13
13
# You should have received a copy of the GNU General Public License
14
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
from cStringIO import StringIO
17
19
from bzrlib.lazy_import import lazy_import
18
20
lazy_import(globals(), """
23
25
from bzrlib import (
42
37
revision as _mod_revision,
49
from bzrlib.bundle import serializer
50
42
from bzrlib.revisiontree import RevisionTree
51
43
from bzrlib.store.versioned import VersionedFileStore
44
from bzrlib.store.text import TextStore
52
45
from bzrlib.testament import Testament
60
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
49
from bzrlib.decorators import needs_read_lock, needs_write_lock
61
50
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)
51
from bzrlib.inventory import Inventory, InventoryDirectory, ROOT_ID
52
from bzrlib.symbol_versioning import (
56
from bzrlib.trace import mutter, note, warning
74
59
# Old formats display a warning, but only once
75
60
_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 it's 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 its
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
63
######################################################################
895
class Repository(_RelockDebugMixin, controldir.ControlComponent):
66
class Repository(object):
896
67
"""Repository holding history for one or more branches.
898
69
The repository holds and retrieves historical information including
899
70
revisions and file history. It's normally accessed only by the Branch,
900
71
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
73
The Repository builds on top of Stores and a Transport, which respectively
74
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 it's 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
78
_file_ids_altered_regex = lazy_regex.lazy_compile(
955
79
r'file_id="(?P<file_id>[^"]+)"'
956
r'.* revision="(?P<revision_id>[^"]+)"'
80
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
84
def add_inventory(self, revision_id, inv, parents):
1032
85
"""Add the inventory inv to the repository as revision_id.
1034
87
:param parents: The revision ids of the parents that revision_id
1035
88
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.
90
returns the sha1 of the serialized inventory.
1040
if not self.is_in_write_group():
1041
raise AssertionError("%r not in write group" % (self,))
92
revision_id = osutils.safe_revision_id(revision_id)
1042
93
_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()
94
assert inv.revision_id is None or inv.revision_id == revision_id, \
95
"Mismatch between inventory revision" \
96
" id and insertion revid (%r, %r)" % (inv.revision_id, revision_id)
97
assert inv.root is not None
98
inv_text = self.serialise_inventory(inv)
99
inv_sha1 = osutils.sha_string(inv_text)
100
inv_vf = self.control_weaves.get_weave('inventory',
101
self.get_transaction())
102
self._inventory_add_lines(inv_vf, revision_id, parents,
103
osutils.split_lines(inv_text))
106
def _inventory_add_lines(self, inv_vf, revision_id, parents, lines):
108
for parent in parents:
110
final_parents.append(parent)
112
inv_vf.add_lines(revision_id, final_parents, lines)
1119
115
def add_revision(self, revision_id, rev, inv=None, config=None):
1120
116
"""Add rev to the revision store as revision_id.
1191
187
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]))
190
for id in revision_ids:
191
if self.has_revision(id):
1313
196
def create(a_bzrdir):
1314
197
"""Construct the current default format repository in a_bzrdir."""
1315
198
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1317
def __init__(self, _format, a_bzrdir, control_files):
200
def __init__(self, _format, a_bzrdir, control_files, _revision_store, control_store, text_store):
1318
201
"""instantiate a Repository.
1320
203
:param _format: The format of the repository on disk.
1321
204
:param a_bzrdir: The BzrDir of the repository.
206
In the future we will have a single api for all stores for
207
getting file texts, inventories and revisions, then
208
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
210
super(Repository, self).__init__()
1327
211
self._format = _format
1328
212
# the following are part of the public API for Repository:
1329
213
self.bzrdir = a_bzrdir
1330
214
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
215
self._revision_store = _revision_store
216
self.text_store = text_store
217
# backwards compatibility
218
self.weave_store = text_store
219
# not right yet - should be more semantically clear ?
221
self.control_store = control_store
222
self.control_weaves = control_store
223
# TODO: make sure to construct the right store classes, etc, depending
224
# on whether escaping is required.
225
self._warn_if_deprecated()
1354
227
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
228
return '%s(%r)' % (self.__class__.__name__,
229
self.bzrdir.transport.base)
1393
231
def is_locked(self):
1394
232
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
234
def lock_write(self, token=None):
1401
235
"""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
237
:param token: if this is already locked, then lock_write will fail
1414
238
unless the token matches the existing lock.
1415
239
:returns: a token if this instance supports tokens, otherwise None.
1843
401
# Most control formats need the repository to be specifically
1844
402
# created, but on some old all-in-one formats it's not needed
1846
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
404
dest_repo = self._format.initialize(a_bzrdir, shared=self.is_shared())
1847
405
except errors.UninitializableFormat:
1848
406
dest_repo = a_bzrdir.open_repository()
407
self.copy_content_into(dest_repo, revision_id)
1849
408
return dest_repo
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)
1859
410
@needs_read_lock
1860
411
def has_revision(self, revision_id):
1861
412
"""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]
413
revision_id = osutils.safe_revision_id(revision_id)
414
return self._revision_store.has_revision_id(revision_id,
415
self.get_transaction())
1884
417
@needs_read_lock
1885
418
def get_revision_reconcile(self, revision_id):
1886
419
"""'reconcile' helper routine that allows access to a revision always.
1888
421
This variant of get_revision does not cross check the weave graph
1889
422
against the revision one as get_revision does: but it should only
1890
423
be used by reconcile, or reconcile-alike commands that are correcting
1891
424
or testing the revision graph.
1893
return self._get_revisions([revision_id])[0]
426
if not revision_id or not isinstance(revision_id, basestring):
427
raise errors.InvalidRevisionId(revision_id=revision_id,
429
return self.get_revisions([revision_id])[0]
1895
431
@needs_read_lock
1896
432
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):
433
revision_ids = [osutils.safe_revision_id(r) for r in revision_ids]
434
revs = self._revision_store.get_revisions(revision_ids,
435
self.get_transaction())
437
assert not isinstance(rev.revision_id, unicode)
438
for parent_id in rev.parent_ids:
439
assert not isinstance(parent_id, unicode)
443
def get_revision_xml(self, revision_id):
444
# TODO: jam 20070210 This shouldn't be necessary since get_revision
445
# would have already do it.
446
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
447
revision_id = osutils.safe_revision_id(revision_id)
448
rev = self.get_revision(revision_id)
450
# the current serializer..
451
self._revision_store._serializer.write_revision(rev, rev_tmp)
453
return rev_tmp.getvalue()
456
def get_revision(self, revision_id):
457
"""Return the Revision object for a named revision"""
458
# TODO: jam 20070210 get_revision_reconcile should do this for us
459
revision_id = osutils.safe_revision_id(revision_id)
460
r = self.get_revision_reconcile(revision_id)
461
# weave corruption can lead to absent revision markers that should be
463
# the following test is reasonably cheap (it needs a single weave read)
464
# and the weave is cached in read transactions. In write transactions
465
# it is not cached but typically we only read a small number of
466
# revisions. For knits when they are introduced we will probably want
467
# to ensure that caching write transactions are in use.
468
inv = self.get_inventory_weave()
469
self._check_revision_parents(r, inv)
473
def get_deltas_for_revisions(self, revisions):
1942
474
"""Produce a generator of revision deltas.
1944
476
Note that the input is a sequence of REVISIONS, not revision_ids.
1945
477
Trees will be held in memory until the generator exits.
1946
478
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
480
required_trees = set()
1954
481
for revision in revisions:
1955
482
required_trees.add(revision.revision_id)
1956
483
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
484
trees = dict((t.get_revision_id(), t) for
485
t in self.revision_trees(required_trees))
1973
486
for revision in revisions:
1974
487
if not revision.parent_ids:
1975
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
488
old_tree = self.revision_tree(None)
1977
490
old_tree = trees[revision.parent_ids[0]]
1978
491
yield trees[revision.revision_id].changes_from(old_tree)
1980
493
@needs_read_lock
1981
def get_revision_delta(self, revision_id, specific_fileids=None):
494
def get_revision_delta(self, revision_id):
1982
495
"""Return the delta for one revision.
1984
497
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
500
r = self.get_revision(revision_id)
1992
return list(self.get_deltas_for_revisions([r],
1993
specific_fileids=specific_fileids))[0]
501
return list(self.get_deltas_for_revisions([r]))[0]
503
def _check_revision_parents(self, revision, inventory):
504
"""Private to Repository and Fetch.
506
This checks the parentage of revision in an inventory weave for
507
consistency and is only applicable to inventory-weave-for-ancestry
508
using repository formats & fetchers.
510
weave_parents = inventory.get_parents(revision.revision_id)
511
weave_names = inventory.versions()
512
for parent_id in revision.parent_ids:
513
if parent_id in weave_names:
514
# this parent must not be a ghost.
515
if not parent_id in weave_parents:
517
raise errors.CorruptRepository(self)
1995
519
@needs_write_lock
1996
520
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
521
revision_id = osutils.safe_revision_id(revision_id)
1997
522
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)
523
self._revision_store.add_revision_signature_text(revision_id,
525
self.get_transaction())
527
def fileids_altered_by_revision_ids(self, revision_ids):
528
"""Find the file ids and versions affected by revisions.
530
:param revisions: an iterable containing revision ids.
531
:return: a dictionary mapping altered file-ids to an iterable of
532
revision_ids. Each altered file-ids has the exact revision_ids that
533
altered it listed explicitly.
535
assert self._serializer.support_altered_by_hack, \
536
("fileids_altered_by_revision_ids only supported for branches "
537
"which store inventory as unnested xml, not on %r" % self)
538
selected_revision_ids = set(osutils.safe_revision_id(r)
539
for r in revision_ids)
540
w = self.get_inventory_weave()
2042
543
# this code needs to read every new line in every inventory for the
2043
544
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2044
# not present in one of those inventories is unnecessary but not
545
# not present in one of those inventories is unnecessary but not
2045
546
# 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
547
# inventory lines : we only select file ids altered in one of those
2047
548
# revisions. We don't need to see all lines in the inventory because
2048
549
# only those added in an inventory in rev X can contain a revision=X
2058
559
search = self._file_ids_altered_regex.search
2059
560
unescape = _unescape_xml
2060
561
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
562
pb = ui.ui_factory.nested_progress_bar()
2238
return self._do_generate_text_key_index(ancestors,
2239
text_key_references, pb)
564
for line in w.iter_lines_added_or_present_in_versions(
565
selected_revision_ids, pb=pb):
569
# One call to match.group() returning multiple items is quite a
570
# bit faster than 2 calls to match.group() each returning 1
571
file_id, revision_id = match.group('file_id', 'revision_id')
573
# Inlining the cache lookups helps a lot when you make 170,000
574
# lines and 350k ids, versus 8.4 unique ids.
575
# Using a cache helps in 2 ways:
576
# 1) Avoids unnecessary decoding calls
577
# 2) Re-uses cached strings, which helps in future set and
579
# (2) is enough that removing encoding entirely along with
580
# the cache (so we are using plain strings) results in no
581
# performance improvement.
583
revision_id = unescape_revid_cache[revision_id]
585
unescaped = unescape(revision_id)
586
unescape_revid_cache[revision_id] = unescaped
587
revision_id = unescaped
589
if revision_id in selected_revision_ids:
591
file_id = unescape_fileid_cache[file_id]
593
unescaped = unescape(file_id)
594
unescape_fileid_cache[file_id] = unescaped
596
setdefault(file_id, set()).add(revision_id)
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)
602
def get_inventory_weave(self):
603
return self.control_weaves.get_weave('inventory',
604
self.get_transaction())
2380
606
@needs_read_lock
2381
607
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.
608
"""Get Inventory object by hash."""
609
# TODO: jam 20070210 Technically we don't need to sanitize, since all
610
# called functions must sanitize.
611
revision_id = osutils.safe_revision_id(revision_id)
612
return self.deserialise_inventory(
613
revision_id, self.get_inventory_xml(revision_id))
615
def deserialise_inventory(self, revision_id, xml):
616
"""Transform the xml into an inventory object.
2448
618
:param revision_id: The expected revision id of the inventory.
2449
619
: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))
621
revision_id = osutils.safe_revision_id(revision_id)
622
result = self._serializer.read_inventory_from_string(xml)
623
result.root.revision = revision_id
2459
def get_serializer_format(self):
2460
return self._serializer.format_num
626
def serialise_inventory(self, inv):
627
return self._serializer.write_inventory_to_string(inv)
2462
629
@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')
630
def get_inventory_xml(self, revision_id):
631
"""Get inventory XML as a file object."""
632
revision_id = osutils.safe_revision_id(revision_id)
2467
text, revision_id = texts.next()
2468
except StopIteration:
634
assert isinstance(revision_id, str), type(revision_id)
635
iw = self.get_inventory_weave()
636
return iw.get_text(revision_id)
2469
638
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])
641
def get_inventory_sha1(self, revision_id):
642
"""Return the sha1 hash of the inventory entry
644
# TODO: jam 20070210 Shouldn't this be deprecated / removed?
645
revision_id = osutils.safe_revision_id(revision_id)
646
return self.get_revision(revision_id).inventory_sha1
649
def get_revision_graph(self, revision_id=None):
650
"""Return a dictionary containing the revision graph.
652
:param revision_id: The revision_id to get a graph from. If None, then
653
the entire revision graph is returned. This is a deprecated mode of
654
operation and will be removed in the future.
655
:return: a dictionary of revision_id->revision_parents_list.
657
# special case NULL_REVISION
658
if revision_id == _mod_revision.NULL_REVISION:
660
revision_id = osutils.safe_revision_id(revision_id)
661
a_weave = self.get_inventory_weave()
662
all_revisions = self._eliminate_revisions_not_present(
664
entire_graph = dict([(node, a_weave.get_parents(node)) for
665
node in all_revisions])
666
if revision_id is None:
668
elif revision_id not in entire_graph:
669
raise errors.NoSuchRevision(self, revision_id)
671
# add what can be reached from revision_id
673
pending = set([revision_id])
674
while len(pending) > 0:
676
result[node] = entire_graph[node]
677
for revision_id in result[node]:
678
if revision_id not in result:
679
pending.add(revision_id)
683
def get_revision_graph_with_ghosts(self, revision_ids=None):
684
"""Return a graph of the revisions with ghosts marked as applicable.
686
:param revision_ids: an iterable of revisions to graph or None for all.
687
:return: a Graph object with the graph reachable from revision_ids.
689
result = graph.Graph()
691
pending = set(self.all_revision_ids())
694
pending = set(osutils.safe_revision_id(r) for r in revision_ids)
695
# special case NULL_REVISION
696
if _mod_revision.NULL_REVISION in pending:
697
pending.remove(_mod_revision.NULL_REVISION)
698
required = set(pending)
701
revision_id = pending.pop()
703
rev = self.get_revision(revision_id)
704
except errors.NoSuchRevision:
705
if revision_id in required:
708
result.add_ghost(revision_id)
710
for parent_id in rev.parent_ids:
711
# is this queued or done ?
712
if (parent_id not in pending and
713
parent_id not in done):
715
pending.add(parent_id)
716
result.add_node(revision_id, rev.parent_ids)
717
done.add(revision_id)
720
def _get_history_vf(self):
721
"""Get a versionedfile whose history graph reflects all revisions.
723
For weave repositories, this is the inventory weave.
725
return self.get_inventory_weave()
2507
727
def iter_reverse_revision_history(self, revision_id):
2508
728
"""Iterate backwards through revision ids in the lefthand history
2525
771
reconciler = RepoReconciler(self, thorough=thorough)
2526
772
reconciler.reconcile()
2527
773
return reconciler
2529
def _refresh_data(self):
2530
"""Helper called from lock_* to ensure coherency with disk.
2532
The default implementation does nothing; it is however possible
2533
for repositories to maintain loaded indices across multiple locks
2534
by checking inside their implementation of this method to see
2535
whether their indices are still valid. This depends of course on
2536
the disk format being validatable in this manner. This method is
2537
also called by the refresh_data() public interface to cause a refresh
2538
to occur while in a write lock so that data inserted by a smart server
2539
push operation is visible on the client's instance of the physical
2543
775
@needs_read_lock
2544
776
def revision_tree(self, revision_id):
2545
777
"""Return Tree for a revision on this branch.
2547
`revision_id` may be NULL_REVISION for the empty tree revision.
779
`revision_id` may be None for the empty tree revision.
2549
revision_id = _mod_revision.ensure_null(revision_id)
2550
781
# TODO: refactor this to use an existing revision object
2551
782
# so we don't need to read it in twice.
2552
if revision_id == _mod_revision.NULL_REVISION:
2553
return RevisionTree(self, Inventory(root_id=None),
783
if revision_id is None or revision_id == _mod_revision.NULL_REVISION:
784
return RevisionTree(self, Inventory(root_id=None),
2554
785
_mod_revision.NULL_REVISION)
2556
inv = self.get_inventory(revision_id)
787
revision_id = osutils.safe_revision_id(revision_id)
788
inv = self.get_revision_inventory(revision_id)
2557
789
return RevisionTree(self, inv, revision_id)
2559
792
def revision_trees(self, revision_ids):
2560
"""Return Trees for revisions in this repository.
2562
:param revision_ids: a sequence of revision-ids;
2563
a revision-id may not be None or 'null:'
2565
inventories = self.iter_inventories(revision_ids)
2566
for inv in inventories:
2567
yield RevisionTree(self, inv, inv.revision_id)
2569
def _filtered_revision_trees(self, revision_ids, file_ids):
2570
"""Return Tree for a revision on this branch with only some files.
2572
:param revision_ids: a sequence of revision-ids;
2573
a revision-id may not be None or 'null:'
2574
:param file_ids: if not None, the result is filtered
2575
so that only those file-ids, their parents and their
2576
children are included.
2578
inventories = self.iter_inventories(revision_ids)
2579
for inv in inventories:
2580
# Should we introduce a FilteredRevisionTree class rather
2581
# than pre-filter the inventory here?
2582
filtered_inv = inv.filter(file_ids)
2583
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
793
"""Return Tree for a revision on this branch.
795
`revision_id` may not be None or 'null:'"""
796
assert None not in revision_ids
797
assert _mod_revision.NULL_REVISION not in revision_ids
798
texts = self.get_inventory_weave().get_texts(revision_ids)
799
for text, revision_id in zip(texts, revision_ids):
800
inv = self.deserialise_inventory(revision_id, text)
801
yield RevisionTree(self, inv, revision_id)
2585
803
@needs_read_lock
2586
def get_ancestry(self, revision_id, topo_sorted=True):
804
def get_ancestry(self, revision_id):
2587
805
"""Return a list of revision-ids integrated by a revision.
2589
The first element of the list is always None, indicating the origin
2590
revision. This might change when we have history horizons, or
807
The first element of the list is always None, indicating the origin
808
revision. This might change when we have history horizons, or
2591
809
perhaps we should have a new API.
2593
811
This is topologically sorted.
2595
if _mod_revision.is_null(revision_id):
813
if revision_id is None:
815
revision_id = osutils.safe_revision_id(revision_id)
2597
816
if not self.has_revision(revision_id):
2598
817
raise errors.NoSuchRevision(self, revision_id)
2599
graph = self.get_graph()
2601
search = graph._make_breadth_first_searcher([revision_id])
2604
found, ghosts = search.next_with_ghosts()
2605
except StopIteration:
2608
if _mod_revision.NULL_REVISION in keys:
2609
keys.remove(_mod_revision.NULL_REVISION)
2611
parent_map = graph.get_parent_map(keys)
2612
keys = tsort.topo_sort(parent_map)
2613
return [None] + list(keys)
2615
def pack(self, hint=None, clean_obsolete_packs=False):
2616
"""Compress the data within the repository.
2618
This operation only makes sense for some repository types. For other
2619
types it should be a no-op that just returns.
2621
This stub method does not require a lock, but subclasses should use
2622
@needs_write_lock as this is a long running call it's reasonable to
2623
implicitly lock for the user.
2625
:param hint: If not supplied, the whole repository is packed.
2626
If supplied, the repository may use the hint parameter as a
2627
hint for the parts of the repository to pack. A hint can be
2628
obtained from the result of commit_write_group(). Out of
2629
date hints are simply ignored, because concurrent operations
2630
can obsolete them rapidly.
2632
:param clean_obsolete_packs: Clean obsolete packs immediately after
818
w = self.get_inventory_weave()
819
candidates = w.get_ancestry(revision_id)
820
return [None] + candidates # self._eliminate_revisions_not_present(candidates)
823
def print_file(self, file, revision_id):
824
"""Print `file` to stdout.
826
FIXME RBC 20060125 as John Meinel points out this is a bad api
827
- it writes to stdout, it assumes that that is valid etc. Fix
828
by creating a new more flexible convenience function.
830
revision_id = osutils.safe_revision_id(revision_id)
831
tree = self.revision_tree(revision_id)
832
# use inventory as it was in that revision
833
file_id = tree.inventory.path2id(file)
835
# TODO: jam 20060427 Write a test for this code path
836
# it had a bug in it, and was raising the wrong
838
raise errors.BzrError("%r is not present in revision %s" % (file, revision_id))
839
tree.print_file(file_id)
2636
841
def get_transaction(self):
2637
842
return self.control_files.get_transaction()
2639
def get_parent_map(self, revision_ids):
2640
"""See graph.StackedParentsProvider.get_parent_map"""
2641
# revisions index works in keys; this just works in revisions
2642
# therefore wrap and unwrap
2645
for revision_id in revision_ids:
2646
if revision_id == _mod_revision.NULL_REVISION:
2647
result[revision_id] = ()
2648
elif revision_id is None:
2649
raise ValueError('get_parent_map(None) is not valid')
2651
query_keys.append((revision_id ,))
2652
for ((revision_id,), parent_keys) in \
2653
self.revisions.get_parent_map(query_keys).iteritems():
2655
result[revision_id] = tuple([parent_revid
2656
for (parent_revid,) in parent_keys])
2658
result[revision_id] = (_mod_revision.NULL_REVISION,)
2661
def _make_parents_provider(self):
2665
def get_known_graph_ancestry(self, revision_ids):
2666
"""Return the known graph for a set of revision ids and their ancestors.
2668
st = static_tuple.StaticTuple
2669
revision_keys = [st(r_id).intern() for r_id in revision_ids]
2670
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
2671
return graph.GraphThunkIdsToKeys(known_graph)
2673
def get_graph(self, other_repository=None):
2674
"""Return the graph walker for this repository format"""
2675
parents_provider = self._make_parents_provider()
2676
if (other_repository is not None and
2677
not self.has_same_location(other_repository)):
2678
parents_provider = graph.StackedParentsProvider(
2679
[parents_provider, other_repository._make_parents_provider()])
2680
return graph.Graph(parents_provider)
2682
def _get_versioned_file_checker(self, text_key_references=None,
2684
"""Return an object suitable for checking versioned files.
2686
:param text_key_references: if non-None, an already built
2687
dictionary mapping text keys ((fileid, revision_id) tuples)
2688
to whether they were referred to by the inventory of the
2689
revision_id that they contain. If None, this will be
2691
:param ancestors: Optional result from
2692
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2695
return _VersionedFileChecker(self,
2696
text_key_references=text_key_references, ancestors=ancestors)
2698
def revision_ids_to_search_result(self, result_set):
2699
"""Convert a set of revision ids to a graph SearchResult."""
2700
result_parents = set()
2701
for parents in self.get_graph().get_parent_map(
2702
result_set).itervalues():
2703
result_parents.update(parents)
2704
included_keys = result_set.intersection(result_parents)
2705
start_keys = result_set.difference(included_keys)
2706
exclude_keys = result_parents.difference(result_set)
2707
result = graph.SearchResult(start_keys, exclude_keys,
2708
len(result_set), result_set)
844
def revision_parents(self, revision_id):
845
revision_id = osutils.safe_revision_id(revision_id)
846
return self.get_inventory_weave().parent_names(revision_id)
2711
848
@needs_write_lock
2712
849
def set_make_working_trees(self, new_value):
3397
1312
InterRepository.get(other).method_name(parameters).
3400
_walk_to_common_revisions_batch_size = 50
3401
1315
_optimisers = []
3402
1316
"""The available optimised InterRepository types."""
3405
1318
def copy_content(self, revision_id=None):
3406
"""Make a complete copy of the content in self into destination.
3408
This is a destructive operation! Do not use it on existing
3411
:param revision_id: Only copy the content needed to construct
3412
revision_id and its parents.
3415
self.target.set_make_working_trees(self.source.make_working_trees())
3416
except NotImplementedError:
3418
self.target.fetch(self.source, revision_id=revision_id)
3421
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
1319
raise NotImplementedError(self.copy_content)
1321
def fetch(self, revision_id=None, pb=None):
3423
1322
"""Fetch the content required to construct revision_id.
3425
1324
The content is copied from self.source to self.target.
3427
1326
:param revision_id: if None all content is copied, if NULL_REVISION no
3428
1327
content is copied.
3432
ui.ui_factory.warn_experimental_format_fetch(self)
3433
from bzrlib.fetch import RepoFetcher
3434
# See <https://launchpad.net/bugs/456077> asking for a warning here
3435
if self.source._format.network_name() != self.target._format.network_name():
3436
ui.ui_factory.show_user_warning('cross_format_fetch',
3437
from_format=self.source._format,
3438
to_format=self.target._format)
3439
f = RepoFetcher(to_repository=self.target,
3440
from_repository=self.source,
3441
last_revision=revision_id,
3442
fetch_spec=fetch_spec,
3443
find_ghosts=find_ghosts)
3445
def _walk_to_common_revisions(self, revision_ids):
3446
"""Walk out from revision_ids in source to revisions target has.
3448
:param revision_ids: The start point for the search.
3449
:return: A set of revision ids.
3451
target_graph = self.target.get_graph()
3452
revision_ids = frozenset(revision_ids)
3453
missing_revs = set()
3454
source_graph = self.source.get_graph()
3455
# ensure we don't pay silly lookup costs.
3456
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3457
null_set = frozenset([_mod_revision.NULL_REVISION])
3458
searcher_exhausted = False
3462
# Iterate the searcher until we have enough next_revs
3463
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3465
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3466
next_revs.update(next_revs_part)
3467
ghosts.update(ghosts_part)
3468
except StopIteration:
3469
searcher_exhausted = True
3471
# If there are ghosts in the source graph, and the caller asked for
3472
# them, make sure that they are present in the target.
3473
# We don't care about other ghosts as we can't fetch them and
3474
# haven't been asked to.
3475
ghosts_to_check = set(revision_ids.intersection(ghosts))
3476
revs_to_get = set(next_revs).union(ghosts_to_check)
3478
have_revs = set(target_graph.get_parent_map(revs_to_get))
3479
# we always have NULL_REVISION present.
3480
have_revs = have_revs.union(null_set)
3481
# Check if the target is missing any ghosts we need.
3482
ghosts_to_check.difference_update(have_revs)
3484
# One of the caller's revision_ids is a ghost in both the
3485
# source and the target.
3486
raise errors.NoSuchRevision(
3487
self.source, ghosts_to_check.pop())
3488
missing_revs.update(next_revs - have_revs)
3489
# Because we may have walked past the original stop point, make
3490
# sure everything is stopped
3491
stop_revs = searcher.find_seen_ancestors(have_revs)
3492
searcher.stop_searching_any(stop_revs)
3493
if searcher_exhausted:
3495
return searcher.get_result()
1328
:param pb: optional progress bar to use for progress reports. If not
1329
provided a default one will be created.
1331
Returns the copied revision count and the failed revisions in a tuple:
1334
raise NotImplementedError(self.fetch)
3497
1336
@needs_read_lock
3498
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1337
def missing_revision_ids(self, revision_id=None):
3499
1338
"""Return the revision ids that source has that target does not.
1340
These are returned in topological order.
3501
1342
:param revision_id: only return revision ids included by this
3503
:param find_ghosts: If True find missing revisions in deep history
3504
rather than just finding the surface difference.
3505
:return: A bzrlib.graph.SearchResult.
3507
# stop searching at found target revisions.
3508
if not find_ghosts and revision_id is not None:
3509
return self._walk_to_common_revisions([revision_id])
3510
1345
# generic, possibly worst case, slow code path.
3511
1346
target_ids = set(self.target.all_revision_ids())
3512
1347
if revision_id is not None:
1348
# TODO: jam 20070210 InterRepository is internal enough that it
1349
# should assume revision_ids are already utf-8
1350
revision_id = osutils.safe_revision_id(revision_id)
3513
1351
source_ids = self.source.get_ancestry(revision_id)
3514
if source_ids[0] is not None:
3515
raise AssertionError()
1352
assert source_ids[0] is None
3516
1353
source_ids.pop(0)
3518
1355
source_ids = self.source.all_revision_ids()
3519
1356
result_set = set(source_ids).difference(target_ids)
3520
return self.source.revision_ids_to_search_result(result_set)
3523
def _same_model(source, target):
3524
"""True if source and target have the same data representation.
3526
Note: this is always called on the base class; overriding it in a
3527
subclass will have no effect.
3530
InterRepository._assert_same_model(source, target)
3532
except errors.IncompatibleRepositories, e:
3536
def _assert_same_model(source, target):
3537
"""Raise an exception if two repositories do not use the same model.
3539
if source.supports_rich_root() != target.supports_rich_root():
3540
raise errors.IncompatibleRepositories(source, target,
3541
"different rich-root support")
3542
if source._serializer != target._serializer:
3543
raise errors.IncompatibleRepositories(source, target,
3544
"different serializers")
1357
# this may look like a no-op: its not. It preserves the ordering
1358
# other_ids had while only returning the members from other_ids
1359
# that we've decided we need.
1360
return [rev_id for rev_id in source_ids if rev_id in result_set]
3547
1363
class InterSameDataRepository(InterRepository):
3548
1364
"""Code for converting between repositories that represent the same data.
3550
1366
Data format and model must match for this to work.
3554
1370
def _get_repo_format_to_test(self):
3555
"""Repository format for testing with.
3557
InterSameData can pull from subtree to subtree and from non-subtree to
3558
non-subtree, so we test this with the richest repository format.
3560
from bzrlib.repofmt import knitrepo
3561
return knitrepo.RepositoryFormatKnit3()
1371
"""Repository format for testing with."""
1372
return RepositoryFormat.get_default_format()
3564
1375
def is_compatible(source, target):
3565
return InterRepository._same_model(source, target)
1376
if source.supports_rich_root() != target.supports_rich_root():
1378
if source._serializer != target._serializer:
1383
def copy_content(self, revision_id=None):
1384
"""Make a complete copy of the content in self into destination.
1386
This is a destructive operation! Do not use it on existing
1389
:param revision_id: Only copy the content needed to construct
1390
revision_id and its parents.
1393
self.target.set_make_working_trees(self.source.make_working_trees())
1394
except NotImplementedError:
1396
# TODO: jam 20070210 This is fairly internal, so we should probably
1397
# just assert that revision_id is not unicode.
1398
revision_id = osutils.safe_revision_id(revision_id)
1399
# but don't bother fetching if we have the needed data now.
1400
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1401
self.target.has_revision(revision_id)):
1403
self.target.fetch(self.source, revision_id=revision_id)
1406
def fetch(self, revision_id=None, pb=None):
1407
"""See InterRepository.fetch()."""
1408
from bzrlib.fetch import GenericRepoFetcher
1409
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1410
self.source, self.source._format, self.target,
1411
self.target._format)
1412
# TODO: jam 20070210 This should be an assert, not a translate
1413
revision_id = osutils.safe_revision_id(revision_id)
1414
f = GenericRepoFetcher(to_repository=self.target,
1415
from_repository=self.source,
1416
last_revision=revision_id,
1418
return f.count_copied, f.failed_revisions
3568
1421
class InterWeaveRepo(InterSameDataRepository):
3569
"""Optimised code paths between Weave based repositories.
3571
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3572
implemented lazy inter-object optimisation.
1422
"""Optimised code paths between Weave based repositories."""
3576
1425
def _get_repo_format_to_test(self):
3745
1615
def is_compatible(source, target):
3746
"""Be compatible with Knit2 source and Knit3 target"""
3747
# This is redundant with format.check_conversion_target(), however that
3748
# raises an exception, and we just want to say "False" as in we won't
3749
# support converting between these formats.
3750
if 'IDS_never' in debug.debug_flags:
3752
if source.supports_rich_root() and not target.supports_rich_root():
3754
if (source._format.supports_tree_reference
3755
and not target._format.supports_tree_reference):
3757
if target._fallback_repositories and target._format.supports_chks:
3758
# IDS doesn't know how to copy CHKs for the parent inventories it
3759
# adds to stacked repos.
3761
if 'IDS_always' in debug.debug_flags:
1616
if not source.supports_rich_root() and target.supports_rich_root():
3763
# Only use this code path for local source and target. IDS does far
3764
# too much IO (both bandwidth and roundtrips) over a network.
3765
if not source.bzrdir.transport.base.startswith('file:///'):
3767
if not target.bzrdir.transport.base.startswith('file:///'):
3771
def _get_trees(self, revision_ids, cache):
3773
for rev_id in revision_ids:
3775
possible_trees.append((rev_id, cache[rev_id]))
3777
# Not cached, but inventory might be present anyway.
3779
tree = self.source.revision_tree(rev_id)
3780
except errors.NoSuchRevision:
3781
# Nope, parent is ghost.
3784
cache[rev_id] = tree
3785
possible_trees.append((rev_id, tree))
3786
return possible_trees
3788
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3789
"""Get the best delta and base for this revision.
3791
:return: (basis_id, delta)
3794
# Generate deltas against each tree, to find the shortest.
3795
texts_possibly_new_in_tree = set()
3796
for basis_id, basis_tree in possible_trees:
3797
delta = tree.inventory._make_delta(basis_tree.inventory)
3798
for old_path, new_path, file_id, new_entry in delta:
3799
if new_path is None:
3800
# This file_id isn't present in the new rev, so we don't
3804
# Rich roots are handled elsewhere...
3806
kind = new_entry.kind
3807
if kind != 'directory' and kind != 'file':
3808
# No text record associated with this inventory entry.
3810
# This is a directory or file that has changed somehow.
3811
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3812
deltas.append((len(delta), basis_id, delta))
3814
return deltas[0][1:]
3816
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3817
"""Find all parent revisions that are absent, but for which the
3818
inventory is present, and copy those inventories.
3820
This is necessary to preserve correctness when the source is stacked
3821
without fallbacks configured. (Note that in cases like upgrade the
3822
source may be not have _fallback_repositories even though it is
3826
for parents in parent_map.values():
3827
parent_revs.update(parents)
3828
present_parents = self.source.get_parent_map(parent_revs)
3829
absent_parents = set(parent_revs).difference(present_parents)
3830
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3831
(rev_id,) for rev_id in absent_parents)
3832
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3833
for parent_tree in self.source.revision_trees(parent_inv_ids):
3834
current_revision_id = parent_tree.get_revision_id()
3835
parents_parents_keys = parent_invs_keys_for_stacking[
3836
(current_revision_id,)]
3837
parents_parents = [key[-1] for key in parents_parents_keys]
3838
basis_id = _mod_revision.NULL_REVISION
3839
basis_tree = self.source.revision_tree(basis_id)
3840
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3841
self.target.add_inventory_by_delta(
3842
basis_id, delta, current_revision_id, parents_parents)
3843
cache[current_revision_id] = parent_tree
3845
def _fetch_batch(self, revision_ids, basis_id, cache):
3846
"""Fetch across a few revisions.
3848
:param revision_ids: The revisions to copy
3849
:param basis_id: The revision_id of a tree that must be in cache, used
3850
as a basis for delta when no other base is available
3851
:param cache: A cache of RevisionTrees that we can use.
3852
:return: The revision_id of the last converted tree. The RevisionTree
3853
for it will be in cache
3855
# Walk though all revisions; get inventory deltas, copy referenced
3856
# texts that delta references, insert the delta, revision and
3858
root_keys_to_create = set()
3861
pending_revisions = []
3862
parent_map = self.source.get_parent_map(revision_ids)
3863
self._fetch_parent_invs_for_stacking(parent_map, cache)
3864
self.source._safe_to_return_from_cache = True
3865
for tree in self.source.revision_trees(revision_ids):
3866
# Find a inventory delta for this revision.
3867
# Find text entries that need to be copied, too.
3868
current_revision_id = tree.get_revision_id()
3869
parent_ids = parent_map.get(current_revision_id, ())
3870
parent_trees = self._get_trees(parent_ids, cache)
3871
possible_trees = list(parent_trees)
3872
if len(possible_trees) == 0:
3873
# There either aren't any parents, or the parents are ghosts,
3874
# so just use the last converted tree.
3875
possible_trees.append((basis_id, cache[basis_id]))
3876
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3878
revision = self.source.get_revision(current_revision_id)
3879
pending_deltas.append((basis_id, delta,
3880
current_revision_id, revision.parent_ids))
3881
if self._converting_to_rich_root:
3882
self._revision_id_to_root_id[current_revision_id] = \
3884
# Determine which texts are in present in this revision but not in
3885
# any of the available parents.
3886
texts_possibly_new_in_tree = set()
3887
for old_path, new_path, file_id, entry in delta:
3888
if new_path is None:
3889
# This file_id isn't present in the new rev
3893
if not self.target.supports_rich_root():
3894
# The target doesn't support rich root, so we don't
3897
if self._converting_to_rich_root:
3898
# This can't be copied normally, we have to insert
3900
root_keys_to_create.add((file_id, entry.revision))
3903
texts_possibly_new_in_tree.add((file_id, entry.revision))
3904
for basis_id, basis_tree in possible_trees:
3905
basis_inv = basis_tree.inventory
3906
for file_key in list(texts_possibly_new_in_tree):
3907
file_id, file_revision = file_key
3909
entry = basis_inv[file_id]
3910
except errors.NoSuchId:
3912
if entry.revision == file_revision:
3913
texts_possibly_new_in_tree.remove(file_key)
3914
text_keys.update(texts_possibly_new_in_tree)
3915
pending_revisions.append(revision)
3916
cache[current_revision_id] = tree
3917
basis_id = current_revision_id
3918
self.source._safe_to_return_from_cache = False
3920
from_texts = self.source.texts
3921
to_texts = self.target.texts
3922
if root_keys_to_create:
3923
root_stream = _mod_fetch._new_root_data_stream(
3924
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3926
to_texts.insert_record_stream(root_stream)
3927
to_texts.insert_record_stream(from_texts.get_record_stream(
3928
text_keys, self.target._format._fetch_order,
3929
not self.target._format._fetch_uses_deltas))
3930
# insert inventory deltas
3931
for delta in pending_deltas:
3932
self.target.add_inventory_by_delta(*delta)
3933
if self.target._fallback_repositories:
3934
# Make sure this stacked repository has all the parent inventories
3935
# for the new revisions that we are about to insert. We do this
3936
# before adding the revisions so that no revision is added until
3937
# all the inventories it may depend on are added.
3938
# Note that this is overzealous, as we may have fetched these in an
3941
revision_ids = set()
3942
for revision in pending_revisions:
3943
revision_ids.add(revision.revision_id)
3944
parent_ids.update(revision.parent_ids)
3945
parent_ids.difference_update(revision_ids)
3946
parent_ids.discard(_mod_revision.NULL_REVISION)
3947
parent_map = self.source.get_parent_map(parent_ids)
3948
# we iterate over parent_map and not parent_ids because we don't
3949
# want to try copying any revision which is a ghost
3950
for parent_tree in self.source.revision_trees(parent_map):
3951
current_revision_id = parent_tree.get_revision_id()
3952
parents_parents = parent_map[current_revision_id]
3953
possible_trees = self._get_trees(parents_parents, cache)
3954
if len(possible_trees) == 0:
3955
# There either aren't any parents, or the parents are
3956
# ghosts, so just use the last converted tree.
3957
possible_trees.append((basis_id, cache[basis_id]))
3958
basis_id, delta = self._get_delta_for_revision(parent_tree,
3959
parents_parents, possible_trees)
3960
self.target.add_inventory_by_delta(
3961
basis_id, delta, current_revision_id, parents_parents)
3962
# insert signatures and revisions
3963
for revision in pending_revisions:
3965
signature = self.source.get_signature_text(
3966
revision.revision_id)
3967
self.target.add_signature_text(revision.revision_id,
3969
except errors.NoSuchRevision:
3971
self.target.add_revision(revision.revision_id, revision)
3974
def _fetch_all_revisions(self, revision_ids, pb):
3975
"""Fetch everything for the list of revisions.
3977
:param revision_ids: The list of revisions to fetch. Must be in
3979
:param pb: A ProgressTask
3982
basis_id, basis_tree = self._get_basis(revision_ids[0])
3984
cache = lru_cache.LRUCache(100)
3985
cache[basis_id] = basis_tree
3986
del basis_tree # We don't want to hang on to it here
3990
for offset in range(0, len(revision_ids), batch_size):
3991
self.target.start_write_group()
3993
pb.update('Transferring revisions', offset,
3995
batch = revision_ids[offset:offset+batch_size]
3996
basis_id = self._fetch_batch(batch, basis_id, cache)
3998
self.source._safe_to_return_from_cache = False
3999
self.target.abort_write_group()
4002
hint = self.target.commit_write_group()
4005
if hints and self.target._format.pack_compresses:
4006
self.target.pack(hint=hints)
4007
pb.update('Transferring revisions', len(revision_ids),
4011
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
4013
"""See InterRepository.fetch()."""
4014
if fetch_spec is not None:
4015
raise AssertionError("Not implemented yet...")
4016
ui.ui_factory.warn_experimental_format_fetch(self)
4017
if (not self.source.supports_rich_root()
4018
and self.target.supports_rich_root()):
4019
self._converting_to_rich_root = True
4020
self._revision_id_to_root_id = {}
4022
self._converting_to_rich_root = False
4023
# See <https://launchpad.net/bugs/456077> asking for a warning here
4024
if self.source._format.network_name() != self.target._format.network_name():
4025
ui.ui_factory.show_user_warning('cross_format_fetch',
4026
from_format=self.source._format,
4027
to_format=self.target._format)
4028
revision_ids = self.target.search_missing_revision_ids(self.source,
4029
revision_id, find_ghosts=find_ghosts).get_keys()
4030
if not revision_ids:
4032
revision_ids = tsort.topo_sort(
4033
self.source.get_graph().get_parent_map(revision_ids))
4034
if not revision_ids:
4036
# Walk though all revisions; get inventory deltas, copy referenced
4037
# texts that delta references, insert the delta, revision and
4040
my_pb = ui.ui_factory.nested_progress_bar()
4043
symbol_versioning.warn(
4044
symbol_versioning.deprecated_in((1, 14, 0))
4045
% "pb parameter to fetch()")
4048
self._fetch_all_revisions(revision_ids, pb)
4050
if my_pb is not None:
4052
return len(revision_ids), 0
4054
def _get_basis(self, first_revision_id):
4055
"""Get a revision and tree which exists in the target.
4057
This assumes that first_revision_id is selected for transmission
4058
because all other ancestors are already present. If we can't find an
4059
ancestor we fall back to NULL_REVISION since we know that is safe.
4061
:return: (basis_id, basis_tree)
4063
first_rev = self.source.get_revision(first_revision_id)
4065
basis_id = first_rev.parent_ids[0]
4066
# only valid as a basis if the target has it
4067
self.target.get_revision(basis_id)
4068
# Try to get a basis tree - if it's a ghost it will hit the
4069
# NoSuchRevision case.
4070
basis_tree = self.source.revision_tree(basis_id)
4071
except (IndexError, errors.NoSuchRevision):
4072
basis_id = _mod_revision.NULL_REVISION
4073
basis_tree = self.source.revision_tree(basis_id)
4074
return basis_id, basis_tree
4077
InterRepository.register_optimiser(InterDifferingSerializer)
1622
def fetch(self, revision_id=None, pb=None):
1623
"""See InterRepository.fetch()."""
1624
from bzrlib.fetch import Model1toKnit2Fetcher
1625
# TODO: jam 20070210 This should be an assert, not a translate
1626
revision_id = osutils.safe_revision_id(revision_id)
1627
f = Model1toKnit2Fetcher(to_repository=self.target,
1628
from_repository=self.source,
1629
last_revision=revision_id,
1631
return f.count_copied, f.failed_revisions
1634
def copy_content(self, revision_id=None):
1635
"""Make a complete copy of the content in self into destination.
1637
This is a destructive operation! Do not use it on existing
1640
:param revision_id: Only copy the content needed to construct
1641
revision_id and its parents.
1644
self.target.set_make_working_trees(self.source.make_working_trees())
1645
except NotImplementedError:
1647
# TODO: jam 20070210 Internal, assert, don't translate
1648
revision_id = osutils.safe_revision_id(revision_id)
1649
# but don't bother fetching if we have the needed data now.
1650
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1651
self.target.has_revision(revision_id)):
1653
self.target.fetch(self.source, revision_id=revision_id)
1656
class InterKnit1and2(InterKnitRepo):
1659
def _get_repo_format_to_test(self):
1663
def is_compatible(source, target):
1664
"""Be compatible with Knit1 source and Knit3 target"""
1665
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit3
1667
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1, \
1668
RepositoryFormatKnit3
1669
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1670
isinstance(target._format, (RepositoryFormatKnit3)))
1671
except AttributeError:
1675
def fetch(self, revision_id=None, pb=None):
1676
"""See InterRepository.fetch()."""
1677
from bzrlib.fetch import Knit1to2Fetcher
1678
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1679
self.source, self.source._format, self.target,
1680
self.target._format)
1681
# TODO: jam 20070210 This should be an assert, not a translate
1682
revision_id = osutils.safe_revision_id(revision_id)
1683
f = Knit1to2Fetcher(to_repository=self.target,
1684
from_repository=self.source,
1685
last_revision=revision_id,
1687
return f.count_copied, f.failed_revisions
4078
1690
InterRepository.register_optimiser(InterSameDataRepository)
4079
1691
InterRepository.register_optimiser(InterWeaveRepo)
4080
1692
InterRepository.register_optimiser(InterKnitRepo)
1693
InterRepository.register_optimiser(InterModel1and2)
1694
InterRepository.register_optimiser(InterKnit1and2)
1697
class RepositoryTestProviderAdapter(object):
1698
"""A tool to generate a suite testing multiple repository formats at once.
1700
This is done by copying the test once for each transport and injecting
1701
the transport_server, transport_readonly_server, and bzrdir_format and
1702
repository_format classes into each copy. Each copy is also given a new id()
1703
to make it easy to identify.
1706
def __init__(self, transport_server, transport_readonly_server, formats):
1707
self._transport_server = transport_server
1708
self._transport_readonly_server = transport_readonly_server
1709
self._formats = formats
1711
def adapt(self, test):
1712
result = unittest.TestSuite()
1713
for repository_format, bzrdir_format in self._formats:
1714
from copy import deepcopy
1715
new_test = deepcopy(test)
1716
new_test.transport_server = self._transport_server
1717
new_test.transport_readonly_server = self._transport_readonly_server
1718
new_test.bzrdir_format = bzrdir_format
1719
new_test.repository_format = repository_format
1720
def make_new_test_id():
1721
new_id = "%s(%s)" % (new_test.id(), repository_format.__class__.__name__)
1722
return lambda: new_id
1723
new_test.id = make_new_test_id()
1724
result.addTest(new_test)
1728
class InterRepositoryTestProviderAdapter(object):
1729
"""A tool to generate a suite testing multiple inter repository formats.
1731
This is done by copying the test once for each interrepo provider and injecting
1732
the transport_server, transport_readonly_server, repository_format and
1733
repository_to_format classes into each copy.
1734
Each copy is also given a new id() to make it easy to identify.
1737
def __init__(self, transport_server, transport_readonly_server, formats):
1738
self._transport_server = transport_server
1739
self._transport_readonly_server = transport_readonly_server
1740
self._formats = formats
1742
def adapt(self, test):
1743
result = unittest.TestSuite()
1744
for interrepo_class, repository_format, repository_format_to in self._formats:
1745
from copy import deepcopy
1746
new_test = deepcopy(test)
1747
new_test.transport_server = self._transport_server
1748
new_test.transport_readonly_server = self._transport_readonly_server
1749
new_test.interrepo_class = interrepo_class
1750
new_test.repository_format = repository_format
1751
new_test.repository_format_to = repository_format_to
1752
def make_new_test_id():
1753
new_id = "%s(%s)" % (new_test.id(), interrepo_class.__name__)
1754
return lambda: new_id
1755
new_test.id = make_new_test_id()
1756
result.addTest(new_test)
1760
def default_test_list():
1761
"""Generate the default list of interrepo permutations to test."""
1762
from bzrlib.repofmt import knitrepo, weaverepo
1764
# test the default InterRepository between format 6 and the current
1766
# XXX: robertc 20060220 reinstate this when there are two supported
1767
# formats which do not have an optimal code path between them.
1768
#result.append((InterRepository,
1769
# RepositoryFormat6(),
1770
# RepositoryFormatKnit1()))
1771
for optimiser_class in InterRepository._optimisers:
1772
format_to_test = optimiser_class._get_repo_format_to_test()
1773
if format_to_test is not None:
1774
result.append((optimiser_class,
1775
format_to_test, format_to_test))
1776
# if there are specific combinations we want to use, we can add them
1778
result.append((InterModel1and2,
1779
weaverepo.RepositoryFormat5(),
1780
knitrepo.RepositoryFormatKnit3()))
1781
result.append((InterKnit1and2,
1782
knitrepo.RepositoryFormatKnit1(),
1783
knitrepo.RepositoryFormatKnit3()))
4083
1787
class CopyConverter(object):
4084
1788
"""A repository conversion tool which just performs a copy of the content.
4086
1790
This is slow but quite reliable.
4092
1796
:param target_format: The format the resulting repository should be.
4094
1798
self.target_format = target_format
4096
1800
def convert(self, repo, pb):
4097
1801
"""Perform the conversion of to_convert, giving feedback via pb.
4099
1803
:param to_convert: The disk object to convert.
4100
1804
:param pb: a progress bar to use for progress information.
4102
pb = ui.ui_factory.nested_progress_bar()
4105
1809
# this is only useful with metadir layouts - separated repo content.
4106
1810
# trigger an assertion if not such
4107
1811
repo._format.get_format_string()
4108
1812
self.repo_dir = repo.bzrdir
4109
pb.update('Moving repository to repository.backup')
1813
self.step('Moving repository to repository.backup')
4110
1814
self.repo_dir.transport.move('repository', 'repository.backup')
4111
1815
backup_transport = self.repo_dir.transport.clone('repository.backup')
4112
1816
repo._format.check_conversion_target(self.target_format)
4113
1817
self.source_repo = repo._format.open(self.repo_dir,
4115
1819
_override_transport=backup_transport)
4116
pb.update('Creating new repository')
1820
self.step('Creating new repository')
4117
1821
converted = self.target_format.initialize(self.repo_dir,
4118
1822
self.source_repo.is_shared())
4119
1823
converted.lock_write()
4121
pb.update('Copying content')
1825
self.step('Copying content into repository.')
4122
1826
self.source_repo.copy_content_into(converted)
4124
1828
converted.unlock()
4125
pb.update('Deleting old repository content')
1829
self.step('Deleting old repository content.')
4126
1830
self.repo_dir.transport.delete_tree('repository.backup')
4127
ui.ui_factory.note('repository converted')
1831
self.pb.note('repository converted')
1833
def step(self, message):
1834
"""Update the pb by a step."""
1836
self.pb.update(message, self.count, self.total)
1839
class CommitBuilder(object):
1840
"""Provides an interface to build up a commit.
1842
This allows describing a tree to be committed without needing to
1843
know the internals of the format of the repository.
1846
record_root_entry = False
1847
def __init__(self, repository, parents, config, timestamp=None,
1848
timezone=None, committer=None, revprops=None,
1850
"""Initiate a CommitBuilder.
1852
:param repository: Repository to commit to.
1853
:param parents: Revision ids of the parents of the new revision.
1854
:param config: Configuration to use.
1855
:param timestamp: Optional timestamp recorded for commit.
1856
:param timezone: Optional timezone for timestamp.
1857
:param committer: Optional committer to set for commit.
1858
:param revprops: Optional dictionary of revision properties.
1859
:param revision_id: Optional revision id.
1861
self._config = config
1863
if committer is None:
1864
self._committer = self._config.username()
1866
assert isinstance(committer, basestring), type(committer)
1867
self._committer = committer
1869
self.new_inventory = Inventory(None)
1870
self._new_revision_id = osutils.safe_revision_id(revision_id)
1871
self.parents = parents
1872
self.repository = repository
1875
if revprops is not None:
1876
self._revprops.update(revprops)
1878
if timestamp is None:
1879
timestamp = time.time()
1880
# Restrict resolution to 1ms
1881
self._timestamp = round(timestamp, 3)
1883
if timezone is None:
1884
self._timezone = osutils.local_time_offset()
1886
self._timezone = int(timezone)
1888
self._generate_revision_if_needed()
1890
def commit(self, message):
1891
"""Make the actual commit.
1893
:return: The revision id of the recorded revision.
1895
rev = _mod_revision.Revision(
1896
timestamp=self._timestamp,
1897
timezone=self._timezone,
1898
committer=self._committer,
1900
inventory_sha1=self.inv_sha1,
1901
revision_id=self._new_revision_id,
1902
properties=self._revprops)
1903
rev.parent_ids = self.parents
1904
self.repository.add_revision(self._new_revision_id, rev,
1905
self.new_inventory, self._config)
1906
return self._new_revision_id
1908
def revision_tree(self):
1909
"""Return the tree that was just committed.
1911
After calling commit() this can be called to get a RevisionTree
1912
representing the newly committed tree. This is preferred to
1913
calling Repository.revision_tree() because that may require
1914
deserializing the inventory, while we already have a copy in
1917
return RevisionTree(self.repository, self.new_inventory,
1918
self._new_revision_id)
1920
def finish_inventory(self):
1921
"""Tell the builder that the inventory is finished."""
1922
if self.new_inventory.root is None:
1923
symbol_versioning.warn('Root entry should be supplied to'
1924
' record_entry_contents, as of bzr 0.10.',
1925
DeprecationWarning, stacklevel=2)
1926
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
1927
self.new_inventory.revision_id = self._new_revision_id
1928
self.inv_sha1 = self.repository.add_inventory(
1929
self._new_revision_id,
1934
def _gen_revision_id(self):
1935
"""Return new revision-id."""
1936
return generate_ids.gen_revision_id(self._config.username(),
1939
def _generate_revision_if_needed(self):
1940
"""Create a revision id if None was supplied.
1942
If the repository can not support user-specified revision ids
1943
they should override this function and raise CannotSetRevisionId
1944
if _new_revision_id is not None.
1946
:raises: CannotSetRevisionId
1948
if self._new_revision_id is None:
1949
self._new_revision_id = self._gen_revision_id()
1951
def record_entry_contents(self, ie, parent_invs, path, tree):
1952
"""Record the content of ie from tree into the commit if needed.
1954
Side effect: sets ie.revision when unchanged
1956
:param ie: An inventory entry present in the commit.
1957
:param parent_invs: The inventories of the parent revisions of the
1959
:param path: The path the entry is at in the tree.
1960
:param tree: The tree which contains this entry and should be used to
1963
if self.new_inventory.root is None and ie.parent_id is not None:
1964
symbol_versioning.warn('Root entry should be supplied to'
1965
' record_entry_contents, as of bzr 0.10.',
1966
DeprecationWarning, stacklevel=2)
1967
self.record_entry_contents(tree.inventory.root.copy(), parent_invs,
1969
self.new_inventory.add(ie)
1971
# ie.revision is always None if the InventoryEntry is considered
1972
# for committing. ie.snapshot will record the correct revision
1973
# which may be the sole parent if it is untouched.
1974
if ie.revision is not None:
1977
# In this revision format, root entries have no knit or weave
1978
if ie is self.new_inventory.root:
1979
# When serializing out to disk and back in
1980
# root.revision is always _new_revision_id
1981
ie.revision = self._new_revision_id
1983
previous_entries = ie.find_previous_heads(
1985
self.repository.weave_store,
1986
self.repository.get_transaction())
1987
# we are creating a new revision for ie in the history store
1989
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
1991
def modified_directory(self, file_id, file_parents):
1992
"""Record the presence of a symbolic link.
1994
:param file_id: The file_id of the link to record.
1995
:param file_parents: The per-file parent revision ids.
1997
self._add_text_to_weave(file_id, [], file_parents.keys())
1999
def modified_reference(self, file_id, file_parents):
2000
"""Record the modification of a reference.
2002
:param file_id: The file_id of the link to record.
2003
:param file_parents: The per-file parent revision ids.
2005
self._add_text_to_weave(file_id, [], file_parents.keys())
2007
def modified_file_text(self, file_id, file_parents,
2008
get_content_byte_lines, text_sha1=None,
2010
"""Record the text of file file_id
2012
:param file_id: The file_id of the file to record the text of.
2013
:param file_parents: The per-file parent revision ids.
2014
:param get_content_byte_lines: A callable which will return the byte
2016
:param text_sha1: Optional SHA1 of the file contents.
2017
:param text_size: Optional size of the file contents.
2019
# mutter('storing text of file {%s} in revision {%s} into %r',
2020
# file_id, self._new_revision_id, self.repository.weave_store)
2021
# special case to avoid diffing on renames or
2023
if (len(file_parents) == 1
2024
and text_sha1 == file_parents.values()[0].text_sha1
2025
and text_size == file_parents.values()[0].text_size):
2026
previous_ie = file_parents.values()[0]
2027
versionedfile = self.repository.weave_store.get_weave(file_id,
2028
self.repository.get_transaction())
2029
versionedfile.clone_text(self._new_revision_id,
2030
previous_ie.revision, file_parents.keys())
2031
return text_sha1, text_size
2033
new_lines = get_content_byte_lines()
2034
# TODO: Rather than invoking sha_strings here, _add_text_to_weave
2035
# should return the SHA1 and size
2036
self._add_text_to_weave(file_id, new_lines, file_parents.keys())
2037
return osutils.sha_strings(new_lines), \
2038
sum(map(len, new_lines))
2040
def modified_link(self, file_id, file_parents, link_target):
2041
"""Record the presence of a symbolic link.
2043
:param file_id: The file_id of the link to record.
2044
:param file_parents: The per-file parent revision ids.
2045
:param link_target: Target location of this link.
2047
self._add_text_to_weave(file_id, [], file_parents.keys())
2049
def _add_text_to_weave(self, file_id, new_lines, parents):
2050
versionedfile = self.repository.weave_store.get_weave_or_empty(
2051
file_id, self.repository.get_transaction())
2052
versionedfile.add_lines(self._new_revision_id, parents, new_lines)
2053
versionedfile.clear_cache()
2056
class _CommitBuilder(CommitBuilder):
2057
"""Temporary class so old CommitBuilders are detected properly
2059
Note: CommitBuilder works whether or not root entry is recorded.
2062
record_root_entry = True
2065
class RootCommitBuilder(CommitBuilder):
2066
"""This commitbuilder actually records the root id"""
2068
record_root_entry = True
2070
def record_entry_contents(self, ie, parent_invs, path, tree):
2071
"""Record the content of ie from tree into the commit if needed.
2073
Side effect: sets ie.revision when unchanged
2075
:param ie: An inventory entry present in the commit.
2076
:param parent_invs: The inventories of the parent revisions of the
2078
:param path: The path the entry is at in the tree.
2079
:param tree: The tree which contains this entry and should be used to
2082
assert self.new_inventory.root is not None or ie.parent_id is None
2083
self.new_inventory.add(ie)
2085
# ie.revision is always None if the InventoryEntry is considered
2086
# for committing. ie.snapshot will record the correct revision
2087
# which may be the sole parent if it is untouched.
2088
if ie.revision is not None:
2091
previous_entries = ie.find_previous_heads(
2093
self.repository.weave_store,
2094
self.repository.get_transaction())
2095
# we are creating a new revision for ie in the history store
2097
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
4131
2100
_unescape_map = {
4156
2125
if _unescape_re is None:
4157
2126
_unescape_re = re.compile('\&([^;]*);')
4158
2127
return _unescape_re.sub(_unescaper, data)
4161
class _VersionedFileChecker(object):
4163
def __init__(self, repository, text_key_references=None, ancestors=None):
4164
self.repository = repository
4165
self.text_index = self.repository._generate_text_key_index(
4166
text_key_references=text_key_references, ancestors=ancestors)
4168
def calculate_file_version_parents(self, text_key):
4169
"""Calculate the correct parents for a file version according to
4172
parent_keys = self.text_index[text_key]
4173
if parent_keys == [_mod_revision.NULL_REVISION]:
4175
return tuple(parent_keys)
4177
def check_file_version_parents(self, texts, progress_bar=None):
4178
"""Check the parents stored in a versioned file are correct.
4180
It also detects file versions that are not referenced by their
4181
corresponding revision's inventory.
4183
:returns: A tuple of (wrong_parents, dangling_file_versions).
4184
wrong_parents is a dict mapping {revision_id: (stored_parents,
4185
correct_parents)} for each revision_id where the stored parents
4186
are not correct. dangling_file_versions is a set of (file_id,
4187
revision_id) tuples for versions that are present in this versioned
4188
file, but not used by the corresponding inventory.
4190
local_progress = None
4191
if progress_bar is None:
4192
local_progress = ui.ui_factory.nested_progress_bar()
4193
progress_bar = local_progress
4195
return self._check_file_version_parents(texts, progress_bar)
4198
local_progress.finished()
4200
def _check_file_version_parents(self, texts, progress_bar):
4201
"""See check_file_version_parents."""
4203
self.file_ids = set([file_id for file_id, _ in
4204
self.text_index.iterkeys()])
4205
# text keys is now grouped by file_id
4206
n_versions = len(self.text_index)
4207
progress_bar.update('loading text store', 0, n_versions)
4208
parent_map = self.repository.texts.get_parent_map(self.text_index)
4209
# On unlistable transports this could well be empty/error...
4210
text_keys = self.repository.texts.keys()
4211
unused_keys = frozenset(text_keys) - set(self.text_index)
4212
for num, key in enumerate(self.text_index.iterkeys()):
4213
progress_bar.update('checking text graph', num, n_versions)
4214
correct_parents = self.calculate_file_version_parents(key)
4216
knit_parents = parent_map[key]
4217
except errors.RevisionNotPresent:
4220
if correct_parents != knit_parents:
4221
wrong_parents[key] = (knit_parents, correct_parents)
4222
return wrong_parents, unused_keys
4225
def _old_get_graph(repository, revision_id):
4226
"""DO NOT USE. That is all. I'm serious."""
4227
graph = repository.get_graph()
4228
revision_graph = dict(((key, value) for key, value in
4229
graph.iter_ancestry([revision_id]) if value is not None))
4230
return _strip_NULL_ghosts(revision_graph)
4233
def _strip_NULL_ghosts(revision_graph):
4234
"""Also don't use this. more compatibility code for unmigrated clients."""
4235
# Filter ghosts, and null:
4236
if _mod_revision.NULL_REVISION in revision_graph:
4237
del revision_graph[_mod_revision.NULL_REVISION]
4238
for key, parents in revision_graph.items():
4239
revision_graph[key] = tuple(parent for parent in parents if parent
4241
return revision_graph
4244
class StreamSink(object):
4245
"""An object that can insert a stream into a repository.
4247
This interface handles the complexity of reserialising inventories and
4248
revisions from different formats, and allows unidirectional insertion into
4249
stacked repositories without looking for the missing basis parents
4253
def __init__(self, target_repo):
4254
self.target_repo = target_repo
4256
def insert_stream(self, stream, src_format, resume_tokens):
4257
"""Insert a stream's content into the target repository.
4259
:param src_format: a bzr repository format.
4261
:return: a list of resume tokens and an iterable of keys additional
4262
items required before the insertion can be completed.
4264
self.target_repo.lock_write()
4267
self.target_repo.resume_write_group(resume_tokens)
4270
self.target_repo.start_write_group()
4273
# locked_insert_stream performs a commit|suspend.
4274
return self._locked_insert_stream(stream, src_format,
4277
self.target_repo.abort_write_group(suppress_errors=True)
4280
self.target_repo.unlock()
4282
def _locked_insert_stream(self, stream, src_format, is_resume):
4283
to_serializer = self.target_repo._format._serializer
4284
src_serializer = src_format._serializer
4286
if to_serializer == src_serializer:
4287
# If serializers match and the target is a pack repository, set the
4288
# write cache size on the new pack. This avoids poor performance
4289
# on transports where append is unbuffered (such as
4290
# RemoteTransport). This is safe to do because nothing should read
4291
# back from the target repository while a stream with matching
4292
# serialization is being inserted.
4293
# The exception is that a delta record from the source that should
4294
# be a fulltext may need to be expanded by the target (see
4295
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4296
# explicitly flush any buffered writes first in that rare case.
4298
new_pack = self.target_repo._pack_collection._new_pack
4299
except AttributeError:
4300
# Not a pack repository
4303
new_pack.set_write_cache_size(1024*1024)
4304
for substream_type, substream in stream:
4305
if 'stream' in debug.debug_flags:
4306
mutter('inserting substream: %s', substream_type)
4307
if substream_type == 'texts':
4308
self.target_repo.texts.insert_record_stream(substream)
4309
elif substream_type == 'inventories':
4310
if src_serializer == to_serializer:
4311
self.target_repo.inventories.insert_record_stream(
4314
self._extract_and_insert_inventories(
4315
substream, src_serializer)
4316
elif substream_type == 'inventory-deltas':
4317
self._extract_and_insert_inventory_deltas(
4318
substream, src_serializer)
4319
elif substream_type == 'chk_bytes':
4320
# XXX: This doesn't support conversions, as it assumes the
4321
# conversion was done in the fetch code.
4322
self.target_repo.chk_bytes.insert_record_stream(substream)
4323
elif substream_type == 'revisions':
4324
# This may fallback to extract-and-insert more often than
4325
# required if the serializers are different only in terms of
4327
if src_serializer == to_serializer:
4328
self.target_repo.revisions.insert_record_stream(substream)
4330
self._extract_and_insert_revisions(substream,
4332
elif substream_type == 'signatures':
4333
self.target_repo.signatures.insert_record_stream(substream)
4335
raise AssertionError('kaboom! %s' % (substream_type,))
4336
# Done inserting data, and the missing_keys calculations will try to
4337
# read back from the inserted data, so flush the writes to the new pack
4338
# (if this is pack format).
4339
if new_pack is not None:
4340
new_pack._write_data('', flush=True)
4341
# Find all the new revisions (including ones from resume_tokens)
4342
missing_keys = self.target_repo.get_missing_parent_inventories(
4343
check_for_missing_texts=is_resume)
4345
for prefix, versioned_file in (
4346
('texts', self.target_repo.texts),
4347
('inventories', self.target_repo.inventories),
4348
('revisions', self.target_repo.revisions),
4349
('signatures', self.target_repo.signatures),
4350
('chk_bytes', self.target_repo.chk_bytes),
4352
if versioned_file is None:
4354
# TODO: key is often going to be a StaticTuple object
4355
# I don't believe we can define a method by which
4356
# (prefix,) + StaticTuple will work, though we could
4357
# define a StaticTuple.sq_concat that would allow you to
4358
# pass in either a tuple or a StaticTuple as the second
4359
# object, so instead we could have:
4360
# StaticTuple(prefix) + key here...
4361
missing_keys.update((prefix,) + key for key in
4362
versioned_file.get_missing_compression_parent_keys())
4363
except NotImplementedError:
4364
# cannot even attempt suspending, and missing would have failed
4365
# during stream insertion.
4366
missing_keys = set()
4369
# suspend the write group and tell the caller what we is
4370
# missing. We know we can suspend or else we would not have
4371
# entered this code path. (All repositories that can handle
4372
# missing keys can handle suspending a write group).
4373
write_group_tokens = self.target_repo.suspend_write_group()
4374
return write_group_tokens, missing_keys
4375
hint = self.target_repo.commit_write_group()
4376
if (to_serializer != src_serializer and
4377
self.target_repo._format.pack_compresses):
4378
self.target_repo.pack(hint=hint)
4381
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4382
target_rich_root = self.target_repo._format.rich_root_data
4383
target_tree_refs = self.target_repo._format.supports_tree_reference
4384
for record in substream:
4385
# Insert the delta directly
4386
inventory_delta_bytes = record.get_bytes_as('fulltext')
4387
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4389
parse_result = deserialiser.parse_text_bytes(
4390
inventory_delta_bytes)
4391
except inventory_delta.IncompatibleInventoryDelta, err:
4392
trace.mutter("Incompatible delta: %s", err.msg)
4393
raise errors.IncompatibleRevision(self.target_repo._format)
4394
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4395
revision_id = new_id
4396
parents = [key[0] for key in record.parents]
4397
self.target_repo.add_inventory_by_delta(
4398
basis_id, inv_delta, revision_id, parents)
4400
def _extract_and_insert_inventories(self, substream, serializer,
4402
"""Generate a new inventory versionedfile in target, converting data.
4404
The inventory is retrieved from the source, (deserializing it), and
4405
stored in the target (reserializing it in a different format).
4407
target_rich_root = self.target_repo._format.rich_root_data
4408
target_tree_refs = self.target_repo._format.supports_tree_reference
4409
for record in substream:
4410
# It's not a delta, so it must be a fulltext in the source
4411
# serializer's format.
4412
bytes = record.get_bytes_as('fulltext')
4413
revision_id = record.key[0]
4414
inv = serializer.read_inventory_from_string(bytes, revision_id)
4415
parents = [key[0] for key in record.parents]
4416
self.target_repo.add_inventory(revision_id, inv, parents)
4417
# No need to keep holding this full inv in memory when the rest of
4418
# the substream is likely to be all deltas.
4421
def _extract_and_insert_revisions(self, substream, serializer):
4422
for record in substream:
4423
bytes = record.get_bytes_as('fulltext')
4424
revision_id = record.key[0]
4425
rev = serializer.read_revision_from_string(bytes)
4426
if rev.revision_id != revision_id:
4427
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4428
self.target_repo.add_revision(revision_id, rev)
4431
if self.target_repo._format._fetch_reconcile:
4432
self.target_repo.reconcile()
4435
class StreamSource(object):
4436
"""A source of a stream for fetching between repositories."""
4438
def __init__(self, from_repository, to_format):
4439
"""Create a StreamSource streaming from from_repository."""
4440
self.from_repository = from_repository
4441
self.to_format = to_format
4442
self._record_counter = RecordCounter()
4444
def delta_on_metadata(self):
4445
"""Return True if delta's are permitted on metadata streams.
4447
That is on revisions and signatures.
4449
src_serializer = self.from_repository._format._serializer
4450
target_serializer = self.to_format._serializer
4451
return (self.to_format._fetch_uses_deltas and
4452
src_serializer == target_serializer)
4454
def _fetch_revision_texts(self, revs):
4455
# fetch signatures first and then the revision texts
4456
# may need to be a InterRevisionStore call here.
4457
from_sf = self.from_repository.signatures
4458
# A missing signature is just skipped.
4459
keys = [(rev_id,) for rev_id in revs]
4460
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4462
self.to_format._fetch_order,
4463
not self.to_format._fetch_uses_deltas))
4464
# If a revision has a delta, this is actually expanded inside the
4465
# insert_record_stream code now, which is an alternate fix for
4467
from_rf = self.from_repository.revisions
4468
revisions = from_rf.get_record_stream(
4470
self.to_format._fetch_order,
4471
not self.delta_on_metadata())
4472
return [('signatures', signatures), ('revisions', revisions)]
4474
def _generate_root_texts(self, revs):
4475
"""This will be called by get_stream between fetching weave texts and
4476
fetching the inventory weave.
4478
if self._rich_root_upgrade():
4479
return _mod_fetch.Inter1and2Helper(
4480
self.from_repository).generate_root_texts(revs)
4484
def get_stream(self, search):
4486
revs = search.get_keys()
4487
graph = self.from_repository.get_graph()
4488
revs = tsort.topo_sort(graph.get_parent_map(revs))
4489
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4491
for knit_kind, file_id, revisions in data_to_fetch:
4492
if knit_kind != phase:
4494
# Make a new progress bar for this phase
4495
if knit_kind == "file":
4496
# Accumulate file texts
4497
text_keys.extend([(file_id, revision) for revision in
4499
elif knit_kind == "inventory":
4500
# Now copy the file texts.
4501
from_texts = self.from_repository.texts
4502
yield ('texts', from_texts.get_record_stream(
4503
text_keys, self.to_format._fetch_order,
4504
not self.to_format._fetch_uses_deltas))
4505
# Cause an error if a text occurs after we have done the
4508
# Before we process the inventory we generate the root
4509
# texts (if necessary) so that the inventories references
4511
for _ in self._generate_root_texts(revs):
4513
# we fetch only the referenced inventories because we do not
4514
# know for unselected inventories whether all their required
4515
# texts are present in the other repository - it could be
4517
for info in self._get_inventory_stream(revs):
4519
elif knit_kind == "signatures":
4520
# Nothing to do here; this will be taken care of when
4521
# _fetch_revision_texts happens.
4523
elif knit_kind == "revisions":
4524
for record in self._fetch_revision_texts(revs):
4527
raise AssertionError("Unknown knit kind %r" % knit_kind)
4529
def get_stream_for_missing_keys(self, missing_keys):
4530
# missing keys can only occur when we are byte copying and not
4531
# translating (because translation means we don't send
4532
# unreconstructable deltas ever).
4534
keys['texts'] = set()
4535
keys['revisions'] = set()
4536
keys['inventories'] = set()
4537
keys['chk_bytes'] = set()
4538
keys['signatures'] = set()
4539
for key in missing_keys:
4540
keys[key[0]].add(key[1:])
4541
if len(keys['revisions']):
4542
# If we allowed copying revisions at this point, we could end up
4543
# copying a revision without copying its required texts: a
4544
# violation of the requirements for repository integrity.
4545
raise AssertionError(
4546
'cannot copy revisions to fill in missing deltas %s' % (
4547
keys['revisions'],))
4548
for substream_kind, keys in keys.iteritems():
4549
vf = getattr(self.from_repository, substream_kind)
4550
if vf is None and keys:
4551
raise AssertionError(
4552
"cannot fill in keys for a versioned file we don't"
4553
" have: %s needs %s" % (substream_kind, keys))
4555
# No need to stream something we don't have
4557
if substream_kind == 'inventories':
4558
# Some missing keys are genuinely ghosts, filter those out.
4559
present = self.from_repository.inventories.get_parent_map(keys)
4560
revs = [key[0] for key in present]
4561
# Get the inventory stream more-or-less as we do for the
4562
# original stream; there's no reason to assume that records
4563
# direct from the source will be suitable for the sink. (Think
4564
# e.g. 2a -> 1.9-rich-root).
4565
for info in self._get_inventory_stream(revs, missing=True):
4569
# Ask for full texts always so that we don't need more round trips
4570
# after this stream.
4571
# Some of the missing keys are genuinely ghosts, so filter absent
4572
# records. The Sink is responsible for doing another check to
4573
# ensure that ghosts don't introduce missing data for future
4575
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4576
self.to_format._fetch_order, True))
4577
yield substream_kind, stream
4579
def inventory_fetch_order(self):
4580
if self._rich_root_upgrade():
4581
return 'topological'
4583
return self.to_format._fetch_order
4585
def _rich_root_upgrade(self):
4586
return (not self.from_repository._format.rich_root_data and
4587
self.to_format.rich_root_data)
4589
def _get_inventory_stream(self, revision_ids, missing=False):
4590
from_format = self.from_repository._format
4591
if (from_format.supports_chks and self.to_format.supports_chks and
4592
from_format.network_name() == self.to_format.network_name()):
4593
raise AssertionError(
4594
"this case should be handled by GroupCHKStreamSource")
4595
elif 'forceinvdeltas' in debug.debug_flags:
4596
return self._get_convertable_inventory_stream(revision_ids,
4597
delta_versus_null=missing)
4598
elif from_format.network_name() == self.to_format.network_name():
4600
return self._get_simple_inventory_stream(revision_ids,
4602
elif (not from_format.supports_chks and not self.to_format.supports_chks
4603
and from_format._serializer == self.to_format._serializer):
4604
# Essentially the same format.
4605
return self._get_simple_inventory_stream(revision_ids,
4608
# Any time we switch serializations, we want to use an
4609
# inventory-delta based approach.
4610
return self._get_convertable_inventory_stream(revision_ids,
4611
delta_versus_null=missing)
4613
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4614
# NB: This currently reopens the inventory weave in source;
4615
# using a single stream interface instead would avoid this.
4616
from_weave = self.from_repository.inventories
4618
delta_closure = True
4620
delta_closure = not self.delta_on_metadata()
4621
yield ('inventories', from_weave.get_record_stream(
4622
[(rev_id,) for rev_id in revision_ids],
4623
self.inventory_fetch_order(), delta_closure))
4625
def _get_convertable_inventory_stream(self, revision_ids,
4626
delta_versus_null=False):
4627
# The two formats are sufficiently different that there is no fast
4628
# path, so we need to send just inventorydeltas, which any
4629
# sufficiently modern client can insert into any repository.
4630
# The StreamSink code expects to be able to
4631
# convert on the target, so we need to put bytes-on-the-wire that can
4632
# be converted. That means inventory deltas (if the remote is <1.19,
4633
# RemoteStreamSink will fallback to VFS to insert the deltas).
4634
yield ('inventory-deltas',
4635
self._stream_invs_as_deltas(revision_ids,
4636
delta_versus_null=delta_versus_null))
4638
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4639
"""Return a stream of inventory-deltas for the given rev ids.
4641
:param revision_ids: The list of inventories to transmit
4642
:param delta_versus_null: Don't try to find a minimal delta for this
4643
entry, instead compute the delta versus the NULL_REVISION. This
4644
effectively streams a complete inventory. Used for stuff like
4645
filling in missing parents, etc.
4647
from_repo = self.from_repository
4648
revision_keys = [(rev_id,) for rev_id in revision_ids]
4649
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4650
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4652
inventories = self.from_repository.iter_inventories(
4653
revision_ids, 'topological')
4654
format = from_repo._format
4655
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4656
inventory_cache = lru_cache.LRUCache(50)
4657
null_inventory = from_repo.revision_tree(
4658
_mod_revision.NULL_REVISION).inventory
4659
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4660
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4661
# repo back into a non-rich-root repo ought to be allowed)
4662
serializer = inventory_delta.InventoryDeltaSerializer(
4663
versioned_root=format.rich_root_data,
4664
tree_references=format.supports_tree_reference)
4665
for inv in inventories:
4666
key = (inv.revision_id,)
4667
parent_keys = parent_map.get(key, ())
4669
if not delta_versus_null and parent_keys:
4670
# The caller did not ask for complete inventories and we have
4671
# some parents that we can delta against. Make a delta against
4672
# each parent so that we can find the smallest.
4673
parent_ids = [parent_key[0] for parent_key in parent_keys]
4674
for parent_id in parent_ids:
4675
if parent_id not in invs_sent_so_far:
4676
# We don't know that the remote side has this basis, so
4679
if parent_id == _mod_revision.NULL_REVISION:
4680
parent_inv = null_inventory
4682
parent_inv = inventory_cache.get(parent_id, None)
4683
if parent_inv is None:
4684
parent_inv = from_repo.get_inventory(parent_id)
4685
candidate_delta = inv._make_delta(parent_inv)
4686
if (delta is None or
4687
len(delta) > len(candidate_delta)):
4688
delta = candidate_delta
4689
basis_id = parent_id
4691
# Either none of the parents ended up being suitable, or we
4692
# were asked to delta against NULL
4693
basis_id = _mod_revision.NULL_REVISION
4694
delta = inv._make_delta(null_inventory)
4695
invs_sent_so_far.add(inv.revision_id)
4696
inventory_cache[inv.revision_id] = inv
4697
delta_serialized = ''.join(
4698
serializer.delta_to_lines(basis_id, key[-1], delta))
4699
yield versionedfile.FulltextContentFactory(
4700
key, parent_keys, None, delta_serialized)
4703
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4704
stop_revision=None):
4705
"""Extend the partial history to include a given index
4707
If a stop_index is supplied, stop when that index has been reached.
4708
If a stop_revision is supplied, stop when that revision is
4709
encountered. Otherwise, stop when the beginning of history is
4712
:param stop_index: The index which should be present. When it is
4713
present, history extension will stop.
4714
:param stop_revision: The revision id which should be present. When
4715
it is encountered, history extension will stop.
4717
start_revision = partial_history_cache[-1]
4718
iterator = repo.iter_reverse_revision_history(start_revision)
4720
#skip the last revision in the list
4723
if (stop_index is not None and
4724
len(partial_history_cache) > stop_index):
4726
if partial_history_cache[-1] == stop_revision:
4728
revision_id = iterator.next()
4729
partial_history_cache.append(revision_id)
4730
except StopIteration:
added
removed
bzrlib/repository.py
