:return: a dictionary mapping altered file-ids to an iterable of
2134
revision_ids. Each altered file-ids has the exact revision_ids that
2135
altered it listed explicitly.
2137
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2138
line_iterator).iterkeys())
2139
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2140
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2141
self._inventory_xml_lines_for_keys(parent_keys)))
2142
new_keys = seen - parent_seen
2144
setdefault = result.setdefault
2145
for key in new_keys:
2146
setdefault(key[0], set()).add(key[-1])
2149
def _find_parent_ids_of_revisions(self, revision_ids):
2150
"""Find all parent ids that are mentioned in the revision graph.
2152
:return: set of revisions that are parents of revision_ids which are
2153
not part of revision_ids themselves
2155
parent_map = self.get_parent_map(revision_ids)
2157
map(parent_ids.update, parent_map.itervalues())
2158
parent_ids.difference_update(revision_ids)
2159
parent_ids.discard(_mod_revision.NULL_REVISION)
2162
def _find_parent_keys_of_revisions(self, revision_keys):
2163
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2165
:param revision_keys: An iterable of revision_keys.
2166
:return: The parents of all revision_keys that are not already in
2169
parent_map = self.revisions.get_parent_map(revision_keys)
2171
map(parent_keys.update, parent_map.itervalues())
2172
parent_keys.difference_update(revision_keys)
2173
parent_keys.discard(_mod_revision.NULL_REVISION)
2176
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2177
"""Find the file ids and versions affected by revisions.
2179
:param revisions: an iterable containing revision ids.
2180
:param _inv_weave: The inventory weave from this repository or None.
2181
If None, the inventory weave will be opened automatically.
2182
:return: a dictionary mapping altered file-ids to an iterable of
2183
revision_ids. Each altered file-ids has the exact revision_ids that
2184
altered it listed explicitly.
2186
selected_keys = set((revid,) for revid in revision_ids)
2187
w = _inv_weave or self.inventories
2188
return self._find_file_ids_from_xml_inventory_lines(
2189
w.iter_lines_added_or_present_in_keys(
2190
selected_keys, pb=None),
2193
def iter_files_bytes(self, desired_files):
2194
"""Iterate through file versions.
2196
Files will not necessarily be returned in the order they occur in
2197
desired_files. No specific order is guaranteed.
2199
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2200
value supplied by the caller as part of desired_files. It should
2201
uniquely identify the file version in the caller's context. (Examples:
2202
an index number or a TreeTransform trans_id.)
2204
bytes_iterator is an iterable of bytestrings for the file. The
2205
kind of iterable and length of the bytestrings are unspecified, but for
2206
this implementation, it is a list of bytes produced by
2207
VersionedFile.get_record_stream().
2209
:param desired_files: a list of (file_id, revision_id, identifier)
2213
for file_id, revision_id, callable_data in desired_files:
2214
text_keys[(file_id, revision_id)] = callable_data
2215
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2216
if record.storage_kind == 'absent':
2217
raise errors.RevisionNotPresent(record.key, self)
2218
yield text_keys[record.key], record.get_bytes_as('chunked')
2220
def _generate_text_key_index(self, text_key_references=None,
2222
"""Generate a new text key index for the repository.
2224
This is an expensive function that will take considerable time to run.
2226
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2227
list of parents, also text keys. When a given key has no parents,
2228
the parents list will be [NULL_REVISION].
2230
# All revisions, to find inventory parents.
2231
if ancestors is None:
2232
graph = self.get_graph()
2233
ancestors = graph.get_parent_map(self.all_revision_ids())
2234
if text_key_references is None:
2235
text_key_references = self.find_text_key_references()
2236
pb = ui.ui_factory.nested_progress_bar()
2238
return self._do_generate_text_key_index(ancestors,
2239
text_key_references, pb)
2243
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2244
"""Helper for _generate_text_key_index to avoid deep nesting."""
2245
revision_order = tsort.topo_sort(ancestors)
2246
invalid_keys = set()
2248
for revision_id in revision_order:
2249
revision_keys[revision_id] = set()
2250
text_count = len(text_key_references)
2251
# a cache of the text keys to allow reuse; costs a dict of all the
2252
# keys, but saves a 2-tuple for every child of a given key.
2254
for text_key, valid in text_key_references.iteritems():
2256
invalid_keys.add(text_key)
2258
revision_keys[text_key[1]].add(text_key)
2259
text_key_cache[text_key] = text_key
2260
del text_key_references
2262
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2263
NULL_REVISION = _mod_revision.NULL_REVISION
2264
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2265
# too small for large or very branchy trees. However, for 55K path
2266
# trees, it would be easy to use too much memory trivially. Ideally we
2267
# could gauge this by looking at available real memory etc, but this is
2268
# always a tricky proposition.
2269
inventory_cache = lru_cache.LRUCache(10)
2270
batch_size = 10 # should be ~150MB on a 55K path tree
2271
batch_count = len(revision_order) / batch_size + 1
2273
pb.update("Calculating text parents", processed_texts, text_count)
2274
for offset in xrange(batch_count):
2275
to_query = revision_order[offset * batch_size:(offset + 1) *
2279
for revision_id in to_query:
2280
parent_ids = ancestors[revision_id]
2281
for text_key in revision_keys[revision_id]:
2282
pb.update("Calculating text parents", processed_texts)
2283
processed_texts += 1
2284
candidate_parents = []
2285
for parent_id in parent_ids:
2286
parent_text_key = (text_key[0], parent_id)
2288
check_parent = parent_text_key not in \
2289
revision_keys[parent_id]
2291
# the parent parent_id is a ghost:
2292
check_parent = False
2293
# truncate the derived graph against this ghost.
2294
parent_text_key = None
2296
# look at the parent commit details inventories to
2297
# determine possible candidates in the per file graph.
2300
inv = inventory_cache[parent_id]
2302
inv = self.revision_tree(parent_id).inventory
2303
inventory_cache[parent_id] = inv
2305
parent_entry = inv[text_key[0]]
2306
except (KeyError, errors.NoSuchId):
2308
if parent_entry is not None:
2310
text_key[0], parent_entry.revision)
2312
parent_text_key = None
2313
if parent_text_key is not None:
2314
candidate_parents.append(
2315
text_key_cache[parent_text_key])
2316
parent_heads = text_graph.heads(candidate_parents)
2317
new_parents = list(parent_heads)
2318
new_parents.sort(key=lambda x:candidate_parents.index(x))
2319
if new_parents == []:
2320
new_parents = [NULL_REVISION]
2321
text_index[text_key] = new_parents
2323
for text_key in invalid_keys:
2324
text_index[text_key] = [NULL_REVISION]
2327
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2328
"""Get an iterable listing the keys of all the data introduced by a set
2331
The keys will be ordered so that the corresponding items can be safely
2332
fetched and inserted in that order.
2334
:returns: An iterable producing tuples of (knit-kind, file-id,
2335
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2336
'revisions'. file-id is None unless knit-kind is 'file'.
2338
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2341
for result in self._find_non_file_keys_to_fetch(revision_ids):
2344
def _find_file_keys_to_fetch(self, revision_ids, pb):
2345
# XXX: it's a bit weird to control the inventory weave caching in this
2346
# generator. Ideally the caching would be done in fetch.py I think. Or
2347
# maybe this generator should explicitly have the contract that it
2348
# should not be iterated until the previously yielded item has been
2350
inv_w = self.inventories
2352
# file ids that changed
2353
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2355
num_file_ids = len(file_ids)
2356
for file_id, altered_versions in file_ids.iteritems():
2358
pb.update("Fetch texts", count, num_file_ids)
2360
yield ("file", file_id, altered_versions)
2362
def _find_non_file_keys_to_fetch(self, revision_ids):
2364
yield ("inventory", None, revision_ids)
2367
# XXX: Note ATM no callers actually pay attention to this return
2368
# instead they just use the list of revision ids and ignore
2369
# missing sigs. Consider removing this work entirely
2370
revisions_with_signatures = set(self.signatures.get_parent_map(
2371
[(r,) for r in revision_ids]))
2372
revisions_with_signatures = set(
2373
[r for (r,) in revisions_with_signatures])
2374
revisions_with_signatures.intersection_update(revision_ids)
2375
yield ("signatures", None, revisions_with_signatures)
2378
yield ("revisions", None, revision_ids)
413
for line in w.iter_lines_added_or_present_in_versions(selected_revision_ids):
414
start = line.find('file_id="')+9
415
if start < 9: continue
416
end = line.find('"', start)
418
file_id = _unescape_xml(line[start:end])
420
start = line.find('revision="')+10
421
if start < 10: continue
422
end = line.find('"', start)
424
revision_id = _unescape_xml(line[start:end])
425
if revision_id in selected_revision_ids:
426
result.setdefault(file_id, set()).add(revision_id)
430
def get_inventory_weave(self):
431
return self.control_weaves.get_weave('inventory',
432
self.get_transaction())
2380
434
@needs_read_lock
2381
435
def get_inventory(self, revision_id):
2382
"""Get Inventory object by revision id."""
2383
return self.iter_inventories([revision_id]).next()
2385
def iter_inventories(self, revision_ids, ordering=None):
2386
"""Get many inventories by revision_ids.
2388
This will buffer some or all of the texts used in constructing the
2389
inventories in memory, but will only parse a single inventory at a
2392
:param revision_ids: The expected revision ids of the inventories.
2393
:param ordering: optional ordering, e.g. 'topological'. If not
2394
specified, the order of revision_ids will be preserved (by
2395
buffering if necessary).
2396
:return: An iterator of inventories.
2398
if ((None in revision_ids)
2399
or (_mod_revision.NULL_REVISION in revision_ids)):
2400
raise ValueError('cannot get null revision inventory')
2401
return self._iter_inventories(revision_ids, ordering)
2403
def _iter_inventories(self, revision_ids, ordering):
2404
"""single-document based inventory iteration."""
2405
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2406
for text, revision_id in inv_xmls:
2407
yield self._deserialise_inventory(revision_id, text)
2409
def _iter_inventory_xmls(self, revision_ids, ordering):
2410
if ordering is None:
2411
order_as_requested = True
2412
ordering = 'unordered'
2414
order_as_requested = False
2415
keys = [(revision_id,) for revision_id in revision_ids]
2418
if order_as_requested:
2419
key_iter = iter(keys)
2420
next_key = key_iter.next()
2421
stream = self.inventories.get_record_stream(keys, ordering, True)
2423
for record in stream:
2424
if record.storage_kind != 'absent':
2425
chunks = record.get_bytes_as('chunked')
2426
if order_as_requested:
2427
text_chunks[record.key] = chunks
2429
yield ''.join(chunks), record.key[-1]
2431
raise errors.NoSuchRevision(self, record.key)
2432
if order_as_requested:
2433
# Yield as many results as we can while preserving order.
2434
while next_key in text_chunks:
2435
chunks = text_chunks.pop(next_key)
2436
yield ''.join(chunks), next_key[-1]
2438
next_key = key_iter.next()
2439
except StopIteration:
2440
# We still want to fully consume the get_record_stream,
2441
# just in case it is not actually finished at this point
2445
def _deserialise_inventory(self, revision_id, xml):
2446
"""Transform the xml into an inventory object.
436
"""Get Inventory object by hash."""
437
return self.deserialise_inventory(
438
revision_id, self.get_inventory_xml(revision_id))
440
def deserialise_inventory(self, revision_id, xml):
441
"""Transform the xml into an inventory object.
2448
443
:param revision_id: The expected revision id of the inventory.
2449
444
:param xml: A serialised inventory.
2451
result = self._serializer.read_inventory_from_string(xml, revision_id,
2452
entry_cache=self._inventory_entry_cache,
2453
return_from_cache=self._safe_to_return_from_cache)
2454
if result.revision_id != revision_id:
2455
raise AssertionError('revision id mismatch %s != %s' % (
2456
result.revision_id, revision_id))
2459
def get_serializer_format(self):
2460
return self._serializer.format_num
446
return xml5.serializer_v5.read_inventory_from_string(xml)
2462
448
@needs_read_lock
2463
def _get_inventory_xml(self, revision_id):
2464
"""Get serialized inventory as a string."""
2465
texts = self._iter_inventory_xmls([revision_id], 'unordered')
449
def get_inventory_xml(self, revision_id):
450
"""Get inventory XML as a file object."""
2467
text, revision_id = texts.next()
2468
except StopIteration:
452
assert isinstance(revision_id, basestring), type(revision_id)
453
iw = self.get_inventory_weave()
454
return iw.get_text(revision_id)
2469
456
raise errors.HistoryMissing(self, 'inventory', revision_id)
2472
def get_rev_id_for_revno(self, revno, known_pair):
2473
"""Return the revision id of a revno, given a later (revno, revid)
2474
pair in the same history.
2476
:return: if found (True, revid). If the available history ran out
2477
before reaching the revno, then this returns
2478
(False, (closest_revno, closest_revid)).
2480
known_revno, known_revid = known_pair
2481
partial_history = [known_revid]
2482
distance_from_known = known_revno - revno
2483
if distance_from_known < 0:
2485
'requested revno (%d) is later than given known revno (%d)'
2486
% (revno, known_revno))
2489
self, partial_history, stop_index=distance_from_known)
2490
except errors.RevisionNotPresent, err:
2491
if err.revision_id == known_revid:
2492
# The start revision (known_revid) wasn't found.
2494
# This is a stacked repository with no fallbacks, or a there's a
2495
# left-hand ghost. Either way, even though the revision named in
2496
# the error isn't in this repo, we know it's the next step in this
2497
# left-hand history.
2498
partial_history.append(err.revision_id)
2499
if len(partial_history) <= distance_from_known:
2500
# Didn't find enough history to get a revid for the revno.
2501
earliest_revno = known_revno - len(partial_history) + 1
2502
return (False, (earliest_revno, partial_history[-1]))
2503
if len(partial_history) - 1 > distance_from_known:
2504
raise AssertionError('_iter_for_revno returned too much history')
2505
return (True, partial_history[-1])
2507
def iter_reverse_revision_history(self, revision_id):
2508
"""Iterate backwards through revision ids in the lefthand history
2510
:param revision_id: The revision id to start with. All its lefthand
2511
ancestors will be traversed.
2513
graph = self.get_graph()
2514
next_id = revision_id
2516
if next_id in (None, _mod_revision.NULL_REVISION):
459
def get_inventory_sha1(self, revision_id):
460
"""Return the sha1 hash of the inventory entry
462
return self.get_revision(revision_id).inventory_sha1
465
def get_revision_graph(self, revision_id=None):
466
"""Return a dictionary containing the revision graph.
468
:param revision_id: The revision_id to get a graph from. If None, then
469
the entire revision graph is returned. This is a deprecated mode of
470
operation and will be removed in the future.
471
:return: a dictionary of revision_id->revision_parents_list.
473
# special case NULL_REVISION
474
if revision_id == NULL_REVISION:
476
weave = self.get_inventory_weave()
477
all_revisions = self._eliminate_revisions_not_present(weave.versions())
478
entire_graph = dict([(node, weave.get_parents(node)) for
479
node in all_revisions])
480
if revision_id is None:
482
elif revision_id not in entire_graph:
483
raise errors.NoSuchRevision(self, revision_id)
485
# add what can be reached from revision_id
487
pending = set([revision_id])
488
while len(pending) > 0:
490
result[node] = entire_graph[node]
491
for revision_id in result[node]:
492
if revision_id not in result:
493
pending.add(revision_id)
497
def get_revision_graph_with_ghosts(self, revision_ids=None):
498
"""Return a graph of the revisions with ghosts marked as applicable.
500
:param revision_ids: an iterable of revisions to graph or None for all.
501
:return: a Graph object with the graph reachable from revision_ids.
505
pending = set(self.all_revision_ids())
508
pending = set(revision_ids)
509
# special case NULL_REVISION
510
if NULL_REVISION in pending:
511
pending.remove(NULL_REVISION)
512
required = set(pending)
515
revision_id = pending.pop()
2519
parents = graph.get_parent_map([next_id])[next_id]
2521
raise errors.RevisionNotPresent(next_id, self)
2523
if len(parents) == 0:
2526
next_id = parents[0]
517
rev = self.get_revision(revision_id)
518
except errors.NoSuchRevision:
519
if revision_id in required:
522
result.add_ghost(revision_id)
524
for parent_id in rev.parent_ids:
525
# is this queued or done ?
526
if (parent_id not in pending and
527
parent_id not in done):
529
pending.add(parent_id)
530
result.add_node(revision_id, rev.parent_ids)
531
done.add(revision_id)
535
def get_revision_inventory(self, revision_id):
536
"""Return inventory of a past revision."""
537
# TODO: Unify this with get_inventory()
538
# bzr 0.0.6 and later imposes the constraint that the inventory_id
539
# must be the same as its revision, so this is trivial.
540
if revision_id is None:
541
# This does not make sense: if there is no revision,
542
# then it is the current tree inventory surely ?!
543
# and thus get_root_id() is something that looks at the last
544
# commit on the branch, and the get_root_id is an inventory check.
545
raise NotImplementedError
546
# return Inventory(self.get_root_id())
548
return self.get_inventory(revision_id)
2528
551
def is_shared(self):
2529
552
"""Return True if this repository is flagged as a shared repository."""
2530
553
raise NotImplementedError(self.is_shared)
2536
559
reconciler = RepoReconciler(self, thorough=thorough)
2537
560
reconciler.reconcile()
2538
561
return reconciler
2540
def _refresh_data(self):
2541
"""Helper called from lock_* to ensure coherency with disk.
2543
The default implementation does nothing; it is however possible
2544
for repositories to maintain loaded indices across multiple locks
2545
by checking inside their implementation of this method to see
2546
whether their indices are still valid. This depends of course on
2547
the disk format being validatable in this manner. This method is
2548
also called by the refresh_data() public interface to cause a refresh
2549
to occur while in a write lock so that data inserted by a smart server
2550
push operation is visible on the client's instance of the physical
2554
563
@needs_read_lock
2555
564
def revision_tree(self, revision_id):
2556
565
"""Return Tree for a revision on this branch.
2558
`revision_id` may be NULL_REVISION for the empty tree revision.
567
`revision_id` may be None for the empty tree revision.
2560
revision_id = _mod_revision.ensure_null(revision_id)
2561
569
# TODO: refactor this to use an existing revision object
2562
570
# so we don't need to read it in twice.
2563
if revision_id == _mod_revision.NULL_REVISION:
2564
return RevisionTree(self, Inventory(root_id=None),
2565
_mod_revision.NULL_REVISION)
571
if revision_id is None or revision_id == NULL_REVISION:
572
return RevisionTree(self, Inventory(), NULL_REVISION)
2567
inv = self.get_inventory(revision_id)
574
inv = self.get_revision_inventory(revision_id)
2568
575
return RevisionTree(self, inv, revision_id)
2570
578
def revision_trees(self, revision_ids):
2571
"""Return Trees for revisions in this repository.
2573
:param revision_ids: a sequence of revision-ids;
2574
a revision-id may not be None or 'null:'
2576
inventories = self.iter_inventories(revision_ids)
2577
for inv in inventories:
2578
yield RevisionTree(self, inv, inv.revision_id)
2580
def _filtered_revision_trees(self, revision_ids, file_ids):
2581
"""Return Tree for a revision on this branch with only some files.
2583
:param revision_ids: a sequence of revision-ids;
2584
a revision-id may not be None or 'null:'
2585
:param file_ids: if not None, the result is filtered
2586
so that only those file-ids, their parents and their
2587
children are included.
2589
inventories = self.iter_inventories(revision_ids)
2590
for inv in inventories:
2591
# Should we introduce a FilteredRevisionTree class rather
2592
# than pre-filter the inventory here?
2593
filtered_inv = inv.filter(file_ids)
2594
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
579
"""Return Tree for a revision on this branch.
581
`revision_id` may not be None or 'null:'"""
582
assert None not in revision_ids
583
assert NULL_REVISION not in revision_ids
584
texts = self.get_inventory_weave().get_texts(revision_ids)
585
for text, revision_id in zip(texts, revision_ids):
586
inv = self.deserialise_inventory(revision_id, text)
587
yield RevisionTree(self, inv, revision_id)
2596
589
@needs_read_lock
2597
def get_ancestry(self, revision_id, topo_sorted=True):
590
def get_ancestry(self, revision_id):
2598
591
"""Return a list of revision-ids integrated by a revision.
2600
The first element of the list is always None, indicating the origin
2601
revision. This might change when we have history horizons, or
593
The first element of the list is always None, indicating the origin
594
revision. This might change when we have history horizons, or
2602
595
perhaps we should have a new API.
2604
597
This is topologically sorted.
2606
if _mod_revision.is_null(revision_id):
599
if revision_id is None:
2608
601
if not self.has_revision(revision_id):
2609
602
raise errors.NoSuchRevision(self, revision_id)
2610
graph = self.get_graph()
2612
search = graph._make_breadth_first_searcher([revision_id])
2615
found, ghosts = search.next_with_ghosts()
2616
except StopIteration:
2619
if _mod_revision.NULL_REVISION in keys:
2620
keys.remove(_mod_revision.NULL_REVISION)
2622
parent_map = graph.get_parent_map(keys)
2623
keys = tsort.topo_sort(parent_map)
2624
return [None] + list(keys)
2626
def pack(self, hint=None, clean_obsolete_packs=False):
2627
"""Compress the data within the repository.
2629
This operation only makes sense for some repository types. For other
2630
types it should be a no-op that just returns.
2632
This stub method does not require a lock, but subclasses should use
2633
@needs_write_lock as this is a long running call its reasonable to
2634
implicitly lock for the user.
2636
:param hint: If not supplied, the whole repository is packed.
2637
If supplied, the repository may use the hint parameter as a
2638
hint for the parts of the repository to pack. A hint can be
2639
obtained from the result of commit_write_group(). Out of
2640
date hints are simply ignored, because concurrent operations
2641
can obsolete them rapidly.
2643
:param clean_obsolete_packs: Clean obsolete packs immediately after
603
w = self.get_inventory_weave()
604
candidates = w.get_ancestry(revision_id)
605
return [None] + candidates # self._eliminate_revisions_not_present(candidates)
608
def print_file(self, file, revision_id):
609
"""Print `file` to stdout.
611
FIXME RBC 20060125 as John Meinel points out this is a bad api
612
- it writes to stdout, it assumes that that is valid etc. Fix
613
by creating a new more flexible convenience function.
615
tree = self.revision_tree(revision_id)
616
# use inventory as it was in that revision
617
file_id = tree.inventory.path2id(file)
619
# TODO: jam 20060427 Write a test for this code path
620
# it had a bug in it, and was raising the wrong
622
raise errors.BzrError("%r is not present in revision %s" % (file, revision_id))
623
tree.print_file(file_id)
2647
625
def get_transaction(self):
2648
626
return self.control_files.get_transaction()
2650
def get_parent_map(self, revision_ids):
2651
"""See graph.StackedParentsProvider.get_parent_map"""
2652
# revisions index works in keys; this just works in revisions
2653
# therefore wrap and unwrap
2656
for revision_id in revision_ids:
2657
if revision_id == _mod_revision.NULL_REVISION:
2658
result[revision_id] = ()
2659
elif revision_id is None:
2660
raise ValueError('get_parent_map(None) is not valid')
2662
query_keys.append((revision_id ,))
2663
for ((revision_id,), parent_keys) in \
2664
self.revisions.get_parent_map(query_keys).iteritems():
2666
result[revision_id] = tuple([parent_revid
2667
for (parent_revid,) in parent_keys])
2669
result[revision_id] = (_mod_revision.NULL_REVISION,)
2672
def _make_parents_provider(self):
2676
def get_known_graph_ancestry(self, revision_ids):
2677
"""Return the known graph for a set of revision ids and their ancestors.
2679
st = static_tuple.StaticTuple
2680
revision_keys = [st(r_id).intern() for r_id in revision_ids]
2681
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
2682
return graph.GraphThunkIdsToKeys(known_graph)
2684
def get_graph(self, other_repository=None):
2685
"""Return the graph walker for this repository format"""
2686
parents_provider = self._make_parents_provider()
2687
if (other_repository is not None and
2688
not self.has_same_location(other_repository)):
2689
parents_provider = graph.StackedParentsProvider(
2690
[parents_provider, other_repository._make_parents_provider()])
2691
return graph.Graph(parents_provider)
2693
def _get_versioned_file_checker(self, text_key_references=None,
2695
"""Return an object suitable for checking versioned files.
2697
:param text_key_references: if non-None, an already built
2698
dictionary mapping text keys ((fileid, revision_id) tuples)
2699
to whether they were referred to by the inventory of the
2700
revision_id that they contain. If None, this will be
2702
:param ancestors: Optional result from
2703
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2706
return _VersionedFileChecker(self,
2707
text_key_references=text_key_references, ancestors=ancestors)
2709
def revision_ids_to_search_result(self, result_set):
2710
"""Convert a set of revision ids to a graph SearchResult."""
2711
result_parents = set()
2712
for parents in self.get_graph().get_parent_map(
2713
result_set).itervalues():
2714
result_parents.update(parents)
2715
included_keys = result_set.intersection(result_parents)
2716
start_keys = result_set.difference(included_keys)
2717
exclude_keys = result_parents.difference(result_set)
2718
result = graph.SearchResult(start_keys, exclude_keys,
2719
len(result_set), result_set)
628
def revision_parents(self, revid):
629
return self.get_inventory_weave().parent_names(revid)
2722
631
@needs_write_lock
2723
632
def set_make_working_trees(self, new_value):
2983
self._transport.delete('no-working-trees')
816
self.control_files._transport.delete('no-working-trees')
2984
817
except errors.NoSuchFile:
2987
self._transport.put_bytes('no-working-trees', '',
2988
mode=self.bzrdir._get_file_mode())
820
self.control_files.put_utf8('no-working-trees', '')
2990
822
def make_working_trees(self):
2991
823
"""Returns the policy for making working trees on new branches."""
2992
return not self._transport.has('no-working-trees')
2995
class MetaDirVersionedFileRepository(MetaDirRepository):
2996
"""Repositories in a meta-dir, that work via versioned file objects."""
2998
def __init__(self, _format, a_bzrdir, control_files):
2999
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
3003
network_format_registry = registry.FormatRegistry()
3004
"""Registry of formats indexed by their network name.
3006
The network name for a repository format is an identifier that can be used when
3007
referring to formats with smart server operations. See
3008
RepositoryFormat.network_name() for more detail.
3012
format_registry = registry.FormatRegistry(network_format_registry)
3013
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
3015
This can contain either format instances themselves, or classes/factories that
3016
can be called to obtain one.
3020
#####################################################################
3021
# Repository Formats
824
return not self.control_files._transport.has('no-working-trees')
827
class KnitRepository(MetaDirRepository):
828
"""Knit format repository."""
830
def _inventory_add_lines(self, inv_vf, revid, parents, lines):
831
inv_vf.add_lines_with_ghosts(revid, parents, lines)
834
def _all_revision_ids(self):
835
"""See Repository.all_revision_ids()."""
836
# Knits get the revision graph from the index of the revision knit, so
837
# it's always possible even if they're on an unlistable transport.
838
return self._revision_store.all_revision_ids(self.get_transaction())
840
def fileid_involved_between_revs(self, from_revid, to_revid):
841
"""Find file_id(s) which are involved in the changes between revisions.
843
This determines the set of revisions which are involved, and then
844
finds all file ids affected by those revisions.
846
vf = self._get_revision_vf()
847
from_set = set(vf.get_ancestry(from_revid))
848
to_set = set(vf.get_ancestry(to_revid))
849
changed = to_set.difference(from_set)
850
return self._fileid_involved_by_set(changed)
852
def fileid_involved(self, last_revid=None):
853
"""Find all file_ids modified in the ancestry of last_revid.
855
:param last_revid: If None, last_revision() will be used.
858
changed = set(self.all_revision_ids())
860
changed = set(self.get_ancestry(last_revid))
863
return self._fileid_involved_by_set(changed)
866
def get_ancestry(self, revision_id):
867
"""Return a list of revision-ids integrated by a revision.
869
This is topologically sorted.
871
if revision_id is None:
873
vf = self._get_revision_vf()
875
return [None] + vf.get_ancestry(revision_id)
876
except errors.RevisionNotPresent:
877
raise errors.NoSuchRevision(self, revision_id)
880
def get_revision(self, revision_id):
881
"""Return the Revision object for a named revision"""
882
return self.get_revision_reconcile(revision_id)
885
def get_revision_graph(self, revision_id=None):
886
"""Return a dictionary containing the revision graph.
888
:param revision_id: The revision_id to get a graph from. If None, then
889
the entire revision graph is returned. This is a deprecated mode of
890
operation and will be removed in the future.
891
:return: a dictionary of revision_id->revision_parents_list.
893
# special case NULL_REVISION
894
if revision_id == NULL_REVISION:
896
weave = self._get_revision_vf()
897
entire_graph = weave.get_graph()
898
if revision_id is None:
899
return weave.get_graph()
900
elif revision_id not in weave:
901
raise errors.NoSuchRevision(self, revision_id)
903
# add what can be reached from revision_id
905
pending = set([revision_id])
906
while len(pending) > 0:
908
result[node] = weave.get_parents(node)
909
for revision_id in result[node]:
910
if revision_id not in result:
911
pending.add(revision_id)
915
def get_revision_graph_with_ghosts(self, revision_ids=None):
916
"""Return a graph of the revisions with ghosts marked as applicable.
918
:param revision_ids: an iterable of revisions to graph or None for all.
919
:return: a Graph object with the graph reachable from revision_ids.
922
vf = self._get_revision_vf()
923
versions = set(vf.versions())
925
pending = set(self.all_revision_ids())
928
pending = set(revision_ids)
929
# special case NULL_REVISION
930
if NULL_REVISION in pending:
931
pending.remove(NULL_REVISION)
932
required = set(pending)
935
revision_id = pending.pop()
936
if not revision_id in versions:
937
if revision_id in required:
938
raise errors.NoSuchRevision(self, revision_id)
940
result.add_ghost(revision_id)
941
# mark it as done so we don't try for it again.
942
done.add(revision_id)
944
parent_ids = vf.get_parents_with_ghosts(revision_id)
945
for parent_id in parent_ids:
946
# is this queued or done ?
947
if (parent_id not in pending and
948
parent_id not in done):
950
pending.add(parent_id)
951
result.add_node(revision_id, parent_ids)
952
done.add(revision_id)
955
def _get_revision_vf(self):
956
""":return: a versioned file containing the revisions."""
957
vf = self._revision_store.get_revision_file(self.get_transaction())
961
def reconcile(self, other=None, thorough=False):
962
"""Reconcile this repository."""
963
from bzrlib.reconcile import KnitReconciler
964
reconciler = KnitReconciler(self, thorough=thorough)
965
reconciler.reconcile()
968
def revision_parents(self, revision_id):
969
return self._get_revision_vf().get_parents(revision_id)
3023
972
class RepositoryFormat(object):
3024
973
"""A repository format.
3026
Formats provide four things:
975
Formats provide three things:
3027
976
* An initialization routine to construct repository data on disk.
3028
* a optional format string which is used when the BzrDir supports
977
* a format string which is used when the BzrDir supports versioned
3030
979
* an open routine which returns a Repository instance.
3031
* A network name for referring to the format in smart server RPC
3034
There is one and only one Format subclass for each on-disk format. But
3035
there can be one Repository subclass that is used for several different
3036
formats. The _format attribute on a Repository instance can be used to
3037
determine the disk format.
3039
Formats are placed in a registry by their format string for reference
3040
during opening. These should be subclasses of RepositoryFormat for
981
Formats are placed in an dict by their format string for reference
982
during opening. These should be subclasses of RepositoryFormat
3043
985
Once a format is deprecated, just deprecate the initialize and open
3044
methods on the format class. Do not deprecate the object, as the
3045
object may be created even when a repository instance hasn't been
986
methods on the format class. Do not deprecate the object, as the
987
object will be created every system load.
3048
989
Common instance attributes:
3049
990
_matchingbzrdir - the bzrdir format that the repository format was
3050
991
originally written to work with. This can be used if manually
3051
992
constructing a bzrdir and repository, or more commonly for test suite
3055
# Set to True or False in derived classes. True indicates that the format
3056
# supports ghosts gracefully.
3057
supports_ghosts = None
3058
# Can this repository be given external locations to lookup additional
3059
# data. Set to True or False in derived classes.
3060
supports_external_lookups = None
3061
# Does this format support CHK bytestring lookups. Set to True or False in
3063
supports_chks = None
3064
# Should commit add an inventory, or an inventory delta to the repository.
3065
_commit_inv_deltas = True
3066
# What order should fetch operations request streams in?
3067
# The default is unordered as that is the cheapest for an origin to
3069
_fetch_order = 'unordered'
3070
# Does this repository format use deltas that can be fetched as-deltas ?
3071
# (E.g. knits, where the knit deltas can be transplanted intact.
3072
# We default to False, which will ensure that enough data to get
3073
# a full text out of any fetch stream will be grabbed.
3074
_fetch_uses_deltas = False
3075
# Should fetch trigger a reconcile after the fetch? Only needed for
3076
# some repository formats that can suffer internal inconsistencies.
3077
_fetch_reconcile = False
3078
# Does this format have < O(tree_size) delta generation. Used to hint what
3079
# code path for commit, amongst other things.
3081
# Does doing a pack operation compress data? Useful for the pack UI command
3082
# (so if there is one pack, the operation can still proceed because it may
3083
# help), and for fetching when data won't have come from the same
3085
pack_compresses = False
3086
# Does the repository inventory storage understand references to trees?
3087
supports_tree_reference = None
3088
# Is the format experimental ?
3089
experimental = False
3092
return "%s()" % self.__class__.__name__
3094
def __eq__(self, other):
3095
# format objects are generally stateless
3096
return isinstance(other, self.__class__)
3098
def __ne__(self, other):
3099
return not self == other
996
_default_format = None
997
"""The default format used for new repositories."""
1000
"""The known formats."""
3102
1003
def find_format(klass, a_bzrdir):
3103
"""Return the format for the repository object in a_bzrdir.
3105
This is used by bzr native formats that have a "format" file in
3106
the repository. Other methods may be used by different types of
1004
"""Return the format for the repository object in a_bzrdir."""
3110
1006
transport = a_bzrdir.get_repository_transport(None)
3111
format_string = transport.get_bytes("format")
3112
return format_registry.get(format_string)
1007
format_string = transport.get("format").read()
1008
return klass._formats[format_string]
3113
1009
except errors.NoSuchFile:
3114
1010
raise errors.NoRepositoryPresent(a_bzrdir)
3115
1011
except KeyError:
3116
raise errors.UnknownFormatError(format=format_string,
3120
def register_format(klass, format):
3121
format_registry.register(format.get_format_string(), format)
3124
def unregister_format(klass, format):
3125
format_registry.remove(format.get_format_string())
1012
raise errors.UnknownFormatError(format=format_string)
1014
def _get_control_store(self, repo_transport, control_files):
1015
"""Return the control store for this repository."""
1016
raise NotImplementedError(self._get_control_store)
3128
1019
def get_default_format(klass):
3129
1020
"""Return the current default format."""
3130
from bzrlib import bzrdir
3131
return bzrdir.format_registry.make_bzrdir('default').repository_format
1021
return klass._default_format
3133
1023
def get_format_string(self):
3134
1024
"""Return the ASCII format string that identifies this format.
3136
Note that in pre format ?? repositories the format string is
1026
Note that in pre format ?? repositories the format string is
3137
1027
not permitted nor written to disk.
3139
1029
raise NotImplementedError(self.get_format_string)
3170
1081
:param a_bzrdir: The bzrdir to put the new repository in it.
3171
1082
:param shared: The repository should be initialized as a sharable one.
3172
:returns: The new repository object.
3174
1084
This may raise UninitializableFormat if shared repository are not
3175
1085
compatible the a_bzrdir.
3177
raise NotImplementedError(self.initialize)
3179
1088
def is_supported(self):
3180
1089
"""Is this format supported?
3182
1091
Supported formats must be initializable and openable.
3183
Unsupported formats may not support initialization or committing or
1092
Unsupported formats may not support initialization or committing or
3184
1093
some other features depending on the reason for not being supported.
3188
def network_name(self):
3189
"""A simple byte string uniquely identifying this format for RPC calls.
3191
MetaDir repository formats use their disk format string to identify the
3192
repository over the wire. All in one formats such as bzr < 0.8, and
3193
foreign formats like svn/git and hg should use some marker which is
3194
unique and immutable.
3196
raise NotImplementedError(self.network_name)
3198
def check_conversion_target(self, target_format):
3199
if self.rich_root_data and not target_format.rich_root_data:
3200
raise errors.BadConversionTarget(
3201
'Does not support rich root data.', target_format,
3203
if (self.supports_tree_reference and
3204
not getattr(target_format, 'supports_tree_reference', False)):
3205
raise errors.BadConversionTarget(
3206
'Does not support nested trees', target_format,
3209
1097
def open(self, a_bzrdir, _found=False):
3210
1098
"""Return an instance of this format for the bzrdir a_bzrdir.
3212
1100
_found is a private parameter, do not use it.
3214
1102
raise NotImplementedError(self.open)
3216
def _run_post_repo_init_hooks(self, repository, a_bzrdir, shared):
3217
from bzrlib.bzrdir import BzrDir, RepoInitHookParams
3218
hooks = BzrDir.hooks['post_repo_init']
3221
params = RepoInitHookParams(repository, self, a_bzrdir, shared)
1105
def register_format(klass, format):
1106
klass._formats[format.get_format_string()] = format
1109
def set_default_format(klass, format):
1110
klass._default_format = format
1113
def unregister_format(klass, format):
1114
assert klass._formats[format.get_format_string()] is format
1115
del klass._formats[format.get_format_string()]
1118
class PreSplitOutRepositoryFormat(RepositoryFormat):
1119
"""Base class for the pre split out repository formats."""
1121
def initialize(self, a_bzrdir, shared=False, _internal=False):
1122
"""Create a weave repository.
1124
TODO: when creating split out bzr branch formats, move this to a common
1125
base for Format5, Format6. or something like that.
1127
from bzrlib.weavefile import write_weave_v5
1128
from bzrlib.weave import Weave
1131
raise errors.IncompatibleFormat(self, a_bzrdir._format)
1134
# always initialized when the bzrdir is.
1135
return self.open(a_bzrdir, _found=True)
1137
# Create an empty weave
1139
write_weave_v5(Weave(), sio)
1140
empty_weave = sio.getvalue()
1142
mutter('creating repository in %s.', a_bzrdir.transport.base)
1143
dirs = ['revision-store', 'weaves']
1144
files = [('inventory.weave', StringIO(empty_weave)),
1147
# FIXME: RBC 20060125 don't peek under the covers
1148
# NB: no need to escape relative paths that are url safe.
1149
control_files = LockableFiles(a_bzrdir.transport, 'branch-lock',
1151
control_files.create_lock()
1152
control_files.lock_write()
1153
control_files._transport.mkdir_multi(dirs,
1154
mode=control_files._dir_mode)
1156
for file, content in files:
1157
control_files.put(file, content)
1159
control_files.unlock()
1160
return self.open(a_bzrdir, _found=True)
1162
def _get_control_store(self, repo_transport, control_files):
1163
"""Return the control store for this repository."""
1164
return self._get_versioned_file_store('',
1169
def _get_text_store(self, transport, control_files):
1170
"""Get a store for file texts for this format."""
1171
raise NotImplementedError(self._get_text_store)
1173
def open(self, a_bzrdir, _found=False):
1174
"""See RepositoryFormat.open()."""
1176
# we are being called directly and must probe.
1177
raise NotImplementedError
1179
repo_transport = a_bzrdir.get_repository_transport(None)
1180
control_files = a_bzrdir._control_files
1181
text_store = self._get_text_store(repo_transport, control_files)
1182
control_store = self._get_control_store(repo_transport, control_files)
1183
_revision_store = self._get_revision_store(repo_transport, control_files)
1184
return AllInOneRepository(_format=self,
1186
_revision_store=_revision_store,
1187
control_store=control_store,
1188
text_store=text_store)
1191
class RepositoryFormat4(PreSplitOutRepositoryFormat):
1192
"""Bzr repository format 4.
1194
This repository format has:
1196
- TextStores for texts, inventories,revisions.
1198
This format is deprecated: it indexes texts using a text id which is
1199
removed in format 5; initialization and write support for this format
1204
super(RepositoryFormat4, self).__init__()
1205
self._matchingbzrdir = bzrdir.BzrDirFormat4()
1207
def get_format_description(self):
1208
"""See RepositoryFormat.get_format_description()."""
1209
return "Repository format 4"
1211
def initialize(self, url, shared=False, _internal=False):
1212
"""Format 4 branches cannot be created."""
1213
raise errors.UninitializableFormat(self)
1215
def is_supported(self):
1216
"""Format 4 is not supported.
1218
It is not supported because the model changed from 4 to 5 and the
1219
conversion logic is expensive - so doing it on the fly was not
1224
def _get_control_store(self, repo_transport, control_files):
1225
"""Format 4 repositories have no formal control store at this point.
1227
This will cause any control-file-needing apis to fail - this is desired.
1231
def _get_revision_store(self, repo_transport, control_files):
1232
"""See RepositoryFormat._get_revision_store()."""
1233
from bzrlib.xml4 import serializer_v4
1234
return self._get_text_rev_store(repo_transport,
1237
serializer=serializer_v4)
1239
def _get_text_store(self, transport, control_files):
1240
"""See RepositoryFormat._get_text_store()."""
1243
class RepositoryFormat5(PreSplitOutRepositoryFormat):
1244
"""Bzr control format 5.
1246
This repository format has:
1247
- weaves for file texts and inventory
1249
- TextStores for revisions and signatures.
1253
super(RepositoryFormat5, self).__init__()
1254
self._matchingbzrdir = bzrdir.BzrDirFormat5()
1256
def get_format_description(self):
1257
"""See RepositoryFormat.get_format_description()."""
1258
return "Weave repository format 5"
1260
def _get_revision_store(self, repo_transport, control_files):
1261
"""See RepositoryFormat._get_revision_store()."""
1262
"""Return the revision store object for this a_bzrdir."""
1263
return self._get_text_rev_store(repo_transport,
1268
def _get_text_store(self, transport, control_files):
1269
"""See RepositoryFormat._get_text_store()."""
1270
return self._get_versioned_file_store('weaves', transport, control_files, prefixed=False)
1273
class RepositoryFormat6(PreSplitOutRepositoryFormat):
1274
"""Bzr control format 6.
1276
This repository format has:
1277
- weaves for file texts and inventory
1278
- hash subdirectory based stores.
1279
- TextStores for revisions and signatures.
1283
super(RepositoryFormat6, self).__init__()
1284
self._matchingbzrdir = bzrdir.BzrDirFormat6()
1286
def get_format_description(self):
1287
"""See RepositoryFormat.get_format_description()."""
1288
return "Weave repository format 6"
1290
def _get_revision_store(self, repo_transport, control_files):
1291
"""See RepositoryFormat._get_revision_store()."""
1292
return self._get_text_rev_store(repo_transport,
1298
def _get_text_store(self, transport, control_files):
1299
"""See RepositoryFormat._get_text_store()."""
1300
return self._get_versioned_file_store('weaves', transport, control_files)
3226
1303
class MetaDirRepositoryFormat(RepositoryFormat):
3227
1304
"""Common base class for the new repositories using the metadir layout."""
3229
rich_root_data = False
3230
supports_tree_reference = False
3231
supports_external_lookups = False
3234
def _matchingbzrdir(self):
3235
matching = bzrdir.BzrDirMetaFormat1()
3236
matching.repository_format = self
3239
1306
def __init__(self):
3240
1307
super(MetaDirRepositoryFormat, self).__init__()
1308
self._matchingbzrdir = bzrdir.BzrDirMetaFormat1()
3242
1310
def _create_control_files(self, a_bzrdir):
3243
1311
"""Create the required files and the initial control_files object."""
3244
1312
# FIXME: RBC 20060125 don't peek under the covers
3245
1313
# NB: no need to escape relative paths that are url safe.
3246
1314
repository_transport = a_bzrdir.get_repository_transport(self)
3247
control_files = lockable_files.LockableFiles(repository_transport,
3248
'lock', lockdir.LockDir)
1315
control_files = LockableFiles(repository_transport, 'lock', LockDir)
3249
1316
control_files.create_lock()
3250
1317
return control_files
3253
1320
"""Upload the initial blank content."""
3254
1321
control_files = self._create_control_files(a_bzrdir)
3255
1322
control_files.lock_write()
3256
transport = control_files._transport
3258
utf8_files += [('shared-storage', '')]
3260
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3261
for (filename, content_stream) in files:
3262
transport.put_file(filename, content_stream,
3263
mode=a_bzrdir._get_file_mode())
3264
for (filename, content_bytes) in utf8_files:
3265
transport.put_bytes_non_atomic(filename, content_bytes,
3266
mode=a_bzrdir._get_file_mode())
1324
control_files._transport.mkdir_multi(dirs,
1325
mode=control_files._dir_mode)
1326
for file, content in files:
1327
control_files.put(file, content)
1328
for file, content in utf8_files:
1329
control_files.put_utf8(file, content)
1331
control_files.put_utf8('shared-storage', '')
3268
1333
control_files.unlock()
3270
def network_name(self):
3271
"""Metadir formats have matching disk and network format strings."""
3272
return self.get_format_string()
3275
# Pre-0.8 formats that don't have a disk format string (because they are
3276
# versioned by the matching control directory). We use the control directories
3277
# disk format string as a key for the network_name because they meet the
3278
# constraints (simple string, unique, immutable).
3279
network_format_registry.register_lazy(
3280
"Bazaar-NG branch, format 5\n",
3281
'bzrlib.repofmt.weaverepo',
3282
'RepositoryFormat5',
3284
network_format_registry.register_lazy(
3285
"Bazaar-NG branch, format 6\n",
3286
'bzrlib.repofmt.weaverepo',
3287
'RepositoryFormat6',
3290
# formats which have no format string are not discoverable or independently
3291
# creatable on disk, so are not registered in format_registry. They're
3292
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3293
# needed, it's constructed directly by the BzrDir. Non-native formats where
3294
# the repository is not separately opened are similar.
3296
format_registry.register_lazy(
3297
'Bazaar-NG Repository format 7',
3298
'bzrlib.repofmt.weaverepo',
3302
format_registry.register_lazy(
3303
'Bazaar-NG Knit Repository Format 1',
3304
'bzrlib.repofmt.knitrepo',
3305
'RepositoryFormatKnit1',
3308
format_registry.register_lazy(
3309
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3310
'bzrlib.repofmt.knitrepo',
3311
'RepositoryFormatKnit3',
3314
format_registry.register_lazy(
3315
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3316
'bzrlib.repofmt.knitrepo',
3317
'RepositoryFormatKnit4',
3320
# Pack-based formats. There is one format for pre-subtrees, and one for
3321
# post-subtrees to allow ease of testing.
3322
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3323
format_registry.register_lazy(
3324
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3325
'bzrlib.repofmt.pack_repo',
3326
'RepositoryFormatKnitPack1',
3328
format_registry.register_lazy(
3329
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3330
'bzrlib.repofmt.pack_repo',
3331
'RepositoryFormatKnitPack3',
3333
format_registry.register_lazy(
3334
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3335
'bzrlib.repofmt.pack_repo',
3336
'RepositoryFormatKnitPack4',
3338
format_registry.register_lazy(
3339
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3340
'bzrlib.repofmt.pack_repo',
3341
'RepositoryFormatKnitPack5',
3343
format_registry.register_lazy(
3344
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3345
'bzrlib.repofmt.pack_repo',
3346
'RepositoryFormatKnitPack5RichRoot',
3348
format_registry.register_lazy(
3349
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3350
'bzrlib.repofmt.pack_repo',
3351
'RepositoryFormatKnitPack5RichRootBroken',
3353
format_registry.register_lazy(
3354
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3355
'bzrlib.repofmt.pack_repo',
3356
'RepositoryFormatKnitPack6',
3358
format_registry.register_lazy(
3359
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3360
'bzrlib.repofmt.pack_repo',
3361
'RepositoryFormatKnitPack6RichRoot',
3364
# Development formats.
3365
# Obsolete but kept pending a CHK based subtree format.
3366
format_registry.register_lazy(
3367
("Bazaar development format 2 with subtree support "
3368
"(needs bzr.dev from before 1.8)\n"),
3369
'bzrlib.repofmt.pack_repo',
3370
'RepositoryFormatPackDevelopment2Subtree',
3373
# 1.14->1.16 go below here
3374
format_registry.register_lazy(
3375
'Bazaar development format - group compression and chk inventory'
3376
' (needs bzr.dev from 1.14)\n',
3377
'bzrlib.repofmt.groupcompress_repo',
3378
'RepositoryFormatCHK1',
3381
format_registry.register_lazy(
3382
'Bazaar development format - chk repository with bencode revision '
3383
'serialization (needs bzr.dev from 1.16)\n',
3384
'bzrlib.repofmt.groupcompress_repo',
3385
'RepositoryFormatCHK2',
3387
format_registry.register_lazy(
3388
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3389
'bzrlib.repofmt.groupcompress_repo',
3390
'RepositoryFormat2a',
1336
class RepositoryFormat7(MetaDirRepositoryFormat):
1337
"""Bzr repository 7.
1339
This repository format has:
1340
- weaves for file texts and inventory
1341
- hash subdirectory based stores.
1342
- TextStores for revisions and signatures.
1343
- a format marker of its own
1344
- an optional 'shared-storage' flag
1345
- an optional 'no-working-trees' flag
1348
def _get_control_store(self, repo_transport, control_files):
1349
"""Return the control store for this repository."""
1350
return self._get_versioned_file_store('',
1355
def get_format_string(self):
1356
"""See RepositoryFormat.get_format_string()."""
1357
return "Bazaar-NG Repository format 7"
1359
def get_format_description(self):
1360
"""See RepositoryFormat.get_format_description()."""
1361
return "Weave repository format 7"
1363
def _get_revision_store(self, repo_transport, control_files):
1364
"""See RepositoryFormat._get_revision_store()."""
1365
return self._get_text_rev_store(repo_transport,
1372
def _get_text_store(self, transport, control_files):
1373
"""See RepositoryFormat._get_text_store()."""
1374
return self._get_versioned_file_store('weaves',
1378
def initialize(self, a_bzrdir, shared=False):
1379
"""Create a weave repository.
1381
:param shared: If true the repository will be initialized as a shared
1384
from bzrlib.weavefile import write_weave_v5
1385
from bzrlib.weave import Weave
1387
# Create an empty weave
1389
write_weave_v5(Weave(), sio)
1390
empty_weave = sio.getvalue()
1392
mutter('creating repository in %s.', a_bzrdir.transport.base)
1393
dirs = ['revision-store', 'weaves']
1394
files = [('inventory.weave', StringIO(empty_weave)),
1396
utf8_files = [('format', self.get_format_string())]
1398
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1399
return self.open(a_bzrdir=a_bzrdir, _found=True)
1401
def open(self, a_bzrdir, _found=False, _override_transport=None):
1402
"""See RepositoryFormat.open().
1404
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1405
repository at a slightly different url
1406
than normal. I.e. during 'upgrade'.
1409
format = RepositoryFormat.find_format(a_bzrdir)
1410
assert format.__class__ == self.__class__
1411
if _override_transport is not None:
1412
repo_transport = _override_transport
1414
repo_transport = a_bzrdir.get_repository_transport(None)
1415
control_files = LockableFiles(repo_transport, 'lock', LockDir)
1416
text_store = self._get_text_store(repo_transport, control_files)
1417
control_store = self._get_control_store(repo_transport, control_files)
1418
_revision_store = self._get_revision_store(repo_transport, control_files)
1419
return MetaDirRepository(_format=self,
1421
control_files=control_files,
1422
_revision_store=_revision_store,
1423
control_store=control_store,
1424
text_store=text_store)
1427
class RepositoryFormatKnit1(MetaDirRepositoryFormat):
1428
"""Bzr repository knit format 1.
1430
This repository format has:
1431
- knits for file texts and inventory
1432
- hash subdirectory based stores.
1433
- knits for revisions and signatures
1434
- TextStores for revisions and signatures.
1435
- a format marker of its own
1436
- an optional 'shared-storage' flag
1437
- an optional 'no-working-trees' flag
1440
This format was introduced in bzr 0.8.
1443
def _get_control_store(self, repo_transport, control_files):
1444
"""Return the control store for this repository."""
1445
return VersionedFileStore(
1448
file_mode=control_files._file_mode,
1449
versionedfile_class=KnitVersionedFile,
1450
versionedfile_kwargs={'factory':KnitPlainFactory()},
1453
def get_format_string(self):
1454
"""See RepositoryFormat.get_format_string()."""
1455
return "Bazaar-NG Knit Repository Format 1"
1457
def get_format_description(self):
1458
"""See RepositoryFormat.get_format_description()."""
1459
return "Knit repository format 1"
1461
def _get_revision_store(self, repo_transport, control_files):
1462
"""See RepositoryFormat._get_revision_store()."""
1463
from bzrlib.store.revision.knit import KnitRevisionStore
1464
versioned_file_store = VersionedFileStore(
1466
file_mode=control_files._file_mode,
1469
versionedfile_class=KnitVersionedFile,
1470
versionedfile_kwargs={'delta':False, 'factory':KnitPlainFactory()},
1473
return KnitRevisionStore(versioned_file_store)
1475
def _get_text_store(self, transport, control_files):
1476
"""See RepositoryFormat._get_text_store()."""
1477
return self._get_versioned_file_store('knits',
1480
versionedfile_class=KnitVersionedFile,
1483
def initialize(self, a_bzrdir, shared=False):
1484
"""Create a knit format 1 repository.
1486
:param a_bzrdir: bzrdir to contain the new repository; must already
1488
:param shared: If true the repository will be initialized as a shared
1491
mutter('creating repository in %s.', a_bzrdir.transport.base)
1492
dirs = ['revision-store', 'knits']
1494
utf8_files = [('format', self.get_format_string())]
1496
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1497
repo_transport = a_bzrdir.get_repository_transport(None)
1498
control_files = LockableFiles(repo_transport, 'lock', LockDir)
1499
control_store = self._get_control_store(repo_transport, control_files)
1500
transaction = transactions.WriteTransaction()
1501
# trigger a write of the inventory store.
1502
control_store.get_weave_or_empty('inventory', transaction)
1503
_revision_store = self._get_revision_store(repo_transport, control_files)
1504
_revision_store.has_revision_id('A', transaction)
1505
_revision_store.get_signature_file(transaction)
1506
return self.open(a_bzrdir=a_bzrdir, _found=True)
1508
def open(self, a_bzrdir, _found=False, _override_transport=None):
1509
"""See RepositoryFormat.open().
1511
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1512
repository at a slightly different url
1513
than normal. I.e. during 'upgrade'.
1516
format = RepositoryFormat.find_format(a_bzrdir)
1517
assert format.__class__ == self.__class__
1518
if _override_transport is not None:
1519
repo_transport = _override_transport
1521
repo_transport = a_bzrdir.get_repository_transport(None)
1522
control_files = LockableFiles(repo_transport, 'lock', LockDir)
1523
text_store = self._get_text_store(repo_transport, control_files)
1524
control_store = self._get_control_store(repo_transport, control_files)
1525
_revision_store = self._get_revision_store(repo_transport, control_files)
1526
return KnitRepository(_format=self,
1528
control_files=control_files,
1529
_revision_store=_revision_store,
1530
control_store=control_store,
1531
text_store=text_store)
1534
# formats which have no format string are not discoverable
1535
# and not independently creatable, so are not registered.
1536
RepositoryFormat.register_format(RepositoryFormat7())
1537
_default_format = RepositoryFormatKnit1()
1538
RepositoryFormat.register_format(_default_format)
1539
RepositoryFormat.set_default_format(_default_format)
1540
_legacy_formats = [RepositoryFormat4(),
1541
RepositoryFormat5(),
1542
RepositoryFormat6()]
3394
1545
class InterRepository(InterObject):
3395
1546
"""This class represents operations taking place between two repositories.
3397
1548
Its instances have methods like copy_content and fetch, and contain
3398
references to the source and target repositories these operations can be
1549
references to the source and target repositories these operations can be
3399
1550
carried out on.
3401
1552
Often we will provide convenience methods on 'repository' which carry out
3666
1731
# we do not have a revision as that would be pointless.
3667
1732
target_ids = set(self.target._all_possible_ids())
3668
1733
possibly_present_revisions = target_ids.intersection(source_ids_set)
3669
actually_present_revisions = set(
3670
self.target._eliminate_revisions_not_present(possibly_present_revisions))
1734
actually_present_revisions = set(self.target._eliminate_revisions_not_present(possibly_present_revisions))
3671
1735
required_revisions = source_ids_set.difference(actually_present_revisions)
1736
required_topo_revisions = [rev_id for rev_id in source_ids if rev_id in required_revisions]
3672
1737
if revision_id is not None:
3673
1738
# we used get_ancestry to determine source_ids then we are assured all
3674
1739
# revisions referenced are present as they are installed in topological order.
3675
1740
# and the tip revision was validated by get_ancestry.
3676
result_set = required_revisions
1741
return required_topo_revisions
3678
# if we just grabbed the possibly available ids, then
1743
# if we just grabbed the possibly available ids, then
3679
1744
# we only have an estimate of whats available and need to validate
3680
1745
# that against the revision records.
3682
self.source._eliminate_revisions_not_present(required_revisions))
3683
return self.source.revision_ids_to_search_result(result_set)
3686
class InterKnitRepo(InterSameDataRepository):
1746
return self.source._eliminate_revisions_not_present(required_topo_revisions)
1749
class InterKnitRepo(InterRepository):
3687
1750
"""Optimised code paths between Knit based repositories."""
3690
def _get_repo_format_to_test(self):
3691
from bzrlib.repofmt import knitrepo
3692
return knitrepo.RepositoryFormatKnit1()
1752
_matching_repo_format = RepositoryFormatKnit1()
1753
"""Repository format for testing with."""
3695
1756
def is_compatible(source, target):
3696
1757
"""Be compatible with known Knit formats.
3698
1759
We don't test for the stores being of specific types because that
3699
could lead to confusing results, and there is no need to be
1760
could lead to confusing results, and there is no need to be
3700
1761
overly general.
3702
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3704
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3705
isinstance(target._format, RepositoryFormatKnit))
1764
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1765
isinstance(target._format, (RepositoryFormatKnit1)))
3706
1766
except AttributeError:
3708
return are_knits and InterRepository._same_model(source, target)
1770
def fetch(self, revision_id=None, pb=None):
1771
"""See InterRepository.fetch()."""
1772
from bzrlib.fetch import KnitRepoFetcher
1773
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1774
self.source, self.source._format, self.target, self.target._format)
1775
f = KnitRepoFetcher(to_repository=self.target,
1776
from_repository=self.source,
1777
last_revision=revision_id,
1779
return f.count_copied, f.failed_revisions
3710
1781
@needs_read_lock
3711
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1782
def missing_revision_ids(self, revision_id=None):
3712
1783
"""See InterRepository.missing_revision_ids()."""
3713
1784
if revision_id is not None:
3714
1785
source_ids = self.source.get_ancestry(revision_id)
3715
if source_ids[0] is not None:
3716
raise AssertionError()
1786
assert source_ids[0] == None
3717
1787
source_ids.pop(0)
3719
source_ids = self.source.all_revision_ids()
1789
source_ids = self.source._all_possible_ids()
3720
1790
source_ids_set = set(source_ids)
3721
1791
# source_ids is the worst possible case we may need to pull.
3722
1792
# now we want to filter source_ids against what we actually
3723
1793
# have in target, but don't try to check for existence where we know
3724
1794
# we do not have a revision as that would be pointless.
3725
target_ids = set(self.target.all_revision_ids())
1795
target_ids = set(self.target._all_possible_ids())
3726
1796
possibly_present_revisions = target_ids.intersection(source_ids_set)
3727
actually_present_revisions = set(
3728
self.target._eliminate_revisions_not_present(possibly_present_revisions))
1797
actually_present_revisions = set(self.target._eliminate_revisions_not_present(possibly_present_revisions))
3729
1798
required_revisions = source_ids_set.difference(actually_present_revisions)
1799
required_topo_revisions = [rev_id for rev_id in source_ids if rev_id in required_revisions]
3730
1800
if revision_id is not None:
3731
1801
# we used get_ancestry to determine source_ids then we are assured all
3732
1802
# revisions referenced are present as they are installed in topological order.
3733
1803
# and the tip revision was validated by get_ancestry.
3734
result_set = required_revisions
1804
return required_topo_revisions
3736
# if we just grabbed the possibly available ids, then
1806
# if we just grabbed the possibly available ids, then
3737
1807
# we only have an estimate of whats available and need to validate
3738
1808
# that against the revision records.
3740
self.source._eliminate_revisions_not_present(required_revisions))
3741
return self.source.revision_ids_to_search_result(result_set)
3744
class InterDifferingSerializer(InterRepository):
3747
def _get_repo_format_to_test(self):
3751
def is_compatible(source, target):
3752
"""Be compatible with Knit2 source and Knit3 target"""
3753
# This is redundant with format.check_conversion_target(), however that
3754
# raises an exception, and we just want to say "False" as in we won't
3755
# support converting between these formats.
3756
if 'IDS_never' in debug.debug_flags:
3758
if source.supports_rich_root() and not target.supports_rich_root():
3760
if (source._format.supports_tree_reference
3761
and not target._format.supports_tree_reference):
3763
if target._fallback_repositories and target._format.supports_chks:
3764
# IDS doesn't know how to copy CHKs for the parent inventories it
3765
# adds to stacked repos.
3767
if 'IDS_always' in debug.debug_flags:
3769
# Only use this code path for local source and target. IDS does far
3770
# too much IO (both bandwidth and roundtrips) over a network.
3771
if not source.bzrdir.transport.base.startswith('file:///'):
3773
if not target.bzrdir.transport.base.startswith('file:///'):
3777
def _get_trees(self, revision_ids, cache):
3779
for rev_id in revision_ids:
3781
possible_trees.append((rev_id, cache[rev_id]))
3783
# Not cached, but inventory might be present anyway.
3785
tree = self.source.revision_tree(rev_id)
3786
except errors.NoSuchRevision:
3787
# Nope, parent is ghost.
3790
cache[rev_id] = tree
3791
possible_trees.append((rev_id, tree))
3792
return possible_trees
3794
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3795
"""Get the best delta and base for this revision.
3797
:return: (basis_id, delta)
3800
# Generate deltas against each tree, to find the shortest.
3801
texts_possibly_new_in_tree = set()
3802
for basis_id, basis_tree in possible_trees:
3803
delta = tree.inventory._make_delta(basis_tree.inventory)
3804
for old_path, new_path, file_id, new_entry in delta:
3805
if new_path is None:
3806
# This file_id isn't present in the new rev, so we don't
3810
# Rich roots are handled elsewhere...
3812
kind = new_entry.kind
3813
if kind != 'directory' and kind != 'file':
3814
# No text record associated with this inventory entry.
3816
# This is a directory or file that has changed somehow.
3817
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3818
deltas.append((len(delta), basis_id, delta))
3820
return deltas[0][1:]
3822
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3823
"""Find all parent revisions that are absent, but for which the
3824
inventory is present, and copy those inventories.
3826
This is necessary to preserve correctness when the source is stacked
3827
without fallbacks configured. (Note that in cases like upgrade the
3828
source may be not have _fallback_repositories even though it is
3832
for parents in parent_map.values():
3833
parent_revs.update(parents)
3834
present_parents = self.source.get_parent_map(parent_revs)
3835
absent_parents = set(parent_revs).difference(present_parents)
3836
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3837
(rev_id,) for rev_id in absent_parents)
3838
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3839
for parent_tree in self.source.revision_trees(parent_inv_ids):
3840
current_revision_id = parent_tree.get_revision_id()
3841
parents_parents_keys = parent_invs_keys_for_stacking[
3842
(current_revision_id,)]
3843
parents_parents = [key[-1] for key in parents_parents_keys]
3844
basis_id = _mod_revision.NULL_REVISION
3845
basis_tree = self.source.revision_tree(basis_id)
3846
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3847
self.target.add_inventory_by_delta(
3848
basis_id, delta, current_revision_id, parents_parents)
3849
cache[current_revision_id] = parent_tree
3851
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3852
"""Fetch across a few revisions.
3854
:param revision_ids: The revisions to copy
3855
:param basis_id: The revision_id of a tree that must be in cache, used
3856
as a basis for delta when no other base is available
3857
:param cache: A cache of RevisionTrees that we can use.
3858
:param a_graph: A Graph object to determine the heads() of the
3859
rich-root data stream.
3860
:return: The revision_id of the last converted tree. The RevisionTree
3861
for it will be in cache
3863
# Walk though all revisions; get inventory deltas, copy referenced
3864
# texts that delta references, insert the delta, revision and
3866
root_keys_to_create = set()
3869
pending_revisions = []
3870
parent_map = self.source.get_parent_map(revision_ids)
3871
self._fetch_parent_invs_for_stacking(parent_map, cache)
3872
self.source._safe_to_return_from_cache = True
3873
for tree in self.source.revision_trees(revision_ids):
3874
# Find a inventory delta for this revision.
3875
# Find text entries that need to be copied, too.
3876
current_revision_id = tree.get_revision_id()
3877
parent_ids = parent_map.get(current_revision_id, ())
3878
parent_trees = self._get_trees(parent_ids, cache)
3879
possible_trees = list(parent_trees)
3880
if len(possible_trees) == 0:
3881
# There either aren't any parents, or the parents are ghosts,
3882
# so just use the last converted tree.
3883
possible_trees.append((basis_id, cache[basis_id]))
3884
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3886
revision = self.source.get_revision(current_revision_id)
3887
pending_deltas.append((basis_id, delta,
3888
current_revision_id, revision.parent_ids))
3889
if self._converting_to_rich_root:
3890
self._revision_id_to_root_id[current_revision_id] = \
3892
# Determine which texts are in present in this revision but not in
3893
# any of the available parents.
3894
texts_possibly_new_in_tree = set()
3895
for old_path, new_path, file_id, entry in delta:
3896
if new_path is None:
3897
# This file_id isn't present in the new rev
3901
if not self.target.supports_rich_root():
3902
# The target doesn't support rich root, so we don't
3905
if self._converting_to_rich_root:
3906
# This can't be copied normally, we have to insert
3908
root_keys_to_create.add((file_id, entry.revision))
3911
texts_possibly_new_in_tree.add((file_id, entry.revision))
3912
for basis_id, basis_tree in possible_trees:
3913
basis_inv = basis_tree.inventory
3914
for file_key in list(texts_possibly_new_in_tree):
3915
file_id, file_revision = file_key
3917
entry = basis_inv[file_id]
3918
except errors.NoSuchId:
3920
if entry.revision == file_revision:
3921
texts_possibly_new_in_tree.remove(file_key)
3922
text_keys.update(texts_possibly_new_in_tree)
3923
pending_revisions.append(revision)
3924
cache[current_revision_id] = tree
3925
basis_id = current_revision_id
3926
self.source._safe_to_return_from_cache = False
3928
from_texts = self.source.texts
3929
to_texts = self.target.texts
3930
if root_keys_to_create:
3931
root_stream = _mod_fetch._new_root_data_stream(
3932
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3933
self.source, graph=a_graph)
3934
to_texts.insert_record_stream(root_stream)
3935
to_texts.insert_record_stream(from_texts.get_record_stream(
3936
text_keys, self.target._format._fetch_order,
3937
not self.target._format._fetch_uses_deltas))
3938
# insert inventory deltas
3939
for delta in pending_deltas:
3940
self.target.add_inventory_by_delta(*delta)
3941
if self.target._fallback_repositories:
3942
# Make sure this stacked repository has all the parent inventories
3943
# for the new revisions that we are about to insert. We do this
3944
# before adding the revisions so that no revision is added until
3945
# all the inventories it may depend on are added.
3946
# Note that this is overzealous, as we may have fetched these in an
3949
revision_ids = set()
3950
for revision in pending_revisions:
3951
revision_ids.add(revision.revision_id)
3952
parent_ids.update(revision.parent_ids)
3953
parent_ids.difference_update(revision_ids)
3954
parent_ids.discard(_mod_revision.NULL_REVISION)
3955
parent_map = self.source.get_parent_map(parent_ids)
3956
# we iterate over parent_map and not parent_ids because we don't
3957
# want to try copying any revision which is a ghost
3958
for parent_tree in self.source.revision_trees(parent_map):
3959
current_revision_id = parent_tree.get_revision_id()
3960
parents_parents = parent_map[current_revision_id]
3961
possible_trees = self._get_trees(parents_parents, cache)
3962
if len(possible_trees) == 0:
3963
# There either aren't any parents, or the parents are
3964
# ghosts, so just use the last converted tree.
3965
possible_trees.append((basis_id, cache[basis_id]))
3966
basis_id, delta = self._get_delta_for_revision(parent_tree,
3967
parents_parents, possible_trees)
3968
self.target.add_inventory_by_delta(
3969
basis_id, delta, current_revision_id, parents_parents)
3970
# insert signatures and revisions
3971
for revision in pending_revisions:
3973
signature = self.source.get_signature_text(
3974
revision.revision_id)
3975
self.target.add_signature_text(revision.revision_id,
3977
except errors.NoSuchRevision:
3979
self.target.add_revision(revision.revision_id, revision)
3982
def _fetch_all_revisions(self, revision_ids, pb):
3983
"""Fetch everything for the list of revisions.
3985
:param revision_ids: The list of revisions to fetch. Must be in
3987
:param pb: A ProgressTask
3990
basis_id, basis_tree = self._get_basis(revision_ids[0])
3992
cache = lru_cache.LRUCache(100)
3993
cache[basis_id] = basis_tree
3994
del basis_tree # We don't want to hang on to it here
3996
if self._converting_to_rich_root and len(revision_ids) > 100:
3997
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
4002
for offset in range(0, len(revision_ids), batch_size):
4003
self.target.start_write_group()
4005
pb.update('Transferring revisions', offset,
4007
batch = revision_ids[offset:offset+batch_size]
4008
basis_id = self._fetch_batch(batch, basis_id, cache,
4011
self.source._safe_to_return_from_cache = False
4012
self.target.abort_write_group()
4015
hint = self.target.commit_write_group()
4018
if hints and self.target._format.pack_compresses:
4019
self.target.pack(hint=hints)
4020
pb.update('Transferring revisions', len(revision_ids),
4024
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
4026
"""See InterRepository.fetch()."""
4027
if fetch_spec is not None:
4028
raise AssertionError("Not implemented yet...")
4029
ui.ui_factory.warn_experimental_format_fetch(self)
4030
if (not self.source.supports_rich_root()
4031
and self.target.supports_rich_root()):
4032
self._converting_to_rich_root = True
4033
self._revision_id_to_root_id = {}
4035
self._converting_to_rich_root = False
4036
# See <https://launchpad.net/bugs/456077> asking for a warning here
4037
if self.source._format.network_name() != self.target._format.network_name():
4038
ui.ui_factory.show_user_warning('cross_format_fetch',
4039
from_format=self.source._format,
4040
to_format=self.target._format)
4041
revision_ids = self.target.search_missing_revision_ids(self.source,
4042
revision_id, find_ghosts=find_ghosts).get_keys()
4043
if not revision_ids:
4045
revision_ids = tsort.topo_sort(
4046
self.source.get_graph().get_parent_map(revision_ids))
4047
if not revision_ids:
4049
# Walk though all revisions; get inventory deltas, copy referenced
4050
# texts that delta references, insert the delta, revision and
4053
my_pb = ui.ui_factory.nested_progress_bar()
4056
symbol_versioning.warn(
4057
symbol_versioning.deprecated_in((1, 14, 0))
4058
% "pb parameter to fetch()")
4061
self._fetch_all_revisions(revision_ids, pb)
4063
if my_pb is not None:
4065
return len(revision_ids), 0
4067
def _get_basis(self, first_revision_id):
4068
"""Get a revision and tree which exists in the target.
4070
This assumes that first_revision_id is selected for transmission
4071
because all other ancestors are already present. If we can't find an
4072
ancestor we fall back to NULL_REVISION since we know that is safe.
4074
:return: (basis_id, basis_tree)
4076
first_rev = self.source.get_revision(first_revision_id)
4078
basis_id = first_rev.parent_ids[0]
4079
# only valid as a basis if the target has it
4080
self.target.get_revision(basis_id)
4081
# Try to get a basis tree - if its a ghost it will hit the
4082
# NoSuchRevision case.
4083
basis_tree = self.source.revision_tree(basis_id)
4084
except (IndexError, errors.NoSuchRevision):
4085
basis_id = _mod_revision.NULL_REVISION
4086
basis_tree = self.source.revision_tree(basis_id)
4087
return basis_id, basis_tree
4090
InterRepository.register_optimiser(InterDifferingSerializer)
4091
InterRepository.register_optimiser(InterSameDataRepository)
1809
return self.source._eliminate_revisions_not_present(required_topo_revisions)
4092
1811
InterRepository.register_optimiser(InterWeaveRepo)
4093
1812
InterRepository.register_optimiser(InterKnitRepo)
1815
class RepositoryTestProviderAdapter(object):
1816
"""A tool to generate a suite testing multiple repository formats at once.
1818
This is done by copying the test once for each transport and injecting
1819
the transport_server, transport_readonly_server, and bzrdir_format and
1820
repository_format classes into each copy. Each copy is also given a new id()
1821
to make it easy to identify.
1824
def __init__(self, transport_server, transport_readonly_server, formats):
1825
self._transport_server = transport_server
1826
self._transport_readonly_server = transport_readonly_server
1827
self._formats = formats
1829
def adapt(self, test):
1830
result = TestSuite()
1831
for repository_format, bzrdir_format in self._formats:
1832
new_test = deepcopy(test)
1833
new_test.transport_server = self._transport_server
1834
new_test.transport_readonly_server = self._transport_readonly_server
1835
new_test.bzrdir_format = bzrdir_format
1836
new_test.repository_format = repository_format
1837
def make_new_test_id():
1838
new_id = "%s(%s)" % (new_test.id(), repository_format.__class__.__name__)
1839
return lambda: new_id
1840
new_test.id = make_new_test_id()
1841
result.addTest(new_test)
1845
class InterRepositoryTestProviderAdapter(object):
1846
"""A tool to generate a suite testing multiple inter repository formats.
1848
This is done by copying the test once for each interrepo provider and injecting
1849
the transport_server, transport_readonly_server, repository_format and
1850
repository_to_format classes into each copy.
1851
Each copy is also given a new id() to make it easy to identify.
1854
def __init__(self, transport_server, transport_readonly_server, formats):
1855
self._transport_server = transport_server
1856
self._transport_readonly_server = transport_readonly_server
1857
self._formats = formats
1859
def adapt(self, test):
1860
result = TestSuite()
1861
for interrepo_class, repository_format, repository_format_to in self._formats:
1862
new_test = deepcopy(test)
1863
new_test.transport_server = self._transport_server
1864
new_test.transport_readonly_server = self._transport_readonly_server
1865
new_test.interrepo_class = interrepo_class
1866
new_test.repository_format = repository_format
1867
new_test.repository_format_to = repository_format_to
1868
def make_new_test_id():
1869
new_id = "%s(%s)" % (new_test.id(), interrepo_class.__name__)
1870
return lambda: new_id
1871
new_test.id = make_new_test_id()
1872
result.addTest(new_test)
1876
def default_test_list():
1877
"""Generate the default list of interrepo permutations to test."""
1879
# test the default InterRepository between format 6 and the current
1881
# XXX: robertc 20060220 reinstate this when there are two supported
1882
# formats which do not have an optimal code path between them.
1883
result.append((InterRepository,
1884
RepositoryFormat6(),
1885
RepositoryFormatKnit1()))
1886
for optimiser in InterRepository._optimisers:
1887
result.append((optimiser,
1888
optimiser._matching_repo_format,
1889
optimiser._matching_repo_format
1891
# if there are specific combinations we want to use, we can add them
4096
1896
class CopyConverter(object):
4097
1897
"""A repository conversion tool which just performs a copy of the content.
4099
1899
This is slow but quite reliable.
4167
2140
"""Unescape predefined XML entities in a string of data."""
4168
2141
global _unescape_re
4169
2142
if _unescape_re is None:
4170
_unescape_re = re.compile('\&([^;]*);')
2143
_unescape_re = re.compile('\&([^;]*);')
4171
2144
return _unescape_re.sub(_unescaper, data)
4174
class _VersionedFileChecker(object):
4176
def __init__(self, repository, text_key_references=None, ancestors=None):
4177
self.repository = repository
4178
self.text_index = self.repository._generate_text_key_index(
4179
text_key_references=text_key_references, ancestors=ancestors)
4181
def calculate_file_version_parents(self, text_key):
4182
"""Calculate the correct parents for a file version according to
4185
parent_keys = self.text_index[text_key]
4186
if parent_keys == [_mod_revision.NULL_REVISION]:
4188
return tuple(parent_keys)
4190
def check_file_version_parents(self, texts, progress_bar=None):
4191
"""Check the parents stored in a versioned file are correct.
4193
It also detects file versions that are not referenced by their
4194
corresponding revision's inventory.
4196
:returns: A tuple of (wrong_parents, dangling_file_versions).
4197
wrong_parents is a dict mapping {revision_id: (stored_parents,
4198
correct_parents)} for each revision_id where the stored parents
4199
are not correct. dangling_file_versions is a set of (file_id,
4200
revision_id) tuples for versions that are present in this versioned
4201
file, but not used by the corresponding inventory.
4203
local_progress = None
4204
if progress_bar is None:
4205
local_progress = ui.ui_factory.nested_progress_bar()
4206
progress_bar = local_progress
4208
return self._check_file_version_parents(texts, progress_bar)
4211
local_progress.finished()
4213
def _check_file_version_parents(self, texts, progress_bar):
4214
"""See check_file_version_parents."""
4216
self.file_ids = set([file_id for file_id, _ in
4217
self.text_index.iterkeys()])
4218
# text keys is now grouped by file_id
4219
n_versions = len(self.text_index)
4220
progress_bar.update('loading text store', 0, n_versions)
4221
parent_map = self.repository.texts.get_parent_map(self.text_index)
4222
# On unlistable transports this could well be empty/error...
4223
text_keys = self.repository.texts.keys()
4224
unused_keys = frozenset(text_keys) - set(self.text_index)
4225
for num, key in enumerate(self.text_index.iterkeys()):
4226
progress_bar.update('checking text graph', num, n_versions)
4227
correct_parents = self.calculate_file_version_parents(key)
4229
knit_parents = parent_map[key]
4230
except errors.RevisionNotPresent:
4233
if correct_parents != knit_parents:
4234
wrong_parents[key] = (knit_parents, correct_parents)
4235
return wrong_parents, unused_keys
4238
def _old_get_graph(repository, revision_id):
4239
"""DO NOT USE. That is all. I'm serious."""
4240
graph = repository.get_graph()
4241
revision_graph = dict(((key, value) for key, value in
4242
graph.iter_ancestry([revision_id]) if value is not None))
4243
return _strip_NULL_ghosts(revision_graph)
4246
def _strip_NULL_ghosts(revision_graph):
4247
"""Also don't use this. more compatibility code for unmigrated clients."""
4248
# Filter ghosts, and null:
4249
if _mod_revision.NULL_REVISION in revision_graph:
4250
del revision_graph[_mod_revision.NULL_REVISION]
4251
for key, parents in revision_graph.items():
4252
revision_graph[key] = tuple(parent for parent in parents if parent
4254
return revision_graph
4257
class StreamSink(object):
4258
"""An object that can insert a stream into a repository.
4260
This interface handles the complexity of reserialising inventories and
4261
revisions from different formats, and allows unidirectional insertion into
4262
stacked repositories without looking for the missing basis parents
4266
def __init__(self, target_repo):
4267
self.target_repo = target_repo
4269
def insert_stream(self, stream, src_format, resume_tokens):
4270
"""Insert a stream's content into the target repository.
4272
:param src_format: a bzr repository format.
4274
:return: a list of resume tokens and an iterable of keys additional
4275
items required before the insertion can be completed.
4277
self.target_repo.lock_write()
4280
self.target_repo.resume_write_group(resume_tokens)
4283
self.target_repo.start_write_group()
4286
# locked_insert_stream performs a commit|suspend.
4287
return self._locked_insert_stream(stream, src_format,
4290
self.target_repo.abort_write_group(suppress_errors=True)
4293
self.target_repo.unlock()
4295
def _locked_insert_stream(self, stream, src_format, is_resume):
4296
to_serializer = self.target_repo._format._serializer
4297
src_serializer = src_format._serializer
4299
if to_serializer == src_serializer:
4300
# If serializers match and the target is a pack repository, set the
4301
# write cache size on the new pack. This avoids poor performance
4302
# on transports where append is unbuffered (such as
4303
# RemoteTransport). This is safe to do because nothing should read
4304
# back from the target repository while a stream with matching
4305
# serialization is being inserted.
4306
# The exception is that a delta record from the source that should
4307
# be a fulltext may need to be expanded by the target (see
4308
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4309
# explicitly flush any buffered writes first in that rare case.
4311
new_pack = self.target_repo._pack_collection._new_pack
4312
except AttributeError:
4313
# Not a pack repository
4316
new_pack.set_write_cache_size(1024*1024)
4317
for substream_type, substream in stream:
4318
if 'stream' in debug.debug_flags:
4319
mutter('inserting substream: %s', substream_type)
4320
if substream_type == 'texts':
4321
self.target_repo.texts.insert_record_stream(substream)
4322
elif substream_type == 'inventories':
4323
if src_serializer == to_serializer:
4324
self.target_repo.inventories.insert_record_stream(
4327
self._extract_and_insert_inventories(
4328
substream, src_serializer)
4329
elif substream_type == 'inventory-deltas':
4330
self._extract_and_insert_inventory_deltas(
4331
substream, src_serializer)
4332
elif substream_type == 'chk_bytes':
4333
# XXX: This doesn't support conversions, as it assumes the
4334
# conversion was done in the fetch code.
4335
self.target_repo.chk_bytes.insert_record_stream(substream)
4336
elif substream_type == 'revisions':
4337
# This may fallback to extract-and-insert more often than
4338
# required if the serializers are different only in terms of
4340
if src_serializer == to_serializer:
4341
self.target_repo.revisions.insert_record_stream(substream)
4343
self._extract_and_insert_revisions(substream,
4345
elif substream_type == 'signatures':
4346
self.target_repo.signatures.insert_record_stream(substream)
4348
raise AssertionError('kaboom! %s' % (substream_type,))
4349
# Done inserting data, and the missing_keys calculations will try to
4350
# read back from the inserted data, so flush the writes to the new pack
4351
# (if this is pack format).
4352
if new_pack is not None:
4353
new_pack._write_data('', flush=True)
4354
# Find all the new revisions (including ones from resume_tokens)
4355
missing_keys = self.target_repo.get_missing_parent_inventories(
4356
check_for_missing_texts=is_resume)
4358
for prefix, versioned_file in (
4359
('texts', self.target_repo.texts),
4360
('inventories', self.target_repo.inventories),
4361
('revisions', self.target_repo.revisions),
4362
('signatures', self.target_repo.signatures),
4363
('chk_bytes', self.target_repo.chk_bytes),
4365
if versioned_file is None:
4367
# TODO: key is often going to be a StaticTuple object
4368
# I don't believe we can define a method by which
4369
# (prefix,) + StaticTuple will work, though we could
4370
# define a StaticTuple.sq_concat that would allow you to
4371
# pass in either a tuple or a StaticTuple as the second
4372
# object, so instead we could have:
4373
# StaticTuple(prefix) + key here...
4374
missing_keys.update((prefix,) + key for key in
4375
versioned_file.get_missing_compression_parent_keys())
4376
except NotImplementedError:
4377
# cannot even attempt suspending, and missing would have failed
4378
# during stream insertion.
4379
missing_keys = set()
4382
# suspend the write group and tell the caller what we is
4383
# missing. We know we can suspend or else we would not have
4384
# entered this code path. (All repositories that can handle
4385
# missing keys can handle suspending a write group).
4386
write_group_tokens = self.target_repo.suspend_write_group()
4387
return write_group_tokens, missing_keys
4388
hint = self.target_repo.commit_write_group()
4389
if (to_serializer != src_serializer and
4390
self.target_repo._format.pack_compresses):
4391
self.target_repo.pack(hint=hint)
4394
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4395
target_rich_root = self.target_repo._format.rich_root_data
4396
target_tree_refs = self.target_repo._format.supports_tree_reference
4397
for record in substream:
4398
# Insert the delta directly
4399
inventory_delta_bytes = record.get_bytes_as('fulltext')
4400
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4402
parse_result = deserialiser.parse_text_bytes(
4403
inventory_delta_bytes)
4404
except inventory_delta.IncompatibleInventoryDelta, err:
4405
trace.mutter("Incompatible delta: %s", err.msg)
4406
raise errors.IncompatibleRevision(self.target_repo._format)
4407
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4408
revision_id = new_id
4409
parents = [key[0] for key in record.parents]
4410
self.target_repo.add_inventory_by_delta(
4411
basis_id, inv_delta, revision_id, parents)
4413
def _extract_and_insert_inventories(self, substream, serializer,
4415
"""Generate a new inventory versionedfile in target, converting data.
4417
The inventory is retrieved from the source, (deserializing it), and
4418
stored in the target (reserializing it in a different format).
4420
target_rich_root = self.target_repo._format.rich_root_data
4421
target_tree_refs = self.target_repo._format.supports_tree_reference
4422
for record in substream:
4423
# It's not a delta, so it must be a fulltext in the source
4424
# serializer's format.
4425
bytes = record.get_bytes_as('fulltext')
4426
revision_id = record.key[0]
4427
inv = serializer.read_inventory_from_string(bytes, revision_id)
4428
parents = [key[0] for key in record.parents]
4429
self.target_repo.add_inventory(revision_id, inv, parents)
4430
# No need to keep holding this full inv in memory when the rest of
4431
# the substream is likely to be all deltas.
4434
def _extract_and_insert_revisions(self, substream, serializer):
4435
for record in substream:
4436
bytes = record.get_bytes_as('fulltext')
4437
revision_id = record.key[0]
4438
rev = serializer.read_revision_from_string(bytes)
4439
if rev.revision_id != revision_id:
4440
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4441
self.target_repo.add_revision(revision_id, rev)
4444
if self.target_repo._format._fetch_reconcile:
4445
self.target_repo.reconcile()
4448
class StreamSource(object):
4449
"""A source of a stream for fetching between repositories."""
4451
def __init__(self, from_repository, to_format):
4452
"""Create a StreamSource streaming from from_repository."""
4453
self.from_repository = from_repository
4454
self.to_format = to_format
4455
self._record_counter = RecordCounter()
4457
def delta_on_metadata(self):
4458
"""Return True if delta's are permitted on metadata streams.
4460
That is on revisions and signatures.
4462
src_serializer = self.from_repository._format._serializer
4463
target_serializer = self.to_format._serializer
4464
return (self.to_format._fetch_uses_deltas and
4465
src_serializer == target_serializer)
4467
def _fetch_revision_texts(self, revs):
4468
# fetch signatures first and then the revision texts
4469
# may need to be a InterRevisionStore call here.
4470
from_sf = self.from_repository.signatures
4471
# A missing signature is just skipped.
4472
keys = [(rev_id,) for rev_id in revs]
4473
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4475
self.to_format._fetch_order,
4476
not self.to_format._fetch_uses_deltas))
4477
# If a revision has a delta, this is actually expanded inside the
4478
# insert_record_stream code now, which is an alternate fix for
4480
from_rf = self.from_repository.revisions
4481
revisions = from_rf.get_record_stream(
4483
self.to_format._fetch_order,
4484
not self.delta_on_metadata())
4485
return [('signatures', signatures), ('revisions', revisions)]
4487
def _generate_root_texts(self, revs):
4488
"""This will be called by get_stream between fetching weave texts and
4489
fetching the inventory weave.
4491
if self._rich_root_upgrade():
4492
return _mod_fetch.Inter1and2Helper(
4493
self.from_repository).generate_root_texts(revs)
4497
def get_stream(self, search):
4499
revs = search.get_keys()
4500
graph = self.from_repository.get_graph()
4501
revs = tsort.topo_sort(graph.get_parent_map(revs))
4502
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4504
for knit_kind, file_id, revisions in data_to_fetch:
4505
if knit_kind != phase:
4507
# Make a new progress bar for this phase
4508
if knit_kind == "file":
4509
# Accumulate file texts
4510
text_keys.extend([(file_id, revision) for revision in
4512
elif knit_kind == "inventory":
4513
# Now copy the file texts.
4514
from_texts = self.from_repository.texts
4515
yield ('texts', from_texts.get_record_stream(
4516
text_keys, self.to_format._fetch_order,
4517
not self.to_format._fetch_uses_deltas))
4518
# Cause an error if a text occurs after we have done the
4521
# Before we process the inventory we generate the root
4522
# texts (if necessary) so that the inventories references
4524
for _ in self._generate_root_texts(revs):
4526
# we fetch only the referenced inventories because we do not
4527
# know for unselected inventories whether all their required
4528
# texts are present in the other repository - it could be
4530
for info in self._get_inventory_stream(revs):
4532
elif knit_kind == "signatures":
4533
# Nothing to do here; this will be taken care of when
4534
# _fetch_revision_texts happens.
4536
elif knit_kind == "revisions":
4537
for record in self._fetch_revision_texts(revs):
4540
raise AssertionError("Unknown knit kind %r" % knit_kind)
4542
def get_stream_for_missing_keys(self, missing_keys):
4543
# missing keys can only occur when we are byte copying and not
4544
# translating (because translation means we don't send
4545
# unreconstructable deltas ever).
4547
keys['texts'] = set()
4548
keys['revisions'] = set()
4549
keys['inventories'] = set()
4550
keys['chk_bytes'] = set()
4551
keys['signatures'] = set()
4552
for key in missing_keys:
4553
keys[key[0]].add(key[1:])
4554
if len(keys['revisions']):
4555
# If we allowed copying revisions at this point, we could end up
4556
# copying a revision without copying its required texts: a
4557
# violation of the requirements for repository integrity.
4558
raise AssertionError(
4559
'cannot copy revisions to fill in missing deltas %s' % (
4560
keys['revisions'],))
4561
for substream_kind, keys in keys.iteritems():
4562
vf = getattr(self.from_repository, substream_kind)
4563
if vf is None and keys:
4564
raise AssertionError(
4565
"cannot fill in keys for a versioned file we don't"
4566
" have: %s needs %s" % (substream_kind, keys))
4568
# No need to stream something we don't have
4570
if substream_kind == 'inventories':
4571
# Some missing keys are genuinely ghosts, filter those out.
4572
present = self.from_repository.inventories.get_parent_map(keys)
4573
revs = [key[0] for key in present]
4574
# Get the inventory stream more-or-less as we do for the
4575
# original stream; there's no reason to assume that records
4576
# direct from the source will be suitable for the sink. (Think
4577
# e.g. 2a -> 1.9-rich-root).
4578
for info in self._get_inventory_stream(revs, missing=True):
4582
# Ask for full texts always so that we don't need more round trips
4583
# after this stream.
4584
# Some of the missing keys are genuinely ghosts, so filter absent
4585
# records. The Sink is responsible for doing another check to
4586
# ensure that ghosts don't introduce missing data for future
4588
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4589
self.to_format._fetch_order, True))
4590
yield substream_kind, stream
4592
def inventory_fetch_order(self):
4593
if self._rich_root_upgrade():
4594
return 'topological'
4596
return self.to_format._fetch_order
4598
def _rich_root_upgrade(self):
4599
return (not self.from_repository._format.rich_root_data and
4600
self.to_format.rich_root_data)
4602
def _get_inventory_stream(self, revision_ids, missing=False):
4603
from_format = self.from_repository._format
4604
if (from_format.supports_chks and self.to_format.supports_chks and
4605
from_format.network_name() == self.to_format.network_name()):
4606
raise AssertionError(
4607
"this case should be handled by GroupCHKStreamSource")
4608
elif 'forceinvdeltas' in debug.debug_flags:
4609
return self._get_convertable_inventory_stream(revision_ids,
4610
delta_versus_null=missing)
4611
elif from_format.network_name() == self.to_format.network_name():
4613
return self._get_simple_inventory_stream(revision_ids,
4615
elif (not from_format.supports_chks and not self.to_format.supports_chks
4616
and from_format._serializer == self.to_format._serializer):
4617
# Essentially the same format.
4618
return self._get_simple_inventory_stream(revision_ids,
4621
# Any time we switch serializations, we want to use an
4622
# inventory-delta based approach.
4623
return self._get_convertable_inventory_stream(revision_ids,
4624
delta_versus_null=missing)
4626
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4627
# NB: This currently reopens the inventory weave in source;
4628
# using a single stream interface instead would avoid this.
4629
from_weave = self.from_repository.inventories
4631
delta_closure = True
4633
delta_closure = not self.delta_on_metadata()
4634
yield ('inventories', from_weave.get_record_stream(
4635
[(rev_id,) for rev_id in revision_ids],
4636
self.inventory_fetch_order(), delta_closure))
4638
def _get_convertable_inventory_stream(self, revision_ids,
4639
delta_versus_null=False):
4640
# The two formats are sufficiently different that there is no fast
4641
# path, so we need to send just inventorydeltas, which any
4642
# sufficiently modern client can insert into any repository.
4643
# The StreamSink code expects to be able to
4644
# convert on the target, so we need to put bytes-on-the-wire that can
4645
# be converted. That means inventory deltas (if the remote is <1.19,
4646
# RemoteStreamSink will fallback to VFS to insert the deltas).
4647
yield ('inventory-deltas',
4648
self._stream_invs_as_deltas(revision_ids,
4649
delta_versus_null=delta_versus_null))
4651
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4652
"""Return a stream of inventory-deltas for the given rev ids.
4654
:param revision_ids: The list of inventories to transmit
4655
:param delta_versus_null: Don't try to find a minimal delta for this
4656
entry, instead compute the delta versus the NULL_REVISION. This
4657
effectively streams a complete inventory. Used for stuff like
4658
filling in missing parents, etc.
4660
from_repo = self.from_repository
4661
revision_keys = [(rev_id,) for rev_id in revision_ids]
4662
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4663
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4665
inventories = self.from_repository.iter_inventories(
4666
revision_ids, 'topological')
4667
format = from_repo._format
4668
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4669
inventory_cache = lru_cache.LRUCache(50)
4670
null_inventory = from_repo.revision_tree(
4671
_mod_revision.NULL_REVISION).inventory
4672
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4673
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4674
# repo back into a non-rich-root repo ought to be allowed)
4675
serializer = inventory_delta.InventoryDeltaSerializer(
4676
versioned_root=format.rich_root_data,
4677
tree_references=format.supports_tree_reference)
4678
for inv in inventories:
4679
key = (inv.revision_id,)
4680
parent_keys = parent_map.get(key, ())
4682
if not delta_versus_null and parent_keys:
4683
# The caller did not ask for complete inventories and we have
4684
# some parents that we can delta against. Make a delta against
4685
# each parent so that we can find the smallest.
4686
parent_ids = [parent_key[0] for parent_key in parent_keys]
4687
for parent_id in parent_ids:
4688
if parent_id not in invs_sent_so_far:
4689
# We don't know that the remote side has this basis, so
4692
if parent_id == _mod_revision.NULL_REVISION:
4693
parent_inv = null_inventory
4695
parent_inv = inventory_cache.get(parent_id, None)
4696
if parent_inv is None:
4697
parent_inv = from_repo.get_inventory(parent_id)
4698
candidate_delta = inv._make_delta(parent_inv)
4699
if (delta is None or
4700
len(delta) > len(candidate_delta)):
4701
delta = candidate_delta
4702
basis_id = parent_id
4704
# Either none of the parents ended up being suitable, or we
4705
# were asked to delta against NULL
4706
basis_id = _mod_revision.NULL_REVISION
4707
delta = inv._make_delta(null_inventory)
4708
invs_sent_so_far.add(inv.revision_id)
4709
inventory_cache[inv.revision_id] = inv
4710
delta_serialized = ''.join(
4711
serializer.delta_to_lines(basis_id, key[-1], delta))
4712
yield versionedfile.FulltextContentFactory(
4713
key, parent_keys, None, delta_serialized)
4716
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4717
stop_revision=None):
4718
"""Extend the partial history to include a given index
4720
If a stop_index is supplied, stop when that index has been reached.
4721
If a stop_revision is supplied, stop when that revision is
4722
encountered. Otherwise, stop when the beginning of history is
4725
:param stop_index: The index which should be present. When it is
4726
present, history extension will stop.
4727
:param stop_revision: The revision id which should be present. When
4728
it is encountered, history extension will stop.
4730
start_revision = partial_history_cache[-1]
4731
iterator = repo.iter_reverse_revision_history(start_revision)
4733
#skip the last revision in the list
4736
if (stop_index is not None and
4737
len(partial_history_cache) > stop_index):
4739
if partial_history_cache[-1] == stop_revision:
4741
revision_id = iterator.next()
4742
partial_history_cache.append(revision_id)
4743
except StopIteration: