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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
from cStringIO import StringIO
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19
17
from bzrlib.lazy_import import lazy_import
20
18
lazy_import(globals(), """
24
23
from bzrlib import (
38
40
revision as _mod_revision,
47
from bzrlib.bundle import serializer
43
48
from bzrlib.revisiontree import RevisionTree
44
49
from bzrlib.store.versioned import VersionedFileStore
45
from bzrlib.store.text import TextStore
46
50
from bzrlib.testament import Testament
50
53
from bzrlib.decorators import needs_read_lock, needs_write_lock
51
54
from bzrlib.inter import InterObject
52
from bzrlib.inventory import Inventory, InventoryDirectory, ROOT_ID
53
from bzrlib.symbol_versioning import (
57
from bzrlib.trace import mutter, note, warning
55
from bzrlib.inventory import (
61
from bzrlib import registry
62
from bzrlib.trace import (
63
log_exception_quietly, note, mutter, mutter_callsite, warning)
60
66
# Old formats display a warning, but only once
61
67
_deprecation_warning_done = False
70
class CommitBuilder(object):
71
"""Provides an interface to build up a commit.
73
This allows describing a tree to be committed without needing to
74
know the internals of the format of the repository.
77
# all clients should supply tree roots.
78
record_root_entry = True
79
# the default CommitBuilder does not manage trees whose root is versioned.
80
_versioned_root = False
82
def __init__(self, repository, parents, config, timestamp=None,
83
timezone=None, committer=None, revprops=None,
85
"""Initiate a CommitBuilder.
87
:param repository: Repository to commit to.
88
:param parents: Revision ids of the parents of the new revision.
89
:param config: Configuration to use.
90
:param timestamp: Optional timestamp recorded for commit.
91
:param timezone: Optional timezone for timestamp.
92
:param committer: Optional committer to set for commit.
93
:param revprops: Optional dictionary of revision properties.
94
:param revision_id: Optional revision id.
99
self._committer = self._config.username()
101
self._committer = committer
103
self.new_inventory = Inventory(None)
104
self._new_revision_id = revision_id
105
self.parents = parents
106
self.repository = repository
109
if revprops is not None:
110
self._validate_revprops(revprops)
111
self._revprops.update(revprops)
113
if timestamp is None:
114
timestamp = time.time()
115
# Restrict resolution to 1ms
116
self._timestamp = round(timestamp, 3)
119
self._timezone = osutils.local_time_offset()
121
self._timezone = int(timezone)
123
self._generate_revision_if_needed()
124
self.__heads = graph.HeadsCache(repository.get_graph()).heads
125
self._basis_delta = []
126
# API compatibility, older code that used CommitBuilder did not call
127
# .record_delete(), which means the delta that is computed would not be
128
# valid. Callers that will call record_delete() should call
129
# .will_record_deletes() to indicate that.
130
self._recording_deletes = False
131
# memo'd check for no-op commits.
132
self._any_changes = False
134
def any_changes(self):
135
"""Return True if any entries were changed.
137
This includes merge-only changes. It is the core for the --unchanged
140
:return: True if any changes have occured.
142
return self._any_changes
144
def _validate_unicode_text(self, text, context):
145
"""Verify things like commit messages don't have bogus characters."""
147
raise ValueError('Invalid value for %s: %r' % (context, text))
149
def _validate_revprops(self, revprops):
150
for key, value in revprops.iteritems():
151
# We know that the XML serializers do not round trip '\r'
152
# correctly, so refuse to accept them
153
if not isinstance(value, basestring):
154
raise ValueError('revision property (%s) is not a valid'
155
' (unicode) string: %r' % (key, value))
156
self._validate_unicode_text(value,
157
'revision property (%s)' % (key,))
159
def commit(self, message):
160
"""Make the actual commit.
162
:return: The revision id of the recorded revision.
164
self._validate_unicode_text(message, 'commit message')
165
rev = _mod_revision.Revision(
166
timestamp=self._timestamp,
167
timezone=self._timezone,
168
committer=self._committer,
170
inventory_sha1=self.inv_sha1,
171
revision_id=self._new_revision_id,
172
properties=self._revprops)
173
rev.parent_ids = self.parents
174
self.repository.add_revision(self._new_revision_id, rev,
175
self.new_inventory, self._config)
176
self.repository.commit_write_group()
177
return self._new_revision_id
180
"""Abort the commit that is being built.
182
self.repository.abort_write_group()
184
def revision_tree(self):
185
"""Return the tree that was just committed.
187
After calling commit() this can be called to get a RevisionTree
188
representing the newly committed tree. This is preferred to
189
calling Repository.revision_tree() because that may require
190
deserializing the inventory, while we already have a copy in
193
if self.new_inventory is None:
194
self.new_inventory = self.repository.get_inventory(
195
self._new_revision_id)
196
return RevisionTree(self.repository, self.new_inventory,
197
self._new_revision_id)
199
def finish_inventory(self):
200
"""Tell the builder that the inventory is finished.
202
:return: The inventory id in the repository, which can be used with
203
repository.get_inventory.
205
if self.new_inventory is None:
206
# an inventory delta was accumulated without creating a new
208
basis_id = self.basis_delta_revision
209
self.inv_sha1 = self.repository.add_inventory_by_delta(
210
basis_id, self._basis_delta, self._new_revision_id,
213
if self.new_inventory.root is None:
214
raise AssertionError('Root entry should be supplied to'
215
' record_entry_contents, as of bzr 0.10.')
216
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
217
self.new_inventory.revision_id = self._new_revision_id
218
self.inv_sha1 = self.repository.add_inventory(
219
self._new_revision_id,
223
return self._new_revision_id
225
def _gen_revision_id(self):
226
"""Return new revision-id."""
227
return generate_ids.gen_revision_id(self._config.username(),
230
def _generate_revision_if_needed(self):
231
"""Create a revision id if None was supplied.
233
If the repository can not support user-specified revision ids
234
they should override this function and raise CannotSetRevisionId
235
if _new_revision_id is not None.
237
:raises: CannotSetRevisionId
239
if self._new_revision_id is None:
240
self._new_revision_id = self._gen_revision_id()
241
self.random_revid = True
243
self.random_revid = False
245
def _heads(self, file_id, revision_ids):
246
"""Calculate the graph heads for revision_ids in the graph of file_id.
248
This can use either a per-file graph or a global revision graph as we
249
have an identity relationship between the two graphs.
251
return self.__heads(revision_ids)
253
def _check_root(self, ie, parent_invs, tree):
254
"""Helper for record_entry_contents.
256
:param ie: An entry being added.
257
:param parent_invs: The inventories of the parent revisions of the
259
:param tree: The tree that is being committed.
261
# In this revision format, root entries have no knit or weave When
262
# serializing out to disk and back in root.revision is always
264
ie.revision = self._new_revision_id
266
def _require_root_change(self, tree):
267
"""Enforce an appropriate root object change.
269
This is called once when record_iter_changes is called, if and only if
270
the root was not in the delta calculated by record_iter_changes.
272
:param tree: The tree which is being committed.
274
# NB: if there are no parents then this method is not called, so no
275
# need to guard on parents having length.
276
entry = entry_factory['directory'](tree.path2id(''), '',
278
entry.revision = self._new_revision_id
279
self._basis_delta.append(('', '', entry.file_id, entry))
281
def _get_delta(self, ie, basis_inv, path):
282
"""Get a delta against the basis inventory for ie."""
283
if ie.file_id not in basis_inv:
285
result = (None, path, ie.file_id, ie)
286
self._basis_delta.append(result)
288
elif ie != basis_inv[ie.file_id]:
290
# TODO: avoid tis id2path call.
291
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
292
self._basis_delta.append(result)
298
def get_basis_delta(self):
299
"""Return the complete inventory delta versus the basis inventory.
301
This has been built up with the calls to record_delete and
302
record_entry_contents. The client must have already called
303
will_record_deletes() to indicate that they will be generating a
306
:return: An inventory delta, suitable for use with apply_delta, or
307
Repository.add_inventory_by_delta, etc.
309
if not self._recording_deletes:
310
raise AssertionError("recording deletes not activated.")
311
return self._basis_delta
313
def record_delete(self, path, file_id):
314
"""Record that a delete occured against a basis tree.
316
This is an optional API - when used it adds items to the basis_delta
317
being accumulated by the commit builder. It cannot be called unless the
318
method will_record_deletes() has been called to inform the builder that
319
a delta is being supplied.
321
:param path: The path of the thing deleted.
322
:param file_id: The file id that was deleted.
324
if not self._recording_deletes:
325
raise AssertionError("recording deletes not activated.")
326
delta = (path, None, file_id, None)
327
self._basis_delta.append(delta)
328
self._any_changes = True
331
def will_record_deletes(self):
332
"""Tell the commit builder that deletes are being notified.
334
This enables the accumulation of an inventory delta; for the resulting
335
commit to be valid, deletes against the basis MUST be recorded via
336
builder.record_delete().
338
self._recording_deletes = True
340
basis_id = self.parents[0]
342
basis_id = _mod_revision.NULL_REVISION
343
self.basis_delta_revision = basis_id
345
def record_entry_contents(self, ie, parent_invs, path, tree,
347
"""Record the content of ie from tree into the commit if needed.
349
Side effect: sets ie.revision when unchanged
351
:param ie: An inventory entry present in the commit.
352
:param parent_invs: The inventories of the parent revisions of the
354
:param path: The path the entry is at in the tree.
355
:param tree: The tree which contains this entry and should be used to
357
:param content_summary: Summary data from the tree about the paths
358
content - stat, length, exec, sha/link target. This is only
359
accessed when the entry has a revision of None - that is when it is
360
a candidate to commit.
361
:return: A tuple (change_delta, version_recorded, fs_hash).
362
change_delta is an inventory_delta change for this entry against
363
the basis tree of the commit, or None if no change occured against
365
version_recorded is True if a new version of the entry has been
366
recorded. For instance, committing a merge where a file was only
367
changed on the other side will return (delta, False).
368
fs_hash is either None, or the hash details for the path (currently
369
a tuple of the contents sha1 and the statvalue returned by
370
tree.get_file_with_stat()).
372
if self.new_inventory.root is None:
373
if ie.parent_id is not None:
374
raise errors.RootMissing()
375
self._check_root(ie, parent_invs, tree)
376
if ie.revision is None:
377
kind = content_summary[0]
379
# ie is carried over from a prior commit
381
# XXX: repository specific check for nested tree support goes here - if
382
# the repo doesn't want nested trees we skip it ?
383
if (kind == 'tree-reference' and
384
not self.repository._format.supports_tree_reference):
385
# mismatch between commit builder logic and repository:
386
# this needs the entry creation pushed down into the builder.
387
raise NotImplementedError('Missing repository subtree support.')
388
self.new_inventory.add(ie)
390
# TODO: slow, take it out of the inner loop.
392
basis_inv = parent_invs[0]
394
basis_inv = Inventory(root_id=None)
396
# ie.revision is always None if the InventoryEntry is considered
397
# for committing. We may record the previous parents revision if the
398
# content is actually unchanged against a sole head.
399
if ie.revision is not None:
400
if not self._versioned_root and path == '':
401
# repositories that do not version the root set the root's
402
# revision to the new commit even when no change occurs (more
403
# specifically, they do not record a revision on the root; and
404
# the rev id is assigned to the root during deserialisation -
405
# this masks when a change may have occurred against the basis.
406
# To match this we always issue a delta, because the revision
407
# of the root will always be changing.
408
if ie.file_id in basis_inv:
409
delta = (basis_inv.id2path(ie.file_id), path,
413
delta = (None, path, ie.file_id, ie)
414
self._basis_delta.append(delta)
415
return delta, False, None
417
# we don't need to commit this, because the caller already
418
# determined that an existing revision of this file is
419
# appropriate. If its not being considered for committing then
420
# it and all its parents to the root must be unaltered so
421
# no-change against the basis.
422
if ie.revision == self._new_revision_id:
423
raise AssertionError("Impossible situation, a skipped "
424
"inventory entry (%r) claims to be modified in this "
425
"commit (%r).", (ie, self._new_revision_id))
426
return None, False, None
427
# XXX: Friction: parent_candidates should return a list not a dict
428
# so that we don't have to walk the inventories again.
429
parent_candiate_entries = ie.parent_candidates(parent_invs)
430
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
432
for inv in parent_invs:
433
if ie.file_id in inv:
434
old_rev = inv[ie.file_id].revision
435
if old_rev in head_set:
436
heads.append(inv[ie.file_id].revision)
437
head_set.remove(inv[ie.file_id].revision)
440
# now we check to see if we need to write a new record to the
442
# We write a new entry unless there is one head to the ancestors, and
443
# the kind-derived content is unchanged.
445
# Cheapest check first: no ancestors, or more the one head in the
446
# ancestors, we write a new node.
450
# There is a single head, look it up for comparison
451
parent_entry = parent_candiate_entries[heads[0]]
452
# if the non-content specific data has changed, we'll be writing a
454
if (parent_entry.parent_id != ie.parent_id or
455
parent_entry.name != ie.name):
457
# now we need to do content specific checks:
459
# if the kind changed the content obviously has
460
if kind != parent_entry.kind:
462
# Stat cache fingerprint feedback for the caller - None as we usually
463
# don't generate one.
466
if content_summary[2] is None:
467
raise ValueError("Files must not have executable = None")
469
# We can't trust a check of the file length because of content
471
if (# if the exec bit has changed we have to store:
472
parent_entry.executable != content_summary[2]):
474
elif parent_entry.text_sha1 == content_summary[3]:
475
# all meta and content is unchanged (using a hash cache
476
# hit to check the sha)
477
ie.revision = parent_entry.revision
478
ie.text_size = parent_entry.text_size
479
ie.text_sha1 = parent_entry.text_sha1
480
ie.executable = parent_entry.executable
481
return self._get_delta(ie, basis_inv, path), False, None
483
# Either there is only a hash change(no hash cache entry,
484
# or same size content change), or there is no change on
486
# Provide the parent's hash to the store layer, so that the
487
# content is unchanged we will not store a new node.
488
nostore_sha = parent_entry.text_sha1
490
# We want to record a new node regardless of the presence or
491
# absence of a content change in the file.
493
ie.executable = content_summary[2]
494
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
496
text = file_obj.read()
500
ie.text_sha1, ie.text_size = self._add_text_to_weave(
501
ie.file_id, text, heads, nostore_sha)
502
# Let the caller know we generated a stat fingerprint.
503
fingerprint = (ie.text_sha1, stat_value)
504
except errors.ExistingContent:
505
# Turns out that the file content was unchanged, and we were
506
# only going to store a new node if it was changed. Carry over
508
ie.revision = parent_entry.revision
509
ie.text_size = parent_entry.text_size
510
ie.text_sha1 = parent_entry.text_sha1
511
ie.executable = parent_entry.executable
512
return self._get_delta(ie, basis_inv, path), False, None
513
elif kind == 'directory':
515
# all data is meta here, nothing specific to directory, so
517
ie.revision = parent_entry.revision
518
return self._get_delta(ie, basis_inv, path), False, None
519
self._add_text_to_weave(ie.file_id, '', heads, None)
520
elif kind == 'symlink':
521
current_link_target = content_summary[3]
523
# symlink target is not generic metadata, check if it has
525
if current_link_target != parent_entry.symlink_target:
528
# unchanged, carry over.
529
ie.revision = parent_entry.revision
530
ie.symlink_target = parent_entry.symlink_target
531
return self._get_delta(ie, basis_inv, path), False, None
532
ie.symlink_target = current_link_target
533
self._add_text_to_weave(ie.file_id, '', heads, None)
534
elif kind == 'tree-reference':
536
if content_summary[3] != parent_entry.reference_revision:
539
# unchanged, carry over.
540
ie.reference_revision = parent_entry.reference_revision
541
ie.revision = parent_entry.revision
542
return self._get_delta(ie, basis_inv, path), False, None
543
ie.reference_revision = content_summary[3]
544
if ie.reference_revision is None:
545
raise AssertionError("invalid content_summary for nested tree: %r"
546
% (content_summary,))
547
self._add_text_to_weave(ie.file_id, '', heads, None)
549
raise NotImplementedError('unknown kind')
550
ie.revision = self._new_revision_id
551
self._any_changes = True
552
return self._get_delta(ie, basis_inv, path), True, fingerprint
554
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
555
_entry_factory=entry_factory):
556
"""Record a new tree via iter_changes.
558
:param tree: The tree to obtain text contents from for changed objects.
559
:param basis_revision_id: The revision id of the tree the iter_changes
560
has been generated against. Currently assumed to be the same
561
as self.parents[0] - if it is not, errors may occur.
562
:param iter_changes: An iter_changes iterator with the changes to apply
563
to basis_revision_id. The iterator must not include any items with
564
a current kind of None - missing items must be either filtered out
565
or errored-on beefore record_iter_changes sees the item.
566
:param _entry_factory: Private method to bind entry_factory locally for
568
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
571
# Create an inventory delta based on deltas between all the parents and
572
# deltas between all the parent inventories. We use inventory delta's
573
# between the inventory objects because iter_changes masks
574
# last-changed-field only changes.
576
# file_id -> change map, change is fileid, paths, changed, versioneds,
577
# parents, names, kinds, executables
579
# {file_id -> revision_id -> inventory entry, for entries in parent
580
# trees that are not parents[0]
584
revtrees = list(self.repository.revision_trees(self.parents))
585
except errors.NoSuchRevision:
586
# one or more ghosts, slow path.
588
for revision_id in self.parents:
590
revtrees.append(self.repository.revision_tree(revision_id))
591
except errors.NoSuchRevision:
593
basis_revision_id = _mod_revision.NULL_REVISION
595
revtrees.append(self.repository.revision_tree(
596
_mod_revision.NULL_REVISION))
597
# The basis inventory from a repository
599
basis_inv = revtrees[0].inventory
601
basis_inv = self.repository.revision_tree(
602
_mod_revision.NULL_REVISION).inventory
603
if len(self.parents) > 0:
604
if basis_revision_id != self.parents[0] and not ghost_basis:
606
"arbitrary basis parents not yet supported with merges")
607
for revtree in revtrees[1:]:
608
for change in revtree.inventory._make_delta(basis_inv):
609
if change[1] is None:
610
# Not present in this parent.
612
if change[2] not in merged_ids:
613
if change[0] is not None:
614
basis_entry = basis_inv[change[2]]
615
merged_ids[change[2]] = [
617
basis_entry.revision,
620
parent_entries[change[2]] = {
622
basis_entry.revision:basis_entry,
624
change[3].revision:change[3],
627
merged_ids[change[2]] = [change[3].revision]
628
parent_entries[change[2]] = {change[3].revision:change[3]}
630
merged_ids[change[2]].append(change[3].revision)
631
parent_entries[change[2]][change[3].revision] = change[3]
634
# Setup the changes from the tree:
635
# changes maps file_id -> (change, [parent revision_ids])
637
for change in iter_changes:
638
# This probably looks up in basis_inv way to much.
639
if change[1][0] is not None:
640
head_candidate = [basis_inv[change[0]].revision]
643
changes[change[0]] = change, merged_ids.get(change[0],
645
unchanged_merged = set(merged_ids) - set(changes)
646
# Extend the changes dict with synthetic changes to record merges of
648
for file_id in unchanged_merged:
649
# Record a merged version of these items that did not change vs the
650
# basis. This can be either identical parallel changes, or a revert
651
# of a specific file after a merge. The recorded content will be
652
# that of the current tree (which is the same as the basis), but
653
# the per-file graph will reflect a merge.
654
# NB:XXX: We are reconstructing path information we had, this
655
# should be preserved instead.
656
# inv delta change: (file_id, (path_in_source, path_in_target),
657
# changed_content, versioned, parent, name, kind,
660
basis_entry = basis_inv[file_id]
661
except errors.NoSuchId:
662
# a change from basis->some_parents but file_id isn't in basis
663
# so was new in the merge, which means it must have changed
664
# from basis -> current, and as it hasn't the add was reverted
665
# by the user. So we discard this change.
669
(basis_inv.id2path(file_id), tree.id2path(file_id)),
671
(basis_entry.parent_id, basis_entry.parent_id),
672
(basis_entry.name, basis_entry.name),
673
(basis_entry.kind, basis_entry.kind),
674
(basis_entry.executable, basis_entry.executable))
675
changes[file_id] = (change, merged_ids[file_id])
676
# changes contains tuples with the change and a set of inventory
677
# candidates for the file.
679
# old_path, new_path, file_id, new_inventory_entry
680
seen_root = False # Is the root in the basis delta?
681
inv_delta = self._basis_delta
682
modified_rev = self._new_revision_id
683
for change, head_candidates in changes.values():
684
if change[3][1]: # versioned in target.
685
# Several things may be happening here:
686
# We may have a fork in the per-file graph
687
# - record a change with the content from tree
688
# We may have a change against < all trees
689
# - carry over the tree that hasn't changed
690
# We may have a change against all trees
691
# - record the change with the content from tree
694
entry = _entry_factory[kind](file_id, change[5][1],
696
head_set = self._heads(change[0], set(head_candidates))
699
for head_candidate in head_candidates:
700
if head_candidate in head_set:
701
heads.append(head_candidate)
702
head_set.remove(head_candidate)
705
# Could be a carry-over situation:
706
parent_entry_revs = parent_entries.get(file_id, None)
707
if parent_entry_revs:
708
parent_entry = parent_entry_revs.get(heads[0], None)
711
if parent_entry is None:
712
# The parent iter_changes was called against is the one
713
# that is the per-file head, so any change is relevant
714
# iter_changes is valid.
715
carry_over_possible = False
717
# could be a carry over situation
718
# A change against the basis may just indicate a merge,
719
# we need to check the content against the source of the
720
# merge to determine if it was changed after the merge
722
if (parent_entry.kind != entry.kind or
723
parent_entry.parent_id != entry.parent_id or
724
parent_entry.name != entry.name):
725
# Metadata common to all entries has changed
726
# against per-file parent
727
carry_over_possible = False
729
carry_over_possible = True
730
# per-type checks for changes against the parent_entry
733
# Cannot be a carry-over situation
734
carry_over_possible = False
735
# Populate the entry in the delta
737
# XXX: There is still a small race here: If someone reverts the content of a file
738
# after iter_changes examines and decides it has changed,
739
# we will unconditionally record a new version even if some
740
# other process reverts it while commit is running (with
741
# the revert happening after iter_changes did it's
744
entry.executable = True
746
entry.executable = False
747
if (carry_over_possible and
748
parent_entry.executable == entry.executable):
749
# Check the file length, content hash after reading
751
nostore_sha = parent_entry.text_sha1
754
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
756
text = file_obj.read()
760
entry.text_sha1, entry.text_size = self._add_text_to_weave(
761
file_id, text, heads, nostore_sha)
762
yield file_id, change[1][1], (entry.text_sha1, stat_value)
763
except errors.ExistingContent:
764
# No content change against a carry_over parent
765
# Perhaps this should also yield a fs hash update?
767
entry.text_size = parent_entry.text_size
768
entry.text_sha1 = parent_entry.text_sha1
769
elif kind == 'symlink':
771
entry.symlink_target = tree.get_symlink_target(file_id)
772
if (carry_over_possible and
773
parent_entry.symlink_target == entry.symlink_target):
776
self._add_text_to_weave(change[0], '', heads, None)
777
elif kind == 'directory':
778
if carry_over_possible:
781
# Nothing to set on the entry.
782
# XXX: split into the Root and nonRoot versions.
783
if change[1][1] != '' or self.repository.supports_rich_root():
784
self._add_text_to_weave(change[0], '', heads, None)
785
elif kind == 'tree-reference':
786
if not self.repository._format.supports_tree_reference:
787
# This isn't quite sane as an error, but we shouldn't
788
# ever see this code path in practice: tree's don't
789
# permit references when the repo doesn't support tree
791
raise errors.UnsupportedOperation(tree.add_reference,
793
reference_revision = tree.get_reference_revision(change[0])
794
entry.reference_revision = reference_revision
795
if (carry_over_possible and
796
parent_entry.reference_revision == reference_revision):
799
self._add_text_to_weave(change[0], '', heads, None)
801
raise AssertionError('unknown kind %r' % kind)
803
entry.revision = modified_rev
805
entry.revision = parent_entry.revision
808
new_path = change[1][1]
809
inv_delta.append((change[1][0], new_path, change[0], entry))
812
self.new_inventory = None
814
# This should perhaps be guarded by a check that the basis we
815
# commit against is the basis for the commit and if not do a delta
817
self._any_changes = True
819
# housekeeping root entry changes do not affect no-change commits.
820
self._require_root_change(tree)
821
self.basis_delta_revision = basis_revision_id
823
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
824
parent_keys = tuple([(file_id, parent) for parent in parents])
825
return self.repository.texts._add_text(
826
(file_id, self._new_revision_id), parent_keys, new_text,
827
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
830
class RootCommitBuilder(CommitBuilder):
831
"""This commitbuilder actually records the root id"""
833
# the root entry gets versioned properly by this builder.
834
_versioned_root = True
836
def _check_root(self, ie, parent_invs, tree):
837
"""Helper for record_entry_contents.
839
:param ie: An entry being added.
840
:param parent_invs: The inventories of the parent revisions of the
842
:param tree: The tree that is being committed.
845
def _require_root_change(self, tree):
846
"""Enforce an appropriate root object change.
848
This is called once when record_iter_changes is called, if and only if
849
the root was not in the delta calculated by record_iter_changes.
851
:param tree: The tree which is being committed.
853
# versioned roots do not change unless the tree found a change.
64
856
######################################################################
67
860
class Repository(object):
68
861
"""Repository holding history for one or more branches.
71
864
revisions and file history. It's normally accessed only by the Branch,
72
865
which views a particular line of development through that history.
74
The Repository builds on top of Stores and a Transport, which respectively
75
describe the disk data format and the way of accessing the (possibly
867
The Repository builds on top of some byte storage facilies (the revisions,
868
signatures, inventories, texts and chk_bytes attributes) and a Transport,
869
which respectively provide byte storage and a means to access the (possibly
872
The byte storage facilities are addressed via tuples, which we refer to
873
as 'keys' throughout the code base. Revision_keys, inventory_keys and
874
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
875
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
876
byte string made up of a hash identifier and a hash value.
877
We use this interface because it allows low friction with the underlying
878
code that implements disk indices, network encoding and other parts of
881
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
882
the serialised revisions for the repository. This can be used to obtain
883
revision graph information or to access raw serialised revisions.
884
The result of trying to insert data into the repository via this store
885
is undefined: it should be considered read-only except for implementors
887
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
888
the serialised signatures for the repository. This can be used to
889
obtain access to raw serialised signatures. The result of trying to
890
insert data into the repository via this store is undefined: it should
891
be considered read-only except for implementors of repositories.
892
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
893
the serialised inventories for the repository. This can be used to
894
obtain unserialised inventories. The result of trying to insert data
895
into the repository via this store is undefined: it should be
896
considered read-only except for implementors of repositories.
897
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
898
texts of files and directories for the repository. This can be used to
899
obtain file texts or file graphs. Note that Repository.iter_file_bytes
900
is usually a better interface for accessing file texts.
901
The result of trying to insert data into the repository via this store
902
is undefined: it should be considered read-only except for implementors
904
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
905
any data the repository chooses to store or have indexed by its hash.
906
The result of trying to insert data into the repository via this store
907
is undefined: it should be considered read-only except for implementors
909
:ivar _transport: Transport for file access to repository, typically
910
pointing to .bzr/repository.
913
# What class to use for a CommitBuilder. Often its simpler to change this
914
# in a Repository class subclass rather than to override
915
# get_commit_builder.
916
_commit_builder_class = CommitBuilder
917
# The search regex used by xml based repositories to determine what things
918
# where changed in a single commit.
79
919
_file_ids_altered_regex = lazy_regex.lazy_compile(
80
920
r'file_id="(?P<file_id>[^"]+)"'
81
r'.*revision="(?P<revision_id>[^"]+)"'
921
r'.* revision="(?P<revision_id>[^"]+)"'
924
def abort_write_group(self, suppress_errors=False):
925
"""Commit the contents accrued within the current write group.
927
:param suppress_errors: if true, abort_write_group will catch and log
928
unexpected errors that happen during the abort, rather than
929
allowing them to propagate. Defaults to False.
931
:seealso: start_write_group.
933
if self._write_group is not self.get_transaction():
934
# has an unlock or relock occured ?
937
'(suppressed) mismatched lock context and write group. %r, %r',
938
self._write_group, self.get_transaction())
940
raise errors.BzrError(
941
'mismatched lock context and write group. %r, %r' %
942
(self._write_group, self.get_transaction()))
944
self._abort_write_group()
945
except Exception, exc:
946
self._write_group = None
947
if not suppress_errors:
949
mutter('abort_write_group failed')
950
log_exception_quietly()
951
note('bzr: ERROR (ignored): %s', exc)
952
self._write_group = None
954
def _abort_write_group(self):
955
"""Template method for per-repository write group cleanup.
957
This is called during abort before the write group is considered to be
958
finished and should cleanup any internal state accrued during the write
959
group. There is no requirement that data handed to the repository be
960
*not* made available - this is not a rollback - but neither should any
961
attempt be made to ensure that data added is fully commited. Abort is
962
invoked when an error has occured so futher disk or network operations
963
may not be possible or may error and if possible should not be
967
def add_fallback_repository(self, repository):
968
"""Add a repository to use for looking up data not held locally.
970
:param repository: A repository.
972
if not self._format.supports_external_lookups:
973
raise errors.UnstackableRepositoryFormat(self._format, self.base)
975
# This repository will call fallback.unlock() when we transition to
976
# the unlocked state, so we make sure to increment the lock count
977
repository.lock_read()
978
self._check_fallback_repository(repository)
979
self._fallback_repositories.append(repository)
980
self.texts.add_fallback_versioned_files(repository.texts)
981
self.inventories.add_fallback_versioned_files(repository.inventories)
982
self.revisions.add_fallback_versioned_files(repository.revisions)
983
self.signatures.add_fallback_versioned_files(repository.signatures)
984
if self.chk_bytes is not None:
985
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
987
def _check_fallback_repository(self, repository):
988
"""Check that this repository can fallback to repository safely.
990
Raise an error if not.
992
:param repository: A repository to fallback to.
994
return InterRepository._assert_same_model(self, repository)
85
996
def add_inventory(self, revision_id, inv, parents):
86
997
"""Add the inventory inv to the repository as revision_id.
88
999
:param parents: The revision ids of the parents that revision_id
89
1000
is known to have and are in the repository already.
91
returns the sha1 of the serialized inventory.
1002
:returns: The validator(which is a sha1 digest, though what is sha'd is
1003
repository format specific) of the serialized inventory.
93
revision_id = osutils.safe_revision_id(revision_id)
1005
if not self.is_in_write_group():
1006
raise AssertionError("%r not in write group" % (self,))
94
1007
_mod_revision.check_not_reserved_id(revision_id)
95
assert inv.revision_id is None or inv.revision_id == revision_id, \
96
"Mismatch between inventory revision" \
97
" id and insertion revid (%r, %r)" % (inv.revision_id, revision_id)
98
assert inv.root is not None
99
inv_text = self.serialise_inventory(inv)
100
inv_sha1 = osutils.sha_string(inv_text)
101
inv_vf = self.control_weaves.get_weave('inventory',
102
self.get_transaction())
103
self._inventory_add_lines(inv_vf, revision_id, parents,
104
osutils.split_lines(inv_text))
107
def _inventory_add_lines(self, inv_vf, revision_id, parents, lines):
109
for parent in parents:
111
final_parents.append(parent)
113
inv_vf.add_lines(revision_id, final_parents, lines)
1008
if not (inv.revision_id is None or inv.revision_id == revision_id):
1009
raise AssertionError(
1010
"Mismatch between inventory revision"
1011
" id and insertion revid (%r, %r)"
1012
% (inv.revision_id, revision_id))
1013
if inv.root is None:
1014
raise AssertionError()
1015
return self._add_inventory_checked(revision_id, inv, parents)
1017
def _add_inventory_checked(self, revision_id, inv, parents):
1018
"""Add inv to the repository after checking the inputs.
1020
This function can be overridden to allow different inventory styles.
1022
:seealso: add_inventory, for the contract.
1024
inv_lines = self._serialise_inventory_to_lines(inv)
1025
return self._inventory_add_lines(revision_id, parents,
1026
inv_lines, check_content=False)
1028
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1029
parents, basis_inv=None, propagate_caches=False):
1030
"""Add a new inventory expressed as a delta against another revision.
1032
See the inventory developers documentation for the theory behind
1035
:param basis_revision_id: The inventory id the delta was created
1036
against. (This does not have to be a direct parent.)
1037
:param delta: The inventory delta (see Inventory.apply_delta for
1039
:param new_revision_id: The revision id that the inventory is being
1041
:param parents: The revision ids of the parents that revision_id is
1042
known to have and are in the repository already. These are supplied
1043
for repositories that depend on the inventory graph for revision
1044
graph access, as well as for those that pun ancestry with delta
1046
:param basis_inv: The basis inventory if it is already known,
1048
:param propagate_caches: If True, the caches for this inventory are
1049
copied to and updated for the result if possible.
1051
:returns: (validator, new_inv)
1052
The validator(which is a sha1 digest, though what is sha'd is
1053
repository format specific) of the serialized inventory, and the
1054
resulting inventory.
1056
if not self.is_in_write_group():
1057
raise AssertionError("%r not in write group" % (self,))
1058
_mod_revision.check_not_reserved_id(new_revision_id)
1059
basis_tree = self.revision_tree(basis_revision_id)
1060
basis_tree.lock_read()
1062
# Note that this mutates the inventory of basis_tree, which not all
1063
# inventory implementations may support: A better idiom would be to
1064
# return a new inventory, but as there is no revision tree cache in
1065
# repository this is safe for now - RBC 20081013
1066
if basis_inv is None:
1067
basis_inv = basis_tree.inventory
1068
basis_inv.apply_delta(delta)
1069
basis_inv.revision_id = new_revision_id
1070
return (self.add_inventory(new_revision_id, basis_inv, parents),
1075
def _inventory_add_lines(self, revision_id, parents, lines,
1076
check_content=True):
1077
"""Store lines in inv_vf and return the sha1 of the inventory."""
1078
parents = [(parent,) for parent in parents]
1079
result = self.inventories.add_lines((revision_id,), parents, lines,
1080
check_content=check_content)[0]
1081
self.inventories._access.flush()
116
1084
def add_revision(self, revision_id, rev, inv=None, config=None):
117
1085
"""Add rev to the revision store as revision_id.
347
1535
def copy_content_into(self, destination, revision_id=None):
348
1536
"""Make a complete copy of the content in self into destination.
350
This is a destructive operation! Do not use it on existing
1538
This is a destructive operation! Do not use it on existing
353
revision_id = osutils.safe_revision_id(revision_id)
354
1541
return InterRepository.get(self, destination).copy_content(revision_id)
356
def fetch(self, source, revision_id=None, pb=None):
1543
def commit_write_group(self):
1544
"""Commit the contents accrued within the current write group.
1546
:seealso: start_write_group.
1548
:return: it may return an opaque hint that can be passed to 'pack'.
1550
if self._write_group is not self.get_transaction():
1551
# has an unlock or relock occured ?
1552
raise errors.BzrError('mismatched lock context %r and '
1554
(self.get_transaction(), self._write_group))
1555
result = self._commit_write_group()
1556
self._write_group = None
1559
def _commit_write_group(self):
1560
"""Template method for per-repository write group cleanup.
1562
This is called before the write group is considered to be
1563
finished and should ensure that all data handed to the repository
1564
for writing during the write group is safely committed (to the
1565
extent possible considering file system caching etc).
1568
def suspend_write_group(self):
1569
raise errors.UnsuspendableWriteGroup(self)
1571
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1572
"""Return the keys of missing inventory parents for revisions added in
1575
A revision is not complete if the inventory delta for that revision
1576
cannot be calculated. Therefore if the parent inventories of a
1577
revision are not present, the revision is incomplete, and e.g. cannot
1578
be streamed by a smart server. This method finds missing inventory
1579
parents for revisions added in this write group.
1581
if not self._format.supports_external_lookups:
1582
# This is only an issue for stacked repositories
1584
if not self.is_in_write_group():
1585
raise AssertionError('not in a write group')
1587
# XXX: We assume that every added revision already has its
1588
# corresponding inventory, so we only check for parent inventories that
1589
# might be missing, rather than all inventories.
1590
parents = set(self.revisions._index.get_missing_parents())
1591
parents.discard(_mod_revision.NULL_REVISION)
1592
unstacked_inventories = self.inventories._index
1593
present_inventories = unstacked_inventories.get_parent_map(
1594
key[-1:] for key in parents)
1595
parents.difference_update(present_inventories)
1596
if len(parents) == 0:
1597
# No missing parent inventories.
1599
if not check_for_missing_texts:
1600
return set(('inventories', rev_id) for (rev_id,) in parents)
1601
# Ok, now we have a list of missing inventories. But these only matter
1602
# if the inventories that reference them are missing some texts they
1603
# appear to introduce.
1604
# XXX: Texts referenced by all added inventories need to be present,
1605
# but at the moment we're only checking for texts referenced by
1606
# inventories at the graph's edge.
1607
key_deps = self.revisions._index._key_dependencies
1608
key_deps.satisfy_refs_for_keys(present_inventories)
1609
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1610
file_ids = self.fileids_altered_by_revision_ids(referrers)
1611
missing_texts = set()
1612
for file_id, version_ids in file_ids.iteritems():
1613
missing_texts.update(
1614
(file_id, version_id) for version_id in version_ids)
1615
present_texts = self.texts.get_parent_map(missing_texts)
1616
missing_texts.difference_update(present_texts)
1617
if not missing_texts:
1618
# No texts are missing, so all revisions and their deltas are
1621
# Alternatively the text versions could be returned as the missing
1622
# keys, but this is likely to be less data.
1623
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1626
def refresh_data(self):
1627
"""Re-read any data needed to to synchronise with disk.
1629
This method is intended to be called after another repository instance
1630
(such as one used by a smart server) has inserted data into the
1631
repository. It may not be called during a write group, but may be
1632
called at any other time.
1634
if self.is_in_write_group():
1635
raise errors.InternalBzrError(
1636
"May not refresh_data while in a write group.")
1637
self._refresh_data()
1639
def resume_write_group(self, tokens):
1640
if not self.is_write_locked():
1641
raise errors.NotWriteLocked(self)
1642
if self._write_group:
1643
raise errors.BzrError('already in a write group')
1644
self._resume_write_group(tokens)
1645
# so we can detect unlock/relock - the write group is now entered.
1646
self._write_group = self.get_transaction()
1648
def _resume_write_group(self, tokens):
1649
raise errors.UnsuspendableWriteGroup(self)
1651
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
357
1653
"""Fetch the content required to construct revision_id from source.
359
If revision_id is None all content is copied.
1655
If revision_id is None and fetch_spec is None, then all content is
1658
fetch() may not be used when the repository is in a write group -
1659
either finish the current write group before using fetch, or use
1660
fetch before starting the write group.
1662
:param find_ghosts: Find and copy revisions in the source that are
1663
ghosts in the target (and not reachable directly by walking out to
1664
the first-present revision in target from revision_id).
1665
:param revision_id: If specified, all the content needed for this
1666
revision ID will be copied to the target. Fetch will determine for
1667
itself which content needs to be copied.
1668
:param fetch_spec: If specified, a SearchResult or
1669
PendingAncestryResult that describes which revisions to copy. This
1670
allows copying multiple heads at once. Mutually exclusive with
361
revision_id = osutils.safe_revision_id(revision_id)
1673
if fetch_spec is not None and revision_id is not None:
1674
raise AssertionError(
1675
"fetch_spec and revision_id are mutually exclusive.")
1676
if self.is_in_write_group():
1677
raise errors.InternalBzrError(
1678
"May not fetch while in a write group.")
1679
# fast path same-url fetch operations
1680
# TODO: lift out to somewhere common with RemoteRepository
1681
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1682
if (self.has_same_location(source)
1683
and fetch_spec is None
1684
and self._has_same_fallbacks(source)):
1685
# check that last_revision is in 'from' and then return a
1687
if (revision_id is not None and
1688
not _mod_revision.is_null(revision_id)):
1689
self.get_revision(revision_id)
1691
# if there is no specific appropriate InterRepository, this will get
1692
# the InterRepository base class, which raises an
1693
# IncompatibleRepositories when asked to fetch.
362
1694
inter = InterRepository.get(source, self)
364
return inter.fetch(revision_id=revision_id, pb=pb)
365
except NotImplementedError:
366
raise errors.IncompatibleRepositories(source, self)
368
def get_commit_builder(self, branch, parents, config, timestamp=None,
369
timezone=None, committer=None, revprops=None,
1695
return inter.fetch(revision_id=revision_id, pb=pb,
1696
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1698
def create_bundle(self, target, base, fileobj, format=None):
1699
return serializer.write_bundle(self, target, base, fileobj, format)
1701
def get_commit_builder(self, branch, parents, config, timestamp=None,
1702
timezone=None, committer=None, revprops=None,
370
1703
revision_id=None):
371
1704
"""Obtain a CommitBuilder for this repository.
373
1706
:param branch: Branch to commit to.
374
1707
:param parents: Revision ids of the parents of the new revision.
375
1708
:param config: Configuration to use.
424
1802
dest_repo = a_bzrdir.open_repository()
425
1803
return dest_repo
1805
def _get_sink(self):
1806
"""Return a sink for streaming into this repository."""
1807
return StreamSink(self)
1809
def _get_source(self, to_format):
1810
"""Return a source for streaming from this repository."""
1811
return StreamSource(self, to_format)
427
1813
@needs_read_lock
428
1814
def has_revision(self, revision_id):
429
1815
"""True if this repository has a copy of the revision."""
430
revision_id = osutils.safe_revision_id(revision_id)
431
return self._revision_store.has_revision_id(revision_id,
432
self.get_transaction())
1816
return revision_id in self.has_revisions((revision_id,))
1819
def has_revisions(self, revision_ids):
1820
"""Probe to find out the presence of multiple revisions.
1822
:param revision_ids: An iterable of revision_ids.
1823
:return: A set of the revision_ids that were present.
1825
parent_map = self.revisions.get_parent_map(
1826
[(rev_id,) for rev_id in revision_ids])
1828
if _mod_revision.NULL_REVISION in revision_ids:
1829
result.add(_mod_revision.NULL_REVISION)
1830
result.update([key[0] for key in parent_map])
1834
def get_revision(self, revision_id):
1835
"""Return the Revision object for a named revision."""
1836
return self.get_revisions([revision_id])[0]
434
1838
@needs_read_lock
435
1839
def get_revision_reconcile(self, revision_id):
436
1840
"""'reconcile' helper routine that allows access to a revision always.
438
1842
This variant of get_revision does not cross check the weave graph
439
1843
against the revision one as get_revision does: but it should only
440
1844
be used by reconcile, or reconcile-alike commands that are correcting
441
1845
or testing the revision graph.
443
if not revision_id or not isinstance(revision_id, basestring):
444
raise errors.InvalidRevisionId(revision_id=revision_id,
446
return self.get_revisions([revision_id])[0]
1847
return self._get_revisions([revision_id])[0]
448
1849
@needs_read_lock
449
1850
def get_revisions(self, revision_ids):
450
revision_ids = [osutils.safe_revision_id(r) for r in revision_ids]
451
revs = self._revision_store.get_revisions(revision_ids,
452
self.get_transaction())
454
assert not isinstance(rev.revision_id, unicode)
455
for parent_id in rev.parent_ids:
456
assert not isinstance(parent_id, unicode)
1851
"""Get many revisions at once.
1853
Repositories that need to check data on every revision read should
1854
subclass this method.
1856
return self._get_revisions(revision_ids)
1859
def _get_revisions(self, revision_ids):
1860
"""Core work logic to get many revisions without sanity checks."""
1862
for revid, rev in self._iter_revisions(revision_ids):
1864
raise errors.NoSuchRevision(self, revid)
1866
return [revs[revid] for revid in revision_ids]
1868
def _iter_revisions(self, revision_ids):
1869
"""Iterate over revision objects.
1871
:param revision_ids: An iterable of revisions to examine. None may be
1872
passed to request all revisions known to the repository. Note that
1873
not all repositories can find unreferenced revisions; for those
1874
repositories only referenced ones will be returned.
1875
:return: An iterator of (revid, revision) tuples. Absent revisions (
1876
those asked for but not available) are returned as (revid, None).
1878
if revision_ids is None:
1879
revision_ids = self.all_revision_ids()
1881
for rev_id in revision_ids:
1882
if not rev_id or not isinstance(rev_id, basestring):
1883
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1884
keys = [(key,) for key in revision_ids]
1885
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1886
for record in stream:
1887
revid = record.key[0]
1888
if record.storage_kind == 'absent':
1891
text = record.get_bytes_as('fulltext')
1892
rev = self._serializer.read_revision_from_string(text)
459
1895
@needs_read_lock
460
1896
def get_revision_xml(self, revision_id):
461
1897
# TODO: jam 20070210 This shouldn't be necessary since get_revision
462
1898
# would have already do it.
463
1899
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
464
revision_id = osutils.safe_revision_id(revision_id)
1900
# TODO: this can't just be replaced by:
1901
# return self._serializer.write_revision_to_string(
1902
# self.get_revision(revision_id))
1903
# as cStringIO preservers the encoding unlike write_revision_to_string
1904
# or some other call down the path.
465
1905
rev = self.get_revision(revision_id)
1906
rev_tmp = cStringIO.StringIO()
467
1907
# the current serializer..
468
self._revision_store._serializer.write_revision(rev, rev_tmp)
1908
self._serializer.write_revision(rev, rev_tmp)
470
1910
return rev_tmp.getvalue()
473
def get_revision(self, revision_id):
474
"""Return the Revision object for a named revision"""
475
# TODO: jam 20070210 get_revision_reconcile should do this for us
476
revision_id = osutils.safe_revision_id(revision_id)
477
r = self.get_revision_reconcile(revision_id)
478
# weave corruption can lead to absent revision markers that should be
480
# the following test is reasonably cheap (it needs a single weave read)
481
# and the weave is cached in read transactions. In write transactions
482
# it is not cached but typically we only read a small number of
483
# revisions. For knits when they are introduced we will probably want
484
# to ensure that caching write transactions are in use.
485
inv = self.get_inventory_weave()
486
self._check_revision_parents(r, inv)
490
def get_deltas_for_revisions(self, revisions):
1912
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
491
1913
"""Produce a generator of revision deltas.
493
1915
Note that the input is a sequence of REVISIONS, not revision_ids.
494
1916
Trees will be held in memory until the generator exits.
495
1917
Each delta is relative to the revision's lefthand predecessor.
1919
:param specific_fileids: if not None, the result is filtered
1920
so that only those file-ids, their parents and their
1921
children are included.
1923
# Get the revision-ids of interest
497
1924
required_trees = set()
498
1925
for revision in revisions:
499
1926
required_trees.add(revision.revision_id)
500
1927
required_trees.update(revision.parent_ids[:1])
501
trees = dict((t.get_revision_id(), t) for
502
t in self.revision_trees(required_trees))
1929
# Get the matching filtered trees. Note that it's more
1930
# efficient to pass filtered trees to changes_from() rather
1931
# than doing the filtering afterwards. changes_from() could
1932
# arguably do the filtering itself but it's path-based, not
1933
# file-id based, so filtering before or afterwards is
1935
if specific_fileids is None:
1936
trees = dict((t.get_revision_id(), t) for
1937
t in self.revision_trees(required_trees))
1939
trees = dict((t.get_revision_id(), t) for
1940
t in self._filtered_revision_trees(required_trees,
1943
# Calculate the deltas
503
1944
for revision in revisions:
504
1945
if not revision.parent_ids:
505
old_tree = self.revision_tree(None)
1946
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
507
1948
old_tree = trees[revision.parent_ids[0]]
508
1949
yield trees[revision.revision_id].changes_from(old_tree)
510
1951
@needs_read_lock
511
def get_revision_delta(self, revision_id):
1952
def get_revision_delta(self, revision_id, specific_fileids=None):
512
1953
"""Return the delta for one revision.
514
1955
The delta is relative to the left-hand predecessor of the
1958
:param specific_fileids: if not None, the result is filtered
1959
so that only those file-ids, their parents and their
1960
children are included.
517
1962
r = self.get_revision(revision_id)
518
return list(self.get_deltas_for_revisions([r]))[0]
520
def _check_revision_parents(self, revision, inventory):
521
"""Private to Repository and Fetch.
523
This checks the parentage of revision in an inventory weave for
524
consistency and is only applicable to inventory-weave-for-ancestry
525
using repository formats & fetchers.
527
weave_parents = inventory.get_parents(revision.revision_id)
528
weave_names = inventory.versions()
529
for parent_id in revision.parent_ids:
530
if parent_id in weave_names:
531
# this parent must not be a ghost.
532
if not parent_id in weave_parents:
534
raise errors.CorruptRepository(self)
1963
return list(self.get_deltas_for_revisions([r],
1964
specific_fileids=specific_fileids))[0]
536
1966
@needs_write_lock
537
1967
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
538
revision_id = osutils.safe_revision_id(revision_id)
539
1968
signature = gpg_strategy.sign(plaintext)
540
self._revision_store.add_revision_signature_text(revision_id,
542
self.get_transaction())
544
def fileids_altered_by_revision_ids(self, revision_ids):
545
"""Find the file ids and versions affected by revisions.
547
:param revisions: an iterable containing revision ids.
548
:return: a dictionary mapping altered file-ids to an iterable of
549
revision_ids. Each altered file-ids has the exact revision_ids that
550
altered it listed explicitly.
552
assert self._serializer.support_altered_by_hack, \
553
("fileids_altered_by_revision_ids only supported for branches "
554
"which store inventory as unnested xml, not on %r" % self)
555
selected_revision_ids = set(osutils.safe_revision_id(r)
556
for r in revision_ids)
557
w = self.get_inventory_weave()
1969
self.add_signature_text(revision_id, signature)
1972
def add_signature_text(self, revision_id, signature):
1973
self.signatures.add_lines((revision_id,), (),
1974
osutils.split_lines(signature))
1976
def find_text_key_references(self):
1977
"""Find the text key references within the repository.
1979
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1980
to whether they were referred to by the inventory of the
1981
revision_id that they contain. The inventory texts from all present
1982
revision ids are assessed to generate this report.
1984
revision_keys = self.revisions.keys()
1985
w = self.inventories
1986
pb = ui.ui_factory.nested_progress_bar()
1988
return self._find_text_key_references_from_xml_inventory_lines(
1989
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1993
def _find_text_key_references_from_xml_inventory_lines(self,
1995
"""Core routine for extracting references to texts from inventories.
1997
This performs the translation of xml lines to revision ids.
1999
:param line_iterator: An iterator of lines, origin_version_id
2000
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2001
to whether they were referred to by the inventory of the
2002
revision_id that they contain. Note that if that revision_id was
2003
not part of the line_iterator's output then False will be given -
2004
even though it may actually refer to that key.
2006
if not self._serializer.support_altered_by_hack:
2007
raise AssertionError(
2008
"_find_text_key_references_from_xml_inventory_lines only "
2009
"supported for branches which store inventory as unnested xml"
2010
", not on %r" % self)
560
2013
# this code needs to read every new line in every inventory for the
561
2014
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
562
# not present in one of those inventories is unnecessary but not
2015
# not present in one of those inventories is unnecessary but not
563
2016
# harmful because we are filtering by the revision id marker in the
564
# inventory lines : we only select file ids altered in one of those
2017
# inventory lines : we only select file ids altered in one of those
565
2018
# revisions. We don't need to see all lines in the inventory because
566
2019
# only those added in an inventory in rev X can contain a revision=X
576
2029
search = self._file_ids_altered_regex.search
577
2030
unescape = _unescape_xml
578
2031
setdefault = result.setdefault
579
pb = ui.ui_factory.nested_progress_bar()
581
for line in w.iter_lines_added_or_present_in_versions(
582
selected_revision_ids, pb=pb):
586
# One call to match.group() returning multiple items is quite a
587
# bit faster than 2 calls to match.group() each returning 1
588
file_id, revision_id = match.group('file_id', 'revision_id')
590
# Inlining the cache lookups helps a lot when you make 170,000
591
# lines and 350k ids, versus 8.4 unique ids.
592
# Using a cache helps in 2 ways:
593
# 1) Avoids unnecessary decoding calls
594
# 2) Re-uses cached strings, which helps in future set and
596
# (2) is enough that removing encoding entirely along with
597
# the cache (so we are using plain strings) results in no
598
# performance improvement.
600
revision_id = unescape_revid_cache[revision_id]
602
unescaped = unescape(revision_id)
603
unescape_revid_cache[revision_id] = unescaped
604
revision_id = unescaped
606
if revision_id in selected_revision_ids:
608
file_id = unescape_fileid_cache[file_id]
610
unescaped = unescape(file_id)
611
unescape_fileid_cache[file_id] = unescaped
613
setdefault(file_id, set()).add(revision_id)
619
def get_inventory_weave(self):
620
return self.control_weaves.get_weave('inventory',
621
self.get_transaction())
2032
for line, line_key in line_iterator:
2033
match = search(line)
2036
# One call to match.group() returning multiple items is quite a
2037
# bit faster than 2 calls to match.group() each returning 1
2038
file_id, revision_id = match.group('file_id', 'revision_id')
2040
# Inlining the cache lookups helps a lot when you make 170,000
2041
# lines and 350k ids, versus 8.4 unique ids.
2042
# Using a cache helps in 2 ways:
2043
# 1) Avoids unnecessary decoding calls
2044
# 2) Re-uses cached strings, which helps in future set and
2046
# (2) is enough that removing encoding entirely along with
2047
# the cache (so we are using plain strings) results in no
2048
# performance improvement.
2050
revision_id = unescape_revid_cache[revision_id]
2052
unescaped = unescape(revision_id)
2053
unescape_revid_cache[revision_id] = unescaped
2054
revision_id = unescaped
2056
# Note that unconditionally unescaping means that we deserialise
2057
# every fileid, which for general 'pull' is not great, but we don't
2058
# really want to have some many fulltexts that this matters anyway.
2061
file_id = unescape_fileid_cache[file_id]
2063
unescaped = unescape(file_id)
2064
unescape_fileid_cache[file_id] = unescaped
2067
key = (file_id, revision_id)
2068
setdefault(key, False)
2069
if revision_id == line_key[-1]:
2073
def _inventory_xml_lines_for_keys(self, keys):
2074
"""Get a line iterator of the sort needed for findind references.
2076
Not relevant for non-xml inventory repositories.
2078
Ghosts in revision_keys are ignored.
2080
:param revision_keys: The revision keys for the inventories to inspect.
2081
:return: An iterator over (inventory line, revid) for the fulltexts of
2082
all of the xml inventories specified by revision_keys.
2084
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2085
for record in stream:
2086
if record.storage_kind != 'absent':
2087
chunks = record.get_bytes_as('chunked')
2088
revid = record.key[-1]
2089
lines = osutils.chunks_to_lines(chunks)
2093
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2095
"""Helper routine for fileids_altered_by_revision_ids.
2097
This performs the translation of xml lines to revision ids.
2099
:param line_iterator: An iterator of lines, origin_version_id
2100
:param revision_keys: The revision ids to filter for. This should be a
2101
set or other type which supports efficient __contains__ lookups, as
2102
the revision key from each parsed line will be looked up in the
2103
revision_keys filter.
2104
:return: a dictionary mapping altered file-ids to an iterable of
2105
revision_ids. Each altered file-ids has the exact revision_ids that
2106
altered it listed explicitly.
2108
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2109
line_iterator).iterkeys())
2110
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2111
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2112
self._inventory_xml_lines_for_keys(parent_keys)))
2113
new_keys = seen - parent_seen
2115
setdefault = result.setdefault
2116
for key in new_keys:
2117
setdefault(key[0], set()).add(key[-1])
2120
def _find_parent_ids_of_revisions(self, revision_ids):
2121
"""Find all parent ids that are mentioned in the revision graph.
2123
:return: set of revisions that are parents of revision_ids which are
2124
not part of revision_ids themselves
2126
parent_map = self.get_parent_map(revision_ids)
2128
map(parent_ids.update, parent_map.itervalues())
2129
parent_ids.difference_update(revision_ids)
2130
parent_ids.discard(_mod_revision.NULL_REVISION)
2133
def _find_parent_keys_of_revisions(self, revision_keys):
2134
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2136
:param revision_keys: An iterable of revision_keys.
2137
:return: The parents of all revision_keys that are not already in
2140
parent_map = self.revisions.get_parent_map(revision_keys)
2142
map(parent_keys.update, parent_map.itervalues())
2143
parent_keys.difference_update(revision_keys)
2144
parent_keys.discard(_mod_revision.NULL_REVISION)
2147
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2148
"""Find the file ids and versions affected by revisions.
2150
:param revisions: an iterable containing revision ids.
2151
:param _inv_weave: The inventory weave from this repository or None.
2152
If None, the inventory weave will be opened automatically.
2153
:return: a dictionary mapping altered file-ids to an iterable of
2154
revision_ids. Each altered file-ids has the exact revision_ids that
2155
altered it listed explicitly.
2157
selected_keys = set((revid,) for revid in revision_ids)
2158
w = _inv_weave or self.inventories
2159
pb = ui.ui_factory.nested_progress_bar()
2161
return self._find_file_ids_from_xml_inventory_lines(
2162
w.iter_lines_added_or_present_in_keys(
2163
selected_keys, pb=pb),
2168
def iter_files_bytes(self, desired_files):
2169
"""Iterate through file versions.
2171
Files will not necessarily be returned in the order they occur in
2172
desired_files. No specific order is guaranteed.
2174
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2175
value supplied by the caller as part of desired_files. It should
2176
uniquely identify the file version in the caller's context. (Examples:
2177
an index number or a TreeTransform trans_id.)
2179
bytes_iterator is an iterable of bytestrings for the file. The
2180
kind of iterable and length of the bytestrings are unspecified, but for
2181
this implementation, it is a list of bytes produced by
2182
VersionedFile.get_record_stream().
2184
:param desired_files: a list of (file_id, revision_id, identifier)
2188
for file_id, revision_id, callable_data in desired_files:
2189
text_keys[(file_id, revision_id)] = callable_data
2190
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2191
if record.storage_kind == 'absent':
2192
raise errors.RevisionNotPresent(record.key, self)
2193
yield text_keys[record.key], record.get_bytes_as('chunked')
2195
def _generate_text_key_index(self, text_key_references=None,
2197
"""Generate a new text key index for the repository.
2199
This is an expensive function that will take considerable time to run.
2201
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2202
list of parents, also text keys. When a given key has no parents,
2203
the parents list will be [NULL_REVISION].
2205
# All revisions, to find inventory parents.
2206
if ancestors is None:
2207
graph = self.get_graph()
2208
ancestors = graph.get_parent_map(self.all_revision_ids())
2209
if text_key_references is None:
2210
text_key_references = self.find_text_key_references()
2211
pb = ui.ui_factory.nested_progress_bar()
2213
return self._do_generate_text_key_index(ancestors,
2214
text_key_references, pb)
2218
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2219
"""Helper for _generate_text_key_index to avoid deep nesting."""
2220
revision_order = tsort.topo_sort(ancestors)
2221
invalid_keys = set()
2223
for revision_id in revision_order:
2224
revision_keys[revision_id] = set()
2225
text_count = len(text_key_references)
2226
# a cache of the text keys to allow reuse; costs a dict of all the
2227
# keys, but saves a 2-tuple for every child of a given key.
2229
for text_key, valid in text_key_references.iteritems():
2231
invalid_keys.add(text_key)
2233
revision_keys[text_key[1]].add(text_key)
2234
text_key_cache[text_key] = text_key
2235
del text_key_references
2237
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2238
NULL_REVISION = _mod_revision.NULL_REVISION
2239
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2240
# too small for large or very branchy trees. However, for 55K path
2241
# trees, it would be easy to use too much memory trivially. Ideally we
2242
# could gauge this by looking at available real memory etc, but this is
2243
# always a tricky proposition.
2244
inventory_cache = lru_cache.LRUCache(10)
2245
batch_size = 10 # should be ~150MB on a 55K path tree
2246
batch_count = len(revision_order) / batch_size + 1
2248
pb.update("Calculating text parents", processed_texts, text_count)
2249
for offset in xrange(batch_count):
2250
to_query = revision_order[offset * batch_size:(offset + 1) *
2254
for revision_id in to_query:
2255
parent_ids = ancestors[revision_id]
2256
for text_key in revision_keys[revision_id]:
2257
pb.update("Calculating text parents", processed_texts)
2258
processed_texts += 1
2259
candidate_parents = []
2260
for parent_id in parent_ids:
2261
parent_text_key = (text_key[0], parent_id)
2263
check_parent = parent_text_key not in \
2264
revision_keys[parent_id]
2266
# the parent parent_id is a ghost:
2267
check_parent = False
2268
# truncate the derived graph against this ghost.
2269
parent_text_key = None
2271
# look at the parent commit details inventories to
2272
# determine possible candidates in the per file graph.
2275
inv = inventory_cache[parent_id]
2277
inv = self.revision_tree(parent_id).inventory
2278
inventory_cache[parent_id] = inv
2280
parent_entry = inv[text_key[0]]
2281
except (KeyError, errors.NoSuchId):
2283
if parent_entry is not None:
2285
text_key[0], parent_entry.revision)
2287
parent_text_key = None
2288
if parent_text_key is not None:
2289
candidate_parents.append(
2290
text_key_cache[parent_text_key])
2291
parent_heads = text_graph.heads(candidate_parents)
2292
new_parents = list(parent_heads)
2293
new_parents.sort(key=lambda x:candidate_parents.index(x))
2294
if new_parents == []:
2295
new_parents = [NULL_REVISION]
2296
text_index[text_key] = new_parents
2298
for text_key in invalid_keys:
2299
text_index[text_key] = [NULL_REVISION]
2302
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2303
"""Get an iterable listing the keys of all the data introduced by a set
2306
The keys will be ordered so that the corresponding items can be safely
2307
fetched and inserted in that order.
2309
:returns: An iterable producing tuples of (knit-kind, file-id,
2310
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2311
'revisions'. file-id is None unless knit-kind is 'file'.
2313
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2316
for result in self._find_non_file_keys_to_fetch(revision_ids):
2319
def _find_file_keys_to_fetch(self, revision_ids, pb):
2320
# XXX: it's a bit weird to control the inventory weave caching in this
2321
# generator. Ideally the caching would be done in fetch.py I think. Or
2322
# maybe this generator should explicitly have the contract that it
2323
# should not be iterated until the previously yielded item has been
2325
inv_w = self.inventories
2327
# file ids that changed
2328
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2330
num_file_ids = len(file_ids)
2331
for file_id, altered_versions in file_ids.iteritems():
2333
pb.update("fetch texts", count, num_file_ids)
2335
yield ("file", file_id, altered_versions)
2337
def _find_non_file_keys_to_fetch(self, revision_ids):
2339
yield ("inventory", None, revision_ids)
2342
# XXX: Note ATM no callers actually pay attention to this return
2343
# instead they just use the list of revision ids and ignore
2344
# missing sigs. Consider removing this work entirely
2345
revisions_with_signatures = set(self.signatures.get_parent_map(
2346
[(r,) for r in revision_ids]))
2347
revisions_with_signatures = set(
2348
[r for (r,) in revisions_with_signatures])
2349
revisions_with_signatures.intersection_update(revision_ids)
2350
yield ("signatures", None, revisions_with_signatures)
2353
yield ("revisions", None, revision_ids)
623
2355
@needs_read_lock
624
2356
def get_inventory(self, revision_id):
625
"""Get Inventory object by hash."""
626
# TODO: jam 20070210 Technically we don't need to sanitize, since all
627
# called functions must sanitize.
628
revision_id = osutils.safe_revision_id(revision_id)
629
return self.deserialise_inventory(
630
revision_id, self.get_inventory_xml(revision_id))
2357
"""Get Inventory object by revision id."""
2358
return self.iter_inventories([revision_id]).next()
2360
def iter_inventories(self, revision_ids, ordering=None):
2361
"""Get many inventories by revision_ids.
2363
This will buffer some or all of the texts used in constructing the
2364
inventories in memory, but will only parse a single inventory at a
2367
:param revision_ids: The expected revision ids of the inventories.
2368
:param ordering: optional ordering, e.g. 'topological'. If not
2369
specified, the order of revision_ids will be preserved (by
2370
buffering if necessary).
2371
:return: An iterator of inventories.
2373
if ((None in revision_ids)
2374
or (_mod_revision.NULL_REVISION in revision_ids)):
2375
raise ValueError('cannot get null revision inventory')
2376
return self._iter_inventories(revision_ids, ordering)
2378
def _iter_inventories(self, revision_ids, ordering):
2379
"""single-document based inventory iteration."""
2380
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2381
for text, revision_id in inv_xmls:
2382
yield self.deserialise_inventory(revision_id, text)
2384
def _iter_inventory_xmls(self, revision_ids, ordering):
2385
if ordering is None:
2386
order_as_requested = True
2387
ordering = 'unordered'
2389
order_as_requested = False
2390
keys = [(revision_id,) for revision_id in revision_ids]
2393
if order_as_requested:
2394
key_iter = iter(keys)
2395
next_key = key_iter.next()
2396
stream = self.inventories.get_record_stream(keys, ordering, True)
2398
for record in stream:
2399
if record.storage_kind != 'absent':
2400
chunks = record.get_bytes_as('chunked')
2401
if order_as_requested:
2402
text_chunks[record.key] = chunks
2404
yield ''.join(chunks), record.key[-1]
2406
raise errors.NoSuchRevision(self, record.key)
2407
if order_as_requested:
2408
# Yield as many results as we can while preserving order.
2409
while next_key in text_chunks:
2410
chunks = text_chunks.pop(next_key)
2411
yield ''.join(chunks), next_key[-1]
2413
next_key = key_iter.next()
2414
except StopIteration:
2415
# We still want to fully consume the get_record_stream,
2416
# just in case it is not actually finished at this point
632
2420
def deserialise_inventory(self, revision_id, xml):
633
"""Transform the xml into an inventory object.
2421
"""Transform the xml into an inventory object.
635
2423
:param revision_id: The expected revision id of the inventory.
636
2424
:param xml: A serialised inventory.
638
revision_id = osutils.safe_revision_id(revision_id)
639
result = self._serializer.read_inventory_from_string(xml)
640
result.root.revision = revision_id
2426
result = self._serializer.read_inventory_from_string(xml, revision_id,
2427
entry_cache=self._inventory_entry_cache)
2428
if result.revision_id != revision_id:
2429
raise AssertionError('revision id mismatch %s != %s' % (
2430
result.revision_id, revision_id))
643
2433
def serialise_inventory(self, inv):
644
2434
return self._serializer.write_inventory_to_string(inv)
2436
def _serialise_inventory_to_lines(self, inv):
2437
return self._serializer.write_inventory_to_lines(inv)
2439
def get_serializer_format(self):
2440
return self._serializer.format_num
646
2442
@needs_read_lock
647
2443
def get_inventory_xml(self, revision_id):
648
2444
"""Get inventory XML as a file object."""
649
revision_id = osutils.safe_revision_id(revision_id)
2445
texts = self._iter_inventory_xmls([revision_id], 'unordered')
651
assert isinstance(revision_id, str), type(revision_id)
652
iw = self.get_inventory_weave()
653
return iw.get_text(revision_id)
2447
text, revision_id = texts.next()
2448
except StopIteration:
655
2449
raise errors.HistoryMissing(self, 'inventory', revision_id)
657
2452
@needs_read_lock
658
2453
def get_inventory_sha1(self, revision_id):
659
2454
"""Return the sha1 hash of the inventory entry
661
# TODO: jam 20070210 Shouldn't this be deprecated / removed?
662
revision_id = osutils.safe_revision_id(revision_id)
663
2456
return self.get_revision(revision_id).inventory_sha1
666
def get_revision_graph(self, revision_id=None):
667
"""Return a dictionary containing the revision graph.
669
:param revision_id: The revision_id to get a graph from. If None, then
670
the entire revision graph is returned. This is a deprecated mode of
671
operation and will be removed in the future.
672
:return: a dictionary of revision_id->revision_parents_list.
674
# special case NULL_REVISION
675
if revision_id == _mod_revision.NULL_REVISION:
677
revision_id = osutils.safe_revision_id(revision_id)
678
a_weave = self.get_inventory_weave()
679
all_revisions = self._eliminate_revisions_not_present(
681
entire_graph = dict([(node, a_weave.get_parents(node)) for
682
node in all_revisions])
683
if revision_id is None:
685
elif revision_id not in entire_graph:
686
raise errors.NoSuchRevision(self, revision_id)
688
# add what can be reached from revision_id
690
pending = set([revision_id])
691
while len(pending) > 0:
693
result[node] = entire_graph[node]
694
for revision_id in result[node]:
695
if revision_id not in result:
696
pending.add(revision_id)
700
def get_revision_graph_with_ghosts(self, revision_ids=None):
701
"""Return a graph of the revisions with ghosts marked as applicable.
703
:param revision_ids: an iterable of revisions to graph or None for all.
704
:return: a Graph object with the graph reachable from revision_ids.
706
result = deprecated_graph.Graph()
708
pending = set(self.all_revision_ids())
711
pending = set(osutils.safe_revision_id(r) for r in revision_ids)
712
# special case NULL_REVISION
713
if _mod_revision.NULL_REVISION in pending:
714
pending.remove(_mod_revision.NULL_REVISION)
715
required = set(pending)
718
revision_id = pending.pop()
720
rev = self.get_revision(revision_id)
721
except errors.NoSuchRevision:
722
if revision_id in required:
725
result.add_ghost(revision_id)
727
for parent_id in rev.parent_ids:
728
# is this queued or done ?
729
if (parent_id not in pending and
730
parent_id not in done):
732
pending.add(parent_id)
733
result.add_node(revision_id, rev.parent_ids)
734
done.add(revision_id)
737
def _get_history_vf(self):
738
"""Get a versionedfile whose history graph reflects all revisions.
740
For weave repositories, this is the inventory weave.
742
return self.get_inventory_weave()
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])
744
2493
def iter_reverse_revision_history(self, revision_id):
745
2494
"""Iterate backwards through revision ids in the lefthand history
789
2539
reconciler.reconcile()
790
2540
return reconciler
2542
def _refresh_data(self):
2543
"""Helper called from lock_* to ensure coherency with disk.
2545
The default implementation does nothing; it is however possible
2546
for repositories to maintain loaded indices across multiple locks
2547
by checking inside their implementation of this method to see
2548
whether their indices are still valid. This depends of course on
2549
the disk format being validatable in this manner. This method is
2550
also called by the refresh_data() public interface to cause a refresh
2551
to occur while in a write lock so that data inserted by a smart server
2552
push operation is visible on the client's instance of the physical
792
2556
@needs_read_lock
793
2557
def revision_tree(self, revision_id):
794
2558
"""Return Tree for a revision on this branch.
796
`revision_id` may be None for the empty tree revision.
2560
`revision_id` may be NULL_REVISION for the empty tree revision.
2562
revision_id = _mod_revision.ensure_null(revision_id)
798
2563
# TODO: refactor this to use an existing revision object
799
2564
# so we don't need to read it in twice.
800
if revision_id is None or revision_id == _mod_revision.NULL_REVISION:
801
return RevisionTree(self, Inventory(root_id=None),
2565
if revision_id == _mod_revision.NULL_REVISION:
2566
return RevisionTree(self, Inventory(root_id=None),
802
2567
_mod_revision.NULL_REVISION)
804
revision_id = osutils.safe_revision_id(revision_id)
805
2569
inv = self.get_revision_inventory(revision_id)
806
2570
return RevisionTree(self, inv, revision_id)
809
2572
def revision_trees(self, revision_ids):
810
"""Return Tree for a revision on this branch.
812
`revision_id` may not be None or 'null:'"""
813
assert None not in revision_ids
814
assert _mod_revision.NULL_REVISION not in revision_ids
815
texts = self.get_inventory_weave().get_texts(revision_ids)
816
for text, revision_id in zip(texts, revision_ids):
817
inv = self.deserialise_inventory(revision_id, text)
818
yield RevisionTree(self, inv, revision_id)
2573
"""Return Trees for revisions in this repository.
2575
:param revision_ids: a sequence of revision-ids;
2576
a revision-id may not be None or 'null:'
2578
inventories = self.iter_inventories(revision_ids)
2579
for inv in inventories:
2580
yield RevisionTree(self, inv, inv.revision_id)
2582
def _filtered_revision_trees(self, revision_ids, file_ids):
2583
"""Return Tree for a revision on this branch with only some files.
2585
:param revision_ids: a sequence of revision-ids;
2586
a revision-id may not be None or 'null:'
2587
:param file_ids: if not None, the result is filtered
2588
so that only those file-ids, their parents and their
2589
children are included.
2591
inventories = self.iter_inventories(revision_ids)
2592
for inv in inventories:
2593
# Should we introduce a FilteredRevisionTree class rather
2594
# than pre-filter the inventory here?
2595
filtered_inv = inv.filter(file_ids)
2596
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
820
2598
@needs_read_lock
821
2599
def get_ancestry(self, revision_id, topo_sorted=True):
822
2600
"""Return a list of revision-ids integrated by a revision.
824
The first element of the list is always None, indicating the origin
825
revision. This might change when we have history horizons, or
2602
The first element of the list is always None, indicating the origin
2603
revision. This might change when we have history horizons, or
826
2604
perhaps we should have a new API.
828
2606
This is topologically sorted.
830
if revision_id is None:
2608
if _mod_revision.is_null(revision_id):
832
revision_id = osutils.safe_revision_id(revision_id)
833
2610
if not self.has_revision(revision_id):
834
2611
raise errors.NoSuchRevision(self, revision_id)
835
w = self.get_inventory_weave()
836
candidates = w.get_ancestry(revision_id, topo_sorted)
837
return [None] + candidates # self._eliminate_revisions_not_present(candidates)
840
def print_file(self, file, revision_id):
841
"""Print `file` to stdout.
843
FIXME RBC 20060125 as John Meinel points out this is a bad api
844
- it writes to stdout, it assumes that that is valid etc. Fix
845
by creating a new more flexible convenience function.
2612
graph = self.get_graph()
2614
search = graph._make_breadth_first_searcher([revision_id])
2617
found, ghosts = search.next_with_ghosts()
2618
except StopIteration:
2621
if _mod_revision.NULL_REVISION in keys:
2622
keys.remove(_mod_revision.NULL_REVISION)
2624
parent_map = graph.get_parent_map(keys)
2625
keys = tsort.topo_sort(parent_map)
2626
return [None] + list(keys)
2628
def pack(self, hint=None):
2629
"""Compress the data within the repository.
2631
This operation only makes sense for some repository types. For other
2632
types it should be a no-op that just returns.
2634
This stub method does not require a lock, but subclasses should use
2635
@needs_write_lock as this is a long running call its reasonable to
2636
implicitly lock for the user.
2638
:param hint: If not supplied, the whole repository is packed.
2639
If supplied, the repository may use the hint parameter as a
2640
hint for the parts of the repository to pack. A hint can be
2641
obtained from the result of commit_write_group(). Out of
2642
date hints are simply ignored, because concurrent operations
2643
can obsolete them rapidly.
847
revision_id = osutils.safe_revision_id(revision_id)
848
tree = self.revision_tree(revision_id)
849
# use inventory as it was in that revision
850
file_id = tree.inventory.path2id(file)
852
# TODO: jam 20060427 Write a test for this code path
853
# it had a bug in it, and was raising the wrong
855
raise errors.BzrError("%r is not present in revision %s" % (file, revision_id))
856
tree.print_file(file_id)
858
2646
def get_transaction(self):
859
2647
return self.control_files.get_transaction()
861
def revision_parents(self, revision_id):
862
revision_id = osutils.safe_revision_id(revision_id)
863
return self.get_inventory_weave().parent_names(revision_id)
865
def get_parents(self, revision_ids):
866
"""See StackedParentsProvider.get_parents"""
2649
def get_parent_map(self, revision_ids):
2650
"""See graph.StackedParentsProvider.get_parent_map"""
2651
# revisions index works in keys; this just works in revisions
2652
# therefore wrap and unwrap
868
2655
for revision_id in revision_ids:
869
2656
if revision_id == _mod_revision.NULL_REVISION:
873
parents = self.get_revision(revision_id).parent_ids
874
except errors.NoSuchRevision:
877
if len(parents) == 0:
878
parents = [_mod_revision.NULL_REVISION]
879
parents_list.append(parents)
2657
result[revision_id] = ()
2658
elif revision_id is None:
2659
raise ValueError('get_parent_map(None) is not valid')
2661
query_keys.append((revision_id ,))
2662
for ((revision_id,), parent_keys) in \
2663
self.revisions.get_parent_map(query_keys).iteritems():
2665
result[revision_id] = tuple(parent_revid
2666
for (parent_revid,) in parent_keys)
2668
result[revision_id] = (_mod_revision.NULL_REVISION,)
882
2671
def _make_parents_provider(self):
1376
3404
content is copied.
1377
3405
:param pb: optional progress bar to use for progress reports. If not
1378
3406
provided a default one will be created.
1380
Returns the copied revision count and the failed revisions in a tuple:
1383
raise NotImplementedError(self.fetch)
3409
from bzrlib.fetch import RepoFetcher
3410
f = RepoFetcher(to_repository=self.target,
3411
from_repository=self.source,
3412
last_revision=revision_id,
3413
fetch_spec=fetch_spec,
3414
pb=pb, find_ghosts=find_ghosts)
3416
def _walk_to_common_revisions(self, revision_ids):
3417
"""Walk out from revision_ids in source to revisions target has.
3419
:param revision_ids: The start point for the search.
3420
:return: A set of revision ids.
3422
target_graph = self.target.get_graph()
3423
revision_ids = frozenset(revision_ids)
3424
missing_revs = set()
3425
source_graph = self.source.get_graph()
3426
# ensure we don't pay silly lookup costs.
3427
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3428
null_set = frozenset([_mod_revision.NULL_REVISION])
3429
searcher_exhausted = False
3433
# Iterate the searcher until we have enough next_revs
3434
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3436
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3437
next_revs.update(next_revs_part)
3438
ghosts.update(ghosts_part)
3439
except StopIteration:
3440
searcher_exhausted = True
3442
# If there are ghosts in the source graph, and the caller asked for
3443
# them, make sure that they are present in the target.
3444
# We don't care about other ghosts as we can't fetch them and
3445
# haven't been asked to.
3446
ghosts_to_check = set(revision_ids.intersection(ghosts))
3447
revs_to_get = set(next_revs).union(ghosts_to_check)
3449
have_revs = set(target_graph.get_parent_map(revs_to_get))
3450
# we always have NULL_REVISION present.
3451
have_revs = have_revs.union(null_set)
3452
# Check if the target is missing any ghosts we need.
3453
ghosts_to_check.difference_update(have_revs)
3455
# One of the caller's revision_ids is a ghost in both the
3456
# source and the target.
3457
raise errors.NoSuchRevision(
3458
self.source, ghosts_to_check.pop())
3459
missing_revs.update(next_revs - have_revs)
3460
# Because we may have walked past the original stop point, make
3461
# sure everything is stopped
3462
stop_revs = searcher.find_seen_ancestors(have_revs)
3463
searcher.stop_searching_any(stop_revs)
3464
if searcher_exhausted:
3466
return searcher.get_result()
1385
3468
@needs_read_lock
1386
def missing_revision_ids(self, revision_id=None):
3469
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1387
3470
"""Return the revision ids that source has that target does not.
1389
These are returned in topological order.
1391
3472
:param revision_id: only return revision ids included by this
3474
:param find_ghosts: If True find missing revisions in deep history
3475
rather than just finding the surface difference.
3476
:return: A bzrlib.graph.SearchResult.
3478
# stop searching at found target revisions.
3479
if not find_ghosts and revision_id is not None:
3480
return self._walk_to_common_revisions([revision_id])
1394
3481
# generic, possibly worst case, slow code path.
1395
3482
target_ids = set(self.target.all_revision_ids())
1396
3483
if revision_id is not None:
1397
# TODO: jam 20070210 InterRepository is internal enough that it
1398
# should assume revision_ids are already utf-8
1399
revision_id = osutils.safe_revision_id(revision_id)
1400
3484
source_ids = self.source.get_ancestry(revision_id)
1401
assert source_ids[0] is None
3485
if source_ids[0] is not None:
3486
raise AssertionError()
1402
3487
source_ids.pop(0)
1404
3489
source_ids = self.source.all_revision_ids()
1405
3490
result_set = set(source_ids).difference(target_ids)
1406
# this may look like a no-op: its not. It preserves the ordering
1407
# other_ids had while only returning the members from other_ids
1408
# that we've decided we need.
1409
return [rev_id for rev_id in source_ids if rev_id in result_set]
3491
return self.source.revision_ids_to_search_result(result_set)
3494
def _same_model(source, target):
3495
"""True if source and target have the same data representation.
3497
Note: this is always called on the base class; overriding it in a
3498
subclass will have no effect.
3501
InterRepository._assert_same_model(source, target)
3503
except errors.IncompatibleRepositories, e:
3507
def _assert_same_model(source, target):
3508
"""Raise an exception if two repositories do not use the same model.
3510
if source.supports_rich_root() != target.supports_rich_root():
3511
raise errors.IncompatibleRepositories(source, target,
3512
"different rich-root support")
3513
if source._serializer != target._serializer:
3514
raise errors.IncompatibleRepositories(source, target,
3515
"different serializers")
1412
3518
class InterSameDataRepository(InterRepository):
1413
3519
"""Code for converting between repositories that represent the same data.
1415
3521
Data format and model must match for this to work.
1419
3525
def _get_repo_format_to_test(self):
1420
"""Repository format for testing with."""
1421
return RepositoryFormat.get_default_format()
3526
"""Repository format for testing with.
3528
InterSameData can pull from subtree to subtree and from non-subtree to
3529
non-subtree, so we test this with the richest repository format.
3531
from bzrlib.repofmt import knitrepo
3532
return knitrepo.RepositoryFormatKnit3()
1424
3535
def is_compatible(source, target):
1425
if source.supports_rich_root() != target.supports_rich_root():
1427
if source._serializer != target._serializer:
1432
def copy_content(self, revision_id=None):
1433
"""Make a complete copy of the content in self into destination.
1435
This copies both the repository's revision data, and configuration information
1436
such as the make_working_trees setting.
1438
This is a destructive operation! Do not use it on existing
1441
:param revision_id: Only copy the content needed to construct
1442
revision_id and its parents.
1445
self.target.set_make_working_trees(self.source.make_working_trees())
1446
except NotImplementedError:
1448
# TODO: jam 20070210 This is fairly internal, so we should probably
1449
# just assert that revision_id is not unicode.
1450
revision_id = osutils.safe_revision_id(revision_id)
1451
# but don't bother fetching if we have the needed data now.
1452
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1453
self.target.has_revision(revision_id)):
1455
self.target.fetch(self.source, revision_id=revision_id)
1458
def fetch(self, revision_id=None, pb=None):
1459
"""See InterRepository.fetch()."""
1460
from bzrlib.fetch import GenericRepoFetcher
1461
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1462
self.source, self.source._format, self.target,
1463
self.target._format)
1464
# TODO: jam 20070210 This should be an assert, not a translate
1465
revision_id = osutils.safe_revision_id(revision_id)
1466
f = GenericRepoFetcher(to_repository=self.target,
1467
from_repository=self.source,
1468
last_revision=revision_id,
1470
return f.count_copied, f.failed_revisions
3536
return InterRepository._same_model(source, target)
1473
3539
class InterWeaveRepo(InterSameDataRepository):
1474
"""Optimised code paths between Weave based repositories."""
3540
"""Optimised code paths between Weave based repositories.
3542
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3543
implemented lazy inter-object optimisation.
1477
3547
def _get_repo_format_to_test(self):
1601
3660
def is_compatible(source, target):
1602
3661
"""Be compatible with known Knit formats.
1604
3663
We don't test for the stores being of specific types because that
1605
could lead to confusing results, and there is no need to be
3664
could lead to confusing results, and there is no need to be
1606
3665
overly general.
1608
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1
3667
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
1610
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1611
isinstance(target._format, (RepositoryFormatKnit1)))
3669
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3670
isinstance(target._format, RepositoryFormatKnit))
1612
3671
except AttributeError:
1616
def fetch(self, revision_id=None, pb=None):
1617
"""See InterRepository.fetch()."""
1618
from bzrlib.fetch import KnitRepoFetcher
1619
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1620
self.source, self.source._format, self.target, self.target._format)
1621
# TODO: jam 20070210 This should be an assert, not a translate
1622
revision_id = osutils.safe_revision_id(revision_id)
1623
f = KnitRepoFetcher(to_repository=self.target,
1624
from_repository=self.source,
1625
last_revision=revision_id,
1627
return f.count_copied, f.failed_revisions
3673
return are_knits and InterRepository._same_model(source, target)
1629
3675
@needs_read_lock
1630
def missing_revision_ids(self, revision_id=None):
3676
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1631
3677
"""See InterRepository.missing_revision_ids()."""
1632
3678
if revision_id is not None:
1633
3679
source_ids = self.source.get_ancestry(revision_id)
1634
assert source_ids[0] is None
3680
if source_ids[0] is not None:
3681
raise AssertionError()
1635
3682
source_ids.pop(0)
1637
source_ids = self.source._all_possible_ids()
3684
source_ids = self.source.all_revision_ids()
1638
3685
source_ids_set = set(source_ids)
1639
3686
# source_ids is the worst possible case we may need to pull.
1640
3687
# now we want to filter source_ids against what we actually
1641
3688
# have in target, but don't try to check for existence where we know
1642
3689
# we do not have a revision as that would be pointless.
1643
target_ids = set(self.target._all_possible_ids())
3690
target_ids = set(self.target.all_revision_ids())
1644
3691
possibly_present_revisions = target_ids.intersection(source_ids_set)
1645
actually_present_revisions = set(self.target._eliminate_revisions_not_present(possibly_present_revisions))
3692
actually_present_revisions = set(
3693
self.target._eliminate_revisions_not_present(possibly_present_revisions))
1646
3694
required_revisions = source_ids_set.difference(actually_present_revisions)
1647
required_topo_revisions = [rev_id for rev_id in source_ids if rev_id in required_revisions]
1648
3695
if revision_id is not None:
1649
3696
# we used get_ancestry to determine source_ids then we are assured all
1650
3697
# revisions referenced are present as they are installed in topological order.
1651
3698
# and the tip revision was validated by get_ancestry.
1652
return required_topo_revisions
3699
result_set = required_revisions
1654
# if we just grabbed the possibly available ids, then
3701
# if we just grabbed the possibly available ids, then
1655
3702
# we only have an estimate of whats available and need to validate
1656
3703
# that against the revision records.
1657
return self.source._eliminate_revisions_not_present(required_topo_revisions)
1660
class InterModel1and2(InterRepository):
1663
def _get_repo_format_to_test(self):
1667
def is_compatible(source, target):
1668
if not source.supports_rich_root() and target.supports_rich_root():
1674
def fetch(self, revision_id=None, pb=None):
1675
"""See InterRepository.fetch()."""
1676
from bzrlib.fetch import Model1toKnit2Fetcher
1677
# TODO: jam 20070210 This should be an assert, not a translate
1678
revision_id = osutils.safe_revision_id(revision_id)
1679
f = Model1toKnit2Fetcher(to_repository=self.target,
1680
from_repository=self.source,
1681
last_revision=revision_id,
1683
return f.count_copied, f.failed_revisions
1686
def copy_content(self, revision_id=None):
1687
"""Make a complete copy of the content in self into destination.
1689
This is a destructive operation! Do not use it on existing
1692
:param revision_id: Only copy the content needed to construct
1693
revision_id and its parents.
1696
self.target.set_make_working_trees(self.source.make_working_trees())
1697
except NotImplementedError:
1699
# TODO: jam 20070210 Internal, assert, don't translate
1700
revision_id = osutils.safe_revision_id(revision_id)
1701
# but don't bother fetching if we have the needed data now.
1702
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1703
self.target.has_revision(revision_id)):
1705
self.target.fetch(self.source, revision_id=revision_id)
1708
class InterKnit1and2(InterKnitRepo):
1711
def _get_repo_format_to_test(self):
1715
def is_compatible(source, target):
1716
"""Be compatible with Knit1 source and Knit3 target"""
1717
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit3
1719
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1, \
1720
RepositoryFormatKnit3
1721
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1722
isinstance(target._format, (RepositoryFormatKnit3)))
1723
except AttributeError:
1727
def fetch(self, revision_id=None, pb=None):
1728
"""See InterRepository.fetch()."""
1729
from bzrlib.fetch import Knit1to2Fetcher
1730
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1731
self.source, self.source._format, self.target,
1732
self.target._format)
1733
# TODO: jam 20070210 This should be an assert, not a translate
1734
revision_id = osutils.safe_revision_id(revision_id)
1735
f = Knit1to2Fetcher(to_repository=self.target,
1736
from_repository=self.source,
1737
last_revision=revision_id,
1739
return f.count_copied, f.failed_revisions
1742
class InterRemoteRepository(InterRepository):
1743
"""Code for converting between RemoteRepository objects.
1745
This just gets an non-remote repository from the RemoteRepository, and calls
1746
InterRepository.get again.
1749
def __init__(self, source, target):
1750
if isinstance(source, remote.RemoteRepository):
1751
source._ensure_real()
1752
real_source = source._real_repository
1754
real_source = source
1755
if isinstance(target, remote.RemoteRepository):
1756
target._ensure_real()
1757
real_target = target._real_repository
1759
real_target = target
1760
self.real_inter = InterRepository.get(real_source, real_target)
1763
def is_compatible(source, target):
1764
if isinstance(source, remote.RemoteRepository):
1766
if isinstance(target, remote.RemoteRepository):
1770
def copy_content(self, revision_id=None):
1771
self.real_inter.copy_content(revision_id=revision_id)
1773
def fetch(self, revision_id=None, pb=None):
1774
self.real_inter.fetch(revision_id=revision_id, pb=pb)
1777
def _get_repo_format_to_test(self):
3705
self.source._eliminate_revisions_not_present(required_revisions))
3706
return self.source.revision_ids_to_search_result(result_set)
3709
class InterDifferingSerializer(InterRepository):
3712
def _get_repo_format_to_test(self):
3716
def is_compatible(source, target):
3717
"""Be compatible with Knit2 source and Knit3 target"""
3718
# This is redundant with format.check_conversion_target(), however that
3719
# raises an exception, and we just want to say "False" as in we won't
3720
# support converting between these formats.
3721
if 'IDS_never' in debug.debug_flags:
3723
if source.supports_rich_root() and not target.supports_rich_root():
3725
if (source._format.supports_tree_reference
3726
and not target._format.supports_tree_reference):
3728
if target._fallback_repositories and target._format.supports_chks:
3729
# IDS doesn't know how to copy CHKs for the parent inventories it
3730
# adds to stacked repos.
3732
if 'IDS_always' in debug.debug_flags:
3734
# Only use this code path for local source and target. IDS does far
3735
# too much IO (both bandwidth and roundtrips) over a network.
3736
if not source.bzrdir.transport.base.startswith('file:///'):
3738
if not target.bzrdir.transport.base.startswith('file:///'):
3742
def _get_trees(self, revision_ids, cache):
3744
for rev_id in revision_ids:
3746
possible_trees.append((rev_id, cache[rev_id]))
3748
# Not cached, but inventory might be present anyway.
3750
tree = self.source.revision_tree(rev_id)
3751
except errors.NoSuchRevision:
3752
# Nope, parent is ghost.
3755
cache[rev_id] = tree
3756
possible_trees.append((rev_id, tree))
3757
return possible_trees
3759
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3760
"""Get the best delta and base for this revision.
3762
:return: (basis_id, delta)
3765
# Generate deltas against each tree, to find the shortest.
3766
texts_possibly_new_in_tree = set()
3767
for basis_id, basis_tree in possible_trees:
3768
delta = tree.inventory._make_delta(basis_tree.inventory)
3769
for old_path, new_path, file_id, new_entry in delta:
3770
if new_path is None:
3771
# This file_id isn't present in the new rev, so we don't
3775
# Rich roots are handled elsewhere...
3777
kind = new_entry.kind
3778
if kind != 'directory' and kind != 'file':
3779
# No text record associated with this inventory entry.
3781
# This is a directory or file that has changed somehow.
3782
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3783
deltas.append((len(delta), basis_id, delta))
3785
return deltas[0][1:]
3787
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3788
"""Find all parent revisions that are absent, but for which the
3789
inventory is present, and copy those inventories.
3791
This is necessary to preserve correctness when the source is stacked
3792
without fallbacks configured. (Note that in cases like upgrade the
3793
source may be not have _fallback_repositories even though it is
3797
for parents in parent_map.values():
3798
parent_revs.update(parents)
3799
present_parents = self.source.get_parent_map(parent_revs)
3800
absent_parents = set(parent_revs).difference(present_parents)
3801
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3802
(rev_id,) for rev_id in absent_parents)
3803
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3804
for parent_tree in self.source.revision_trees(parent_inv_ids):
3805
current_revision_id = parent_tree.get_revision_id()
3806
parents_parents_keys = parent_invs_keys_for_stacking[
3807
(current_revision_id,)]
3808
parents_parents = [key[-1] for key in parents_parents_keys]
3809
basis_id = _mod_revision.NULL_REVISION
3810
basis_tree = self.source.revision_tree(basis_id)
3811
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3812
self.target.add_inventory_by_delta(
3813
basis_id, delta, current_revision_id, parents_parents)
3814
cache[current_revision_id] = parent_tree
3816
def _fetch_batch(self, revision_ids, basis_id, cache):
3817
"""Fetch across a few revisions.
3819
:param revision_ids: The revisions to copy
3820
:param basis_id: The revision_id of a tree that must be in cache, used
3821
as a basis for delta when no other base is available
3822
:param cache: A cache of RevisionTrees that we can use.
3823
:return: The revision_id of the last converted tree. The RevisionTree
3824
for it will be in cache
3826
# Walk though all revisions; get inventory deltas, copy referenced
3827
# texts that delta references, insert the delta, revision and
3829
root_keys_to_create = set()
3832
pending_revisions = []
3833
parent_map = self.source.get_parent_map(revision_ids)
3834
self._fetch_parent_invs_for_stacking(parent_map, cache)
3835
for tree in self.source.revision_trees(revision_ids):
3836
# Find a inventory delta for this revision.
3837
# Find text entries that need to be copied, too.
3838
current_revision_id = tree.get_revision_id()
3839
parent_ids = parent_map.get(current_revision_id, ())
3840
parent_trees = self._get_trees(parent_ids, cache)
3841
possible_trees = list(parent_trees)
3842
if len(possible_trees) == 0:
3843
# There either aren't any parents, or the parents are ghosts,
3844
# so just use the last converted tree.
3845
possible_trees.append((basis_id, cache[basis_id]))
3846
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3848
revision = self.source.get_revision(current_revision_id)
3849
pending_deltas.append((basis_id, delta,
3850
current_revision_id, revision.parent_ids))
3851
if self._converting_to_rich_root:
3852
self._revision_id_to_root_id[current_revision_id] = \
3854
# Determine which texts are in present in this revision but not in
3855
# any of the available parents.
3856
texts_possibly_new_in_tree = set()
3857
for old_path, new_path, file_id, entry in delta:
3858
if new_path is None:
3859
# This file_id isn't present in the new rev
3863
if not self.target.supports_rich_root():
3864
# The target doesn't support rich root, so we don't
3867
if self._converting_to_rich_root:
3868
# This can't be copied normally, we have to insert
3870
root_keys_to_create.add((file_id, entry.revision))
3873
texts_possibly_new_in_tree.add((file_id, entry.revision))
3874
for basis_id, basis_tree in possible_trees:
3875
basis_inv = basis_tree.inventory
3876
for file_key in list(texts_possibly_new_in_tree):
3877
file_id, file_revision = file_key
3879
entry = basis_inv[file_id]
3880
except errors.NoSuchId:
3882
if entry.revision == file_revision:
3883
texts_possibly_new_in_tree.remove(file_key)
3884
text_keys.update(texts_possibly_new_in_tree)
3885
pending_revisions.append(revision)
3886
cache[current_revision_id] = tree
3887
basis_id = current_revision_id
3889
from_texts = self.source.texts
3890
to_texts = self.target.texts
3891
if root_keys_to_create:
3892
from bzrlib.fetch import _new_root_data_stream
3893
root_stream = _new_root_data_stream(
3894
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3896
to_texts.insert_record_stream(root_stream)
3897
to_texts.insert_record_stream(from_texts.get_record_stream(
3898
text_keys, self.target._format._fetch_order,
3899
not self.target._format._fetch_uses_deltas))
3900
# insert inventory deltas
3901
for delta in pending_deltas:
3902
self.target.add_inventory_by_delta(*delta)
3903
if self.target._fallback_repositories:
3904
# Make sure this stacked repository has all the parent inventories
3905
# for the new revisions that we are about to insert. We do this
3906
# before adding the revisions so that no revision is added until
3907
# all the inventories it may depend on are added.
3908
# Note that this is overzealous, as we may have fetched these in an
3911
revision_ids = set()
3912
for revision in pending_revisions:
3913
revision_ids.add(revision.revision_id)
3914
parent_ids.update(revision.parent_ids)
3915
parent_ids.difference_update(revision_ids)
3916
parent_ids.discard(_mod_revision.NULL_REVISION)
3917
parent_map = self.source.get_parent_map(parent_ids)
3918
# we iterate over parent_map and not parent_ids because we don't
3919
# want to try copying any revision which is a ghost
3920
for parent_tree in self.source.revision_trees(parent_map):
3921
current_revision_id = parent_tree.get_revision_id()
3922
parents_parents = parent_map[current_revision_id]
3923
possible_trees = self._get_trees(parents_parents, cache)
3924
if len(possible_trees) == 0:
3925
# There either aren't any parents, or the parents are
3926
# ghosts, so just use the last converted tree.
3927
possible_trees.append((basis_id, cache[basis_id]))
3928
basis_id, delta = self._get_delta_for_revision(parent_tree,
3929
parents_parents, possible_trees)
3930
self.target.add_inventory_by_delta(
3931
basis_id, delta, current_revision_id, parents_parents)
3932
# insert signatures and revisions
3933
for revision in pending_revisions:
3935
signature = self.source.get_signature_text(
3936
revision.revision_id)
3937
self.target.add_signature_text(revision.revision_id,
3939
except errors.NoSuchRevision:
3941
self.target.add_revision(revision.revision_id, revision)
3944
def _fetch_all_revisions(self, revision_ids, pb):
3945
"""Fetch everything for the list of revisions.
3947
:param revision_ids: The list of revisions to fetch. Must be in
3949
:param pb: A ProgressTask
3952
basis_id, basis_tree = self._get_basis(revision_ids[0])
3954
cache = lru_cache.LRUCache(100)
3955
cache[basis_id] = basis_tree
3956
del basis_tree # We don't want to hang on to it here
3958
for offset in range(0, len(revision_ids), batch_size):
3959
self.target.start_write_group()
3961
pb.update('Transferring revisions', offset,
3963
batch = revision_ids[offset:offset+batch_size]
3964
basis_id = self._fetch_batch(batch, basis_id, cache)
3966
self.target.abort_write_group()
3969
hint = self.target.commit_write_group()
3972
if hints and self.target._format.pack_compresses:
3973
self.target.pack(hint=hints)
3974
pb.update('Transferring revisions', len(revision_ids),
3978
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3980
"""See InterRepository.fetch()."""
3981
if fetch_spec is not None:
3982
raise AssertionError("Not implemented yet...")
3983
# See <https://launchpad.net/bugs/456077> asking for a warning here
3985
# nb this is only active for local-local fetches; other things using
3987
ui.ui_factory.warn_cross_format_fetch(self.source._format,
3988
self.target._format)
3989
if (not self.source.supports_rich_root()
3990
and self.target.supports_rich_root()):
3991
self._converting_to_rich_root = True
3992
self._revision_id_to_root_id = {}
3994
self._converting_to_rich_root = False
3995
revision_ids = self.target.search_missing_revision_ids(self.source,
3996
revision_id, find_ghosts=find_ghosts).get_keys()
3997
if not revision_ids:
3999
revision_ids = tsort.topo_sort(
4000
self.source.get_graph().get_parent_map(revision_ids))
4001
if not revision_ids:
4003
# Walk though all revisions; get inventory deltas, copy referenced
4004
# texts that delta references, insert the delta, revision and
4007
my_pb = ui.ui_factory.nested_progress_bar()
4010
symbol_versioning.warn(
4011
symbol_versioning.deprecated_in((1, 14, 0))
4012
% "pb parameter to fetch()")
4015
self._fetch_all_revisions(revision_ids, pb)
4017
if my_pb is not None:
4019
return len(revision_ids), 0
4021
def _get_basis(self, first_revision_id):
4022
"""Get a revision and tree which exists in the target.
4024
This assumes that first_revision_id is selected for transmission
4025
because all other ancestors are already present. If we can't find an
4026
ancestor we fall back to NULL_REVISION since we know that is safe.
4028
:return: (basis_id, basis_tree)
4030
first_rev = self.source.get_revision(first_revision_id)
4032
basis_id = first_rev.parent_ids[0]
4033
# only valid as a basis if the target has it
4034
self.target.get_revision(basis_id)
4035
# Try to get a basis tree - if its a ghost it will hit the
4036
# NoSuchRevision case.
4037
basis_tree = self.source.revision_tree(basis_id)
4038
except (IndexError, errors.NoSuchRevision):
4039
basis_id = _mod_revision.NULL_REVISION
4040
basis_tree = self.source.revision_tree(basis_id)
4041
return basis_id, basis_tree
4044
InterRepository.register_optimiser(InterDifferingSerializer)
1781
4045
InterRepository.register_optimiser(InterSameDataRepository)
1782
4046
InterRepository.register_optimiser(InterWeaveRepo)
1783
4047
InterRepository.register_optimiser(InterKnitRepo)
1784
InterRepository.register_optimiser(InterModel1and2)
1785
InterRepository.register_optimiser(InterKnit1and2)
1786
InterRepository.register_optimiser(InterRemoteRepository)
1789
4050
class CopyConverter(object):
1790
4051
"""A repository conversion tool which just performs a copy of the content.
1792
4053
This is slow but quite reliable.
1838
4099
self.pb.update(message, self.count, self.total)
1841
class CommitBuilder(object):
1842
"""Provides an interface to build up a commit.
1844
This allows describing a tree to be committed without needing to
1845
know the internals of the format of the repository.
1848
record_root_entry = False
1849
def __init__(self, repository, parents, config, timestamp=None,
1850
timezone=None, committer=None, revprops=None,
1852
"""Initiate a CommitBuilder.
1854
:param repository: Repository to commit to.
1855
:param parents: Revision ids of the parents of the new revision.
1856
:param config: Configuration to use.
1857
:param timestamp: Optional timestamp recorded for commit.
1858
:param timezone: Optional timezone for timestamp.
1859
:param committer: Optional committer to set for commit.
1860
:param revprops: Optional dictionary of revision properties.
1861
:param revision_id: Optional revision id.
1863
self._config = config
1865
if committer is None:
1866
self._committer = self._config.username()
1868
assert isinstance(committer, basestring), type(committer)
1869
self._committer = committer
1871
self.new_inventory = Inventory(None)
1872
self._new_revision_id = osutils.safe_revision_id(revision_id)
1873
self.parents = parents
1874
self.repository = repository
1877
if revprops is not None:
1878
self._revprops.update(revprops)
1880
if timestamp is None:
1881
timestamp = time.time()
1882
# Restrict resolution to 1ms
1883
self._timestamp = round(timestamp, 3)
1885
if timezone is None:
1886
self._timezone = osutils.local_time_offset()
1888
self._timezone = int(timezone)
1890
self._generate_revision_if_needed()
1892
def commit(self, message):
1893
"""Make the actual commit.
1895
:return: The revision id of the recorded revision.
1897
rev = _mod_revision.Revision(
1898
timestamp=self._timestamp,
1899
timezone=self._timezone,
1900
committer=self._committer,
1902
inventory_sha1=self.inv_sha1,
1903
revision_id=self._new_revision_id,
1904
properties=self._revprops)
1905
rev.parent_ids = self.parents
1906
self.repository.add_revision(self._new_revision_id, rev,
1907
self.new_inventory, self._config)
1908
return self._new_revision_id
1910
def revision_tree(self):
1911
"""Return the tree that was just committed.
1913
After calling commit() this can be called to get a RevisionTree
1914
representing the newly committed tree. This is preferred to
1915
calling Repository.revision_tree() because that may require
1916
deserializing the inventory, while we already have a copy in
1919
return RevisionTree(self.repository, self.new_inventory,
1920
self._new_revision_id)
1922
def finish_inventory(self):
1923
"""Tell the builder that the inventory is finished."""
1924
if self.new_inventory.root is None:
1925
symbol_versioning.warn('Root entry should be supplied to'
1926
' record_entry_contents, as of bzr 0.10.',
1927
DeprecationWarning, stacklevel=2)
1928
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
1929
self.new_inventory.revision_id = self._new_revision_id
1930
self.inv_sha1 = self.repository.add_inventory(
1931
self._new_revision_id,
1936
def _gen_revision_id(self):
1937
"""Return new revision-id."""
1938
return generate_ids.gen_revision_id(self._config.username(),
1941
def _generate_revision_if_needed(self):
1942
"""Create a revision id if None was supplied.
1944
If the repository can not support user-specified revision ids
1945
they should override this function and raise CannotSetRevisionId
1946
if _new_revision_id is not None.
1948
:raises: CannotSetRevisionId
1950
if self._new_revision_id is None:
1951
self._new_revision_id = self._gen_revision_id()
1953
def record_entry_contents(self, ie, parent_invs, path, tree):
1954
"""Record the content of ie from tree into the commit if needed.
1956
Side effect: sets ie.revision when unchanged
1958
:param ie: An inventory entry present in the commit.
1959
:param parent_invs: The inventories of the parent revisions of the
1961
:param path: The path the entry is at in the tree.
1962
:param tree: The tree which contains this entry and should be used to
1965
if self.new_inventory.root is None and ie.parent_id is not None:
1966
symbol_versioning.warn('Root entry should be supplied to'
1967
' record_entry_contents, as of bzr 0.10.',
1968
DeprecationWarning, stacklevel=2)
1969
self.record_entry_contents(tree.inventory.root.copy(), parent_invs,
1971
self.new_inventory.add(ie)
1973
# ie.revision is always None if the InventoryEntry is considered
1974
# for committing. ie.snapshot will record the correct revision
1975
# which may be the sole parent if it is untouched.
1976
if ie.revision is not None:
1979
# In this revision format, root entries have no knit or weave
1980
if ie is self.new_inventory.root:
1981
# When serializing out to disk and back in
1982
# root.revision is always _new_revision_id
1983
ie.revision = self._new_revision_id
1985
previous_entries = ie.find_previous_heads(
1987
self.repository.weave_store,
1988
self.repository.get_transaction())
1989
# we are creating a new revision for ie in the history store
1991
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
1993
def modified_directory(self, file_id, file_parents):
1994
"""Record the presence of a symbolic link.
1996
:param file_id: The file_id of the link to record.
1997
:param file_parents: The per-file parent revision ids.
1999
self._add_text_to_weave(file_id, [], file_parents.keys())
2001
def modified_reference(self, file_id, file_parents):
2002
"""Record the modification of a reference.
2004
:param file_id: The file_id of the link to record.
2005
:param file_parents: The per-file parent revision ids.
2007
self._add_text_to_weave(file_id, [], file_parents.keys())
2009
def modified_file_text(self, file_id, file_parents,
2010
get_content_byte_lines, text_sha1=None,
2012
"""Record the text of file file_id
2014
:param file_id: The file_id of the file to record the text of.
2015
:param file_parents: The per-file parent revision ids.
2016
:param get_content_byte_lines: A callable which will return the byte
2018
:param text_sha1: Optional SHA1 of the file contents.
2019
:param text_size: Optional size of the file contents.
2021
# mutter('storing text of file {%s} in revision {%s} into %r',
2022
# file_id, self._new_revision_id, self.repository.weave_store)
2023
# special case to avoid diffing on renames or
2025
if (len(file_parents) == 1
2026
and text_sha1 == file_parents.values()[0].text_sha1
2027
and text_size == file_parents.values()[0].text_size):
2028
previous_ie = file_parents.values()[0]
2029
versionedfile = self.repository.weave_store.get_weave(file_id,
2030
self.repository.get_transaction())
2031
versionedfile.clone_text(self._new_revision_id,
2032
previous_ie.revision, file_parents.keys())
2033
return text_sha1, text_size
2035
new_lines = get_content_byte_lines()
2036
# TODO: Rather than invoking sha_strings here, _add_text_to_weave
2037
# should return the SHA1 and size
2038
self._add_text_to_weave(file_id, new_lines, file_parents.keys())
2039
return osutils.sha_strings(new_lines), \
2040
sum(map(len, new_lines))
2042
def modified_link(self, file_id, file_parents, link_target):
2043
"""Record the presence of a symbolic link.
2045
:param file_id: The file_id of the link to record.
2046
:param file_parents: The per-file parent revision ids.
2047
:param link_target: Target location of this link.
2049
self._add_text_to_weave(file_id, [], file_parents.keys())
2051
def _add_text_to_weave(self, file_id, new_lines, parents):
2052
versionedfile = self.repository.weave_store.get_weave_or_empty(
2053
file_id, self.repository.get_transaction())
2054
versionedfile.add_lines(self._new_revision_id, parents, new_lines)
2055
versionedfile.clear_cache()
2058
class _CommitBuilder(CommitBuilder):
2059
"""Temporary class so old CommitBuilders are detected properly
2061
Note: CommitBuilder works whether or not root entry is recorded.
2064
record_root_entry = True
2067
class RootCommitBuilder(CommitBuilder):
2068
"""This commitbuilder actually records the root id"""
2070
record_root_entry = True
2072
def record_entry_contents(self, ie, parent_invs, path, tree):
2073
"""Record the content of ie from tree into the commit if needed.
2075
Side effect: sets ie.revision when unchanged
2077
:param ie: An inventory entry present in the commit.
2078
:param parent_invs: The inventories of the parent revisions of the
2080
:param path: The path the entry is at in the tree.
2081
:param tree: The tree which contains this entry and should be used to
2084
assert self.new_inventory.root is not None or ie.parent_id is None
2085
self.new_inventory.add(ie)
2087
# ie.revision is always None if the InventoryEntry is considered
2088
# for committing. ie.snapshot will record the correct revision
2089
# which may be the sole parent if it is untouched.
2090
if ie.revision is not None:
2093
previous_entries = ie.find_previous_heads(
2095
self.repository.weave_store,
2096
self.repository.get_transaction())
2097
# we are creating a new revision for ie in the history store
2099
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
2102
4102
_unescape_map = {
2127
4127
if _unescape_re is None:
2128
4128
_unescape_re = re.compile('\&([^;]*);')
2129
4129
return _unescape_re.sub(_unescaper, data)
4132
class _VersionedFileChecker(object):
4134
def __init__(self, repository, text_key_references=None, ancestors=None):
4135
self.repository = repository
4136
self.text_index = self.repository._generate_text_key_index(
4137
text_key_references=text_key_references, ancestors=ancestors)
4139
def calculate_file_version_parents(self, text_key):
4140
"""Calculate the correct parents for a file version according to
4143
parent_keys = self.text_index[text_key]
4144
if parent_keys == [_mod_revision.NULL_REVISION]:
4146
return tuple(parent_keys)
4148
def check_file_version_parents(self, texts, progress_bar=None):
4149
"""Check the parents stored in a versioned file are correct.
4151
It also detects file versions that are not referenced by their
4152
corresponding revision's inventory.
4154
:returns: A tuple of (wrong_parents, dangling_file_versions).
4155
wrong_parents is a dict mapping {revision_id: (stored_parents,
4156
correct_parents)} for each revision_id where the stored parents
4157
are not correct. dangling_file_versions is a set of (file_id,
4158
revision_id) tuples for versions that are present in this versioned
4159
file, but not used by the corresponding inventory.
4161
local_progress = None
4162
if progress_bar is None:
4163
local_progress = ui.ui_factory.nested_progress_bar()
4164
progress_bar = local_progress
4166
return self._check_file_version_parents(texts, progress_bar)
4169
local_progress.finished()
4171
def _check_file_version_parents(self, texts, progress_bar):
4172
"""See check_file_version_parents."""
4174
self.file_ids = set([file_id for file_id, _ in
4175
self.text_index.iterkeys()])
4176
# text keys is now grouped by file_id
4177
n_versions = len(self.text_index)
4178
progress_bar.update('loading text store', 0, n_versions)
4179
parent_map = self.repository.texts.get_parent_map(self.text_index)
4180
# On unlistable transports this could well be empty/error...
4181
text_keys = self.repository.texts.keys()
4182
unused_keys = frozenset(text_keys) - set(self.text_index)
4183
for num, key in enumerate(self.text_index.iterkeys()):
4184
progress_bar.update('checking text graph', num, n_versions)
4185
correct_parents = self.calculate_file_version_parents(key)
4187
knit_parents = parent_map[key]
4188
except errors.RevisionNotPresent:
4191
if correct_parents != knit_parents:
4192
wrong_parents[key] = (knit_parents, correct_parents)
4193
return wrong_parents, unused_keys
4196
def _old_get_graph(repository, revision_id):
4197
"""DO NOT USE. That is all. I'm serious."""
4198
graph = repository.get_graph()
4199
revision_graph = dict(((key, value) for key, value in
4200
graph.iter_ancestry([revision_id]) if value is not None))
4201
return _strip_NULL_ghosts(revision_graph)
4204
def _strip_NULL_ghosts(revision_graph):
4205
"""Also don't use this. more compatibility code for unmigrated clients."""
4206
# Filter ghosts, and null:
4207
if _mod_revision.NULL_REVISION in revision_graph:
4208
del revision_graph[_mod_revision.NULL_REVISION]
4209
for key, parents in revision_graph.items():
4210
revision_graph[key] = tuple(parent for parent in parents if parent
4212
return revision_graph
4215
class StreamSink(object):
4216
"""An object that can insert a stream into a repository.
4218
This interface handles the complexity of reserialising inventories and
4219
revisions from different formats, and allows unidirectional insertion into
4220
stacked repositories without looking for the missing basis parents
4224
def __init__(self, target_repo):
4225
self.target_repo = target_repo
4227
def insert_stream(self, stream, src_format, resume_tokens):
4228
"""Insert a stream's content into the target repository.
4230
:param src_format: a bzr repository format.
4232
:return: a list of resume tokens and an iterable of keys additional
4233
items required before the insertion can be completed.
4235
self.target_repo.lock_write()
4238
self.target_repo.resume_write_group(resume_tokens)
4241
self.target_repo.start_write_group()
4244
# locked_insert_stream performs a commit|suspend.
4245
return self._locked_insert_stream(stream, src_format, is_resume)
4247
self.target_repo.abort_write_group(suppress_errors=True)
4250
self.target_repo.unlock()
4252
def _locked_insert_stream(self, stream, src_format, is_resume):
4253
to_serializer = self.target_repo._format._serializer
4254
src_serializer = src_format._serializer
4256
if to_serializer == src_serializer:
4257
# If serializers match and the target is a pack repository, set the
4258
# write cache size on the new pack. This avoids poor performance
4259
# on transports where append is unbuffered (such as
4260
# RemoteTransport). This is safe to do because nothing should read
4261
# back from the target repository while a stream with matching
4262
# serialization is being inserted.
4263
# The exception is that a delta record from the source that should
4264
# be a fulltext may need to be expanded by the target (see
4265
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4266
# explicitly flush any buffered writes first in that rare case.
4268
new_pack = self.target_repo._pack_collection._new_pack
4269
except AttributeError:
4270
# Not a pack repository
4273
new_pack.set_write_cache_size(1024*1024)
4274
for substream_type, substream in stream:
4275
if 'stream' in debug.debug_flags:
4276
mutter('inserting substream: %s', substream_type)
4277
if substream_type == 'texts':
4278
self.target_repo.texts.insert_record_stream(substream)
4279
elif substream_type == 'inventories':
4280
if src_serializer == to_serializer:
4281
self.target_repo.inventories.insert_record_stream(
4284
self._extract_and_insert_inventories(
4285
substream, src_serializer)
4286
elif substream_type == 'inventory-deltas':
4287
ui.ui_factory.warn_cross_format_fetch(src_format,
4288
self.target_repo._format)
4289
self._extract_and_insert_inventory_deltas(
4290
substream, src_serializer)
4291
elif substream_type == 'chk_bytes':
4292
# XXX: This doesn't support conversions, as it assumes the
4293
# conversion was done in the fetch code.
4294
self.target_repo.chk_bytes.insert_record_stream(substream)
4295
elif substream_type == 'revisions':
4296
# This may fallback to extract-and-insert more often than
4297
# required if the serializers are different only in terms of
4299
if src_serializer == to_serializer:
4300
self.target_repo.revisions.insert_record_stream(
4303
self._extract_and_insert_revisions(substream,
4305
elif substream_type == 'signatures':
4306
self.target_repo.signatures.insert_record_stream(substream)
4308
raise AssertionError('kaboom! %s' % (substream_type,))
4309
# Done inserting data, and the missing_keys calculations will try to
4310
# read back from the inserted data, so flush the writes to the new pack
4311
# (if this is pack format).
4312
if new_pack is not None:
4313
new_pack._write_data('', flush=True)
4314
# Find all the new revisions (including ones from resume_tokens)
4315
missing_keys = self.target_repo.get_missing_parent_inventories(
4316
check_for_missing_texts=is_resume)
4318
for prefix, versioned_file in (
4319
('texts', self.target_repo.texts),
4320
('inventories', self.target_repo.inventories),
4321
('revisions', self.target_repo.revisions),
4322
('signatures', self.target_repo.signatures),
4323
('chk_bytes', self.target_repo.chk_bytes),
4325
if versioned_file is None:
4327
missing_keys.update((prefix,) + key for key in
4328
versioned_file.get_missing_compression_parent_keys())
4329
except NotImplementedError:
4330
# cannot even attempt suspending, and missing would have failed
4331
# during stream insertion.
4332
missing_keys = set()
4335
# suspend the write group and tell the caller what we is
4336
# missing. We know we can suspend or else we would not have
4337
# entered this code path. (All repositories that can handle
4338
# missing keys can handle suspending a write group).
4339
write_group_tokens = self.target_repo.suspend_write_group()
4340
return write_group_tokens, missing_keys
4341
hint = self.target_repo.commit_write_group()
4342
if (to_serializer != src_serializer and
4343
self.target_repo._format.pack_compresses):
4344
self.target_repo.pack(hint=hint)
4347
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4348
target_rich_root = self.target_repo._format.rich_root_data
4349
target_tree_refs = self.target_repo._format.supports_tree_reference
4350
for record in substream:
4351
# Insert the delta directly
4352
inventory_delta_bytes = record.get_bytes_as('fulltext')
4353
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4355
parse_result = deserialiser.parse_text_bytes(
4356
inventory_delta_bytes)
4357
except inventory_delta.IncompatibleInventoryDelta, err:
4358
trace.mutter("Incompatible delta: %s", err.msg)
4359
raise errors.IncompatibleRevision(self.target_repo._format)
4360
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4361
revision_id = new_id
4362
parents = [key[0] for key in record.parents]
4363
self.target_repo.add_inventory_by_delta(
4364
basis_id, inv_delta, revision_id, parents)
4366
def _extract_and_insert_inventories(self, substream, serializer,
4368
"""Generate a new inventory versionedfile in target, converting data.
4370
The inventory is retrieved from the source, (deserializing it), and
4371
stored in the target (reserializing it in a different format).
4373
target_rich_root = self.target_repo._format.rich_root_data
4374
target_tree_refs = self.target_repo._format.supports_tree_reference
4375
for record in substream:
4376
# It's not a delta, so it must be a fulltext in the source
4377
# serializer's format.
4378
bytes = record.get_bytes_as('fulltext')
4379
revision_id = record.key[0]
4380
inv = serializer.read_inventory_from_string(bytes, revision_id)
4381
parents = [key[0] for key in record.parents]
4382
self.target_repo.add_inventory(revision_id, inv, parents)
4383
# No need to keep holding this full inv in memory when the rest of
4384
# the substream is likely to be all deltas.
4387
def _extract_and_insert_revisions(self, substream, serializer):
4388
for record in substream:
4389
bytes = record.get_bytes_as('fulltext')
4390
revision_id = record.key[0]
4391
rev = serializer.read_revision_from_string(bytes)
4392
if rev.revision_id != revision_id:
4393
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4394
self.target_repo.add_revision(revision_id, rev)
4397
if self.target_repo._format._fetch_reconcile:
4398
self.target_repo.reconcile()
4401
class StreamSource(object):
4402
"""A source of a stream for fetching between repositories."""
4404
def __init__(self, from_repository, to_format):
4405
"""Create a StreamSource streaming from from_repository."""
4406
self.from_repository = from_repository
4407
self.to_format = to_format
4409
def delta_on_metadata(self):
4410
"""Return True if delta's are permitted on metadata streams.
4412
That is on revisions and signatures.
4414
src_serializer = self.from_repository._format._serializer
4415
target_serializer = self.to_format._serializer
4416
return (self.to_format._fetch_uses_deltas and
4417
src_serializer == target_serializer)
4419
def _fetch_revision_texts(self, revs):
4420
# fetch signatures first and then the revision texts
4421
# may need to be a InterRevisionStore call here.
4422
from_sf = self.from_repository.signatures
4423
# A missing signature is just skipped.
4424
keys = [(rev_id,) for rev_id in revs]
4425
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4427
self.to_format._fetch_order,
4428
not self.to_format._fetch_uses_deltas))
4429
# If a revision has a delta, this is actually expanded inside the
4430
# insert_record_stream code now, which is an alternate fix for
4432
from_rf = self.from_repository.revisions
4433
revisions = from_rf.get_record_stream(
4435
self.to_format._fetch_order,
4436
not self.delta_on_metadata())
4437
return [('signatures', signatures), ('revisions', revisions)]
4439
def _generate_root_texts(self, revs):
4440
"""This will be called by get_stream between fetching weave texts and
4441
fetching the inventory weave.
4443
if self._rich_root_upgrade():
4445
return bzrlib.fetch.Inter1and2Helper(
4446
self.from_repository).generate_root_texts(revs)
4450
def get_stream(self, search):
4452
revs = search.get_keys()
4453
graph = self.from_repository.get_graph()
4454
revs = tsort.topo_sort(graph.get_parent_map(revs))
4455
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4457
for knit_kind, file_id, revisions in data_to_fetch:
4458
if knit_kind != phase:
4460
# Make a new progress bar for this phase
4461
if knit_kind == "file":
4462
# Accumulate file texts
4463
text_keys.extend([(file_id, revision) for revision in
4465
elif knit_kind == "inventory":
4466
# Now copy the file texts.
4467
from_texts = self.from_repository.texts
4468
yield ('texts', from_texts.get_record_stream(
4469
text_keys, self.to_format._fetch_order,
4470
not self.to_format._fetch_uses_deltas))
4471
# Cause an error if a text occurs after we have done the
4474
# Before we process the inventory we generate the root
4475
# texts (if necessary) so that the inventories references
4477
for _ in self._generate_root_texts(revs):
4479
# we fetch only the referenced inventories because we do not
4480
# know for unselected inventories whether all their required
4481
# texts are present in the other repository - it could be
4483
for info in self._get_inventory_stream(revs):
4485
elif knit_kind == "signatures":
4486
# Nothing to do here; this will be taken care of when
4487
# _fetch_revision_texts happens.
4489
elif knit_kind == "revisions":
4490
for record in self._fetch_revision_texts(revs):
4493
raise AssertionError("Unknown knit kind %r" % knit_kind)
4495
def get_stream_for_missing_keys(self, missing_keys):
4496
# missing keys can only occur when we are byte copying and not
4497
# translating (because translation means we don't send
4498
# unreconstructable deltas ever).
4500
keys['texts'] = set()
4501
keys['revisions'] = set()
4502
keys['inventories'] = set()
4503
keys['chk_bytes'] = set()
4504
keys['signatures'] = set()
4505
for key in missing_keys:
4506
keys[key[0]].add(key[1:])
4507
if len(keys['revisions']):
4508
# If we allowed copying revisions at this point, we could end up
4509
# copying a revision without copying its required texts: a
4510
# violation of the requirements for repository integrity.
4511
raise AssertionError(
4512
'cannot copy revisions to fill in missing deltas %s' % (
4513
keys['revisions'],))
4514
for substream_kind, keys in keys.iteritems():
4515
vf = getattr(self.from_repository, substream_kind)
4516
if vf is None and keys:
4517
raise AssertionError(
4518
"cannot fill in keys for a versioned file we don't"
4519
" have: %s needs %s" % (substream_kind, keys))
4521
# No need to stream something we don't have
4523
if substream_kind == 'inventories':
4524
# Some missing keys are genuinely ghosts, filter those out.
4525
present = self.from_repository.inventories.get_parent_map(keys)
4526
revs = [key[0] for key in present]
4527
# Get the inventory stream more-or-less as we do for the
4528
# original stream; there's no reason to assume that records
4529
# direct from the source will be suitable for the sink. (Think
4530
# e.g. 2a -> 1.9-rich-root).
4531
for info in self._get_inventory_stream(revs, missing=True):
4535
# Ask for full texts always so that we don't need more round trips
4536
# after this stream.
4537
# Some of the missing keys are genuinely ghosts, so filter absent
4538
# records. The Sink is responsible for doing another check to
4539
# ensure that ghosts don't introduce missing data for future
4541
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4542
self.to_format._fetch_order, True))
4543
yield substream_kind, stream
4545
def inventory_fetch_order(self):
4546
if self._rich_root_upgrade():
4547
return 'topological'
4549
return self.to_format._fetch_order
4551
def _rich_root_upgrade(self):
4552
return (not self.from_repository._format.rich_root_data and
4553
self.to_format.rich_root_data)
4555
def _get_inventory_stream(self, revision_ids, missing=False):
4556
from_format = self.from_repository._format
4557
if (from_format.supports_chks and self.to_format.supports_chks and
4558
from_format.network_name() == self.to_format.network_name()):
4559
raise AssertionError(
4560
"this case should be handled by GroupCHKStreamSource")
4561
elif 'forceinvdeltas' in debug.debug_flags:
4562
return self._get_convertable_inventory_stream(revision_ids,
4563
delta_versus_null=missing)
4564
elif from_format.network_name() == self.to_format.network_name():
4566
return self._get_simple_inventory_stream(revision_ids,
4568
elif (not from_format.supports_chks and not self.to_format.supports_chks
4569
and from_format._serializer == self.to_format._serializer):
4570
# Essentially the same format.
4571
return self._get_simple_inventory_stream(revision_ids,
4574
# Any time we switch serializations, we want to use an
4575
# inventory-delta based approach.
4576
return self._get_convertable_inventory_stream(revision_ids,
4577
delta_versus_null=missing)
4579
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4580
# NB: This currently reopens the inventory weave in source;
4581
# using a single stream interface instead would avoid this.
4582
from_weave = self.from_repository.inventories
4584
delta_closure = True
4586
delta_closure = not self.delta_on_metadata()
4587
yield ('inventories', from_weave.get_record_stream(
4588
[(rev_id,) for rev_id in revision_ids],
4589
self.inventory_fetch_order(), delta_closure))
4591
def _get_convertable_inventory_stream(self, revision_ids,
4592
delta_versus_null=False):
4593
# The two formats are sufficiently different that there is no fast
4594
# path, so we need to send just inventorydeltas, which any
4595
# sufficiently modern client can insert into any repository.
4596
# The StreamSink code expects to be able to
4597
# convert on the target, so we need to put bytes-on-the-wire that can
4598
# be converted. That means inventory deltas (if the remote is <1.19,
4599
# RemoteStreamSink will fallback to VFS to insert the deltas).
4600
yield ('inventory-deltas',
4601
self._stream_invs_as_deltas(revision_ids,
4602
delta_versus_null=delta_versus_null))
4604
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4605
"""Return a stream of inventory-deltas for the given rev ids.
4607
:param revision_ids: The list of inventories to transmit
4608
:param delta_versus_null: Don't try to find a minimal delta for this
4609
entry, instead compute the delta versus the NULL_REVISION. This
4610
effectively streams a complete inventory. Used for stuff like
4611
filling in missing parents, etc.
4613
from_repo = self.from_repository
4614
revision_keys = [(rev_id,) for rev_id in revision_ids]
4615
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4616
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4618
inventories = self.from_repository.iter_inventories(
4619
revision_ids, 'topological')
4620
format = from_repo._format
4621
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4622
inventory_cache = lru_cache.LRUCache(50)
4623
null_inventory = from_repo.revision_tree(
4624
_mod_revision.NULL_REVISION).inventory
4625
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4626
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4627
# repo back into a non-rich-root repo ought to be allowed)
4628
serializer = inventory_delta.InventoryDeltaSerializer(
4629
versioned_root=format.rich_root_data,
4630
tree_references=format.supports_tree_reference)
4631
for inv in inventories:
4632
key = (inv.revision_id,)
4633
parent_keys = parent_map.get(key, ())
4635
if not delta_versus_null and parent_keys:
4636
# The caller did not ask for complete inventories and we have
4637
# some parents that we can delta against. Make a delta against
4638
# each parent so that we can find the smallest.
4639
parent_ids = [parent_key[0] for parent_key in parent_keys]
4640
for parent_id in parent_ids:
4641
if parent_id not in invs_sent_so_far:
4642
# We don't know that the remote side has this basis, so
4645
if parent_id == _mod_revision.NULL_REVISION:
4646
parent_inv = null_inventory
4648
parent_inv = inventory_cache.get(parent_id, None)
4649
if parent_inv is None:
4650
parent_inv = from_repo.get_inventory(parent_id)
4651
candidate_delta = inv._make_delta(parent_inv)
4652
if (delta is None or
4653
len(delta) > len(candidate_delta)):
4654
delta = candidate_delta
4655
basis_id = parent_id
4657
# Either none of the parents ended up being suitable, or we
4658
# were asked to delta against NULL
4659
basis_id = _mod_revision.NULL_REVISION
4660
delta = inv._make_delta(null_inventory)
4661
invs_sent_so_far.add(inv.revision_id)
4662
inventory_cache[inv.revision_id] = inv
4663
delta_serialized = ''.join(
4664
serializer.delta_to_lines(basis_id, key[-1], delta))
4665
yield versionedfile.FulltextContentFactory(
4666
key, parent_keys, None, delta_serialized)
4669
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4670
stop_revision=None):
4671
"""Extend the partial history to include a given index
4673
If a stop_index is supplied, stop when that index has been reached.
4674
If a stop_revision is supplied, stop when that revision is
4675
encountered. Otherwise, stop when the beginning of history is
4678
:param stop_index: The index which should be present. When it is
4679
present, history extension will stop.
4680
:param stop_revision: The revision id which should be present. When
4681
it is encountered, history extension will stop.
4683
start_revision = partial_history_cache[-1]
4684
iterator = repo.iter_reverse_revision_history(start_revision)
4686
#skip the last revision in the list
4689
if (stop_index is not None and
4690
len(partial_history_cache) > stop_index):
4692
if partial_history_cache[-1] == stop_revision:
4694
revision_id = iterator.next()
4695
partial_history_cache.append(revision_id)
4696
except StopIteration: