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,
50
from bzrlib.bundle import serializer
51
42
from bzrlib.revisiontree import RevisionTree
52
43
from bzrlib.store.versioned import VersionedFileStore
44
from bzrlib.store.text import TextStore
53
45
from bzrlib.testament import Testament
56
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
49
from bzrlib.decorators import needs_read_lock, needs_write_lock
57
50
from bzrlib.inter import InterObject
58
from bzrlib.inventory import (
64
from bzrlib.lock import _RelockDebugMixin
65
from bzrlib import registry
66
from bzrlib.trace import (
67
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
70
59
# Old formats display a warning, but only once
71
60
_deprecation_warning_done = False
74
class CommitBuilder(object):
75
"""Provides an interface to build up a commit.
77
This allows describing a tree to be committed without needing to
78
know the internals of the format of the repository.
81
# all clients should supply tree roots.
82
record_root_entry = True
83
# the default CommitBuilder does not manage trees whose root is versioned.
84
_versioned_root = False
86
def __init__(self, repository, parents, config, timestamp=None,
87
timezone=None, committer=None, revprops=None,
89
"""Initiate a CommitBuilder.
91
:param repository: Repository to commit to.
92
:param parents: Revision ids of the parents of the new revision.
93
:param config: Configuration to use.
94
:param timestamp: Optional timestamp recorded for commit.
95
:param timezone: Optional timezone for timestamp.
96
:param committer: Optional committer to set for commit.
97
:param revprops: Optional dictionary of revision properties.
98
:param revision_id: Optional revision id.
100
self._config = config
102
if committer is None:
103
self._committer = self._config.username()
105
self._committer = committer
107
self.new_inventory = Inventory(None)
108
self._new_revision_id = revision_id
109
self.parents = parents
110
self.repository = repository
113
if revprops is not None:
114
self._validate_revprops(revprops)
115
self._revprops.update(revprops)
117
if timestamp is None:
118
timestamp = time.time()
119
# Restrict resolution to 1ms
120
self._timestamp = round(timestamp, 3)
123
self._timezone = osutils.local_time_offset()
125
self._timezone = int(timezone)
127
self._generate_revision_if_needed()
128
self.__heads = graph.HeadsCache(repository.get_graph()).heads
129
self._basis_delta = []
130
# API compatibility, older code that used CommitBuilder did not call
131
# .record_delete(), which means the delta that is computed would not be
132
# valid. Callers that will call record_delete() should call
133
# .will_record_deletes() to indicate that.
134
self._recording_deletes = False
135
# memo'd check for no-op commits.
136
self._any_changes = False
138
def any_changes(self):
139
"""Return True if any entries were changed.
141
This includes merge-only changes. It is the core for the --unchanged
144
:return: True if any changes have occured.
146
return self._any_changes
148
def _validate_unicode_text(self, text, context):
149
"""Verify things like commit messages don't have bogus characters."""
151
raise ValueError('Invalid value for %s: %r' % (context, text))
153
def _validate_revprops(self, revprops):
154
for key, value in revprops.iteritems():
155
# We know that the XML serializers do not round trip '\r'
156
# correctly, so refuse to accept them
157
if not isinstance(value, basestring):
158
raise ValueError('revision property (%s) is not a valid'
159
' (unicode) string: %r' % (key, value))
160
self._validate_unicode_text(value,
161
'revision property (%s)' % (key,))
163
def commit(self, message):
164
"""Make the actual commit.
166
:return: The revision id of the recorded revision.
168
self._validate_unicode_text(message, 'commit message')
169
rev = _mod_revision.Revision(
170
timestamp=self._timestamp,
171
timezone=self._timezone,
172
committer=self._committer,
174
inventory_sha1=self.inv_sha1,
175
revision_id=self._new_revision_id,
176
properties=self._revprops)
177
rev.parent_ids = self.parents
178
self.repository.add_revision(self._new_revision_id, rev,
179
self.new_inventory, self._config)
180
self.repository.commit_write_group()
181
return self._new_revision_id
184
"""Abort the commit that is being built.
186
self.repository.abort_write_group()
188
def revision_tree(self):
189
"""Return the tree that was just committed.
191
After calling commit() this can be called to get a RevisionTree
192
representing the newly committed tree. This is preferred to
193
calling Repository.revision_tree() because that may require
194
deserializing the inventory, while we already have a copy in
197
if self.new_inventory is None:
198
self.new_inventory = self.repository.get_inventory(
199
self._new_revision_id)
200
return RevisionTree(self.repository, self.new_inventory,
201
self._new_revision_id)
203
def finish_inventory(self):
204
"""Tell the builder that the inventory is finished.
206
:return: The inventory id in the repository, which can be used with
207
repository.get_inventory.
209
if self.new_inventory is None:
210
# an inventory delta was accumulated without creating a new
212
basis_id = self.basis_delta_revision
213
# We ignore the 'inventory' returned by add_inventory_by_delta
214
# because self.new_inventory is used to hint to the rest of the
215
# system what code path was taken
216
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
217
basis_id, self._basis_delta, self._new_revision_id,
220
if self.new_inventory.root is None:
221
raise AssertionError('Root entry should be supplied to'
222
' record_entry_contents, as of bzr 0.10.')
223
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
224
self.new_inventory.revision_id = self._new_revision_id
225
self.inv_sha1 = self.repository.add_inventory(
226
self._new_revision_id,
230
return self._new_revision_id
232
def _gen_revision_id(self):
233
"""Return new revision-id."""
234
return generate_ids.gen_revision_id(self._config.username(),
237
def _generate_revision_if_needed(self):
238
"""Create a revision id if None was supplied.
240
If the repository can not support user-specified revision ids
241
they should override this function and raise CannotSetRevisionId
242
if _new_revision_id is not None.
244
:raises: CannotSetRevisionId
246
if self._new_revision_id is None:
247
self._new_revision_id = self._gen_revision_id()
248
self.random_revid = True
250
self.random_revid = False
252
def _heads(self, file_id, revision_ids):
253
"""Calculate the graph heads for revision_ids in the graph of file_id.
255
This can use either a per-file graph or a global revision graph as we
256
have an identity relationship between the two graphs.
258
return self.__heads(revision_ids)
260
def _check_root(self, ie, parent_invs, tree):
261
"""Helper for record_entry_contents.
263
:param ie: An entry being added.
264
:param parent_invs: The inventories of the parent revisions of the
266
:param tree: The tree that is being committed.
268
# In this revision format, root entries have no knit or weave When
269
# serializing out to disk and back in root.revision is always
271
ie.revision = self._new_revision_id
273
def _require_root_change(self, tree):
274
"""Enforce an appropriate root object change.
276
This is called once when record_iter_changes is called, if and only if
277
the root was not in the delta calculated by record_iter_changes.
279
:param tree: The tree which is being committed.
281
# NB: if there are no parents then this method is not called, so no
282
# need to guard on parents having length.
283
entry = entry_factory['directory'](tree.path2id(''), '',
285
entry.revision = self._new_revision_id
286
self._basis_delta.append(('', '', entry.file_id, entry))
288
def _get_delta(self, ie, basis_inv, path):
289
"""Get a delta against the basis inventory for ie."""
290
if ie.file_id not in basis_inv:
292
result = (None, path, ie.file_id, ie)
293
self._basis_delta.append(result)
295
elif ie != basis_inv[ie.file_id]:
297
# TODO: avoid tis id2path call.
298
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
299
self._basis_delta.append(result)
305
def get_basis_delta(self):
306
"""Return the complete inventory delta versus the basis inventory.
308
This has been built up with the calls to record_delete and
309
record_entry_contents. The client must have already called
310
will_record_deletes() to indicate that they will be generating a
313
:return: An inventory delta, suitable for use with apply_delta, or
314
Repository.add_inventory_by_delta, etc.
316
if not self._recording_deletes:
317
raise AssertionError("recording deletes not activated.")
318
return self._basis_delta
320
def record_delete(self, path, file_id):
321
"""Record that a delete occured against a basis tree.
323
This is an optional API - when used it adds items to the basis_delta
324
being accumulated by the commit builder. It cannot be called unless the
325
method will_record_deletes() has been called to inform the builder that
326
a delta is being supplied.
328
:param path: The path of the thing deleted.
329
:param file_id: The file id that was deleted.
331
if not self._recording_deletes:
332
raise AssertionError("recording deletes not activated.")
333
delta = (path, None, file_id, None)
334
self._basis_delta.append(delta)
335
self._any_changes = True
338
def will_record_deletes(self):
339
"""Tell the commit builder that deletes are being notified.
341
This enables the accumulation of an inventory delta; for the resulting
342
commit to be valid, deletes against the basis MUST be recorded via
343
builder.record_delete().
345
self._recording_deletes = True
347
basis_id = self.parents[0]
349
basis_id = _mod_revision.NULL_REVISION
350
self.basis_delta_revision = basis_id
352
def record_entry_contents(self, ie, parent_invs, path, tree,
354
"""Record the content of ie from tree into the commit if needed.
356
Side effect: sets ie.revision when unchanged
358
:param ie: An inventory entry present in the commit.
359
:param parent_invs: The inventories of the parent revisions of the
361
:param path: The path the entry is at in the tree.
362
:param tree: The tree which contains this entry and should be used to
364
:param content_summary: Summary data from the tree about the paths
365
content - stat, length, exec, sha/link target. This is only
366
accessed when the entry has a revision of None - that is when it is
367
a candidate to commit.
368
:return: A tuple (change_delta, version_recorded, fs_hash).
369
change_delta is an inventory_delta change for this entry against
370
the basis tree of the commit, or None if no change occured against
372
version_recorded is True if a new version of the entry has been
373
recorded. For instance, committing a merge where a file was only
374
changed on the other side will return (delta, False).
375
fs_hash is either None, or the hash details for the path (currently
376
a tuple of the contents sha1 and the statvalue returned by
377
tree.get_file_with_stat()).
379
if self.new_inventory.root is None:
380
if ie.parent_id is not None:
381
raise errors.RootMissing()
382
self._check_root(ie, parent_invs, tree)
383
if ie.revision is None:
384
kind = content_summary[0]
386
# ie is carried over from a prior commit
388
# XXX: repository specific check for nested tree support goes here - if
389
# the repo doesn't want nested trees we skip it ?
390
if (kind == 'tree-reference' and
391
not self.repository._format.supports_tree_reference):
392
# mismatch between commit builder logic and repository:
393
# this needs the entry creation pushed down into the builder.
394
raise NotImplementedError('Missing repository subtree support.')
395
self.new_inventory.add(ie)
397
# TODO: slow, take it out of the inner loop.
399
basis_inv = parent_invs[0]
401
basis_inv = Inventory(root_id=None)
403
# ie.revision is always None if the InventoryEntry is considered
404
# for committing. We may record the previous parents revision if the
405
# content is actually unchanged against a sole head.
406
if ie.revision is not None:
407
if not self._versioned_root and path == '':
408
# repositories that do not version the root set the root's
409
# revision to the new commit even when no change occurs (more
410
# specifically, they do not record a revision on the root; and
411
# the rev id is assigned to the root during deserialisation -
412
# this masks when a change may have occurred against the basis.
413
# To match this we always issue a delta, because the revision
414
# of the root will always be changing.
415
if ie.file_id in basis_inv:
416
delta = (basis_inv.id2path(ie.file_id), path,
420
delta = (None, path, ie.file_id, ie)
421
self._basis_delta.append(delta)
422
return delta, False, None
424
# we don't need to commit this, because the caller already
425
# determined that an existing revision of this file is
426
# appropriate. If its not being considered for committing then
427
# it and all its parents to the root must be unaltered so
428
# no-change against the basis.
429
if ie.revision == self._new_revision_id:
430
raise AssertionError("Impossible situation, a skipped "
431
"inventory entry (%r) claims to be modified in this "
432
"commit (%r).", (ie, self._new_revision_id))
433
return None, False, None
434
# XXX: Friction: parent_candidates should return a list not a dict
435
# so that we don't have to walk the inventories again.
436
parent_candiate_entries = ie.parent_candidates(parent_invs)
437
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
439
for inv in parent_invs:
440
if ie.file_id in inv:
441
old_rev = inv[ie.file_id].revision
442
if old_rev in head_set:
443
heads.append(inv[ie.file_id].revision)
444
head_set.remove(inv[ie.file_id].revision)
447
# now we check to see if we need to write a new record to the
449
# We write a new entry unless there is one head to the ancestors, and
450
# the kind-derived content is unchanged.
452
# Cheapest check first: no ancestors, or more the one head in the
453
# ancestors, we write a new node.
457
# There is a single head, look it up for comparison
458
parent_entry = parent_candiate_entries[heads[0]]
459
# if the non-content specific data has changed, we'll be writing a
461
if (parent_entry.parent_id != ie.parent_id or
462
parent_entry.name != ie.name):
464
# now we need to do content specific checks:
466
# if the kind changed the content obviously has
467
if kind != parent_entry.kind:
469
# Stat cache fingerprint feedback for the caller - None as we usually
470
# don't generate one.
473
if content_summary[2] is None:
474
raise ValueError("Files must not have executable = None")
476
# We can't trust a check of the file length because of content
478
if (# if the exec bit has changed we have to store:
479
parent_entry.executable != content_summary[2]):
481
elif parent_entry.text_sha1 == content_summary[3]:
482
# all meta and content is unchanged (using a hash cache
483
# hit to check the sha)
484
ie.revision = parent_entry.revision
485
ie.text_size = parent_entry.text_size
486
ie.text_sha1 = parent_entry.text_sha1
487
ie.executable = parent_entry.executable
488
return self._get_delta(ie, basis_inv, path), False, None
490
# Either there is only a hash change(no hash cache entry,
491
# or same size content change), or there is no change on
493
# Provide the parent's hash to the store layer, so that the
494
# content is unchanged we will not store a new node.
495
nostore_sha = parent_entry.text_sha1
497
# We want to record a new node regardless of the presence or
498
# absence of a content change in the file.
500
ie.executable = content_summary[2]
501
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
503
text = file_obj.read()
507
ie.text_sha1, ie.text_size = self._add_text_to_weave(
508
ie.file_id, text, heads, nostore_sha)
509
# Let the caller know we generated a stat fingerprint.
510
fingerprint = (ie.text_sha1, stat_value)
511
except errors.ExistingContent:
512
# Turns out that the file content was unchanged, and we were
513
# only going to store a new node if it was changed. Carry over
515
ie.revision = parent_entry.revision
516
ie.text_size = parent_entry.text_size
517
ie.text_sha1 = parent_entry.text_sha1
518
ie.executable = parent_entry.executable
519
return self._get_delta(ie, basis_inv, path), False, None
520
elif kind == 'directory':
522
# all data is meta here, nothing specific to directory, so
524
ie.revision = parent_entry.revision
525
return self._get_delta(ie, basis_inv, path), False, None
526
self._add_text_to_weave(ie.file_id, '', heads, None)
527
elif kind == 'symlink':
528
current_link_target = content_summary[3]
530
# symlink target is not generic metadata, check if it has
532
if current_link_target != parent_entry.symlink_target:
535
# unchanged, carry over.
536
ie.revision = parent_entry.revision
537
ie.symlink_target = parent_entry.symlink_target
538
return self._get_delta(ie, basis_inv, path), False, None
539
ie.symlink_target = current_link_target
540
self._add_text_to_weave(ie.file_id, '', heads, None)
541
elif kind == 'tree-reference':
543
if content_summary[3] != parent_entry.reference_revision:
546
# unchanged, carry over.
547
ie.reference_revision = parent_entry.reference_revision
548
ie.revision = parent_entry.revision
549
return self._get_delta(ie, basis_inv, path), False, None
550
ie.reference_revision = content_summary[3]
551
if ie.reference_revision is None:
552
raise AssertionError("invalid content_summary for nested tree: %r"
553
% (content_summary,))
554
self._add_text_to_weave(ie.file_id, '', heads, None)
556
raise NotImplementedError('unknown kind')
557
ie.revision = self._new_revision_id
558
self._any_changes = True
559
return self._get_delta(ie, basis_inv, path), True, fingerprint
561
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
562
_entry_factory=entry_factory):
563
"""Record a new tree via iter_changes.
565
:param tree: The tree to obtain text contents from for changed objects.
566
:param basis_revision_id: The revision id of the tree the iter_changes
567
has been generated against. Currently assumed to be the same
568
as self.parents[0] - if it is not, errors may occur.
569
:param iter_changes: An iter_changes iterator with the changes to apply
570
to basis_revision_id. The iterator must not include any items with
571
a current kind of None - missing items must be either filtered out
572
or errored-on beefore record_iter_changes sees the item.
573
:param _entry_factory: Private method to bind entry_factory locally for
575
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
578
# Create an inventory delta based on deltas between all the parents and
579
# deltas between all the parent inventories. We use inventory delta's
580
# between the inventory objects because iter_changes masks
581
# last-changed-field only changes.
583
# file_id -> change map, change is fileid, paths, changed, versioneds,
584
# parents, names, kinds, executables
586
# {file_id -> revision_id -> inventory entry, for entries in parent
587
# trees that are not parents[0]
591
revtrees = list(self.repository.revision_trees(self.parents))
592
except errors.NoSuchRevision:
593
# one or more ghosts, slow path.
595
for revision_id in self.parents:
597
revtrees.append(self.repository.revision_tree(revision_id))
598
except errors.NoSuchRevision:
600
basis_revision_id = _mod_revision.NULL_REVISION
602
revtrees.append(self.repository.revision_tree(
603
_mod_revision.NULL_REVISION))
604
# The basis inventory from a repository
606
basis_inv = revtrees[0].inventory
608
basis_inv = self.repository.revision_tree(
609
_mod_revision.NULL_REVISION).inventory
610
if len(self.parents) > 0:
611
if basis_revision_id != self.parents[0] and not ghost_basis:
613
"arbitrary basis parents not yet supported with merges")
614
for revtree in revtrees[1:]:
615
for change in revtree.inventory._make_delta(basis_inv):
616
if change[1] is None:
617
# Not present in this parent.
619
if change[2] not in merged_ids:
620
if change[0] is not None:
621
basis_entry = basis_inv[change[2]]
622
merged_ids[change[2]] = [
624
basis_entry.revision,
627
parent_entries[change[2]] = {
629
basis_entry.revision:basis_entry,
631
change[3].revision:change[3],
634
merged_ids[change[2]] = [change[3].revision]
635
parent_entries[change[2]] = {change[3].revision:change[3]}
637
merged_ids[change[2]].append(change[3].revision)
638
parent_entries[change[2]][change[3].revision] = change[3]
641
# Setup the changes from the tree:
642
# changes maps file_id -> (change, [parent revision_ids])
644
for change in iter_changes:
645
# This probably looks up in basis_inv way to much.
646
if change[1][0] is not None:
647
head_candidate = [basis_inv[change[0]].revision]
650
changes[change[0]] = change, merged_ids.get(change[0],
652
unchanged_merged = set(merged_ids) - set(changes)
653
# Extend the changes dict with synthetic changes to record merges of
655
for file_id in unchanged_merged:
656
# Record a merged version of these items that did not change vs the
657
# basis. This can be either identical parallel changes, or a revert
658
# of a specific file after a merge. The recorded content will be
659
# that of the current tree (which is the same as the basis), but
660
# the per-file graph will reflect a merge.
661
# NB:XXX: We are reconstructing path information we had, this
662
# should be preserved instead.
663
# inv delta change: (file_id, (path_in_source, path_in_target),
664
# changed_content, versioned, parent, name, kind,
667
basis_entry = basis_inv[file_id]
668
except errors.NoSuchId:
669
# a change from basis->some_parents but file_id isn't in basis
670
# so was new in the merge, which means it must have changed
671
# from basis -> current, and as it hasn't the add was reverted
672
# by the user. So we discard this change.
676
(basis_inv.id2path(file_id), tree.id2path(file_id)),
678
(basis_entry.parent_id, basis_entry.parent_id),
679
(basis_entry.name, basis_entry.name),
680
(basis_entry.kind, basis_entry.kind),
681
(basis_entry.executable, basis_entry.executable))
682
changes[file_id] = (change, merged_ids[file_id])
683
# changes contains tuples with the change and a set of inventory
684
# candidates for the file.
686
# old_path, new_path, file_id, new_inventory_entry
687
seen_root = False # Is the root in the basis delta?
688
inv_delta = self._basis_delta
689
modified_rev = self._new_revision_id
690
for change, head_candidates in changes.values():
691
if change[3][1]: # versioned in target.
692
# Several things may be happening here:
693
# We may have a fork in the per-file graph
694
# - record a change with the content from tree
695
# We may have a change against < all trees
696
# - carry over the tree that hasn't changed
697
# We may have a change against all trees
698
# - record the change with the content from tree
701
entry = _entry_factory[kind](file_id, change[5][1],
703
head_set = self._heads(change[0], set(head_candidates))
706
for head_candidate in head_candidates:
707
if head_candidate in head_set:
708
heads.append(head_candidate)
709
head_set.remove(head_candidate)
712
# Could be a carry-over situation:
713
parent_entry_revs = parent_entries.get(file_id, None)
714
if parent_entry_revs:
715
parent_entry = parent_entry_revs.get(heads[0], None)
718
if parent_entry is None:
719
# The parent iter_changes was called against is the one
720
# that is the per-file head, so any change is relevant
721
# iter_changes is valid.
722
carry_over_possible = False
724
# could be a carry over situation
725
# A change against the basis may just indicate a merge,
726
# we need to check the content against the source of the
727
# merge to determine if it was changed after the merge
729
if (parent_entry.kind != entry.kind or
730
parent_entry.parent_id != entry.parent_id or
731
parent_entry.name != entry.name):
732
# Metadata common to all entries has changed
733
# against per-file parent
734
carry_over_possible = False
736
carry_over_possible = True
737
# per-type checks for changes against the parent_entry
740
# Cannot be a carry-over situation
741
carry_over_possible = False
742
# Populate the entry in the delta
744
# XXX: There is still a small race here: If someone reverts the content of a file
745
# after iter_changes examines and decides it has changed,
746
# we will unconditionally record a new version even if some
747
# other process reverts it while commit is running (with
748
# the revert happening after iter_changes did it's
751
entry.executable = True
753
entry.executable = False
754
if (carry_over_possible and
755
parent_entry.executable == entry.executable):
756
# Check the file length, content hash after reading
758
nostore_sha = parent_entry.text_sha1
761
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
763
text = file_obj.read()
767
entry.text_sha1, entry.text_size = self._add_text_to_weave(
768
file_id, text, heads, nostore_sha)
769
yield file_id, change[1][1], (entry.text_sha1, stat_value)
770
except errors.ExistingContent:
771
# No content change against a carry_over parent
772
# Perhaps this should also yield a fs hash update?
774
entry.text_size = parent_entry.text_size
775
entry.text_sha1 = parent_entry.text_sha1
776
elif kind == 'symlink':
778
entry.symlink_target = tree.get_symlink_target(file_id)
779
if (carry_over_possible and
780
parent_entry.symlink_target == entry.symlink_target):
783
self._add_text_to_weave(change[0], '', heads, None)
784
elif kind == 'directory':
785
if carry_over_possible:
788
# Nothing to set on the entry.
789
# XXX: split into the Root and nonRoot versions.
790
if change[1][1] != '' or self.repository.supports_rich_root():
791
self._add_text_to_weave(change[0], '', heads, None)
792
elif kind == 'tree-reference':
793
if not self.repository._format.supports_tree_reference:
794
# This isn't quite sane as an error, but we shouldn't
795
# ever see this code path in practice: tree's don't
796
# permit references when the repo doesn't support tree
798
raise errors.UnsupportedOperation(tree.add_reference,
800
reference_revision = tree.get_reference_revision(change[0])
801
entry.reference_revision = reference_revision
802
if (carry_over_possible and
803
parent_entry.reference_revision == reference_revision):
806
self._add_text_to_weave(change[0], '', heads, None)
808
raise AssertionError('unknown kind %r' % kind)
810
entry.revision = modified_rev
812
entry.revision = parent_entry.revision
815
new_path = change[1][1]
816
inv_delta.append((change[1][0], new_path, change[0], entry))
819
self.new_inventory = None
821
# This should perhaps be guarded by a check that the basis we
822
# commit against is the basis for the commit and if not do a delta
824
self._any_changes = True
826
# housekeeping root entry changes do not affect no-change commits.
827
self._require_root_change(tree)
828
self.basis_delta_revision = basis_revision_id
830
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
831
parent_keys = tuple([(file_id, parent) for parent in parents])
832
return self.repository.texts._add_text(
833
(file_id, self._new_revision_id), parent_keys, new_text,
834
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
837
class RootCommitBuilder(CommitBuilder):
838
"""This commitbuilder actually records the root id"""
840
# the root entry gets versioned properly by this builder.
841
_versioned_root = True
843
def _check_root(self, ie, parent_invs, tree):
844
"""Helper for record_entry_contents.
846
:param ie: An entry being added.
847
:param parent_invs: The inventories of the parent revisions of the
849
:param tree: The tree that is being committed.
852
def _require_root_change(self, tree):
853
"""Enforce an appropriate root object change.
855
This is called once when record_iter_changes is called, if and only if
856
the root was not in the delta calculated by record_iter_changes.
858
:param tree: The tree which is being committed.
860
# versioned roots do not change unless the tree found a change.
863
class RepositoryWriteLockResult(object):
864
"""The result of write locking a repository.
866
:ivar repository_token: The token obtained from the underlying lock, or
868
:ivar unlock: A callable which will unlock the lock.
871
def __init__(self, unlock, repository_token):
872
self.repository_token = repository_token
876
return "RepositoryWriteLockResult(%s, %s)" % (self.repository_token,
880
63
######################################################################
884
class Repository(_RelockDebugMixin, bzrdir.ControlComponent):
66
class Repository(object):
885
67
"""Repository holding history for one or more branches.
887
69
The repository holds and retrieves historical information including
888
70
revisions and file history. It's normally accessed only by the Branch,
889
71
which views a particular line of development through that history.
891
The Repository builds on top of some byte storage facilies (the revisions,
892
signatures, inventories, texts and chk_bytes attributes) and a Transport,
893
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
896
The byte storage facilities are addressed via tuples, which we refer to
897
as 'keys' throughout the code base. Revision_keys, inventory_keys and
898
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
899
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
900
byte string made up of a hash identifier and a hash value.
901
We use this interface because it allows low friction with the underlying
902
code that implements disk indices, network encoding and other parts of
905
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
906
the serialised revisions for the repository. This can be used to obtain
907
revision graph information or to access raw serialised revisions.
908
The result of trying to insert data into the repository via this store
909
is undefined: it should be considered read-only except for implementors
911
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
912
the serialised signatures for the repository. This can be used to
913
obtain access to raw serialised signatures. The result of trying to
914
insert data into the repository via this store is undefined: it should
915
be considered read-only except for implementors of repositories.
916
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
917
the serialised inventories for the repository. This can be used to
918
obtain unserialised inventories. The result of trying to insert data
919
into the repository via this store is undefined: it should be
920
considered read-only except for implementors of repositories.
921
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
922
texts of files and directories for the repository. This can be used to
923
obtain file texts or file graphs. Note that Repository.iter_file_bytes
924
is usually a better interface for accessing file texts.
925
The result of trying to insert data into the repository via this store
926
is undefined: it should be considered read-only except for implementors
928
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
929
any data the repository chooses to store or have indexed by its hash.
930
The result of trying to insert data into the repository via this store
931
is undefined: it should be considered read-only except for implementors
933
:ivar _transport: Transport for file access to repository, typically
934
pointing to .bzr/repository.
937
# What class to use for a CommitBuilder. Often its simpler to change this
938
# in a Repository class subclass rather than to override
939
# get_commit_builder.
940
_commit_builder_class = CommitBuilder
941
# The search regex used by xml based repositories to determine what things
942
# where changed in a single commit.
943
78
_file_ids_altered_regex = lazy_regex.lazy_compile(
944
79
r'file_id="(?P<file_id>[^"]+)"'
945
r'.* revision="(?P<revision_id>[^"]+)"'
80
r'.*revision="(?P<revision_id>[^"]+)"'
948
def abort_write_group(self, suppress_errors=False):
949
"""Commit the contents accrued within the current write group.
951
:param suppress_errors: if true, abort_write_group will catch and log
952
unexpected errors that happen during the abort, rather than
953
allowing them to propagate. Defaults to False.
955
:seealso: start_write_group.
957
if self._write_group is not self.get_transaction():
958
# has an unlock or relock occured ?
961
'(suppressed) mismatched lock context and write group. %r, %r',
962
self._write_group, self.get_transaction())
964
raise errors.BzrError(
965
'mismatched lock context and write group. %r, %r' %
966
(self._write_group, self.get_transaction()))
968
self._abort_write_group()
969
except Exception, exc:
970
self._write_group = None
971
if not suppress_errors:
973
mutter('abort_write_group failed')
974
log_exception_quietly()
975
note('bzr: ERROR (ignored): %s', exc)
976
self._write_group = None
978
def _abort_write_group(self):
979
"""Template method for per-repository write group cleanup.
981
This is called during abort before the write group is considered to be
982
finished and should cleanup any internal state accrued during the write
983
group. There is no requirement that data handed to the repository be
984
*not* made available - this is not a rollback - but neither should any
985
attempt be made to ensure that data added is fully commited. Abort is
986
invoked when an error has occured so futher disk or network operations
987
may not be possible or may error and if possible should not be
991
def add_fallback_repository(self, repository):
992
"""Add a repository to use for looking up data not held locally.
994
:param repository: A repository.
996
if not self._format.supports_external_lookups:
997
raise errors.UnstackableRepositoryFormat(self._format, self.base)
999
# This repository will call fallback.unlock() when we transition to
1000
# the unlocked state, so we make sure to increment the lock count
1001
repository.lock_read()
1002
self._check_fallback_repository(repository)
1003
self._fallback_repositories.append(repository)
1004
self.texts.add_fallback_versioned_files(repository.texts)
1005
self.inventories.add_fallback_versioned_files(repository.inventories)
1006
self.revisions.add_fallback_versioned_files(repository.revisions)
1007
self.signatures.add_fallback_versioned_files(repository.signatures)
1008
if self.chk_bytes is not None:
1009
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
1011
def _check_fallback_repository(self, repository):
1012
"""Check that this repository can fallback to repository safely.
1014
Raise an error if not.
1016
:param repository: A repository to fallback to.
1018
return InterRepository._assert_same_model(self, repository)
1020
84
def add_inventory(self, revision_id, inv, parents):
1021
85
"""Add the inventory inv to the repository as revision_id.
1023
87
:param parents: The revision ids of the parents that revision_id
1024
88
is known to have and are in the repository already.
1026
:returns: The validator(which is a sha1 digest, though what is sha'd is
1027
repository format specific) of the serialized inventory.
90
returns the sha1 of the serialized inventory.
1029
if not self.is_in_write_group():
1030
raise AssertionError("%r not in write group" % (self,))
92
revision_id = osutils.safe_revision_id(revision_id)
1031
93
_mod_revision.check_not_reserved_id(revision_id)
1032
if not (inv.revision_id is None or inv.revision_id == revision_id):
1033
raise AssertionError(
1034
"Mismatch between inventory revision"
1035
" id and insertion revid (%r, %r)"
1036
% (inv.revision_id, revision_id))
1037
if inv.root is None:
1038
raise AssertionError()
1039
return self._add_inventory_checked(revision_id, inv, parents)
1041
def _add_inventory_checked(self, revision_id, inv, parents):
1042
"""Add inv to the repository after checking the inputs.
1044
This function can be overridden to allow different inventory styles.
1046
:seealso: add_inventory, for the contract.
1048
inv_lines = self._serializer.write_inventory_to_lines(inv)
1049
return self._inventory_add_lines(revision_id, parents,
1050
inv_lines, check_content=False)
1052
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1053
parents, basis_inv=None, propagate_caches=False):
1054
"""Add a new inventory expressed as a delta against another revision.
1056
See the inventory developers documentation for the theory behind
1059
:param basis_revision_id: The inventory id the delta was created
1060
against. (This does not have to be a direct parent.)
1061
:param delta: The inventory delta (see Inventory.apply_delta for
1063
:param new_revision_id: The revision id that the inventory is being
1065
:param parents: The revision ids of the parents that revision_id is
1066
known to have and are in the repository already. These are supplied
1067
for repositories that depend on the inventory graph for revision
1068
graph access, as well as for those that pun ancestry with delta
1070
:param basis_inv: The basis inventory if it is already known,
1072
:param propagate_caches: If True, the caches for this inventory are
1073
copied to and updated for the result if possible.
1075
:returns: (validator, new_inv)
1076
The validator(which is a sha1 digest, though what is sha'd is
1077
repository format specific) of the serialized inventory, and the
1078
resulting inventory.
1080
if not self.is_in_write_group():
1081
raise AssertionError("%r not in write group" % (self,))
1082
_mod_revision.check_not_reserved_id(new_revision_id)
1083
basis_tree = self.revision_tree(basis_revision_id)
1084
basis_tree.lock_read()
1086
# Note that this mutates the inventory of basis_tree, which not all
1087
# inventory implementations may support: A better idiom would be to
1088
# return a new inventory, but as there is no revision tree cache in
1089
# repository this is safe for now - RBC 20081013
1090
if basis_inv is None:
1091
basis_inv = basis_tree.inventory
1092
basis_inv.apply_delta(delta)
1093
basis_inv.revision_id = new_revision_id
1094
return (self.add_inventory(new_revision_id, basis_inv, parents),
1099
def _inventory_add_lines(self, revision_id, parents, lines,
1100
check_content=True):
1101
"""Store lines in inv_vf and return the sha1 of the inventory."""
1102
parents = [(parent,) for parent in parents]
1103
result = self.inventories.add_lines((revision_id,), parents, lines,
1104
check_content=check_content)[0]
1105
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)
1108
115
def add_revision(self, revision_id, rev, inv=None, config=None):
1109
116
"""Add rev to the revision store as revision_id.
1180
187
Returns a set of the present revisions.
1183
graph = self.get_graph()
1184
parent_map = graph.get_parent_map(revision_ids)
1185
# The old API returned a list, should this actually be a set?
1186
return parent_map.keys()
1188
def _check_inventories(self, checker):
1189
"""Check the inventories found from the revision scan.
1191
This is responsible for verifying the sha1 of inventories and
1192
creating a pending_keys set that covers data referenced by inventories.
1194
bar = ui.ui_factory.nested_progress_bar()
1196
self._do_check_inventories(checker, bar)
1200
def _do_check_inventories(self, checker, bar):
1201
"""Helper for _check_inventories."""
1203
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1204
kinds = ['chk_bytes', 'texts']
1205
count = len(checker.pending_keys)
1206
bar.update("inventories", 0, 2)
1207
current_keys = checker.pending_keys
1208
checker.pending_keys = {}
1209
# Accumulate current checks.
1210
for key in current_keys:
1211
if key[0] != 'inventories' and key[0] not in kinds:
1212
checker._report_items.append('unknown key type %r' % (key,))
1213
keys[key[0]].add(key[1:])
1214
if keys['inventories']:
1215
# NB: output order *should* be roughly sorted - topo or
1216
# inverse topo depending on repository - either way decent
1217
# to just delta against. However, pre-CHK formats didn't
1218
# try to optimise inventory layout on disk. As such the
1219
# pre-CHK code path does not use inventory deltas.
1221
for record in self.inventories.check(keys=keys['inventories']):
1222
if record.storage_kind == 'absent':
1223
checker._report_items.append(
1224
'Missing inventory {%s}' % (record.key,))
1226
last_object = self._check_record('inventories', record,
1227
checker, last_object,
1228
current_keys[('inventories',) + record.key])
1229
del keys['inventories']
1232
bar.update("texts", 1)
1233
while (checker.pending_keys or keys['chk_bytes']
1235
# Something to check.
1236
current_keys = checker.pending_keys
1237
checker.pending_keys = {}
1238
# Accumulate current checks.
1239
for key in current_keys:
1240
if key[0] not in kinds:
1241
checker._report_items.append('unknown key type %r' % (key,))
1242
keys[key[0]].add(key[1:])
1243
# Check the outermost kind only - inventories || chk_bytes || texts
1247
for record in getattr(self, kind).check(keys=keys[kind]):
1248
if record.storage_kind == 'absent':
1249
checker._report_items.append(
1250
'Missing %s {%s}' % (kind, record.key,))
1252
last_object = self._check_record(kind, record,
1253
checker, last_object, current_keys[(kind,) + record.key])
1257
def _check_record(self, kind, record, checker, last_object, item_data):
1258
"""Check a single text from this repository."""
1259
if kind == 'inventories':
1260
rev_id = record.key[0]
1261
inv = self._deserialise_inventory(rev_id,
1262
record.get_bytes_as('fulltext'))
1263
if last_object is not None:
1264
delta = inv._make_delta(last_object)
1265
for old_path, path, file_id, ie in delta:
1268
ie.check(checker, rev_id, inv)
1270
for path, ie in inv.iter_entries():
1271
ie.check(checker, rev_id, inv)
1272
if self._format.fast_deltas:
1274
elif kind == 'chk_bytes':
1275
# No code written to check chk_bytes for this repo format.
1276
checker._report_items.append(
1277
'unsupported key type chk_bytes for %s' % (record.key,))
1278
elif kind == 'texts':
1279
self._check_text(record, checker, item_data)
1281
checker._report_items.append(
1282
'unknown key type %s for %s' % (kind, record.key))
1284
def _check_text(self, record, checker, item_data):
1285
"""Check a single text."""
1286
# Check it is extractable.
1287
# TODO: check length.
1288
if record.storage_kind == 'chunked':
1289
chunks = record.get_bytes_as(record.storage_kind)
1290
sha1 = osutils.sha_strings(chunks)
1291
length = sum(map(len, chunks))
1293
content = record.get_bytes_as('fulltext')
1294
sha1 = osutils.sha_string(content)
1295
length = len(content)
1296
if item_data and sha1 != item_data[1]:
1297
checker._report_items.append(
1298
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1299
(record.key, sha1, item_data[1], item_data[2]))
190
for id in revision_ids:
191
if self.has_revision(id):
1302
196
def create(a_bzrdir):
1303
197
"""Construct the current default format repository in a_bzrdir."""
1304
198
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1306
def __init__(self, _format, a_bzrdir, control_files):
200
def __init__(self, _format, a_bzrdir, control_files, _revision_store, control_store, text_store):
1307
201
"""instantiate a Repository.
1309
203
:param _format: The format of the repository on disk.
1310
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.
1312
# In the future we will have a single api for all stores for
1313
# getting file texts, inventories and revisions, then
1314
# this construct will accept instances of those things.
1315
210
super(Repository, self).__init__()
1316
211
self._format = _format
1317
212
# the following are part of the public API for Repository:
1318
213
self.bzrdir = a_bzrdir
1319
214
self.control_files = control_files
1320
self._transport = control_files._transport
1321
self.base = self._transport.base
1323
self._reconcile_does_inventory_gc = True
1324
self._reconcile_fixes_text_parents = False
1325
self._reconcile_backsup_inventory = True
1326
self._write_group = None
1327
# Additional places to query for data.
1328
self._fallback_repositories = []
1329
# An InventoryEntry cache, used during deserialization
1330
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1331
# Is it safe to return inventory entries directly from the entry cache,
1332
# rather copying them?
1333
self._safe_to_return_from_cache = False
1336
def user_transport(self):
1337
return self.bzrdir.user_transport
1340
def control_transport(self):
1341
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()
1343
227
def __repr__(self):
1344
if self._fallback_repositories:
1345
return '%s(%r, fallback_repositories=%r)' % (
1346
self.__class__.__name__,
1348
self._fallback_repositories)
1350
return '%s(%r)' % (self.__class__.__name__,
1353
def _has_same_fallbacks(self, other_repo):
1354
"""Returns true if the repositories have the same fallbacks."""
1355
my_fb = self._fallback_repositories
1356
other_fb = other_repo._fallback_repositories
1357
if len(my_fb) != len(other_fb):
1359
for f, g in zip(my_fb, other_fb):
1360
if not f.has_same_location(g):
1364
def has_same_location(self, other):
1365
"""Returns a boolean indicating if this repository is at the same
1366
location as another repository.
1368
This might return False even when two repository objects are accessing
1369
the same physical repository via different URLs.
1371
if self.__class__ is not other.__class__:
1373
return (self._transport.base == other._transport.base)
1375
def is_in_write_group(self):
1376
"""Return True if there is an open write group.
1378
:seealso: start_write_group.
1380
return self._write_group is not None
228
return '%s(%r)' % (self.__class__.__name__,
229
self.bzrdir.transport.base)
1382
231
def is_locked(self):
1383
232
return self.control_files.is_locked()
1385
def is_write_locked(self):
1386
"""Return True if this object is write locked."""
1387
return self.is_locked() and self.control_files._lock_mode == 'w'
1389
def lock_write(self, token=None):
1390
"""Lock this repository for writing.
1392
This causes caching within the repository obejct to start accumlating
1393
data during reads, and allows a 'write_group' to be obtained. Write
1394
groups must be used for actual data insertion.
1396
A token should be passed in if you know that you have locked the object
1397
some other way, and need to synchronise this object's state with that
1400
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1402
:param token: if this is already locked, then lock_write will fail
1403
unless the token matches the existing lock.
1404
:returns: a token if this instance supports tokens, otherwise None.
1405
:raises TokenLockingNotSupported: when a token is given but this
1406
instance doesn't support using token locks.
1407
:raises MismatchedToken: if the specified token doesn't match the token
1408
of the existing lock.
1409
:seealso: start_write_group.
1410
:return: A RepositoryWriteLockResult.
1412
locked = self.is_locked()
1413
token = self.control_files.lock_write(token=token)
1415
self._warn_if_deprecated()
1416
self._note_lock('w')
1417
for repo in self._fallback_repositories:
1418
# Writes don't affect fallback repos
1420
self._refresh_data()
1421
return RepositoryWriteLockResult(self.unlock, token)
234
def lock_write(self):
235
self.control_files.lock_write()
1423
237
def lock_read(self):
1424
"""Lock the repository for read operations.
1426
:return: An object with an unlock method which will release the lock
1429
locked = self.is_locked()
1430
238
self.control_files.lock_read()
1432
self._warn_if_deprecated()
1433
self._note_lock('r')
1434
for repo in self._fallback_repositories:
1436
self._refresh_data()
1439
240
def get_physical_lock_status(self):
1440
241
return self.control_files.get_physical_lock_status()
1442
def leave_lock_in_place(self):
1443
"""Tell this repository not to release the physical lock when this
1446
If lock_write doesn't return a token, then this method is not supported.
1448
self.control_files.leave_in_place()
1450
def dont_leave_lock_in_place(self):
1451
"""Tell this repository to release the physical lock when this
1452
object is unlocked, even if it didn't originally acquire it.
1454
If lock_write doesn't return a token, then this method is not supported.
1456
self.control_files.dont_leave_in_place()
1458
243
@needs_read_lock
1459
244
def gather_stats(self, revid=None, committers=None):
1460
245
"""Gather statistics from a revision id.
1832
369
# Most control formats need the repository to be specifically
1833
370
# created, but on some old all-in-one formats it's not needed
1835
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
372
dest_repo = self._format.initialize(a_bzrdir, shared=self.is_shared())
1836
373
except errors.UninitializableFormat:
1837
374
dest_repo = a_bzrdir.open_repository()
375
self.copy_content_into(dest_repo, revision_id)
1838
376
return dest_repo
1840
def _get_sink(self):
1841
"""Return a sink for streaming into this repository."""
1842
return StreamSink(self)
1844
def _get_source(self, to_format):
1845
"""Return a source for streaming from this repository."""
1846
return StreamSource(self, to_format)
1848
378
@needs_read_lock
1849
379
def has_revision(self, revision_id):
1850
380
"""True if this repository has a copy of the revision."""
1851
return revision_id in self.has_revisions((revision_id,))
1854
def has_revisions(self, revision_ids):
1855
"""Probe to find out the presence of multiple revisions.
1857
:param revision_ids: An iterable of revision_ids.
1858
:return: A set of the revision_ids that were present.
1860
parent_map = self.revisions.get_parent_map(
1861
[(rev_id,) for rev_id in revision_ids])
1863
if _mod_revision.NULL_REVISION in revision_ids:
1864
result.add(_mod_revision.NULL_REVISION)
1865
result.update([key[0] for key in parent_map])
1869
def get_revision(self, revision_id):
1870
"""Return the Revision object for a named revision."""
1871
return self.get_revisions([revision_id])[0]
381
revision_id = osutils.safe_revision_id(revision_id)
382
return self._revision_store.has_revision_id(revision_id,
383
self.get_transaction())
1873
385
@needs_read_lock
1874
386
def get_revision_reconcile(self, revision_id):
1875
387
"""'reconcile' helper routine that allows access to a revision always.
1877
389
This variant of get_revision does not cross check the weave graph
1878
390
against the revision one as get_revision does: but it should only
1879
391
be used by reconcile, or reconcile-alike commands that are correcting
1880
392
or testing the revision graph.
1882
return self._get_revisions([revision_id])[0]
394
if not revision_id or not isinstance(revision_id, basestring):
395
raise errors.InvalidRevisionId(revision_id=revision_id,
397
return self.get_revisions([revision_id])[0]
1884
399
@needs_read_lock
1885
400
def get_revisions(self, revision_ids):
1886
"""Get many revisions at once.
1888
Repositories that need to check data on every revision read should
1889
subclass this method.
1891
return self._get_revisions(revision_ids)
1894
def _get_revisions(self, revision_ids):
1895
"""Core work logic to get many revisions without sanity checks."""
1897
for revid, rev in self._iter_revisions(revision_ids):
1899
raise errors.NoSuchRevision(self, revid)
1901
return [revs[revid] for revid in revision_ids]
1903
def _iter_revisions(self, revision_ids):
1904
"""Iterate over revision objects.
1906
:param revision_ids: An iterable of revisions to examine. None may be
1907
passed to request all revisions known to the repository. Note that
1908
not all repositories can find unreferenced revisions; for those
1909
repositories only referenced ones will be returned.
1910
:return: An iterator of (revid, revision) tuples. Absent revisions (
1911
those asked for but not available) are returned as (revid, None).
1913
if revision_ids is None:
1914
revision_ids = self.all_revision_ids()
1916
for rev_id in revision_ids:
1917
if not rev_id or not isinstance(rev_id, basestring):
1918
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1919
keys = [(key,) for key in revision_ids]
1920
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1921
for record in stream:
1922
revid = record.key[0]
1923
if record.storage_kind == 'absent':
1926
text = record.get_bytes_as('fulltext')
1927
rev = self._serializer.read_revision_from_string(text)
1930
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
401
revision_ids = [osutils.safe_revision_id(r) for r in revision_ids]
402
revs = self._revision_store.get_revisions(revision_ids,
403
self.get_transaction())
405
assert not isinstance(rev.revision_id, unicode)
406
for parent_id in rev.parent_ids:
407
assert not isinstance(parent_id, unicode)
411
def get_revision_xml(self, revision_id):
412
# TODO: jam 20070210 This shouldn't be necessary since get_revision
413
# would have already do it.
414
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
415
revision_id = osutils.safe_revision_id(revision_id)
416
rev = self.get_revision(revision_id)
418
# the current serializer..
419
self._revision_store._serializer.write_revision(rev, rev_tmp)
421
return rev_tmp.getvalue()
424
def get_revision(self, revision_id):
425
"""Return the Revision object for a named revision"""
426
# TODO: jam 20070210 get_revision_reconcile should do this for us
427
revision_id = osutils.safe_revision_id(revision_id)
428
r = self.get_revision_reconcile(revision_id)
429
# weave corruption can lead to absent revision markers that should be
431
# the following test is reasonably cheap (it needs a single weave read)
432
# and the weave is cached in read transactions. In write transactions
433
# it is not cached but typically we only read a small number of
434
# revisions. For knits when they are introduced we will probably want
435
# to ensure that caching write transactions are in use.
436
inv = self.get_inventory_weave()
437
self._check_revision_parents(r, inv)
441
def get_deltas_for_revisions(self, revisions):
1931
442
"""Produce a generator of revision deltas.
1933
444
Note that the input is a sequence of REVISIONS, not revision_ids.
1934
445
Trees will be held in memory until the generator exits.
1935
446
Each delta is relative to the revision's lefthand predecessor.
1937
:param specific_fileids: if not None, the result is filtered
1938
so that only those file-ids, their parents and their
1939
children are included.
1941
# Get the revision-ids of interest
1942
448
required_trees = set()
1943
449
for revision in revisions:
1944
450
required_trees.add(revision.revision_id)
1945
451
required_trees.update(revision.parent_ids[:1])
1947
# Get the matching filtered trees. Note that it's more
1948
# efficient to pass filtered trees to changes_from() rather
1949
# than doing the filtering afterwards. changes_from() could
1950
# arguably do the filtering itself but it's path-based, not
1951
# file-id based, so filtering before or afterwards is
1953
if specific_fileids is None:
1954
trees = dict((t.get_revision_id(), t) for
1955
t in self.revision_trees(required_trees))
1957
trees = dict((t.get_revision_id(), t) for
1958
t in self._filtered_revision_trees(required_trees,
1961
# Calculate the deltas
452
trees = dict((t.get_revision_id(), t) for
453
t in self.revision_trees(required_trees))
1962
454
for revision in revisions:
1963
455
if not revision.parent_ids:
1964
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
456
old_tree = self.revision_tree(None)
1966
458
old_tree = trees[revision.parent_ids[0]]
1967
459
yield trees[revision.revision_id].changes_from(old_tree)
1969
461
@needs_read_lock
1970
def get_revision_delta(self, revision_id, specific_fileids=None):
462
def get_revision_delta(self, revision_id):
1971
463
"""Return the delta for one revision.
1973
465
The delta is relative to the left-hand predecessor of the
1976
:param specific_fileids: if not None, the result is filtered
1977
so that only those file-ids, their parents and their
1978
children are included.
1980
468
r = self.get_revision(revision_id)
1981
return list(self.get_deltas_for_revisions([r],
1982
specific_fileids=specific_fileids))[0]
469
return list(self.get_deltas_for_revisions([r]))[0]
471
def _check_revision_parents(self, revision, inventory):
472
"""Private to Repository and Fetch.
474
This checks the parentage of revision in an inventory weave for
475
consistency and is only applicable to inventory-weave-for-ancestry
476
using repository formats & fetchers.
478
weave_parents = inventory.get_parents(revision.revision_id)
479
weave_names = inventory.versions()
480
for parent_id in revision.parent_ids:
481
if parent_id in weave_names:
482
# this parent must not be a ghost.
483
if not parent_id in weave_parents:
485
raise errors.CorruptRepository(self)
1984
487
@needs_write_lock
1985
488
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
489
revision_id = osutils.safe_revision_id(revision_id)
1986
490
signature = gpg_strategy.sign(plaintext)
1987
self.add_signature_text(revision_id, signature)
1990
def add_signature_text(self, revision_id, signature):
1991
self.signatures.add_lines((revision_id,), (),
1992
osutils.split_lines(signature))
1994
def find_text_key_references(self):
1995
"""Find the text key references within the repository.
1997
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1998
to whether they were referred to by the inventory of the
1999
revision_id that they contain. The inventory texts from all present
2000
revision ids are assessed to generate this report.
2002
revision_keys = self.revisions.keys()
2003
w = self.inventories
2004
pb = ui.ui_factory.nested_progress_bar()
2006
return self._find_text_key_references_from_xml_inventory_lines(
2007
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2011
def _find_text_key_references_from_xml_inventory_lines(self,
2013
"""Core routine for extracting references to texts from inventories.
2015
This performs the translation of xml lines to revision ids.
2017
:param line_iterator: An iterator of lines, origin_version_id
2018
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2019
to whether they were referred to by the inventory of the
2020
revision_id that they contain. Note that if that revision_id was
2021
not part of the line_iterator's output then False will be given -
2022
even though it may actually refer to that key.
2024
if not self._serializer.support_altered_by_hack:
2025
raise AssertionError(
2026
"_find_text_key_references_from_xml_inventory_lines only "
2027
"supported for branches which store inventory as unnested xml"
2028
", not on %r" % self)
491
self._revision_store.add_revision_signature_text(revision_id,
493
self.get_transaction())
495
def fileids_altered_by_revision_ids(self, revision_ids):
496
"""Find the file ids and versions affected by revisions.
498
:param revisions: an iterable containing revision ids.
499
:return: a dictionary mapping altered file-ids to an iterable of
500
revision_ids. Each altered file-ids has the exact revision_ids that
501
altered it listed explicitly.
503
assert self._serializer.support_altered_by_hack, \
504
("fileids_altered_by_revision_ids only supported for branches "
505
"which store inventory as unnested xml, not on %r" % self)
506
selected_revision_ids = set(osutils.safe_revision_id(r)
507
for r in revision_ids)
508
w = self.get_inventory_weave()
2031
511
# this code needs to read every new line in every inventory for the
2032
512
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2033
# not present in one of those inventories is unnecessary but not
513
# not present in one of those inventories is unnecessary but not
2034
514
# harmful because we are filtering by the revision id marker in the
2035
# inventory lines : we only select file ids altered in one of those
515
# inventory lines : we only select file ids altered in one of those
2036
516
# revisions. We don't need to see all lines in the inventory because
2037
517
# only those added in an inventory in rev X can contain a revision=X
2047
527
search = self._file_ids_altered_regex.search
2048
528
unescape = _unescape_xml
2049
529
setdefault = result.setdefault
2050
for line, line_key in line_iterator:
2051
match = search(line)
2054
# One call to match.group() returning multiple items is quite a
2055
# bit faster than 2 calls to match.group() each returning 1
2056
file_id, revision_id = match.group('file_id', 'revision_id')
2058
# Inlining the cache lookups helps a lot when you make 170,000
2059
# lines and 350k ids, versus 8.4 unique ids.
2060
# Using a cache helps in 2 ways:
2061
# 1) Avoids unnecessary decoding calls
2062
# 2) Re-uses cached strings, which helps in future set and
2064
# (2) is enough that removing encoding entirely along with
2065
# the cache (so we are using plain strings) results in no
2066
# performance improvement.
2068
revision_id = unescape_revid_cache[revision_id]
2070
unescaped = unescape(revision_id)
2071
unescape_revid_cache[revision_id] = unescaped
2072
revision_id = unescaped
2074
# Note that unconditionally unescaping means that we deserialise
2075
# every fileid, which for general 'pull' is not great, but we don't
2076
# really want to have some many fulltexts that this matters anyway.
2079
file_id = unescape_fileid_cache[file_id]
2081
unescaped = unescape(file_id)
2082
unescape_fileid_cache[file_id] = unescaped
2085
key = (file_id, revision_id)
2086
setdefault(key, False)
2087
if revision_id == line_key[-1]:
2091
def _inventory_xml_lines_for_keys(self, keys):
2092
"""Get a line iterator of the sort needed for findind references.
2094
Not relevant for non-xml inventory repositories.
2096
Ghosts in revision_keys are ignored.
2098
:param revision_keys: The revision keys for the inventories to inspect.
2099
:return: An iterator over (inventory line, revid) for the fulltexts of
2100
all of the xml inventories specified by revision_keys.
2102
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2103
for record in stream:
2104
if record.storage_kind != 'absent':
2105
chunks = record.get_bytes_as('chunked')
2106
revid = record.key[-1]
2107
lines = osutils.chunks_to_lines(chunks)
2111
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2113
"""Helper routine for fileids_altered_by_revision_ids.
2115
This performs the translation of xml lines to revision ids.
2117
:param line_iterator: An iterator of lines, origin_version_id
2118
:param revision_keys: The revision ids to filter for. This should be a
2119
set or other type which supports efficient __contains__ lookups, as
2120
the revision key from each parsed line will be looked up in the
2121
revision_keys filter.
2122
:return: a dictionary mapping altered file-ids to an iterable of
2123
revision_ids. Each altered file-ids has the exact revision_ids that
2124
altered it listed explicitly.
2126
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2127
line_iterator).iterkeys())
2128
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2129
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2130
self._inventory_xml_lines_for_keys(parent_keys)))
2131
new_keys = seen - parent_seen
2133
setdefault = result.setdefault
2134
for key in new_keys:
2135
setdefault(key[0], set()).add(key[-1])
2138
def _find_parent_ids_of_revisions(self, revision_ids):
2139
"""Find all parent ids that are mentioned in the revision graph.
2141
:return: set of revisions that are parents of revision_ids which are
2142
not part of revision_ids themselves
2144
parent_map = self.get_parent_map(revision_ids)
2146
map(parent_ids.update, parent_map.itervalues())
2147
parent_ids.difference_update(revision_ids)
2148
parent_ids.discard(_mod_revision.NULL_REVISION)
2151
def _find_parent_keys_of_revisions(self, revision_keys):
2152
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2154
:param revision_keys: An iterable of revision_keys.
2155
:return: The parents of all revision_keys that are not already in
2158
parent_map = self.revisions.get_parent_map(revision_keys)
2160
map(parent_keys.update, parent_map.itervalues())
2161
parent_keys.difference_update(revision_keys)
2162
parent_keys.discard(_mod_revision.NULL_REVISION)
2165
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2166
"""Find the file ids and versions affected by revisions.
2168
:param revisions: an iterable containing revision ids.
2169
:param _inv_weave: The inventory weave from this repository or None.
2170
If None, the inventory weave will be opened automatically.
2171
:return: a dictionary mapping altered file-ids to an iterable of
2172
revision_ids. Each altered file-ids has the exact revision_ids that
2173
altered it listed explicitly.
2175
selected_keys = set((revid,) for revid in revision_ids)
2176
w = _inv_weave or self.inventories
2177
return self._find_file_ids_from_xml_inventory_lines(
2178
w.iter_lines_added_or_present_in_keys(
2179
selected_keys, pb=None),
2182
def iter_files_bytes(self, desired_files):
2183
"""Iterate through file versions.
2185
Files will not necessarily be returned in the order they occur in
2186
desired_files. No specific order is guaranteed.
2188
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2189
value supplied by the caller as part of desired_files. It should
2190
uniquely identify the file version in the caller's context. (Examples:
2191
an index number or a TreeTransform trans_id.)
2193
bytes_iterator is an iterable of bytestrings for the file. The
2194
kind of iterable and length of the bytestrings are unspecified, but for
2195
this implementation, it is a list of bytes produced by
2196
VersionedFile.get_record_stream().
2198
:param desired_files: a list of (file_id, revision_id, identifier)
2202
for file_id, revision_id, callable_data in desired_files:
2203
text_keys[(file_id, revision_id)] = callable_data
2204
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2205
if record.storage_kind == 'absent':
2206
raise errors.RevisionNotPresent(record.key, self)
2207
yield text_keys[record.key], record.get_bytes_as('chunked')
2209
def _generate_text_key_index(self, text_key_references=None,
2211
"""Generate a new text key index for the repository.
2213
This is an expensive function that will take considerable time to run.
2215
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2216
list of parents, also text keys. When a given key has no parents,
2217
the parents list will be [NULL_REVISION].
2219
# All revisions, to find inventory parents.
2220
if ancestors is None:
2221
graph = self.get_graph()
2222
ancestors = graph.get_parent_map(self.all_revision_ids())
2223
if text_key_references is None:
2224
text_key_references = self.find_text_key_references()
2225
530
pb = ui.ui_factory.nested_progress_bar()
2227
return self._do_generate_text_key_index(ancestors,
2228
text_key_references, pb)
532
for line in w.iter_lines_added_or_present_in_versions(
533
selected_revision_ids, pb=pb):
537
# One call to match.group() returning multiple items is quite a
538
# bit faster than 2 calls to match.group() each returning 1
539
file_id, revision_id = match.group('file_id', 'revision_id')
541
# Inlining the cache lookups helps a lot when you make 170,000
542
# lines and 350k ids, versus 8.4 unique ids.
543
# Using a cache helps in 2 ways:
544
# 1) Avoids unnecessary decoding calls
545
# 2) Re-uses cached strings, which helps in future set and
547
# (2) is enough that removing encoding entirely along with
548
# the cache (so we are using plain strings) results in no
549
# performance improvement.
551
revision_id = unescape_revid_cache[revision_id]
553
unescaped = unescape(revision_id)
554
unescape_revid_cache[revision_id] = unescaped
555
revision_id = unescaped
557
if revision_id in selected_revision_ids:
559
file_id = unescape_fileid_cache[file_id]
561
unescaped = unescape(file_id)
562
unescape_fileid_cache[file_id] = unescaped
564
setdefault(file_id, set()).add(revision_id)
2232
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2233
"""Helper for _generate_text_key_index to avoid deep nesting."""
2234
revision_order = tsort.topo_sort(ancestors)
2235
invalid_keys = set()
2237
for revision_id in revision_order:
2238
revision_keys[revision_id] = set()
2239
text_count = len(text_key_references)
2240
# a cache of the text keys to allow reuse; costs a dict of all the
2241
# keys, but saves a 2-tuple for every child of a given key.
2243
for text_key, valid in text_key_references.iteritems():
2245
invalid_keys.add(text_key)
2247
revision_keys[text_key[1]].add(text_key)
2248
text_key_cache[text_key] = text_key
2249
del text_key_references
2251
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2252
NULL_REVISION = _mod_revision.NULL_REVISION
2253
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2254
# too small for large or very branchy trees. However, for 55K path
2255
# trees, it would be easy to use too much memory trivially. Ideally we
2256
# could gauge this by looking at available real memory etc, but this is
2257
# always a tricky proposition.
2258
inventory_cache = lru_cache.LRUCache(10)
2259
batch_size = 10 # should be ~150MB on a 55K path tree
2260
batch_count = len(revision_order) / batch_size + 1
2262
pb.update("Calculating text parents", processed_texts, text_count)
2263
for offset in xrange(batch_count):
2264
to_query = revision_order[offset * batch_size:(offset + 1) *
2268
for revision_id in to_query:
2269
parent_ids = ancestors[revision_id]
2270
for text_key in revision_keys[revision_id]:
2271
pb.update("Calculating text parents", processed_texts)
2272
processed_texts += 1
2273
candidate_parents = []
2274
for parent_id in parent_ids:
2275
parent_text_key = (text_key[0], parent_id)
2277
check_parent = parent_text_key not in \
2278
revision_keys[parent_id]
2280
# the parent parent_id is a ghost:
2281
check_parent = False
2282
# truncate the derived graph against this ghost.
2283
parent_text_key = None
2285
# look at the parent commit details inventories to
2286
# determine possible candidates in the per file graph.
2289
inv = inventory_cache[parent_id]
2291
inv = self.revision_tree(parent_id).inventory
2292
inventory_cache[parent_id] = inv
2294
parent_entry = inv[text_key[0]]
2295
except (KeyError, errors.NoSuchId):
2297
if parent_entry is not None:
2299
text_key[0], parent_entry.revision)
2301
parent_text_key = None
2302
if parent_text_key is not None:
2303
candidate_parents.append(
2304
text_key_cache[parent_text_key])
2305
parent_heads = text_graph.heads(candidate_parents)
2306
new_parents = list(parent_heads)
2307
new_parents.sort(key=lambda x:candidate_parents.index(x))
2308
if new_parents == []:
2309
new_parents = [NULL_REVISION]
2310
text_index[text_key] = new_parents
2312
for text_key in invalid_keys:
2313
text_index[text_key] = [NULL_REVISION]
2316
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2317
"""Get an iterable listing the keys of all the data introduced by a set
2320
The keys will be ordered so that the corresponding items can be safely
2321
fetched and inserted in that order.
2323
:returns: An iterable producing tuples of (knit-kind, file-id,
2324
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2325
'revisions'. file-id is None unless knit-kind is 'file'.
2327
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2330
for result in self._find_non_file_keys_to_fetch(revision_ids):
2333
def _find_file_keys_to_fetch(self, revision_ids, pb):
2334
# XXX: it's a bit weird to control the inventory weave caching in this
2335
# generator. Ideally the caching would be done in fetch.py I think. Or
2336
# maybe this generator should explicitly have the contract that it
2337
# should not be iterated until the previously yielded item has been
2339
inv_w = self.inventories
2341
# file ids that changed
2342
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2344
num_file_ids = len(file_ids)
2345
for file_id, altered_versions in file_ids.iteritems():
2347
pb.update("Fetch texts", count, num_file_ids)
2349
yield ("file", file_id, altered_versions)
2351
def _find_non_file_keys_to_fetch(self, revision_ids):
2353
yield ("inventory", None, revision_ids)
2356
# XXX: Note ATM no callers actually pay attention to this return
2357
# instead they just use the list of revision ids and ignore
2358
# missing sigs. Consider removing this work entirely
2359
revisions_with_signatures = set(self.signatures.get_parent_map(
2360
[(r,) for r in revision_ids]))
2361
revisions_with_signatures = set(
2362
[r for (r,) in revisions_with_signatures])
2363
revisions_with_signatures.intersection_update(revision_ids)
2364
yield ("signatures", None, revisions_with_signatures)
2367
yield ("revisions", None, revision_ids)
570
def get_inventory_weave(self):
571
return self.control_weaves.get_weave('inventory',
572
self.get_transaction())
2369
574
@needs_read_lock
2370
575
def get_inventory(self, revision_id):
2371
"""Get Inventory object by revision id."""
2372
return self.iter_inventories([revision_id]).next()
2374
def iter_inventories(self, revision_ids, ordering=None):
2375
"""Get many inventories by revision_ids.
2377
This will buffer some or all of the texts used in constructing the
2378
inventories in memory, but will only parse a single inventory at a
2381
:param revision_ids: The expected revision ids of the inventories.
2382
:param ordering: optional ordering, e.g. 'topological'. If not
2383
specified, the order of revision_ids will be preserved (by
2384
buffering if necessary).
2385
:return: An iterator of inventories.
2387
if ((None in revision_ids)
2388
or (_mod_revision.NULL_REVISION in revision_ids)):
2389
raise ValueError('cannot get null revision inventory')
2390
return self._iter_inventories(revision_ids, ordering)
2392
def _iter_inventories(self, revision_ids, ordering):
2393
"""single-document based inventory iteration."""
2394
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2395
for text, revision_id in inv_xmls:
2396
yield self._deserialise_inventory(revision_id, text)
2398
def _iter_inventory_xmls(self, revision_ids, ordering):
2399
if ordering is None:
2400
order_as_requested = True
2401
ordering = 'unordered'
2403
order_as_requested = False
2404
keys = [(revision_id,) for revision_id in revision_ids]
2407
if order_as_requested:
2408
key_iter = iter(keys)
2409
next_key = key_iter.next()
2410
stream = self.inventories.get_record_stream(keys, ordering, True)
2412
for record in stream:
2413
if record.storage_kind != 'absent':
2414
chunks = record.get_bytes_as('chunked')
2415
if order_as_requested:
2416
text_chunks[record.key] = chunks
2418
yield ''.join(chunks), record.key[-1]
2420
raise errors.NoSuchRevision(self, record.key)
2421
if order_as_requested:
2422
# Yield as many results as we can while preserving order.
2423
while next_key in text_chunks:
2424
chunks = text_chunks.pop(next_key)
2425
yield ''.join(chunks), next_key[-1]
2427
next_key = key_iter.next()
2428
except StopIteration:
2429
# We still want to fully consume the get_record_stream,
2430
# just in case it is not actually finished at this point
2434
def _deserialise_inventory(self, revision_id, xml):
2435
"""Transform the xml into an inventory object.
576
"""Get Inventory object by hash."""
577
# TODO: jam 20070210 Technically we don't need to sanitize, since all
578
# called functions must sanitize.
579
revision_id = osutils.safe_revision_id(revision_id)
580
return self.deserialise_inventory(
581
revision_id, self.get_inventory_xml(revision_id))
583
def deserialise_inventory(self, revision_id, xml):
584
"""Transform the xml into an inventory object.
2437
586
:param revision_id: The expected revision id of the inventory.
2438
587
:param xml: A serialised inventory.
2440
result = self._serializer.read_inventory_from_string(xml, revision_id,
2441
entry_cache=self._inventory_entry_cache,
2442
return_from_cache=self._safe_to_return_from_cache)
2443
if result.revision_id != revision_id:
2444
raise AssertionError('revision id mismatch %s != %s' % (
2445
result.revision_id, revision_id))
589
revision_id = osutils.safe_revision_id(revision_id)
590
result = self._serializer.read_inventory_from_string(xml)
591
result.root.revision = revision_id
2448
def get_serializer_format(self):
2449
return self._serializer.format_num
594
def serialise_inventory(self, inv):
595
return self._serializer.write_inventory_to_string(inv)
2451
597
@needs_read_lock
2452
def _get_inventory_xml(self, revision_id):
2453
"""Get serialized inventory as a string."""
2454
texts = self._iter_inventory_xmls([revision_id], 'unordered')
598
def get_inventory_xml(self, revision_id):
599
"""Get inventory XML as a file object."""
600
revision_id = osutils.safe_revision_id(revision_id)
2456
text, revision_id = texts.next()
2457
except StopIteration:
602
assert isinstance(revision_id, str), type(revision_id)
603
iw = self.get_inventory_weave()
604
return iw.get_text(revision_id)
2458
606
raise errors.HistoryMissing(self, 'inventory', revision_id)
2461
def get_rev_id_for_revno(self, revno, known_pair):
2462
"""Return the revision id of a revno, given a later (revno, revid)
2463
pair in the same history.
2465
:return: if found (True, revid). If the available history ran out
2466
before reaching the revno, then this returns
2467
(False, (closest_revno, closest_revid)).
2469
known_revno, known_revid = known_pair
2470
partial_history = [known_revid]
2471
distance_from_known = known_revno - revno
2472
if distance_from_known < 0:
2474
'requested revno (%d) is later than given known revno (%d)'
2475
% (revno, known_revno))
2478
self, partial_history, stop_index=distance_from_known)
2479
except errors.RevisionNotPresent, err:
2480
if err.revision_id == known_revid:
2481
# The start revision (known_revid) wasn't found.
2483
# This is a stacked repository with no fallbacks, or a there's a
2484
# left-hand ghost. Either way, even though the revision named in
2485
# the error isn't in this repo, we know it's the next step in this
2486
# left-hand history.
2487
partial_history.append(err.revision_id)
2488
if len(partial_history) <= distance_from_known:
2489
# Didn't find enough history to get a revid for the revno.
2490
earliest_revno = known_revno - len(partial_history) + 1
2491
return (False, (earliest_revno, partial_history[-1]))
2492
if len(partial_history) - 1 > distance_from_known:
2493
raise AssertionError('_iter_for_revno returned too much history')
2494
return (True, partial_history[-1])
609
def get_inventory_sha1(self, revision_id):
610
"""Return the sha1 hash of the inventory entry
612
# TODO: jam 20070210 Shouldn't this be deprecated / removed?
613
revision_id = osutils.safe_revision_id(revision_id)
614
return self.get_revision(revision_id).inventory_sha1
617
def get_revision_graph(self, revision_id=None):
618
"""Return a dictionary containing the revision graph.
620
:param revision_id: The revision_id to get a graph from. If None, then
621
the entire revision graph is returned. This is a deprecated mode of
622
operation and will be removed in the future.
623
:return: a dictionary of revision_id->revision_parents_list.
625
# special case NULL_REVISION
626
if revision_id == _mod_revision.NULL_REVISION:
628
revision_id = osutils.safe_revision_id(revision_id)
629
a_weave = self.get_inventory_weave()
630
all_revisions = self._eliminate_revisions_not_present(
632
entire_graph = dict([(node, a_weave.get_parents(node)) for
633
node in all_revisions])
634
if revision_id is None:
636
elif revision_id not in entire_graph:
637
raise errors.NoSuchRevision(self, revision_id)
639
# add what can be reached from revision_id
641
pending = set([revision_id])
642
while len(pending) > 0:
644
result[node] = entire_graph[node]
645
for revision_id in result[node]:
646
if revision_id not in result:
647
pending.add(revision_id)
651
def get_revision_graph_with_ghosts(self, revision_ids=None):
652
"""Return a graph of the revisions with ghosts marked as applicable.
654
:param revision_ids: an iterable of revisions to graph or None for all.
655
:return: a Graph object with the graph reachable from revision_ids.
657
result = graph.Graph()
659
pending = set(self.all_revision_ids())
662
pending = set(osutils.safe_revision_id(r) for r in revision_ids)
663
# special case NULL_REVISION
664
if _mod_revision.NULL_REVISION in pending:
665
pending.remove(_mod_revision.NULL_REVISION)
666
required = set(pending)
669
revision_id = pending.pop()
671
rev = self.get_revision(revision_id)
672
except errors.NoSuchRevision:
673
if revision_id in required:
676
result.add_ghost(revision_id)
678
for parent_id in rev.parent_ids:
679
# is this queued or done ?
680
if (parent_id not in pending and
681
parent_id not in done):
683
pending.add(parent_id)
684
result.add_node(revision_id, rev.parent_ids)
685
done.add(revision_id)
688
def _get_history_vf(self):
689
"""Get a versionedfile whose history graph reflects all revisions.
691
For weave repositories, this is the inventory weave.
693
return self.get_inventory_weave()
2496
695
def iter_reverse_revision_history(self, revision_id):
2497
696
"""Iterate backwards through revision ids in the lefthand history
2525
739
reconciler = RepoReconciler(self, thorough=thorough)
2526
740
reconciler.reconcile()
2527
741
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
743
@needs_read_lock
2544
744
def revision_tree(self, revision_id):
2545
745
"""Return Tree for a revision on this branch.
2547
`revision_id` may be NULL_REVISION for the empty tree revision.
747
`revision_id` may be None for the empty tree revision.
2549
revision_id = _mod_revision.ensure_null(revision_id)
2550
749
# TODO: refactor this to use an existing revision object
2551
750
# 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),
751
if revision_id is None or revision_id == _mod_revision.NULL_REVISION:
752
return RevisionTree(self, Inventory(root_id=None),
2554
753
_mod_revision.NULL_REVISION)
2556
inv = self.get_inventory(revision_id)
755
revision_id = osutils.safe_revision_id(revision_id)
756
inv = self.get_revision_inventory(revision_id)
2557
757
return RevisionTree(self, inv, revision_id)
2559
760
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)
761
"""Return Tree for a revision on this branch.
763
`revision_id` may not be None or 'null:'"""
764
assert None not in revision_ids
765
assert _mod_revision.NULL_REVISION not in revision_ids
766
texts = self.get_inventory_weave().get_texts(revision_ids)
767
for text, revision_id in zip(texts, revision_ids):
768
inv = self.deserialise_inventory(revision_id, text)
769
yield RevisionTree(self, inv, revision_id)
2585
771
@needs_read_lock
2586
def get_ancestry(self, revision_id, topo_sorted=True):
772
def get_ancestry(self, revision_id):
2587
773
"""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
775
The first element of the list is always None, indicating the origin
776
revision. This might change when we have history horizons, or
2591
777
perhaps we should have a new API.
2593
779
This is topologically sorted.
2595
if _mod_revision.is_null(revision_id):
781
if revision_id is None:
783
revision_id = osutils.safe_revision_id(revision_id)
2597
784
if not self.has_revision(revision_id):
2598
785
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 its 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
786
w = self.get_inventory_weave()
787
candidates = w.get_ancestry(revision_id)
788
return [None] + candidates # self._eliminate_revisions_not_present(candidates)
791
def print_file(self, file, revision_id):
792
"""Print `file` to stdout.
794
FIXME RBC 20060125 as John Meinel points out this is a bad api
795
- it writes to stdout, it assumes that that is valid etc. Fix
796
by creating a new more flexible convenience function.
798
revision_id = osutils.safe_revision_id(revision_id)
799
tree = self.revision_tree(revision_id)
800
# use inventory as it was in that revision
801
file_id = tree.inventory.path2id(file)
803
# TODO: jam 20060427 Write a test for this code path
804
# it had a bug in it, and was raising the wrong
806
raise errors.BzrError("%r is not present in revision %s" % (file, revision_id))
807
tree.print_file(file_id)
2636
809
def get_transaction(self):
2637
810
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)
812
def revision_parents(self, revision_id):
813
revision_id = osutils.safe_revision_id(revision_id)
814
return self.get_inventory_weave().parent_names(revision_id)
2711
816
@needs_write_lock
2712
817
def set_make_working_trees(self, new_value):
3392
1280
InterRepository.get(other).method_name(parameters).
3395
_walk_to_common_revisions_batch_size = 50
3396
1283
_optimisers = []
3397
1284
"""The available optimised InterRepository types."""
3400
1286
def copy_content(self, revision_id=None):
3401
"""Make a complete copy of the content in self into destination.
3403
This is a destructive operation! Do not use it on existing
3406
:param revision_id: Only copy the content needed to construct
3407
revision_id and its parents.
3410
self.target.set_make_working_trees(self.source.make_working_trees())
3411
except NotImplementedError:
3413
self.target.fetch(self.source, revision_id=revision_id)
3416
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
1287
raise NotImplementedError(self.copy_content)
1289
def fetch(self, revision_id=None, pb=None):
3418
1290
"""Fetch the content required to construct revision_id.
3420
1292
The content is copied from self.source to self.target.
3422
1294
:param revision_id: if None all content is copied, if NULL_REVISION no
3423
1295
content is copied.
3427
ui.ui_factory.warn_experimental_format_fetch(self)
3428
from bzrlib.fetch import RepoFetcher
3429
# See <https://launchpad.net/bugs/456077> asking for a warning here
3430
if self.source._format.network_name() != self.target._format.network_name():
3431
ui.ui_factory.show_user_warning('cross_format_fetch',
3432
from_format=self.source._format,
3433
to_format=self.target._format)
3434
f = RepoFetcher(to_repository=self.target,
3435
from_repository=self.source,
3436
last_revision=revision_id,
3437
fetch_spec=fetch_spec,
3438
find_ghosts=find_ghosts)
3440
def _walk_to_common_revisions(self, revision_ids):
3441
"""Walk out from revision_ids in source to revisions target has.
3443
:param revision_ids: The start point for the search.
3444
:return: A set of revision ids.
3446
target_graph = self.target.get_graph()
3447
revision_ids = frozenset(revision_ids)
3448
missing_revs = set()
3449
source_graph = self.source.get_graph()
3450
# ensure we don't pay silly lookup costs.
3451
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3452
null_set = frozenset([_mod_revision.NULL_REVISION])
3453
searcher_exhausted = False
3457
# Iterate the searcher until we have enough next_revs
3458
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3460
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3461
next_revs.update(next_revs_part)
3462
ghosts.update(ghosts_part)
3463
except StopIteration:
3464
searcher_exhausted = True
3466
# If there are ghosts in the source graph, and the caller asked for
3467
# them, make sure that they are present in the target.
3468
# We don't care about other ghosts as we can't fetch them and
3469
# haven't been asked to.
3470
ghosts_to_check = set(revision_ids.intersection(ghosts))
3471
revs_to_get = set(next_revs).union(ghosts_to_check)
3473
have_revs = set(target_graph.get_parent_map(revs_to_get))
3474
# we always have NULL_REVISION present.
3475
have_revs = have_revs.union(null_set)
3476
# Check if the target is missing any ghosts we need.
3477
ghosts_to_check.difference_update(have_revs)
3479
# One of the caller's revision_ids is a ghost in both the
3480
# source and the target.
3481
raise errors.NoSuchRevision(
3482
self.source, ghosts_to_check.pop())
3483
missing_revs.update(next_revs - have_revs)
3484
# Because we may have walked past the original stop point, make
3485
# sure everything is stopped
3486
stop_revs = searcher.find_seen_ancestors(have_revs)
3487
searcher.stop_searching_any(stop_revs)
3488
if searcher_exhausted:
3490
return searcher.get_result()
1296
:param pb: optional progress bar to use for progress reports. If not
1297
provided a default one will be created.
1299
Returns the copied revision count and the failed revisions in a tuple:
1302
raise NotImplementedError(self.fetch)
3492
1304
@needs_read_lock
3493
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1305
def missing_revision_ids(self, revision_id=None):
3494
1306
"""Return the revision ids that source has that target does not.
1308
These are returned in topological order.
3496
1310
:param revision_id: only return revision ids included by this
3498
:param find_ghosts: If True find missing revisions in deep history
3499
rather than just finding the surface difference.
3500
:return: A bzrlib.graph.SearchResult.
3502
# stop searching at found target revisions.
3503
if not find_ghosts and revision_id is not None:
3504
return self._walk_to_common_revisions([revision_id])
3505
1313
# generic, possibly worst case, slow code path.
3506
1314
target_ids = set(self.target.all_revision_ids())
3507
1315
if revision_id is not None:
1316
# TODO: jam 20070210 InterRepository is internal enough that it
1317
# should assume revision_ids are already utf-8
1318
revision_id = osutils.safe_revision_id(revision_id)
3508
1319
source_ids = self.source.get_ancestry(revision_id)
3509
if source_ids[0] is not None:
3510
raise AssertionError()
1320
assert source_ids[0] is None
3511
1321
source_ids.pop(0)
3513
1323
source_ids = self.source.all_revision_ids()
3514
1324
result_set = set(source_ids).difference(target_ids)
3515
return self.source.revision_ids_to_search_result(result_set)
3518
def _same_model(source, target):
3519
"""True if source and target have the same data representation.
3521
Note: this is always called on the base class; overriding it in a
3522
subclass will have no effect.
3525
InterRepository._assert_same_model(source, target)
3527
except errors.IncompatibleRepositories, e:
3531
def _assert_same_model(source, target):
3532
"""Raise an exception if two repositories do not use the same model.
3534
if source.supports_rich_root() != target.supports_rich_root():
3535
raise errors.IncompatibleRepositories(source, target,
3536
"different rich-root support")
3537
if source._serializer != target._serializer:
3538
raise errors.IncompatibleRepositories(source, target,
3539
"different serializers")
1325
# this may look like a no-op: its not. It preserves the ordering
1326
# other_ids had while only returning the members from other_ids
1327
# that we've decided we need.
1328
return [rev_id for rev_id in source_ids if rev_id in result_set]
3542
1331
class InterSameDataRepository(InterRepository):
3543
1332
"""Code for converting between repositories that represent the same data.
3545
1334
Data format and model must match for this to work.
3549
1338
def _get_repo_format_to_test(self):
3550
"""Repository format for testing with.
3552
InterSameData can pull from subtree to subtree and from non-subtree to
3553
non-subtree, so we test this with the richest repository format.
3555
from bzrlib.repofmt import knitrepo
3556
return knitrepo.RepositoryFormatKnit3()
1339
"""Repository format for testing with."""
1340
return RepositoryFormat.get_default_format()
3559
1343
def is_compatible(source, target):
3560
return InterRepository._same_model(source, target)
1344
if not isinstance(source, Repository):
1346
if not isinstance(target, Repository):
1348
if source._format.rich_root_data != target._format.rich_root_data:
1350
if source._serializer != target._serializer:
1356
def copy_content(self, revision_id=None):
1357
"""Make a complete copy of the content in self into destination.
1359
This is a destructive operation! Do not use it on existing
1362
:param revision_id: Only copy the content needed to construct
1363
revision_id and its parents.
1366
self.target.set_make_working_trees(self.source.make_working_trees())
1367
except NotImplementedError:
1369
# TODO: jam 20070210 This is fairly internal, so we should probably
1370
# just assert that revision_id is not unicode.
1371
revision_id = osutils.safe_revision_id(revision_id)
1372
# but don't bother fetching if we have the needed data now.
1373
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1374
self.target.has_revision(revision_id)):
1376
self.target.fetch(self.source, revision_id=revision_id)
1379
def fetch(self, revision_id=None, pb=None):
1380
"""See InterRepository.fetch()."""
1381
from bzrlib.fetch import GenericRepoFetcher
1382
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1383
self.source, self.source._format, self.target,
1384
self.target._format)
1385
# TODO: jam 20070210 This should be an assert, not a translate
1386
revision_id = osutils.safe_revision_id(revision_id)
1387
f = GenericRepoFetcher(to_repository=self.target,
1388
from_repository=self.source,
1389
last_revision=revision_id,
1391
return f.count_copied, f.failed_revisions
3563
1394
class InterWeaveRepo(InterSameDataRepository):
3564
"""Optimised code paths between Weave based repositories.
3566
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3567
implemented lazy inter-object optimisation.
1395
"""Optimised code paths between Weave based repositories."""
3571
1398
def _get_repo_format_to_test(self):
3740
1588
def is_compatible(source, target):
3741
"""Be compatible with Knit2 source and Knit3 target"""
3742
# This is redundant with format.check_conversion_target(), however that
3743
# raises an exception, and we just want to say "False" as in we won't
3744
# support converting between these formats.
3745
if 'IDS_never' in debug.debug_flags:
3747
if source.supports_rich_root() and not target.supports_rich_root():
3749
if (source._format.supports_tree_reference
3750
and not target._format.supports_tree_reference):
3752
if target._fallback_repositories and target._format.supports_chks:
3753
# IDS doesn't know how to copy CHKs for the parent inventories it
3754
# adds to stacked repos.
3756
if 'IDS_always' in debug.debug_flags:
1589
if not isinstance(source, Repository):
1591
if not isinstance(target, Repository):
1593
if not source._format.rich_root_data and target._format.rich_root_data:
3758
# Only use this code path for local source and target. IDS does far
3759
# too much IO (both bandwidth and roundtrips) over a network.
3760
if not source.bzrdir.transport.base.startswith('file:///'):
3762
if not target.bzrdir.transport.base.startswith('file:///'):
3766
def _get_trees(self, revision_ids, cache):
3768
for rev_id in revision_ids:
3770
possible_trees.append((rev_id, cache[rev_id]))
3772
# Not cached, but inventory might be present anyway.
3774
tree = self.source.revision_tree(rev_id)
3775
except errors.NoSuchRevision:
3776
# Nope, parent is ghost.
3779
cache[rev_id] = tree
3780
possible_trees.append((rev_id, tree))
3781
return possible_trees
3783
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3784
"""Get the best delta and base for this revision.
3786
:return: (basis_id, delta)
3789
# Generate deltas against each tree, to find the shortest.
3790
texts_possibly_new_in_tree = set()
3791
for basis_id, basis_tree in possible_trees:
3792
delta = tree.inventory._make_delta(basis_tree.inventory)
3793
for old_path, new_path, file_id, new_entry in delta:
3794
if new_path is None:
3795
# This file_id isn't present in the new rev, so we don't
3799
# Rich roots are handled elsewhere...
3801
kind = new_entry.kind
3802
if kind != 'directory' and kind != 'file':
3803
# No text record associated with this inventory entry.
3805
# This is a directory or file that has changed somehow.
3806
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3807
deltas.append((len(delta), basis_id, delta))
3809
return deltas[0][1:]
3811
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3812
"""Find all parent revisions that are absent, but for which the
3813
inventory is present, and copy those inventories.
3815
This is necessary to preserve correctness when the source is stacked
3816
without fallbacks configured. (Note that in cases like upgrade the
3817
source may be not have _fallback_repositories even though it is
3821
for parents in parent_map.values():
3822
parent_revs.update(parents)
3823
present_parents = self.source.get_parent_map(parent_revs)
3824
absent_parents = set(parent_revs).difference(present_parents)
3825
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3826
(rev_id,) for rev_id in absent_parents)
3827
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3828
for parent_tree in self.source.revision_trees(parent_inv_ids):
3829
current_revision_id = parent_tree.get_revision_id()
3830
parents_parents_keys = parent_invs_keys_for_stacking[
3831
(current_revision_id,)]
3832
parents_parents = [key[-1] for key in parents_parents_keys]
3833
basis_id = _mod_revision.NULL_REVISION
3834
basis_tree = self.source.revision_tree(basis_id)
3835
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3836
self.target.add_inventory_by_delta(
3837
basis_id, delta, current_revision_id, parents_parents)
3838
cache[current_revision_id] = parent_tree
3840
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3841
"""Fetch across a few revisions.
3843
:param revision_ids: The revisions to copy
3844
:param basis_id: The revision_id of a tree that must be in cache, used
3845
as a basis for delta when no other base is available
3846
:param cache: A cache of RevisionTrees that we can use.
3847
:param a_graph: A Graph object to determine the heads() of the
3848
rich-root data stream.
3849
:return: The revision_id of the last converted tree. The RevisionTree
3850
for it will be in cache
3852
# Walk though all revisions; get inventory deltas, copy referenced
3853
# texts that delta references, insert the delta, revision and
3855
root_keys_to_create = set()
3858
pending_revisions = []
3859
parent_map = self.source.get_parent_map(revision_ids)
3860
self._fetch_parent_invs_for_stacking(parent_map, cache)
3861
self.source._safe_to_return_from_cache = True
3862
for tree in self.source.revision_trees(revision_ids):
3863
# Find a inventory delta for this revision.
3864
# Find text entries that need to be copied, too.
3865
current_revision_id = tree.get_revision_id()
3866
parent_ids = parent_map.get(current_revision_id, ())
3867
parent_trees = self._get_trees(parent_ids, cache)
3868
possible_trees = list(parent_trees)
3869
if len(possible_trees) == 0:
3870
# There either aren't any parents, or the parents are ghosts,
3871
# so just use the last converted tree.
3872
possible_trees.append((basis_id, cache[basis_id]))
3873
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3875
revision = self.source.get_revision(current_revision_id)
3876
pending_deltas.append((basis_id, delta,
3877
current_revision_id, revision.parent_ids))
3878
if self._converting_to_rich_root:
3879
self._revision_id_to_root_id[current_revision_id] = \
3881
# Determine which texts are in present in this revision but not in
3882
# any of the available parents.
3883
texts_possibly_new_in_tree = set()
3884
for old_path, new_path, file_id, entry in delta:
3885
if new_path is None:
3886
# This file_id isn't present in the new rev
3890
if not self.target.supports_rich_root():
3891
# The target doesn't support rich root, so we don't
3894
if self._converting_to_rich_root:
3895
# This can't be copied normally, we have to insert
3897
root_keys_to_create.add((file_id, entry.revision))
3900
texts_possibly_new_in_tree.add((file_id, entry.revision))
3901
for basis_id, basis_tree in possible_trees:
3902
basis_inv = basis_tree.inventory
3903
for file_key in list(texts_possibly_new_in_tree):
3904
file_id, file_revision = file_key
3906
entry = basis_inv[file_id]
3907
except errors.NoSuchId:
3909
if entry.revision == file_revision:
3910
texts_possibly_new_in_tree.remove(file_key)
3911
text_keys.update(texts_possibly_new_in_tree)
3912
pending_revisions.append(revision)
3913
cache[current_revision_id] = tree
3914
basis_id = current_revision_id
3915
self.source._safe_to_return_from_cache = False
3917
from_texts = self.source.texts
3918
to_texts = self.target.texts
3919
if root_keys_to_create:
3920
root_stream = _mod_fetch._new_root_data_stream(
3921
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3922
self.source, graph=a_graph)
3923
to_texts.insert_record_stream(root_stream)
3924
to_texts.insert_record_stream(from_texts.get_record_stream(
3925
text_keys, self.target._format._fetch_order,
3926
not self.target._format._fetch_uses_deltas))
3927
# insert inventory deltas
3928
for delta in pending_deltas:
3929
self.target.add_inventory_by_delta(*delta)
3930
if self.target._fallback_repositories:
3931
# Make sure this stacked repository has all the parent inventories
3932
# for the new revisions that we are about to insert. We do this
3933
# before adding the revisions so that no revision is added until
3934
# all the inventories it may depend on are added.
3935
# Note that this is overzealous, as we may have fetched these in an
3938
revision_ids = set()
3939
for revision in pending_revisions:
3940
revision_ids.add(revision.revision_id)
3941
parent_ids.update(revision.parent_ids)
3942
parent_ids.difference_update(revision_ids)
3943
parent_ids.discard(_mod_revision.NULL_REVISION)
3944
parent_map = self.source.get_parent_map(parent_ids)
3945
# we iterate over parent_map and not parent_ids because we don't
3946
# want to try copying any revision which is a ghost
3947
for parent_tree in self.source.revision_trees(parent_map):
3948
current_revision_id = parent_tree.get_revision_id()
3949
parents_parents = parent_map[current_revision_id]
3950
possible_trees = self._get_trees(parents_parents, cache)
3951
if len(possible_trees) == 0:
3952
# There either aren't any parents, or the parents are
3953
# ghosts, so just use the last converted tree.
3954
possible_trees.append((basis_id, cache[basis_id]))
3955
basis_id, delta = self._get_delta_for_revision(parent_tree,
3956
parents_parents, possible_trees)
3957
self.target.add_inventory_by_delta(
3958
basis_id, delta, current_revision_id, parents_parents)
3959
# insert signatures and revisions
3960
for revision in pending_revisions:
3962
signature = self.source.get_signature_text(
3963
revision.revision_id)
3964
self.target.add_signature_text(revision.revision_id,
3966
except errors.NoSuchRevision:
3968
self.target.add_revision(revision.revision_id, revision)
3971
def _fetch_all_revisions(self, revision_ids, pb):
3972
"""Fetch everything for the list of revisions.
3974
:param revision_ids: The list of revisions to fetch. Must be in
3976
:param pb: A ProgressTask
3979
basis_id, basis_tree = self._get_basis(revision_ids[0])
3981
cache = lru_cache.LRUCache(100)
3982
cache[basis_id] = basis_tree
3983
del basis_tree # We don't want to hang on to it here
3985
if self._converting_to_rich_root and len(revision_ids) > 100:
3986
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3991
for offset in range(0, len(revision_ids), batch_size):
3992
self.target.start_write_group()
3994
pb.update('Transferring revisions', offset,
3996
batch = revision_ids[offset:offset+batch_size]
3997
basis_id = self._fetch_batch(batch, basis_id, cache,
4000
self.source._safe_to_return_from_cache = False
4001
self.target.abort_write_group()
4004
hint = self.target.commit_write_group()
4007
if hints and self.target._format.pack_compresses:
4008
self.target.pack(hint=hints)
4009
pb.update('Transferring revisions', len(revision_ids),
4013
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
4015
"""See InterRepository.fetch()."""
4016
if fetch_spec is not None:
4017
raise AssertionError("Not implemented yet...")
4018
ui.ui_factory.warn_experimental_format_fetch(self)
4019
if (not self.source.supports_rich_root()
4020
and self.target.supports_rich_root()):
4021
self._converting_to_rich_root = True
4022
self._revision_id_to_root_id = {}
4024
self._converting_to_rich_root = False
4025
# See <https://launchpad.net/bugs/456077> asking for a warning here
4026
if self.source._format.network_name() != self.target._format.network_name():
4027
ui.ui_factory.show_user_warning('cross_format_fetch',
4028
from_format=self.source._format,
4029
to_format=self.target._format)
4030
revision_ids = self.target.search_missing_revision_ids(self.source,
4031
revision_id, find_ghosts=find_ghosts).get_keys()
4032
if not revision_ids:
4034
revision_ids = tsort.topo_sort(
4035
self.source.get_graph().get_parent_map(revision_ids))
4036
if not revision_ids:
4038
# Walk though all revisions; get inventory deltas, copy referenced
4039
# texts that delta references, insert the delta, revision and
4042
my_pb = ui.ui_factory.nested_progress_bar()
4045
symbol_versioning.warn(
4046
symbol_versioning.deprecated_in((1, 14, 0))
4047
% "pb parameter to fetch()")
4050
self._fetch_all_revisions(revision_ids, pb)
4052
if my_pb is not None:
4054
return len(revision_ids), 0
4056
def _get_basis(self, first_revision_id):
4057
"""Get a revision and tree which exists in the target.
4059
This assumes that first_revision_id is selected for transmission
4060
because all other ancestors are already present. If we can't find an
4061
ancestor we fall back to NULL_REVISION since we know that is safe.
4063
:return: (basis_id, basis_tree)
4065
first_rev = self.source.get_revision(first_revision_id)
4067
basis_id = first_rev.parent_ids[0]
4068
# only valid as a basis if the target has it
4069
self.target.get_revision(basis_id)
4070
# Try to get a basis tree - if its a ghost it will hit the
4071
# NoSuchRevision case.
4072
basis_tree = self.source.revision_tree(basis_id)
4073
except (IndexError, errors.NoSuchRevision):
4074
basis_id = _mod_revision.NULL_REVISION
4075
basis_tree = self.source.revision_tree(basis_id)
4076
return basis_id, basis_tree
4079
InterRepository.register_optimiser(InterDifferingSerializer)
1599
def fetch(self, revision_id=None, pb=None):
1600
"""See InterRepository.fetch()."""
1601
from bzrlib.fetch import Model1toKnit2Fetcher
1602
# TODO: jam 20070210 This should be an assert, not a translate
1603
revision_id = osutils.safe_revision_id(revision_id)
1604
f = Model1toKnit2Fetcher(to_repository=self.target,
1605
from_repository=self.source,
1606
last_revision=revision_id,
1608
return f.count_copied, f.failed_revisions
1611
def copy_content(self, revision_id=None):
1612
"""Make a complete copy of the content in self into destination.
1614
This is a destructive operation! Do not use it on existing
1617
:param revision_id: Only copy the content needed to construct
1618
revision_id and its parents.
1621
self.target.set_make_working_trees(self.source.make_working_trees())
1622
except NotImplementedError:
1624
# TODO: jam 20070210 Internal, assert, don't translate
1625
revision_id = osutils.safe_revision_id(revision_id)
1626
# but don't bother fetching if we have the needed data now.
1627
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1628
self.target.has_revision(revision_id)):
1630
self.target.fetch(self.source, revision_id=revision_id)
1633
class InterKnit1and2(InterKnitRepo):
1636
def _get_repo_format_to_test(self):
1640
def is_compatible(source, target):
1641
"""Be compatible with Knit1 source and Knit3 target"""
1642
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit3
1644
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1, \
1645
RepositoryFormatKnit3
1646
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1647
isinstance(target._format, (RepositoryFormatKnit3)))
1648
except AttributeError:
1652
def fetch(self, revision_id=None, pb=None):
1653
"""See InterRepository.fetch()."""
1654
from bzrlib.fetch import Knit1to2Fetcher
1655
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1656
self.source, self.source._format, self.target,
1657
self.target._format)
1658
# TODO: jam 20070210 This should be an assert, not a translate
1659
revision_id = osutils.safe_revision_id(revision_id)
1660
f = Knit1to2Fetcher(to_repository=self.target,
1661
from_repository=self.source,
1662
last_revision=revision_id,
1664
return f.count_copied, f.failed_revisions
4080
1667
InterRepository.register_optimiser(InterSameDataRepository)
4081
1668
InterRepository.register_optimiser(InterWeaveRepo)
4082
1669
InterRepository.register_optimiser(InterKnitRepo)
1670
InterRepository.register_optimiser(InterModel1and2)
1671
InterRepository.register_optimiser(InterKnit1and2)
1674
class RepositoryTestProviderAdapter(object):
1675
"""A tool to generate a suite testing multiple repository formats at once.
1677
This is done by copying the test once for each transport and injecting
1678
the transport_server, transport_readonly_server, and bzrdir_format and
1679
repository_format classes into each copy. Each copy is also given a new id()
1680
to make it easy to identify.
1683
def __init__(self, transport_server, transport_readonly_server, formats):
1684
self._transport_server = transport_server
1685
self._transport_readonly_server = transport_readonly_server
1686
self._formats = formats
1688
def adapt(self, test):
1689
result = unittest.TestSuite()
1690
for repository_format, bzrdir_format in self._formats:
1691
from copy import deepcopy
1692
new_test = deepcopy(test)
1693
new_test.transport_server = self._transport_server
1694
new_test.transport_readonly_server = self._transport_readonly_server
1695
new_test.bzrdir_format = bzrdir_format
1696
new_test.repository_format = repository_format
1697
def make_new_test_id():
1698
new_id = "%s(%s)" % (new_test.id(), repository_format.__class__.__name__)
1699
return lambda: new_id
1700
new_test.id = make_new_test_id()
1701
result.addTest(new_test)
1705
class InterRepositoryTestProviderAdapter(object):
1706
"""A tool to generate a suite testing multiple inter repository formats.
1708
This is done by copying the test once for each interrepo provider and injecting
1709
the transport_server, transport_readonly_server, repository_format and
1710
repository_to_format classes into each copy.
1711
Each copy is also given a new id() to make it easy to identify.
1714
def __init__(self, transport_server, transport_readonly_server, formats):
1715
self._transport_server = transport_server
1716
self._transport_readonly_server = transport_readonly_server
1717
self._formats = formats
1719
def adapt(self, test):
1720
result = unittest.TestSuite()
1721
for interrepo_class, repository_format, repository_format_to in self._formats:
1722
from copy import deepcopy
1723
new_test = deepcopy(test)
1724
new_test.transport_server = self._transport_server
1725
new_test.transport_readonly_server = self._transport_readonly_server
1726
new_test.interrepo_class = interrepo_class
1727
new_test.repository_format = repository_format
1728
new_test.repository_format_to = repository_format_to
1729
def make_new_test_id():
1730
new_id = "%s(%s)" % (new_test.id(), interrepo_class.__name__)
1731
return lambda: new_id
1732
new_test.id = make_new_test_id()
1733
result.addTest(new_test)
1737
def default_test_list():
1738
"""Generate the default list of interrepo permutations to test."""
1739
from bzrlib.repofmt import knitrepo, weaverepo
1741
# test the default InterRepository between format 6 and the current
1743
# XXX: robertc 20060220 reinstate this when there are two supported
1744
# formats which do not have an optimal code path between them.
1745
#result.append((InterRepository,
1746
# RepositoryFormat6(),
1747
# RepositoryFormatKnit1()))
1748
for optimiser_class in InterRepository._optimisers:
1749
format_to_test = optimiser_class._get_repo_format_to_test()
1750
if format_to_test is not None:
1751
result.append((optimiser_class,
1752
format_to_test, format_to_test))
1753
# if there are specific combinations we want to use, we can add them
1755
result.append((InterModel1and2,
1756
weaverepo.RepositoryFormat5(),
1757
knitrepo.RepositoryFormatKnit3()))
1758
result.append((InterKnit1and2,
1759
knitrepo.RepositoryFormatKnit1(),
1760
knitrepo.RepositoryFormatKnit3()))
4085
1764
class CopyConverter(object):
4086
1765
"""A repository conversion tool which just performs a copy of the content.
4088
1767
This is slow but quite reliable.
4094
1773
:param target_format: The format the resulting repository should be.
4096
1775
self.target_format = target_format
4098
1777
def convert(self, repo, pb):
4099
1778
"""Perform the conversion of to_convert, giving feedback via pb.
4101
1780
:param to_convert: The disk object to convert.
4102
1781
:param pb: a progress bar to use for progress information.
4104
pb = ui.ui_factory.nested_progress_bar()
4107
1786
# this is only useful with metadir layouts - separated repo content.
4108
1787
# trigger an assertion if not such
4109
1788
repo._format.get_format_string()
4110
1789
self.repo_dir = repo.bzrdir
4111
pb.update('Moving repository to repository.backup')
1790
self.step('Moving repository to repository.backup')
4112
1791
self.repo_dir.transport.move('repository', 'repository.backup')
4113
1792
backup_transport = self.repo_dir.transport.clone('repository.backup')
4114
1793
repo._format.check_conversion_target(self.target_format)
4115
1794
self.source_repo = repo._format.open(self.repo_dir,
4117
1796
_override_transport=backup_transport)
4118
pb.update('Creating new repository')
1797
self.step('Creating new repository')
4119
1798
converted = self.target_format.initialize(self.repo_dir,
4120
1799
self.source_repo.is_shared())
4121
1800
converted.lock_write()
4123
pb.update('Copying content')
1802
self.step('Copying content into repository.')
4124
1803
self.source_repo.copy_content_into(converted)
4126
1805
converted.unlock()
4127
pb.update('Deleting old repository content')
1806
self.step('Deleting old repository content.')
4128
1807
self.repo_dir.transport.delete_tree('repository.backup')
4129
ui.ui_factory.note('repository converted')
1808
self.pb.note('repository converted')
1810
def step(self, message):
1811
"""Update the pb by a step."""
1813
self.pb.update(message, self.count, self.total)
1816
class CommitBuilder(object):
1817
"""Provides an interface to build up a commit.
1819
This allows describing a tree to be committed without needing to
1820
know the internals of the format of the repository.
1823
record_root_entry = False
1824
def __init__(self, repository, parents, config, timestamp=None,
1825
timezone=None, committer=None, revprops=None,
1827
"""Initiate a CommitBuilder.
1829
:param repository: Repository to commit to.
1830
:param parents: Revision ids of the parents of the new revision.
1831
:param config: Configuration to use.
1832
:param timestamp: Optional timestamp recorded for commit.
1833
:param timezone: Optional timezone for timestamp.
1834
:param committer: Optional committer to set for commit.
1835
:param revprops: Optional dictionary of revision properties.
1836
:param revision_id: Optional revision id.
1838
self._config = config
1840
if committer is None:
1841
self._committer = self._config.username()
1843
assert isinstance(committer, basestring), type(committer)
1844
self._committer = committer
1846
self.new_inventory = Inventory(None)
1847
self._new_revision_id = osutils.safe_revision_id(revision_id)
1848
self.parents = parents
1849
self.repository = repository
1852
if revprops is not None:
1853
self._revprops.update(revprops)
1855
if timestamp is None:
1856
timestamp = time.time()
1857
# Restrict resolution to 1ms
1858
self._timestamp = round(timestamp, 3)
1860
if timezone is None:
1861
self._timezone = osutils.local_time_offset()
1863
self._timezone = int(timezone)
1865
self._generate_revision_if_needed()
1867
def commit(self, message):
1868
"""Make the actual commit.
1870
:return: The revision id of the recorded revision.
1872
rev = _mod_revision.Revision(
1873
timestamp=self._timestamp,
1874
timezone=self._timezone,
1875
committer=self._committer,
1877
inventory_sha1=self.inv_sha1,
1878
revision_id=self._new_revision_id,
1879
properties=self._revprops)
1880
rev.parent_ids = self.parents
1881
self.repository.add_revision(self._new_revision_id, rev,
1882
self.new_inventory, self._config)
1883
return self._new_revision_id
1885
def revision_tree(self):
1886
"""Return the tree that was just committed.
1888
After calling commit() this can be called to get a RevisionTree
1889
representing the newly committed tree. This is preferred to
1890
calling Repository.revision_tree() because that may require
1891
deserializing the inventory, while we already have a copy in
1894
return RevisionTree(self.repository, self.new_inventory,
1895
self._new_revision_id)
1897
def finish_inventory(self):
1898
"""Tell the builder that the inventory is finished."""
1899
if self.new_inventory.root is None:
1900
symbol_versioning.warn('Root entry should be supplied to'
1901
' record_entry_contents, as of bzr 0.10.',
1902
DeprecationWarning, stacklevel=2)
1903
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
1904
self.new_inventory.revision_id = self._new_revision_id
1905
self.inv_sha1 = self.repository.add_inventory(
1906
self._new_revision_id,
1911
def _gen_revision_id(self):
1912
"""Return new revision-id."""
1913
return generate_ids.gen_revision_id(self._config.username(),
1916
def _generate_revision_if_needed(self):
1917
"""Create a revision id if None was supplied.
1919
If the repository can not support user-specified revision ids
1920
they should override this function and raise CannotSetRevisionId
1921
if _new_revision_id is not None.
1923
:raises: CannotSetRevisionId
1925
if self._new_revision_id is None:
1926
self._new_revision_id = self._gen_revision_id()
1928
def record_entry_contents(self, ie, parent_invs, path, tree):
1929
"""Record the content of ie from tree into the commit if needed.
1931
Side effect: sets ie.revision when unchanged
1933
:param ie: An inventory entry present in the commit.
1934
:param parent_invs: The inventories of the parent revisions of the
1936
:param path: The path the entry is at in the tree.
1937
:param tree: The tree which contains this entry and should be used to
1940
if self.new_inventory.root is None and ie.parent_id is not None:
1941
symbol_versioning.warn('Root entry should be supplied to'
1942
' record_entry_contents, as of bzr 0.10.',
1943
DeprecationWarning, stacklevel=2)
1944
self.record_entry_contents(tree.inventory.root.copy(), parent_invs,
1946
self.new_inventory.add(ie)
1948
# ie.revision is always None if the InventoryEntry is considered
1949
# for committing. ie.snapshot will record the correct revision
1950
# which may be the sole parent if it is untouched.
1951
if ie.revision is not None:
1954
# In this revision format, root entries have no knit or weave
1955
if ie is self.new_inventory.root:
1956
# When serializing out to disk and back in
1957
# root.revision is always _new_revision_id
1958
ie.revision = self._new_revision_id
1960
previous_entries = ie.find_previous_heads(
1962
self.repository.weave_store,
1963
self.repository.get_transaction())
1964
# we are creating a new revision for ie in the history store
1966
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
1968
def modified_directory(self, file_id, file_parents):
1969
"""Record the presence of a symbolic link.
1971
:param file_id: The file_id of the link to record.
1972
:param file_parents: The per-file parent revision ids.
1974
self._add_text_to_weave(file_id, [], file_parents.keys())
1976
def modified_reference(self, file_id, file_parents):
1977
"""Record the modification of a reference.
1979
:param file_id: The file_id of the link to record.
1980
:param file_parents: The per-file parent revision ids.
1982
self._add_text_to_weave(file_id, [], file_parents.keys())
1984
def modified_file_text(self, file_id, file_parents,
1985
get_content_byte_lines, text_sha1=None,
1987
"""Record the text of file file_id
1989
:param file_id: The file_id of the file to record the text of.
1990
:param file_parents: The per-file parent revision ids.
1991
:param get_content_byte_lines: A callable which will return the byte
1993
:param text_sha1: Optional SHA1 of the file contents.
1994
:param text_size: Optional size of the file contents.
1996
# mutter('storing text of file {%s} in revision {%s} into %r',
1997
# file_id, self._new_revision_id, self.repository.weave_store)
1998
# special case to avoid diffing on renames or
2000
if (len(file_parents) == 1
2001
and text_sha1 == file_parents.values()[0].text_sha1
2002
and text_size == file_parents.values()[0].text_size):
2003
previous_ie = file_parents.values()[0]
2004
versionedfile = self.repository.weave_store.get_weave(file_id,
2005
self.repository.get_transaction())
2006
versionedfile.clone_text(self._new_revision_id,
2007
previous_ie.revision, file_parents.keys())
2008
return text_sha1, text_size
2010
new_lines = get_content_byte_lines()
2011
# TODO: Rather than invoking sha_strings here, _add_text_to_weave
2012
# should return the SHA1 and size
2013
self._add_text_to_weave(file_id, new_lines, file_parents.keys())
2014
return osutils.sha_strings(new_lines), \
2015
sum(map(len, new_lines))
2017
def modified_link(self, file_id, file_parents, link_target):
2018
"""Record the presence of a symbolic link.
2020
:param file_id: The file_id of the link to record.
2021
:param file_parents: The per-file parent revision ids.
2022
:param link_target: Target location of this link.
2024
self._add_text_to_weave(file_id, [], file_parents.keys())
2026
def _add_text_to_weave(self, file_id, new_lines, parents):
2027
versionedfile = self.repository.weave_store.get_weave_or_empty(
2028
file_id, self.repository.get_transaction())
2029
versionedfile.add_lines(self._new_revision_id, parents, new_lines)
2030
versionedfile.clear_cache()
2033
class _CommitBuilder(CommitBuilder):
2034
"""Temporary class so old CommitBuilders are detected properly
2036
Note: CommitBuilder works whether or not root entry is recorded.
2039
record_root_entry = True
2042
class RootCommitBuilder(CommitBuilder):
2043
"""This commitbuilder actually records the root id"""
2045
record_root_entry = True
2047
def record_entry_contents(self, ie, parent_invs, path, tree):
2048
"""Record the content of ie from tree into the commit if needed.
2050
Side effect: sets ie.revision when unchanged
2052
:param ie: An inventory entry present in the commit.
2053
:param parent_invs: The inventories of the parent revisions of the
2055
:param path: The path the entry is at in the tree.
2056
:param tree: The tree which contains this entry and should be used to
2059
assert self.new_inventory.root is not None or ie.parent_id is None
2060
self.new_inventory.add(ie)
2062
# ie.revision is always None if the InventoryEntry is considered
2063
# for committing. ie.snapshot will record the correct revision
2064
# which may be the sole parent if it is untouched.
2065
if ie.revision is not None:
2068
previous_entries = ie.find_previous_heads(
2070
self.repository.weave_store,
2071
self.repository.get_transaction())
2072
# we are creating a new revision for ie in the history store
2074
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
4133
2077
_unescape_map = {
4158
2102
if _unescape_re is None:
4159
2103
_unescape_re = re.compile('\&([^;]*);')
4160
2104
return _unescape_re.sub(_unescaper, data)
4163
class _VersionedFileChecker(object):
4165
def __init__(self, repository, text_key_references=None, ancestors=None):
4166
self.repository = repository
4167
self.text_index = self.repository._generate_text_key_index(
4168
text_key_references=text_key_references, ancestors=ancestors)
4170
def calculate_file_version_parents(self, text_key):
4171
"""Calculate the correct parents for a file version according to
4174
parent_keys = self.text_index[text_key]
4175
if parent_keys == [_mod_revision.NULL_REVISION]:
4177
return tuple(parent_keys)
4179
def check_file_version_parents(self, texts, progress_bar=None):
4180
"""Check the parents stored in a versioned file are correct.
4182
It also detects file versions that are not referenced by their
4183
corresponding revision's inventory.
4185
:returns: A tuple of (wrong_parents, dangling_file_versions).
4186
wrong_parents is a dict mapping {revision_id: (stored_parents,
4187
correct_parents)} for each revision_id where the stored parents
4188
are not correct. dangling_file_versions is a set of (file_id,
4189
revision_id) tuples for versions that are present in this versioned
4190
file, but not used by the corresponding inventory.
4192
local_progress = None
4193
if progress_bar is None:
4194
local_progress = ui.ui_factory.nested_progress_bar()
4195
progress_bar = local_progress
4197
return self._check_file_version_parents(texts, progress_bar)
4200
local_progress.finished()
4202
def _check_file_version_parents(self, texts, progress_bar):
4203
"""See check_file_version_parents."""
4205
self.file_ids = set([file_id for file_id, _ in
4206
self.text_index.iterkeys()])
4207
# text keys is now grouped by file_id
4208
n_versions = len(self.text_index)
4209
progress_bar.update('loading text store', 0, n_versions)
4210
parent_map = self.repository.texts.get_parent_map(self.text_index)
4211
# On unlistable transports this could well be empty/error...
4212
text_keys = self.repository.texts.keys()
4213
unused_keys = frozenset(text_keys) - set(self.text_index)
4214
for num, key in enumerate(self.text_index.iterkeys()):
4215
progress_bar.update('checking text graph', num, n_versions)
4216
correct_parents = self.calculate_file_version_parents(key)
4218
knit_parents = parent_map[key]
4219
except errors.RevisionNotPresent:
4222
if correct_parents != knit_parents:
4223
wrong_parents[key] = (knit_parents, correct_parents)
4224
return wrong_parents, unused_keys
4227
def _old_get_graph(repository, revision_id):
4228
"""DO NOT USE. That is all. I'm serious."""
4229
graph = repository.get_graph()
4230
revision_graph = dict(((key, value) for key, value in
4231
graph.iter_ancestry([revision_id]) if value is not None))
4232
return _strip_NULL_ghosts(revision_graph)
4235
def _strip_NULL_ghosts(revision_graph):
4236
"""Also don't use this. more compatibility code for unmigrated clients."""
4237
# Filter ghosts, and null:
4238
if _mod_revision.NULL_REVISION in revision_graph:
4239
del revision_graph[_mod_revision.NULL_REVISION]
4240
for key, parents in revision_graph.items():
4241
revision_graph[key] = tuple(parent for parent in parents if parent
4243
return revision_graph
4246
class StreamSink(object):
4247
"""An object that can insert a stream into a repository.
4249
This interface handles the complexity of reserialising inventories and
4250
revisions from different formats, and allows unidirectional insertion into
4251
stacked repositories without looking for the missing basis parents
4255
def __init__(self, target_repo):
4256
self.target_repo = target_repo
4258
def insert_stream(self, stream, src_format, resume_tokens):
4259
"""Insert a stream's content into the target repository.
4261
:param src_format: a bzr repository format.
4263
:return: a list of resume tokens and an iterable of keys additional
4264
items required before the insertion can be completed.
4266
self.target_repo.lock_write()
4269
self.target_repo.resume_write_group(resume_tokens)
4272
self.target_repo.start_write_group()
4275
# locked_insert_stream performs a commit|suspend.
4276
return self._locked_insert_stream(stream, src_format, is_resume)
4278
self.target_repo.abort_write_group(suppress_errors=True)
4281
self.target_repo.unlock()
4283
def _locked_insert_stream(self, stream, src_format, is_resume):
4284
to_serializer = self.target_repo._format._serializer
4285
src_serializer = src_format._serializer
4287
if to_serializer == src_serializer:
4288
# If serializers match and the target is a pack repository, set the
4289
# write cache size on the new pack. This avoids poor performance
4290
# on transports where append is unbuffered (such as
4291
# RemoteTransport). This is safe to do because nothing should read
4292
# back from the target repository while a stream with matching
4293
# serialization is being inserted.
4294
# The exception is that a delta record from the source that should
4295
# be a fulltext may need to be expanded by the target (see
4296
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4297
# explicitly flush any buffered writes first in that rare case.
4299
new_pack = self.target_repo._pack_collection._new_pack
4300
except AttributeError:
4301
# Not a pack repository
4304
new_pack.set_write_cache_size(1024*1024)
4305
for substream_type, substream in stream:
4306
if 'stream' in debug.debug_flags:
4307
mutter('inserting substream: %s', substream_type)
4308
if substream_type == 'texts':
4309
self.target_repo.texts.insert_record_stream(substream)
4310
elif substream_type == 'inventories':
4311
if src_serializer == to_serializer:
4312
self.target_repo.inventories.insert_record_stream(
4315
self._extract_and_insert_inventories(
4316
substream, src_serializer)
4317
elif substream_type == 'inventory-deltas':
4318
self._extract_and_insert_inventory_deltas(
4319
substream, src_serializer)
4320
elif substream_type == 'chk_bytes':
4321
# XXX: This doesn't support conversions, as it assumes the
4322
# conversion was done in the fetch code.
4323
self.target_repo.chk_bytes.insert_record_stream(substream)
4324
elif substream_type == 'revisions':
4325
# This may fallback to extract-and-insert more often than
4326
# required if the serializers are different only in terms of
4328
if src_serializer == to_serializer:
4329
self.target_repo.revisions.insert_record_stream(
4332
self._extract_and_insert_revisions(substream,
4334
elif substream_type == 'signatures':
4335
self.target_repo.signatures.insert_record_stream(substream)
4337
raise AssertionError('kaboom! %s' % (substream_type,))
4338
# Done inserting data, and the missing_keys calculations will try to
4339
# read back from the inserted data, so flush the writes to the new pack
4340
# (if this is pack format).
4341
if new_pack is not None:
4342
new_pack._write_data('', flush=True)
4343
# Find all the new revisions (including ones from resume_tokens)
4344
missing_keys = self.target_repo.get_missing_parent_inventories(
4345
check_for_missing_texts=is_resume)
4347
for prefix, versioned_file in (
4348
('texts', self.target_repo.texts),
4349
('inventories', self.target_repo.inventories),
4350
('revisions', self.target_repo.revisions),
4351
('signatures', self.target_repo.signatures),
4352
('chk_bytes', self.target_repo.chk_bytes),
4354
if versioned_file is None:
4356
# TODO: key is often going to be a StaticTuple object
4357
# I don't believe we can define a method by which
4358
# (prefix,) + StaticTuple will work, though we could
4359
# define a StaticTuple.sq_concat that would allow you to
4360
# pass in either a tuple or a StaticTuple as the second
4361
# object, so instead we could have:
4362
# StaticTuple(prefix) + key here...
4363
missing_keys.update((prefix,) + key for key in
4364
versioned_file.get_missing_compression_parent_keys())
4365
except NotImplementedError:
4366
# cannot even attempt suspending, and missing would have failed
4367
# during stream insertion.
4368
missing_keys = set()
4371
# suspend the write group and tell the caller what we is
4372
# missing. We know we can suspend or else we would not have
4373
# entered this code path. (All repositories that can handle
4374
# missing keys can handle suspending a write group).
4375
write_group_tokens = self.target_repo.suspend_write_group()
4376
return write_group_tokens, missing_keys
4377
hint = self.target_repo.commit_write_group()
4378
if (to_serializer != src_serializer and
4379
self.target_repo._format.pack_compresses):
4380
self.target_repo.pack(hint=hint)
4383
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4384
target_rich_root = self.target_repo._format.rich_root_data
4385
target_tree_refs = self.target_repo._format.supports_tree_reference
4386
for record in substream:
4387
# Insert the delta directly
4388
inventory_delta_bytes = record.get_bytes_as('fulltext')
4389
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4391
parse_result = deserialiser.parse_text_bytes(
4392
inventory_delta_bytes)
4393
except inventory_delta.IncompatibleInventoryDelta, err:
4394
trace.mutter("Incompatible delta: %s", err.msg)
4395
raise errors.IncompatibleRevision(self.target_repo._format)
4396
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4397
revision_id = new_id
4398
parents = [key[0] for key in record.parents]
4399
self.target_repo.add_inventory_by_delta(
4400
basis_id, inv_delta, revision_id, parents)
4402
def _extract_and_insert_inventories(self, substream, serializer,
4404
"""Generate a new inventory versionedfile in target, converting data.
4406
The inventory is retrieved from the source, (deserializing it), and
4407
stored in the target (reserializing it in a different format).
4409
target_rich_root = self.target_repo._format.rich_root_data
4410
target_tree_refs = self.target_repo._format.supports_tree_reference
4411
for record in substream:
4412
# It's not a delta, so it must be a fulltext in the source
4413
# serializer's format.
4414
bytes = record.get_bytes_as('fulltext')
4415
revision_id = record.key[0]
4416
inv = serializer.read_inventory_from_string(bytes, revision_id)
4417
parents = [key[0] for key in record.parents]
4418
self.target_repo.add_inventory(revision_id, inv, parents)
4419
# No need to keep holding this full inv in memory when the rest of
4420
# the substream is likely to be all deltas.
4423
def _extract_and_insert_revisions(self, substream, serializer):
4424
for record in substream:
4425
bytes = record.get_bytes_as('fulltext')
4426
revision_id = record.key[0]
4427
rev = serializer.read_revision_from_string(bytes)
4428
if rev.revision_id != revision_id:
4429
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4430
self.target_repo.add_revision(revision_id, rev)
4433
if self.target_repo._format._fetch_reconcile:
4434
self.target_repo.reconcile()
4437
class StreamSource(object):
4438
"""A source of a stream for fetching between repositories."""
4440
def __init__(self, from_repository, to_format):
4441
"""Create a StreamSource streaming from from_repository."""
4442
self.from_repository = from_repository
4443
self.to_format = to_format
4445
def delta_on_metadata(self):
4446
"""Return True if delta's are permitted on metadata streams.
4448
That is on revisions and signatures.
4450
src_serializer = self.from_repository._format._serializer
4451
target_serializer = self.to_format._serializer
4452
return (self.to_format._fetch_uses_deltas and
4453
src_serializer == target_serializer)
4455
def _fetch_revision_texts(self, revs):
4456
# fetch signatures first and then the revision texts
4457
# may need to be a InterRevisionStore call here.
4458
from_sf = self.from_repository.signatures
4459
# A missing signature is just skipped.
4460
keys = [(rev_id,) for rev_id in revs]
4461
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4463
self.to_format._fetch_order,
4464
not self.to_format._fetch_uses_deltas))
4465
# If a revision has a delta, this is actually expanded inside the
4466
# insert_record_stream code now, which is an alternate fix for
4468
from_rf = self.from_repository.revisions
4469
revisions = from_rf.get_record_stream(
4471
self.to_format._fetch_order,
4472
not self.delta_on_metadata())
4473
return [('signatures', signatures), ('revisions', revisions)]
4475
def _generate_root_texts(self, revs):
4476
"""This will be called by get_stream between fetching weave texts and
4477
fetching the inventory weave.
4479
if self._rich_root_upgrade():
4480
return _mod_fetch.Inter1and2Helper(
4481
self.from_repository).generate_root_texts(revs)
4485
def get_stream(self, search):
4487
revs = search.get_keys()
4488
graph = self.from_repository.get_graph()
4489
revs = tsort.topo_sort(graph.get_parent_map(revs))
4490
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4492
for knit_kind, file_id, revisions in data_to_fetch:
4493
if knit_kind != phase:
4495
# Make a new progress bar for this phase
4496
if knit_kind == "file":
4497
# Accumulate file texts
4498
text_keys.extend([(file_id, revision) for revision in
4500
elif knit_kind == "inventory":
4501
# Now copy the file texts.
4502
from_texts = self.from_repository.texts
4503
yield ('texts', from_texts.get_record_stream(
4504
text_keys, self.to_format._fetch_order,
4505
not self.to_format._fetch_uses_deltas))
4506
# Cause an error if a text occurs after we have done the
4509
# Before we process the inventory we generate the root
4510
# texts (if necessary) so that the inventories references
4512
for _ in self._generate_root_texts(revs):
4514
# we fetch only the referenced inventories because we do not
4515
# know for unselected inventories whether all their required
4516
# texts are present in the other repository - it could be
4518
for info in self._get_inventory_stream(revs):
4520
elif knit_kind == "signatures":
4521
# Nothing to do here; this will be taken care of when
4522
# _fetch_revision_texts happens.
4524
elif knit_kind == "revisions":
4525
for record in self._fetch_revision_texts(revs):
4528
raise AssertionError("Unknown knit kind %r" % knit_kind)
4530
def get_stream_for_missing_keys(self, missing_keys):
4531
# missing keys can only occur when we are byte copying and not
4532
# translating (because translation means we don't send
4533
# unreconstructable deltas ever).
4535
keys['texts'] = set()
4536
keys['revisions'] = set()
4537
keys['inventories'] = set()
4538
keys['chk_bytes'] = set()
4539
keys['signatures'] = set()
4540
for key in missing_keys:
4541
keys[key[0]].add(key[1:])
4542
if len(keys['revisions']):
4543
# If we allowed copying revisions at this point, we could end up
4544
# copying a revision without copying its required texts: a
4545
# violation of the requirements for repository integrity.
4546
raise AssertionError(
4547
'cannot copy revisions to fill in missing deltas %s' % (
4548
keys['revisions'],))
4549
for substream_kind, keys in keys.iteritems():
4550
vf = getattr(self.from_repository, substream_kind)
4551
if vf is None and keys:
4552
raise AssertionError(
4553
"cannot fill in keys for a versioned file we don't"
4554
" have: %s needs %s" % (substream_kind, keys))
4556
# No need to stream something we don't have
4558
if substream_kind == 'inventories':
4559
# Some missing keys are genuinely ghosts, filter those out.
4560
present = self.from_repository.inventories.get_parent_map(keys)
4561
revs = [key[0] for key in present]
4562
# Get the inventory stream more-or-less as we do for the
4563
# original stream; there's no reason to assume that records
4564
# direct from the source will be suitable for the sink. (Think
4565
# e.g. 2a -> 1.9-rich-root).
4566
for info in self._get_inventory_stream(revs, missing=True):
4570
# Ask for full texts always so that we don't need more round trips
4571
# after this stream.
4572
# Some of the missing keys are genuinely ghosts, so filter absent
4573
# records. The Sink is responsible for doing another check to
4574
# ensure that ghosts don't introduce missing data for future
4576
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4577
self.to_format._fetch_order, True))
4578
yield substream_kind, stream
4580
def inventory_fetch_order(self):
4581
if self._rich_root_upgrade():
4582
return 'topological'
4584
return self.to_format._fetch_order
4586
def _rich_root_upgrade(self):
4587
return (not self.from_repository._format.rich_root_data and
4588
self.to_format.rich_root_data)
4590
def _get_inventory_stream(self, revision_ids, missing=False):
4591
from_format = self.from_repository._format
4592
if (from_format.supports_chks and self.to_format.supports_chks and
4593
from_format.network_name() == self.to_format.network_name()):
4594
raise AssertionError(
4595
"this case should be handled by GroupCHKStreamSource")
4596
elif 'forceinvdeltas' in debug.debug_flags:
4597
return self._get_convertable_inventory_stream(revision_ids,
4598
delta_versus_null=missing)
4599
elif from_format.network_name() == self.to_format.network_name():
4601
return self._get_simple_inventory_stream(revision_ids,
4603
elif (not from_format.supports_chks and not self.to_format.supports_chks
4604
and from_format._serializer == self.to_format._serializer):
4605
# Essentially the same format.
4606
return self._get_simple_inventory_stream(revision_ids,
4609
# Any time we switch serializations, we want to use an
4610
# inventory-delta based approach.
4611
return self._get_convertable_inventory_stream(revision_ids,
4612
delta_versus_null=missing)
4614
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4615
# NB: This currently reopens the inventory weave in source;
4616
# using a single stream interface instead would avoid this.
4617
from_weave = self.from_repository.inventories
4619
delta_closure = True
4621
delta_closure = not self.delta_on_metadata()
4622
yield ('inventories', from_weave.get_record_stream(
4623
[(rev_id,) for rev_id in revision_ids],
4624
self.inventory_fetch_order(), delta_closure))
4626
def _get_convertable_inventory_stream(self, revision_ids,
4627
delta_versus_null=False):
4628
# The two formats are sufficiently different that there is no fast
4629
# path, so we need to send just inventorydeltas, which any
4630
# sufficiently modern client can insert into any repository.
4631
# The StreamSink code expects to be able to
4632
# convert on the target, so we need to put bytes-on-the-wire that can
4633
# be converted. That means inventory deltas (if the remote is <1.19,
4634
# RemoteStreamSink will fallback to VFS to insert the deltas).
4635
yield ('inventory-deltas',
4636
self._stream_invs_as_deltas(revision_ids,
4637
delta_versus_null=delta_versus_null))
4639
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4640
"""Return a stream of inventory-deltas for the given rev ids.
4642
:param revision_ids: The list of inventories to transmit
4643
:param delta_versus_null: Don't try to find a minimal delta for this
4644
entry, instead compute the delta versus the NULL_REVISION. This
4645
effectively streams a complete inventory. Used for stuff like
4646
filling in missing parents, etc.
4648
from_repo = self.from_repository
4649
revision_keys = [(rev_id,) for rev_id in revision_ids]
4650
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4651
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4653
inventories = self.from_repository.iter_inventories(
4654
revision_ids, 'topological')
4655
format = from_repo._format
4656
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4657
inventory_cache = lru_cache.LRUCache(50)
4658
null_inventory = from_repo.revision_tree(
4659
_mod_revision.NULL_REVISION).inventory
4660
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4661
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4662
# repo back into a non-rich-root repo ought to be allowed)
4663
serializer = inventory_delta.InventoryDeltaSerializer(
4664
versioned_root=format.rich_root_data,
4665
tree_references=format.supports_tree_reference)
4666
for inv in inventories:
4667
key = (inv.revision_id,)
4668
parent_keys = parent_map.get(key, ())
4670
if not delta_versus_null and parent_keys:
4671
# The caller did not ask for complete inventories and we have
4672
# some parents that we can delta against. Make a delta against
4673
# each parent so that we can find the smallest.
4674
parent_ids = [parent_key[0] for parent_key in parent_keys]
4675
for parent_id in parent_ids:
4676
if parent_id not in invs_sent_so_far:
4677
# We don't know that the remote side has this basis, so
4680
if parent_id == _mod_revision.NULL_REVISION:
4681
parent_inv = null_inventory
4683
parent_inv = inventory_cache.get(parent_id, None)
4684
if parent_inv is None:
4685
parent_inv = from_repo.get_inventory(parent_id)
4686
candidate_delta = inv._make_delta(parent_inv)
4687
if (delta is None or
4688
len(delta) > len(candidate_delta)):
4689
delta = candidate_delta
4690
basis_id = parent_id
4692
# Either none of the parents ended up being suitable, or we
4693
# were asked to delta against NULL
4694
basis_id = _mod_revision.NULL_REVISION
4695
delta = inv._make_delta(null_inventory)
4696
invs_sent_so_far.add(inv.revision_id)
4697
inventory_cache[inv.revision_id] = inv
4698
delta_serialized = ''.join(
4699
serializer.delta_to_lines(basis_id, key[-1], delta))
4700
yield versionedfile.FulltextContentFactory(
4701
key, parent_keys, None, delta_serialized)
4704
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4705
stop_revision=None):
4706
"""Extend the partial history to include a given index
4708
If a stop_index is supplied, stop when that index has been reached.
4709
If a stop_revision is supplied, stop when that revision is
4710
encountered. Otherwise, stop when the beginning of history is
4713
:param stop_index: The index which should be present. When it is
4714
present, history extension will stop.
4715
:param stop_revision: The revision id which should be present. When
4716
it is encountered, history extension will stop.
4718
start_revision = partial_history_cache[-1]
4719
iterator = repo.iter_reverse_revision_history(start_revision)
4721
#skip the last revision in the list
4724
if (stop_index is not None and
4725
len(partial_history_cache) > stop_index):
4727
if partial_history_cache[-1] == stop_revision:
4729
revision_id = iterator.next()
4730
partial_history_cache.append(revision_id)
4731
except StopIteration: