13
13
# You should have received a copy of the GNU General Public License
14
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
from cStringIO import StringIO
17
19
from bzrlib.lazy_import import lazy_import
18
20
lazy_import(globals(), """
23
24
from bzrlib import (
42
38
revision as _mod_revision,
49
43
from bzrlib.bundle import serializer
50
44
from bzrlib.revisiontree import RevisionTree
51
45
from bzrlib.store.versioned import VersionedFileStore
46
from bzrlib.store.text import TextStore
52
47
from bzrlib.testament import Testament
55
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
51
from bzrlib.decorators import needs_read_lock, needs_write_lock
56
52
from bzrlib.inter import InterObject
57
from bzrlib.inventory import (
63
from bzrlib.lock import _RelockDebugMixin
64
from bzrlib import registry
65
from bzrlib.trace import (
66
log_exception_quietly, note, mutter, mutter_callsite, warning)
53
from bzrlib.inventory import Inventory, InventoryDirectory, ROOT_ID
54
from bzrlib.symbol_versioning import (
58
from bzrlib.trace import mutter, note, warning
69
61
# Old formats display a warning, but only once
70
62
_deprecation_warning_done = False
73
class CommitBuilder(object):
74
"""Provides an interface to build up a commit.
76
This allows describing a tree to be committed without needing to
77
know the internals of the format of the repository.
80
# all clients should supply tree roots.
81
record_root_entry = True
82
# the default CommitBuilder does not manage trees whose root is versioned.
83
_versioned_root = False
85
def __init__(self, repository, parents, config, timestamp=None,
86
timezone=None, committer=None, revprops=None,
88
"""Initiate a CommitBuilder.
90
:param repository: Repository to commit to.
91
:param parents: Revision ids of the parents of the new revision.
92
:param config: Configuration to use.
93
:param timestamp: Optional timestamp recorded for commit.
94
:param timezone: Optional timezone for timestamp.
95
:param committer: Optional committer to set for commit.
96
:param revprops: Optional dictionary of revision properties.
97
:param revision_id: Optional revision id.
101
if committer is None:
102
self._committer = self._config.username()
104
self._committer = committer
106
self.new_inventory = Inventory(None)
107
self._new_revision_id = revision_id
108
self.parents = parents
109
self.repository = repository
112
if revprops is not None:
113
self._validate_revprops(revprops)
114
self._revprops.update(revprops)
116
if timestamp is None:
117
timestamp = time.time()
118
# Restrict resolution to 1ms
119
self._timestamp = round(timestamp, 3)
122
self._timezone = osutils.local_time_offset()
124
self._timezone = int(timezone)
126
self._generate_revision_if_needed()
127
self.__heads = graph.HeadsCache(repository.get_graph()).heads
128
self._basis_delta = []
129
# API compatibility, older code that used CommitBuilder did not call
130
# .record_delete(), which means the delta that is computed would not be
131
# valid. Callers that will call record_delete() should call
132
# .will_record_deletes() to indicate that.
133
self._recording_deletes = False
134
# memo'd check for no-op commits.
135
self._any_changes = False
137
def any_changes(self):
138
"""Return True if any entries were changed.
140
This includes merge-only changes. It is the core for the --unchanged
143
:return: True if any changes have occured.
145
return self._any_changes
147
def _validate_unicode_text(self, text, context):
148
"""Verify things like commit messages don't have bogus characters."""
150
raise ValueError('Invalid value for %s: %r' % (context, text))
152
def _validate_revprops(self, revprops):
153
for key, value in revprops.iteritems():
154
# We know that the XML serializers do not round trip '\r'
155
# correctly, so refuse to accept them
156
if not isinstance(value, basestring):
157
raise ValueError('revision property (%s) is not a valid'
158
' (unicode) string: %r' % (key, value))
159
self._validate_unicode_text(value,
160
'revision property (%s)' % (key,))
162
def commit(self, message):
163
"""Make the actual commit.
165
:return: The revision id of the recorded revision.
167
self._validate_unicode_text(message, 'commit message')
168
rev = _mod_revision.Revision(
169
timestamp=self._timestamp,
170
timezone=self._timezone,
171
committer=self._committer,
173
inventory_sha1=self.inv_sha1,
174
revision_id=self._new_revision_id,
175
properties=self._revprops)
176
rev.parent_ids = self.parents
177
self.repository.add_revision(self._new_revision_id, rev,
178
self.new_inventory, self._config)
179
self.repository.commit_write_group()
180
return self._new_revision_id
183
"""Abort the commit that is being built.
185
self.repository.abort_write_group()
187
def revision_tree(self):
188
"""Return the tree that was just committed.
190
After calling commit() this can be called to get a RevisionTree
191
representing the newly committed tree. This is preferred to
192
calling Repository.revision_tree() because that may require
193
deserializing the inventory, while we already have a copy in
196
if self.new_inventory is None:
197
self.new_inventory = self.repository.get_inventory(
198
self._new_revision_id)
199
return RevisionTree(self.repository, self.new_inventory,
200
self._new_revision_id)
202
def finish_inventory(self):
203
"""Tell the builder that the inventory is finished.
205
:return: The inventory id in the repository, which can be used with
206
repository.get_inventory.
208
if self.new_inventory is None:
209
# an inventory delta was accumulated without creating a new
211
basis_id = self.basis_delta_revision
212
# We ignore the 'inventory' returned by add_inventory_by_delta
213
# because self.new_inventory is used to hint to the rest of the
214
# system what code path was taken
215
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
216
basis_id, self._basis_delta, self._new_revision_id,
219
if self.new_inventory.root is None:
220
raise AssertionError('Root entry should be supplied to'
221
' record_entry_contents, as of bzr 0.10.')
222
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
223
self.new_inventory.revision_id = self._new_revision_id
224
self.inv_sha1 = self.repository.add_inventory(
225
self._new_revision_id,
229
return self._new_revision_id
231
def _gen_revision_id(self):
232
"""Return new revision-id."""
233
return generate_ids.gen_revision_id(self._config.username(),
236
def _generate_revision_if_needed(self):
237
"""Create a revision id if None was supplied.
239
If the repository can not support user-specified revision ids
240
they should override this function and raise CannotSetRevisionId
241
if _new_revision_id is not None.
243
:raises: CannotSetRevisionId
245
if self._new_revision_id is None:
246
self._new_revision_id = self._gen_revision_id()
247
self.random_revid = True
249
self.random_revid = False
251
def _heads(self, file_id, revision_ids):
252
"""Calculate the graph heads for revision_ids in the graph of file_id.
254
This can use either a per-file graph or a global revision graph as we
255
have an identity relationship between the two graphs.
257
return self.__heads(revision_ids)
259
def _check_root(self, ie, parent_invs, tree):
260
"""Helper for record_entry_contents.
262
:param ie: An entry being added.
263
:param parent_invs: The inventories of the parent revisions of the
265
:param tree: The tree that is being committed.
267
# In this revision format, root entries have no knit or weave When
268
# serializing out to disk and back in root.revision is always
270
ie.revision = self._new_revision_id
272
def _require_root_change(self, tree):
273
"""Enforce an appropriate root object change.
275
This is called once when record_iter_changes is called, if and only if
276
the root was not in the delta calculated by record_iter_changes.
278
:param tree: The tree which is being committed.
280
# NB: if there are no parents then this method is not called, so no
281
# need to guard on parents having length.
282
entry = entry_factory['directory'](tree.path2id(''), '',
284
entry.revision = self._new_revision_id
285
self._basis_delta.append(('', '', entry.file_id, entry))
287
def _get_delta(self, ie, basis_inv, path):
288
"""Get a delta against the basis inventory for ie."""
289
if ie.file_id not in basis_inv:
291
result = (None, path, ie.file_id, ie)
292
self._basis_delta.append(result)
294
elif ie != basis_inv[ie.file_id]:
296
# TODO: avoid tis id2path call.
297
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
298
self._basis_delta.append(result)
304
def get_basis_delta(self):
305
"""Return the complete inventory delta versus the basis inventory.
307
This has been built up with the calls to record_delete and
308
record_entry_contents. The client must have already called
309
will_record_deletes() to indicate that they will be generating a
312
:return: An inventory delta, suitable for use with apply_delta, or
313
Repository.add_inventory_by_delta, etc.
315
if not self._recording_deletes:
316
raise AssertionError("recording deletes not activated.")
317
return self._basis_delta
319
def record_delete(self, path, file_id):
320
"""Record that a delete occured against a basis tree.
322
This is an optional API - when used it adds items to the basis_delta
323
being accumulated by the commit builder. It cannot be called unless the
324
method will_record_deletes() has been called to inform the builder that
325
a delta is being supplied.
327
:param path: The path of the thing deleted.
328
:param file_id: The file id that was deleted.
330
if not self._recording_deletes:
331
raise AssertionError("recording deletes not activated.")
332
delta = (path, None, file_id, None)
333
self._basis_delta.append(delta)
334
self._any_changes = True
337
def will_record_deletes(self):
338
"""Tell the commit builder that deletes are being notified.
340
This enables the accumulation of an inventory delta; for the resulting
341
commit to be valid, deletes against the basis MUST be recorded via
342
builder.record_delete().
344
self._recording_deletes = True
346
basis_id = self.parents[0]
348
basis_id = _mod_revision.NULL_REVISION
349
self.basis_delta_revision = basis_id
351
def record_entry_contents(self, ie, parent_invs, path, tree,
353
"""Record the content of ie from tree into the commit if needed.
355
Side effect: sets ie.revision when unchanged
357
:param ie: An inventory entry present in the commit.
358
:param parent_invs: The inventories of the parent revisions of the
360
:param path: The path the entry is at in the tree.
361
:param tree: The tree which contains this entry and should be used to
363
:param content_summary: Summary data from the tree about the paths
364
content - stat, length, exec, sha/link target. This is only
365
accessed when the entry has a revision of None - that is when it is
366
a candidate to commit.
367
:return: A tuple (change_delta, version_recorded, fs_hash).
368
change_delta is an inventory_delta change for this entry against
369
the basis tree of the commit, or None if no change occured against
371
version_recorded is True if a new version of the entry has been
372
recorded. For instance, committing a merge where a file was only
373
changed on the other side will return (delta, False).
374
fs_hash is either None, or the hash details for the path (currently
375
a tuple of the contents sha1 and the statvalue returned by
376
tree.get_file_with_stat()).
378
if self.new_inventory.root is None:
379
if ie.parent_id is not None:
380
raise errors.RootMissing()
381
self._check_root(ie, parent_invs, tree)
382
if ie.revision is None:
383
kind = content_summary[0]
385
# ie is carried over from a prior commit
387
# XXX: repository specific check for nested tree support goes here - if
388
# the repo doesn't want nested trees we skip it ?
389
if (kind == 'tree-reference' and
390
not self.repository._format.supports_tree_reference):
391
# mismatch between commit builder logic and repository:
392
# this needs the entry creation pushed down into the builder.
393
raise NotImplementedError('Missing repository subtree support.')
394
self.new_inventory.add(ie)
396
# TODO: slow, take it out of the inner loop.
398
basis_inv = parent_invs[0]
400
basis_inv = Inventory(root_id=None)
402
# ie.revision is always None if the InventoryEntry is considered
403
# for committing. We may record the previous parents revision if the
404
# content is actually unchanged against a sole head.
405
if ie.revision is not None:
406
if not self._versioned_root and path == '':
407
# repositories that do not version the root set the root's
408
# revision to the new commit even when no change occurs (more
409
# specifically, they do not record a revision on the root; and
410
# the rev id is assigned to the root during deserialisation -
411
# this masks when a change may have occurred against the basis.
412
# To match this we always issue a delta, because the revision
413
# of the root will always be changing.
414
if ie.file_id in basis_inv:
415
delta = (basis_inv.id2path(ie.file_id), path,
419
delta = (None, path, ie.file_id, ie)
420
self._basis_delta.append(delta)
421
return delta, False, None
423
# we don't need to commit this, because the caller already
424
# determined that an existing revision of this file is
425
# appropriate. If its not being considered for committing then
426
# it and all its parents to the root must be unaltered so
427
# no-change against the basis.
428
if ie.revision == self._new_revision_id:
429
raise AssertionError("Impossible situation, a skipped "
430
"inventory entry (%r) claims to be modified in this "
431
"commit (%r).", (ie, self._new_revision_id))
432
return None, False, None
433
# XXX: Friction: parent_candidates should return a list not a dict
434
# so that we don't have to walk the inventories again.
435
parent_candiate_entries = ie.parent_candidates(parent_invs)
436
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
438
for inv in parent_invs:
439
if ie.file_id in inv:
440
old_rev = inv[ie.file_id].revision
441
if old_rev in head_set:
442
heads.append(inv[ie.file_id].revision)
443
head_set.remove(inv[ie.file_id].revision)
446
# now we check to see if we need to write a new record to the
448
# We write a new entry unless there is one head to the ancestors, and
449
# the kind-derived content is unchanged.
451
# Cheapest check first: no ancestors, or more the one head in the
452
# ancestors, we write a new node.
456
# There is a single head, look it up for comparison
457
parent_entry = parent_candiate_entries[heads[0]]
458
# if the non-content specific data has changed, we'll be writing a
460
if (parent_entry.parent_id != ie.parent_id or
461
parent_entry.name != ie.name):
463
# now we need to do content specific checks:
465
# if the kind changed the content obviously has
466
if kind != parent_entry.kind:
468
# Stat cache fingerprint feedback for the caller - None as we usually
469
# don't generate one.
472
if content_summary[2] is None:
473
raise ValueError("Files must not have executable = None")
475
# We can't trust a check of the file length because of content
477
if (# if the exec bit has changed we have to store:
478
parent_entry.executable != content_summary[2]):
480
elif parent_entry.text_sha1 == content_summary[3]:
481
# all meta and content is unchanged (using a hash cache
482
# hit to check the sha)
483
ie.revision = parent_entry.revision
484
ie.text_size = parent_entry.text_size
485
ie.text_sha1 = parent_entry.text_sha1
486
ie.executable = parent_entry.executable
487
return self._get_delta(ie, basis_inv, path), False, None
489
# Either there is only a hash change(no hash cache entry,
490
# or same size content change), or there is no change on
492
# Provide the parent's hash to the store layer, so that the
493
# content is unchanged we will not store a new node.
494
nostore_sha = parent_entry.text_sha1
496
# We want to record a new node regardless of the presence or
497
# absence of a content change in the file.
499
ie.executable = content_summary[2]
500
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
502
text = file_obj.read()
506
ie.text_sha1, ie.text_size = self._add_text_to_weave(
507
ie.file_id, text, heads, nostore_sha)
508
# Let the caller know we generated a stat fingerprint.
509
fingerprint = (ie.text_sha1, stat_value)
510
except errors.ExistingContent:
511
# Turns out that the file content was unchanged, and we were
512
# only going to store a new node if it was changed. Carry over
514
ie.revision = parent_entry.revision
515
ie.text_size = parent_entry.text_size
516
ie.text_sha1 = parent_entry.text_sha1
517
ie.executable = parent_entry.executable
518
return self._get_delta(ie, basis_inv, path), False, None
519
elif kind == 'directory':
521
# all data is meta here, nothing specific to directory, so
523
ie.revision = parent_entry.revision
524
return self._get_delta(ie, basis_inv, path), False, None
525
self._add_text_to_weave(ie.file_id, '', heads, None)
526
elif kind == 'symlink':
527
current_link_target = content_summary[3]
529
# symlink target is not generic metadata, check if it has
531
if current_link_target != parent_entry.symlink_target:
534
# unchanged, carry over.
535
ie.revision = parent_entry.revision
536
ie.symlink_target = parent_entry.symlink_target
537
return self._get_delta(ie, basis_inv, path), False, None
538
ie.symlink_target = current_link_target
539
self._add_text_to_weave(ie.file_id, '', heads, None)
540
elif kind == 'tree-reference':
542
if content_summary[3] != parent_entry.reference_revision:
545
# unchanged, carry over.
546
ie.reference_revision = parent_entry.reference_revision
547
ie.revision = parent_entry.revision
548
return self._get_delta(ie, basis_inv, path), False, None
549
ie.reference_revision = content_summary[3]
550
if ie.reference_revision is None:
551
raise AssertionError("invalid content_summary for nested tree: %r"
552
% (content_summary,))
553
self._add_text_to_weave(ie.file_id, '', heads, None)
555
raise NotImplementedError('unknown kind')
556
ie.revision = self._new_revision_id
557
self._any_changes = True
558
return self._get_delta(ie, basis_inv, path), True, fingerprint
560
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
561
_entry_factory=entry_factory):
562
"""Record a new tree via iter_changes.
564
:param tree: The tree to obtain text contents from for changed objects.
565
:param basis_revision_id: The revision id of the tree the iter_changes
566
has been generated against. Currently assumed to be the same
567
as self.parents[0] - if it is not, errors may occur.
568
:param iter_changes: An iter_changes iterator with the changes to apply
569
to basis_revision_id. The iterator must not include any items with
570
a current kind of None - missing items must be either filtered out
571
or errored-on beefore record_iter_changes sees the item.
572
:param _entry_factory: Private method to bind entry_factory locally for
574
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
577
# Create an inventory delta based on deltas between all the parents and
578
# deltas between all the parent inventories. We use inventory delta's
579
# between the inventory objects because iter_changes masks
580
# last-changed-field only changes.
582
# file_id -> change map, change is fileid, paths, changed, versioneds,
583
# parents, names, kinds, executables
585
# {file_id -> revision_id -> inventory entry, for entries in parent
586
# trees that are not parents[0]
590
revtrees = list(self.repository.revision_trees(self.parents))
591
except errors.NoSuchRevision:
592
# one or more ghosts, slow path.
594
for revision_id in self.parents:
596
revtrees.append(self.repository.revision_tree(revision_id))
597
except errors.NoSuchRevision:
599
basis_revision_id = _mod_revision.NULL_REVISION
601
revtrees.append(self.repository.revision_tree(
602
_mod_revision.NULL_REVISION))
603
# The basis inventory from a repository
605
basis_inv = revtrees[0].inventory
607
basis_inv = self.repository.revision_tree(
608
_mod_revision.NULL_REVISION).inventory
609
if len(self.parents) > 0:
610
if basis_revision_id != self.parents[0] and not ghost_basis:
612
"arbitrary basis parents not yet supported with merges")
613
for revtree in revtrees[1:]:
614
for change in revtree.inventory._make_delta(basis_inv):
615
if change[1] is None:
616
# Not present in this parent.
618
if change[2] not in merged_ids:
619
if change[0] is not None:
620
basis_entry = basis_inv[change[2]]
621
merged_ids[change[2]] = [
623
basis_entry.revision,
626
parent_entries[change[2]] = {
628
basis_entry.revision:basis_entry,
630
change[3].revision:change[3],
633
merged_ids[change[2]] = [change[3].revision]
634
parent_entries[change[2]] = {change[3].revision:change[3]}
636
merged_ids[change[2]].append(change[3].revision)
637
parent_entries[change[2]][change[3].revision] = change[3]
640
# Setup the changes from the tree:
641
# changes maps file_id -> (change, [parent revision_ids])
643
for change in iter_changes:
644
# This probably looks up in basis_inv way to much.
645
if change[1][0] is not None:
646
head_candidate = [basis_inv[change[0]].revision]
649
changes[change[0]] = change, merged_ids.get(change[0],
651
unchanged_merged = set(merged_ids) - set(changes)
652
# Extend the changes dict with synthetic changes to record merges of
654
for file_id in unchanged_merged:
655
# Record a merged version of these items that did not change vs the
656
# basis. This can be either identical parallel changes, or a revert
657
# of a specific file after a merge. The recorded content will be
658
# that of the current tree (which is the same as the basis), but
659
# the per-file graph will reflect a merge.
660
# NB:XXX: We are reconstructing path information we had, this
661
# should be preserved instead.
662
# inv delta change: (file_id, (path_in_source, path_in_target),
663
# changed_content, versioned, parent, name, kind,
666
basis_entry = basis_inv[file_id]
667
except errors.NoSuchId:
668
# a change from basis->some_parents but file_id isn't in basis
669
# so was new in the merge, which means it must have changed
670
# from basis -> current, and as it hasn't the add was reverted
671
# by the user. So we discard this change.
675
(basis_inv.id2path(file_id), tree.id2path(file_id)),
677
(basis_entry.parent_id, basis_entry.parent_id),
678
(basis_entry.name, basis_entry.name),
679
(basis_entry.kind, basis_entry.kind),
680
(basis_entry.executable, basis_entry.executable))
681
changes[file_id] = (change, merged_ids[file_id])
682
# changes contains tuples with the change and a set of inventory
683
# candidates for the file.
685
# old_path, new_path, file_id, new_inventory_entry
686
seen_root = False # Is the root in the basis delta?
687
inv_delta = self._basis_delta
688
modified_rev = self._new_revision_id
689
for change, head_candidates in changes.values():
690
if change[3][1]: # versioned in target.
691
# Several things may be happening here:
692
# We may have a fork in the per-file graph
693
# - record a change with the content from tree
694
# We may have a change against < all trees
695
# - carry over the tree that hasn't changed
696
# We may have a change against all trees
697
# - record the change with the content from tree
700
entry = _entry_factory[kind](file_id, change[5][1],
702
head_set = self._heads(change[0], set(head_candidates))
705
for head_candidate in head_candidates:
706
if head_candidate in head_set:
707
heads.append(head_candidate)
708
head_set.remove(head_candidate)
711
# Could be a carry-over situation:
712
parent_entry_revs = parent_entries.get(file_id, None)
713
if parent_entry_revs:
714
parent_entry = parent_entry_revs.get(heads[0], None)
717
if parent_entry is None:
718
# The parent iter_changes was called against is the one
719
# that is the per-file head, so any change is relevant
720
# iter_changes is valid.
721
carry_over_possible = False
723
# could be a carry over situation
724
# A change against the basis may just indicate a merge,
725
# we need to check the content against the source of the
726
# merge to determine if it was changed after the merge
728
if (parent_entry.kind != entry.kind or
729
parent_entry.parent_id != entry.parent_id or
730
parent_entry.name != entry.name):
731
# Metadata common to all entries has changed
732
# against per-file parent
733
carry_over_possible = False
735
carry_over_possible = True
736
# per-type checks for changes against the parent_entry
739
# Cannot be a carry-over situation
740
carry_over_possible = False
741
# Populate the entry in the delta
743
# XXX: There is still a small race here: If someone reverts the content of a file
744
# after iter_changes examines and decides it has changed,
745
# we will unconditionally record a new version even if some
746
# other process reverts it while commit is running (with
747
# the revert happening after iter_changes did it's
750
entry.executable = True
752
entry.executable = False
753
if (carry_over_possible and
754
parent_entry.executable == entry.executable):
755
# Check the file length, content hash after reading
757
nostore_sha = parent_entry.text_sha1
760
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
762
text = file_obj.read()
766
entry.text_sha1, entry.text_size = self._add_text_to_weave(
767
file_id, text, heads, nostore_sha)
768
yield file_id, change[1][1], (entry.text_sha1, stat_value)
769
except errors.ExistingContent:
770
# No content change against a carry_over parent
771
# Perhaps this should also yield a fs hash update?
773
entry.text_size = parent_entry.text_size
774
entry.text_sha1 = parent_entry.text_sha1
775
elif kind == 'symlink':
777
entry.symlink_target = tree.get_symlink_target(file_id)
778
if (carry_over_possible and
779
parent_entry.symlink_target == entry.symlink_target):
782
self._add_text_to_weave(change[0], '', heads, None)
783
elif kind == 'directory':
784
if carry_over_possible:
787
# Nothing to set on the entry.
788
# XXX: split into the Root and nonRoot versions.
789
if change[1][1] != '' or self.repository.supports_rich_root():
790
self._add_text_to_weave(change[0], '', heads, None)
791
elif kind == 'tree-reference':
792
if not self.repository._format.supports_tree_reference:
793
# This isn't quite sane as an error, but we shouldn't
794
# ever see this code path in practice: tree's don't
795
# permit references when the repo doesn't support tree
797
raise errors.UnsupportedOperation(tree.add_reference,
799
reference_revision = tree.get_reference_revision(change[0])
800
entry.reference_revision = reference_revision
801
if (carry_over_possible and
802
parent_entry.reference_revision == reference_revision):
805
self._add_text_to_weave(change[0], '', heads, None)
807
raise AssertionError('unknown kind %r' % kind)
809
entry.revision = modified_rev
811
entry.revision = parent_entry.revision
814
new_path = change[1][1]
815
inv_delta.append((change[1][0], new_path, change[0], entry))
818
self.new_inventory = None
820
# This should perhaps be guarded by a check that the basis we
821
# commit against is the basis for the commit and if not do a delta
823
self._any_changes = True
825
# housekeeping root entry changes do not affect no-change commits.
826
self._require_root_change(tree)
827
self.basis_delta_revision = basis_revision_id
829
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
830
parent_keys = tuple([(file_id, parent) for parent in parents])
831
return self.repository.texts._add_text(
832
(file_id, self._new_revision_id), parent_keys, new_text,
833
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
836
class RootCommitBuilder(CommitBuilder):
837
"""This commitbuilder actually records the root id"""
839
# the root entry gets versioned properly by this builder.
840
_versioned_root = True
842
def _check_root(self, ie, parent_invs, tree):
843
"""Helper for record_entry_contents.
845
:param ie: An entry being added.
846
:param parent_invs: The inventories of the parent revisions of the
848
:param tree: The tree that is being committed.
851
def _require_root_change(self, tree):
852
"""Enforce an appropriate root object change.
854
This is called once when record_iter_changes is called, if and only if
855
the root was not in the delta calculated by record_iter_changes.
857
:param tree: The tree which is being committed.
859
# versioned roots do not change unless the tree found a change.
862
65
######################################################################
866
class Repository(_RelockDebugMixin):
68
class Repository(object):
867
69
"""Repository holding history for one or more branches.
869
71
The repository holds and retrieves historical information including
870
72
revisions and file history. It's normally accessed only by the Branch,
871
73
which views a particular line of development through that history.
873
The Repository builds on top of some byte storage facilies (the revisions,
874
signatures, inventories, texts and chk_bytes attributes) and a Transport,
875
which respectively provide byte storage and a means to access the (possibly
75
The Repository builds on top of Stores and a Transport, which respectively
76
describe the disk data format and the way of accessing the (possibly
878
The byte storage facilities are addressed via tuples, which we refer to
879
as 'keys' throughout the code base. Revision_keys, inventory_keys and
880
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
881
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
882
byte string made up of a hash identifier and a hash value.
883
We use this interface because it allows low friction with the underlying
884
code that implements disk indices, network encoding and other parts of
887
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
888
the serialised revisions for the repository. This can be used to obtain
889
revision graph information or to access raw serialised revisions.
890
The result of trying to insert data into the repository via this store
891
is undefined: it should be considered read-only except for implementors
893
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
894
the serialised signatures for the repository. This can be used to
895
obtain access to raw serialised signatures. The result of trying to
896
insert data into the repository via this store is undefined: it should
897
be considered read-only except for implementors of repositories.
898
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
899
the serialised inventories for the repository. This can be used to
900
obtain unserialised inventories. The result of trying to insert data
901
into the repository via this store is undefined: it should be
902
considered read-only except for implementors of repositories.
903
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
904
texts of files and directories for the repository. This can be used to
905
obtain file texts or file graphs. Note that Repository.iter_file_bytes
906
is usually a better interface for accessing file texts.
907
The result of trying to insert data into the repository via this store
908
is undefined: it should be considered read-only except for implementors
910
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
911
any data the repository chooses to store or have indexed by its hash.
912
The result of trying to insert data into the repository via this store
913
is undefined: it should be considered read-only except for implementors
915
:ivar _transport: Transport for file access to repository, typically
916
pointing to .bzr/repository.
919
# What class to use for a CommitBuilder. Often its simpler to change this
920
# in a Repository class subclass rather than to override
921
# get_commit_builder.
922
_commit_builder_class = CommitBuilder
923
# The search regex used by xml based repositories to determine what things
924
# where changed in a single commit.
925
80
_file_ids_altered_regex = lazy_regex.lazy_compile(
926
81
r'file_id="(?P<file_id>[^"]+)"'
927
r'.* revision="(?P<revision_id>[^"]+)"'
82
r'.*revision="(?P<revision_id>[^"]+)"'
930
def abort_write_group(self, suppress_errors=False):
931
"""Commit the contents accrued within the current write group.
933
:param suppress_errors: if true, abort_write_group will catch and log
934
unexpected errors that happen during the abort, rather than
935
allowing them to propagate. Defaults to False.
937
:seealso: start_write_group.
939
if self._write_group is not self.get_transaction():
940
# has an unlock or relock occured ?
943
'(suppressed) mismatched lock context and write group. %r, %r',
944
self._write_group, self.get_transaction())
946
raise errors.BzrError(
947
'mismatched lock context and write group. %r, %r' %
948
(self._write_group, self.get_transaction()))
950
self._abort_write_group()
951
except Exception, exc:
952
self._write_group = None
953
if not suppress_errors:
955
mutter('abort_write_group failed')
956
log_exception_quietly()
957
note('bzr: ERROR (ignored): %s', exc)
958
self._write_group = None
960
def _abort_write_group(self):
961
"""Template method for per-repository write group cleanup.
963
This is called during abort before the write group is considered to be
964
finished and should cleanup any internal state accrued during the write
965
group. There is no requirement that data handed to the repository be
966
*not* made available - this is not a rollback - but neither should any
967
attempt be made to ensure that data added is fully commited. Abort is
968
invoked when an error has occured so futher disk or network operations
969
may not be possible or may error and if possible should not be
973
def add_fallback_repository(self, repository):
974
"""Add a repository to use for looking up data not held locally.
976
:param repository: A repository.
978
if not self._format.supports_external_lookups:
979
raise errors.UnstackableRepositoryFormat(self._format, self.base)
981
# This repository will call fallback.unlock() when we transition to
982
# the unlocked state, so we make sure to increment the lock count
983
repository.lock_read()
984
self._check_fallback_repository(repository)
985
self._fallback_repositories.append(repository)
986
self.texts.add_fallback_versioned_files(repository.texts)
987
self.inventories.add_fallback_versioned_files(repository.inventories)
988
self.revisions.add_fallback_versioned_files(repository.revisions)
989
self.signatures.add_fallback_versioned_files(repository.signatures)
990
if self.chk_bytes is not None:
991
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
993
def _check_fallback_repository(self, repository):
994
"""Check that this repository can fallback to repository safely.
996
Raise an error if not.
998
:param repository: A repository to fallback to.
1000
return InterRepository._assert_same_model(self, repository)
1002
86
def add_inventory(self, revision_id, inv, parents):
1003
87
"""Add the inventory inv to the repository as revision_id.
1005
89
:param parents: The revision ids of the parents that revision_id
1006
90
is known to have and are in the repository already.
1008
:returns: The validator(which is a sha1 digest, though what is sha'd is
1009
repository format specific) of the serialized inventory.
92
returns the sha1 of the serialized inventory.
1011
if not self.is_in_write_group():
1012
raise AssertionError("%r not in write group" % (self,))
94
revision_id = osutils.safe_revision_id(revision_id)
1013
95
_mod_revision.check_not_reserved_id(revision_id)
1014
if not (inv.revision_id is None or inv.revision_id == revision_id):
1015
raise AssertionError(
1016
"Mismatch between inventory revision"
1017
" id and insertion revid (%r, %r)"
1018
% (inv.revision_id, revision_id))
1019
if inv.root is None:
1020
raise AssertionError()
1021
return self._add_inventory_checked(revision_id, inv, parents)
1023
def _add_inventory_checked(self, revision_id, inv, parents):
1024
"""Add inv to the repository after checking the inputs.
1026
This function can be overridden to allow different inventory styles.
1028
:seealso: add_inventory, for the contract.
1030
inv_lines = self._serialise_inventory_to_lines(inv)
1031
return self._inventory_add_lines(revision_id, parents,
1032
inv_lines, check_content=False)
1034
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1035
parents, basis_inv=None, propagate_caches=False):
1036
"""Add a new inventory expressed as a delta against another revision.
1038
See the inventory developers documentation for the theory behind
1041
:param basis_revision_id: The inventory id the delta was created
1042
against. (This does not have to be a direct parent.)
1043
:param delta: The inventory delta (see Inventory.apply_delta for
1045
:param new_revision_id: The revision id that the inventory is being
1047
:param parents: The revision ids of the parents that revision_id is
1048
known to have and are in the repository already. These are supplied
1049
for repositories that depend on the inventory graph for revision
1050
graph access, as well as for those that pun ancestry with delta
1052
:param basis_inv: The basis inventory if it is already known,
1054
:param propagate_caches: If True, the caches for this inventory are
1055
copied to and updated for the result if possible.
1057
:returns: (validator, new_inv)
1058
The validator(which is a sha1 digest, though what is sha'd is
1059
repository format specific) of the serialized inventory, and the
1060
resulting inventory.
1062
if not self.is_in_write_group():
1063
raise AssertionError("%r not in write group" % (self,))
1064
_mod_revision.check_not_reserved_id(new_revision_id)
1065
basis_tree = self.revision_tree(basis_revision_id)
1066
basis_tree.lock_read()
1068
# Note that this mutates the inventory of basis_tree, which not all
1069
# inventory implementations may support: A better idiom would be to
1070
# return a new inventory, but as there is no revision tree cache in
1071
# repository this is safe for now - RBC 20081013
1072
if basis_inv is None:
1073
basis_inv = basis_tree.inventory
1074
basis_inv.apply_delta(delta)
1075
basis_inv.revision_id = new_revision_id
1076
return (self.add_inventory(new_revision_id, basis_inv, parents),
1081
def _inventory_add_lines(self, revision_id, parents, lines,
1082
check_content=True):
1083
"""Store lines in inv_vf and return the sha1 of the inventory."""
1084
parents = [(parent,) for parent in parents]
1085
result = self.inventories.add_lines((revision_id,), parents, lines,
1086
check_content=check_content)[0]
1087
self.inventories._access.flush()
96
assert inv.revision_id is None or inv.revision_id == revision_id, \
97
"Mismatch between inventory revision" \
98
" id and insertion revid (%r, %r)" % (inv.revision_id, revision_id)
99
assert inv.root is not None
100
inv_text = self.serialise_inventory(inv)
101
inv_sha1 = osutils.sha_string(inv_text)
102
inv_vf = self.control_weaves.get_weave('inventory',
103
self.get_transaction())
104
self._inventory_add_lines(inv_vf, revision_id, parents,
105
osutils.split_lines(inv_text))
108
def _inventory_add_lines(self, inv_vf, revision_id, parents, lines):
110
for parent in parents:
112
final_parents.append(parent)
114
inv_vf.add_lines(revision_id, final_parents, lines)
1090
117
def add_revision(self, revision_id, rev, inv=None, config=None):
1091
118
"""Add rev to the revision store as revision_id.
1811
434
dest_repo = a_bzrdir.open_repository()
1812
435
return dest_repo
1814
def _get_sink(self):
1815
"""Return a sink for streaming into this repository."""
1816
return StreamSink(self)
1818
def _get_source(self, to_format):
1819
"""Return a source for streaming from this repository."""
1820
return StreamSource(self, to_format)
1822
437
@needs_read_lock
1823
438
def has_revision(self, revision_id):
1824
439
"""True if this repository has a copy of the revision."""
1825
return revision_id in self.has_revisions((revision_id,))
1828
def has_revisions(self, revision_ids):
1829
"""Probe to find out the presence of multiple revisions.
1831
:param revision_ids: An iterable of revision_ids.
1832
:return: A set of the revision_ids that were present.
1834
parent_map = self.revisions.get_parent_map(
1835
[(rev_id,) for rev_id in revision_ids])
1837
if _mod_revision.NULL_REVISION in revision_ids:
1838
result.add(_mod_revision.NULL_REVISION)
1839
result.update([key[0] for key in parent_map])
1843
def get_revision(self, revision_id):
1844
"""Return the Revision object for a named revision."""
1845
return self.get_revisions([revision_id])[0]
440
revision_id = osutils.safe_revision_id(revision_id)
441
return self._revision_store.has_revision_id(revision_id,
442
self.get_transaction())
1847
444
@needs_read_lock
1848
445
def get_revision_reconcile(self, revision_id):
1849
446
"""'reconcile' helper routine that allows access to a revision always.
1851
448
This variant of get_revision does not cross check the weave graph
1852
449
against the revision one as get_revision does: but it should only
1853
450
be used by reconcile, or reconcile-alike commands that are correcting
1854
451
or testing the revision graph.
1856
return self._get_revisions([revision_id])[0]
453
if not revision_id or not isinstance(revision_id, basestring):
454
raise errors.InvalidRevisionId(revision_id=revision_id,
456
return self.get_revisions([revision_id])[0]
1858
458
@needs_read_lock
1859
459
def get_revisions(self, revision_ids):
1860
"""Get many revisions at once.
1862
Repositories that need to check data on every revision read should
1863
subclass this method.
1865
return self._get_revisions(revision_ids)
1868
def _get_revisions(self, revision_ids):
1869
"""Core work logic to get many revisions without sanity checks."""
1871
for revid, rev in self._iter_revisions(revision_ids):
1873
raise errors.NoSuchRevision(self, revid)
1875
return [revs[revid] for revid in revision_ids]
1877
def _iter_revisions(self, revision_ids):
1878
"""Iterate over revision objects.
1880
:param revision_ids: An iterable of revisions to examine. None may be
1881
passed to request all revisions known to the repository. Note that
1882
not all repositories can find unreferenced revisions; for those
1883
repositories only referenced ones will be returned.
1884
:return: An iterator of (revid, revision) tuples. Absent revisions (
1885
those asked for but not available) are returned as (revid, None).
1887
if revision_ids is None:
1888
revision_ids = self.all_revision_ids()
1890
for rev_id in revision_ids:
1891
if not rev_id or not isinstance(rev_id, basestring):
1892
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1893
keys = [(key,) for key in revision_ids]
1894
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1895
for record in stream:
1896
revid = record.key[0]
1897
if record.storage_kind == 'absent':
1900
text = record.get_bytes_as('fulltext')
1901
rev = self._serializer.read_revision_from_string(text)
460
revision_ids = [osutils.safe_revision_id(r) for r in revision_ids]
461
revs = self._revision_store.get_revisions(revision_ids,
462
self.get_transaction())
464
assert not isinstance(rev.revision_id, unicode)
465
for parent_id in rev.parent_ids:
466
assert not isinstance(parent_id, unicode)
1904
469
@needs_read_lock
1905
470
def get_revision_xml(self, revision_id):
1906
471
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1907
472
# would have already do it.
1908
473
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1909
# TODO: this can't just be replaced by:
1910
# return self._serializer.write_revision_to_string(
1911
# self.get_revision(revision_id))
1912
# as cStringIO preservers the encoding unlike write_revision_to_string
1913
# or some other call down the path.
474
revision_id = osutils.safe_revision_id(revision_id)
1914
475
rev = self.get_revision(revision_id)
1915
rev_tmp = cStringIO.StringIO()
1916
477
# the current serializer..
1917
self._serializer.write_revision(rev, rev_tmp)
478
self._revision_store._serializer.write_revision(rev, rev_tmp)
1919
480
return rev_tmp.getvalue()
1921
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
483
def get_revision(self, revision_id):
484
"""Return the Revision object for a named revision"""
485
# TODO: jam 20070210 get_revision_reconcile should do this for us
486
revision_id = osutils.safe_revision_id(revision_id)
487
r = self.get_revision_reconcile(revision_id)
488
# weave corruption can lead to absent revision markers that should be
490
# the following test is reasonably cheap (it needs a single weave read)
491
# and the weave is cached in read transactions. In write transactions
492
# it is not cached but typically we only read a small number of
493
# revisions. For knits when they are introduced we will probably want
494
# to ensure that caching write transactions are in use.
495
inv = self.get_inventory_weave()
496
self._check_revision_parents(r, inv)
500
def get_deltas_for_revisions(self, revisions):
1922
501
"""Produce a generator of revision deltas.
1924
503
Note that the input is a sequence of REVISIONS, not revision_ids.
1925
504
Trees will be held in memory until the generator exits.
1926
505
Each delta is relative to the revision's lefthand predecessor.
1928
:param specific_fileids: if not None, the result is filtered
1929
so that only those file-ids, their parents and their
1930
children are included.
1932
# Get the revision-ids of interest
1933
507
required_trees = set()
1934
508
for revision in revisions:
1935
509
required_trees.add(revision.revision_id)
1936
510
required_trees.update(revision.parent_ids[:1])
1938
# Get the matching filtered trees. Note that it's more
1939
# efficient to pass filtered trees to changes_from() rather
1940
# than doing the filtering afterwards. changes_from() could
1941
# arguably do the filtering itself but it's path-based, not
1942
# file-id based, so filtering before or afterwards is
1944
if specific_fileids is None:
1945
trees = dict((t.get_revision_id(), t) for
1946
t in self.revision_trees(required_trees))
1948
trees = dict((t.get_revision_id(), t) for
1949
t in self._filtered_revision_trees(required_trees,
1952
# Calculate the deltas
511
trees = dict((t.get_revision_id(), t) for
512
t in self.revision_trees(required_trees))
1953
513
for revision in revisions:
1954
514
if not revision.parent_ids:
1955
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
515
old_tree = self.revision_tree(None)
1957
517
old_tree = trees[revision.parent_ids[0]]
1958
518
yield trees[revision.revision_id].changes_from(old_tree)
1960
520
@needs_read_lock
1961
def get_revision_delta(self, revision_id, specific_fileids=None):
521
def get_revision_delta(self, revision_id):
1962
522
"""Return the delta for one revision.
1964
524
The delta is relative to the left-hand predecessor of the
1967
:param specific_fileids: if not None, the result is filtered
1968
so that only those file-ids, their parents and their
1969
children are included.
1971
527
r = self.get_revision(revision_id)
1972
return list(self.get_deltas_for_revisions([r],
1973
specific_fileids=specific_fileids))[0]
528
return list(self.get_deltas_for_revisions([r]))[0]
530
def _check_revision_parents(self, revision, inventory):
531
"""Private to Repository and Fetch.
533
This checks the parentage of revision in an inventory weave for
534
consistency and is only applicable to inventory-weave-for-ancestry
535
using repository formats & fetchers.
537
weave_parents = inventory.get_parents(revision.revision_id)
538
weave_names = inventory.versions()
539
for parent_id in revision.parent_ids:
540
if parent_id in weave_names:
541
# this parent must not be a ghost.
542
if not parent_id in weave_parents:
544
raise errors.CorruptRepository(self)
1975
546
@needs_write_lock
1976
547
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
548
revision_id = osutils.safe_revision_id(revision_id)
1977
549
signature = gpg_strategy.sign(plaintext)
1978
self.add_signature_text(revision_id, signature)
1981
def add_signature_text(self, revision_id, signature):
1982
self.signatures.add_lines((revision_id,), (),
1983
osutils.split_lines(signature))
1985
def find_text_key_references(self):
1986
"""Find the text key references within the repository.
1988
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1989
to whether they were referred to by the inventory of the
1990
revision_id that they contain. The inventory texts from all present
1991
revision ids are assessed to generate this report.
1993
revision_keys = self.revisions.keys()
1994
w = self.inventories
1995
pb = ui.ui_factory.nested_progress_bar()
1997
return self._find_text_key_references_from_xml_inventory_lines(
1998
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2002
def _find_text_key_references_from_xml_inventory_lines(self,
2004
"""Core routine for extracting references to texts from inventories.
2006
This performs the translation of xml lines to revision ids.
2008
:param line_iterator: An iterator of lines, origin_version_id
2009
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2010
to whether they were referred to by the inventory of the
2011
revision_id that they contain. Note that if that revision_id was
2012
not part of the line_iterator's output then False will be given -
2013
even though it may actually refer to that key.
2015
if not self._serializer.support_altered_by_hack:
2016
raise AssertionError(
2017
"_find_text_key_references_from_xml_inventory_lines only "
2018
"supported for branches which store inventory as unnested xml"
2019
", not on %r" % self)
550
self._revision_store.add_revision_signature_text(revision_id,
552
self.get_transaction())
554
def fileids_altered_by_revision_ids(self, revision_ids):
555
"""Find the file ids and versions affected by revisions.
557
:param revisions: an iterable containing revision ids.
558
:return: a dictionary mapping altered file-ids to an iterable of
559
revision_ids. Each altered file-ids has the exact revision_ids that
560
altered it listed explicitly.
562
assert self._serializer.support_altered_by_hack, \
563
("fileids_altered_by_revision_ids only supported for branches "
564
"which store inventory as unnested xml, not on %r" % self)
565
selected_revision_ids = set(osutils.safe_revision_id(r)
566
for r in revision_ids)
567
w = self.get_inventory_weave()
2022
570
# this code needs to read every new line in every inventory for the
2023
571
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2024
# not present in one of those inventories is unnecessary but not
572
# not present in one of those inventories is unnecessary but not
2025
573
# harmful because we are filtering by the revision id marker in the
2026
# inventory lines : we only select file ids altered in one of those
574
# inventory lines : we only select file ids altered in one of those
2027
575
# revisions. We don't need to see all lines in the inventory because
2028
576
# only those added in an inventory in rev X can contain a revision=X
2038
586
search = self._file_ids_altered_regex.search
2039
587
unescape = _unescape_xml
2040
588
setdefault = result.setdefault
2041
for line, line_key in line_iterator:
2042
match = search(line)
2045
# One call to match.group() returning multiple items is quite a
2046
# bit faster than 2 calls to match.group() each returning 1
2047
file_id, revision_id = match.group('file_id', 'revision_id')
2049
# Inlining the cache lookups helps a lot when you make 170,000
2050
# lines and 350k ids, versus 8.4 unique ids.
2051
# Using a cache helps in 2 ways:
2052
# 1) Avoids unnecessary decoding calls
2053
# 2) Re-uses cached strings, which helps in future set and
2055
# (2) is enough that removing encoding entirely along with
2056
# the cache (so we are using plain strings) results in no
2057
# performance improvement.
2059
revision_id = unescape_revid_cache[revision_id]
2061
unescaped = unescape(revision_id)
2062
unescape_revid_cache[revision_id] = unescaped
2063
revision_id = unescaped
2065
# Note that unconditionally unescaping means that we deserialise
2066
# every fileid, which for general 'pull' is not great, but we don't
2067
# really want to have some many fulltexts that this matters anyway.
2070
file_id = unescape_fileid_cache[file_id]
2072
unescaped = unescape(file_id)
2073
unescape_fileid_cache[file_id] = unescaped
2076
key = (file_id, revision_id)
2077
setdefault(key, False)
2078
if revision_id == line_key[-1]:
2082
def _inventory_xml_lines_for_keys(self, keys):
2083
"""Get a line iterator of the sort needed for findind references.
2085
Not relevant for non-xml inventory repositories.
2087
Ghosts in revision_keys are ignored.
2089
:param revision_keys: The revision keys for the inventories to inspect.
2090
:return: An iterator over (inventory line, revid) for the fulltexts of
2091
all of the xml inventories specified by revision_keys.
2093
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2094
for record in stream:
2095
if record.storage_kind != 'absent':
2096
chunks = record.get_bytes_as('chunked')
2097
revid = record.key[-1]
2098
lines = osutils.chunks_to_lines(chunks)
2102
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2104
"""Helper routine for fileids_altered_by_revision_ids.
2106
This performs the translation of xml lines to revision ids.
2108
:param line_iterator: An iterator of lines, origin_version_id
2109
:param revision_keys: The revision ids to filter for. This should be a
2110
set or other type which supports efficient __contains__ lookups, as
2111
the revision key from each parsed line will be looked up in the
2112
revision_keys filter.
2113
:return: a dictionary mapping altered file-ids to an iterable of
2114
revision_ids. Each altered file-ids has the exact revision_ids that
2115
altered it listed explicitly.
2117
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2118
line_iterator).iterkeys())
2119
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2120
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2121
self._inventory_xml_lines_for_keys(parent_keys)))
2122
new_keys = seen - parent_seen
2124
setdefault = result.setdefault
2125
for key in new_keys:
2126
setdefault(key[0], set()).add(key[-1])
2129
def _find_parent_ids_of_revisions(self, revision_ids):
2130
"""Find all parent ids that are mentioned in the revision graph.
2132
:return: set of revisions that are parents of revision_ids which are
2133
not part of revision_ids themselves
2135
parent_map = self.get_parent_map(revision_ids)
2137
map(parent_ids.update, parent_map.itervalues())
2138
parent_ids.difference_update(revision_ids)
2139
parent_ids.discard(_mod_revision.NULL_REVISION)
2142
def _find_parent_keys_of_revisions(self, revision_keys):
2143
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2145
:param revision_keys: An iterable of revision_keys.
2146
:return: The parents of all revision_keys that are not already in
2149
parent_map = self.revisions.get_parent_map(revision_keys)
2151
map(parent_keys.update, parent_map.itervalues())
2152
parent_keys.difference_update(revision_keys)
2153
parent_keys.discard(_mod_revision.NULL_REVISION)
2156
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2157
"""Find the file ids and versions affected by revisions.
2159
:param revisions: an iterable containing revision ids.
2160
:param _inv_weave: The inventory weave from this repository or None.
2161
If None, the inventory weave will be opened automatically.
2162
:return: a dictionary mapping altered file-ids to an iterable of
2163
revision_ids. Each altered file-ids has the exact revision_ids that
2164
altered it listed explicitly.
2166
selected_keys = set((revid,) for revid in revision_ids)
2167
w = _inv_weave or self.inventories
2168
return self._find_file_ids_from_xml_inventory_lines(
2169
w.iter_lines_added_or_present_in_keys(
2170
selected_keys, pb=None),
2173
def iter_files_bytes(self, desired_files):
2174
"""Iterate through file versions.
2176
Files will not necessarily be returned in the order they occur in
2177
desired_files. No specific order is guaranteed.
2179
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2180
value supplied by the caller as part of desired_files. It should
2181
uniquely identify the file version in the caller's context. (Examples:
2182
an index number or a TreeTransform trans_id.)
2184
bytes_iterator is an iterable of bytestrings for the file. The
2185
kind of iterable and length of the bytestrings are unspecified, but for
2186
this implementation, it is a list of bytes produced by
2187
VersionedFile.get_record_stream().
2189
:param desired_files: a list of (file_id, revision_id, identifier)
2193
for file_id, revision_id, callable_data in desired_files:
2194
text_keys[(file_id, revision_id)] = callable_data
2195
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2196
if record.storage_kind == 'absent':
2197
raise errors.RevisionNotPresent(record.key, self)
2198
yield text_keys[record.key], record.get_bytes_as('chunked')
2200
def _generate_text_key_index(self, text_key_references=None,
2202
"""Generate a new text key index for the repository.
2204
This is an expensive function that will take considerable time to run.
2206
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2207
list of parents, also text keys. When a given key has no parents,
2208
the parents list will be [NULL_REVISION].
2210
# All revisions, to find inventory parents.
2211
if ancestors is None:
2212
graph = self.get_graph()
2213
ancestors = graph.get_parent_map(self.all_revision_ids())
2214
if text_key_references is None:
2215
text_key_references = self.find_text_key_references()
2216
589
pb = ui.ui_factory.nested_progress_bar()
2218
return self._do_generate_text_key_index(ancestors,
2219
text_key_references, pb)
591
for line in w.iter_lines_added_or_present_in_versions(
592
selected_revision_ids, pb=pb):
596
# One call to match.group() returning multiple items is quite a
597
# bit faster than 2 calls to match.group() each returning 1
598
file_id, revision_id = match.group('file_id', 'revision_id')
600
# Inlining the cache lookups helps a lot when you make 170,000
601
# lines and 350k ids, versus 8.4 unique ids.
602
# Using a cache helps in 2 ways:
603
# 1) Avoids unnecessary decoding calls
604
# 2) Re-uses cached strings, which helps in future set and
606
# (2) is enough that removing encoding entirely along with
607
# the cache (so we are using plain strings) results in no
608
# performance improvement.
610
revision_id = unescape_revid_cache[revision_id]
612
unescaped = unescape(revision_id)
613
unescape_revid_cache[revision_id] = unescaped
614
revision_id = unescaped
616
if revision_id in selected_revision_ids:
618
file_id = unescape_fileid_cache[file_id]
620
unescaped = unescape(file_id)
621
unescape_fileid_cache[file_id] = unescaped
623
setdefault(file_id, set()).add(revision_id)
2223
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2224
"""Helper for _generate_text_key_index to avoid deep nesting."""
2225
revision_order = tsort.topo_sort(ancestors)
2226
invalid_keys = set()
2228
for revision_id in revision_order:
2229
revision_keys[revision_id] = set()
2230
text_count = len(text_key_references)
2231
# a cache of the text keys to allow reuse; costs a dict of all the
2232
# keys, but saves a 2-tuple for every child of a given key.
2234
for text_key, valid in text_key_references.iteritems():
2236
invalid_keys.add(text_key)
2238
revision_keys[text_key[1]].add(text_key)
2239
text_key_cache[text_key] = text_key
2240
del text_key_references
2242
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2243
NULL_REVISION = _mod_revision.NULL_REVISION
2244
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2245
# too small for large or very branchy trees. However, for 55K path
2246
# trees, it would be easy to use too much memory trivially. Ideally we
2247
# could gauge this by looking at available real memory etc, but this is
2248
# always a tricky proposition.
2249
inventory_cache = lru_cache.LRUCache(10)
2250
batch_size = 10 # should be ~150MB on a 55K path tree
2251
batch_count = len(revision_order) / batch_size + 1
2253
pb.update("Calculating text parents", processed_texts, text_count)
2254
for offset in xrange(batch_count):
2255
to_query = revision_order[offset * batch_size:(offset + 1) *
2259
for revision_id in to_query:
2260
parent_ids = ancestors[revision_id]
2261
for text_key in revision_keys[revision_id]:
2262
pb.update("Calculating text parents", processed_texts)
2263
processed_texts += 1
2264
candidate_parents = []
2265
for parent_id in parent_ids:
2266
parent_text_key = (text_key[0], parent_id)
2268
check_parent = parent_text_key not in \
2269
revision_keys[parent_id]
2271
# the parent parent_id is a ghost:
2272
check_parent = False
2273
# truncate the derived graph against this ghost.
2274
parent_text_key = None
2276
# look at the parent commit details inventories to
2277
# determine possible candidates in the per file graph.
2280
inv = inventory_cache[parent_id]
2282
inv = self.revision_tree(parent_id).inventory
2283
inventory_cache[parent_id] = inv
2285
parent_entry = inv[text_key[0]]
2286
except (KeyError, errors.NoSuchId):
2288
if parent_entry is not None:
2290
text_key[0], parent_entry.revision)
2292
parent_text_key = None
2293
if parent_text_key is not None:
2294
candidate_parents.append(
2295
text_key_cache[parent_text_key])
2296
parent_heads = text_graph.heads(candidate_parents)
2297
new_parents = list(parent_heads)
2298
new_parents.sort(key=lambda x:candidate_parents.index(x))
2299
if new_parents == []:
2300
new_parents = [NULL_REVISION]
2301
text_index[text_key] = new_parents
2303
for text_key in invalid_keys:
2304
text_index[text_key] = [NULL_REVISION]
2307
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2308
"""Get an iterable listing the keys of all the data introduced by a set
2311
The keys will be ordered so that the corresponding items can be safely
2312
fetched and inserted in that order.
2314
:returns: An iterable producing tuples of (knit-kind, file-id,
2315
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2316
'revisions'. file-id is None unless knit-kind is 'file'.
2318
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2321
for result in self._find_non_file_keys_to_fetch(revision_ids):
2324
def _find_file_keys_to_fetch(self, revision_ids, pb):
2325
# XXX: it's a bit weird to control the inventory weave caching in this
2326
# generator. Ideally the caching would be done in fetch.py I think. Or
2327
# maybe this generator should explicitly have the contract that it
2328
# should not be iterated until the previously yielded item has been
2330
inv_w = self.inventories
2332
# file ids that changed
2333
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2335
num_file_ids = len(file_ids)
2336
for file_id, altered_versions in file_ids.iteritems():
2338
pb.update("Fetch texts", count, num_file_ids)
2340
yield ("file", file_id, altered_versions)
2342
def _find_non_file_keys_to_fetch(self, revision_ids):
2344
yield ("inventory", None, revision_ids)
2347
# XXX: Note ATM no callers actually pay attention to this return
2348
# instead they just use the list of revision ids and ignore
2349
# missing sigs. Consider removing this work entirely
2350
revisions_with_signatures = set(self.signatures.get_parent_map(
2351
[(r,) for r in revision_ids]))
2352
revisions_with_signatures = set(
2353
[r for (r,) in revisions_with_signatures])
2354
revisions_with_signatures.intersection_update(revision_ids)
2355
yield ("signatures", None, revisions_with_signatures)
2358
yield ("revisions", None, revision_ids)
629
def get_inventory_weave(self):
630
return self.control_weaves.get_weave('inventory',
631
self.get_transaction())
2360
633
@needs_read_lock
2361
634
def get_inventory(self, revision_id):
2362
"""Get Inventory object by revision id."""
2363
return self.iter_inventories([revision_id]).next()
2365
def iter_inventories(self, revision_ids, ordering=None):
2366
"""Get many inventories by revision_ids.
2368
This will buffer some or all of the texts used in constructing the
2369
inventories in memory, but will only parse a single inventory at a
2372
:param revision_ids: The expected revision ids of the inventories.
2373
:param ordering: optional ordering, e.g. 'topological'. If not
2374
specified, the order of revision_ids will be preserved (by
2375
buffering if necessary).
2376
:return: An iterator of inventories.
2378
if ((None in revision_ids)
2379
or (_mod_revision.NULL_REVISION in revision_ids)):
2380
raise ValueError('cannot get null revision inventory')
2381
return self._iter_inventories(revision_ids, ordering)
2383
def _iter_inventories(self, revision_ids, ordering):
2384
"""single-document based inventory iteration."""
2385
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2386
for text, revision_id in inv_xmls:
2387
yield self._deserialise_inventory(revision_id, text)
2389
def _iter_inventory_xmls(self, revision_ids, ordering):
2390
if ordering is None:
2391
order_as_requested = True
2392
ordering = 'unordered'
2394
order_as_requested = False
2395
keys = [(revision_id,) for revision_id in revision_ids]
2398
if order_as_requested:
2399
key_iter = iter(keys)
2400
next_key = key_iter.next()
2401
stream = self.inventories.get_record_stream(keys, ordering, True)
2403
for record in stream:
2404
if record.storage_kind != 'absent':
2405
chunks = record.get_bytes_as('chunked')
2406
if order_as_requested:
2407
text_chunks[record.key] = chunks
2409
yield ''.join(chunks), record.key[-1]
2411
raise errors.NoSuchRevision(self, record.key)
2412
if order_as_requested:
2413
# Yield as many results as we can while preserving order.
2414
while next_key in text_chunks:
2415
chunks = text_chunks.pop(next_key)
2416
yield ''.join(chunks), next_key[-1]
2418
next_key = key_iter.next()
2419
except StopIteration:
2420
# We still want to fully consume the get_record_stream,
2421
# just in case it is not actually finished at this point
2425
def _deserialise_inventory(self, revision_id, xml):
2426
"""Transform the xml into an inventory object.
635
"""Get Inventory object by hash."""
636
# TODO: jam 20070210 Technically we don't need to sanitize, since all
637
# called functions must sanitize.
638
revision_id = osutils.safe_revision_id(revision_id)
639
return self.deserialise_inventory(
640
revision_id, self.get_inventory_xml(revision_id))
642
def deserialise_inventory(self, revision_id, xml):
643
"""Transform the xml into an inventory object.
2428
645
:param revision_id: The expected revision id of the inventory.
2429
646
:param xml: A serialised inventory.
2431
result = self._serializer.read_inventory_from_string(xml, revision_id,
2432
entry_cache=self._inventory_entry_cache,
2433
return_from_cache=self._safe_to_return_from_cache)
2434
if result.revision_id != revision_id:
2435
raise AssertionError('revision id mismatch %s != %s' % (
2436
result.revision_id, revision_id))
648
revision_id = osutils.safe_revision_id(revision_id)
649
result = self._serializer.read_inventory_from_string(xml)
650
result.root.revision = revision_id
2439
def _serialise_inventory(self, inv):
653
def serialise_inventory(self, inv):
2440
654
return self._serializer.write_inventory_to_string(inv)
2442
def _serialise_inventory_to_lines(self, inv):
2443
return self._serializer.write_inventory_to_lines(inv)
2445
656
def get_serializer_format(self):
2446
657
return self._serializer.format_num
2448
659
@needs_read_lock
2449
def _get_inventory_xml(self, revision_id):
2450
"""Get serialized inventory as a string."""
2451
texts = self._iter_inventory_xmls([revision_id], 'unordered')
660
def get_inventory_xml(self, revision_id):
661
"""Get inventory XML as a file object."""
662
revision_id = osutils.safe_revision_id(revision_id)
2453
text, revision_id = texts.next()
2454
except StopIteration:
664
assert isinstance(revision_id, str), type(revision_id)
665
iw = self.get_inventory_weave()
666
return iw.get_text(revision_id)
2455
668
raise errors.HistoryMissing(self, 'inventory', revision_id)
2458
def get_rev_id_for_revno(self, revno, known_pair):
2459
"""Return the revision id of a revno, given a later (revno, revid)
2460
pair in the same history.
2462
:return: if found (True, revid). If the available history ran out
2463
before reaching the revno, then this returns
2464
(False, (closest_revno, closest_revid)).
2466
known_revno, known_revid = known_pair
2467
partial_history = [known_revid]
2468
distance_from_known = known_revno - revno
2469
if distance_from_known < 0:
2471
'requested revno (%d) is later than given known revno (%d)'
2472
% (revno, known_revno))
2475
self, partial_history, stop_index=distance_from_known)
2476
except errors.RevisionNotPresent, err:
2477
if err.revision_id == known_revid:
2478
# The start revision (known_revid) wasn't found.
2480
# This is a stacked repository with no fallbacks, or a there's a
2481
# left-hand ghost. Either way, even though the revision named in
2482
# the error isn't in this repo, we know it's the next step in this
2483
# left-hand history.
2484
partial_history.append(err.revision_id)
2485
if len(partial_history) <= distance_from_known:
2486
# Didn't find enough history to get a revid for the revno.
2487
earliest_revno = known_revno - len(partial_history) + 1
2488
return (False, (earliest_revno, partial_history[-1]))
2489
if len(partial_history) - 1 > distance_from_known:
2490
raise AssertionError('_iter_for_revno returned too much history')
2491
return (True, partial_history[-1])
671
def get_inventory_sha1(self, revision_id):
672
"""Return the sha1 hash of the inventory entry
674
# TODO: jam 20070210 Shouldn't this be deprecated / removed?
675
revision_id = osutils.safe_revision_id(revision_id)
676
return self.get_revision(revision_id).inventory_sha1
679
def get_revision_graph(self, revision_id=None):
680
"""Return a dictionary containing the revision graph.
682
:param revision_id: The revision_id to get a graph from. If None, then
683
the entire revision graph is returned. This is a deprecated mode of
684
operation and will be removed in the future.
685
:return: a dictionary of revision_id->revision_parents_list.
687
# special case NULL_REVISION
688
if revision_id == _mod_revision.NULL_REVISION:
690
revision_id = osutils.safe_revision_id(revision_id)
691
a_weave = self.get_inventory_weave()
692
all_revisions = self._eliminate_revisions_not_present(
694
entire_graph = dict([(node, a_weave.get_parents(node)) for
695
node in all_revisions])
696
if revision_id is None:
698
elif revision_id not in entire_graph:
699
raise errors.NoSuchRevision(self, revision_id)
701
# add what can be reached from revision_id
703
pending = set([revision_id])
704
while len(pending) > 0:
706
result[node] = entire_graph[node]
707
for revision_id in result[node]:
708
if revision_id not in result:
709
pending.add(revision_id)
713
def get_revision_graph_with_ghosts(self, revision_ids=None):
714
"""Return a graph of the revisions with ghosts marked as applicable.
716
:param revision_ids: an iterable of revisions to graph or None for all.
717
:return: a Graph object with the graph reachable from revision_ids.
719
result = deprecated_graph.Graph()
721
pending = set(self.all_revision_ids())
724
pending = set(osutils.safe_revision_id(r) for r in revision_ids)
725
# special case NULL_REVISION
726
if _mod_revision.NULL_REVISION in pending:
727
pending.remove(_mod_revision.NULL_REVISION)
728
required = set(pending)
731
revision_id = pending.pop()
733
rev = self.get_revision(revision_id)
734
except errors.NoSuchRevision:
735
if revision_id in required:
738
result.add_ghost(revision_id)
740
for parent_id in rev.parent_ids:
741
# is this queued or done ?
742
if (parent_id not in pending and
743
parent_id not in done):
745
pending.add(parent_id)
746
result.add_node(revision_id, rev.parent_ids)
747
done.add(revision_id)
750
def _get_history_vf(self):
751
"""Get a versionedfile whose history graph reflects all revisions.
753
For weave repositories, this is the inventory weave.
755
return self.get_inventory_weave()
2493
757
def iter_reverse_revision_history(self, revision_id):
2494
758
"""Iterate backwards through revision ids in the lefthand history
3382
1385
InterRepository.get(other).method_name(parameters).
3385
_walk_to_common_revisions_batch_size = 50
3386
1388
_optimisers = []
3387
1389
"""The available optimised InterRepository types."""
3390
1391
def copy_content(self, revision_id=None):
3391
"""Make a complete copy of the content in self into destination.
3393
This is a destructive operation! Do not use it on existing
3396
:param revision_id: Only copy the content needed to construct
3397
revision_id and its parents.
3400
self.target.set_make_working_trees(self.source.make_working_trees())
3401
except NotImplementedError:
3403
self.target.fetch(self.source, revision_id=revision_id)
3406
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
1392
raise NotImplementedError(self.copy_content)
1394
def fetch(self, revision_id=None, pb=None):
3408
1395
"""Fetch the content required to construct revision_id.
3410
1397
The content is copied from self.source to self.target.
3412
1399
:param revision_id: if None all content is copied, if NULL_REVISION no
3413
1400
content is copied.
3417
f = _mod_fetch.RepoFetcher(to_repository=self.target,
3418
from_repository=self.source,
3419
last_revision=revision_id,
3420
fetch_spec=fetch_spec,
3421
find_ghosts=find_ghosts)
3423
def _walk_to_common_revisions(self, revision_ids):
3424
"""Walk out from revision_ids in source to revisions target has.
3426
:param revision_ids: The start point for the search.
3427
:return: A set of revision ids.
3429
target_graph = self.target.get_graph()
3430
revision_ids = frozenset(revision_ids)
3431
missing_revs = set()
3432
source_graph = self.source.get_graph()
3433
# ensure we don't pay silly lookup costs.
3434
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3435
null_set = frozenset([_mod_revision.NULL_REVISION])
3436
searcher_exhausted = False
3440
# Iterate the searcher until we have enough next_revs
3441
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3443
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3444
next_revs.update(next_revs_part)
3445
ghosts.update(ghosts_part)
3446
except StopIteration:
3447
searcher_exhausted = True
3449
# If there are ghosts in the source graph, and the caller asked for
3450
# them, make sure that they are present in the target.
3451
# We don't care about other ghosts as we can't fetch them and
3452
# haven't been asked to.
3453
ghosts_to_check = set(revision_ids.intersection(ghosts))
3454
revs_to_get = set(next_revs).union(ghosts_to_check)
3456
have_revs = set(target_graph.get_parent_map(revs_to_get))
3457
# we always have NULL_REVISION present.
3458
have_revs = have_revs.union(null_set)
3459
# Check if the target is missing any ghosts we need.
3460
ghosts_to_check.difference_update(have_revs)
3462
# One of the caller's revision_ids is a ghost in both the
3463
# source and the target.
3464
raise errors.NoSuchRevision(
3465
self.source, ghosts_to_check.pop())
3466
missing_revs.update(next_revs - have_revs)
3467
# Because we may have walked past the original stop point, make
3468
# sure everything is stopped
3469
stop_revs = searcher.find_seen_ancestors(have_revs)
3470
searcher.stop_searching_any(stop_revs)
3471
if searcher_exhausted:
3473
return searcher.get_result()
1401
:param pb: optional progress bar to use for progress reports. If not
1402
provided a default one will be created.
1404
Returns the copied revision count and the failed revisions in a tuple:
1407
raise NotImplementedError(self.fetch)
3475
1409
@needs_read_lock
3476
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1410
def missing_revision_ids(self, revision_id=None):
3477
1411
"""Return the revision ids that source has that target does not.
1413
These are returned in topological order.
3479
1415
:param revision_id: only return revision ids included by this
3481
:param find_ghosts: If True find missing revisions in deep history
3482
rather than just finding the surface difference.
3483
:return: A bzrlib.graph.SearchResult.
3485
# stop searching at found target revisions.
3486
if not find_ghosts and revision_id is not None:
3487
return self._walk_to_common_revisions([revision_id])
3488
1418
# generic, possibly worst case, slow code path.
3489
1419
target_ids = set(self.target.all_revision_ids())
3490
1420
if revision_id is not None:
1421
# TODO: jam 20070210 InterRepository is internal enough that it
1422
# should assume revision_ids are already utf-8
1423
revision_id = osutils.safe_revision_id(revision_id)
3491
1424
source_ids = self.source.get_ancestry(revision_id)
3492
if source_ids[0] is not None:
3493
raise AssertionError()
1425
assert source_ids[0] is None
3494
1426
source_ids.pop(0)
3496
1428
source_ids = self.source.all_revision_ids()
3497
1429
result_set = set(source_ids).difference(target_ids)
3498
return self.source.revision_ids_to_search_result(result_set)
3501
def _same_model(source, target):
3502
"""True if source and target have the same data representation.
3504
Note: this is always called on the base class; overriding it in a
3505
subclass will have no effect.
3508
InterRepository._assert_same_model(source, target)
3510
except errors.IncompatibleRepositories, e:
3514
def _assert_same_model(source, target):
3515
"""Raise an exception if two repositories do not use the same model.
3517
if source.supports_rich_root() != target.supports_rich_root():
3518
raise errors.IncompatibleRepositories(source, target,
3519
"different rich-root support")
3520
if source._serializer != target._serializer:
3521
raise errors.IncompatibleRepositories(source, target,
3522
"different serializers")
1430
# this may look like a no-op: its not. It preserves the ordering
1431
# other_ids had while only returning the members from other_ids
1432
# that we've decided we need.
1433
return [rev_id for rev_id in source_ids if rev_id in result_set]
3525
1436
class InterSameDataRepository(InterRepository):
3526
1437
"""Code for converting between repositories that represent the same data.
3528
1439
Data format and model must match for this to work.
3532
1443
def _get_repo_format_to_test(self):
3533
"""Repository format for testing with.
3535
InterSameData can pull from subtree to subtree and from non-subtree to
3536
non-subtree, so we test this with the richest repository format.
3538
from bzrlib.repofmt import knitrepo
3539
return knitrepo.RepositoryFormatKnit3()
1444
"""Repository format for testing with."""
1445
return RepositoryFormat.get_default_format()
3542
1448
def is_compatible(source, target):
3543
return InterRepository._same_model(source, target)
1449
if source.supports_rich_root() != target.supports_rich_root():
1451
if source._serializer != target._serializer:
1456
def copy_content(self, revision_id=None):
1457
"""Make a complete copy of the content in self into destination.
1459
This copies both the repository's revision data, and configuration information
1460
such as the make_working_trees setting.
1462
This is a destructive operation! Do not use it on existing
1465
:param revision_id: Only copy the content needed to construct
1466
revision_id and its parents.
1469
self.target.set_make_working_trees(self.source.make_working_trees())
1470
except NotImplementedError:
1472
# TODO: jam 20070210 This is fairly internal, so we should probably
1473
# just assert that revision_id is not unicode.
1474
revision_id = osutils.safe_revision_id(revision_id)
1475
# but don't bother fetching if we have the needed data now.
1476
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1477
self.target.has_revision(revision_id)):
1479
self.target.fetch(self.source, revision_id=revision_id)
1482
def fetch(self, revision_id=None, pb=None):
1483
"""See InterRepository.fetch()."""
1484
from bzrlib.fetch import GenericRepoFetcher
1485
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1486
self.source, self.source._format, self.target,
1487
self.target._format)
1488
# TODO: jam 20070210 This should be an assert, not a translate
1489
revision_id = osutils.safe_revision_id(revision_id)
1490
f = GenericRepoFetcher(to_repository=self.target,
1491
from_repository=self.source,
1492
last_revision=revision_id,
1494
return f.count_copied, f.failed_revisions
3546
1497
class InterWeaveRepo(InterSameDataRepository):
3547
"""Optimised code paths between Weave based repositories.
3549
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3550
implemented lazy inter-object optimisation.
1498
"""Optimised code paths between Weave based repositories."""
3554
1501
def _get_repo_format_to_test(self):
3667
1625
def is_compatible(source, target):
3668
1626
"""Be compatible with known Knit formats.
3670
1628
We don't test for the stores being of specific types because that
3671
could lead to confusing results, and there is no need to be
1629
could lead to confusing results, and there is no need to be
3672
1630
overly general.
3674
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
1632
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1
3676
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3677
isinstance(target._format, RepositoryFormatKnit))
1634
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1635
isinstance(target._format, (RepositoryFormatKnit1)))
3678
1636
except AttributeError:
3680
return are_knits and InterRepository._same_model(source, target)
1640
def fetch(self, revision_id=None, pb=None):
1641
"""See InterRepository.fetch()."""
1642
from bzrlib.fetch import KnitRepoFetcher
1643
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1644
self.source, self.source._format, self.target, self.target._format)
1645
# TODO: jam 20070210 This should be an assert, not a translate
1646
revision_id = osutils.safe_revision_id(revision_id)
1647
f = KnitRepoFetcher(to_repository=self.target,
1648
from_repository=self.source,
1649
last_revision=revision_id,
1651
return f.count_copied, f.failed_revisions
3682
1653
@needs_read_lock
3683
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1654
def missing_revision_ids(self, revision_id=None):
3684
1655
"""See InterRepository.missing_revision_ids()."""
3685
1656
if revision_id is not None:
3686
1657
source_ids = self.source.get_ancestry(revision_id)
3687
if source_ids[0] is not None:
3688
raise AssertionError()
1658
assert source_ids[0] is None
3689
1659
source_ids.pop(0)
3691
source_ids = self.source.all_revision_ids()
1661
source_ids = self.source._all_possible_ids()
3692
1662
source_ids_set = set(source_ids)
3693
1663
# source_ids is the worst possible case we may need to pull.
3694
1664
# now we want to filter source_ids against what we actually
3695
1665
# have in target, but don't try to check for existence where we know
3696
1666
# we do not have a revision as that would be pointless.
3697
target_ids = set(self.target.all_revision_ids())
1667
target_ids = set(self.target._all_possible_ids())
3698
1668
possibly_present_revisions = target_ids.intersection(source_ids_set)
3699
actually_present_revisions = set(
3700
self.target._eliminate_revisions_not_present(possibly_present_revisions))
1669
actually_present_revisions = set(self.target._eliminate_revisions_not_present(possibly_present_revisions))
3701
1670
required_revisions = source_ids_set.difference(actually_present_revisions)
1671
required_topo_revisions = [rev_id for rev_id in source_ids if rev_id in required_revisions]
3702
1672
if revision_id is not None:
3703
1673
# we used get_ancestry to determine source_ids then we are assured all
3704
1674
# revisions referenced are present as they are installed in topological order.
3705
1675
# and the tip revision was validated by get_ancestry.
3706
result_set = required_revisions
1676
return required_topo_revisions
3708
# if we just grabbed the possibly available ids, then
1678
# if we just grabbed the possibly available ids, then
3709
1679
# we only have an estimate of whats available and need to validate
3710
1680
# that against the revision records.
3712
self.source._eliminate_revisions_not_present(required_revisions))
3713
return self.source.revision_ids_to_search_result(result_set)
3716
class InterDifferingSerializer(InterRepository):
3719
def _get_repo_format_to_test(self):
3723
def is_compatible(source, target):
3724
"""Be compatible with Knit2 source and Knit3 target"""
3725
# This is redundant with format.check_conversion_target(), however that
3726
# raises an exception, and we just want to say "False" as in we won't
3727
# support converting between these formats.
3728
if 'IDS_never' in debug.debug_flags:
3730
if source.supports_rich_root() and not target.supports_rich_root():
3732
if (source._format.supports_tree_reference
3733
and not target._format.supports_tree_reference):
3735
if target._fallback_repositories and target._format.supports_chks:
3736
# IDS doesn't know how to copy CHKs for the parent inventories it
3737
# adds to stacked repos.
3739
if 'IDS_always' in debug.debug_flags:
3741
# Only use this code path for local source and target. IDS does far
3742
# too much IO (both bandwidth and roundtrips) over a network.
3743
if not source.bzrdir.transport.base.startswith('file:///'):
3745
if not target.bzrdir.transport.base.startswith('file:///'):
3749
def _get_trees(self, revision_ids, cache):
3751
for rev_id in revision_ids:
3753
possible_trees.append((rev_id, cache[rev_id]))
3755
# Not cached, but inventory might be present anyway.
3757
tree = self.source.revision_tree(rev_id)
3758
except errors.NoSuchRevision:
3759
# Nope, parent is ghost.
3762
cache[rev_id] = tree
3763
possible_trees.append((rev_id, tree))
3764
return possible_trees
3766
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3767
"""Get the best delta and base for this revision.
3769
:return: (basis_id, delta)
3772
# Generate deltas against each tree, to find the shortest.
3773
texts_possibly_new_in_tree = set()
3774
for basis_id, basis_tree in possible_trees:
3775
delta = tree.inventory._make_delta(basis_tree.inventory)
3776
for old_path, new_path, file_id, new_entry in delta:
3777
if new_path is None:
3778
# This file_id isn't present in the new rev, so we don't
3782
# Rich roots are handled elsewhere...
3784
kind = new_entry.kind
3785
if kind != 'directory' and kind != 'file':
3786
# No text record associated with this inventory entry.
3788
# This is a directory or file that has changed somehow.
3789
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3790
deltas.append((len(delta), basis_id, delta))
3792
return deltas[0][1:]
3794
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3795
"""Find all parent revisions that are absent, but for which the
3796
inventory is present, and copy those inventories.
3798
This is necessary to preserve correctness when the source is stacked
3799
without fallbacks configured. (Note that in cases like upgrade the
3800
source may be not have _fallback_repositories even though it is
3804
for parents in parent_map.values():
3805
parent_revs.update(parents)
3806
present_parents = self.source.get_parent_map(parent_revs)
3807
absent_parents = set(parent_revs).difference(present_parents)
3808
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3809
(rev_id,) for rev_id in absent_parents)
3810
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3811
for parent_tree in self.source.revision_trees(parent_inv_ids):
3812
current_revision_id = parent_tree.get_revision_id()
3813
parents_parents_keys = parent_invs_keys_for_stacking[
3814
(current_revision_id,)]
3815
parents_parents = [key[-1] for key in parents_parents_keys]
3816
basis_id = _mod_revision.NULL_REVISION
3817
basis_tree = self.source.revision_tree(basis_id)
3818
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3819
self.target.add_inventory_by_delta(
3820
basis_id, delta, current_revision_id, parents_parents)
3821
cache[current_revision_id] = parent_tree
3823
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3824
"""Fetch across a few revisions.
3826
:param revision_ids: The revisions to copy
3827
:param basis_id: The revision_id of a tree that must be in cache, used
3828
as a basis for delta when no other base is available
3829
:param cache: A cache of RevisionTrees that we can use.
3830
:param a_graph: A Graph object to determine the heads() of the
3831
rich-root data stream.
3832
:return: The revision_id of the last converted tree. The RevisionTree
3833
for it will be in cache
3835
# Walk though all revisions; get inventory deltas, copy referenced
3836
# texts that delta references, insert the delta, revision and
3838
root_keys_to_create = set()
3841
pending_revisions = []
3842
parent_map = self.source.get_parent_map(revision_ids)
3843
self._fetch_parent_invs_for_stacking(parent_map, cache)
3844
self.source._safe_to_return_from_cache = True
3845
for tree in self.source.revision_trees(revision_ids):
3846
# Find a inventory delta for this revision.
3847
# Find text entries that need to be copied, too.
3848
current_revision_id = tree.get_revision_id()
3849
parent_ids = parent_map.get(current_revision_id, ())
3850
parent_trees = self._get_trees(parent_ids, cache)
3851
possible_trees = list(parent_trees)
3852
if len(possible_trees) == 0:
3853
# There either aren't any parents, or the parents are ghosts,
3854
# so just use the last converted tree.
3855
possible_trees.append((basis_id, cache[basis_id]))
3856
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3858
revision = self.source.get_revision(current_revision_id)
3859
pending_deltas.append((basis_id, delta,
3860
current_revision_id, revision.parent_ids))
3861
if self._converting_to_rich_root:
3862
self._revision_id_to_root_id[current_revision_id] = \
3864
# Determine which texts are in present in this revision but not in
3865
# any of the available parents.
3866
texts_possibly_new_in_tree = set()
3867
for old_path, new_path, file_id, entry in delta:
3868
if new_path is None:
3869
# This file_id isn't present in the new rev
3873
if not self.target.supports_rich_root():
3874
# The target doesn't support rich root, so we don't
3877
if self._converting_to_rich_root:
3878
# This can't be copied normally, we have to insert
3880
root_keys_to_create.add((file_id, entry.revision))
3883
texts_possibly_new_in_tree.add((file_id, entry.revision))
3884
for basis_id, basis_tree in possible_trees:
3885
basis_inv = basis_tree.inventory
3886
for file_key in list(texts_possibly_new_in_tree):
3887
file_id, file_revision = file_key
3889
entry = basis_inv[file_id]
3890
except errors.NoSuchId:
3892
if entry.revision == file_revision:
3893
texts_possibly_new_in_tree.remove(file_key)
3894
text_keys.update(texts_possibly_new_in_tree)
3895
pending_revisions.append(revision)
3896
cache[current_revision_id] = tree
3897
basis_id = current_revision_id
3898
self.source._safe_to_return_from_cache = False
3900
from_texts = self.source.texts
3901
to_texts = self.target.texts
3902
if root_keys_to_create:
3903
root_stream = _mod_fetch._new_root_data_stream(
3904
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3905
self.source, graph=a_graph)
3906
to_texts.insert_record_stream(root_stream)
3907
to_texts.insert_record_stream(from_texts.get_record_stream(
3908
text_keys, self.target._format._fetch_order,
3909
not self.target._format._fetch_uses_deltas))
3910
# insert inventory deltas
3911
for delta in pending_deltas:
3912
self.target.add_inventory_by_delta(*delta)
3913
if self.target._fallback_repositories:
3914
# Make sure this stacked repository has all the parent inventories
3915
# for the new revisions that we are about to insert. We do this
3916
# before adding the revisions so that no revision is added until
3917
# all the inventories it may depend on are added.
3918
# Note that this is overzealous, as we may have fetched these in an
3921
revision_ids = set()
3922
for revision in pending_revisions:
3923
revision_ids.add(revision.revision_id)
3924
parent_ids.update(revision.parent_ids)
3925
parent_ids.difference_update(revision_ids)
3926
parent_ids.discard(_mod_revision.NULL_REVISION)
3927
parent_map = self.source.get_parent_map(parent_ids)
3928
# we iterate over parent_map and not parent_ids because we don't
3929
# want to try copying any revision which is a ghost
3930
for parent_tree in self.source.revision_trees(parent_map):
3931
current_revision_id = parent_tree.get_revision_id()
3932
parents_parents = parent_map[current_revision_id]
3933
possible_trees = self._get_trees(parents_parents, cache)
3934
if len(possible_trees) == 0:
3935
# There either aren't any parents, or the parents are
3936
# ghosts, so just use the last converted tree.
3937
possible_trees.append((basis_id, cache[basis_id]))
3938
basis_id, delta = self._get_delta_for_revision(parent_tree,
3939
parents_parents, possible_trees)
3940
self.target.add_inventory_by_delta(
3941
basis_id, delta, current_revision_id, parents_parents)
3942
# insert signatures and revisions
3943
for revision in pending_revisions:
3945
signature = self.source.get_signature_text(
3946
revision.revision_id)
3947
self.target.add_signature_text(revision.revision_id,
3949
except errors.NoSuchRevision:
3951
self.target.add_revision(revision.revision_id, revision)
3954
def _fetch_all_revisions(self, revision_ids, pb):
3955
"""Fetch everything for the list of revisions.
3957
:param revision_ids: The list of revisions to fetch. Must be in
3959
:param pb: A ProgressTask
3962
basis_id, basis_tree = self._get_basis(revision_ids[0])
3964
cache = lru_cache.LRUCache(100)
3965
cache[basis_id] = basis_tree
3966
del basis_tree # We don't want to hang on to it here
3968
if self._converting_to_rich_root and len(revision_ids) > 100:
3969
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3974
for offset in range(0, len(revision_ids), batch_size):
3975
self.target.start_write_group()
3977
pb.update('Transferring revisions', offset,
3979
batch = revision_ids[offset:offset+batch_size]
3980
basis_id = self._fetch_batch(batch, basis_id, cache,
3983
self.source._safe_to_return_from_cache = False
3984
self.target.abort_write_group()
3987
hint = self.target.commit_write_group()
3990
if hints and self.target._format.pack_compresses:
3991
self.target.pack(hint=hints)
3992
pb.update('Transferring revisions', len(revision_ids),
3996
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3998
"""See InterRepository.fetch()."""
3999
if fetch_spec is not None:
4000
raise AssertionError("Not implemented yet...")
4001
# See <https://launchpad.net/bugs/456077> asking for a warning here
4003
# nb this is only active for local-local fetches; other things using
4005
ui.ui_factory.warn_cross_format_fetch(self.source._format,
4006
self.target._format)
4007
if (not self.source.supports_rich_root()
4008
and self.target.supports_rich_root()):
4009
self._converting_to_rich_root = True
4010
self._revision_id_to_root_id = {}
4012
self._converting_to_rich_root = False
4013
revision_ids = self.target.search_missing_revision_ids(self.source,
4014
revision_id, find_ghosts=find_ghosts).get_keys()
4015
if not revision_ids:
4017
revision_ids = tsort.topo_sort(
4018
self.source.get_graph().get_parent_map(revision_ids))
4019
if not revision_ids:
4021
# Walk though all revisions; get inventory deltas, copy referenced
4022
# texts that delta references, insert the delta, revision and
4025
my_pb = ui.ui_factory.nested_progress_bar()
4028
symbol_versioning.warn(
4029
symbol_versioning.deprecated_in((1, 14, 0))
4030
% "pb parameter to fetch()")
4033
self._fetch_all_revisions(revision_ids, pb)
4035
if my_pb is not None:
4037
return len(revision_ids), 0
4039
def _get_basis(self, first_revision_id):
4040
"""Get a revision and tree which exists in the target.
4042
This assumes that first_revision_id is selected for transmission
4043
because all other ancestors are already present. If we can't find an
4044
ancestor we fall back to NULL_REVISION since we know that is safe.
4046
:return: (basis_id, basis_tree)
4048
first_rev = self.source.get_revision(first_revision_id)
4050
basis_id = first_rev.parent_ids[0]
4051
# only valid as a basis if the target has it
4052
self.target.get_revision(basis_id)
4053
# Try to get a basis tree - if its a ghost it will hit the
4054
# NoSuchRevision case.
4055
basis_tree = self.source.revision_tree(basis_id)
4056
except (IndexError, errors.NoSuchRevision):
4057
basis_id = _mod_revision.NULL_REVISION
4058
basis_tree = self.source.revision_tree(basis_id)
4059
return basis_id, basis_tree
4062
InterRepository.register_optimiser(InterDifferingSerializer)
1681
return self.source._eliminate_revisions_not_present(required_topo_revisions)
1684
class InterModel1and2(InterRepository):
1687
def _get_repo_format_to_test(self):
1691
def is_compatible(source, target):
1692
if not source.supports_rich_root() and target.supports_rich_root():
1698
def fetch(self, revision_id=None, pb=None):
1699
"""See InterRepository.fetch()."""
1700
from bzrlib.fetch import Model1toKnit2Fetcher
1701
# TODO: jam 20070210 This should be an assert, not a translate
1702
revision_id = osutils.safe_revision_id(revision_id)
1703
f = Model1toKnit2Fetcher(to_repository=self.target,
1704
from_repository=self.source,
1705
last_revision=revision_id,
1707
return f.count_copied, f.failed_revisions
1710
def copy_content(self, revision_id=None):
1711
"""Make a complete copy of the content in self into destination.
1713
This is a destructive operation! Do not use it on existing
1716
:param revision_id: Only copy the content needed to construct
1717
revision_id and its parents.
1720
self.target.set_make_working_trees(self.source.make_working_trees())
1721
except NotImplementedError:
1723
# TODO: jam 20070210 Internal, assert, don't translate
1724
revision_id = osutils.safe_revision_id(revision_id)
1725
# but don't bother fetching if we have the needed data now.
1726
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1727
self.target.has_revision(revision_id)):
1729
self.target.fetch(self.source, revision_id=revision_id)
1732
class InterKnit1and2(InterKnitRepo):
1735
def _get_repo_format_to_test(self):
1739
def is_compatible(source, target):
1740
"""Be compatible with Knit1 source and Knit3 target"""
1741
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit3
1743
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1, \
1744
RepositoryFormatKnit3
1745
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1746
isinstance(target._format, (RepositoryFormatKnit3)))
1747
except AttributeError:
1751
def fetch(self, revision_id=None, pb=None):
1752
"""See InterRepository.fetch()."""
1753
from bzrlib.fetch import Knit1to2Fetcher
1754
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1755
self.source, self.source._format, self.target,
1756
self.target._format)
1757
# TODO: jam 20070210 This should be an assert, not a translate
1758
revision_id = osutils.safe_revision_id(revision_id)
1759
f = Knit1to2Fetcher(to_repository=self.target,
1760
from_repository=self.source,
1761
last_revision=revision_id,
1763
return f.count_copied, f.failed_revisions
1766
class InterRemoteRepository(InterRepository):
1767
"""Code for converting between RemoteRepository objects.
1769
This just gets an non-remote repository from the RemoteRepository, and calls
1770
InterRepository.get again.
1773
def __init__(self, source, target):
1774
if isinstance(source, remote.RemoteRepository):
1775
source._ensure_real()
1776
real_source = source._real_repository
1778
real_source = source
1779
if isinstance(target, remote.RemoteRepository):
1780
target._ensure_real()
1781
real_target = target._real_repository
1783
real_target = target
1784
self.real_inter = InterRepository.get(real_source, real_target)
1787
def is_compatible(source, target):
1788
if isinstance(source, remote.RemoteRepository):
1790
if isinstance(target, remote.RemoteRepository):
1794
def copy_content(self, revision_id=None):
1795
self.real_inter.copy_content(revision_id=revision_id)
1797
def fetch(self, revision_id=None, pb=None):
1798
self.real_inter.fetch(revision_id=revision_id, pb=pb)
1801
def _get_repo_format_to_test(self):
4063
1805
InterRepository.register_optimiser(InterSameDataRepository)
4064
1806
InterRepository.register_optimiser(InterWeaveRepo)
4065
1807
InterRepository.register_optimiser(InterKnitRepo)
1808
InterRepository.register_optimiser(InterModel1and2)
1809
InterRepository.register_optimiser(InterKnit1and2)
1810
InterRepository.register_optimiser(InterRemoteRepository)
4068
1813
class CopyConverter(object):
4069
1814
"""A repository conversion tool which just performs a copy of the content.
4071
1816
This is slow but quite reliable.
4077
1822
:param target_format: The format the resulting repository should be.
4079
1824
self.target_format = target_format
4081
1826
def convert(self, repo, pb):
4082
1827
"""Perform the conversion of to_convert, giving feedback via pb.
4084
1829
:param to_convert: The disk object to convert.
4085
1830
:param pb: a progress bar to use for progress information.
4087
pb = ui.ui_factory.nested_progress_bar()
4090
1835
# this is only useful with metadir layouts - separated repo content.
4091
1836
# trigger an assertion if not such
4092
1837
repo._format.get_format_string()
4093
1838
self.repo_dir = repo.bzrdir
4094
pb.update('Moving repository to repository.backup')
1839
self.step('Moving repository to repository.backup')
4095
1840
self.repo_dir.transport.move('repository', 'repository.backup')
4096
1841
backup_transport = self.repo_dir.transport.clone('repository.backup')
4097
1842
repo._format.check_conversion_target(self.target_format)
4098
1843
self.source_repo = repo._format.open(self.repo_dir,
4100
1845
_override_transport=backup_transport)
4101
pb.update('Creating new repository')
1846
self.step('Creating new repository')
4102
1847
converted = self.target_format.initialize(self.repo_dir,
4103
1848
self.source_repo.is_shared())
4104
1849
converted.lock_write()
4106
pb.update('Copying content')
1851
self.step('Copying content into repository.')
4107
1852
self.source_repo.copy_content_into(converted)
4109
1854
converted.unlock()
4110
pb.update('Deleting old repository content')
1855
self.step('Deleting old repository content.')
4111
1856
self.repo_dir.transport.delete_tree('repository.backup')
4112
ui.ui_factory.note('repository converted')
1857
self.pb.note('repository converted')
1859
def step(self, message):
1860
"""Update the pb by a step."""
1862
self.pb.update(message, self.count, self.total)
1865
class CommitBuilder(object):
1866
"""Provides an interface to build up a commit.
1868
This allows describing a tree to be committed without needing to
1869
know the internals of the format of the repository.
1872
record_root_entry = False
1873
def __init__(self, repository, parents, config, timestamp=None,
1874
timezone=None, committer=None, revprops=None,
1876
"""Initiate a CommitBuilder.
1878
:param repository: Repository to commit to.
1879
:param parents: Revision ids of the parents of the new revision.
1880
:param config: Configuration to use.
1881
:param timestamp: Optional timestamp recorded for commit.
1882
:param timezone: Optional timezone for timestamp.
1883
:param committer: Optional committer to set for commit.
1884
:param revprops: Optional dictionary of revision properties.
1885
:param revision_id: Optional revision id.
1887
self._config = config
1889
if committer is None:
1890
self._committer = self._config.username()
1892
assert isinstance(committer, basestring), type(committer)
1893
self._committer = committer
1895
self.new_inventory = Inventory(None)
1896
self._new_revision_id = osutils.safe_revision_id(revision_id)
1897
self.parents = parents
1898
self.repository = repository
1901
if revprops is not None:
1902
self._revprops.update(revprops)
1904
if timestamp is None:
1905
timestamp = time.time()
1906
# Restrict resolution to 1ms
1907
self._timestamp = round(timestamp, 3)
1909
if timezone is None:
1910
self._timezone = osutils.local_time_offset()
1912
self._timezone = int(timezone)
1914
self._generate_revision_if_needed()
1916
def commit(self, message):
1917
"""Make the actual commit.
1919
:return: The revision id of the recorded revision.
1921
rev = _mod_revision.Revision(
1922
timestamp=self._timestamp,
1923
timezone=self._timezone,
1924
committer=self._committer,
1926
inventory_sha1=self.inv_sha1,
1927
revision_id=self._new_revision_id,
1928
properties=self._revprops)
1929
rev.parent_ids = self.parents
1930
self.repository.add_revision(self._new_revision_id, rev,
1931
self.new_inventory, self._config)
1932
return self._new_revision_id
1934
def revision_tree(self):
1935
"""Return the tree that was just committed.
1937
After calling commit() this can be called to get a RevisionTree
1938
representing the newly committed tree. This is preferred to
1939
calling Repository.revision_tree() because that may require
1940
deserializing the inventory, while we already have a copy in
1943
return RevisionTree(self.repository, self.new_inventory,
1944
self._new_revision_id)
1946
def finish_inventory(self):
1947
"""Tell the builder that the inventory is finished."""
1948
if self.new_inventory.root is None:
1949
symbol_versioning.warn('Root entry should be supplied to'
1950
' record_entry_contents, as of bzr 0.10.',
1951
DeprecationWarning, stacklevel=2)
1952
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
1953
self.new_inventory.revision_id = self._new_revision_id
1954
self.inv_sha1 = self.repository.add_inventory(
1955
self._new_revision_id,
1960
def _gen_revision_id(self):
1961
"""Return new revision-id."""
1962
return generate_ids.gen_revision_id(self._config.username(),
1965
def _generate_revision_if_needed(self):
1966
"""Create a revision id if None was supplied.
1968
If the repository can not support user-specified revision ids
1969
they should override this function and raise CannotSetRevisionId
1970
if _new_revision_id is not None.
1972
:raises: CannotSetRevisionId
1974
if self._new_revision_id is None:
1975
self._new_revision_id = self._gen_revision_id()
1977
def record_entry_contents(self, ie, parent_invs, path, tree):
1978
"""Record the content of ie from tree into the commit if needed.
1980
Side effect: sets ie.revision when unchanged
1982
:param ie: An inventory entry present in the commit.
1983
:param parent_invs: The inventories of the parent revisions of the
1985
:param path: The path the entry is at in the tree.
1986
:param tree: The tree which contains this entry and should be used to
1989
if self.new_inventory.root is None and ie.parent_id is not None:
1990
symbol_versioning.warn('Root entry should be supplied to'
1991
' record_entry_contents, as of bzr 0.10.',
1992
DeprecationWarning, stacklevel=2)
1993
self.record_entry_contents(tree.inventory.root.copy(), parent_invs,
1995
self.new_inventory.add(ie)
1997
# ie.revision is always None if the InventoryEntry is considered
1998
# for committing. ie.snapshot will record the correct revision
1999
# which may be the sole parent if it is untouched.
2000
if ie.revision is not None:
2003
# In this revision format, root entries have no knit or weave
2004
if ie is self.new_inventory.root:
2005
# When serializing out to disk and back in
2006
# root.revision is always _new_revision_id
2007
ie.revision = self._new_revision_id
2009
previous_entries = ie.find_previous_heads(
2011
self.repository.weave_store,
2012
self.repository.get_transaction())
2013
# we are creating a new revision for ie in the history store
2015
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
2017
def modified_directory(self, file_id, file_parents):
2018
"""Record the presence of a symbolic link.
2020
:param file_id: The file_id of the link to record.
2021
:param file_parents: The per-file parent revision ids.
2023
self._add_text_to_weave(file_id, [], file_parents.keys())
2025
def modified_reference(self, file_id, file_parents):
2026
"""Record the modification of a reference.
2028
:param file_id: The file_id of the link to record.
2029
:param file_parents: The per-file parent revision ids.
2031
self._add_text_to_weave(file_id, [], file_parents.keys())
2033
def modified_file_text(self, file_id, file_parents,
2034
get_content_byte_lines, text_sha1=None,
2036
"""Record the text of file file_id
2038
:param file_id: The file_id of the file to record the text of.
2039
:param file_parents: The per-file parent revision ids.
2040
:param get_content_byte_lines: A callable which will return the byte
2042
:param text_sha1: Optional SHA1 of the file contents.
2043
:param text_size: Optional size of the file contents.
2045
# mutter('storing text of file {%s} in revision {%s} into %r',
2046
# file_id, self._new_revision_id, self.repository.weave_store)
2047
# special case to avoid diffing on renames or
2049
if (len(file_parents) == 1
2050
and text_sha1 == file_parents.values()[0].text_sha1
2051
and text_size == file_parents.values()[0].text_size):
2052
previous_ie = file_parents.values()[0]
2053
versionedfile = self.repository.weave_store.get_weave(file_id,
2054
self.repository.get_transaction())
2055
versionedfile.clone_text(self._new_revision_id,
2056
previous_ie.revision, file_parents.keys())
2057
return text_sha1, text_size
2059
new_lines = get_content_byte_lines()
2060
# TODO: Rather than invoking sha_strings here, _add_text_to_weave
2061
# should return the SHA1 and size
2062
self._add_text_to_weave(file_id, new_lines, file_parents.keys())
2063
return osutils.sha_strings(new_lines), \
2064
sum(map(len, new_lines))
2066
def modified_link(self, file_id, file_parents, link_target):
2067
"""Record the presence of a symbolic link.
2069
:param file_id: The file_id of the link to record.
2070
:param file_parents: The per-file parent revision ids.
2071
:param link_target: Target location of this link.
2073
self._add_text_to_weave(file_id, [], file_parents.keys())
2075
def _add_text_to_weave(self, file_id, new_lines, parents):
2076
versionedfile = self.repository.weave_store.get_weave_or_empty(
2077
file_id, self.repository.get_transaction())
2078
versionedfile.add_lines(self._new_revision_id, parents, new_lines)
2079
versionedfile.clear_cache()
2082
class _CommitBuilder(CommitBuilder):
2083
"""Temporary class so old CommitBuilders are detected properly
2085
Note: CommitBuilder works whether or not root entry is recorded.
2088
record_root_entry = True
2091
class RootCommitBuilder(CommitBuilder):
2092
"""This commitbuilder actually records the root id"""
2094
record_root_entry = True
2096
def record_entry_contents(self, ie, parent_invs, path, tree):
2097
"""Record the content of ie from tree into the commit if needed.
2099
Side effect: sets ie.revision when unchanged
2101
:param ie: An inventory entry present in the commit.
2102
:param parent_invs: The inventories of the parent revisions of the
2104
:param path: The path the entry is at in the tree.
2105
:param tree: The tree which contains this entry and should be used to
2108
assert self.new_inventory.root is not None or ie.parent_id is None
2109
self.new_inventory.add(ie)
2111
# ie.revision is always None if the InventoryEntry is considered
2112
# for committing. ie.snapshot will record the correct revision
2113
# which may be the sole parent if it is untouched.
2114
if ie.revision is not None:
2117
previous_entries = ie.find_previous_heads(
2119
self.repository.weave_store,
2120
self.repository.get_transaction())
2121
# we are creating a new revision for ie in the history store
2123
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
4116
2126
_unescape_map = {
4141
2151
if _unescape_re is None:
4142
2152
_unescape_re = re.compile('\&([^;]*);')
4143
2153
return _unescape_re.sub(_unescaper, data)
4146
class _VersionedFileChecker(object):
4148
def __init__(self, repository, text_key_references=None, ancestors=None):
4149
self.repository = repository
4150
self.text_index = self.repository._generate_text_key_index(
4151
text_key_references=text_key_references, ancestors=ancestors)
4153
def calculate_file_version_parents(self, text_key):
4154
"""Calculate the correct parents for a file version according to
4157
parent_keys = self.text_index[text_key]
4158
if parent_keys == [_mod_revision.NULL_REVISION]:
4160
return tuple(parent_keys)
4162
def check_file_version_parents(self, texts, progress_bar=None):
4163
"""Check the parents stored in a versioned file are correct.
4165
It also detects file versions that are not referenced by their
4166
corresponding revision's inventory.
4168
:returns: A tuple of (wrong_parents, dangling_file_versions).
4169
wrong_parents is a dict mapping {revision_id: (stored_parents,
4170
correct_parents)} for each revision_id where the stored parents
4171
are not correct. dangling_file_versions is a set of (file_id,
4172
revision_id) tuples for versions that are present in this versioned
4173
file, but not used by the corresponding inventory.
4175
local_progress = None
4176
if progress_bar is None:
4177
local_progress = ui.ui_factory.nested_progress_bar()
4178
progress_bar = local_progress
4180
return self._check_file_version_parents(texts, progress_bar)
4183
local_progress.finished()
4185
def _check_file_version_parents(self, texts, progress_bar):
4186
"""See check_file_version_parents."""
4188
self.file_ids = set([file_id for file_id, _ in
4189
self.text_index.iterkeys()])
4190
# text keys is now grouped by file_id
4191
n_versions = len(self.text_index)
4192
progress_bar.update('loading text store', 0, n_versions)
4193
parent_map = self.repository.texts.get_parent_map(self.text_index)
4194
# On unlistable transports this could well be empty/error...
4195
text_keys = self.repository.texts.keys()
4196
unused_keys = frozenset(text_keys) - set(self.text_index)
4197
for num, key in enumerate(self.text_index.iterkeys()):
4198
progress_bar.update('checking text graph', num, n_versions)
4199
correct_parents = self.calculate_file_version_parents(key)
4201
knit_parents = parent_map[key]
4202
except errors.RevisionNotPresent:
4205
if correct_parents != knit_parents:
4206
wrong_parents[key] = (knit_parents, correct_parents)
4207
return wrong_parents, unused_keys
4210
def _old_get_graph(repository, revision_id):
4211
"""DO NOT USE. That is all. I'm serious."""
4212
graph = repository.get_graph()
4213
revision_graph = dict(((key, value) for key, value in
4214
graph.iter_ancestry([revision_id]) if value is not None))
4215
return _strip_NULL_ghosts(revision_graph)
4218
def _strip_NULL_ghosts(revision_graph):
4219
"""Also don't use this. more compatibility code for unmigrated clients."""
4220
# Filter ghosts, and null:
4221
if _mod_revision.NULL_REVISION in revision_graph:
4222
del revision_graph[_mod_revision.NULL_REVISION]
4223
for key, parents in revision_graph.items():
4224
revision_graph[key] = tuple(parent for parent in parents if parent
4226
return revision_graph
4229
class StreamSink(object):
4230
"""An object that can insert a stream into a repository.
4232
This interface handles the complexity of reserialising inventories and
4233
revisions from different formats, and allows unidirectional insertion into
4234
stacked repositories without looking for the missing basis parents
4238
def __init__(self, target_repo):
4239
self.target_repo = target_repo
4241
def insert_stream(self, stream, src_format, resume_tokens):
4242
"""Insert a stream's content into the target repository.
4244
:param src_format: a bzr repository format.
4246
:return: a list of resume tokens and an iterable of keys additional
4247
items required before the insertion can be completed.
4249
self.target_repo.lock_write()
4252
self.target_repo.resume_write_group(resume_tokens)
4255
self.target_repo.start_write_group()
4258
# locked_insert_stream performs a commit|suspend.
4259
return self._locked_insert_stream(stream, src_format, is_resume)
4261
self.target_repo.abort_write_group(suppress_errors=True)
4264
self.target_repo.unlock()
4266
def _locked_insert_stream(self, stream, src_format, is_resume):
4267
to_serializer = self.target_repo._format._serializer
4268
src_serializer = src_format._serializer
4270
if to_serializer == src_serializer:
4271
# If serializers match and the target is a pack repository, set the
4272
# write cache size on the new pack. This avoids poor performance
4273
# on transports where append is unbuffered (such as
4274
# RemoteTransport). This is safe to do because nothing should read
4275
# back from the target repository while a stream with matching
4276
# serialization is being inserted.
4277
# The exception is that a delta record from the source that should
4278
# be a fulltext may need to be expanded by the target (see
4279
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4280
# explicitly flush any buffered writes first in that rare case.
4282
new_pack = self.target_repo._pack_collection._new_pack
4283
except AttributeError:
4284
# Not a pack repository
4287
new_pack.set_write_cache_size(1024*1024)
4288
for substream_type, substream in stream:
4289
if 'stream' in debug.debug_flags:
4290
mutter('inserting substream: %s', substream_type)
4291
if substream_type == 'texts':
4292
self.target_repo.texts.insert_record_stream(substream)
4293
elif substream_type == 'inventories':
4294
if src_serializer == to_serializer:
4295
self.target_repo.inventories.insert_record_stream(
4298
self._extract_and_insert_inventories(
4299
substream, src_serializer)
4300
elif substream_type == 'inventory-deltas':
4301
ui.ui_factory.warn_cross_format_fetch(src_format,
4302
self.target_repo._format)
4303
self._extract_and_insert_inventory_deltas(
4304
substream, src_serializer)
4305
elif substream_type == 'chk_bytes':
4306
# XXX: This doesn't support conversions, as it assumes the
4307
# conversion was done in the fetch code.
4308
self.target_repo.chk_bytes.insert_record_stream(substream)
4309
elif substream_type == 'revisions':
4310
# This may fallback to extract-and-insert more often than
4311
# required if the serializers are different only in terms of
4313
if src_serializer == to_serializer:
4314
self.target_repo.revisions.insert_record_stream(
4317
self._extract_and_insert_revisions(substream,
4319
elif substream_type == 'signatures':
4320
self.target_repo.signatures.insert_record_stream(substream)
4322
raise AssertionError('kaboom! %s' % (substream_type,))
4323
# Done inserting data, and the missing_keys calculations will try to
4324
# read back from the inserted data, so flush the writes to the new pack
4325
# (if this is pack format).
4326
if new_pack is not None:
4327
new_pack._write_data('', flush=True)
4328
# Find all the new revisions (including ones from resume_tokens)
4329
missing_keys = self.target_repo.get_missing_parent_inventories(
4330
check_for_missing_texts=is_resume)
4332
for prefix, versioned_file in (
4333
('texts', self.target_repo.texts),
4334
('inventories', self.target_repo.inventories),
4335
('revisions', self.target_repo.revisions),
4336
('signatures', self.target_repo.signatures),
4337
('chk_bytes', self.target_repo.chk_bytes),
4339
if versioned_file is None:
4341
# TODO: key is often going to be a StaticTuple object
4342
# I don't believe we can define a method by which
4343
# (prefix,) + StaticTuple will work, though we could
4344
# define a StaticTuple.sq_concat that would allow you to
4345
# pass in either a tuple or a StaticTuple as the second
4346
# object, so instead we could have:
4347
# StaticTuple(prefix) + key here...
4348
missing_keys.update((prefix,) + key for key in
4349
versioned_file.get_missing_compression_parent_keys())
4350
except NotImplementedError:
4351
# cannot even attempt suspending, and missing would have failed
4352
# during stream insertion.
4353
missing_keys = set()
4356
# suspend the write group and tell the caller what we is
4357
# missing. We know we can suspend or else we would not have
4358
# entered this code path. (All repositories that can handle
4359
# missing keys can handle suspending a write group).
4360
write_group_tokens = self.target_repo.suspend_write_group()
4361
return write_group_tokens, missing_keys
4362
hint = self.target_repo.commit_write_group()
4363
if (to_serializer != src_serializer and
4364
self.target_repo._format.pack_compresses):
4365
self.target_repo.pack(hint=hint)
4368
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4369
target_rich_root = self.target_repo._format.rich_root_data
4370
target_tree_refs = self.target_repo._format.supports_tree_reference
4371
for record in substream:
4372
# Insert the delta directly
4373
inventory_delta_bytes = record.get_bytes_as('fulltext')
4374
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4376
parse_result = deserialiser.parse_text_bytes(
4377
inventory_delta_bytes)
4378
except inventory_delta.IncompatibleInventoryDelta, err:
4379
trace.mutter("Incompatible delta: %s", err.msg)
4380
raise errors.IncompatibleRevision(self.target_repo._format)
4381
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4382
revision_id = new_id
4383
parents = [key[0] for key in record.parents]
4384
self.target_repo.add_inventory_by_delta(
4385
basis_id, inv_delta, revision_id, parents)
4387
def _extract_and_insert_inventories(self, substream, serializer,
4389
"""Generate a new inventory versionedfile in target, converting data.
4391
The inventory is retrieved from the source, (deserializing it), and
4392
stored in the target (reserializing it in a different format).
4394
target_rich_root = self.target_repo._format.rich_root_data
4395
target_tree_refs = self.target_repo._format.supports_tree_reference
4396
for record in substream:
4397
# It's not a delta, so it must be a fulltext in the source
4398
# serializer's format.
4399
bytes = record.get_bytes_as('fulltext')
4400
revision_id = record.key[0]
4401
inv = serializer.read_inventory_from_string(bytes, revision_id)
4402
parents = [key[0] for key in record.parents]
4403
self.target_repo.add_inventory(revision_id, inv, parents)
4404
# No need to keep holding this full inv in memory when the rest of
4405
# the substream is likely to be all deltas.
4408
def _extract_and_insert_revisions(self, substream, serializer):
4409
for record in substream:
4410
bytes = record.get_bytes_as('fulltext')
4411
revision_id = record.key[0]
4412
rev = serializer.read_revision_from_string(bytes)
4413
if rev.revision_id != revision_id:
4414
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4415
self.target_repo.add_revision(revision_id, rev)
4418
if self.target_repo._format._fetch_reconcile:
4419
self.target_repo.reconcile()
4422
class StreamSource(object):
4423
"""A source of a stream for fetching between repositories."""
4425
def __init__(self, from_repository, to_format):
4426
"""Create a StreamSource streaming from from_repository."""
4427
self.from_repository = from_repository
4428
self.to_format = to_format
4430
def delta_on_metadata(self):
4431
"""Return True if delta's are permitted on metadata streams.
4433
That is on revisions and signatures.
4435
src_serializer = self.from_repository._format._serializer
4436
target_serializer = self.to_format._serializer
4437
return (self.to_format._fetch_uses_deltas and
4438
src_serializer == target_serializer)
4440
def _fetch_revision_texts(self, revs):
4441
# fetch signatures first and then the revision texts
4442
# may need to be a InterRevisionStore call here.
4443
from_sf = self.from_repository.signatures
4444
# A missing signature is just skipped.
4445
keys = [(rev_id,) for rev_id in revs]
4446
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4448
self.to_format._fetch_order,
4449
not self.to_format._fetch_uses_deltas))
4450
# If a revision has a delta, this is actually expanded inside the
4451
# insert_record_stream code now, which is an alternate fix for
4453
from_rf = self.from_repository.revisions
4454
revisions = from_rf.get_record_stream(
4456
self.to_format._fetch_order,
4457
not self.delta_on_metadata())
4458
return [('signatures', signatures), ('revisions', revisions)]
4460
def _generate_root_texts(self, revs):
4461
"""This will be called by get_stream between fetching weave texts and
4462
fetching the inventory weave.
4464
if self._rich_root_upgrade():
4465
return _mod_fetch.Inter1and2Helper(
4466
self.from_repository).generate_root_texts(revs)
4470
def get_stream(self, search):
4472
revs = search.get_keys()
4473
graph = self.from_repository.get_graph()
4474
revs = tsort.topo_sort(graph.get_parent_map(revs))
4475
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4477
for knit_kind, file_id, revisions in data_to_fetch:
4478
if knit_kind != phase:
4480
# Make a new progress bar for this phase
4481
if knit_kind == "file":
4482
# Accumulate file texts
4483
text_keys.extend([(file_id, revision) for revision in
4485
elif knit_kind == "inventory":
4486
# Now copy the file texts.
4487
from_texts = self.from_repository.texts
4488
yield ('texts', from_texts.get_record_stream(
4489
text_keys, self.to_format._fetch_order,
4490
not self.to_format._fetch_uses_deltas))
4491
# Cause an error if a text occurs after we have done the
4494
# Before we process the inventory we generate the root
4495
# texts (if necessary) so that the inventories references
4497
for _ in self._generate_root_texts(revs):
4499
# we fetch only the referenced inventories because we do not
4500
# know for unselected inventories whether all their required
4501
# texts are present in the other repository - it could be
4503
for info in self._get_inventory_stream(revs):
4505
elif knit_kind == "signatures":
4506
# Nothing to do here; this will be taken care of when
4507
# _fetch_revision_texts happens.
4509
elif knit_kind == "revisions":
4510
for record in self._fetch_revision_texts(revs):
4513
raise AssertionError("Unknown knit kind %r" % knit_kind)
4515
def get_stream_for_missing_keys(self, missing_keys):
4516
# missing keys can only occur when we are byte copying and not
4517
# translating (because translation means we don't send
4518
# unreconstructable deltas ever).
4520
keys['texts'] = set()
4521
keys['revisions'] = set()
4522
keys['inventories'] = set()
4523
keys['chk_bytes'] = set()
4524
keys['signatures'] = set()
4525
for key in missing_keys:
4526
keys[key[0]].add(key[1:])
4527
if len(keys['revisions']):
4528
# If we allowed copying revisions at this point, we could end up
4529
# copying a revision without copying its required texts: a
4530
# violation of the requirements for repository integrity.
4531
raise AssertionError(
4532
'cannot copy revisions to fill in missing deltas %s' % (
4533
keys['revisions'],))
4534
for substream_kind, keys in keys.iteritems():
4535
vf = getattr(self.from_repository, substream_kind)
4536
if vf is None and keys:
4537
raise AssertionError(
4538
"cannot fill in keys for a versioned file we don't"
4539
" have: %s needs %s" % (substream_kind, keys))
4541
# No need to stream something we don't have
4543
if substream_kind == 'inventories':
4544
# Some missing keys are genuinely ghosts, filter those out.
4545
present = self.from_repository.inventories.get_parent_map(keys)
4546
revs = [key[0] for key in present]
4547
# Get the inventory stream more-or-less as we do for the
4548
# original stream; there's no reason to assume that records
4549
# direct from the source will be suitable for the sink. (Think
4550
# e.g. 2a -> 1.9-rich-root).
4551
for info in self._get_inventory_stream(revs, missing=True):
4555
# Ask for full texts always so that we don't need more round trips
4556
# after this stream.
4557
# Some of the missing keys are genuinely ghosts, so filter absent
4558
# records. The Sink is responsible for doing another check to
4559
# ensure that ghosts don't introduce missing data for future
4561
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4562
self.to_format._fetch_order, True))
4563
yield substream_kind, stream
4565
def inventory_fetch_order(self):
4566
if self._rich_root_upgrade():
4567
return 'topological'
4569
return self.to_format._fetch_order
4571
def _rich_root_upgrade(self):
4572
return (not self.from_repository._format.rich_root_data and
4573
self.to_format.rich_root_data)
4575
def _get_inventory_stream(self, revision_ids, missing=False):
4576
from_format = self.from_repository._format
4577
if (from_format.supports_chks and self.to_format.supports_chks and
4578
from_format.network_name() == self.to_format.network_name()):
4579
raise AssertionError(
4580
"this case should be handled by GroupCHKStreamSource")
4581
elif 'forceinvdeltas' in debug.debug_flags:
4582
return self._get_convertable_inventory_stream(revision_ids,
4583
delta_versus_null=missing)
4584
elif from_format.network_name() == self.to_format.network_name():
4586
return self._get_simple_inventory_stream(revision_ids,
4588
elif (not from_format.supports_chks and not self.to_format.supports_chks
4589
and from_format._serializer == self.to_format._serializer):
4590
# Essentially the same format.
4591
return self._get_simple_inventory_stream(revision_ids,
4594
# Any time we switch serializations, we want to use an
4595
# inventory-delta based approach.
4596
return self._get_convertable_inventory_stream(revision_ids,
4597
delta_versus_null=missing)
4599
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4600
# NB: This currently reopens the inventory weave in source;
4601
# using a single stream interface instead would avoid this.
4602
from_weave = self.from_repository.inventories
4604
delta_closure = True
4606
delta_closure = not self.delta_on_metadata()
4607
yield ('inventories', from_weave.get_record_stream(
4608
[(rev_id,) for rev_id in revision_ids],
4609
self.inventory_fetch_order(), delta_closure))
4611
def _get_convertable_inventory_stream(self, revision_ids,
4612
delta_versus_null=False):
4613
# The two formats are sufficiently different that there is no fast
4614
# path, so we need to send just inventorydeltas, which any
4615
# sufficiently modern client can insert into any repository.
4616
# The StreamSink code expects to be able to
4617
# convert on the target, so we need to put bytes-on-the-wire that can
4618
# be converted. That means inventory deltas (if the remote is <1.19,
4619
# RemoteStreamSink will fallback to VFS to insert the deltas).
4620
yield ('inventory-deltas',
4621
self._stream_invs_as_deltas(revision_ids,
4622
delta_versus_null=delta_versus_null))
4624
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4625
"""Return a stream of inventory-deltas for the given rev ids.
4627
:param revision_ids: The list of inventories to transmit
4628
:param delta_versus_null: Don't try to find a minimal delta for this
4629
entry, instead compute the delta versus the NULL_REVISION. This
4630
effectively streams a complete inventory. Used for stuff like
4631
filling in missing parents, etc.
4633
from_repo = self.from_repository
4634
revision_keys = [(rev_id,) for rev_id in revision_ids]
4635
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4636
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4638
inventories = self.from_repository.iter_inventories(
4639
revision_ids, 'topological')
4640
format = from_repo._format
4641
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4642
inventory_cache = lru_cache.LRUCache(50)
4643
null_inventory = from_repo.revision_tree(
4644
_mod_revision.NULL_REVISION).inventory
4645
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4646
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4647
# repo back into a non-rich-root repo ought to be allowed)
4648
serializer = inventory_delta.InventoryDeltaSerializer(
4649
versioned_root=format.rich_root_data,
4650
tree_references=format.supports_tree_reference)
4651
for inv in inventories:
4652
key = (inv.revision_id,)
4653
parent_keys = parent_map.get(key, ())
4655
if not delta_versus_null and parent_keys:
4656
# The caller did not ask for complete inventories and we have
4657
# some parents that we can delta against. Make a delta against
4658
# each parent so that we can find the smallest.
4659
parent_ids = [parent_key[0] for parent_key in parent_keys]
4660
for parent_id in parent_ids:
4661
if parent_id not in invs_sent_so_far:
4662
# We don't know that the remote side has this basis, so
4665
if parent_id == _mod_revision.NULL_REVISION:
4666
parent_inv = null_inventory
4668
parent_inv = inventory_cache.get(parent_id, None)
4669
if parent_inv is None:
4670
parent_inv = from_repo.get_inventory(parent_id)
4671
candidate_delta = inv._make_delta(parent_inv)
4672
if (delta is None or
4673
len(delta) > len(candidate_delta)):
4674
delta = candidate_delta
4675
basis_id = parent_id
4677
# Either none of the parents ended up being suitable, or we
4678
# were asked to delta against NULL
4679
basis_id = _mod_revision.NULL_REVISION
4680
delta = inv._make_delta(null_inventory)
4681
invs_sent_so_far.add(inv.revision_id)
4682
inventory_cache[inv.revision_id] = inv
4683
delta_serialized = ''.join(
4684
serializer.delta_to_lines(basis_id, key[-1], delta))
4685
yield versionedfile.FulltextContentFactory(
4686
key, parent_keys, None, delta_serialized)
4689
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4690
stop_revision=None):
4691
"""Extend the partial history to include a given index
4693
If a stop_index is supplied, stop when that index has been reached.
4694
If a stop_revision is supplied, stop when that revision is
4695
encountered. Otherwise, stop when the beginning of history is
4698
:param stop_index: The index which should be present. When it is
4699
present, history extension will stop.
4700
:param stop_revision: The revision id which should be present. When
4701
it is encountered, history extension will stop.
4703
start_revision = partial_history_cache[-1]
4704
iterator = repo.iter_reverse_revision_history(start_revision)
4706
#skip the last revision in the list
4709
if (stop_index is not None and
4710
len(partial_history_cache) > stop_index):
4712
if partial_history_cache[-1] == stop_revision:
4714
revision_id = iterator.next()
4715
partial_history_cache.append(revision_id)
4716
except StopIteration:
added
removed
bzrlib/repository.py
