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(), """
21
from binascii import hexlify
22
from copy import deepcopy
23
27
from bzrlib import (
42
41
revision as _mod_revision,
50
from bzrlib.bundle import serializer
50
from bzrlib.osutils import (
51
55
from bzrlib.revisiontree import RevisionTree
52
56
from bzrlib.store.versioned import VersionedFileStore
57
from bzrlib.store.text import TextStore
53
58
from bzrlib.testament import Testament
56
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
61
from bzrlib.decorators import needs_read_lock, needs_write_lock
57
62
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)
63
from bzrlib.inventory import Inventory, InventoryDirectory, ROOT_ID
64
from bzrlib.symbol_versioning import (
68
from bzrlib.trace import mutter, note, warning
70
71
# Old formats display a warning, but only once
71
72
_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
######################################################################
867
class Repository(_RelockDebugMixin, bzrdir.ControlComponent):
75
class Repository(object):
868
76
"""Repository holding history for one or more branches.
870
78
The repository holds and retrieves historical information including
871
79
revisions and file history. It's normally accessed only by the Branch,
872
80
which views a particular line of development through that history.
874
The Repository builds on top of some byte storage facilies (the revisions,
875
signatures, inventories, texts and chk_bytes attributes) and a Transport,
876
which respectively provide byte storage and a means to access the (possibly
82
The Repository builds on top of Stores and a Transport, which respectively
83
describe the disk data format and the way of accessing the (possibly
879
The byte storage facilities are addressed via tuples, which we refer to
880
as 'keys' throughout the code base. Revision_keys, inventory_keys and
881
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
882
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
883
byte string made up of a hash identifier and a hash value.
884
We use this interface because it allows low friction with the underlying
885
code that implements disk indices, network encoding and other parts of
888
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
889
the serialised revisions for the repository. This can be used to obtain
890
revision graph information or to access raw serialised revisions.
891
The result of trying to insert data into the repository via this store
892
is undefined: it should be considered read-only except for implementors
894
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
895
the serialised signatures for the repository. This can be used to
896
obtain access to raw serialised signatures. The result of trying to
897
insert data into the repository via this store is undefined: it should
898
be considered read-only except for implementors of repositories.
899
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
900
the serialised inventories for the repository. This can be used to
901
obtain unserialised inventories. The result of trying to insert data
902
into the repository via this store is undefined: it should be
903
considered read-only except for implementors of repositories.
904
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
905
texts of files and directories for the repository. This can be used to
906
obtain file texts or file graphs. Note that Repository.iter_file_bytes
907
is usually a better interface for accessing file texts.
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 chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
912
any data the repository chooses to store or have indexed by its hash.
913
The result of trying to insert data into the repository via this store
914
is undefined: it should be considered read-only except for implementors
916
:ivar _transport: Transport for file access to repository, typically
917
pointing to .bzr/repository.
920
# What class to use for a CommitBuilder. Often its simpler to change this
921
# in a Repository class subclass rather than to override
922
# get_commit_builder.
923
_commit_builder_class = CommitBuilder
924
# The search regex used by xml based repositories to determine what things
925
# where changed in a single commit.
926
87
_file_ids_altered_regex = lazy_regex.lazy_compile(
927
88
r'file_id="(?P<file_id>[^"]+)"'
928
r'.* revision="(?P<revision_id>[^"]+)"'
89
r'.*revision="(?P<revision_id>[^"]+)"'
931
def abort_write_group(self, suppress_errors=False):
932
"""Commit the contents accrued within the current write group.
934
:param suppress_errors: if true, abort_write_group will catch and log
935
unexpected errors that happen during the abort, rather than
936
allowing them to propagate. Defaults to False.
938
:seealso: start_write_group.
940
if self._write_group is not self.get_transaction():
941
# has an unlock or relock occured ?
944
'(suppressed) mismatched lock context and write group. %r, %r',
945
self._write_group, self.get_transaction())
947
raise errors.BzrError(
948
'mismatched lock context and write group. %r, %r' %
949
(self._write_group, self.get_transaction()))
951
self._abort_write_group()
952
except Exception, exc:
953
self._write_group = None
954
if not suppress_errors:
956
mutter('abort_write_group failed')
957
log_exception_quietly()
958
note('bzr: ERROR (ignored): %s', exc)
959
self._write_group = None
961
def _abort_write_group(self):
962
"""Template method for per-repository write group cleanup.
964
This is called during abort before the write group is considered to be
965
finished and should cleanup any internal state accrued during the write
966
group. There is no requirement that data handed to the repository be
967
*not* made available - this is not a rollback - but neither should any
968
attempt be made to ensure that data added is fully commited. Abort is
969
invoked when an error has occured so futher disk or network operations
970
may not be possible or may error and if possible should not be
974
def add_fallback_repository(self, repository):
975
"""Add a repository to use for looking up data not held locally.
977
:param repository: A repository.
979
if not self._format.supports_external_lookups:
980
raise errors.UnstackableRepositoryFormat(self._format, self.base)
982
# This repository will call fallback.unlock() when we transition to
983
# the unlocked state, so we make sure to increment the lock count
984
repository.lock_read()
985
self._check_fallback_repository(repository)
986
self._fallback_repositories.append(repository)
987
self.texts.add_fallback_versioned_files(repository.texts)
988
self.inventories.add_fallback_versioned_files(repository.inventories)
989
self.revisions.add_fallback_versioned_files(repository.revisions)
990
self.signatures.add_fallback_versioned_files(repository.signatures)
991
if self.chk_bytes is not None:
992
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
994
def _check_fallback_repository(self, repository):
995
"""Check that this repository can fallback to repository safely.
997
Raise an error if not.
999
:param repository: A repository to fallback to.
1001
return InterRepository._assert_same_model(self, repository)
1003
def add_inventory(self, revision_id, inv, parents):
1004
"""Add the inventory inv to the repository as revision_id.
1006
:param parents: The revision ids of the parents that revision_id
93
def add_inventory(self, revid, inv, parents):
94
"""Add the inventory inv to the repository as revid.
96
:param parents: The revision ids of the parents that revid
1007
97
is known to have and are in the repository already.
1009
:returns: The validator(which is a sha1 digest, though what is sha'd is
1010
repository format specific) of the serialized inventory.
1012
if not self.is_in_write_group():
1013
raise AssertionError("%r not in write group" % (self,))
1014
_mod_revision.check_not_reserved_id(revision_id)
1015
if not (inv.revision_id is None or inv.revision_id == revision_id):
1016
raise AssertionError(
1017
"Mismatch between inventory revision"
1018
" id and insertion revid (%r, %r)"
1019
% (inv.revision_id, revision_id))
1020
if inv.root is None:
1021
raise AssertionError()
1022
return self._add_inventory_checked(revision_id, inv, parents)
1024
def _add_inventory_checked(self, revision_id, inv, parents):
1025
"""Add inv to the repository after checking the inputs.
1027
This function can be overridden to allow different inventory styles.
1029
:seealso: add_inventory, for the contract.
1031
inv_lines = self._serializer.write_inventory_to_lines(inv)
1032
return self._inventory_add_lines(revision_id, parents,
1033
inv_lines, check_content=False)
1035
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1036
parents, basis_inv=None, propagate_caches=False):
1037
"""Add a new inventory expressed as a delta against another revision.
1039
See the inventory developers documentation for the theory behind
1042
:param basis_revision_id: The inventory id the delta was created
1043
against. (This does not have to be a direct parent.)
1044
:param delta: The inventory delta (see Inventory.apply_delta for
1046
:param new_revision_id: The revision id that the inventory is being
1048
:param parents: The revision ids of the parents that revision_id is
1049
known to have and are in the repository already. These are supplied
1050
for repositories that depend on the inventory graph for revision
1051
graph access, as well as for those that pun ancestry with delta
1053
:param basis_inv: The basis inventory if it is already known,
1055
:param propagate_caches: If True, the caches for this inventory are
1056
copied to and updated for the result if possible.
1058
:returns: (validator, new_inv)
1059
The validator(which is a sha1 digest, though what is sha'd is
1060
repository format specific) of the serialized inventory, and the
1061
resulting inventory.
1063
if not self.is_in_write_group():
1064
raise AssertionError("%r not in write group" % (self,))
1065
_mod_revision.check_not_reserved_id(new_revision_id)
1066
basis_tree = self.revision_tree(basis_revision_id)
1067
basis_tree.lock_read()
1069
# Note that this mutates the inventory of basis_tree, which not all
1070
# inventory implementations may support: A better idiom would be to
1071
# return a new inventory, but as there is no revision tree cache in
1072
# repository this is safe for now - RBC 20081013
1073
if basis_inv is None:
1074
basis_inv = basis_tree.inventory
1075
basis_inv.apply_delta(delta)
1076
basis_inv.revision_id = new_revision_id
1077
return (self.add_inventory(new_revision_id, basis_inv, parents),
1082
def _inventory_add_lines(self, revision_id, parents, lines,
1083
check_content=True):
1084
"""Store lines in inv_vf and return the sha1 of the inventory."""
1085
parents = [(parent,) for parent in parents]
1086
result = self.inventories.add_lines((revision_id,), parents, lines,
1087
check_content=check_content)[0]
1088
self.inventories._access.flush()
1091
def add_revision(self, revision_id, rev, inv=None, config=None):
1092
"""Add rev to the revision store as revision_id.
1094
:param revision_id: the revision id to use.
99
returns the sha1 of the serialized inventory.
101
_mod_revision.check_not_reserved_id(revid)
102
assert inv.revision_id is None or inv.revision_id == revid, \
103
"Mismatch between inventory revision" \
104
" id and insertion revid (%r, %r)" % (inv.revision_id, revid)
105
assert inv.root is not None
106
inv_text = self.serialise_inventory(inv)
107
inv_sha1 = osutils.sha_string(inv_text)
108
inv_vf = self.control_weaves.get_weave('inventory',
109
self.get_transaction())
110
self._inventory_add_lines(inv_vf, revid, parents, osutils.split_lines(inv_text))
113
def _inventory_add_lines(self, inv_vf, revid, parents, lines):
115
for parent in parents:
117
final_parents.append(parent)
119
inv_vf.add_lines(revid, final_parents, lines)
122
def add_revision(self, rev_id, rev, inv=None, config=None):
123
"""Add rev to the revision store as rev_id.
125
:param rev_id: the revision id to use.
1095
126
:param rev: The revision object.
1096
127
:param inv: The inventory for the revision. if None, it will be looked
1097
128
up in the inventory storer
1163
188
Returns a set of the present revisions.
1166
graph = self.get_graph()
1167
parent_map = graph.get_parent_map(revision_ids)
1168
# The old API returned a list, should this actually be a set?
1169
return parent_map.keys()
1171
def _check_inventories(self, checker):
1172
"""Check the inventories found from the revision scan.
1174
This is responsible for verifying the sha1 of inventories and
1175
creating a pending_keys set that covers data referenced by inventories.
1177
bar = ui.ui_factory.nested_progress_bar()
1179
self._do_check_inventories(checker, bar)
1183
def _do_check_inventories(self, checker, bar):
1184
"""Helper for _check_inventories."""
1186
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1187
kinds = ['chk_bytes', 'texts']
1188
count = len(checker.pending_keys)
1189
bar.update("inventories", 0, 2)
1190
current_keys = checker.pending_keys
1191
checker.pending_keys = {}
1192
# Accumulate current checks.
1193
for key in current_keys:
1194
if key[0] != 'inventories' and key[0] not in kinds:
1195
checker._report_items.append('unknown key type %r' % (key,))
1196
keys[key[0]].add(key[1:])
1197
if keys['inventories']:
1198
# NB: output order *should* be roughly sorted - topo or
1199
# inverse topo depending on repository - either way decent
1200
# to just delta against. However, pre-CHK formats didn't
1201
# try to optimise inventory layout on disk. As such the
1202
# pre-CHK code path does not use inventory deltas.
1204
for record in self.inventories.check(keys=keys['inventories']):
1205
if record.storage_kind == 'absent':
1206
checker._report_items.append(
1207
'Missing inventory {%s}' % (record.key,))
1209
last_object = self._check_record('inventories', record,
1210
checker, last_object,
1211
current_keys[('inventories',) + record.key])
1212
del keys['inventories']
1215
bar.update("texts", 1)
1216
while (checker.pending_keys or keys['chk_bytes']
1218
# Something to check.
1219
current_keys = checker.pending_keys
1220
checker.pending_keys = {}
1221
# Accumulate current checks.
1222
for key in current_keys:
1223
if key[0] not in kinds:
1224
checker._report_items.append('unknown key type %r' % (key,))
1225
keys[key[0]].add(key[1:])
1226
# Check the outermost kind only - inventories || chk_bytes || texts
1230
for record in getattr(self, kind).check(keys=keys[kind]):
1231
if record.storage_kind == 'absent':
1232
checker._report_items.append(
1233
'Missing %s {%s}' % (kind, record.key,))
1235
last_object = self._check_record(kind, record,
1236
checker, last_object, current_keys[(kind,) + record.key])
1240
def _check_record(self, kind, record, checker, last_object, item_data):
1241
"""Check a single text from this repository."""
1242
if kind == 'inventories':
1243
rev_id = record.key[0]
1244
inv = self._deserialise_inventory(rev_id,
1245
record.get_bytes_as('fulltext'))
1246
if last_object is not None:
1247
delta = inv._make_delta(last_object)
1248
for old_path, path, file_id, ie in delta:
1251
ie.check(checker, rev_id, inv)
1253
for path, ie in inv.iter_entries():
1254
ie.check(checker, rev_id, inv)
1255
if self._format.fast_deltas:
1257
elif kind == 'chk_bytes':
1258
# No code written to check chk_bytes for this repo format.
1259
checker._report_items.append(
1260
'unsupported key type chk_bytes for %s' % (record.key,))
1261
elif kind == 'texts':
1262
self._check_text(record, checker, item_data)
1264
checker._report_items.append(
1265
'unknown key type %s for %s' % (kind, record.key))
1267
def _check_text(self, record, checker, item_data):
1268
"""Check a single text."""
1269
# Check it is extractable.
1270
# TODO: check length.
1271
if record.storage_kind == 'chunked':
1272
chunks = record.get_bytes_as(record.storage_kind)
1273
sha1 = osutils.sha_strings(chunks)
1274
length = sum(map(len, chunks))
1276
content = record.get_bytes_as('fulltext')
1277
sha1 = osutils.sha_string(content)
1278
length = len(content)
1279
if item_data and sha1 != item_data[1]:
1280
checker._report_items.append(
1281
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1282
(record.key, sha1, item_data[1], item_data[2]))
191
for id in revision_ids:
192
if self.has_revision(id):
1285
197
def create(a_bzrdir):
1286
198
"""Construct the current default format repository in a_bzrdir."""
1287
199
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1289
def __init__(self, _format, a_bzrdir, control_files):
201
def __init__(self, _format, a_bzrdir, control_files, _revision_store, control_store, text_store):
1290
202
"""instantiate a Repository.
1292
204
:param _format: The format of the repository on disk.
1293
205
:param a_bzrdir: The BzrDir of the repository.
207
In the future we will have a single api for all stores for
208
getting file texts, inventories and revisions, then
209
this construct will accept instances of those things.
1295
# In the future we will have a single api for all stores for
1296
# getting file texts, inventories and revisions, then
1297
# this construct will accept instances of those things.
1298
211
super(Repository, self).__init__()
1299
212
self._format = _format
1300
213
# the following are part of the public API for Repository:
1301
214
self.bzrdir = a_bzrdir
1302
215
self.control_files = control_files
1303
self._transport = control_files._transport
1304
self.base = self._transport.base
1306
self._reconcile_does_inventory_gc = True
1307
self._reconcile_fixes_text_parents = False
1308
self._reconcile_backsup_inventory = True
1309
self._write_group = None
1310
# Additional places to query for data.
1311
self._fallback_repositories = []
1312
# An InventoryEntry cache, used during deserialization
1313
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1314
# Is it safe to return inventory entries directly from the entry cache,
1315
# rather copying them?
1316
self._safe_to_return_from_cache = False
1319
def user_transport(self):
1320
return self.bzrdir.user_transport
1323
def control_transport(self):
1324
return self._transport
216
self._revision_store = _revision_store
217
self.text_store = text_store
218
# backwards compatibility
219
self.weave_store = text_store
220
# not right yet - should be more semantically clear ?
222
self.control_store = control_store
223
self.control_weaves = control_store
224
# TODO: make sure to construct the right store classes, etc, depending
225
# on whether escaping is required.
226
self._warn_if_deprecated()
227
self._serializer = xml5.serializer_v5
1326
229
def __repr__(self):
1327
if self._fallback_repositories:
1328
return '%s(%r, fallback_repositories=%r)' % (
1329
self.__class__.__name__,
1331
self._fallback_repositories)
1333
return '%s(%r)' % (self.__class__.__name__,
1336
def _has_same_fallbacks(self, other_repo):
1337
"""Returns true if the repositories have the same fallbacks."""
1338
my_fb = self._fallback_repositories
1339
other_fb = other_repo._fallback_repositories
1340
if len(my_fb) != len(other_fb):
1342
for f, g in zip(my_fb, other_fb):
1343
if not f.has_same_location(g):
1347
def has_same_location(self, other):
1348
"""Returns a boolean indicating if this repository is at the same
1349
location as another repository.
1351
This might return False even when two repository objects are accessing
1352
the same physical repository via different URLs.
1354
if self.__class__ is not other.__class__:
1356
return (self._transport.base == other._transport.base)
1358
def is_in_write_group(self):
1359
"""Return True if there is an open write group.
1361
:seealso: start_write_group.
1363
return self._write_group is not None
230
return '%s(%r)' % (self.__class__.__name__,
231
self.bzrdir.transport.base)
1365
233
def is_locked(self):
1366
234
return self.control_files.is_locked()
1368
def is_write_locked(self):
1369
"""Return True if this object is write locked."""
1370
return self.is_locked() and self.control_files._lock_mode == 'w'
1372
def lock_write(self, token=None):
1373
"""Lock this repository for writing.
1375
This causes caching within the repository obejct to start accumlating
1376
data during reads, and allows a 'write_group' to be obtained. Write
1377
groups must be used for actual data insertion.
1379
:param token: if this is already locked, then lock_write will fail
1380
unless the token matches the existing lock.
1381
:returns: a token if this instance supports tokens, otherwise None.
1382
:raises TokenLockingNotSupported: when a token is given but this
1383
instance doesn't support using token locks.
1384
:raises MismatchedToken: if the specified token doesn't match the token
1385
of the existing lock.
1386
:seealso: start_write_group.
1388
A token should be passed in if you know that you have locked the object
1389
some other way, and need to synchronise this object's state with that
1392
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1394
locked = self.is_locked()
1395
result = self.control_files.lock_write(token=token)
1397
self._warn_if_deprecated()
1398
self._note_lock('w')
1399
for repo in self._fallback_repositories:
1400
# Writes don't affect fallback repos
1402
self._refresh_data()
236
def lock_write(self):
237
self.control_files.lock_write()
1405
239
def lock_read(self):
1406
locked = self.is_locked()
1407
240
self.control_files.lock_read()
1409
self._warn_if_deprecated()
1410
self._note_lock('r')
1411
for repo in self._fallback_repositories:
1413
self._refresh_data()
1415
242
def get_physical_lock_status(self):
1416
243
return self.control_files.get_physical_lock_status()
1418
def leave_lock_in_place(self):
1419
"""Tell this repository not to release the physical lock when this
1422
If lock_write doesn't return a token, then this method is not supported.
1424
self.control_files.leave_in_place()
1426
def dont_leave_lock_in_place(self):
1427
"""Tell this repository to release the physical lock when this
1428
object is unlocked, even if it didn't originally acquire it.
1430
If lock_write doesn't return a token, then this method is not supported.
1432
self.control_files.dont_leave_in_place()
1435
def gather_stats(self, revid=None, committers=None):
1436
"""Gather statistics from a revision id.
1438
:param revid: The revision id to gather statistics from, if None, then
1439
no revision specific statistics are gathered.
1440
:param committers: Optional parameter controlling whether to grab
1441
a count of committers from the revision specific statistics.
1442
:return: A dictionary of statistics. Currently this contains:
1443
committers: The number of committers if requested.
1444
firstrev: A tuple with timestamp, timezone for the penultimate left
1445
most ancestor of revid, if revid is not the NULL_REVISION.
1446
latestrev: A tuple with timestamp, timezone for revid, if revid is
1447
not the NULL_REVISION.
1448
revisions: The total revision count in the repository.
1449
size: An estimate disk size of the repository in bytes.
1452
if revid and committers:
1453
result['committers'] = 0
1454
if revid and revid != _mod_revision.NULL_REVISION:
1456
all_committers = set()
1457
revisions = self.get_ancestry(revid)
1458
# pop the leading None
1460
first_revision = None
1462
# ignore the revisions in the middle - just grab first and last
1463
revisions = revisions[0], revisions[-1]
1464
for revision in self.get_revisions(revisions):
1465
if not first_revision:
1466
first_revision = revision
1468
all_committers.add(revision.committer)
1469
last_revision = revision
1471
result['committers'] = len(all_committers)
1472
result['firstrev'] = (first_revision.timestamp,
1473
first_revision.timezone)
1474
result['latestrev'] = (last_revision.timestamp,
1475
last_revision.timezone)
1477
# now gather global repository information
1478
# XXX: This is available for many repos regardless of listability.
1479
if self.user_transport.listable():
1480
# XXX: do we want to __define len__() ?
1481
# Maybe the versionedfiles object should provide a different
1482
# method to get the number of keys.
1483
result['revisions'] = len(self.revisions.keys())
1484
# result['size'] = t
1487
def find_branches(self, using=False):
1488
"""Find branches underneath this repository.
1490
This will include branches inside other branches.
1492
:param using: If True, list only branches using this repository.
1494
if using and not self.is_shared():
1495
return self.bzrdir.list_branches()
1496
class Evaluator(object):
1499
self.first_call = True
1501
def __call__(self, bzrdir):
1502
# On the first call, the parameter is always the bzrdir
1503
# containing the current repo.
1504
if not self.first_call:
1506
repository = bzrdir.open_repository()
1507
except errors.NoRepositoryPresent:
1510
return False, ([], repository)
1511
self.first_call = False
1512
value = (bzrdir.list_branches(), None)
1516
for branches, repository in bzrdir.BzrDir.find_bzrdirs(
1517
self.user_transport, evaluate=Evaluator()):
1518
if branches is not None:
1519
ret.extend(branches)
1520
if not using and repository is not None:
1521
ret.extend(repository.find_branches())
1525
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
246
def missing_revision_ids(self, other, revision_id=None):
1526
247
"""Return the revision ids that other has that this does not.
1528
249
These are returned in topological order.
1530
251
revision_id: only return revision ids included by revision_id.
1532
return InterRepository.get(other, self).search_missing_revision_ids(
1533
revision_id, find_ghosts)
253
return InterRepository.get(other, self).missing_revision_ids(revision_id)
1808
314
# Most control formats need the repository to be specifically
1809
315
# created, but on some old all-in-one formats it's not needed
1811
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
317
dest_repo = self._format.initialize(a_bzrdir, shared=self.is_shared())
1812
318
except errors.UninitializableFormat:
1813
319
dest_repo = a_bzrdir.open_repository()
320
self.copy_content_into(dest_repo, revision_id, basis)
1814
321
return dest_repo
1816
def _get_sink(self):
1817
"""Return a sink for streaming into this repository."""
1818
return StreamSink(self)
1820
def _get_source(self, to_format):
1821
"""Return a source for streaming from this repository."""
1822
return StreamSource(self, to_format)
1824
323
@needs_read_lock
1825
324
def has_revision(self, revision_id):
1826
325
"""True if this repository has a copy of the revision."""
1827
return revision_id in self.has_revisions((revision_id,))
1830
def has_revisions(self, revision_ids):
1831
"""Probe to find out the presence of multiple revisions.
1833
:param revision_ids: An iterable of revision_ids.
1834
:return: A set of the revision_ids that were present.
1836
parent_map = self.revisions.get_parent_map(
1837
[(rev_id,) for rev_id in revision_ids])
1839
if _mod_revision.NULL_REVISION in revision_ids:
1840
result.add(_mod_revision.NULL_REVISION)
1841
result.update([key[0] for key in parent_map])
1845
def get_revision(self, revision_id):
1846
"""Return the Revision object for a named revision."""
1847
return self.get_revisions([revision_id])[0]
326
return self._revision_store.has_revision_id(revision_id,
327
self.get_transaction())
1849
329
@needs_read_lock
1850
330
def get_revision_reconcile(self, revision_id):
1851
331
"""'reconcile' helper routine that allows access to a revision always.
1853
333
This variant of get_revision does not cross check the weave graph
1854
334
against the revision one as get_revision does: but it should only
1855
335
be used by reconcile, or reconcile-alike commands that are correcting
1856
336
or testing the revision graph.
1858
return self._get_revisions([revision_id])[0]
338
if not revision_id or not isinstance(revision_id, basestring):
339
raise errors.InvalidRevisionId(revision_id=revision_id,
341
return self._revision_store.get_revisions([revision_id],
342
self.get_transaction())[0]
1860
343
@needs_read_lock
1861
344
def get_revisions(self, revision_ids):
1862
"""Get many revisions at once.
1864
Repositories that need to check data on every revision read should
1865
subclass this method.
1867
return self._get_revisions(revision_ids)
1870
def _get_revisions(self, revision_ids):
1871
"""Core work logic to get many revisions without sanity checks."""
1873
for revid, rev in self._iter_revisions(revision_ids):
1875
raise errors.NoSuchRevision(self, revid)
1877
return [revs[revid] for revid in revision_ids]
1879
def _iter_revisions(self, revision_ids):
1880
"""Iterate over revision objects.
1882
:param revision_ids: An iterable of revisions to examine. None may be
1883
passed to request all revisions known to the repository. Note that
1884
not all repositories can find unreferenced revisions; for those
1885
repositories only referenced ones will be returned.
1886
:return: An iterator of (revid, revision) tuples. Absent revisions (
1887
those asked for but not available) are returned as (revid, None).
1889
if revision_ids is None:
1890
revision_ids = self.all_revision_ids()
1892
for rev_id in revision_ids:
1893
if not rev_id or not isinstance(rev_id, basestring):
1894
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1895
keys = [(key,) for key in revision_ids]
1896
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1897
for record in stream:
1898
revid = record.key[0]
1899
if record.storage_kind == 'absent':
1902
text = record.get_bytes_as('fulltext')
1903
rev = self._serializer.read_revision_from_string(text)
1906
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
345
return self._revision_store.get_revisions(revision_ids,
346
self.get_transaction())
349
def get_revision_xml(self, revision_id):
350
rev = self.get_revision(revision_id)
352
# the current serializer..
353
self._revision_store._serializer.write_revision(rev, rev_tmp)
355
return rev_tmp.getvalue()
358
def get_revision(self, revision_id):
359
"""Return the Revision object for a named revision"""
360
r = self.get_revision_reconcile(revision_id)
361
# weave corruption can lead to absent revision markers that should be
363
# the following test is reasonably cheap (it needs a single weave read)
364
# and the weave is cached in read transactions. In write transactions
365
# it is not cached but typically we only read a small number of
366
# revisions. For knits when they are introduced we will probably want
367
# to ensure that caching write transactions are in use.
368
inv = self.get_inventory_weave()
369
self._check_revision_parents(r, inv)
373
def get_deltas_for_revisions(self, revisions):
1907
374
"""Produce a generator of revision deltas.
1909
376
Note that the input is a sequence of REVISIONS, not revision_ids.
1910
377
Trees will be held in memory until the generator exits.
1911
378
Each delta is relative to the revision's lefthand predecessor.
1913
:param specific_fileids: if not None, the result is filtered
1914
so that only those file-ids, their parents and their
1915
children are included.
1917
# Get the revision-ids of interest
1918
380
required_trees = set()
1919
381
for revision in revisions:
1920
382
required_trees.add(revision.revision_id)
1921
383
required_trees.update(revision.parent_ids[:1])
1923
# Get the matching filtered trees. Note that it's more
1924
# efficient to pass filtered trees to changes_from() rather
1925
# than doing the filtering afterwards. changes_from() could
1926
# arguably do the filtering itself but it's path-based, not
1927
# file-id based, so filtering before or afterwards is
1929
if specific_fileids is None:
1930
trees = dict((t.get_revision_id(), t) for
1931
t in self.revision_trees(required_trees))
1933
trees = dict((t.get_revision_id(), t) for
1934
t in self._filtered_revision_trees(required_trees,
1937
# Calculate the deltas
384
trees = dict((t.get_revision_id(), t) for
385
t in self.revision_trees(required_trees))
1938
386
for revision in revisions:
1939
387
if not revision.parent_ids:
1940
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
388
old_tree = self.revision_tree(None)
1942
390
old_tree = trees[revision.parent_ids[0]]
1943
391
yield trees[revision.revision_id].changes_from(old_tree)
1945
393
@needs_read_lock
1946
def get_revision_delta(self, revision_id, specific_fileids=None):
394
def get_revision_delta(self, revision_id):
1947
395
"""Return the delta for one revision.
1949
397
The delta is relative to the left-hand predecessor of the
1952
:param specific_fileids: if not None, the result is filtered
1953
so that only those file-ids, their parents and their
1954
children are included.
1956
400
r = self.get_revision(revision_id)
1957
return list(self.get_deltas_for_revisions([r],
1958
specific_fileids=specific_fileids))[0]
401
return list(self.get_deltas_for_revisions([r]))[0]
403
def _check_revision_parents(self, revision, inventory):
404
"""Private to Repository and Fetch.
406
This checks the parentage of revision in an inventory weave for
407
consistency and is only applicable to inventory-weave-for-ancestry
408
using repository formats & fetchers.
410
weave_parents = inventory.get_parents(revision.revision_id)
411
weave_names = inventory.versions()
412
for parent_id in revision.parent_ids:
413
if parent_id in weave_names:
414
# this parent must not be a ghost.
415
if not parent_id in weave_parents:
417
raise errors.CorruptRepository(self)
1960
419
@needs_write_lock
1961
420
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1962
421
signature = gpg_strategy.sign(plaintext)
1963
self.add_signature_text(revision_id, signature)
1966
def add_signature_text(self, revision_id, signature):
1967
self.signatures.add_lines((revision_id,), (),
1968
osutils.split_lines(signature))
1970
def find_text_key_references(self):
1971
"""Find the text key references within the repository.
1973
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1974
to whether they were referred to by the inventory of the
1975
revision_id that they contain. The inventory texts from all present
1976
revision ids are assessed to generate this report.
1978
revision_keys = self.revisions.keys()
1979
w = self.inventories
1980
pb = ui.ui_factory.nested_progress_bar()
1982
return self._find_text_key_references_from_xml_inventory_lines(
1983
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1987
def _find_text_key_references_from_xml_inventory_lines(self,
1989
"""Core routine for extracting references to texts from inventories.
1991
This performs the translation of xml lines to revision ids.
1993
:param line_iterator: An iterator of lines, origin_version_id
1994
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1995
to whether they were referred to by the inventory of the
1996
revision_id that they contain. Note that if that revision_id was
1997
not part of the line_iterator's output then False will be given -
1998
even though it may actually refer to that key.
2000
if not self._serializer.support_altered_by_hack:
2001
raise AssertionError(
2002
"_find_text_key_references_from_xml_inventory_lines only "
2003
"supported for branches which store inventory as unnested xml"
2004
", not on %r" % self)
422
self._revision_store.add_revision_signature_text(revision_id,
424
self.get_transaction())
426
def fileids_altered_by_revision_ids(self, revision_ids):
427
"""Find the file ids and versions affected by revisions.
429
:param revisions: an iterable containing revision ids.
430
:return: a dictionary mapping altered file-ids to an iterable of
431
revision_ids. Each altered file-ids has the exact revision_ids that
432
altered it listed explicitly.
434
assert self._serializer.support_altered_by_hack, \
435
("fileids_altered_by_revision_ids only supported for branches "
436
"which store inventory as unnested xml, not on %r" % self)
437
selected_revision_ids = set(revision_ids)
438
w = self.get_inventory_weave()
2007
441
# this code needs to read every new line in every inventory for the
2008
442
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2009
# not present in one of those inventories is unnecessary but not
443
# not present in one of those inventories is unnecessary but not
2010
444
# harmful because we are filtering by the revision id marker in the
2011
# inventory lines : we only select file ids altered in one of those
445
# inventory lines : we only select file ids altered in one of those
2012
446
# revisions. We don't need to see all lines in the inventory because
2013
447
# only those added in an inventory in rev X can contain a revision=X
2023
457
search = self._file_ids_altered_regex.search
2024
458
unescape = _unescape_xml
2025
459
setdefault = result.setdefault
2026
for line, line_key in line_iterator:
2027
match = search(line)
2030
# One call to match.group() returning multiple items is quite a
2031
# bit faster than 2 calls to match.group() each returning 1
2032
file_id, revision_id = match.group('file_id', 'revision_id')
2034
# Inlining the cache lookups helps a lot when you make 170,000
2035
# lines and 350k ids, versus 8.4 unique ids.
2036
# Using a cache helps in 2 ways:
2037
# 1) Avoids unnecessary decoding calls
2038
# 2) Re-uses cached strings, which helps in future set and
2040
# (2) is enough that removing encoding entirely along with
2041
# the cache (so we are using plain strings) results in no
2042
# performance improvement.
2044
revision_id = unescape_revid_cache[revision_id]
2046
unescaped = unescape(revision_id)
2047
unescape_revid_cache[revision_id] = unescaped
2048
revision_id = unescaped
2050
# Note that unconditionally unescaping means that we deserialise
2051
# every fileid, which for general 'pull' is not great, but we don't
2052
# really want to have some many fulltexts that this matters anyway.
2055
file_id = unescape_fileid_cache[file_id]
2057
unescaped = unescape(file_id)
2058
unescape_fileid_cache[file_id] = unescaped
2061
key = (file_id, revision_id)
2062
setdefault(key, False)
2063
if revision_id == line_key[-1]:
2067
def _inventory_xml_lines_for_keys(self, keys):
2068
"""Get a line iterator of the sort needed for findind references.
2070
Not relevant for non-xml inventory repositories.
2072
Ghosts in revision_keys are ignored.
2074
:param revision_keys: The revision keys for the inventories to inspect.
2075
:return: An iterator over (inventory line, revid) for the fulltexts of
2076
all of the xml inventories specified by revision_keys.
2078
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2079
for record in stream:
2080
if record.storage_kind != 'absent':
2081
chunks = record.get_bytes_as('chunked')
2082
revid = record.key[-1]
2083
lines = osutils.chunks_to_lines(chunks)
2087
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2089
"""Helper routine for fileids_altered_by_revision_ids.
2091
This performs the translation of xml lines to revision ids.
2093
:param line_iterator: An iterator of lines, origin_version_id
2094
:param revision_keys: The revision ids to filter for. This should be a
2095
set or other type which supports efficient __contains__ lookups, as
2096
the revision key from each parsed line will be looked up in the
2097
revision_keys filter.
2098
:return: a dictionary mapping altered file-ids to an iterable of
2099
revision_ids. Each altered file-ids has the exact revision_ids that
2100
altered it listed explicitly.
2102
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2103
line_iterator).iterkeys())
2104
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2105
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2106
self._inventory_xml_lines_for_keys(parent_keys)))
2107
new_keys = seen - parent_seen
2109
setdefault = result.setdefault
2110
for key in new_keys:
2111
setdefault(key[0], set()).add(key[-1])
2114
def _find_parent_ids_of_revisions(self, revision_ids):
2115
"""Find all parent ids that are mentioned in the revision graph.
2117
:return: set of revisions that are parents of revision_ids which are
2118
not part of revision_ids themselves
2120
parent_map = self.get_parent_map(revision_ids)
2122
map(parent_ids.update, parent_map.itervalues())
2123
parent_ids.difference_update(revision_ids)
2124
parent_ids.discard(_mod_revision.NULL_REVISION)
2127
def _find_parent_keys_of_revisions(self, revision_keys):
2128
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2130
:param revision_keys: An iterable of revision_keys.
2131
:return: The parents of all revision_keys that are not already in
2134
parent_map = self.revisions.get_parent_map(revision_keys)
2136
map(parent_keys.update, parent_map.itervalues())
2137
parent_keys.difference_update(revision_keys)
2138
parent_keys.discard(_mod_revision.NULL_REVISION)
2141
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2142
"""Find the file ids and versions affected by revisions.
2144
:param revisions: an iterable containing revision ids.
2145
:param _inv_weave: The inventory weave from this repository or None.
2146
If None, the inventory weave will be opened automatically.
2147
:return: a dictionary mapping altered file-ids to an iterable of
2148
revision_ids. Each altered file-ids has the exact revision_ids that
2149
altered it listed explicitly.
2151
selected_keys = set((revid,) for revid in revision_ids)
2152
w = _inv_weave or self.inventories
2153
return self._find_file_ids_from_xml_inventory_lines(
2154
w.iter_lines_added_or_present_in_keys(
2155
selected_keys, pb=None),
2158
def iter_files_bytes(self, desired_files):
2159
"""Iterate through file versions.
2161
Files will not necessarily be returned in the order they occur in
2162
desired_files. No specific order is guaranteed.
2164
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2165
value supplied by the caller as part of desired_files. It should
2166
uniquely identify the file version in the caller's context. (Examples:
2167
an index number or a TreeTransform trans_id.)
2169
bytes_iterator is an iterable of bytestrings for the file. The
2170
kind of iterable and length of the bytestrings are unspecified, but for
2171
this implementation, it is a list of bytes produced by
2172
VersionedFile.get_record_stream().
2174
:param desired_files: a list of (file_id, revision_id, identifier)
2178
for file_id, revision_id, callable_data in desired_files:
2179
text_keys[(file_id, revision_id)] = callable_data
2180
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2181
if record.storage_kind == 'absent':
2182
raise errors.RevisionNotPresent(record.key, self)
2183
yield text_keys[record.key], record.get_bytes_as('chunked')
2185
def _generate_text_key_index(self, text_key_references=None,
2187
"""Generate a new text key index for the repository.
2189
This is an expensive function that will take considerable time to run.
2191
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2192
list of parents, also text keys. When a given key has no parents,
2193
the parents list will be [NULL_REVISION].
2195
# All revisions, to find inventory parents.
2196
if ancestors is None:
2197
graph = self.get_graph()
2198
ancestors = graph.get_parent_map(self.all_revision_ids())
2199
if text_key_references is None:
2200
text_key_references = self.find_text_key_references()
2201
460
pb = ui.ui_factory.nested_progress_bar()
2203
return self._do_generate_text_key_index(ancestors,
2204
text_key_references, pb)
462
for line in w.iter_lines_added_or_present_in_versions(
463
selected_revision_ids, pb=pb):
467
# One call to match.group() returning multiple items is quite a
468
# bit faster than 2 calls to match.group() each returning 1
469
file_id, revision_id = match.group('file_id', 'revision_id')
471
# Inlining the cache lookups helps a lot when you make 170,000
472
# lines and 350k ids, versus 8.4 unique ids.
473
# Using a cache helps in 2 ways:
474
# 1) Avoids unnecessary decoding calls
475
# 2) Re-uses cached strings, which helps in future set and
477
# (2) is enough that removing encoding entirely along with
478
# the cache (so we are using plain strings) results in no
479
# performance improvement.
481
revision_id = unescape_revid_cache[revision_id]
483
unescaped = unescape(revision_id)
484
unescape_revid_cache[revision_id] = unescaped
485
revision_id = unescaped
487
if revision_id in selected_revision_ids:
489
file_id = unescape_fileid_cache[file_id]
491
unescaped = unescape(file_id)
492
unescape_fileid_cache[file_id] = unescaped
494
setdefault(file_id, set()).add(revision_id)
2208
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2209
"""Helper for _generate_text_key_index to avoid deep nesting."""
2210
revision_order = tsort.topo_sort(ancestors)
2211
invalid_keys = set()
2213
for revision_id in revision_order:
2214
revision_keys[revision_id] = set()
2215
text_count = len(text_key_references)
2216
# a cache of the text keys to allow reuse; costs a dict of all the
2217
# keys, but saves a 2-tuple for every child of a given key.
2219
for text_key, valid in text_key_references.iteritems():
2221
invalid_keys.add(text_key)
2223
revision_keys[text_key[1]].add(text_key)
2224
text_key_cache[text_key] = text_key
2225
del text_key_references
2227
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2228
NULL_REVISION = _mod_revision.NULL_REVISION
2229
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2230
# too small for large or very branchy trees. However, for 55K path
2231
# trees, it would be easy to use too much memory trivially. Ideally we
2232
# could gauge this by looking at available real memory etc, but this is
2233
# always a tricky proposition.
2234
inventory_cache = lru_cache.LRUCache(10)
2235
batch_size = 10 # should be ~150MB on a 55K path tree
2236
batch_count = len(revision_order) / batch_size + 1
2238
pb.update("Calculating text parents", processed_texts, text_count)
2239
for offset in xrange(batch_count):
2240
to_query = revision_order[offset * batch_size:(offset + 1) *
2244
for revision_id in to_query:
2245
parent_ids = ancestors[revision_id]
2246
for text_key in revision_keys[revision_id]:
2247
pb.update("Calculating text parents", processed_texts)
2248
processed_texts += 1
2249
candidate_parents = []
2250
for parent_id in parent_ids:
2251
parent_text_key = (text_key[0], parent_id)
2253
check_parent = parent_text_key not in \
2254
revision_keys[parent_id]
2256
# the parent parent_id is a ghost:
2257
check_parent = False
2258
# truncate the derived graph against this ghost.
2259
parent_text_key = None
2261
# look at the parent commit details inventories to
2262
# determine possible candidates in the per file graph.
2265
inv = inventory_cache[parent_id]
2267
inv = self.revision_tree(parent_id).inventory
2268
inventory_cache[parent_id] = inv
2270
parent_entry = inv[text_key[0]]
2271
except (KeyError, errors.NoSuchId):
2273
if parent_entry is not None:
2275
text_key[0], parent_entry.revision)
2277
parent_text_key = None
2278
if parent_text_key is not None:
2279
candidate_parents.append(
2280
text_key_cache[parent_text_key])
2281
parent_heads = text_graph.heads(candidate_parents)
2282
new_parents = list(parent_heads)
2283
new_parents.sort(key=lambda x:candidate_parents.index(x))
2284
if new_parents == []:
2285
new_parents = [NULL_REVISION]
2286
text_index[text_key] = new_parents
2288
for text_key in invalid_keys:
2289
text_index[text_key] = [NULL_REVISION]
2292
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2293
"""Get an iterable listing the keys of all the data introduced by a set
2296
The keys will be ordered so that the corresponding items can be safely
2297
fetched and inserted in that order.
2299
:returns: An iterable producing tuples of (knit-kind, file-id,
2300
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2301
'revisions'. file-id is None unless knit-kind is 'file'.
2303
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2306
for result in self._find_non_file_keys_to_fetch(revision_ids):
2309
def _find_file_keys_to_fetch(self, revision_ids, pb):
2310
# XXX: it's a bit weird to control the inventory weave caching in this
2311
# generator. Ideally the caching would be done in fetch.py I think. Or
2312
# maybe this generator should explicitly have the contract that it
2313
# should not be iterated until the previously yielded item has been
2315
inv_w = self.inventories
2317
# file ids that changed
2318
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2320
num_file_ids = len(file_ids)
2321
for file_id, altered_versions in file_ids.iteritems():
2323
pb.update("Fetch texts", count, num_file_ids)
2325
yield ("file", file_id, altered_versions)
2327
def _find_non_file_keys_to_fetch(self, revision_ids):
2329
yield ("inventory", None, revision_ids)
2332
# XXX: Note ATM no callers actually pay attention to this return
2333
# instead they just use the list of revision ids and ignore
2334
# missing sigs. Consider removing this work entirely
2335
revisions_with_signatures = set(self.signatures.get_parent_map(
2336
[(r,) for r in revision_ids]))
2337
revisions_with_signatures = set(
2338
[r for (r,) in revisions_with_signatures])
2339
revisions_with_signatures.intersection_update(revision_ids)
2340
yield ("signatures", None, revisions_with_signatures)
2343
yield ("revisions", None, revision_ids)
500
def get_inventory_weave(self):
501
return self.control_weaves.get_weave('inventory',
502
self.get_transaction())
2345
504
@needs_read_lock
2346
505
def get_inventory(self, revision_id):
2347
"""Get Inventory object by revision id."""
2348
return self.iter_inventories([revision_id]).next()
2350
def iter_inventories(self, revision_ids, ordering=None):
2351
"""Get many inventories by revision_ids.
2353
This will buffer some or all of the texts used in constructing the
2354
inventories in memory, but will only parse a single inventory at a
2357
:param revision_ids: The expected revision ids of the inventories.
2358
:param ordering: optional ordering, e.g. 'topological'. If not
2359
specified, the order of revision_ids will be preserved (by
2360
buffering if necessary).
2361
:return: An iterator of inventories.
2363
if ((None in revision_ids)
2364
or (_mod_revision.NULL_REVISION in revision_ids)):
2365
raise ValueError('cannot get null revision inventory')
2366
return self._iter_inventories(revision_ids, ordering)
2368
def _iter_inventories(self, revision_ids, ordering):
2369
"""single-document based inventory iteration."""
2370
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2371
for text, revision_id in inv_xmls:
2372
yield self._deserialise_inventory(revision_id, text)
2374
def _iter_inventory_xmls(self, revision_ids, ordering):
2375
if ordering is None:
2376
order_as_requested = True
2377
ordering = 'unordered'
2379
order_as_requested = False
2380
keys = [(revision_id,) for revision_id in revision_ids]
2383
if order_as_requested:
2384
key_iter = iter(keys)
2385
next_key = key_iter.next()
2386
stream = self.inventories.get_record_stream(keys, ordering, True)
2388
for record in stream:
2389
if record.storage_kind != 'absent':
2390
chunks = record.get_bytes_as('chunked')
2391
if order_as_requested:
2392
text_chunks[record.key] = chunks
2394
yield ''.join(chunks), record.key[-1]
2396
raise errors.NoSuchRevision(self, record.key)
2397
if order_as_requested:
2398
# Yield as many results as we can while preserving order.
2399
while next_key in text_chunks:
2400
chunks = text_chunks.pop(next_key)
2401
yield ''.join(chunks), next_key[-1]
2403
next_key = key_iter.next()
2404
except StopIteration:
2405
# We still want to fully consume the get_record_stream,
2406
# just in case it is not actually finished at this point
2410
def _deserialise_inventory(self, revision_id, xml):
2411
"""Transform the xml into an inventory object.
506
"""Get Inventory object by hash."""
507
return self.deserialise_inventory(
508
revision_id, self.get_inventory_xml(revision_id))
510
def deserialise_inventory(self, revision_id, xml):
511
"""Transform the xml into an inventory object.
2413
513
:param revision_id: The expected revision id of the inventory.
2414
514
:param xml: A serialised inventory.
2416
result = self._serializer.read_inventory_from_string(xml, revision_id,
2417
entry_cache=self._inventory_entry_cache,
2418
return_from_cache=self._safe_to_return_from_cache)
2419
if result.revision_id != revision_id:
2420
raise AssertionError('revision id mismatch %s != %s' % (
2421
result.revision_id, revision_id))
516
result = self._serializer.read_inventory_from_string(xml)
517
result.root.revision = revision_id
2424
def get_serializer_format(self):
2425
return self._serializer.format_num
520
def serialise_inventory(self, inv):
521
return self._serializer.write_inventory_to_string(inv)
2427
523
@needs_read_lock
2428
def _get_inventory_xml(self, revision_id):
2429
"""Get serialized inventory as a string."""
2430
texts = self._iter_inventory_xmls([revision_id], 'unordered')
524
def get_inventory_xml(self, revision_id):
525
"""Get inventory XML as a file object."""
2432
text, revision_id = texts.next()
2433
except StopIteration:
527
assert isinstance(revision_id, basestring), type(revision_id)
528
iw = self.get_inventory_weave()
529
return iw.get_text(revision_id)
2434
531
raise errors.HistoryMissing(self, 'inventory', revision_id)
2437
def get_rev_id_for_revno(self, revno, known_pair):
2438
"""Return the revision id of a revno, given a later (revno, revid)
2439
pair in the same history.
2441
:return: if found (True, revid). If the available history ran out
2442
before reaching the revno, then this returns
2443
(False, (closest_revno, closest_revid)).
2445
known_revno, known_revid = known_pair
2446
partial_history = [known_revid]
2447
distance_from_known = known_revno - revno
2448
if distance_from_known < 0:
2450
'requested revno (%d) is later than given known revno (%d)'
2451
% (revno, known_revno))
2454
self, partial_history, stop_index=distance_from_known)
2455
except errors.RevisionNotPresent, err:
2456
if err.revision_id == known_revid:
2457
# The start revision (known_revid) wasn't found.
2459
# This is a stacked repository with no fallbacks, or a there's a
2460
# left-hand ghost. Either way, even though the revision named in
2461
# the error isn't in this repo, we know it's the next step in this
2462
# left-hand history.
2463
partial_history.append(err.revision_id)
2464
if len(partial_history) <= distance_from_known:
2465
# Didn't find enough history to get a revid for the revno.
2466
earliest_revno = known_revno - len(partial_history) + 1
2467
return (False, (earliest_revno, partial_history[-1]))
2468
if len(partial_history) - 1 > distance_from_known:
2469
raise AssertionError('_iter_for_revno returned too much history')
2470
return (True, partial_history[-1])
2472
def iter_reverse_revision_history(self, revision_id):
2473
"""Iterate backwards through revision ids in the lefthand history
2475
:param revision_id: The revision id to start with. All its lefthand
2476
ancestors will be traversed.
2478
graph = self.get_graph()
2479
next_id = revision_id
2481
if next_id in (None, _mod_revision.NULL_REVISION):
534
def get_inventory_sha1(self, revision_id):
535
"""Return the sha1 hash of the inventory entry
537
return self.get_revision(revision_id).inventory_sha1
540
def get_revision_graph(self, revision_id=None):
541
"""Return a dictionary containing the revision graph.
543
:param revision_id: The revision_id to get a graph from. If None, then
544
the entire revision graph is returned. This is a deprecated mode of
545
operation and will be removed in the future.
546
:return: a dictionary of revision_id->revision_parents_list.
548
# special case NULL_REVISION
549
if revision_id == _mod_revision.NULL_REVISION:
551
a_weave = self.get_inventory_weave()
552
all_revisions = self._eliminate_revisions_not_present(
554
entire_graph = dict([(node, a_weave.get_parents(node)) for
555
node in all_revisions])
556
if revision_id is None:
558
elif revision_id not in entire_graph:
559
raise errors.NoSuchRevision(self, revision_id)
561
# add what can be reached from revision_id
563
pending = set([revision_id])
564
while len(pending) > 0:
566
result[node] = entire_graph[node]
567
for revision_id in result[node]:
568
if revision_id not in result:
569
pending.add(revision_id)
573
def get_revision_graph_with_ghosts(self, revision_ids=None):
574
"""Return a graph of the revisions with ghosts marked as applicable.
576
:param revision_ids: an iterable of revisions to graph or None for all.
577
:return: a Graph object with the graph reachable from revision_ids.
579
result = graph.Graph()
581
pending = set(self.all_revision_ids())
584
pending = set(revision_ids)
585
# special case NULL_REVISION
586
if _mod_revision.NULL_REVISION in pending:
587
pending.remove(_mod_revision.NULL_REVISION)
588
required = set(pending)
591
revision_id = pending.pop()
2484
parents = graph.get_parent_map([next_id])[next_id]
2486
raise errors.RevisionNotPresent(next_id, self)
2488
if len(parents) == 0:
2491
next_id = parents[0]
593
rev = self.get_revision(revision_id)
594
except errors.NoSuchRevision:
595
if revision_id in required:
598
result.add_ghost(revision_id)
600
for parent_id in rev.parent_ids:
601
# is this queued or done ?
602
if (parent_id not in pending and
603
parent_id not in done):
605
pending.add(parent_id)
606
result.add_node(revision_id, rev.parent_ids)
607
done.add(revision_id)
611
def get_revision_inventory(self, revision_id):
612
"""Return inventory of a past revision."""
613
# TODO: Unify this with get_inventory()
614
# bzr 0.0.6 and later imposes the constraint that the inventory_id
615
# must be the same as its revision, so this is trivial.
616
if revision_id is None:
617
# This does not make sense: if there is no revision,
618
# then it is the current tree inventory surely ?!
619
# and thus get_root_id() is something that looks at the last
620
# commit on the branch, and the get_root_id is an inventory check.
621
raise NotImplementedError
622
# return Inventory(self.get_root_id())
624
return self.get_inventory(revision_id)
2493
627
def is_shared(self):
2494
628
"""Return True if this repository is flagged as a shared repository."""
2495
629
raise NotImplementedError(self.is_shared)
2501
635
reconciler = RepoReconciler(self, thorough=thorough)
2502
636
reconciler.reconcile()
2503
637
return reconciler
2505
def _refresh_data(self):
2506
"""Helper called from lock_* to ensure coherency with disk.
2508
The default implementation does nothing; it is however possible
2509
for repositories to maintain loaded indices across multiple locks
2510
by checking inside their implementation of this method to see
2511
whether their indices are still valid. This depends of course on
2512
the disk format being validatable in this manner. This method is
2513
also called by the refresh_data() public interface to cause a refresh
2514
to occur while in a write lock so that data inserted by a smart server
2515
push operation is visible on the client's instance of the physical
2519
639
@needs_read_lock
2520
640
def revision_tree(self, revision_id):
2521
641
"""Return Tree for a revision on this branch.
2523
`revision_id` may be NULL_REVISION for the empty tree revision.
643
`revision_id` may be None for the empty tree revision.
2525
revision_id = _mod_revision.ensure_null(revision_id)
2526
645
# TODO: refactor this to use an existing revision object
2527
646
# so we don't need to read it in twice.
2528
if revision_id == _mod_revision.NULL_REVISION:
2529
return RevisionTree(self, Inventory(root_id=None),
647
if revision_id is None or revision_id == _mod_revision.NULL_REVISION:
648
return RevisionTree(self, Inventory(root_id=None),
2530
649
_mod_revision.NULL_REVISION)
2532
inv = self.get_inventory(revision_id)
651
inv = self.get_revision_inventory(revision_id)
2533
652
return RevisionTree(self, inv, revision_id)
2535
655
def revision_trees(self, revision_ids):
2536
"""Return Trees for revisions in this repository.
2538
:param revision_ids: a sequence of revision-ids;
2539
a revision-id may not be None or 'null:'
2541
inventories = self.iter_inventories(revision_ids)
2542
for inv in inventories:
2543
yield RevisionTree(self, inv, inv.revision_id)
2545
def _filtered_revision_trees(self, revision_ids, file_ids):
2546
"""Return Tree for a revision on this branch with only some files.
2548
:param revision_ids: a sequence of revision-ids;
2549
a revision-id may not be None or 'null:'
2550
:param file_ids: if not None, the result is filtered
2551
so that only those file-ids, their parents and their
2552
children are included.
2554
inventories = self.iter_inventories(revision_ids)
2555
for inv in inventories:
2556
# Should we introduce a FilteredRevisionTree class rather
2557
# than pre-filter the inventory here?
2558
filtered_inv = inv.filter(file_ids)
2559
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
656
"""Return Tree for a revision on this branch.
658
`revision_id` may not be None or 'null:'"""
659
assert None not in revision_ids
660
assert _mod_revision.NULL_REVISION not in revision_ids
661
texts = self.get_inventory_weave().get_texts(revision_ids)
662
for text, revision_id in zip(texts, revision_ids):
663
inv = self.deserialise_inventory(revision_id, text)
664
yield RevisionTree(self, inv, revision_id)
2561
666
@needs_read_lock
2562
def get_ancestry(self, revision_id, topo_sorted=True):
667
def get_ancestry(self, revision_id):
2563
668
"""Return a list of revision-ids integrated by a revision.
2565
The first element of the list is always None, indicating the origin
2566
revision. This might change when we have history horizons, or
670
The first element of the list is always None, indicating the origin
671
revision. This might change when we have history horizons, or
2567
672
perhaps we should have a new API.
2569
674
This is topologically sorted.
2571
if _mod_revision.is_null(revision_id):
676
if revision_id is None:
2573
678
if not self.has_revision(revision_id):
2574
679
raise errors.NoSuchRevision(self, revision_id)
2575
graph = self.get_graph()
2577
search = graph._make_breadth_first_searcher([revision_id])
2580
found, ghosts = search.next_with_ghosts()
2581
except StopIteration:
2584
if _mod_revision.NULL_REVISION in keys:
2585
keys.remove(_mod_revision.NULL_REVISION)
2587
parent_map = graph.get_parent_map(keys)
2588
keys = tsort.topo_sort(parent_map)
2589
return [None] + list(keys)
2591
def pack(self, hint=None, clean_obsolete_packs=False):
2592
"""Compress the data within the repository.
2594
This operation only makes sense for some repository types. For other
2595
types it should be a no-op that just returns.
2597
This stub method does not require a lock, but subclasses should use
2598
@needs_write_lock as this is a long running call its reasonable to
2599
implicitly lock for the user.
2601
:param hint: If not supplied, the whole repository is packed.
2602
If supplied, the repository may use the hint parameter as a
2603
hint for the parts of the repository to pack. A hint can be
2604
obtained from the result of commit_write_group(). Out of
2605
date hints are simply ignored, because concurrent operations
2606
can obsolete them rapidly.
2608
:param clean_obsolete_packs: Clean obsolete packs immediately after
680
w = self.get_inventory_weave()
681
candidates = w.get_ancestry(revision_id)
682
return [None] + candidates # self._eliminate_revisions_not_present(candidates)
685
def print_file(self, file, revision_id):
686
"""Print `file` to stdout.
688
FIXME RBC 20060125 as John Meinel points out this is a bad api
689
- it writes to stdout, it assumes that that is valid etc. Fix
690
by creating a new more flexible convenience function.
692
tree = self.revision_tree(revision_id)
693
# use inventory as it was in that revision
694
file_id = tree.inventory.path2id(file)
696
# TODO: jam 20060427 Write a test for this code path
697
# it had a bug in it, and was raising the wrong
699
raise errors.BzrError("%r is not present in revision %s" % (file, revision_id))
700
tree.print_file(file_id)
2612
702
def get_transaction(self):
2613
703
return self.control_files.get_transaction()
2615
def get_parent_map(self, revision_ids):
2616
"""See graph.StackedParentsProvider.get_parent_map"""
2617
# revisions index works in keys; this just works in revisions
2618
# therefore wrap and unwrap
2621
for revision_id in revision_ids:
2622
if revision_id == _mod_revision.NULL_REVISION:
2623
result[revision_id] = ()
2624
elif revision_id is None:
2625
raise ValueError('get_parent_map(None) is not valid')
2627
query_keys.append((revision_id ,))
2628
for ((revision_id,), parent_keys) in \
2629
self.revisions.get_parent_map(query_keys).iteritems():
2631
result[revision_id] = tuple([parent_revid
2632
for (parent_revid,) in parent_keys])
2634
result[revision_id] = (_mod_revision.NULL_REVISION,)
2637
def _make_parents_provider(self):
2641
def get_known_graph_ancestry(self, revision_ids):
2642
"""Return the known graph for a set of revision ids and their ancestors.
2644
st = static_tuple.StaticTuple
2645
revision_keys = [st(r_id).intern() for r_id in revision_ids]
2646
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
2647
return graph.GraphThunkIdsToKeys(known_graph)
2649
def get_graph(self, other_repository=None):
2650
"""Return the graph walker for this repository format"""
2651
parents_provider = self._make_parents_provider()
2652
if (other_repository is not None and
2653
not self.has_same_location(other_repository)):
2654
parents_provider = graph.StackedParentsProvider(
2655
[parents_provider, other_repository._make_parents_provider()])
2656
return graph.Graph(parents_provider)
2658
def _get_versioned_file_checker(self, text_key_references=None,
2660
"""Return an object suitable for checking versioned files.
2662
:param text_key_references: if non-None, an already built
2663
dictionary mapping text keys ((fileid, revision_id) tuples)
2664
to whether they were referred to by the inventory of the
2665
revision_id that they contain. If None, this will be
2667
:param ancestors: Optional result from
2668
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2671
return _VersionedFileChecker(self,
2672
text_key_references=text_key_references, ancestors=ancestors)
2674
def revision_ids_to_search_result(self, result_set):
2675
"""Convert a set of revision ids to a graph SearchResult."""
2676
result_parents = set()
2677
for parents in self.get_graph().get_parent_map(
2678
result_set).itervalues():
2679
result_parents.update(parents)
2680
included_keys = result_set.intersection(result_parents)
2681
start_keys = result_set.difference(included_keys)
2682
exclude_keys = result_parents.difference(result_set)
2683
result = graph.SearchResult(start_keys, exclude_keys,
2684
len(result_set), result_set)
705
def revision_parents(self, revid):
706
return self.get_inventory_weave().parent_names(revid)
2687
708
@needs_write_lock
2688
709
def set_make_working_trees(self, new_value):
3368
1114
InterRepository.get(other).method_name(parameters).
3371
_walk_to_common_revisions_batch_size = 50
3372
1117
_optimisers = []
3373
1118
"""The available optimised InterRepository types."""
3376
def copy_content(self, revision_id=None):
3377
"""Make a complete copy of the content in self into destination.
3379
This is a destructive operation! Do not use it on existing
3382
:param revision_id: Only copy the content needed to construct
3383
revision_id and its parents.
3386
self.target.set_make_working_trees(self.source.make_working_trees())
3387
except NotImplementedError:
3389
self.target.fetch(self.source, revision_id=revision_id)
3392
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
1120
def copy_content(self, revision_id=None, basis=None):
1121
raise NotImplementedError(self.copy_content)
1123
def fetch(self, revision_id=None, pb=None):
3394
1124
"""Fetch the content required to construct revision_id.
3396
1126
The content is copied from self.source to self.target.
3398
1128
:param revision_id: if None all content is copied, if NULL_REVISION no
3399
1129
content is copied.
3403
ui.ui_factory.warn_experimental_format_fetch(self)
3404
from bzrlib.fetch import RepoFetcher
3405
# See <https://launchpad.net/bugs/456077> asking for a warning here
3406
if self.source._format.network_name() != self.target._format.network_name():
3407
ui.ui_factory.show_user_warning('cross_format_fetch',
3408
from_format=self.source._format,
3409
to_format=self.target._format)
3410
f = RepoFetcher(to_repository=self.target,
3411
from_repository=self.source,
3412
last_revision=revision_id,
3413
fetch_spec=fetch_spec,
3414
find_ghosts=find_ghosts)
3416
def _walk_to_common_revisions(self, revision_ids):
3417
"""Walk out from revision_ids in source to revisions target has.
3419
:param revision_ids: The start point for the search.
3420
:return: A set of revision ids.
3422
target_graph = self.target.get_graph()
3423
revision_ids = frozenset(revision_ids)
3424
missing_revs = set()
3425
source_graph = self.source.get_graph()
3426
# ensure we don't pay silly lookup costs.
3427
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3428
null_set = frozenset([_mod_revision.NULL_REVISION])
3429
searcher_exhausted = False
3433
# Iterate the searcher until we have enough next_revs
3434
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3436
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3437
next_revs.update(next_revs_part)
3438
ghosts.update(ghosts_part)
3439
except StopIteration:
3440
searcher_exhausted = True
3442
# If there are ghosts in the source graph, and the caller asked for
3443
# them, make sure that they are present in the target.
3444
# We don't care about other ghosts as we can't fetch them and
3445
# haven't been asked to.
3446
ghosts_to_check = set(revision_ids.intersection(ghosts))
3447
revs_to_get = set(next_revs).union(ghosts_to_check)
3449
have_revs = set(target_graph.get_parent_map(revs_to_get))
3450
# we always have NULL_REVISION present.
3451
have_revs = have_revs.union(null_set)
3452
# Check if the target is missing any ghosts we need.
3453
ghosts_to_check.difference_update(have_revs)
3455
# One of the caller's revision_ids is a ghost in both the
3456
# source and the target.
3457
raise errors.NoSuchRevision(
3458
self.source, ghosts_to_check.pop())
3459
missing_revs.update(next_revs - have_revs)
3460
# Because we may have walked past the original stop point, make
3461
# sure everything is stopped
3462
stop_revs = searcher.find_seen_ancestors(have_revs)
3463
searcher.stop_searching_any(stop_revs)
3464
if searcher_exhausted:
3466
return searcher.get_result()
1130
:param pb: optional progress bar to use for progress reports. If not
1131
provided a default one will be created.
1133
Returns the copied revision count and the failed revisions in a tuple:
1136
raise NotImplementedError(self.fetch)
3468
1138
@needs_read_lock
3469
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1139
def missing_revision_ids(self, revision_id=None):
3470
1140
"""Return the revision ids that source has that target does not.
1142
These are returned in topological order.
3472
1144
:param revision_id: only return revision ids included by this
3474
:param find_ghosts: If True find missing revisions in deep history
3475
rather than just finding the surface difference.
3476
:return: A bzrlib.graph.SearchResult.
3478
# stop searching at found target revisions.
3479
if not find_ghosts and revision_id is not None:
3480
return self._walk_to_common_revisions([revision_id])
3481
1147
# generic, possibly worst case, slow code path.
3482
1148
target_ids = set(self.target.all_revision_ids())
3483
1149
if revision_id is not None:
3484
1150
source_ids = self.source.get_ancestry(revision_id)
3485
if source_ids[0] is not None:
3486
raise AssertionError()
1151
assert source_ids[0] is None
3487
1152
source_ids.pop(0)
3489
1154
source_ids = self.source.all_revision_ids()
3490
1155
result_set = set(source_ids).difference(target_ids)
3491
return self.source.revision_ids_to_search_result(result_set)
3494
def _same_model(source, target):
3495
"""True if source and target have the same data representation.
3497
Note: this is always called on the base class; overriding it in a
3498
subclass will have no effect.
3501
InterRepository._assert_same_model(source, target)
3503
except errors.IncompatibleRepositories, e:
3507
def _assert_same_model(source, target):
3508
"""Raise an exception if two repositories do not use the same model.
3510
if source.supports_rich_root() != target.supports_rich_root():
3511
raise errors.IncompatibleRepositories(source, target,
3512
"different rich-root support")
3513
if source._serializer != target._serializer:
3514
raise errors.IncompatibleRepositories(source, target,
3515
"different serializers")
1156
# this may look like a no-op: its not. It preserves the ordering
1157
# other_ids had while only returning the members from other_ids
1158
# that we've decided we need.
1159
return [rev_id for rev_id in source_ids if rev_id in result_set]
3518
1162
class InterSameDataRepository(InterRepository):
3519
1163
"""Code for converting between repositories that represent the same data.
3521
1165
Data format and model must match for this to work.
3525
def _get_repo_format_to_test(self):
3526
"""Repository format for testing with.
3528
InterSameData can pull from subtree to subtree and from non-subtree to
3529
non-subtree, so we test this with the richest repository format.
3531
from bzrlib.repofmt import knitrepo
3532
return knitrepo.RepositoryFormatKnit3()
3535
def is_compatible(source, target):
3536
return InterRepository._same_model(source, target)
3539
class InterWeaveRepo(InterSameDataRepository):
3540
"""Optimised code paths between Weave based repositories.
3542
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3543
implemented lazy inter-object optimisation.
3547
def _get_repo_format_to_test(self):
3548
from bzrlib.repofmt import weaverepo
3549
return weaverepo.RepositoryFormat7()
3552
def is_compatible(source, target):
3553
"""Be compatible with known Weave formats.
3555
We don't test for the stores being of specific types because that
3556
could lead to confusing results, and there is no need to be
3559
from bzrlib.repofmt.weaverepo import (
3565
return (isinstance(source._format, (RepositoryFormat5,
3567
RepositoryFormat7)) and
3568
isinstance(target._format, (RepositoryFormat5,
3570
RepositoryFormat7)))
3571
except AttributeError:
1169
def _get_matching_repo_format(self):
1170
"""Repository format for testing with."""
1171
return RepositoryFormat.get_default_format()
1174
def is_compatible(source, target):
1175
if not isinstance(source, Repository):
1177
if not isinstance(target, Repository):
1179
if source._format.rich_root_data == target._format.rich_root_data:
3574
1184
@needs_write_lock
3575
def copy_content(self, revision_id=None):
3576
"""See InterRepository.copy_content()."""
3577
# weave specific optimised path:
1185
def copy_content(self, revision_id=None, basis=None):
1186
"""Make a complete copy of the content in self into destination.
1188
This is a destructive operation! Do not use it on existing
1191
:param revision_id: Only copy the content needed to construct
1192
revision_id and its parents.
1193
:param basis: Copy the needed data preferentially from basis.
3579
1196
self.target.set_make_working_trees(self.source.make_working_trees())
3580
except (errors.RepositoryUpgradeRequired, NotImplemented):
1197
except NotImplementedError:
3582
# FIXME do not peek!
3583
if self.source._transport.listable():
3584
pb = ui.ui_factory.nested_progress_bar()
3586
self.target.texts.insert_record_stream(
3587
self.source.texts.get_record_stream(
3588
self.source.texts.keys(), 'topological', False))
3589
pb.update('Copying inventory', 0, 1)
3590
self.target.inventories.insert_record_stream(
3591
self.source.inventories.get_record_stream(
3592
self.source.inventories.keys(), 'topological', False))
3593
self.target.signatures.insert_record_stream(
3594
self.source.signatures.get_record_stream(
3595
self.source.signatures.keys(),
3597
self.target.revisions.insert_record_stream(
3598
self.source.revisions.get_record_stream(
3599
self.source.revisions.keys(),
3600
'topological', True))
3604
self.target.fetch(self.source, revision_id=revision_id)
1199
# grab the basis available data
1200
if basis is not None:
1201
self.target.fetch(basis, revision_id=revision_id)
1202
# but don't bother fetching if we have the needed data now.
1203
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1204
self.target.has_revision(revision_id)):
1206
self.target.fetch(self.source, revision_id=revision_id)
3607
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3608
"""See InterRepository.missing_revision_ids()."""
3609
# we want all revisions to satisfy revision_id in source.
3610
# but we don't want to stat every file here and there.
3611
# we want then, all revisions other needs to satisfy revision_id
3612
# checked, but not those that we have locally.
3613
# so the first thing is to get a subset of the revisions to
3614
# satisfy revision_id in source, and then eliminate those that
3615
# we do already have.
3616
# this is slow on high latency connection to self, but as this
3617
# disk format scales terribly for push anyway due to rewriting
3618
# inventory.weave, this is considered acceptable.
3620
if revision_id is not None:
3621
source_ids = self.source.get_ancestry(revision_id)
3622
if source_ids[0] is not None:
3623
raise AssertionError()
3626
source_ids = self.source._all_possible_ids()
3627
source_ids_set = set(source_ids)
3628
# source_ids is the worst possible case we may need to pull.
3629
# now we want to filter source_ids against what we actually
3630
# have in target, but don't try to check for existence where we know
3631
# we do not have a revision as that would be pointless.
3632
target_ids = set(self.target._all_possible_ids())
3633
possibly_present_revisions = target_ids.intersection(source_ids_set)
3634
actually_present_revisions = set(
3635
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3636
required_revisions = source_ids_set.difference(actually_present_revisions)
3637
if revision_id is not None:
3638
# we used get_ancestry to determine source_ids then we are assured all
3639
# revisions referenced are present as they are installed in topological order.
3640
# and the tip revision was validated by get_ancestry.
3641
result_set = required_revisions
3643
# if we just grabbed the possibly available ids, then
3644
# we only have an estimate of whats available and need to validate
3645
# that against the revision records.
3647
self.source._eliminate_revisions_not_present(required_revisions))
3648
return self.source.revision_ids_to_search_result(result_set)
1209
def fetch(self, revision_id=None, pb=None):
1210
"""See InterRepository.fetch()."""
1211
from bzrlib.fetch import GenericRepoFetcher
1212
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1213
self.source, self.source._format, self.target,
1214
self.target._format)
1215
f = GenericRepoFetcher(to_repository=self.target,
1216
from_repository=self.source,
1217
last_revision=revision_id,
1219
return f.count_copied, f.failed_revisions
3651
1222
class InterKnitRepo(InterSameDataRepository):
3652
1223
"""Optimised code paths between Knit based repositories."""
3655
def _get_repo_format_to_test(self):
1226
def _get_matching_repo_format(self):
3656
1227
from bzrlib.repofmt import knitrepo
3657
1228
return knitrepo.RepositoryFormatKnit1()
3660
1231
def is_compatible(source, target):
3661
1232
"""Be compatible with known Knit formats.
3663
1234
We don't test for the stores being of specific types because that
3664
could lead to confusing results, and there is no need to be
1235
could lead to confusing results, and there is no need to be
3665
1236
overly general.
3667
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
1238
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1
3669
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3670
isinstance(target._format, RepositoryFormatKnit))
1240
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1241
isinstance(target._format, (RepositoryFormatKnit1)))
3671
1242
except AttributeError:
3673
return are_knits and InterRepository._same_model(source, target)
1246
def fetch(self, revision_id=None, pb=None):
1247
"""See InterRepository.fetch()."""
1248
from bzrlib.fetch import KnitRepoFetcher
1249
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1250
self.source, self.source._format, self.target, self.target._format)
1251
f = KnitRepoFetcher(to_repository=self.target,
1252
from_repository=self.source,
1253
last_revision=revision_id,
1255
return f.count_copied, f.failed_revisions
3675
1257
@needs_read_lock
3676
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1258
def missing_revision_ids(self, revision_id=None):
3677
1259
"""See InterRepository.missing_revision_ids()."""
3678
1260
if revision_id is not None:
3679
1261
source_ids = self.source.get_ancestry(revision_id)
3680
if source_ids[0] is not None:
3681
raise AssertionError()
1262
assert source_ids[0] is None
3682
1263
source_ids.pop(0)
3684
source_ids = self.source.all_revision_ids()
1265
source_ids = self.source._all_possible_ids()
3685
1266
source_ids_set = set(source_ids)
3686
1267
# source_ids is the worst possible case we may need to pull.
3687
1268
# now we want to filter source_ids against what we actually
3688
1269
# have in target, but don't try to check for existence where we know
3689
1270
# we do not have a revision as that would be pointless.
3690
target_ids = set(self.target.all_revision_ids())
1271
target_ids = set(self.target._all_possible_ids())
3691
1272
possibly_present_revisions = target_ids.intersection(source_ids_set)
3692
actually_present_revisions = set(
3693
self.target._eliminate_revisions_not_present(possibly_present_revisions))
1273
actually_present_revisions = set(self.target._eliminate_revisions_not_present(possibly_present_revisions))
3694
1274
required_revisions = source_ids_set.difference(actually_present_revisions)
1275
required_topo_revisions = [rev_id for rev_id in source_ids if rev_id in required_revisions]
3695
1276
if revision_id is not None:
3696
1277
# we used get_ancestry to determine source_ids then we are assured all
3697
1278
# revisions referenced are present as they are installed in topological order.
3698
1279
# and the tip revision was validated by get_ancestry.
3699
result_set = required_revisions
1280
return required_topo_revisions
3701
# if we just grabbed the possibly available ids, then
1282
# if we just grabbed the possibly available ids, then
3702
1283
# we only have an estimate of whats available and need to validate
3703
1284
# that against the revision records.
3705
self.source._eliminate_revisions_not_present(required_revisions))
3706
return self.source.revision_ids_to_search_result(result_set)
3709
class InterDifferingSerializer(InterRepository):
1285
return self.source._eliminate_revisions_not_present(required_topo_revisions)
1288
class InterModel1and2(InterRepository):
3712
def _get_repo_format_to_test(self):
1291
def _get_matching_repo_format(self):
3716
1295
def is_compatible(source, target):
3717
"""Be compatible with Knit2 source and Knit3 target"""
3718
# This is redundant with format.check_conversion_target(), however that
3719
# raises an exception, and we just want to say "False" as in we won't
3720
# support converting between these formats.
3721
if 'IDS_never' in debug.debug_flags:
3723
if source.supports_rich_root() and not target.supports_rich_root():
3725
if (source._format.supports_tree_reference
3726
and not target._format.supports_tree_reference):
3728
if target._fallback_repositories and target._format.supports_chks:
3729
# IDS doesn't know how to copy CHKs for the parent inventories it
3730
# adds to stacked repos.
3732
if 'IDS_always' in debug.debug_flags:
1296
if not isinstance(source, Repository):
1298
if not isinstance(target, Repository):
1300
if not source._format.rich_root_data and target._format.rich_root_data:
3734
# Only use this code path for local source and target. IDS does far
3735
# too much IO (both bandwidth and roundtrips) over a network.
3736
if not source.bzrdir.transport.base.startswith('file:///'):
3738
if not target.bzrdir.transport.base.startswith('file:///'):
3742
def _get_trees(self, revision_ids, cache):
3744
for rev_id in revision_ids:
3746
possible_trees.append((rev_id, cache[rev_id]))
3748
# Not cached, but inventory might be present anyway.
3750
tree = self.source.revision_tree(rev_id)
3751
except errors.NoSuchRevision:
3752
# Nope, parent is ghost.
3755
cache[rev_id] = tree
3756
possible_trees.append((rev_id, tree))
3757
return possible_trees
3759
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3760
"""Get the best delta and base for this revision.
3762
:return: (basis_id, delta)
3765
# Generate deltas against each tree, to find the shortest.
3766
texts_possibly_new_in_tree = set()
3767
for basis_id, basis_tree in possible_trees:
3768
delta = tree.inventory._make_delta(basis_tree.inventory)
3769
for old_path, new_path, file_id, new_entry in delta:
3770
if new_path is None:
3771
# This file_id isn't present in the new rev, so we don't
3775
# Rich roots are handled elsewhere...
3777
kind = new_entry.kind
3778
if kind != 'directory' and kind != 'file':
3779
# No text record associated with this inventory entry.
3781
# This is a directory or file that has changed somehow.
3782
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3783
deltas.append((len(delta), basis_id, delta))
3785
return deltas[0][1:]
3787
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3788
"""Find all parent revisions that are absent, but for which the
3789
inventory is present, and copy those inventories.
3791
This is necessary to preserve correctness when the source is stacked
3792
without fallbacks configured. (Note that in cases like upgrade the
3793
source may be not have _fallback_repositories even though it is
3797
for parents in parent_map.values():
3798
parent_revs.update(parents)
3799
present_parents = self.source.get_parent_map(parent_revs)
3800
absent_parents = set(parent_revs).difference(present_parents)
3801
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3802
(rev_id,) for rev_id in absent_parents)
3803
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3804
for parent_tree in self.source.revision_trees(parent_inv_ids):
3805
current_revision_id = parent_tree.get_revision_id()
3806
parents_parents_keys = parent_invs_keys_for_stacking[
3807
(current_revision_id,)]
3808
parents_parents = [key[-1] for key in parents_parents_keys]
3809
basis_id = _mod_revision.NULL_REVISION
3810
basis_tree = self.source.revision_tree(basis_id)
3811
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3812
self.target.add_inventory_by_delta(
3813
basis_id, delta, current_revision_id, parents_parents)
3814
cache[current_revision_id] = parent_tree
3816
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3817
"""Fetch across a few revisions.
3819
:param revision_ids: The revisions to copy
3820
:param basis_id: The revision_id of a tree that must be in cache, used
3821
as a basis for delta when no other base is available
3822
:param cache: A cache of RevisionTrees that we can use.
3823
:param a_graph: A Graph object to determine the heads() of the
3824
rich-root data stream.
3825
:return: The revision_id of the last converted tree. The RevisionTree
3826
for it will be in cache
3828
# Walk though all revisions; get inventory deltas, copy referenced
3829
# texts that delta references, insert the delta, revision and
3831
root_keys_to_create = set()
3834
pending_revisions = []
3835
parent_map = self.source.get_parent_map(revision_ids)
3836
self._fetch_parent_invs_for_stacking(parent_map, cache)
3837
self.source._safe_to_return_from_cache = True
3838
for tree in self.source.revision_trees(revision_ids):
3839
# Find a inventory delta for this revision.
3840
# Find text entries that need to be copied, too.
3841
current_revision_id = tree.get_revision_id()
3842
parent_ids = parent_map.get(current_revision_id, ())
3843
parent_trees = self._get_trees(parent_ids, cache)
3844
possible_trees = list(parent_trees)
3845
if len(possible_trees) == 0:
3846
# There either aren't any parents, or the parents are ghosts,
3847
# so just use the last converted tree.
3848
possible_trees.append((basis_id, cache[basis_id]))
3849
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3851
revision = self.source.get_revision(current_revision_id)
3852
pending_deltas.append((basis_id, delta,
3853
current_revision_id, revision.parent_ids))
3854
if self._converting_to_rich_root:
3855
self._revision_id_to_root_id[current_revision_id] = \
3857
# Determine which texts are in present in this revision but not in
3858
# any of the available parents.
3859
texts_possibly_new_in_tree = set()
3860
for old_path, new_path, file_id, entry in delta:
3861
if new_path is None:
3862
# This file_id isn't present in the new rev
3866
if not self.target.supports_rich_root():
3867
# The target doesn't support rich root, so we don't
3870
if self._converting_to_rich_root:
3871
# This can't be copied normally, we have to insert
3873
root_keys_to_create.add((file_id, entry.revision))
3876
texts_possibly_new_in_tree.add((file_id, entry.revision))
3877
for basis_id, basis_tree in possible_trees:
3878
basis_inv = basis_tree.inventory
3879
for file_key in list(texts_possibly_new_in_tree):
3880
file_id, file_revision = file_key
3882
entry = basis_inv[file_id]
3883
except errors.NoSuchId:
3885
if entry.revision == file_revision:
3886
texts_possibly_new_in_tree.remove(file_key)
3887
text_keys.update(texts_possibly_new_in_tree)
3888
pending_revisions.append(revision)
3889
cache[current_revision_id] = tree
3890
basis_id = current_revision_id
3891
self.source._safe_to_return_from_cache = False
3893
from_texts = self.source.texts
3894
to_texts = self.target.texts
3895
if root_keys_to_create:
3896
root_stream = _mod_fetch._new_root_data_stream(
3897
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3898
self.source, graph=a_graph)
3899
to_texts.insert_record_stream(root_stream)
3900
to_texts.insert_record_stream(from_texts.get_record_stream(
3901
text_keys, self.target._format._fetch_order,
3902
not self.target._format._fetch_uses_deltas))
3903
# insert inventory deltas
3904
for delta in pending_deltas:
3905
self.target.add_inventory_by_delta(*delta)
3906
if self.target._fallback_repositories:
3907
# Make sure this stacked repository has all the parent inventories
3908
# for the new revisions that we are about to insert. We do this
3909
# before adding the revisions so that no revision is added until
3910
# all the inventories it may depend on are added.
3911
# Note that this is overzealous, as we may have fetched these in an
3914
revision_ids = set()
3915
for revision in pending_revisions:
3916
revision_ids.add(revision.revision_id)
3917
parent_ids.update(revision.parent_ids)
3918
parent_ids.difference_update(revision_ids)
3919
parent_ids.discard(_mod_revision.NULL_REVISION)
3920
parent_map = self.source.get_parent_map(parent_ids)
3921
# we iterate over parent_map and not parent_ids because we don't
3922
# want to try copying any revision which is a ghost
3923
for parent_tree in self.source.revision_trees(parent_map):
3924
current_revision_id = parent_tree.get_revision_id()
3925
parents_parents = parent_map[current_revision_id]
3926
possible_trees = self._get_trees(parents_parents, cache)
3927
if len(possible_trees) == 0:
3928
# There either aren't any parents, or the parents are
3929
# ghosts, so just use the last converted tree.
3930
possible_trees.append((basis_id, cache[basis_id]))
3931
basis_id, delta = self._get_delta_for_revision(parent_tree,
3932
parents_parents, possible_trees)
3933
self.target.add_inventory_by_delta(
3934
basis_id, delta, current_revision_id, parents_parents)
3935
# insert signatures and revisions
3936
for revision in pending_revisions:
3938
signature = self.source.get_signature_text(
3939
revision.revision_id)
3940
self.target.add_signature_text(revision.revision_id,
3942
except errors.NoSuchRevision:
3944
self.target.add_revision(revision.revision_id, revision)
3947
def _fetch_all_revisions(self, revision_ids, pb):
3948
"""Fetch everything for the list of revisions.
3950
:param revision_ids: The list of revisions to fetch. Must be in
3952
:param pb: A ProgressTask
3955
basis_id, basis_tree = self._get_basis(revision_ids[0])
3957
cache = lru_cache.LRUCache(100)
3958
cache[basis_id] = basis_tree
3959
del basis_tree # We don't want to hang on to it here
3961
if self._converting_to_rich_root and len(revision_ids) > 100:
3962
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3967
for offset in range(0, len(revision_ids), batch_size):
3968
self.target.start_write_group()
3970
pb.update('Transferring revisions', offset,
3972
batch = revision_ids[offset:offset+batch_size]
3973
basis_id = self._fetch_batch(batch, basis_id, cache,
3976
self.source._safe_to_return_from_cache = False
3977
self.target.abort_write_group()
3980
hint = self.target.commit_write_group()
3983
if hints and self.target._format.pack_compresses:
3984
self.target.pack(hint=hints)
3985
pb.update('Transferring revisions', len(revision_ids),
3989
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3991
"""See InterRepository.fetch()."""
3992
if fetch_spec is not None:
3993
raise AssertionError("Not implemented yet...")
3994
ui.ui_factory.warn_experimental_format_fetch(self)
3995
if (not self.source.supports_rich_root()
3996
and self.target.supports_rich_root()):
3997
self._converting_to_rich_root = True
3998
self._revision_id_to_root_id = {}
4000
self._converting_to_rich_root = False
4001
# See <https://launchpad.net/bugs/456077> asking for a warning here
4002
if self.source._format.network_name() != self.target._format.network_name():
4003
ui.ui_factory.show_user_warning('cross_format_fetch',
4004
from_format=self.source._format,
4005
to_format=self.target._format)
4006
revision_ids = self.target.search_missing_revision_ids(self.source,
4007
revision_id, find_ghosts=find_ghosts).get_keys()
4008
if not revision_ids:
4010
revision_ids = tsort.topo_sort(
4011
self.source.get_graph().get_parent_map(revision_ids))
4012
if not revision_ids:
4014
# Walk though all revisions; get inventory deltas, copy referenced
4015
# texts that delta references, insert the delta, revision and
4018
my_pb = ui.ui_factory.nested_progress_bar()
4021
symbol_versioning.warn(
4022
symbol_versioning.deprecated_in((1, 14, 0))
4023
% "pb parameter to fetch()")
4026
self._fetch_all_revisions(revision_ids, pb)
4028
if my_pb is not None:
4030
return len(revision_ids), 0
4032
def _get_basis(self, first_revision_id):
4033
"""Get a revision and tree which exists in the target.
4035
This assumes that first_revision_id is selected for transmission
4036
because all other ancestors are already present. If we can't find an
4037
ancestor we fall back to NULL_REVISION since we know that is safe.
4039
:return: (basis_id, basis_tree)
4041
first_rev = self.source.get_revision(first_revision_id)
4043
basis_id = first_rev.parent_ids[0]
4044
# only valid as a basis if the target has it
4045
self.target.get_revision(basis_id)
4046
# Try to get a basis tree - if its a ghost it will hit the
4047
# NoSuchRevision case.
4048
basis_tree = self.source.revision_tree(basis_id)
4049
except (IndexError, errors.NoSuchRevision):
4050
basis_id = _mod_revision.NULL_REVISION
4051
basis_tree = self.source.revision_tree(basis_id)
4052
return basis_id, basis_tree
4055
InterRepository.register_optimiser(InterDifferingSerializer)
1306
def fetch(self, revision_id=None, pb=None):
1307
"""See InterRepository.fetch()."""
1308
from bzrlib.fetch import Model1toKnit2Fetcher
1309
f = Model1toKnit2Fetcher(to_repository=self.target,
1310
from_repository=self.source,
1311
last_revision=revision_id,
1313
return f.count_copied, f.failed_revisions
1316
def copy_content(self, revision_id=None, basis=None):
1317
"""Make a complete copy of the content in self into destination.
1319
This is a destructive operation! Do not use it on existing
1322
:param revision_id: Only copy the content needed to construct
1323
revision_id and its parents.
1324
:param basis: Copy the needed data preferentially from basis.
1327
self.target.set_make_working_trees(self.source.make_working_trees())
1328
except NotImplementedError:
1330
# grab the basis available data
1331
if basis is not None:
1332
self.target.fetch(basis, revision_id=revision_id)
1333
# but don't bother fetching if we have the needed data now.
1334
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1335
self.target.has_revision(revision_id)):
1337
self.target.fetch(self.source, revision_id=revision_id)
1340
class InterKnit1and2(InterKnitRepo):
1343
def _get_matching_repo_format(self):
1347
def is_compatible(source, target):
1348
"""Be compatible with Knit1 source and Knit2 target"""
1349
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit2
1351
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1, \
1352
RepositoryFormatKnit2
1353
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1354
isinstance(target._format, (RepositoryFormatKnit2)))
1355
except AttributeError:
1359
def fetch(self, revision_id=None, pb=None):
1360
"""See InterRepository.fetch()."""
1361
from bzrlib.fetch import Knit1to2Fetcher
1362
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1363
self.source, self.source._format, self.target,
1364
self.target._format)
1365
f = Knit1to2Fetcher(to_repository=self.target,
1366
from_repository=self.source,
1367
last_revision=revision_id,
1369
return f.count_copied, f.failed_revisions
4056
1372
InterRepository.register_optimiser(InterSameDataRepository)
4057
InterRepository.register_optimiser(InterWeaveRepo)
4058
1373
InterRepository.register_optimiser(InterKnitRepo)
1374
InterRepository.register_optimiser(InterModel1and2)
1375
InterRepository.register_optimiser(InterKnit1and2)
1378
class RepositoryTestProviderAdapter(object):
1379
"""A tool to generate a suite testing multiple repository formats at once.
1381
This is done by copying the test once for each transport and injecting
1382
the transport_server, transport_readonly_server, and bzrdir_format and
1383
repository_format classes into each copy. Each copy is also given a new id()
1384
to make it easy to identify.
1387
def __init__(self, transport_server, transport_readonly_server, formats):
1388
self._transport_server = transport_server
1389
self._transport_readonly_server = transport_readonly_server
1390
self._formats = formats
1392
def adapt(self, test):
1393
result = unittest.TestSuite()
1394
for repository_format, bzrdir_format in self._formats:
1395
new_test = deepcopy(test)
1396
new_test.transport_server = self._transport_server
1397
new_test.transport_readonly_server = self._transport_readonly_server
1398
new_test.bzrdir_format = bzrdir_format
1399
new_test.repository_format = repository_format
1400
def make_new_test_id():
1401
new_id = "%s(%s)" % (new_test.id(), repository_format.__class__.__name__)
1402
return lambda: new_id
1403
new_test.id = make_new_test_id()
1404
result.addTest(new_test)
1408
class InterRepositoryTestProviderAdapter(object):
1409
"""A tool to generate a suite testing multiple inter repository formats.
1411
This is done by copying the test once for each interrepo provider and injecting
1412
the transport_server, transport_readonly_server, repository_format and
1413
repository_to_format classes into each copy.
1414
Each copy is also given a new id() to make it easy to identify.
1417
def __init__(self, transport_server, transport_readonly_server, formats):
1418
self._transport_server = transport_server
1419
self._transport_readonly_server = transport_readonly_server
1420
self._formats = formats
1422
def adapt(self, test):
1423
result = unittest.TestSuite()
1424
for interrepo_class, repository_format, repository_format_to in self._formats:
1425
new_test = deepcopy(test)
1426
new_test.transport_server = self._transport_server
1427
new_test.transport_readonly_server = self._transport_readonly_server
1428
new_test.interrepo_class = interrepo_class
1429
new_test.repository_format = repository_format
1430
new_test.repository_format_to = repository_format_to
1431
def make_new_test_id():
1432
new_id = "%s(%s)" % (new_test.id(), interrepo_class.__name__)
1433
return lambda: new_id
1434
new_test.id = make_new_test_id()
1435
result.addTest(new_test)
1439
def default_test_list():
1440
"""Generate the default list of interrepo permutations to test."""
1441
from bzrlib.repofmt import knitrepo, weaverepo
1443
# test the default InterRepository between format 6 and the current
1445
# XXX: robertc 20060220 reinstate this when there are two supported
1446
# formats which do not have an optimal code path between them.
1447
#result.append((InterRepository,
1448
# RepositoryFormat6(),
1449
# RepositoryFormatKnit1()))
1450
for optimiser_class in InterRepository._optimisers:
1451
format_to_test = optimiser_class._get_matching_repo_format()
1452
if format_to_test is not None:
1453
result.append((optimiser_class,
1454
format_to_test, format_to_test))
1455
# if there are specific combinations we want to use, we can add them
1457
result.append((InterModel1and2,
1458
weaverepo.RepositoryFormat5(),
1459
knitrepo.RepositoryFormatKnit2()))
1460
result.append((InterKnit1and2,
1461
knitrepo.RepositoryFormatKnit1(),
1462
knitrepo.RepositoryFormatKnit2()))
4061
1466
class CopyConverter(object):
4062
1467
"""A repository conversion tool which just performs a copy of the content.
4064
1469
This is slow but quite reliable.
4070
1475
:param target_format: The format the resulting repository should be.
4072
1477
self.target_format = target_format
4074
1479
def convert(self, repo, pb):
4075
1480
"""Perform the conversion of to_convert, giving feedback via pb.
4077
1482
:param to_convert: The disk object to convert.
4078
1483
:param pb: a progress bar to use for progress information.
4080
pb = ui.ui_factory.nested_progress_bar()
4083
1488
# this is only useful with metadir layouts - separated repo content.
4084
1489
# trigger an assertion if not such
4085
1490
repo._format.get_format_string()
4086
1491
self.repo_dir = repo.bzrdir
4087
pb.update('Moving repository to repository.backup')
1492
self.step('Moving repository to repository.backup')
4088
1493
self.repo_dir.transport.move('repository', 'repository.backup')
4089
1494
backup_transport = self.repo_dir.transport.clone('repository.backup')
4090
1495
repo._format.check_conversion_target(self.target_format)
4091
1496
self.source_repo = repo._format.open(self.repo_dir,
4093
1498
_override_transport=backup_transport)
4094
pb.update('Creating new repository')
1499
self.step('Creating new repository')
4095
1500
converted = self.target_format.initialize(self.repo_dir,
4096
1501
self.source_repo.is_shared())
4097
1502
converted.lock_write()
4099
pb.update('Copying content')
1504
self.step('Copying content into repository.')
4100
1505
self.source_repo.copy_content_into(converted)
4102
1507
converted.unlock()
4103
pb.update('Deleting old repository content')
1508
self.step('Deleting old repository content.')
4104
1509
self.repo_dir.transport.delete_tree('repository.backup')
4105
ui.ui_factory.note('repository converted')
1510
self.pb.note('repository converted')
1512
def step(self, message):
1513
"""Update the pb by a step."""
1515
self.pb.update(message, self.count, self.total)
1518
class CommitBuilder(object):
1519
"""Provides an interface to build up a commit.
1521
This allows describing a tree to be committed without needing to
1522
know the internals of the format of the repository.
1525
record_root_entry = False
1526
def __init__(self, repository, parents, config, timestamp=None,
1527
timezone=None, committer=None, revprops=None,
1529
"""Initiate a CommitBuilder.
1531
:param repository: Repository to commit to.
1532
:param parents: Revision ids of the parents of the new revision.
1533
:param config: Configuration to use.
1534
:param timestamp: Optional timestamp recorded for commit.
1535
:param timezone: Optional timezone for timestamp.
1536
:param committer: Optional committer to set for commit.
1537
:param revprops: Optional dictionary of revision properties.
1538
:param revision_id: Optional revision id.
1540
self._config = config
1542
if committer is None:
1543
self._committer = self._config.username()
1545
assert isinstance(committer, basestring), type(committer)
1546
self._committer = committer
1548
self.new_inventory = Inventory(None)
1549
self._new_revision_id = revision_id
1550
self.parents = parents
1551
self.repository = repository
1554
if revprops is not None:
1555
self._revprops.update(revprops)
1557
if timestamp is None:
1558
timestamp = time.time()
1559
# Restrict resolution to 1ms
1560
self._timestamp = round(timestamp, 3)
1562
if timezone is None:
1563
self._timezone = local_time_offset()
1565
self._timezone = int(timezone)
1567
self._generate_revision_if_needed()
1569
def commit(self, message):
1570
"""Make the actual commit.
1572
:return: The revision id of the recorded revision.
1574
rev = _mod_revision.Revision(
1575
timestamp=self._timestamp,
1576
timezone=self._timezone,
1577
committer=self._committer,
1579
inventory_sha1=self.inv_sha1,
1580
revision_id=self._new_revision_id,
1581
properties=self._revprops)
1582
rev.parent_ids = self.parents
1583
self.repository.add_revision(self._new_revision_id, rev,
1584
self.new_inventory, self._config)
1585
return self._new_revision_id
1587
def revision_tree(self):
1588
"""Return the tree that was just committed.
1590
After calling commit() this can be called to get a RevisionTree
1591
representing the newly committed tree. This is preferred to
1592
calling Repository.revision_tree() because that may require
1593
deserializing the inventory, while we already have a copy in
1596
return RevisionTree(self.repository, self.new_inventory,
1597
self._new_revision_id)
1599
def finish_inventory(self):
1600
"""Tell the builder that the inventory is finished."""
1601
if self.new_inventory.root is None:
1602
symbol_versioning.warn('Root entry should be supplied to'
1603
' record_entry_contents, as of bzr 0.10.',
1604
DeprecationWarning, stacklevel=2)
1605
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
1606
self.new_inventory.revision_id = self._new_revision_id
1607
self.inv_sha1 = self.repository.add_inventory(
1608
self._new_revision_id,
1613
def _gen_revision_id(self):
1614
"""Return new revision-id."""
1615
return generate_ids.gen_revision_id(self._config.username(),
1618
def _generate_revision_if_needed(self):
1619
"""Create a revision id if None was supplied.
1621
If the repository can not support user-specified revision ids
1622
they should override this function and raise CannotSetRevisionId
1623
if _new_revision_id is not None.
1625
:raises: CannotSetRevisionId
1627
if self._new_revision_id is None:
1628
self._new_revision_id = self._gen_revision_id()
1630
def record_entry_contents(self, ie, parent_invs, path, tree):
1631
"""Record the content of ie from tree into the commit if needed.
1633
Side effect: sets ie.revision when unchanged
1635
:param ie: An inventory entry present in the commit.
1636
:param parent_invs: The inventories of the parent revisions of the
1638
:param path: The path the entry is at in the tree.
1639
:param tree: The tree which contains this entry and should be used to
1642
if self.new_inventory.root is None and ie.parent_id is not None:
1643
symbol_versioning.warn('Root entry should be supplied to'
1644
' record_entry_contents, as of bzr 0.10.',
1645
DeprecationWarning, stacklevel=2)
1646
self.record_entry_contents(tree.inventory.root.copy(), parent_invs,
1648
self.new_inventory.add(ie)
1650
# ie.revision is always None if the InventoryEntry is considered
1651
# for committing. ie.snapshot will record the correct revision
1652
# which may be the sole parent if it is untouched.
1653
if ie.revision is not None:
1656
# In this revision format, root entries have no knit or weave
1657
if ie is self.new_inventory.root:
1658
# When serializing out to disk and back in
1659
# root.revision is always _new_revision_id
1660
ie.revision = self._new_revision_id
1662
previous_entries = ie.find_previous_heads(
1664
self.repository.weave_store,
1665
self.repository.get_transaction())
1666
# we are creating a new revision for ie in the history store
1668
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
1670
def modified_directory(self, file_id, file_parents):
1671
"""Record the presence of a symbolic link.
1673
:param file_id: The file_id of the link to record.
1674
:param file_parents: The per-file parent revision ids.
1676
self._add_text_to_weave(file_id, [], file_parents.keys())
1678
def modified_file_text(self, file_id, file_parents,
1679
get_content_byte_lines, text_sha1=None,
1681
"""Record the text of file file_id
1683
:param file_id: The file_id of the file to record the text of.
1684
:param file_parents: The per-file parent revision ids.
1685
:param get_content_byte_lines: A callable which will return the byte
1687
:param text_sha1: Optional SHA1 of the file contents.
1688
:param text_size: Optional size of the file contents.
1690
# mutter('storing text of file {%s} in revision {%s} into %r',
1691
# file_id, self._new_revision_id, self.repository.weave_store)
1692
# special case to avoid diffing on renames or
1694
if (len(file_parents) == 1
1695
and text_sha1 == file_parents.values()[0].text_sha1
1696
and text_size == file_parents.values()[0].text_size):
1697
previous_ie = file_parents.values()[0]
1698
versionedfile = self.repository.weave_store.get_weave(file_id,
1699
self.repository.get_transaction())
1700
versionedfile.clone_text(self._new_revision_id,
1701
previous_ie.revision, file_parents.keys())
1702
return text_sha1, text_size
1704
new_lines = get_content_byte_lines()
1705
# TODO: Rather than invoking sha_strings here, _add_text_to_weave
1706
# should return the SHA1 and size
1707
self._add_text_to_weave(file_id, new_lines, file_parents.keys())
1708
return osutils.sha_strings(new_lines), \
1709
sum(map(len, new_lines))
1711
def modified_link(self, file_id, file_parents, link_target):
1712
"""Record the presence of a symbolic link.
1714
:param file_id: The file_id of the link to record.
1715
:param file_parents: The per-file parent revision ids.
1716
:param link_target: Target location of this link.
1718
self._add_text_to_weave(file_id, [], file_parents.keys())
1720
def _add_text_to_weave(self, file_id, new_lines, parents):
1721
versionedfile = self.repository.weave_store.get_weave_or_empty(
1722
file_id, self.repository.get_transaction())
1723
versionedfile.add_lines(self._new_revision_id, parents, new_lines)
1724
versionedfile.clear_cache()
1727
class _CommitBuilder(CommitBuilder):
1728
"""Temporary class so old CommitBuilders are detected properly
1730
Note: CommitBuilder works whether or not root entry is recorded.
1733
record_root_entry = True
1736
class RootCommitBuilder(CommitBuilder):
1737
"""This commitbuilder actually records the root id"""
1739
record_root_entry = True
1741
def record_entry_contents(self, ie, parent_invs, path, tree):
1742
"""Record the content of ie from tree into the commit if needed.
1744
Side effect: sets ie.revision when unchanged
1746
:param ie: An inventory entry present in the commit.
1747
:param parent_invs: The inventories of the parent revisions of the
1749
:param path: The path the entry is at in the tree.
1750
:param tree: The tree which contains this entry and should be used to
1753
assert self.new_inventory.root is not None or ie.parent_id is None
1754
self.new_inventory.add(ie)
1756
# ie.revision is always None if the InventoryEntry is considered
1757
# for committing. ie.snapshot will record the correct revision
1758
# which may be the sole parent if it is untouched.
1759
if ie.revision is not None:
1762
previous_entries = ie.find_previous_heads(
1764
self.repository.weave_store,
1765
self.repository.get_transaction())
1766
# we are creating a new revision for ie in the history store
1768
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
4109
1771
_unescape_map = {
4134
1790
if _unescape_re is None:
4135
1791
_unescape_re = re.compile('\&([^;]*);')
4136
1792
return _unescape_re.sub(_unescaper, data)
4139
class _VersionedFileChecker(object):
4141
def __init__(self, repository, text_key_references=None, ancestors=None):
4142
self.repository = repository
4143
self.text_index = self.repository._generate_text_key_index(
4144
text_key_references=text_key_references, ancestors=ancestors)
4146
def calculate_file_version_parents(self, text_key):
4147
"""Calculate the correct parents for a file version according to
4150
parent_keys = self.text_index[text_key]
4151
if parent_keys == [_mod_revision.NULL_REVISION]:
4153
return tuple(parent_keys)
4155
def check_file_version_parents(self, texts, progress_bar=None):
4156
"""Check the parents stored in a versioned file are correct.
4158
It also detects file versions that are not referenced by their
4159
corresponding revision's inventory.
4161
:returns: A tuple of (wrong_parents, dangling_file_versions).
4162
wrong_parents is a dict mapping {revision_id: (stored_parents,
4163
correct_parents)} for each revision_id where the stored parents
4164
are not correct. dangling_file_versions is a set of (file_id,
4165
revision_id) tuples for versions that are present in this versioned
4166
file, but not used by the corresponding inventory.
4168
local_progress = None
4169
if progress_bar is None:
4170
local_progress = ui.ui_factory.nested_progress_bar()
4171
progress_bar = local_progress
4173
return self._check_file_version_parents(texts, progress_bar)
4176
local_progress.finished()
4178
def _check_file_version_parents(self, texts, progress_bar):
4179
"""See check_file_version_parents."""
4181
self.file_ids = set([file_id for file_id, _ in
4182
self.text_index.iterkeys()])
4183
# text keys is now grouped by file_id
4184
n_versions = len(self.text_index)
4185
progress_bar.update('loading text store', 0, n_versions)
4186
parent_map = self.repository.texts.get_parent_map(self.text_index)
4187
# On unlistable transports this could well be empty/error...
4188
text_keys = self.repository.texts.keys()
4189
unused_keys = frozenset(text_keys) - set(self.text_index)
4190
for num, key in enumerate(self.text_index.iterkeys()):
4191
progress_bar.update('checking text graph', num, n_versions)
4192
correct_parents = self.calculate_file_version_parents(key)
4194
knit_parents = parent_map[key]
4195
except errors.RevisionNotPresent:
4198
if correct_parents != knit_parents:
4199
wrong_parents[key] = (knit_parents, correct_parents)
4200
return wrong_parents, unused_keys
4203
def _old_get_graph(repository, revision_id):
4204
"""DO NOT USE. That is all. I'm serious."""
4205
graph = repository.get_graph()
4206
revision_graph = dict(((key, value) for key, value in
4207
graph.iter_ancestry([revision_id]) if value is not None))
4208
return _strip_NULL_ghosts(revision_graph)
4211
def _strip_NULL_ghosts(revision_graph):
4212
"""Also don't use this. more compatibility code for unmigrated clients."""
4213
# Filter ghosts, and null:
4214
if _mod_revision.NULL_REVISION in revision_graph:
4215
del revision_graph[_mod_revision.NULL_REVISION]
4216
for key, parents in revision_graph.items():
4217
revision_graph[key] = tuple(parent for parent in parents if parent
4219
return revision_graph
4222
class StreamSink(object):
4223
"""An object that can insert a stream into a repository.
4225
This interface handles the complexity of reserialising inventories and
4226
revisions from different formats, and allows unidirectional insertion into
4227
stacked repositories without looking for the missing basis parents
4231
def __init__(self, target_repo):
4232
self.target_repo = target_repo
4234
def insert_stream(self, stream, src_format, resume_tokens):
4235
"""Insert a stream's content into the target repository.
4237
:param src_format: a bzr repository format.
4239
:return: a list of resume tokens and an iterable of keys additional
4240
items required before the insertion can be completed.
4242
self.target_repo.lock_write()
4245
self.target_repo.resume_write_group(resume_tokens)
4248
self.target_repo.start_write_group()
4251
# locked_insert_stream performs a commit|suspend.
4252
return self._locked_insert_stream(stream, src_format, is_resume)
4254
self.target_repo.abort_write_group(suppress_errors=True)
4257
self.target_repo.unlock()
4259
def _locked_insert_stream(self, stream, src_format, is_resume):
4260
to_serializer = self.target_repo._format._serializer
4261
src_serializer = src_format._serializer
4263
if to_serializer == src_serializer:
4264
# If serializers match and the target is a pack repository, set the
4265
# write cache size on the new pack. This avoids poor performance
4266
# on transports where append is unbuffered (such as
4267
# RemoteTransport). This is safe to do because nothing should read
4268
# back from the target repository while a stream with matching
4269
# serialization is being inserted.
4270
# The exception is that a delta record from the source that should
4271
# be a fulltext may need to be expanded by the target (see
4272
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4273
# explicitly flush any buffered writes first in that rare case.
4275
new_pack = self.target_repo._pack_collection._new_pack
4276
except AttributeError:
4277
# Not a pack repository
4280
new_pack.set_write_cache_size(1024*1024)
4281
for substream_type, substream in stream:
4282
if 'stream' in debug.debug_flags:
4283
mutter('inserting substream: %s', substream_type)
4284
if substream_type == 'texts':
4285
self.target_repo.texts.insert_record_stream(substream)
4286
elif substream_type == 'inventories':
4287
if src_serializer == to_serializer:
4288
self.target_repo.inventories.insert_record_stream(
4291
self._extract_and_insert_inventories(
4292
substream, src_serializer)
4293
elif substream_type == 'inventory-deltas':
4294
self._extract_and_insert_inventory_deltas(
4295
substream, src_serializer)
4296
elif substream_type == 'chk_bytes':
4297
# XXX: This doesn't support conversions, as it assumes the
4298
# conversion was done in the fetch code.
4299
self.target_repo.chk_bytes.insert_record_stream(substream)
4300
elif substream_type == 'revisions':
4301
# This may fallback to extract-and-insert more often than
4302
# required if the serializers are different only in terms of
4304
if src_serializer == to_serializer:
4305
self.target_repo.revisions.insert_record_stream(
4308
self._extract_and_insert_revisions(substream,
4310
elif substream_type == 'signatures':
4311
self.target_repo.signatures.insert_record_stream(substream)
4313
raise AssertionError('kaboom! %s' % (substream_type,))
4314
# Done inserting data, and the missing_keys calculations will try to
4315
# read back from the inserted data, so flush the writes to the new pack
4316
# (if this is pack format).
4317
if new_pack is not None:
4318
new_pack._write_data('', flush=True)
4319
# Find all the new revisions (including ones from resume_tokens)
4320
missing_keys = self.target_repo.get_missing_parent_inventories(
4321
check_for_missing_texts=is_resume)
4323
for prefix, versioned_file in (
4324
('texts', self.target_repo.texts),
4325
('inventories', self.target_repo.inventories),
4326
('revisions', self.target_repo.revisions),
4327
('signatures', self.target_repo.signatures),
4328
('chk_bytes', self.target_repo.chk_bytes),
4330
if versioned_file is None:
4332
# TODO: key is often going to be a StaticTuple object
4333
# I don't believe we can define a method by which
4334
# (prefix,) + StaticTuple will work, though we could
4335
# define a StaticTuple.sq_concat that would allow you to
4336
# pass in either a tuple or a StaticTuple as the second
4337
# object, so instead we could have:
4338
# StaticTuple(prefix) + key here...
4339
missing_keys.update((prefix,) + key for key in
4340
versioned_file.get_missing_compression_parent_keys())
4341
except NotImplementedError:
4342
# cannot even attempt suspending, and missing would have failed
4343
# during stream insertion.
4344
missing_keys = set()
4347
# suspend the write group and tell the caller what we is
4348
# missing. We know we can suspend or else we would not have
4349
# entered this code path. (All repositories that can handle
4350
# missing keys can handle suspending a write group).
4351
write_group_tokens = self.target_repo.suspend_write_group()
4352
return write_group_tokens, missing_keys
4353
hint = self.target_repo.commit_write_group()
4354
if (to_serializer != src_serializer and
4355
self.target_repo._format.pack_compresses):
4356
self.target_repo.pack(hint=hint)
4359
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4360
target_rich_root = self.target_repo._format.rich_root_data
4361
target_tree_refs = self.target_repo._format.supports_tree_reference
4362
for record in substream:
4363
# Insert the delta directly
4364
inventory_delta_bytes = record.get_bytes_as('fulltext')
4365
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4367
parse_result = deserialiser.parse_text_bytes(
4368
inventory_delta_bytes)
4369
except inventory_delta.IncompatibleInventoryDelta, err:
4370
trace.mutter("Incompatible delta: %s", err.msg)
4371
raise errors.IncompatibleRevision(self.target_repo._format)
4372
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4373
revision_id = new_id
4374
parents = [key[0] for key in record.parents]
4375
self.target_repo.add_inventory_by_delta(
4376
basis_id, inv_delta, revision_id, parents)
4378
def _extract_and_insert_inventories(self, substream, serializer,
4380
"""Generate a new inventory versionedfile in target, converting data.
4382
The inventory is retrieved from the source, (deserializing it), and
4383
stored in the target (reserializing it in a different format).
4385
target_rich_root = self.target_repo._format.rich_root_data
4386
target_tree_refs = self.target_repo._format.supports_tree_reference
4387
for record in substream:
4388
# It's not a delta, so it must be a fulltext in the source
4389
# serializer's format.
4390
bytes = record.get_bytes_as('fulltext')
4391
revision_id = record.key[0]
4392
inv = serializer.read_inventory_from_string(bytes, revision_id)
4393
parents = [key[0] for key in record.parents]
4394
self.target_repo.add_inventory(revision_id, inv, parents)
4395
# No need to keep holding this full inv in memory when the rest of
4396
# the substream is likely to be all deltas.
4399
def _extract_and_insert_revisions(self, substream, serializer):
4400
for record in substream:
4401
bytes = record.get_bytes_as('fulltext')
4402
revision_id = record.key[0]
4403
rev = serializer.read_revision_from_string(bytes)
4404
if rev.revision_id != revision_id:
4405
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4406
self.target_repo.add_revision(revision_id, rev)
4409
if self.target_repo._format._fetch_reconcile:
4410
self.target_repo.reconcile()
4413
class StreamSource(object):
4414
"""A source of a stream for fetching between repositories."""
4416
def __init__(self, from_repository, to_format):
4417
"""Create a StreamSource streaming from from_repository."""
4418
self.from_repository = from_repository
4419
self.to_format = to_format
4421
def delta_on_metadata(self):
4422
"""Return True if delta's are permitted on metadata streams.
4424
That is on revisions and signatures.
4426
src_serializer = self.from_repository._format._serializer
4427
target_serializer = self.to_format._serializer
4428
return (self.to_format._fetch_uses_deltas and
4429
src_serializer == target_serializer)
4431
def _fetch_revision_texts(self, revs):
4432
# fetch signatures first and then the revision texts
4433
# may need to be a InterRevisionStore call here.
4434
from_sf = self.from_repository.signatures
4435
# A missing signature is just skipped.
4436
keys = [(rev_id,) for rev_id in revs]
4437
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4439
self.to_format._fetch_order,
4440
not self.to_format._fetch_uses_deltas))
4441
# If a revision has a delta, this is actually expanded inside the
4442
# insert_record_stream code now, which is an alternate fix for
4444
from_rf = self.from_repository.revisions
4445
revisions = from_rf.get_record_stream(
4447
self.to_format._fetch_order,
4448
not self.delta_on_metadata())
4449
return [('signatures', signatures), ('revisions', revisions)]
4451
def _generate_root_texts(self, revs):
4452
"""This will be called by get_stream between fetching weave texts and
4453
fetching the inventory weave.
4455
if self._rich_root_upgrade():
4456
return _mod_fetch.Inter1and2Helper(
4457
self.from_repository).generate_root_texts(revs)
4461
def get_stream(self, search):
4463
revs = search.get_keys()
4464
graph = self.from_repository.get_graph()
4465
revs = tsort.topo_sort(graph.get_parent_map(revs))
4466
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4468
for knit_kind, file_id, revisions in data_to_fetch:
4469
if knit_kind != phase:
4471
# Make a new progress bar for this phase
4472
if knit_kind == "file":
4473
# Accumulate file texts
4474
text_keys.extend([(file_id, revision) for revision in
4476
elif knit_kind == "inventory":
4477
# Now copy the file texts.
4478
from_texts = self.from_repository.texts
4479
yield ('texts', from_texts.get_record_stream(
4480
text_keys, self.to_format._fetch_order,
4481
not self.to_format._fetch_uses_deltas))
4482
# Cause an error if a text occurs after we have done the
4485
# Before we process the inventory we generate the root
4486
# texts (if necessary) so that the inventories references
4488
for _ in self._generate_root_texts(revs):
4490
# we fetch only the referenced inventories because we do not
4491
# know for unselected inventories whether all their required
4492
# texts are present in the other repository - it could be
4494
for info in self._get_inventory_stream(revs):
4496
elif knit_kind == "signatures":
4497
# Nothing to do here; this will be taken care of when
4498
# _fetch_revision_texts happens.
4500
elif knit_kind == "revisions":
4501
for record in self._fetch_revision_texts(revs):
4504
raise AssertionError("Unknown knit kind %r" % knit_kind)
4506
def get_stream_for_missing_keys(self, missing_keys):
4507
# missing keys can only occur when we are byte copying and not
4508
# translating (because translation means we don't send
4509
# unreconstructable deltas ever).
4511
keys['texts'] = set()
4512
keys['revisions'] = set()
4513
keys['inventories'] = set()
4514
keys['chk_bytes'] = set()
4515
keys['signatures'] = set()
4516
for key in missing_keys:
4517
keys[key[0]].add(key[1:])
4518
if len(keys['revisions']):
4519
# If we allowed copying revisions at this point, we could end up
4520
# copying a revision without copying its required texts: a
4521
# violation of the requirements for repository integrity.
4522
raise AssertionError(
4523
'cannot copy revisions to fill in missing deltas %s' % (
4524
keys['revisions'],))
4525
for substream_kind, keys in keys.iteritems():
4526
vf = getattr(self.from_repository, substream_kind)
4527
if vf is None and keys:
4528
raise AssertionError(
4529
"cannot fill in keys for a versioned file we don't"
4530
" have: %s needs %s" % (substream_kind, keys))
4532
# No need to stream something we don't have
4534
if substream_kind == 'inventories':
4535
# Some missing keys are genuinely ghosts, filter those out.
4536
present = self.from_repository.inventories.get_parent_map(keys)
4537
revs = [key[0] for key in present]
4538
# Get the inventory stream more-or-less as we do for the
4539
# original stream; there's no reason to assume that records
4540
# direct from the source will be suitable for the sink. (Think
4541
# e.g. 2a -> 1.9-rich-root).
4542
for info in self._get_inventory_stream(revs, missing=True):
4546
# Ask for full texts always so that we don't need more round trips
4547
# after this stream.
4548
# Some of the missing keys are genuinely ghosts, so filter absent
4549
# records. The Sink is responsible for doing another check to
4550
# ensure that ghosts don't introduce missing data for future
4552
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4553
self.to_format._fetch_order, True))
4554
yield substream_kind, stream
4556
def inventory_fetch_order(self):
4557
if self._rich_root_upgrade():
4558
return 'topological'
4560
return self.to_format._fetch_order
4562
def _rich_root_upgrade(self):
4563
return (not self.from_repository._format.rich_root_data and
4564
self.to_format.rich_root_data)
4566
def _get_inventory_stream(self, revision_ids, missing=False):
4567
from_format = self.from_repository._format
4568
if (from_format.supports_chks and self.to_format.supports_chks and
4569
from_format.network_name() == self.to_format.network_name()):
4570
raise AssertionError(
4571
"this case should be handled by GroupCHKStreamSource")
4572
elif 'forceinvdeltas' in debug.debug_flags:
4573
return self._get_convertable_inventory_stream(revision_ids,
4574
delta_versus_null=missing)
4575
elif from_format.network_name() == self.to_format.network_name():
4577
return self._get_simple_inventory_stream(revision_ids,
4579
elif (not from_format.supports_chks and not self.to_format.supports_chks
4580
and from_format._serializer == self.to_format._serializer):
4581
# Essentially the same format.
4582
return self._get_simple_inventory_stream(revision_ids,
4585
# Any time we switch serializations, we want to use an
4586
# inventory-delta based approach.
4587
return self._get_convertable_inventory_stream(revision_ids,
4588
delta_versus_null=missing)
4590
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4591
# NB: This currently reopens the inventory weave in source;
4592
# using a single stream interface instead would avoid this.
4593
from_weave = self.from_repository.inventories
4595
delta_closure = True
4597
delta_closure = not self.delta_on_metadata()
4598
yield ('inventories', from_weave.get_record_stream(
4599
[(rev_id,) for rev_id in revision_ids],
4600
self.inventory_fetch_order(), delta_closure))
4602
def _get_convertable_inventory_stream(self, revision_ids,
4603
delta_versus_null=False):
4604
# The two formats are sufficiently different that there is no fast
4605
# path, so we need to send just inventorydeltas, which any
4606
# sufficiently modern client can insert into any repository.
4607
# The StreamSink code expects to be able to
4608
# convert on the target, so we need to put bytes-on-the-wire that can
4609
# be converted. That means inventory deltas (if the remote is <1.19,
4610
# RemoteStreamSink will fallback to VFS to insert the deltas).
4611
yield ('inventory-deltas',
4612
self._stream_invs_as_deltas(revision_ids,
4613
delta_versus_null=delta_versus_null))
4615
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4616
"""Return a stream of inventory-deltas for the given rev ids.
4618
:param revision_ids: The list of inventories to transmit
4619
:param delta_versus_null: Don't try to find a minimal delta for this
4620
entry, instead compute the delta versus the NULL_REVISION. This
4621
effectively streams a complete inventory. Used for stuff like
4622
filling in missing parents, etc.
4624
from_repo = self.from_repository
4625
revision_keys = [(rev_id,) for rev_id in revision_ids]
4626
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4627
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4629
inventories = self.from_repository.iter_inventories(
4630
revision_ids, 'topological')
4631
format = from_repo._format
4632
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4633
inventory_cache = lru_cache.LRUCache(50)
4634
null_inventory = from_repo.revision_tree(
4635
_mod_revision.NULL_REVISION).inventory
4636
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4637
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4638
# repo back into a non-rich-root repo ought to be allowed)
4639
serializer = inventory_delta.InventoryDeltaSerializer(
4640
versioned_root=format.rich_root_data,
4641
tree_references=format.supports_tree_reference)
4642
for inv in inventories:
4643
key = (inv.revision_id,)
4644
parent_keys = parent_map.get(key, ())
4646
if not delta_versus_null and parent_keys:
4647
# The caller did not ask for complete inventories and we have
4648
# some parents that we can delta against. Make a delta against
4649
# each parent so that we can find the smallest.
4650
parent_ids = [parent_key[0] for parent_key in parent_keys]
4651
for parent_id in parent_ids:
4652
if parent_id not in invs_sent_so_far:
4653
# We don't know that the remote side has this basis, so
4656
if parent_id == _mod_revision.NULL_REVISION:
4657
parent_inv = null_inventory
4659
parent_inv = inventory_cache.get(parent_id, None)
4660
if parent_inv is None:
4661
parent_inv = from_repo.get_inventory(parent_id)
4662
candidate_delta = inv._make_delta(parent_inv)
4663
if (delta is None or
4664
len(delta) > len(candidate_delta)):
4665
delta = candidate_delta
4666
basis_id = parent_id
4668
# Either none of the parents ended up being suitable, or we
4669
# were asked to delta against NULL
4670
basis_id = _mod_revision.NULL_REVISION
4671
delta = inv._make_delta(null_inventory)
4672
invs_sent_so_far.add(inv.revision_id)
4673
inventory_cache[inv.revision_id] = inv
4674
delta_serialized = ''.join(
4675
serializer.delta_to_lines(basis_id, key[-1], delta))
4676
yield versionedfile.FulltextContentFactory(
4677
key, parent_keys, None, delta_serialized)
4680
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4681
stop_revision=None):
4682
"""Extend the partial history to include a given index
4684
If a stop_index is supplied, stop when that index has been reached.
4685
If a stop_revision is supplied, stop when that revision is
4686
encountered. Otherwise, stop when the beginning of history is
4689
:param stop_index: The index which should be present. When it is
4690
present, history extension will stop.
4691
:param stop_revision: The revision id which should be present. When
4692
it is encountered, history extension will stop.
4694
start_revision = partial_history_cache[-1]
4695
iterator = repo.iter_reverse_revision_history(start_revision)
4697
#skip the last revision in the list
4700
if (stop_index is not None and
4701
len(partial_history_cache) > stop_index):
4703
if partial_history_cache[-1] == stop_revision:
4705
revision_id = iterator.next()
4706
partial_history_cache.append(revision_id)
4707
except StopIteration: