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
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from binascii import hexlify
18
from copy import deepcopy
19
from cStringIO import StringIO
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
22
from unittest import TestSuite
24
from bzrlib import bzrdir, check, delta, gpg, errors, xml5, ui, transactions, osutils
25
from bzrlib.decorators import needs_read_lock, needs_write_lock
26
from bzrlib.errors import InvalidRevisionId
27
from bzrlib.graph import Graph
28
from bzrlib.inter import InterObject
29
from bzrlib.inventory import Inventory
30
from bzrlib.knit import KnitVersionedFile, KnitPlainFactory
31
from bzrlib.lockable_files import LockableFiles, TransportLock
32
from bzrlib.lockdir import LockDir
33
from bzrlib.osutils import (safe_unicode, rand_bytes, compact_date,
35
from bzrlib.revision import NULL_REVISION, Revision
42
revision as _mod_revision,
49
from bzrlib.bundle import serializer
36
50
from bzrlib.revisiontree import RevisionTree
37
from bzrlib.store.versioned import VersionedFileStore, WeaveStore
38
from bzrlib.store.text import TextStore
39
from bzrlib.symbol_versioning import (deprecated_method,
51
from bzrlib.store.versioned import VersionedFileStore
42
52
from bzrlib.testament import Testament
43
from bzrlib.trace import mutter, note
44
from bzrlib.tsort import topo_sort
45
from bzrlib.weave import WeaveFile
48
class Repository(object):
55
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
56
from bzrlib.inter import InterObject
57
from bzrlib.inventory import (
63
from bzrlib.lock import _RelockDebugMixin
64
from bzrlib import registry
65
from bzrlib.trace import (
66
log_exception_quietly, note, mutter, mutter_callsite, warning)
69
# Old formats display a warning, but only once
70
_deprecation_warning_done = False
73
class CommitBuilder(object):
74
"""Provides an interface to build up a commit.
76
This allows describing a tree to be committed without needing to
77
know the internals of the format of the repository.
80
# all clients should supply tree roots.
81
record_root_entry = True
82
# the default CommitBuilder does not manage trees whose root is versioned.
83
_versioned_root = False
85
def __init__(self, repository, parents, config, timestamp=None,
86
timezone=None, committer=None, revprops=None,
88
"""Initiate a CommitBuilder.
90
:param repository: Repository to commit to.
91
:param parents: Revision ids of the parents of the new revision.
92
:param config: Configuration to use.
93
:param timestamp: Optional timestamp recorded for commit.
94
:param timezone: Optional timezone for timestamp.
95
:param committer: Optional committer to set for commit.
96
:param revprops: Optional dictionary of revision properties.
97
:param revision_id: Optional revision id.
101
if committer is None:
102
self._committer = self._config.username()
104
self._committer = committer
106
self.new_inventory = Inventory(None)
107
self._new_revision_id = revision_id
108
self.parents = parents
109
self.repository = repository
112
if revprops is not None:
113
self._validate_revprops(revprops)
114
self._revprops.update(revprops)
116
if timestamp is None:
117
timestamp = time.time()
118
# Restrict resolution to 1ms
119
self._timestamp = round(timestamp, 3)
122
self._timezone = osutils.local_time_offset()
124
self._timezone = int(timezone)
126
self._generate_revision_if_needed()
127
self.__heads = graph.HeadsCache(repository.get_graph()).heads
128
self._basis_delta = []
129
# API compatibility, older code that used CommitBuilder did not call
130
# .record_delete(), which means the delta that is computed would not be
131
# valid. Callers that will call record_delete() should call
132
# .will_record_deletes() to indicate that.
133
self._recording_deletes = False
134
# memo'd check for no-op commits.
135
self._any_changes = False
137
def any_changes(self):
138
"""Return True if any entries were changed.
140
This includes merge-only changes. It is the core for the --unchanged
143
:return: True if any changes have occured.
145
return self._any_changes
147
def _validate_unicode_text(self, text, context):
148
"""Verify things like commit messages don't have bogus characters."""
150
raise ValueError('Invalid value for %s: %r' % (context, text))
152
def _validate_revprops(self, revprops):
153
for key, value in revprops.iteritems():
154
# We know that the XML serializers do not round trip '\r'
155
# correctly, so refuse to accept them
156
if not isinstance(value, basestring):
157
raise ValueError('revision property (%s) is not a valid'
158
' (unicode) string: %r' % (key, value))
159
self._validate_unicode_text(value,
160
'revision property (%s)' % (key,))
162
def commit(self, message):
163
"""Make the actual commit.
165
:return: The revision id of the recorded revision.
167
self._validate_unicode_text(message, 'commit message')
168
rev = _mod_revision.Revision(
169
timestamp=self._timestamp,
170
timezone=self._timezone,
171
committer=self._committer,
173
inventory_sha1=self.inv_sha1,
174
revision_id=self._new_revision_id,
175
properties=self._revprops)
176
rev.parent_ids = self.parents
177
self.repository.add_revision(self._new_revision_id, rev,
178
self.new_inventory, self._config)
179
self.repository.commit_write_group()
180
return self._new_revision_id
183
"""Abort the commit that is being built.
185
self.repository.abort_write_group()
187
def revision_tree(self):
188
"""Return the tree that was just committed.
190
After calling commit() this can be called to get a RevisionTree
191
representing the newly committed tree. This is preferred to
192
calling Repository.revision_tree() because that may require
193
deserializing the inventory, while we already have a copy in
196
if self.new_inventory is None:
197
self.new_inventory = self.repository.get_inventory(
198
self._new_revision_id)
199
return RevisionTree(self.repository, self.new_inventory,
200
self._new_revision_id)
202
def finish_inventory(self):
203
"""Tell the builder that the inventory is finished.
205
:return: The inventory id in the repository, which can be used with
206
repository.get_inventory.
208
if self.new_inventory is None:
209
# an inventory delta was accumulated without creating a new
211
basis_id = self.basis_delta_revision
212
# We ignore the 'inventory' returned by add_inventory_by_delta
213
# because self.new_inventory is used to hint to the rest of the
214
# system what code path was taken
215
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
216
basis_id, self._basis_delta, self._new_revision_id,
219
if self.new_inventory.root is None:
220
raise AssertionError('Root entry should be supplied to'
221
' record_entry_contents, as of bzr 0.10.')
222
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
223
self.new_inventory.revision_id = self._new_revision_id
224
self.inv_sha1 = self.repository.add_inventory(
225
self._new_revision_id,
229
return self._new_revision_id
231
def _gen_revision_id(self):
232
"""Return new revision-id."""
233
return generate_ids.gen_revision_id(self._config.username(),
236
def _generate_revision_if_needed(self):
237
"""Create a revision id if None was supplied.
239
If the repository can not support user-specified revision ids
240
they should override this function and raise CannotSetRevisionId
241
if _new_revision_id is not None.
243
:raises: CannotSetRevisionId
245
if self._new_revision_id is None:
246
self._new_revision_id = self._gen_revision_id()
247
self.random_revid = True
249
self.random_revid = False
251
def _heads(self, file_id, revision_ids):
252
"""Calculate the graph heads for revision_ids in the graph of file_id.
254
This can use either a per-file graph or a global revision graph as we
255
have an identity relationship between the two graphs.
257
return self.__heads(revision_ids)
259
def _check_root(self, ie, parent_invs, tree):
260
"""Helper for record_entry_contents.
262
:param ie: An entry being added.
263
:param parent_invs: The inventories of the parent revisions of the
265
:param tree: The tree that is being committed.
267
# In this revision format, root entries have no knit or weave When
268
# serializing out to disk and back in root.revision is always
270
ie.revision = self._new_revision_id
272
def _require_root_change(self, tree):
273
"""Enforce an appropriate root object change.
275
This is called once when record_iter_changes is called, if and only if
276
the root was not in the delta calculated by record_iter_changes.
278
:param tree: The tree which is being committed.
280
# NB: if there are no parents then this method is not called, so no
281
# need to guard on parents having length.
282
entry = entry_factory['directory'](tree.path2id(''), '',
284
entry.revision = self._new_revision_id
285
self._basis_delta.append(('', '', entry.file_id, entry))
287
def _get_delta(self, ie, basis_inv, path):
288
"""Get a delta against the basis inventory for ie."""
289
if ie.file_id not in basis_inv:
291
result = (None, path, ie.file_id, ie)
292
self._basis_delta.append(result)
294
elif ie != basis_inv[ie.file_id]:
296
# TODO: avoid tis id2path call.
297
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
298
self._basis_delta.append(result)
304
def get_basis_delta(self):
305
"""Return the complete inventory delta versus the basis inventory.
307
This has been built up with the calls to record_delete and
308
record_entry_contents. The client must have already called
309
will_record_deletes() to indicate that they will be generating a
312
:return: An inventory delta, suitable for use with apply_delta, or
313
Repository.add_inventory_by_delta, etc.
315
if not self._recording_deletes:
316
raise AssertionError("recording deletes not activated.")
317
return self._basis_delta
319
def record_delete(self, path, file_id):
320
"""Record that a delete occured against a basis tree.
322
This is an optional API - when used it adds items to the basis_delta
323
being accumulated by the commit builder. It cannot be called unless the
324
method will_record_deletes() has been called to inform the builder that
325
a delta is being supplied.
327
:param path: The path of the thing deleted.
328
:param file_id: The file id that was deleted.
330
if not self._recording_deletes:
331
raise AssertionError("recording deletes not activated.")
332
delta = (path, None, file_id, None)
333
self._basis_delta.append(delta)
334
self._any_changes = True
337
def will_record_deletes(self):
338
"""Tell the commit builder that deletes are being notified.
340
This enables the accumulation of an inventory delta; for the resulting
341
commit to be valid, deletes against the basis MUST be recorded via
342
builder.record_delete().
344
self._recording_deletes = True
346
basis_id = self.parents[0]
348
basis_id = _mod_revision.NULL_REVISION
349
self.basis_delta_revision = basis_id
351
def record_entry_contents(self, ie, parent_invs, path, tree,
353
"""Record the content of ie from tree into the commit if needed.
355
Side effect: sets ie.revision when unchanged
357
:param ie: An inventory entry present in the commit.
358
:param parent_invs: The inventories of the parent revisions of the
360
:param path: The path the entry is at in the tree.
361
:param tree: The tree which contains this entry and should be used to
363
:param content_summary: Summary data from the tree about the paths
364
content - stat, length, exec, sha/link target. This is only
365
accessed when the entry has a revision of None - that is when it is
366
a candidate to commit.
367
:return: A tuple (change_delta, version_recorded, fs_hash).
368
change_delta is an inventory_delta change for this entry against
369
the basis tree of the commit, or None if no change occured against
371
version_recorded is True if a new version of the entry has been
372
recorded. For instance, committing a merge where a file was only
373
changed on the other side will return (delta, False).
374
fs_hash is either None, or the hash details for the path (currently
375
a tuple of the contents sha1 and the statvalue returned by
376
tree.get_file_with_stat()).
378
if self.new_inventory.root is None:
379
if ie.parent_id is not None:
380
raise errors.RootMissing()
381
self._check_root(ie, parent_invs, tree)
382
if ie.revision is None:
383
kind = content_summary[0]
385
# ie is carried over from a prior commit
387
# XXX: repository specific check for nested tree support goes here - if
388
# the repo doesn't want nested trees we skip it ?
389
if (kind == 'tree-reference' and
390
not self.repository._format.supports_tree_reference):
391
# mismatch between commit builder logic and repository:
392
# this needs the entry creation pushed down into the builder.
393
raise NotImplementedError('Missing repository subtree support.')
394
self.new_inventory.add(ie)
396
# TODO: slow, take it out of the inner loop.
398
basis_inv = parent_invs[0]
400
basis_inv = Inventory(root_id=None)
402
# ie.revision is always None if the InventoryEntry is considered
403
# for committing. We may record the previous parents revision if the
404
# content is actually unchanged against a sole head.
405
if ie.revision is not None:
406
if not self._versioned_root and path == '':
407
# repositories that do not version the root set the root's
408
# revision to the new commit even when no change occurs (more
409
# specifically, they do not record a revision on the root; and
410
# the rev id is assigned to the root during deserialisation -
411
# this masks when a change may have occurred against the basis.
412
# To match this we always issue a delta, because the revision
413
# of the root will always be changing.
414
if ie.file_id in basis_inv:
415
delta = (basis_inv.id2path(ie.file_id), path,
419
delta = (None, path, ie.file_id, ie)
420
self._basis_delta.append(delta)
421
return delta, False, None
423
# we don't need to commit this, because the caller already
424
# determined that an existing revision of this file is
425
# appropriate. If its not being considered for committing then
426
# it and all its parents to the root must be unaltered so
427
# no-change against the basis.
428
if ie.revision == self._new_revision_id:
429
raise AssertionError("Impossible situation, a skipped "
430
"inventory entry (%r) claims to be modified in this "
431
"commit (%r).", (ie, self._new_revision_id))
432
return None, False, None
433
# XXX: Friction: parent_candidates should return a list not a dict
434
# so that we don't have to walk the inventories again.
435
parent_candiate_entries = ie.parent_candidates(parent_invs)
436
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
438
for inv in parent_invs:
439
if ie.file_id in inv:
440
old_rev = inv[ie.file_id].revision
441
if old_rev in head_set:
442
heads.append(inv[ie.file_id].revision)
443
head_set.remove(inv[ie.file_id].revision)
446
# now we check to see if we need to write a new record to the
448
# We write a new entry unless there is one head to the ancestors, and
449
# the kind-derived content is unchanged.
451
# Cheapest check first: no ancestors, or more the one head in the
452
# ancestors, we write a new node.
456
# There is a single head, look it up for comparison
457
parent_entry = parent_candiate_entries[heads[0]]
458
# if the non-content specific data has changed, we'll be writing a
460
if (parent_entry.parent_id != ie.parent_id or
461
parent_entry.name != ie.name):
463
# now we need to do content specific checks:
465
# if the kind changed the content obviously has
466
if kind != parent_entry.kind:
468
# Stat cache fingerprint feedback for the caller - None as we usually
469
# don't generate one.
472
if content_summary[2] is None:
473
raise ValueError("Files must not have executable = None")
475
# We can't trust a check of the file length because of content
477
if (# if the exec bit has changed we have to store:
478
parent_entry.executable != content_summary[2]):
480
elif parent_entry.text_sha1 == content_summary[3]:
481
# all meta and content is unchanged (using a hash cache
482
# hit to check the sha)
483
ie.revision = parent_entry.revision
484
ie.text_size = parent_entry.text_size
485
ie.text_sha1 = parent_entry.text_sha1
486
ie.executable = parent_entry.executable
487
return self._get_delta(ie, basis_inv, path), False, None
489
# Either there is only a hash change(no hash cache entry,
490
# or same size content change), or there is no change on
492
# Provide the parent's hash to the store layer, so that the
493
# content is unchanged we will not store a new node.
494
nostore_sha = parent_entry.text_sha1
496
# We want to record a new node regardless of the presence or
497
# absence of a content change in the file.
499
ie.executable = content_summary[2]
500
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
502
text = file_obj.read()
506
ie.text_sha1, ie.text_size = self._add_text_to_weave(
507
ie.file_id, text, heads, nostore_sha)
508
# Let the caller know we generated a stat fingerprint.
509
fingerprint = (ie.text_sha1, stat_value)
510
except errors.ExistingContent:
511
# Turns out that the file content was unchanged, and we were
512
# only going to store a new node if it was changed. Carry over
514
ie.revision = parent_entry.revision
515
ie.text_size = parent_entry.text_size
516
ie.text_sha1 = parent_entry.text_sha1
517
ie.executable = parent_entry.executable
518
return self._get_delta(ie, basis_inv, path), False, None
519
elif kind == 'directory':
521
# all data is meta here, nothing specific to directory, so
523
ie.revision = parent_entry.revision
524
return self._get_delta(ie, basis_inv, path), False, None
525
self._add_text_to_weave(ie.file_id, '', heads, None)
526
elif kind == 'symlink':
527
current_link_target = content_summary[3]
529
# symlink target is not generic metadata, check if it has
531
if current_link_target != parent_entry.symlink_target:
534
# unchanged, carry over.
535
ie.revision = parent_entry.revision
536
ie.symlink_target = parent_entry.symlink_target
537
return self._get_delta(ie, basis_inv, path), False, None
538
ie.symlink_target = current_link_target
539
self._add_text_to_weave(ie.file_id, '', heads, None)
540
elif kind == 'tree-reference':
542
if content_summary[3] != parent_entry.reference_revision:
545
# unchanged, carry over.
546
ie.reference_revision = parent_entry.reference_revision
547
ie.revision = parent_entry.revision
548
return self._get_delta(ie, basis_inv, path), False, None
549
ie.reference_revision = content_summary[3]
550
if ie.reference_revision is None:
551
raise AssertionError("invalid content_summary for nested tree: %r"
552
% (content_summary,))
553
self._add_text_to_weave(ie.file_id, '', heads, None)
555
raise NotImplementedError('unknown kind')
556
ie.revision = self._new_revision_id
557
self._any_changes = True
558
return self._get_delta(ie, basis_inv, path), True, fingerprint
560
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
561
_entry_factory=entry_factory):
562
"""Record a new tree via iter_changes.
564
:param tree: The tree to obtain text contents from for changed objects.
565
:param basis_revision_id: The revision id of the tree the iter_changes
566
has been generated against. Currently assumed to be the same
567
as self.parents[0] - if it is not, errors may occur.
568
:param iter_changes: An iter_changes iterator with the changes to apply
569
to basis_revision_id. The iterator must not include any items with
570
a current kind of None - missing items must be either filtered out
571
or errored-on beefore record_iter_changes sees the item.
572
:param _entry_factory: Private method to bind entry_factory locally for
574
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
577
# Create an inventory delta based on deltas between all the parents and
578
# deltas between all the parent inventories. We use inventory delta's
579
# between the inventory objects because iter_changes masks
580
# last-changed-field only changes.
582
# file_id -> change map, change is fileid, paths, changed, versioneds,
583
# parents, names, kinds, executables
585
# {file_id -> revision_id -> inventory entry, for entries in parent
586
# trees that are not parents[0]
590
revtrees = list(self.repository.revision_trees(self.parents))
591
except errors.NoSuchRevision:
592
# one or more ghosts, slow path.
594
for revision_id in self.parents:
596
revtrees.append(self.repository.revision_tree(revision_id))
597
except errors.NoSuchRevision:
599
basis_revision_id = _mod_revision.NULL_REVISION
601
revtrees.append(self.repository.revision_tree(
602
_mod_revision.NULL_REVISION))
603
# The basis inventory from a repository
605
basis_inv = revtrees[0].inventory
607
basis_inv = self.repository.revision_tree(
608
_mod_revision.NULL_REVISION).inventory
609
if len(self.parents) > 0:
610
if basis_revision_id != self.parents[0] and not ghost_basis:
612
"arbitrary basis parents not yet supported with merges")
613
for revtree in revtrees[1:]:
614
for change in revtree.inventory._make_delta(basis_inv):
615
if change[1] is None:
616
# Not present in this parent.
618
if change[2] not in merged_ids:
619
if change[0] is not None:
620
basis_entry = basis_inv[change[2]]
621
merged_ids[change[2]] = [
623
basis_entry.revision,
626
parent_entries[change[2]] = {
628
basis_entry.revision:basis_entry,
630
change[3].revision:change[3],
633
merged_ids[change[2]] = [change[3].revision]
634
parent_entries[change[2]] = {change[3].revision:change[3]}
636
merged_ids[change[2]].append(change[3].revision)
637
parent_entries[change[2]][change[3].revision] = change[3]
640
# Setup the changes from the tree:
641
# changes maps file_id -> (change, [parent revision_ids])
643
for change in iter_changes:
644
# This probably looks up in basis_inv way to much.
645
if change[1][0] is not None:
646
head_candidate = [basis_inv[change[0]].revision]
649
changes[change[0]] = change, merged_ids.get(change[0],
651
unchanged_merged = set(merged_ids) - set(changes)
652
# Extend the changes dict with synthetic changes to record merges of
654
for file_id in unchanged_merged:
655
# Record a merged version of these items that did not change vs the
656
# basis. This can be either identical parallel changes, or a revert
657
# of a specific file after a merge. The recorded content will be
658
# that of the current tree (which is the same as the basis), but
659
# the per-file graph will reflect a merge.
660
# NB:XXX: We are reconstructing path information we had, this
661
# should be preserved instead.
662
# inv delta change: (file_id, (path_in_source, path_in_target),
663
# changed_content, versioned, parent, name, kind,
666
basis_entry = basis_inv[file_id]
667
except errors.NoSuchId:
668
# a change from basis->some_parents but file_id isn't in basis
669
# so was new in the merge, which means it must have changed
670
# from basis -> current, and as it hasn't the add was reverted
671
# by the user. So we discard this change.
675
(basis_inv.id2path(file_id), tree.id2path(file_id)),
677
(basis_entry.parent_id, basis_entry.parent_id),
678
(basis_entry.name, basis_entry.name),
679
(basis_entry.kind, basis_entry.kind),
680
(basis_entry.executable, basis_entry.executable))
681
changes[file_id] = (change, merged_ids[file_id])
682
# changes contains tuples with the change and a set of inventory
683
# candidates for the file.
685
# old_path, new_path, file_id, new_inventory_entry
686
seen_root = False # Is the root in the basis delta?
687
inv_delta = self._basis_delta
688
modified_rev = self._new_revision_id
689
for change, head_candidates in changes.values():
690
if change[3][1]: # versioned in target.
691
# Several things may be happening here:
692
# We may have a fork in the per-file graph
693
# - record a change with the content from tree
694
# We may have a change against < all trees
695
# - carry over the tree that hasn't changed
696
# We may have a change against all trees
697
# - record the change with the content from tree
700
entry = _entry_factory[kind](file_id, change[5][1],
702
head_set = self._heads(change[0], set(head_candidates))
705
for head_candidate in head_candidates:
706
if head_candidate in head_set:
707
heads.append(head_candidate)
708
head_set.remove(head_candidate)
711
# Could be a carry-over situation:
712
parent_entry_revs = parent_entries.get(file_id, None)
713
if parent_entry_revs:
714
parent_entry = parent_entry_revs.get(heads[0], None)
717
if parent_entry is None:
718
# The parent iter_changes was called against is the one
719
# that is the per-file head, so any change is relevant
720
# iter_changes is valid.
721
carry_over_possible = False
723
# could be a carry over situation
724
# A change against the basis may just indicate a merge,
725
# we need to check the content against the source of the
726
# merge to determine if it was changed after the merge
728
if (parent_entry.kind != entry.kind or
729
parent_entry.parent_id != entry.parent_id or
730
parent_entry.name != entry.name):
731
# Metadata common to all entries has changed
732
# against per-file parent
733
carry_over_possible = False
735
carry_over_possible = True
736
# per-type checks for changes against the parent_entry
739
# Cannot be a carry-over situation
740
carry_over_possible = False
741
# Populate the entry in the delta
743
# XXX: There is still a small race here: If someone reverts the content of a file
744
# after iter_changes examines and decides it has changed,
745
# we will unconditionally record a new version even if some
746
# other process reverts it while commit is running (with
747
# the revert happening after iter_changes did it's
750
entry.executable = True
752
entry.executable = False
753
if (carry_over_possible and
754
parent_entry.executable == entry.executable):
755
# Check the file length, content hash after reading
757
nostore_sha = parent_entry.text_sha1
760
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
762
text = file_obj.read()
766
entry.text_sha1, entry.text_size = self._add_text_to_weave(
767
file_id, text, heads, nostore_sha)
768
yield file_id, change[1][1], (entry.text_sha1, stat_value)
769
except errors.ExistingContent:
770
# No content change against a carry_over parent
771
# Perhaps this should also yield a fs hash update?
773
entry.text_size = parent_entry.text_size
774
entry.text_sha1 = parent_entry.text_sha1
775
elif kind == 'symlink':
777
entry.symlink_target = tree.get_symlink_target(file_id)
778
if (carry_over_possible and
779
parent_entry.symlink_target == entry.symlink_target):
782
self._add_text_to_weave(change[0], '', heads, None)
783
elif kind == 'directory':
784
if carry_over_possible:
787
# Nothing to set on the entry.
788
# XXX: split into the Root and nonRoot versions.
789
if change[1][1] != '' or self.repository.supports_rich_root():
790
self._add_text_to_weave(change[0], '', heads, None)
791
elif kind == 'tree-reference':
792
if not self.repository._format.supports_tree_reference:
793
# This isn't quite sane as an error, but we shouldn't
794
# ever see this code path in practice: tree's don't
795
# permit references when the repo doesn't support tree
797
raise errors.UnsupportedOperation(tree.add_reference,
799
reference_revision = tree.get_reference_revision(change[0])
800
entry.reference_revision = reference_revision
801
if (carry_over_possible and
802
parent_entry.reference_revision == reference_revision):
805
self._add_text_to_weave(change[0], '', heads, None)
807
raise AssertionError('unknown kind %r' % kind)
809
entry.revision = modified_rev
811
entry.revision = parent_entry.revision
814
new_path = change[1][1]
815
inv_delta.append((change[1][0], new_path, change[0], entry))
818
self.new_inventory = None
820
# This should perhaps be guarded by a check that the basis we
821
# commit against is the basis for the commit and if not do a delta
823
self._any_changes = True
825
# housekeeping root entry changes do not affect no-change commits.
826
self._require_root_change(tree)
827
self.basis_delta_revision = basis_revision_id
829
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
830
parent_keys = tuple([(file_id, parent) for parent in parents])
831
return self.repository.texts._add_text(
832
(file_id, self._new_revision_id), parent_keys, new_text,
833
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
836
class RootCommitBuilder(CommitBuilder):
837
"""This commitbuilder actually records the root id"""
839
# the root entry gets versioned properly by this builder.
840
_versioned_root = True
842
def _check_root(self, ie, parent_invs, tree):
843
"""Helper for record_entry_contents.
845
:param ie: An entry being added.
846
:param parent_invs: The inventories of the parent revisions of the
848
:param tree: The tree that is being committed.
851
def _require_root_change(self, tree):
852
"""Enforce an appropriate root object change.
854
This is called once when record_iter_changes is called, if and only if
855
the root was not in the delta calculated by record_iter_changes.
857
:param tree: The tree which is being committed.
859
# versioned roots do not change unless the tree found a change.
862
######################################################################
866
class Repository(_RelockDebugMixin):
49
867
"""Repository holding history for one or more branches.
51
869
The repository holds and retrieves historical information including
52
870
revisions and file history. It's normally accessed only by the Branch,
53
871
which views a particular line of development through that history.
55
The Repository builds on top of Stores and a Transport, which respectively
56
describe the disk data format and the way of accessing the (possibly
873
The Repository builds on top of some byte storage facilies (the revisions,
874
signatures, inventories, texts and chk_bytes attributes) and a Transport,
875
which respectively provide byte storage and a means to access the (possibly
878
The byte storage facilities are addressed via tuples, which we refer to
879
as 'keys' throughout the code base. Revision_keys, inventory_keys and
880
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
881
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
882
byte string made up of a hash identifier and a hash value.
883
We use this interface because it allows low friction with the underlying
884
code that implements disk indices, network encoding and other parts of
887
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
888
the serialised revisions for the repository. This can be used to obtain
889
revision graph information or to access raw serialised revisions.
890
The result of trying to insert data into the repository via this store
891
is undefined: it should be considered read-only except for implementors
893
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
894
the serialised signatures for the repository. This can be used to
895
obtain access to raw serialised signatures. The result of trying to
896
insert data into the repository via this store is undefined: it should
897
be considered read-only except for implementors of repositories.
898
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
899
the serialised inventories for the repository. This can be used to
900
obtain unserialised inventories. The result of trying to insert data
901
into the repository via this store is undefined: it should be
902
considered read-only except for implementors of repositories.
903
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
904
texts of files and directories for the repository. This can be used to
905
obtain file texts or file graphs. Note that Repository.iter_file_bytes
906
is usually a better interface for accessing file texts.
907
The result of trying to insert data into the repository via this store
908
is undefined: it should be considered read-only except for implementors
910
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
911
any data the repository chooses to store or have indexed by its hash.
912
The result of trying to insert data into the repository via this store
913
is undefined: it should be considered read-only except for implementors
915
:ivar _transport: Transport for file access to repository, typically
916
pointing to .bzr/repository.
61
def add_inventory(self, revid, inv, parents):
62
"""Add the inventory inv to the repository as revid.
64
:param parents: The revision ids of the parents that revid
919
# What class to use for a CommitBuilder. Often its simpler to change this
920
# in a Repository class subclass rather than to override
921
# get_commit_builder.
922
_commit_builder_class = CommitBuilder
923
# The search regex used by xml based repositories to determine what things
924
# where changed in a single commit.
925
_file_ids_altered_regex = lazy_regex.lazy_compile(
926
r'file_id="(?P<file_id>[^"]+)"'
927
r'.* revision="(?P<revision_id>[^"]+)"'
930
def abort_write_group(self, suppress_errors=False):
931
"""Commit the contents accrued within the current write group.
933
:param suppress_errors: if true, abort_write_group will catch and log
934
unexpected errors that happen during the abort, rather than
935
allowing them to propagate. Defaults to False.
937
:seealso: start_write_group.
939
if self._write_group is not self.get_transaction():
940
# has an unlock or relock occured ?
943
'(suppressed) mismatched lock context and write group. %r, %r',
944
self._write_group, self.get_transaction())
946
raise errors.BzrError(
947
'mismatched lock context and write group. %r, %r' %
948
(self._write_group, self.get_transaction()))
950
self._abort_write_group()
951
except Exception, exc:
952
self._write_group = None
953
if not suppress_errors:
955
mutter('abort_write_group failed')
956
log_exception_quietly()
957
note('bzr: ERROR (ignored): %s', exc)
958
self._write_group = None
960
def _abort_write_group(self):
961
"""Template method for per-repository write group cleanup.
963
This is called during abort before the write group is considered to be
964
finished and should cleanup any internal state accrued during the write
965
group. There is no requirement that data handed to the repository be
966
*not* made available - this is not a rollback - but neither should any
967
attempt be made to ensure that data added is fully commited. Abort is
968
invoked when an error has occured so futher disk or network operations
969
may not be possible or may error and if possible should not be
973
def add_fallback_repository(self, repository):
974
"""Add a repository to use for looking up data not held locally.
976
:param repository: A repository.
978
if not self._format.supports_external_lookups:
979
raise errors.UnstackableRepositoryFormat(self._format, self.base)
981
# This repository will call fallback.unlock() when we transition to
982
# the unlocked state, so we make sure to increment the lock count
983
repository.lock_read()
984
self._check_fallback_repository(repository)
985
self._fallback_repositories.append(repository)
986
self.texts.add_fallback_versioned_files(repository.texts)
987
self.inventories.add_fallback_versioned_files(repository.inventories)
988
self.revisions.add_fallback_versioned_files(repository.revisions)
989
self.signatures.add_fallback_versioned_files(repository.signatures)
990
if self.chk_bytes is not None:
991
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
993
def _check_fallback_repository(self, repository):
994
"""Check that this repository can fallback to repository safely.
996
Raise an error if not.
998
:param repository: A repository to fallback to.
1000
return InterRepository._assert_same_model(self, repository)
1002
def add_inventory(self, revision_id, inv, parents):
1003
"""Add the inventory inv to the repository as revision_id.
1005
:param parents: The revision ids of the parents that revision_id
65
1006
is known to have and are in the repository already.
67
returns the sha1 of the serialized inventory.
69
assert inv.revision_id is None or inv.revision_id == revid, \
70
"Mismatch between inventory revision" \
71
" id and insertion revid (%r, %r)" % (inv.revision_id, revid)
72
inv_text = xml5.serializer_v5.write_inventory_to_string(inv)
73
inv_sha1 = osutils.sha_string(inv_text)
74
inv_vf = self.control_weaves.get_weave('inventory',
75
self.get_transaction())
76
self._inventory_add_lines(inv_vf, revid, parents, osutils.split_lines(inv_text))
79
def _inventory_add_lines(self, inv_vf, revid, parents, lines):
81
for parent in parents:
83
final_parents.append(parent)
85
inv_vf.add_lines(revid, final_parents, lines)
88
def add_revision(self, rev_id, rev, inv=None, config=None):
89
"""Add rev to the revision store as rev_id.
91
:param rev_id: the revision id to use.
1008
:returns: The validator(which is a sha1 digest, though what is sha'd is
1009
repository format specific) of the serialized inventory.
1011
if not self.is_in_write_group():
1012
raise AssertionError("%r not in write group" % (self,))
1013
_mod_revision.check_not_reserved_id(revision_id)
1014
if not (inv.revision_id is None or inv.revision_id == revision_id):
1015
raise AssertionError(
1016
"Mismatch between inventory revision"
1017
" id and insertion revid (%r, %r)"
1018
% (inv.revision_id, revision_id))
1019
if inv.root is None:
1020
raise AssertionError()
1021
return self._add_inventory_checked(revision_id, inv, parents)
1023
def _add_inventory_checked(self, revision_id, inv, parents):
1024
"""Add inv to the repository after checking the inputs.
1026
This function can be overridden to allow different inventory styles.
1028
:seealso: add_inventory, for the contract.
1030
inv_lines = self._serialise_inventory_to_lines(inv)
1031
return self._inventory_add_lines(revision_id, parents,
1032
inv_lines, check_content=False)
1034
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1035
parents, basis_inv=None, propagate_caches=False):
1036
"""Add a new inventory expressed as a delta against another revision.
1038
See the inventory developers documentation for the theory behind
1041
:param basis_revision_id: The inventory id the delta was created
1042
against. (This does not have to be a direct parent.)
1043
:param delta: The inventory delta (see Inventory.apply_delta for
1045
:param new_revision_id: The revision id that the inventory is being
1047
:param parents: The revision ids of the parents that revision_id is
1048
known to have and are in the repository already. These are supplied
1049
for repositories that depend on the inventory graph for revision
1050
graph access, as well as for those that pun ancestry with delta
1052
:param basis_inv: The basis inventory if it is already known,
1054
:param propagate_caches: If True, the caches for this inventory are
1055
copied to and updated for the result if possible.
1057
:returns: (validator, new_inv)
1058
The validator(which is a sha1 digest, though what is sha'd is
1059
repository format specific) of the serialized inventory, and the
1060
resulting inventory.
1062
if not self.is_in_write_group():
1063
raise AssertionError("%r not in write group" % (self,))
1064
_mod_revision.check_not_reserved_id(new_revision_id)
1065
basis_tree = self.revision_tree(basis_revision_id)
1066
basis_tree.lock_read()
1068
# Note that this mutates the inventory of basis_tree, which not all
1069
# inventory implementations may support: A better idiom would be to
1070
# return a new inventory, but as there is no revision tree cache in
1071
# repository this is safe for now - RBC 20081013
1072
if basis_inv is None:
1073
basis_inv = basis_tree.inventory
1074
basis_inv.apply_delta(delta)
1075
basis_inv.revision_id = new_revision_id
1076
return (self.add_inventory(new_revision_id, basis_inv, parents),
1081
def _inventory_add_lines(self, revision_id, parents, lines,
1082
check_content=True):
1083
"""Store lines in inv_vf and return the sha1 of the inventory."""
1084
parents = [(parent,) for parent in parents]
1085
result = self.inventories.add_lines((revision_id,), parents, lines,
1086
check_content=check_content)[0]
1087
self.inventories._access.flush()
1090
def add_revision(self, revision_id, rev, inv=None, config=None):
1091
"""Add rev to the revision store as revision_id.
1093
:param revision_id: the revision id to use.
92
1094
:param rev: The revision object.
93
1095
:param inv: The inventory for the revision. if None, it will be looked
94
1096
up in the inventory storer
255
1720
:param revprops: Optional dictionary of revision properties.
256
1721
:param revision_id: Optional revision id.
258
return CommitBuilder(self, parents, config, timestamp, timezone,
259
committer, revprops, revision_id)
1723
if self._fallback_repositories:
1724
raise errors.BzrError("Cannot commit from a lightweight checkout "
1725
"to a stacked branch. See "
1726
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1727
result = self._commit_builder_class(self, parents, config,
1728
timestamp, timezone, committer, revprops, revision_id)
1729
self.start_write_group()
1732
@only_raises(errors.LockNotHeld, errors.LockBroken)
261
1733
def unlock(self):
1734
if (self.control_files._lock_count == 1 and
1735
self.control_files._lock_mode == 'w'):
1736
if self._write_group is not None:
1737
self.abort_write_group()
1738
self.control_files.unlock()
1739
raise errors.BzrError(
1740
'Must end write groups before releasing write locks.')
262
1741
self.control_files.unlock()
1742
if self.control_files._lock_count == 0:
1743
self._inventory_entry_cache.clear()
1744
for repo in self._fallback_repositories:
264
1747
@needs_read_lock
265
def clone(self, a_bzrdir, revision_id=None, basis=None):
1748
def clone(self, a_bzrdir, revision_id=None):
266
1749
"""Clone this repository into a_bzrdir using the current format.
268
1751
Currently no check is made that the format of this repository and
269
1752
the bzrdir format are compatible. FIXME RBC 20060201.
1754
:return: The newly created destination repository.
1756
# TODO: deprecate after 0.16; cloning this with all its settings is
1757
# probably not very useful -- mbp 20070423
1758
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1759
self.copy_content_into(dest_repo, revision_id)
1762
def start_write_group(self):
1763
"""Start a write group in the repository.
1765
Write groups are used by repositories which do not have a 1:1 mapping
1766
between file ids and backend store to manage the insertion of data from
1767
both fetch and commit operations.
1769
A write lock is required around the start_write_group/commit_write_group
1770
for the support of lock-requiring repository formats.
1772
One can only insert data into a repository inside a write group.
1776
if not self.is_write_locked():
1777
raise errors.NotWriteLocked(self)
1778
if self._write_group:
1779
raise errors.BzrError('already in a write group')
1780
self._start_write_group()
1781
# so we can detect unlock/relock - the write group is now entered.
1782
self._write_group = self.get_transaction()
1784
def _start_write_group(self):
1785
"""Template method for per-repository write group startup.
1787
This is called before the write group is considered to be
1792
def sprout(self, to_bzrdir, revision_id=None):
1793
"""Create a descendent repository for new development.
1795
Unlike clone, this does not copy the settings of the repository.
1797
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1798
dest_repo.fetch(self, revision_id=revision_id)
1801
def _create_sprouting_repo(self, a_bzrdir, shared):
271
1802
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
272
1803
# use target default format.
273
result = a_bzrdir.create_repository()
274
# FIXME RBC 20060209 split out the repository type to avoid this check ?
275
elif isinstance(a_bzrdir._format,
276
(bzrdir.BzrDirFormat4,
277
bzrdir.BzrDirFormat5,
278
bzrdir.BzrDirFormat6)):
279
result = a_bzrdir.open_repository()
1804
dest_repo = a_bzrdir.create_repository()
281
result = self._format.initialize(a_bzrdir, shared=self.is_shared())
282
self.copy_content_into(result, revision_id, basis)
1806
# Most control formats need the repository to be specifically
1807
# created, but on some old all-in-one formats it's not needed
1809
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1810
except errors.UninitializableFormat:
1811
dest_repo = a_bzrdir.open_repository()
1814
def _get_sink(self):
1815
"""Return a sink for streaming into this repository."""
1816
return StreamSink(self)
1818
def _get_source(self, to_format):
1819
"""Return a source for streaming from this repository."""
1820
return StreamSource(self, to_format)
285
1822
@needs_read_lock
286
1823
def has_revision(self, revision_id):
287
1824
"""True if this repository has a copy of the revision."""
288
return self._revision_store.has_revision_id(revision_id,
289
self.get_transaction())
1825
return revision_id in self.has_revisions((revision_id,))
1828
def has_revisions(self, revision_ids):
1829
"""Probe to find out the presence of multiple revisions.
1831
:param revision_ids: An iterable of revision_ids.
1832
:return: A set of the revision_ids that were present.
1834
parent_map = self.revisions.get_parent_map(
1835
[(rev_id,) for rev_id in revision_ids])
1837
if _mod_revision.NULL_REVISION in revision_ids:
1838
result.add(_mod_revision.NULL_REVISION)
1839
result.update([key[0] for key in parent_map])
1843
def get_revision(self, revision_id):
1844
"""Return the Revision object for a named revision."""
1845
return self.get_revisions([revision_id])[0]
291
1847
@needs_read_lock
292
1848
def get_revision_reconcile(self, revision_id):
293
1849
"""'reconcile' helper routine that allows access to a revision always.
295
1851
This variant of get_revision does not cross check the weave graph
296
1852
against the revision one as get_revision does: but it should only
297
1853
be used by reconcile, or reconcile-alike commands that are correcting
298
1854
or testing the revision graph.
300
if not revision_id or not isinstance(revision_id, basestring):
301
raise InvalidRevisionId(revision_id=revision_id, branch=self)
302
return self._revision_store.get_revisions([revision_id],
303
self.get_transaction())[0]
1856
return self._get_revisions([revision_id])[0]
304
1858
@needs_read_lock
305
1859
def get_revisions(self, revision_ids):
306
return self._revision_store.get_revisions(revision_ids,
307
self.get_transaction())
1860
"""Get many revisions at once.
1862
Repositories that need to check data on every revision read should
1863
subclass this method.
1865
return self._get_revisions(revision_ids)
1868
def _get_revisions(self, revision_ids):
1869
"""Core work logic to get many revisions without sanity checks."""
1871
for revid, rev in self._iter_revisions(revision_ids):
1873
raise errors.NoSuchRevision(self, revid)
1875
return [revs[revid] for revid in revision_ids]
1877
def _iter_revisions(self, revision_ids):
1878
"""Iterate over revision objects.
1880
:param revision_ids: An iterable of revisions to examine. None may be
1881
passed to request all revisions known to the repository. Note that
1882
not all repositories can find unreferenced revisions; for those
1883
repositories only referenced ones will be returned.
1884
:return: An iterator of (revid, revision) tuples. Absent revisions (
1885
those asked for but not available) are returned as (revid, None).
1887
if revision_ids is None:
1888
revision_ids = self.all_revision_ids()
1890
for rev_id in revision_ids:
1891
if not rev_id or not isinstance(rev_id, basestring):
1892
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1893
keys = [(key,) for key in revision_ids]
1894
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1895
for record in stream:
1896
revid = record.key[0]
1897
if record.storage_kind == 'absent':
1900
text = record.get_bytes_as('fulltext')
1901
rev = self._serializer.read_revision_from_string(text)
309
1904
@needs_read_lock
310
1905
def get_revision_xml(self, revision_id):
311
rev = self.get_revision(revision_id)
1906
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1907
# would have already do it.
1908
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1909
# TODO: this can't just be replaced by:
1910
# return self._serializer.write_revision_to_string(
1911
# self.get_revision(revision_id))
1912
# as cStringIO preservers the encoding unlike write_revision_to_string
1913
# or some other call down the path.
1914
rev = self.get_revision(revision_id)
1915
rev_tmp = cStringIO.StringIO()
313
1916
# the current serializer..
314
self._revision_store._serializer.write_revision(rev, rev_tmp)
1917
self._serializer.write_revision(rev, rev_tmp)
316
1919
return rev_tmp.getvalue()
319
def get_revision(self, revision_id):
320
"""Return the Revision object for a named revision"""
321
r = self.get_revision_reconcile(revision_id)
322
# weave corruption can lead to absent revision markers that should be
324
# the following test is reasonably cheap (it needs a single weave read)
325
# and the weave is cached in read transactions. In write transactions
326
# it is not cached but typically we only read a small number of
327
# revisions. For knits when they are introduced we will probably want
328
# to ensure that caching write transactions are in use.
329
inv = self.get_inventory_weave()
330
self._check_revision_parents(r, inv)
334
def get_deltas_for_revisions(self, revisions):
1921
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
335
1922
"""Produce a generator of revision deltas.
337
1924
Note that the input is a sequence of REVISIONS, not revision_ids.
338
1925
Trees will be held in memory until the generator exits.
339
1926
Each delta is relative to the revision's lefthand predecessor.
1928
:param specific_fileids: if not None, the result is filtered
1929
so that only those file-ids, their parents and their
1930
children are included.
1932
# Get the revision-ids of interest
341
1933
required_trees = set()
342
1934
for revision in revisions:
343
1935
required_trees.add(revision.revision_id)
344
1936
required_trees.update(revision.parent_ids[:1])
345
trees = dict((t.get_revision_id(), t) for
346
t in self.revision_trees(required_trees))
1938
# Get the matching filtered trees. Note that it's more
1939
# efficient to pass filtered trees to changes_from() rather
1940
# than doing the filtering afterwards. changes_from() could
1941
# arguably do the filtering itself but it's path-based, not
1942
# file-id based, so filtering before or afterwards is
1944
if specific_fileids is None:
1945
trees = dict((t.get_revision_id(), t) for
1946
t in self.revision_trees(required_trees))
1948
trees = dict((t.get_revision_id(), t) for
1949
t in self._filtered_revision_trees(required_trees,
1952
# Calculate the deltas
347
1953
for revision in revisions:
348
1954
if not revision.parent_ids:
349
old_tree = self.revision_tree(None)
1955
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
351
1957
old_tree = trees[revision.parent_ids[0]]
352
1958
yield trees[revision.revision_id].changes_from(old_tree)
354
1960
@needs_read_lock
355
def get_revision_delta(self, revision_id):
1961
def get_revision_delta(self, revision_id, specific_fileids=None):
356
1962
"""Return the delta for one revision.
358
1964
The delta is relative to the left-hand predecessor of the
1967
:param specific_fileids: if not None, the result is filtered
1968
so that only those file-ids, their parents and their
1969
children are included.
361
1971
r = self.get_revision(revision_id)
362
return list(self.get_deltas_for_revisions([r]))[0]
364
def _check_revision_parents(self, revision, inventory):
365
"""Private to Repository and Fetch.
367
This checks the parentage of revision in an inventory weave for
368
consistency and is only applicable to inventory-weave-for-ancestry
369
using repository formats & fetchers.
371
weave_parents = inventory.get_parents(revision.revision_id)
372
weave_names = inventory.versions()
373
for parent_id in revision.parent_ids:
374
if parent_id in weave_names:
375
# this parent must not be a ghost.
376
if not parent_id in weave_parents:
378
raise errors.CorruptRepository(self)
1972
return list(self.get_deltas_for_revisions([r],
1973
specific_fileids=specific_fileids))[0]
380
1975
@needs_write_lock
381
1976
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
382
1977
signature = gpg_strategy.sign(plaintext)
383
self._revision_store.add_revision_signature_text(revision_id,
385
self.get_transaction())
387
def fileids_altered_by_revision_ids(self, revision_ids):
388
"""Find the file ids and versions affected by revisions.
390
:param revisions: an iterable containing revision ids.
391
:return: a dictionary mapping altered file-ids to an iterable of
392
revision_ids. Each altered file-ids has the exact revision_ids that
393
altered it listed explicitly.
395
assert isinstance(self._format, (RepositoryFormat5,
398
RepositoryFormatKnit1)), \
399
("fileids_altered_by_revision_ids only supported for branches "
400
"which store inventory as unnested xml, not on %r" % self)
401
selected_revision_ids = set(revision_ids)
402
w = self.get_inventory_weave()
1978
self.add_signature_text(revision_id, signature)
1981
def add_signature_text(self, revision_id, signature):
1982
self.signatures.add_lines((revision_id,), (),
1983
osutils.split_lines(signature))
1985
def find_text_key_references(self):
1986
"""Find the text key references within the repository.
1988
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1989
to whether they were referred to by the inventory of the
1990
revision_id that they contain. The inventory texts from all present
1991
revision ids are assessed to generate this report.
1993
revision_keys = self.revisions.keys()
1994
w = self.inventories
1995
pb = ui.ui_factory.nested_progress_bar()
1997
return self._find_text_key_references_from_xml_inventory_lines(
1998
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2002
def _find_text_key_references_from_xml_inventory_lines(self,
2004
"""Core routine for extracting references to texts from inventories.
2006
This performs the translation of xml lines to revision ids.
2008
:param line_iterator: An iterator of lines, origin_version_id
2009
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2010
to whether they were referred to by the inventory of the
2011
revision_id that they contain. Note that if that revision_id was
2012
not part of the line_iterator's output then False will be given -
2013
even though it may actually refer to that key.
2015
if not self._serializer.support_altered_by_hack:
2016
raise AssertionError(
2017
"_find_text_key_references_from_xml_inventory_lines only "
2018
"supported for branches which store inventory as unnested xml"
2019
", not on %r" % self)
405
2022
# this code needs to read every new line in every inventory for the
406
2023
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
407
# not present in one of those inventories is unnecessary but not
2024
# not present in one of those inventories is unnecessary but not
408
2025
# harmful because we are filtering by the revision id marker in the
409
# inventory lines : we only select file ids altered in one of those
2026
# inventory lines : we only select file ids altered in one of those
410
2027
# revisions. We don't need to see all lines in the inventory because
411
2028
# only those added in an inventory in rev X can contain a revision=X
413
for line in w.iter_lines_added_or_present_in_versions(selected_revision_ids):
414
start = line.find('file_id="')+9
415
if start < 9: continue
416
end = line.find('"', start)
418
file_id = _unescape_xml(line[start:end])
420
start = line.find('revision="')+10
421
if start < 10: continue
422
end = line.find('"', start)
424
revision_id = _unescape_xml(line[start:end])
425
if revision_id in selected_revision_ids:
426
result.setdefault(file_id, set()).add(revision_id)
430
def get_inventory_weave(self):
431
return self.control_weaves.get_weave('inventory',
432
self.get_transaction())
2030
unescape_revid_cache = {}
2031
unescape_fileid_cache = {}
2033
# jam 20061218 In a big fetch, this handles hundreds of thousands
2034
# of lines, so it has had a lot of inlining and optimizing done.
2035
# Sorry that it is a little bit messy.
2036
# Move several functions to be local variables, since this is a long
2038
search = self._file_ids_altered_regex.search
2039
unescape = _unescape_xml
2040
setdefault = result.setdefault
2041
for line, line_key in line_iterator:
2042
match = search(line)
2045
# One call to match.group() returning multiple items is quite a
2046
# bit faster than 2 calls to match.group() each returning 1
2047
file_id, revision_id = match.group('file_id', 'revision_id')
2049
# Inlining the cache lookups helps a lot when you make 170,000
2050
# lines and 350k ids, versus 8.4 unique ids.
2051
# Using a cache helps in 2 ways:
2052
# 1) Avoids unnecessary decoding calls
2053
# 2) Re-uses cached strings, which helps in future set and
2055
# (2) is enough that removing encoding entirely along with
2056
# the cache (so we are using plain strings) results in no
2057
# performance improvement.
2059
revision_id = unescape_revid_cache[revision_id]
2061
unescaped = unescape(revision_id)
2062
unescape_revid_cache[revision_id] = unescaped
2063
revision_id = unescaped
2065
# Note that unconditionally unescaping means that we deserialise
2066
# every fileid, which for general 'pull' is not great, but we don't
2067
# really want to have some many fulltexts that this matters anyway.
2070
file_id = unescape_fileid_cache[file_id]
2072
unescaped = unescape(file_id)
2073
unescape_fileid_cache[file_id] = unescaped
2076
key = (file_id, revision_id)
2077
setdefault(key, False)
2078
if revision_id == line_key[-1]:
2082
def _inventory_xml_lines_for_keys(self, keys):
2083
"""Get a line iterator of the sort needed for findind references.
2085
Not relevant for non-xml inventory repositories.
2087
Ghosts in revision_keys are ignored.
2089
:param revision_keys: The revision keys for the inventories to inspect.
2090
:return: An iterator over (inventory line, revid) for the fulltexts of
2091
all of the xml inventories specified by revision_keys.
2093
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2094
for record in stream:
2095
if record.storage_kind != 'absent':
2096
chunks = record.get_bytes_as('chunked')
2097
revid = record.key[-1]
2098
lines = osutils.chunks_to_lines(chunks)
2102
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2104
"""Helper routine for fileids_altered_by_revision_ids.
2106
This performs the translation of xml lines to revision ids.
2108
:param line_iterator: An iterator of lines, origin_version_id
2109
:param revision_keys: The revision ids to filter for. This should be a
2110
set or other type which supports efficient __contains__ lookups, as
2111
the revision key from each parsed line will be looked up in the
2112
revision_keys filter.
2113
:return: a dictionary mapping altered file-ids to an iterable of
2114
revision_ids. Each altered file-ids has the exact revision_ids that
2115
altered it listed explicitly.
2117
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2118
line_iterator).iterkeys())
2119
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2120
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2121
self._inventory_xml_lines_for_keys(parent_keys)))
2122
new_keys = seen - parent_seen
2124
setdefault = result.setdefault
2125
for key in new_keys:
2126
setdefault(key[0], set()).add(key[-1])
2129
def _find_parent_ids_of_revisions(self, revision_ids):
2130
"""Find all parent ids that are mentioned in the revision graph.
2132
:return: set of revisions that are parents of revision_ids which are
2133
not part of revision_ids themselves
2135
parent_map = self.get_parent_map(revision_ids)
2137
map(parent_ids.update, parent_map.itervalues())
2138
parent_ids.difference_update(revision_ids)
2139
parent_ids.discard(_mod_revision.NULL_REVISION)
2142
def _find_parent_keys_of_revisions(self, revision_keys):
2143
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2145
:param revision_keys: An iterable of revision_keys.
2146
:return: The parents of all revision_keys that are not already in
2149
parent_map = self.revisions.get_parent_map(revision_keys)
2151
map(parent_keys.update, parent_map.itervalues())
2152
parent_keys.difference_update(revision_keys)
2153
parent_keys.discard(_mod_revision.NULL_REVISION)
2156
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2157
"""Find the file ids and versions affected by revisions.
2159
:param revisions: an iterable containing revision ids.
2160
:param _inv_weave: The inventory weave from this repository or None.
2161
If None, the inventory weave will be opened automatically.
2162
:return: a dictionary mapping altered file-ids to an iterable of
2163
revision_ids. Each altered file-ids has the exact revision_ids that
2164
altered it listed explicitly.
2166
selected_keys = set((revid,) for revid in revision_ids)
2167
w = _inv_weave or self.inventories
2168
pb = ui.ui_factory.nested_progress_bar()
2170
return self._find_file_ids_from_xml_inventory_lines(
2171
w.iter_lines_added_or_present_in_keys(
2172
selected_keys, pb=pb),
2177
def iter_files_bytes(self, desired_files):
2178
"""Iterate through file versions.
2180
Files will not necessarily be returned in the order they occur in
2181
desired_files. No specific order is guaranteed.
2183
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2184
value supplied by the caller as part of desired_files. It should
2185
uniquely identify the file version in the caller's context. (Examples:
2186
an index number or a TreeTransform trans_id.)
2188
bytes_iterator is an iterable of bytestrings for the file. The
2189
kind of iterable and length of the bytestrings are unspecified, but for
2190
this implementation, it is a list of bytes produced by
2191
VersionedFile.get_record_stream().
2193
:param desired_files: a list of (file_id, revision_id, identifier)
2197
for file_id, revision_id, callable_data in desired_files:
2198
text_keys[(file_id, revision_id)] = callable_data
2199
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2200
if record.storage_kind == 'absent':
2201
raise errors.RevisionNotPresent(record.key, self)
2202
yield text_keys[record.key], record.get_bytes_as('chunked')
2204
def _generate_text_key_index(self, text_key_references=None,
2206
"""Generate a new text key index for the repository.
2208
This is an expensive function that will take considerable time to run.
2210
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2211
list of parents, also text keys. When a given key has no parents,
2212
the parents list will be [NULL_REVISION].
2214
# All revisions, to find inventory parents.
2215
if ancestors is None:
2216
graph = self.get_graph()
2217
ancestors = graph.get_parent_map(self.all_revision_ids())
2218
if text_key_references is None:
2219
text_key_references = self.find_text_key_references()
2220
pb = ui.ui_factory.nested_progress_bar()
2222
return self._do_generate_text_key_index(ancestors,
2223
text_key_references, pb)
2227
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2228
"""Helper for _generate_text_key_index to avoid deep nesting."""
2229
revision_order = tsort.topo_sort(ancestors)
2230
invalid_keys = set()
2232
for revision_id in revision_order:
2233
revision_keys[revision_id] = set()
2234
text_count = len(text_key_references)
2235
# a cache of the text keys to allow reuse; costs a dict of all the
2236
# keys, but saves a 2-tuple for every child of a given key.
2238
for text_key, valid in text_key_references.iteritems():
2240
invalid_keys.add(text_key)
2242
revision_keys[text_key[1]].add(text_key)
2243
text_key_cache[text_key] = text_key
2244
del text_key_references
2246
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2247
NULL_REVISION = _mod_revision.NULL_REVISION
2248
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2249
# too small for large or very branchy trees. However, for 55K path
2250
# trees, it would be easy to use too much memory trivially. Ideally we
2251
# could gauge this by looking at available real memory etc, but this is
2252
# always a tricky proposition.
2253
inventory_cache = lru_cache.LRUCache(10)
2254
batch_size = 10 # should be ~150MB on a 55K path tree
2255
batch_count = len(revision_order) / batch_size + 1
2257
pb.update("Calculating text parents", processed_texts, text_count)
2258
for offset in xrange(batch_count):
2259
to_query = revision_order[offset * batch_size:(offset + 1) *
2263
for revision_id in to_query:
2264
parent_ids = ancestors[revision_id]
2265
for text_key in revision_keys[revision_id]:
2266
pb.update("Calculating text parents", processed_texts)
2267
processed_texts += 1
2268
candidate_parents = []
2269
for parent_id in parent_ids:
2270
parent_text_key = (text_key[0], parent_id)
2272
check_parent = parent_text_key not in \
2273
revision_keys[parent_id]
2275
# the parent parent_id is a ghost:
2276
check_parent = False
2277
# truncate the derived graph against this ghost.
2278
parent_text_key = None
2280
# look at the parent commit details inventories to
2281
# determine possible candidates in the per file graph.
2284
inv = inventory_cache[parent_id]
2286
inv = self.revision_tree(parent_id).inventory
2287
inventory_cache[parent_id] = inv
2289
parent_entry = inv[text_key[0]]
2290
except (KeyError, errors.NoSuchId):
2292
if parent_entry is not None:
2294
text_key[0], parent_entry.revision)
2296
parent_text_key = None
2297
if parent_text_key is not None:
2298
candidate_parents.append(
2299
text_key_cache[parent_text_key])
2300
parent_heads = text_graph.heads(candidate_parents)
2301
new_parents = list(parent_heads)
2302
new_parents.sort(key=lambda x:candidate_parents.index(x))
2303
if new_parents == []:
2304
new_parents = [NULL_REVISION]
2305
text_index[text_key] = new_parents
2307
for text_key in invalid_keys:
2308
text_index[text_key] = [NULL_REVISION]
2311
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2312
"""Get an iterable listing the keys of all the data introduced by a set
2315
The keys will be ordered so that the corresponding items can be safely
2316
fetched and inserted in that order.
2318
:returns: An iterable producing tuples of (knit-kind, file-id,
2319
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2320
'revisions'. file-id is None unless knit-kind is 'file'.
2322
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2325
for result in self._find_non_file_keys_to_fetch(revision_ids):
2328
def _find_file_keys_to_fetch(self, revision_ids, pb):
2329
# XXX: it's a bit weird to control the inventory weave caching in this
2330
# generator. Ideally the caching would be done in fetch.py I think. Or
2331
# maybe this generator should explicitly have the contract that it
2332
# should not be iterated until the previously yielded item has been
2334
inv_w = self.inventories
2336
# file ids that changed
2337
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2339
num_file_ids = len(file_ids)
2340
for file_id, altered_versions in file_ids.iteritems():
2342
pb.update("Fetch texts", count, num_file_ids)
2344
yield ("file", file_id, altered_versions)
2346
def _find_non_file_keys_to_fetch(self, revision_ids):
2348
yield ("inventory", None, revision_ids)
2351
# XXX: Note ATM no callers actually pay attention to this return
2352
# instead they just use the list of revision ids and ignore
2353
# missing sigs. Consider removing this work entirely
2354
revisions_with_signatures = set(self.signatures.get_parent_map(
2355
[(r,) for r in revision_ids]))
2356
revisions_with_signatures = set(
2357
[r for (r,) in revisions_with_signatures])
2358
revisions_with_signatures.intersection_update(revision_ids)
2359
yield ("signatures", None, revisions_with_signatures)
2362
yield ("revisions", None, revision_ids)
434
2364
@needs_read_lock
435
2365
def get_inventory(self, revision_id):
436
"""Get Inventory object by hash."""
437
return self.deserialise_inventory(
438
revision_id, self.get_inventory_xml(revision_id))
440
def deserialise_inventory(self, revision_id, xml):
441
"""Transform the xml into an inventory object.
2366
"""Get Inventory object by revision id."""
2367
return self.iter_inventories([revision_id]).next()
2369
def iter_inventories(self, revision_ids, ordering=None):
2370
"""Get many inventories by revision_ids.
2372
This will buffer some or all of the texts used in constructing the
2373
inventories in memory, but will only parse a single inventory at a
2376
:param revision_ids: The expected revision ids of the inventories.
2377
:param ordering: optional ordering, e.g. 'topological'. If not
2378
specified, the order of revision_ids will be preserved (by
2379
buffering if necessary).
2380
:return: An iterator of inventories.
2382
if ((None in revision_ids)
2383
or (_mod_revision.NULL_REVISION in revision_ids)):
2384
raise ValueError('cannot get null revision inventory')
2385
return self._iter_inventories(revision_ids, ordering)
2387
def _iter_inventories(self, revision_ids, ordering):
2388
"""single-document based inventory iteration."""
2389
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2390
for text, revision_id in inv_xmls:
2391
yield self._deserialise_inventory(revision_id, text)
2393
def _iter_inventory_xmls(self, revision_ids, ordering):
2394
if ordering is None:
2395
order_as_requested = True
2396
ordering = 'unordered'
2398
order_as_requested = False
2399
keys = [(revision_id,) for revision_id in revision_ids]
2402
if order_as_requested:
2403
key_iter = iter(keys)
2404
next_key = key_iter.next()
2405
stream = self.inventories.get_record_stream(keys, ordering, True)
2407
for record in stream:
2408
if record.storage_kind != 'absent':
2409
chunks = record.get_bytes_as('chunked')
2410
if order_as_requested:
2411
text_chunks[record.key] = chunks
2413
yield ''.join(chunks), record.key[-1]
2415
raise errors.NoSuchRevision(self, record.key)
2416
if order_as_requested:
2417
# Yield as many results as we can while preserving order.
2418
while next_key in text_chunks:
2419
chunks = text_chunks.pop(next_key)
2420
yield ''.join(chunks), next_key[-1]
2422
next_key = key_iter.next()
2423
except StopIteration:
2424
# We still want to fully consume the get_record_stream,
2425
# just in case it is not actually finished at this point
2429
def _deserialise_inventory(self, revision_id, xml):
2430
"""Transform the xml into an inventory object.
443
2432
:param revision_id: The expected revision id of the inventory.
444
2433
:param xml: A serialised inventory.
446
return xml5.serializer_v5.read_inventory_from_string(xml)
2435
result = self._serializer.read_inventory_from_string(xml, revision_id,
2436
entry_cache=self._inventory_entry_cache,
2437
return_from_cache=self._safe_to_return_from_cache)
2438
if result.revision_id != revision_id:
2439
raise AssertionError('revision id mismatch %s != %s' % (
2440
result.revision_id, revision_id))
2443
def _serialise_inventory(self, inv):
2444
return self._serializer.write_inventory_to_string(inv)
2446
def _serialise_inventory_to_lines(self, inv):
2447
return self._serializer.write_inventory_to_lines(inv)
2449
def get_serializer_format(self):
2450
return self._serializer.format_num
448
2452
@needs_read_lock
449
def get_inventory_xml(self, revision_id):
450
"""Get inventory XML as a file object."""
2453
def _get_inventory_xml(self, revision_id):
2454
"""Get serialized inventory as a string."""
2455
texts = self._iter_inventory_xmls([revision_id], 'unordered')
452
assert isinstance(revision_id, basestring), type(revision_id)
453
iw = self.get_inventory_weave()
454
return iw.get_text(revision_id)
2457
text, revision_id = texts.next()
2458
except StopIteration:
456
2459
raise errors.HistoryMissing(self, 'inventory', revision_id)
459
def get_inventory_sha1(self, revision_id):
460
"""Return the sha1 hash of the inventory entry
462
return self.get_revision(revision_id).inventory_sha1
465
def get_revision_graph(self, revision_id=None):
466
"""Return a dictionary containing the revision graph.
468
:param revision_id: The revision_id to get a graph from. If None, then
469
the entire revision graph is returned. This is a deprecated mode of
470
operation and will be removed in the future.
471
:return: a dictionary of revision_id->revision_parents_list.
473
# special case NULL_REVISION
474
if revision_id == NULL_REVISION:
476
weave = self.get_inventory_weave()
477
all_revisions = self._eliminate_revisions_not_present(weave.versions())
478
entire_graph = dict([(node, weave.get_parents(node)) for
479
node in all_revisions])
480
if revision_id is None:
482
elif revision_id not in entire_graph:
483
raise errors.NoSuchRevision(self, revision_id)
485
# add what can be reached from revision_id
487
pending = set([revision_id])
488
while len(pending) > 0:
490
result[node] = entire_graph[node]
491
for revision_id in result[node]:
492
if revision_id not in result:
493
pending.add(revision_id)
497
def get_revision_graph_with_ghosts(self, revision_ids=None):
498
"""Return a graph of the revisions with ghosts marked as applicable.
500
:param revision_ids: an iterable of revisions to graph or None for all.
501
:return: a Graph object with the graph reachable from revision_ids.
505
pending = set(self.all_revision_ids())
508
pending = set(revision_ids)
509
# special case NULL_REVISION
510
if NULL_REVISION in pending:
511
pending.remove(NULL_REVISION)
512
required = set(pending)
515
revision_id = pending.pop()
2462
def get_rev_id_for_revno(self, revno, known_pair):
2463
"""Return the revision id of a revno, given a later (revno, revid)
2464
pair in the same history.
2466
:return: if found (True, revid). If the available history ran out
2467
before reaching the revno, then this returns
2468
(False, (closest_revno, closest_revid)).
2470
known_revno, known_revid = known_pair
2471
partial_history = [known_revid]
2472
distance_from_known = known_revno - revno
2473
if distance_from_known < 0:
2475
'requested revno (%d) is later than given known revno (%d)'
2476
% (revno, known_revno))
2479
self, partial_history, stop_index=distance_from_known)
2480
except errors.RevisionNotPresent, err:
2481
if err.revision_id == known_revid:
2482
# The start revision (known_revid) wasn't found.
2484
# This is a stacked repository with no fallbacks, or a there's a
2485
# left-hand ghost. Either way, even though the revision named in
2486
# the error isn't in this repo, we know it's the next step in this
2487
# left-hand history.
2488
partial_history.append(err.revision_id)
2489
if len(partial_history) <= distance_from_known:
2490
# Didn't find enough history to get a revid for the revno.
2491
earliest_revno = known_revno - len(partial_history) + 1
2492
return (False, (earliest_revno, partial_history[-1]))
2493
if len(partial_history) - 1 > distance_from_known:
2494
raise AssertionError('_iter_for_revno returned too much history')
2495
return (True, partial_history[-1])
2497
def iter_reverse_revision_history(self, revision_id):
2498
"""Iterate backwards through revision ids in the lefthand history
2500
:param revision_id: The revision id to start with. All its lefthand
2501
ancestors will be traversed.
2503
graph = self.get_graph()
2504
next_id = revision_id
2506
if next_id in (None, _mod_revision.NULL_REVISION):
517
rev = self.get_revision(revision_id)
518
except errors.NoSuchRevision:
519
if revision_id in required:
522
result.add_ghost(revision_id)
524
for parent_id in rev.parent_ids:
525
# is this queued or done ?
526
if (parent_id not in pending and
527
parent_id not in done):
529
pending.add(parent_id)
530
result.add_node(revision_id, rev.parent_ids)
531
done.add(revision_id)
2509
parents = graph.get_parent_map([next_id])[next_id]
2511
raise errors.RevisionNotPresent(next_id, self)
2513
if len(parents) == 0:
2516
next_id = parents[0]
534
2518
@needs_read_lock
535
2519
def get_revision_inventory(self, revision_id):
816
self.control_files._transport.delete('no-working-trees')
2977
self._transport.delete('no-working-trees')
817
2978
except errors.NoSuchFile:
820
self.control_files.put_utf8('no-working-trees', '')
2981
self._transport.put_bytes('no-working-trees', '',
2982
mode=self.bzrdir._get_file_mode())
822
2984
def make_working_trees(self):
823
2985
"""Returns the policy for making working trees on new branches."""
824
return not self.control_files._transport.has('no-working-trees')
827
class KnitRepository(MetaDirRepository):
828
"""Knit format repository."""
830
def _inventory_add_lines(self, inv_vf, revid, parents, lines):
831
inv_vf.add_lines_with_ghosts(revid, parents, lines)
834
def _all_revision_ids(self):
835
"""See Repository.all_revision_ids()."""
836
# Knits get the revision graph from the index of the revision knit, so
837
# it's always possible even if they're on an unlistable transport.
838
return self._revision_store.all_revision_ids(self.get_transaction())
840
def fileid_involved_between_revs(self, from_revid, to_revid):
841
"""Find file_id(s) which are involved in the changes between revisions.
843
This determines the set of revisions which are involved, and then
844
finds all file ids affected by those revisions.
846
vf = self._get_revision_vf()
847
from_set = set(vf.get_ancestry(from_revid))
848
to_set = set(vf.get_ancestry(to_revid))
849
changed = to_set.difference(from_set)
850
return self._fileid_involved_by_set(changed)
852
def fileid_involved(self, last_revid=None):
853
"""Find all file_ids modified in the ancestry of last_revid.
855
:param last_revid: If None, last_revision() will be used.
858
changed = set(self.all_revision_ids())
860
changed = set(self.get_ancestry(last_revid))
863
return self._fileid_involved_by_set(changed)
866
def get_ancestry(self, revision_id):
867
"""Return a list of revision-ids integrated by a revision.
869
This is topologically sorted.
871
if revision_id is None:
873
vf = self._get_revision_vf()
875
return [None] + vf.get_ancestry(revision_id)
876
except errors.RevisionNotPresent:
877
raise errors.NoSuchRevision(self, revision_id)
880
def get_revision(self, revision_id):
881
"""Return the Revision object for a named revision"""
882
return self.get_revision_reconcile(revision_id)
885
def get_revision_graph(self, revision_id=None):
886
"""Return a dictionary containing the revision graph.
888
:param revision_id: The revision_id to get a graph from. If None, then
889
the entire revision graph is returned. This is a deprecated mode of
890
operation and will be removed in the future.
891
:return: a dictionary of revision_id->revision_parents_list.
893
# special case NULL_REVISION
894
if revision_id == NULL_REVISION:
896
weave = self._get_revision_vf()
897
entire_graph = weave.get_graph()
898
if revision_id is None:
899
return weave.get_graph()
900
elif revision_id not in weave:
901
raise errors.NoSuchRevision(self, revision_id)
903
# add what can be reached from revision_id
905
pending = set([revision_id])
906
while len(pending) > 0:
908
result[node] = weave.get_parents(node)
909
for revision_id in result[node]:
910
if revision_id not in result:
911
pending.add(revision_id)
915
def get_revision_graph_with_ghosts(self, revision_ids=None):
916
"""Return a graph of the revisions with ghosts marked as applicable.
918
:param revision_ids: an iterable of revisions to graph or None for all.
919
:return: a Graph object with the graph reachable from revision_ids.
922
vf = self._get_revision_vf()
923
versions = set(vf.versions())
925
pending = set(self.all_revision_ids())
928
pending = set(revision_ids)
929
# special case NULL_REVISION
930
if NULL_REVISION in pending:
931
pending.remove(NULL_REVISION)
932
required = set(pending)
935
revision_id = pending.pop()
936
if not revision_id in versions:
937
if revision_id in required:
938
raise errors.NoSuchRevision(self, revision_id)
940
result.add_ghost(revision_id)
941
# mark it as done so we don't try for it again.
942
done.add(revision_id)
944
parent_ids = vf.get_parents_with_ghosts(revision_id)
945
for parent_id in parent_ids:
946
# is this queued or done ?
947
if (parent_id not in pending and
948
parent_id not in done):
950
pending.add(parent_id)
951
result.add_node(revision_id, parent_ids)
952
done.add(revision_id)
955
def _get_revision_vf(self):
956
""":return: a versioned file containing the revisions."""
957
vf = self._revision_store.get_revision_file(self.get_transaction())
961
def reconcile(self, other=None, thorough=False):
962
"""Reconcile this repository."""
963
from bzrlib.reconcile import KnitReconciler
964
reconciler = KnitReconciler(self, thorough=thorough)
965
reconciler.reconcile()
968
def revision_parents(self, revision_id):
969
return self._get_revision_vf().get_parents(revision_id)
2986
return not self._transport.has('no-working-trees')
2989
class MetaDirVersionedFileRepository(MetaDirRepository):
2990
"""Repositories in a meta-dir, that work via versioned file objects."""
2992
def __init__(self, _format, a_bzrdir, control_files):
2993
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2997
network_format_registry = registry.FormatRegistry()
2998
"""Registry of formats indexed by their network name.
3000
The network name for a repository format is an identifier that can be used when
3001
referring to formats with smart server operations. See
3002
RepositoryFormat.network_name() for more detail.
3006
format_registry = registry.FormatRegistry(network_format_registry)
3007
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
3009
This can contain either format instances themselves, or classes/factories that
3010
can be called to obtain one.
3014
#####################################################################
3015
# Repository Formats
972
3017
class RepositoryFormat(object):
973
3018
"""A repository format.
975
Formats provide three things:
3020
Formats provide four things:
976
3021
* An initialization routine to construct repository data on disk.
977
* a format string which is used when the BzrDir supports versioned
3022
* a optional format string which is used when the BzrDir supports
979
3024
* an open routine which returns a Repository instance.
981
Formats are placed in an dict by their format string for reference
982
during opening. These should be subclasses of RepositoryFormat
3025
* A network name for referring to the format in smart server RPC
3028
There is one and only one Format subclass for each on-disk format. But
3029
there can be one Repository subclass that is used for several different
3030
formats. The _format attribute on a Repository instance can be used to
3031
determine the disk format.
3033
Formats are placed in a registry by their format string for reference
3034
during opening. These should be subclasses of RepositoryFormat for
985
3037
Once a format is deprecated, just deprecate the initialize and open
986
methods on the format class. Do not deprecate the object, as the
987
object will be created every system load.
3038
methods on the format class. Do not deprecate the object, as the
3039
object may be created even when a repository instance hasn't been
989
3042
Common instance attributes:
990
3043
_matchingbzrdir - the bzrdir format that the repository format was
991
3044
originally written to work with. This can be used if manually
992
3045
constructing a bzrdir and repository, or more commonly for test suite
996
_default_format = None
997
"""The default format used for new repositories."""
1000
"""The known formats."""
3049
# Set to True or False in derived classes. True indicates that the format
3050
# supports ghosts gracefully.
3051
supports_ghosts = None
3052
# Can this repository be given external locations to lookup additional
3053
# data. Set to True or False in derived classes.
3054
supports_external_lookups = None
3055
# Does this format support CHK bytestring lookups. Set to True or False in
3057
supports_chks = None
3058
# Should commit add an inventory, or an inventory delta to the repository.
3059
_commit_inv_deltas = True
3060
# What order should fetch operations request streams in?
3061
# The default is unordered as that is the cheapest for an origin to
3063
_fetch_order = 'unordered'
3064
# Does this repository format use deltas that can be fetched as-deltas ?
3065
# (E.g. knits, where the knit deltas can be transplanted intact.
3066
# We default to False, which will ensure that enough data to get
3067
# a full text out of any fetch stream will be grabbed.
3068
_fetch_uses_deltas = False
3069
# Should fetch trigger a reconcile after the fetch? Only needed for
3070
# some repository formats that can suffer internal inconsistencies.
3071
_fetch_reconcile = False
3072
# Does this format have < O(tree_size) delta generation. Used to hint what
3073
# code path for commit, amongst other things.
3075
# Does doing a pack operation compress data? Useful for the pack UI command
3076
# (so if there is one pack, the operation can still proceed because it may
3077
# help), and for fetching when data won't have come from the same
3079
pack_compresses = False
3080
# Does the repository inventory storage understand references to trees?
3081
supports_tree_reference = None
3084
return "<%s>" % self.__class__.__name__
3086
def __eq__(self, other):
3087
# format objects are generally stateless
3088
return isinstance(other, self.__class__)
3090
def __ne__(self, other):
3091
return not self == other
1003
3094
def find_format(klass, a_bzrdir):
1004
"""Return the format for the repository object in a_bzrdir."""
3095
"""Return the format for the repository object in a_bzrdir.
3097
This is used by bzr native formats that have a "format" file in
3098
the repository. Other methods may be used by different types of
1006
3102
transport = a_bzrdir.get_repository_transport(None)
1007
format_string = transport.get("format").read()
1008
return klass._formats[format_string]
3103
format_string = transport.get_bytes("format")
3104
return format_registry.get(format_string)
1009
3105
except errors.NoSuchFile:
1010
3106
raise errors.NoRepositoryPresent(a_bzrdir)
1011
3107
except KeyError:
1012
raise errors.UnknownFormatError(format=format_string)
1014
def _get_control_store(self, repo_transport, control_files):
1015
"""Return the control store for this repository."""
1016
raise NotImplementedError(self._get_control_store)
3108
raise errors.UnknownFormatError(format=format_string,
3112
def register_format(klass, format):
3113
format_registry.register(format.get_format_string(), format)
3116
def unregister_format(klass, format):
3117
format_registry.remove(format.get_format_string())
1019
3120
def get_default_format(klass):
1020
3121
"""Return the current default format."""
1021
return klass._default_format
3122
from bzrlib import bzrdir
3123
return bzrdir.format_registry.make_bzrdir('default').repository_format
1023
3125
def get_format_string(self):
1024
3126
"""Return the ASCII format string that identifies this format.
1026
Note that in pre format ?? repositories the format string is
3128
Note that in pre format ?? repositories the format string is
1027
3129
not permitted nor written to disk.
1029
3131
raise NotImplementedError(self.get_format_string)
1081
3162
:param a_bzrdir: The bzrdir to put the new repository in it.
1082
3163
:param shared: The repository should be initialized as a sharable one.
3164
:returns: The new repository object.
1084
3166
This may raise UninitializableFormat if shared repository are not
1085
3167
compatible the a_bzrdir.
3169
raise NotImplementedError(self.initialize)
1088
3171
def is_supported(self):
1089
3172
"""Is this format supported?
1091
3174
Supported formats must be initializable and openable.
1092
Unsupported formats may not support initialization or committing or
3175
Unsupported formats may not support initialization or committing or
1093
3176
some other features depending on the reason for not being supported.
3180
def network_name(self):
3181
"""A simple byte string uniquely identifying this format for RPC calls.
3183
MetaDir repository formats use their disk format string to identify the
3184
repository over the wire. All in one formats such as bzr < 0.8, and
3185
foreign formats like svn/git and hg should use some marker which is
3186
unique and immutable.
3188
raise NotImplementedError(self.network_name)
3190
def check_conversion_target(self, target_format):
3191
if self.rich_root_data and not target_format.rich_root_data:
3192
raise errors.BadConversionTarget(
3193
'Does not support rich root data.', target_format,
3195
if (self.supports_tree_reference and
3196
not getattr(target_format, 'supports_tree_reference', False)):
3197
raise errors.BadConversionTarget(
3198
'Does not support nested trees', target_format,
1097
3201
def open(self, a_bzrdir, _found=False):
1098
3202
"""Return an instance of this format for the bzrdir a_bzrdir.
1100
3204
_found is a private parameter, do not use it.
1102
3206
raise NotImplementedError(self.open)
1105
def register_format(klass, format):
1106
klass._formats[format.get_format_string()] = format
1109
def set_default_format(klass, format):
1110
klass._default_format = format
1113
def unregister_format(klass, format):
1114
assert klass._formats[format.get_format_string()] is format
1115
del klass._formats[format.get_format_string()]
1118
class PreSplitOutRepositoryFormat(RepositoryFormat):
1119
"""Base class for the pre split out repository formats."""
1121
def initialize(self, a_bzrdir, shared=False, _internal=False):
1122
"""Create a weave repository.
1124
TODO: when creating split out bzr branch formats, move this to a common
1125
base for Format5, Format6. or something like that.
1127
from bzrlib.weavefile import write_weave_v5
1128
from bzrlib.weave import Weave
1131
raise errors.IncompatibleFormat(self, a_bzrdir._format)
1134
# always initialized when the bzrdir is.
1135
return self.open(a_bzrdir, _found=True)
1137
# Create an empty weave
1139
write_weave_v5(Weave(), sio)
1140
empty_weave = sio.getvalue()
1142
mutter('creating repository in %s.', a_bzrdir.transport.base)
1143
dirs = ['revision-store', 'weaves']
1144
files = [('inventory.weave', StringIO(empty_weave)),
1147
# FIXME: RBC 20060125 don't peek under the covers
1148
# NB: no need to escape relative paths that are url safe.
1149
control_files = LockableFiles(a_bzrdir.transport, 'branch-lock',
1151
control_files.create_lock()
1152
control_files.lock_write()
1153
control_files._transport.mkdir_multi(dirs,
1154
mode=control_files._dir_mode)
1156
for file, content in files:
1157
control_files.put(file, content)
1159
control_files.unlock()
1160
return self.open(a_bzrdir, _found=True)
1162
def _get_control_store(self, repo_transport, control_files):
1163
"""Return the control store for this repository."""
1164
return self._get_versioned_file_store('',
1169
def _get_text_store(self, transport, control_files):
1170
"""Get a store for file texts for this format."""
1171
raise NotImplementedError(self._get_text_store)
1173
def open(self, a_bzrdir, _found=False):
1174
"""See RepositoryFormat.open()."""
1176
# we are being called directly and must probe.
1177
raise NotImplementedError
1179
repo_transport = a_bzrdir.get_repository_transport(None)
1180
control_files = a_bzrdir._control_files
1181
text_store = self._get_text_store(repo_transport, control_files)
1182
control_store = self._get_control_store(repo_transport, control_files)
1183
_revision_store = self._get_revision_store(repo_transport, control_files)
1184
return AllInOneRepository(_format=self,
1186
_revision_store=_revision_store,
1187
control_store=control_store,
1188
text_store=text_store)
1191
class RepositoryFormat4(PreSplitOutRepositoryFormat):
1192
"""Bzr repository format 4.
1194
This repository format has:
1196
- TextStores for texts, inventories,revisions.
1198
This format is deprecated: it indexes texts using a text id which is
1199
removed in format 5; initialization and write support for this format
1204
super(RepositoryFormat4, self).__init__()
1205
self._matchingbzrdir = bzrdir.BzrDirFormat4()
1207
def get_format_description(self):
1208
"""See RepositoryFormat.get_format_description()."""
1209
return "Repository format 4"
1211
def initialize(self, url, shared=False, _internal=False):
1212
"""Format 4 branches cannot be created."""
1213
raise errors.UninitializableFormat(self)
1215
def is_supported(self):
1216
"""Format 4 is not supported.
1218
It is not supported because the model changed from 4 to 5 and the
1219
conversion logic is expensive - so doing it on the fly was not
1224
def _get_control_store(self, repo_transport, control_files):
1225
"""Format 4 repositories have no formal control store at this point.
1227
This will cause any control-file-needing apis to fail - this is desired.
1231
def _get_revision_store(self, repo_transport, control_files):
1232
"""See RepositoryFormat._get_revision_store()."""
1233
from bzrlib.xml4 import serializer_v4
1234
return self._get_text_rev_store(repo_transport,
1237
serializer=serializer_v4)
1239
def _get_text_store(self, transport, control_files):
1240
"""See RepositoryFormat._get_text_store()."""
1243
class RepositoryFormat5(PreSplitOutRepositoryFormat):
1244
"""Bzr control format 5.
1246
This repository format has:
1247
- weaves for file texts and inventory
1249
- TextStores for revisions and signatures.
1253
super(RepositoryFormat5, self).__init__()
1254
self._matchingbzrdir = bzrdir.BzrDirFormat5()
1256
def get_format_description(self):
1257
"""See RepositoryFormat.get_format_description()."""
1258
return "Weave repository format 5"
1260
def _get_revision_store(self, repo_transport, control_files):
1261
"""See RepositoryFormat._get_revision_store()."""
1262
"""Return the revision store object for this a_bzrdir."""
1263
return self._get_text_rev_store(repo_transport,
1268
def _get_text_store(self, transport, control_files):
1269
"""See RepositoryFormat._get_text_store()."""
1270
return self._get_versioned_file_store('weaves', transport, control_files, prefixed=False)
1273
class RepositoryFormat6(PreSplitOutRepositoryFormat):
1274
"""Bzr control format 6.
1276
This repository format has:
1277
- weaves for file texts and inventory
1278
- hash subdirectory based stores.
1279
- TextStores for revisions and signatures.
1283
super(RepositoryFormat6, self).__init__()
1284
self._matchingbzrdir = bzrdir.BzrDirFormat6()
1286
def get_format_description(self):
1287
"""See RepositoryFormat.get_format_description()."""
1288
return "Weave repository format 6"
1290
def _get_revision_store(self, repo_transport, control_files):
1291
"""See RepositoryFormat._get_revision_store()."""
1292
return self._get_text_rev_store(repo_transport,
1298
def _get_text_store(self, transport, control_files):
1299
"""See RepositoryFormat._get_text_store()."""
1300
return self._get_versioned_file_store('weaves', transport, control_files)
1303
3209
class MetaDirRepositoryFormat(RepositoryFormat):
1304
3210
"""Common base class for the new repositories using the metadir layout."""
3212
rich_root_data = False
3213
supports_tree_reference = False
3214
supports_external_lookups = False
3217
def _matchingbzrdir(self):
3218
matching = bzrdir.BzrDirMetaFormat1()
3219
matching.repository_format = self
1306
3222
def __init__(self):
1307
3223
super(MetaDirRepositoryFormat, self).__init__()
1308
self._matchingbzrdir = bzrdir.BzrDirMetaFormat1()
1310
3225
def _create_control_files(self, a_bzrdir):
1311
3226
"""Create the required files and the initial control_files object."""
1312
3227
# FIXME: RBC 20060125 don't peek under the covers
1313
3228
# NB: no need to escape relative paths that are url safe.
1314
3229
repository_transport = a_bzrdir.get_repository_transport(self)
1315
control_files = LockableFiles(repository_transport, 'lock', LockDir)
3230
control_files = lockable_files.LockableFiles(repository_transport,
3231
'lock', lockdir.LockDir)
1316
3232
control_files.create_lock()
1317
3233
return control_files
1320
3236
"""Upload the initial blank content."""
1321
3237
control_files = self._create_control_files(a_bzrdir)
1322
3238
control_files.lock_write()
3239
transport = control_files._transport
3241
utf8_files += [('shared-storage', '')]
1324
control_files._transport.mkdir_multi(dirs,
1325
mode=control_files._dir_mode)
1326
for file, content in files:
1327
control_files.put(file, content)
1328
for file, content in utf8_files:
1329
control_files.put_utf8(file, content)
1331
control_files.put_utf8('shared-storage', '')
3243
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3244
for (filename, content_stream) in files:
3245
transport.put_file(filename, content_stream,
3246
mode=a_bzrdir._get_file_mode())
3247
for (filename, content_bytes) in utf8_files:
3248
transport.put_bytes_non_atomic(filename, content_bytes,
3249
mode=a_bzrdir._get_file_mode())
1333
3251
control_files.unlock()
1336
class RepositoryFormat7(MetaDirRepositoryFormat):
1337
"""Bzr repository 7.
1339
This repository format has:
1340
- weaves for file texts and inventory
1341
- hash subdirectory based stores.
1342
- TextStores for revisions and signatures.
1343
- a format marker of its own
1344
- an optional 'shared-storage' flag
1345
- an optional 'no-working-trees' flag
1348
def _get_control_store(self, repo_transport, control_files):
1349
"""Return the control store for this repository."""
1350
return self._get_versioned_file_store('',
1355
def get_format_string(self):
1356
"""See RepositoryFormat.get_format_string()."""
1357
return "Bazaar-NG Repository format 7"
1359
def get_format_description(self):
1360
"""See RepositoryFormat.get_format_description()."""
1361
return "Weave repository format 7"
1363
def _get_revision_store(self, repo_transport, control_files):
1364
"""See RepositoryFormat._get_revision_store()."""
1365
return self._get_text_rev_store(repo_transport,
1372
def _get_text_store(self, transport, control_files):
1373
"""See RepositoryFormat._get_text_store()."""
1374
return self._get_versioned_file_store('weaves',
1378
def initialize(self, a_bzrdir, shared=False):
1379
"""Create a weave repository.
1381
:param shared: If true the repository will be initialized as a shared
1384
from bzrlib.weavefile import write_weave_v5
1385
from bzrlib.weave import Weave
1387
# Create an empty weave
1389
write_weave_v5(Weave(), sio)
1390
empty_weave = sio.getvalue()
1392
mutter('creating repository in %s.', a_bzrdir.transport.base)
1393
dirs = ['revision-store', 'weaves']
1394
files = [('inventory.weave', StringIO(empty_weave)),
1396
utf8_files = [('format', self.get_format_string())]
1398
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1399
return self.open(a_bzrdir=a_bzrdir, _found=True)
1401
def open(self, a_bzrdir, _found=False, _override_transport=None):
1402
"""See RepositoryFormat.open().
1404
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1405
repository at a slightly different url
1406
than normal. I.e. during 'upgrade'.
1409
format = RepositoryFormat.find_format(a_bzrdir)
1410
assert format.__class__ == self.__class__
1411
if _override_transport is not None:
1412
repo_transport = _override_transport
1414
repo_transport = a_bzrdir.get_repository_transport(None)
1415
control_files = LockableFiles(repo_transport, 'lock', LockDir)
1416
text_store = self._get_text_store(repo_transport, control_files)
1417
control_store = self._get_control_store(repo_transport, control_files)
1418
_revision_store = self._get_revision_store(repo_transport, control_files)
1419
return MetaDirRepository(_format=self,
1421
control_files=control_files,
1422
_revision_store=_revision_store,
1423
control_store=control_store,
1424
text_store=text_store)
1427
class RepositoryFormatKnit1(MetaDirRepositoryFormat):
1428
"""Bzr repository knit format 1.
1430
This repository format has:
1431
- knits for file texts and inventory
1432
- hash subdirectory based stores.
1433
- knits for revisions and signatures
1434
- TextStores for revisions and signatures.
1435
- a format marker of its own
1436
- an optional 'shared-storage' flag
1437
- an optional 'no-working-trees' flag
1440
This format was introduced in bzr 0.8.
1443
def _get_control_store(self, repo_transport, control_files):
1444
"""Return the control store for this repository."""
1445
return VersionedFileStore(
1448
file_mode=control_files._file_mode,
1449
versionedfile_class=KnitVersionedFile,
1450
versionedfile_kwargs={'factory':KnitPlainFactory()},
1453
def get_format_string(self):
1454
"""See RepositoryFormat.get_format_string()."""
1455
return "Bazaar-NG Knit Repository Format 1"
1457
def get_format_description(self):
1458
"""See RepositoryFormat.get_format_description()."""
1459
return "Knit repository format 1"
1461
def _get_revision_store(self, repo_transport, control_files):
1462
"""See RepositoryFormat._get_revision_store()."""
1463
from bzrlib.store.revision.knit import KnitRevisionStore
1464
versioned_file_store = VersionedFileStore(
1466
file_mode=control_files._file_mode,
1469
versionedfile_class=KnitVersionedFile,
1470
versionedfile_kwargs={'delta':False, 'factory':KnitPlainFactory()},
1473
return KnitRevisionStore(versioned_file_store)
1475
def _get_text_store(self, transport, control_files):
1476
"""See RepositoryFormat._get_text_store()."""
1477
return self._get_versioned_file_store('knits',
1480
versionedfile_class=KnitVersionedFile,
1483
def initialize(self, a_bzrdir, shared=False):
1484
"""Create a knit format 1 repository.
1486
:param a_bzrdir: bzrdir to contain the new repository; must already
1488
:param shared: If true the repository will be initialized as a shared
1491
mutter('creating repository in %s.', a_bzrdir.transport.base)
1492
dirs = ['revision-store', 'knits']
1494
utf8_files = [('format', self.get_format_string())]
1496
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1497
repo_transport = a_bzrdir.get_repository_transport(None)
1498
control_files = LockableFiles(repo_transport, 'lock', LockDir)
1499
control_store = self._get_control_store(repo_transport, control_files)
1500
transaction = transactions.WriteTransaction()
1501
# trigger a write of the inventory store.
1502
control_store.get_weave_or_empty('inventory', transaction)
1503
_revision_store = self._get_revision_store(repo_transport, control_files)
1504
_revision_store.has_revision_id('A', transaction)
1505
_revision_store.get_signature_file(transaction)
1506
return self.open(a_bzrdir=a_bzrdir, _found=True)
1508
def open(self, a_bzrdir, _found=False, _override_transport=None):
1509
"""See RepositoryFormat.open().
1511
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1512
repository at a slightly different url
1513
than normal. I.e. during 'upgrade'.
1516
format = RepositoryFormat.find_format(a_bzrdir)
1517
assert format.__class__ == self.__class__
1518
if _override_transport is not None:
1519
repo_transport = _override_transport
1521
repo_transport = a_bzrdir.get_repository_transport(None)
1522
control_files = LockableFiles(repo_transport, 'lock', LockDir)
1523
text_store = self._get_text_store(repo_transport, control_files)
1524
control_store = self._get_control_store(repo_transport, control_files)
1525
_revision_store = self._get_revision_store(repo_transport, control_files)
1526
return KnitRepository(_format=self,
1528
control_files=control_files,
1529
_revision_store=_revision_store,
1530
control_store=control_store,
1531
text_store=text_store)
1534
# formats which have no format string are not discoverable
1535
# and not independently creatable, so are not registered.
1536
RepositoryFormat.register_format(RepositoryFormat7())
1537
_default_format = RepositoryFormatKnit1()
1538
RepositoryFormat.register_format(_default_format)
1539
RepositoryFormat.set_default_format(_default_format)
1540
_legacy_formats = [RepositoryFormat4(),
1541
RepositoryFormat5(),
1542
RepositoryFormat6()]
3253
def network_name(self):
3254
"""Metadir formats have matching disk and network format strings."""
3255
return self.get_format_string()
3258
# Pre-0.8 formats that don't have a disk format string (because they are
3259
# versioned by the matching control directory). We use the control directories
3260
# disk format string as a key for the network_name because they meet the
3261
# constraints (simple string, unique, immutable).
3262
network_format_registry.register_lazy(
3263
"Bazaar-NG branch, format 5\n",
3264
'bzrlib.repofmt.weaverepo',
3265
'RepositoryFormat5',
3267
network_format_registry.register_lazy(
3268
"Bazaar-NG branch, format 6\n",
3269
'bzrlib.repofmt.weaverepo',
3270
'RepositoryFormat6',
3273
# formats which have no format string are not discoverable or independently
3274
# creatable on disk, so are not registered in format_registry. They're
3275
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3276
# needed, it's constructed directly by the BzrDir. Non-native formats where
3277
# the repository is not separately opened are similar.
3279
format_registry.register_lazy(
3280
'Bazaar-NG Repository format 7',
3281
'bzrlib.repofmt.weaverepo',
3285
format_registry.register_lazy(
3286
'Bazaar-NG Knit Repository Format 1',
3287
'bzrlib.repofmt.knitrepo',
3288
'RepositoryFormatKnit1',
3291
format_registry.register_lazy(
3292
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3293
'bzrlib.repofmt.knitrepo',
3294
'RepositoryFormatKnit3',
3297
format_registry.register_lazy(
3298
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3299
'bzrlib.repofmt.knitrepo',
3300
'RepositoryFormatKnit4',
3303
# Pack-based formats. There is one format for pre-subtrees, and one for
3304
# post-subtrees to allow ease of testing.
3305
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3306
format_registry.register_lazy(
3307
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3308
'bzrlib.repofmt.pack_repo',
3309
'RepositoryFormatKnitPack1',
3311
format_registry.register_lazy(
3312
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3313
'bzrlib.repofmt.pack_repo',
3314
'RepositoryFormatKnitPack3',
3316
format_registry.register_lazy(
3317
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3318
'bzrlib.repofmt.pack_repo',
3319
'RepositoryFormatKnitPack4',
3321
format_registry.register_lazy(
3322
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3323
'bzrlib.repofmt.pack_repo',
3324
'RepositoryFormatKnitPack5',
3326
format_registry.register_lazy(
3327
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3328
'bzrlib.repofmt.pack_repo',
3329
'RepositoryFormatKnitPack5RichRoot',
3331
format_registry.register_lazy(
3332
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3333
'bzrlib.repofmt.pack_repo',
3334
'RepositoryFormatKnitPack5RichRootBroken',
3336
format_registry.register_lazy(
3337
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3338
'bzrlib.repofmt.pack_repo',
3339
'RepositoryFormatKnitPack6',
3341
format_registry.register_lazy(
3342
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3343
'bzrlib.repofmt.pack_repo',
3344
'RepositoryFormatKnitPack6RichRoot',
3347
# Development formats.
3348
# Obsolete but kept pending a CHK based subtree format.
3349
format_registry.register_lazy(
3350
("Bazaar development format 2 with subtree support "
3351
"(needs bzr.dev from before 1.8)\n"),
3352
'bzrlib.repofmt.pack_repo',
3353
'RepositoryFormatPackDevelopment2Subtree',
3356
# 1.14->1.16 go below here
3357
format_registry.register_lazy(
3358
'Bazaar development format - group compression and chk inventory'
3359
' (needs bzr.dev from 1.14)\n',
3360
'bzrlib.repofmt.groupcompress_repo',
3361
'RepositoryFormatCHK1',
3364
format_registry.register_lazy(
3365
'Bazaar development format - chk repository with bencode revision '
3366
'serialization (needs bzr.dev from 1.16)\n',
3367
'bzrlib.repofmt.groupcompress_repo',
3368
'RepositoryFormatCHK2',
3370
format_registry.register_lazy(
3371
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3372
'bzrlib.repofmt.groupcompress_repo',
3373
'RepositoryFormat2a',
1545
3377
class InterRepository(InterObject):
1546
3378
"""This class represents operations taking place between two repositories.
1548
3380
Its instances have methods like copy_content and fetch, and contain
1549
references to the source and target repositories these operations can be
3381
references to the source and target repositories these operations can be
1550
3382
carried out on.
1552
3384
Often we will provide convenience methods on 'repository' which carry out
1731
3643
# we do not have a revision as that would be pointless.
1732
3644
target_ids = set(self.target._all_possible_ids())
1733
3645
possibly_present_revisions = target_ids.intersection(source_ids_set)
1734
actually_present_revisions = set(self.target._eliminate_revisions_not_present(possibly_present_revisions))
3646
actually_present_revisions = set(
3647
self.target._eliminate_revisions_not_present(possibly_present_revisions))
1735
3648
required_revisions = source_ids_set.difference(actually_present_revisions)
1736
required_topo_revisions = [rev_id for rev_id in source_ids if rev_id in required_revisions]
1737
3649
if revision_id is not None:
1738
3650
# we used get_ancestry to determine source_ids then we are assured all
1739
3651
# revisions referenced are present as they are installed in topological order.
1740
3652
# and the tip revision was validated by get_ancestry.
1741
return required_topo_revisions
3653
result_set = required_revisions
1743
# if we just grabbed the possibly available ids, then
3655
# if we just grabbed the possibly available ids, then
1744
3656
# we only have an estimate of whats available and need to validate
1745
3657
# that against the revision records.
1746
return self.source._eliminate_revisions_not_present(required_topo_revisions)
1749
class InterKnitRepo(InterRepository):
3659
self.source._eliminate_revisions_not_present(required_revisions))
3660
return self.source.revision_ids_to_search_result(result_set)
3663
class InterKnitRepo(InterSameDataRepository):
1750
3664
"""Optimised code paths between Knit based repositories."""
1752
_matching_repo_format = RepositoryFormatKnit1()
1753
"""Repository format for testing with."""
3667
def _get_repo_format_to_test(self):
3668
from bzrlib.repofmt import knitrepo
3669
return knitrepo.RepositoryFormatKnit1()
1756
3672
def is_compatible(source, target):
1757
3673
"""Be compatible with known Knit formats.
1759
3675
We don't test for the stores being of specific types because that
1760
could lead to confusing results, and there is no need to be
3676
could lead to confusing results, and there is no need to be
1761
3677
overly general.
3679
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
1764
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1765
isinstance(target._format, (RepositoryFormatKnit1)))
3681
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3682
isinstance(target._format, RepositoryFormatKnit))
1766
3683
except AttributeError:
1770
def fetch(self, revision_id=None, pb=None):
1771
"""See InterRepository.fetch()."""
1772
from bzrlib.fetch import KnitRepoFetcher
1773
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1774
self.source, self.source._format, self.target, self.target._format)
1775
f = KnitRepoFetcher(to_repository=self.target,
1776
from_repository=self.source,
1777
last_revision=revision_id,
1779
return f.count_copied, f.failed_revisions
3685
return are_knits and InterRepository._same_model(source, target)
1781
3687
@needs_read_lock
1782
def missing_revision_ids(self, revision_id=None):
3688
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1783
3689
"""See InterRepository.missing_revision_ids()."""
1784
3690
if revision_id is not None:
1785
3691
source_ids = self.source.get_ancestry(revision_id)
1786
assert source_ids[0] == None
3692
if source_ids[0] is not None:
3693
raise AssertionError()
1787
3694
source_ids.pop(0)
1789
source_ids = self.source._all_possible_ids()
3696
source_ids = self.source.all_revision_ids()
1790
3697
source_ids_set = set(source_ids)
1791
3698
# source_ids is the worst possible case we may need to pull.
1792
3699
# now we want to filter source_ids against what we actually
1793
3700
# have in target, but don't try to check for existence where we know
1794
3701
# we do not have a revision as that would be pointless.
1795
target_ids = set(self.target._all_possible_ids())
3702
target_ids = set(self.target.all_revision_ids())
1796
3703
possibly_present_revisions = target_ids.intersection(source_ids_set)
1797
actually_present_revisions = set(self.target._eliminate_revisions_not_present(possibly_present_revisions))
3704
actually_present_revisions = set(
3705
self.target._eliminate_revisions_not_present(possibly_present_revisions))
1798
3706
required_revisions = source_ids_set.difference(actually_present_revisions)
1799
required_topo_revisions = [rev_id for rev_id in source_ids if rev_id in required_revisions]
1800
3707
if revision_id is not None:
1801
3708
# we used get_ancestry to determine source_ids then we are assured all
1802
3709
# revisions referenced are present as they are installed in topological order.
1803
3710
# and the tip revision was validated by get_ancestry.
1804
return required_topo_revisions
3711
result_set = required_revisions
1806
# if we just grabbed the possibly available ids, then
3713
# if we just grabbed the possibly available ids, then
1807
3714
# we only have an estimate of whats available and need to validate
1808
3715
# that against the revision records.
1809
return self.source._eliminate_revisions_not_present(required_topo_revisions)
3717
self.source._eliminate_revisions_not_present(required_revisions))
3718
return self.source.revision_ids_to_search_result(result_set)
3721
class InterDifferingSerializer(InterRepository):
3724
def _get_repo_format_to_test(self):
3728
def is_compatible(source, target):
3729
"""Be compatible with Knit2 source and Knit3 target"""
3730
# This is redundant with format.check_conversion_target(), however that
3731
# raises an exception, and we just want to say "False" as in we won't
3732
# support converting between these formats.
3733
if 'IDS_never' in debug.debug_flags:
3735
if source.supports_rich_root() and not target.supports_rich_root():
3737
if (source._format.supports_tree_reference
3738
and not target._format.supports_tree_reference):
3740
if target._fallback_repositories and target._format.supports_chks:
3741
# IDS doesn't know how to copy CHKs for the parent inventories it
3742
# adds to stacked repos.
3744
if 'IDS_always' in debug.debug_flags:
3746
# Only use this code path for local source and target. IDS does far
3747
# too much IO (both bandwidth and roundtrips) over a network.
3748
if not source.bzrdir.transport.base.startswith('file:///'):
3750
if not target.bzrdir.transport.base.startswith('file:///'):
3754
def _get_trees(self, revision_ids, cache):
3756
for rev_id in revision_ids:
3758
possible_trees.append((rev_id, cache[rev_id]))
3760
# Not cached, but inventory might be present anyway.
3762
tree = self.source.revision_tree(rev_id)
3763
except errors.NoSuchRevision:
3764
# Nope, parent is ghost.
3767
cache[rev_id] = tree
3768
possible_trees.append((rev_id, tree))
3769
return possible_trees
3771
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3772
"""Get the best delta and base for this revision.
3774
:return: (basis_id, delta)
3777
# Generate deltas against each tree, to find the shortest.
3778
texts_possibly_new_in_tree = set()
3779
for basis_id, basis_tree in possible_trees:
3780
delta = tree.inventory._make_delta(basis_tree.inventory)
3781
for old_path, new_path, file_id, new_entry in delta:
3782
if new_path is None:
3783
# This file_id isn't present in the new rev, so we don't
3787
# Rich roots are handled elsewhere...
3789
kind = new_entry.kind
3790
if kind != 'directory' and kind != 'file':
3791
# No text record associated with this inventory entry.
3793
# This is a directory or file that has changed somehow.
3794
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3795
deltas.append((len(delta), basis_id, delta))
3797
return deltas[0][1:]
3799
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3800
"""Find all parent revisions that are absent, but for which the
3801
inventory is present, and copy those inventories.
3803
This is necessary to preserve correctness when the source is stacked
3804
without fallbacks configured. (Note that in cases like upgrade the
3805
source may be not have _fallback_repositories even though it is
3809
for parents in parent_map.values():
3810
parent_revs.update(parents)
3811
present_parents = self.source.get_parent_map(parent_revs)
3812
absent_parents = set(parent_revs).difference(present_parents)
3813
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3814
(rev_id,) for rev_id in absent_parents)
3815
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3816
for parent_tree in self.source.revision_trees(parent_inv_ids):
3817
current_revision_id = parent_tree.get_revision_id()
3818
parents_parents_keys = parent_invs_keys_for_stacking[
3819
(current_revision_id,)]
3820
parents_parents = [key[-1] for key in parents_parents_keys]
3821
basis_id = _mod_revision.NULL_REVISION
3822
basis_tree = self.source.revision_tree(basis_id)
3823
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3824
self.target.add_inventory_by_delta(
3825
basis_id, delta, current_revision_id, parents_parents)
3826
cache[current_revision_id] = parent_tree
3828
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3829
"""Fetch across a few revisions.
3831
:param revision_ids: The revisions to copy
3832
:param basis_id: The revision_id of a tree that must be in cache, used
3833
as a basis for delta when no other base is available
3834
:param cache: A cache of RevisionTrees that we can use.
3835
:param a_graph: A Graph object to determine the heads() of the
3836
rich-root data stream.
3837
:return: The revision_id of the last converted tree. The RevisionTree
3838
for it will be in cache
3840
# Walk though all revisions; get inventory deltas, copy referenced
3841
# texts that delta references, insert the delta, revision and
3843
root_keys_to_create = set()
3846
pending_revisions = []
3847
parent_map = self.source.get_parent_map(revision_ids)
3848
self._fetch_parent_invs_for_stacking(parent_map, cache)
3849
self.source._safe_to_return_from_cache = True
3850
for tree in self.source.revision_trees(revision_ids):
3851
# Find a inventory delta for this revision.
3852
# Find text entries that need to be copied, too.
3853
current_revision_id = tree.get_revision_id()
3854
parent_ids = parent_map.get(current_revision_id, ())
3855
parent_trees = self._get_trees(parent_ids, cache)
3856
possible_trees = list(parent_trees)
3857
if len(possible_trees) == 0:
3858
# There either aren't any parents, or the parents are ghosts,
3859
# so just use the last converted tree.
3860
possible_trees.append((basis_id, cache[basis_id]))
3861
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3863
revision = self.source.get_revision(current_revision_id)
3864
pending_deltas.append((basis_id, delta,
3865
current_revision_id, revision.parent_ids))
3866
if self._converting_to_rich_root:
3867
self._revision_id_to_root_id[current_revision_id] = \
3869
# Determine which texts are in present in this revision but not in
3870
# any of the available parents.
3871
texts_possibly_new_in_tree = set()
3872
for old_path, new_path, file_id, entry in delta:
3873
if new_path is None:
3874
# This file_id isn't present in the new rev
3878
if not self.target.supports_rich_root():
3879
# The target doesn't support rich root, so we don't
3882
if self._converting_to_rich_root:
3883
# This can't be copied normally, we have to insert
3885
root_keys_to_create.add((file_id, entry.revision))
3888
texts_possibly_new_in_tree.add((file_id, entry.revision))
3889
for basis_id, basis_tree in possible_trees:
3890
basis_inv = basis_tree.inventory
3891
for file_key in list(texts_possibly_new_in_tree):
3892
file_id, file_revision = file_key
3894
entry = basis_inv[file_id]
3895
except errors.NoSuchId:
3897
if entry.revision == file_revision:
3898
texts_possibly_new_in_tree.remove(file_key)
3899
text_keys.update(texts_possibly_new_in_tree)
3900
pending_revisions.append(revision)
3901
cache[current_revision_id] = tree
3902
basis_id = current_revision_id
3903
self.source._safe_to_return_from_cache = False
3905
from_texts = self.source.texts
3906
to_texts = self.target.texts
3907
if root_keys_to_create:
3908
root_stream = _mod_fetch._new_root_data_stream(
3909
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3910
self.source, graph=a_graph)
3911
to_texts.insert_record_stream(root_stream)
3912
to_texts.insert_record_stream(from_texts.get_record_stream(
3913
text_keys, self.target._format._fetch_order,
3914
not self.target._format._fetch_uses_deltas))
3915
# insert inventory deltas
3916
for delta in pending_deltas:
3917
self.target.add_inventory_by_delta(*delta)
3918
if self.target._fallback_repositories:
3919
# Make sure this stacked repository has all the parent inventories
3920
# for the new revisions that we are about to insert. We do this
3921
# before adding the revisions so that no revision is added until
3922
# all the inventories it may depend on are added.
3923
# Note that this is overzealous, as we may have fetched these in an
3926
revision_ids = set()
3927
for revision in pending_revisions:
3928
revision_ids.add(revision.revision_id)
3929
parent_ids.update(revision.parent_ids)
3930
parent_ids.difference_update(revision_ids)
3931
parent_ids.discard(_mod_revision.NULL_REVISION)
3932
parent_map = self.source.get_parent_map(parent_ids)
3933
# we iterate over parent_map and not parent_ids because we don't
3934
# want to try copying any revision which is a ghost
3935
for parent_tree in self.source.revision_trees(parent_map):
3936
current_revision_id = parent_tree.get_revision_id()
3937
parents_parents = parent_map[current_revision_id]
3938
possible_trees = self._get_trees(parents_parents, cache)
3939
if len(possible_trees) == 0:
3940
# There either aren't any parents, or the parents are
3941
# ghosts, so just use the last converted tree.
3942
possible_trees.append((basis_id, cache[basis_id]))
3943
basis_id, delta = self._get_delta_for_revision(parent_tree,
3944
parents_parents, possible_trees)
3945
self.target.add_inventory_by_delta(
3946
basis_id, delta, current_revision_id, parents_parents)
3947
# insert signatures and revisions
3948
for revision in pending_revisions:
3950
signature = self.source.get_signature_text(
3951
revision.revision_id)
3952
self.target.add_signature_text(revision.revision_id,
3954
except errors.NoSuchRevision:
3956
self.target.add_revision(revision.revision_id, revision)
3959
def _fetch_all_revisions(self, revision_ids, pb):
3960
"""Fetch everything for the list of revisions.
3962
:param revision_ids: The list of revisions to fetch. Must be in
3964
:param pb: A ProgressTask
3967
basis_id, basis_tree = self._get_basis(revision_ids[0])
3969
cache = lru_cache.LRUCache(100)
3970
cache[basis_id] = basis_tree
3971
del basis_tree # We don't want to hang on to it here
3973
if self._converting_to_rich_root and len(revision_ids) > 100:
3974
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3979
for offset in range(0, len(revision_ids), batch_size):
3980
self.target.start_write_group()
3982
pb.update('Transferring revisions', offset,
3984
batch = revision_ids[offset:offset+batch_size]
3985
basis_id = self._fetch_batch(batch, basis_id, cache,
3988
self.source._safe_to_return_from_cache = False
3989
self.target.abort_write_group()
3992
hint = self.target.commit_write_group()
3995
if hints and self.target._format.pack_compresses:
3996
self.target.pack(hint=hints)
3997
pb.update('Transferring revisions', len(revision_ids),
4001
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
4003
"""See InterRepository.fetch()."""
4004
if fetch_spec is not None:
4005
raise AssertionError("Not implemented yet...")
4006
# See <https://launchpad.net/bugs/456077> asking for a warning here
4008
# nb this is only active for local-local fetches; other things using
4010
ui.ui_factory.warn_cross_format_fetch(self.source._format,
4011
self.target._format)
4012
if (not self.source.supports_rich_root()
4013
and self.target.supports_rich_root()):
4014
self._converting_to_rich_root = True
4015
self._revision_id_to_root_id = {}
4017
self._converting_to_rich_root = False
4018
revision_ids = self.target.search_missing_revision_ids(self.source,
4019
revision_id, find_ghosts=find_ghosts).get_keys()
4020
if not revision_ids:
4022
revision_ids = tsort.topo_sort(
4023
self.source.get_graph().get_parent_map(revision_ids))
4024
if not revision_ids:
4026
# Walk though all revisions; get inventory deltas, copy referenced
4027
# texts that delta references, insert the delta, revision and
4030
my_pb = ui.ui_factory.nested_progress_bar()
4033
symbol_versioning.warn(
4034
symbol_versioning.deprecated_in((1, 14, 0))
4035
% "pb parameter to fetch()")
4038
self._fetch_all_revisions(revision_ids, pb)
4040
if my_pb is not None:
4042
return len(revision_ids), 0
4044
def _get_basis(self, first_revision_id):
4045
"""Get a revision and tree which exists in the target.
4047
This assumes that first_revision_id is selected for transmission
4048
because all other ancestors are already present. If we can't find an
4049
ancestor we fall back to NULL_REVISION since we know that is safe.
4051
:return: (basis_id, basis_tree)
4053
first_rev = self.source.get_revision(first_revision_id)
4055
basis_id = first_rev.parent_ids[0]
4056
# only valid as a basis if the target has it
4057
self.target.get_revision(basis_id)
4058
# Try to get a basis tree - if its a ghost it will hit the
4059
# NoSuchRevision case.
4060
basis_tree = self.source.revision_tree(basis_id)
4061
except (IndexError, errors.NoSuchRevision):
4062
basis_id = _mod_revision.NULL_REVISION
4063
basis_tree = self.source.revision_tree(basis_id)
4064
return basis_id, basis_tree
4067
InterRepository.register_optimiser(InterDifferingSerializer)
4068
InterRepository.register_optimiser(InterSameDataRepository)
1811
4069
InterRepository.register_optimiser(InterWeaveRepo)
1812
4070
InterRepository.register_optimiser(InterKnitRepo)
1815
class RepositoryTestProviderAdapter(object):
1816
"""A tool to generate a suite testing multiple repository formats at once.
1818
This is done by copying the test once for each transport and injecting
1819
the transport_server, transport_readonly_server, and bzrdir_format and
1820
repository_format classes into each copy. Each copy is also given a new id()
1821
to make it easy to identify.
1824
def __init__(self, transport_server, transport_readonly_server, formats):
1825
self._transport_server = transport_server
1826
self._transport_readonly_server = transport_readonly_server
1827
self._formats = formats
1829
def adapt(self, test):
1830
result = TestSuite()
1831
for repository_format, bzrdir_format in self._formats:
1832
new_test = deepcopy(test)
1833
new_test.transport_server = self._transport_server
1834
new_test.transport_readonly_server = self._transport_readonly_server
1835
new_test.bzrdir_format = bzrdir_format
1836
new_test.repository_format = repository_format
1837
def make_new_test_id():
1838
new_id = "%s(%s)" % (new_test.id(), repository_format.__class__.__name__)
1839
return lambda: new_id
1840
new_test.id = make_new_test_id()
1841
result.addTest(new_test)
1845
class InterRepositoryTestProviderAdapter(object):
1846
"""A tool to generate a suite testing multiple inter repository formats.
1848
This is done by copying the test once for each interrepo provider and injecting
1849
the transport_server, transport_readonly_server, repository_format and
1850
repository_to_format classes into each copy.
1851
Each copy is also given a new id() to make it easy to identify.
1854
def __init__(self, transport_server, transport_readonly_server, formats):
1855
self._transport_server = transport_server
1856
self._transport_readonly_server = transport_readonly_server
1857
self._formats = formats
1859
def adapt(self, test):
1860
result = TestSuite()
1861
for interrepo_class, repository_format, repository_format_to in self._formats:
1862
new_test = deepcopy(test)
1863
new_test.transport_server = self._transport_server
1864
new_test.transport_readonly_server = self._transport_readonly_server
1865
new_test.interrepo_class = interrepo_class
1866
new_test.repository_format = repository_format
1867
new_test.repository_format_to = repository_format_to
1868
def make_new_test_id():
1869
new_id = "%s(%s)" % (new_test.id(), interrepo_class.__name__)
1870
return lambda: new_id
1871
new_test.id = make_new_test_id()
1872
result.addTest(new_test)
1876
def default_test_list():
1877
"""Generate the default list of interrepo permutations to test."""
1879
# test the default InterRepository between format 6 and the current
1881
# XXX: robertc 20060220 reinstate this when there are two supported
1882
# formats which do not have an optimal code path between them.
1883
result.append((InterRepository,
1884
RepositoryFormat6(),
1885
RepositoryFormatKnit1()))
1886
for optimiser in InterRepository._optimisers:
1887
result.append((optimiser,
1888
optimiser._matching_repo_format,
1889
optimiser._matching_repo_format
1891
# if there are specific combinations we want to use, we can add them
1896
4073
class CopyConverter(object):
1897
4074
"""A repository conversion tool which just performs a copy of the content.
1899
4076
This is slow but quite reliable.
2140
4148
"""Unescape predefined XML entities in a string of data."""
2141
4149
global _unescape_re
2142
4150
if _unescape_re is None:
2143
_unescape_re = re.compile('\&([^;]*);')
4151
_unescape_re = re.compile('\&([^;]*);')
2144
4152
return _unescape_re.sub(_unescaper, data)
4155
class _VersionedFileChecker(object):
4157
def __init__(self, repository, text_key_references=None, ancestors=None):
4158
self.repository = repository
4159
self.text_index = self.repository._generate_text_key_index(
4160
text_key_references=text_key_references, ancestors=ancestors)
4162
def calculate_file_version_parents(self, text_key):
4163
"""Calculate the correct parents for a file version according to
4166
parent_keys = self.text_index[text_key]
4167
if parent_keys == [_mod_revision.NULL_REVISION]:
4169
return tuple(parent_keys)
4171
def check_file_version_parents(self, texts, progress_bar=None):
4172
"""Check the parents stored in a versioned file are correct.
4174
It also detects file versions that are not referenced by their
4175
corresponding revision's inventory.
4177
:returns: A tuple of (wrong_parents, dangling_file_versions).
4178
wrong_parents is a dict mapping {revision_id: (stored_parents,
4179
correct_parents)} for each revision_id where the stored parents
4180
are not correct. dangling_file_versions is a set of (file_id,
4181
revision_id) tuples for versions that are present in this versioned
4182
file, but not used by the corresponding inventory.
4184
local_progress = None
4185
if progress_bar is None:
4186
local_progress = ui.ui_factory.nested_progress_bar()
4187
progress_bar = local_progress
4189
return self._check_file_version_parents(texts, progress_bar)
4192
local_progress.finished()
4194
def _check_file_version_parents(self, texts, progress_bar):
4195
"""See check_file_version_parents."""
4197
self.file_ids = set([file_id for file_id, _ in
4198
self.text_index.iterkeys()])
4199
# text keys is now grouped by file_id
4200
n_versions = len(self.text_index)
4201
progress_bar.update('loading text store', 0, n_versions)
4202
parent_map = self.repository.texts.get_parent_map(self.text_index)
4203
# On unlistable transports this could well be empty/error...
4204
text_keys = self.repository.texts.keys()
4205
unused_keys = frozenset(text_keys) - set(self.text_index)
4206
for num, key in enumerate(self.text_index.iterkeys()):
4207
progress_bar.update('checking text graph', num, n_versions)
4208
correct_parents = self.calculate_file_version_parents(key)
4210
knit_parents = parent_map[key]
4211
except errors.RevisionNotPresent:
4214
if correct_parents != knit_parents:
4215
wrong_parents[key] = (knit_parents, correct_parents)
4216
return wrong_parents, unused_keys
4219
def _old_get_graph(repository, revision_id):
4220
"""DO NOT USE. That is all. I'm serious."""
4221
graph = repository.get_graph()
4222
revision_graph = dict(((key, value) for key, value in
4223
graph.iter_ancestry([revision_id]) if value is not None))
4224
return _strip_NULL_ghosts(revision_graph)
4227
def _strip_NULL_ghosts(revision_graph):
4228
"""Also don't use this. more compatibility code for unmigrated clients."""
4229
# Filter ghosts, and null:
4230
if _mod_revision.NULL_REVISION in revision_graph:
4231
del revision_graph[_mod_revision.NULL_REVISION]
4232
for key, parents in revision_graph.items():
4233
revision_graph[key] = tuple(parent for parent in parents if parent
4235
return revision_graph
4238
class StreamSink(object):
4239
"""An object that can insert a stream into a repository.
4241
This interface handles the complexity of reserialising inventories and
4242
revisions from different formats, and allows unidirectional insertion into
4243
stacked repositories without looking for the missing basis parents
4247
def __init__(self, target_repo):
4248
self.target_repo = target_repo
4250
def insert_stream(self, stream, src_format, resume_tokens):
4251
"""Insert a stream's content into the target repository.
4253
:param src_format: a bzr repository format.
4255
:return: a list of resume tokens and an iterable of keys additional
4256
items required before the insertion can be completed.
4258
self.target_repo.lock_write()
4261
self.target_repo.resume_write_group(resume_tokens)
4264
self.target_repo.start_write_group()
4267
# locked_insert_stream performs a commit|suspend.
4268
return self._locked_insert_stream(stream, src_format, is_resume)
4270
self.target_repo.abort_write_group(suppress_errors=True)
4273
self.target_repo.unlock()
4275
def _locked_insert_stream(self, stream, src_format, is_resume):
4276
to_serializer = self.target_repo._format._serializer
4277
src_serializer = src_format._serializer
4279
if to_serializer == src_serializer:
4280
# If serializers match and the target is a pack repository, set the
4281
# write cache size on the new pack. This avoids poor performance
4282
# on transports where append is unbuffered (such as
4283
# RemoteTransport). This is safe to do because nothing should read
4284
# back from the target repository while a stream with matching
4285
# serialization is being inserted.
4286
# The exception is that a delta record from the source that should
4287
# be a fulltext may need to be expanded by the target (see
4288
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4289
# explicitly flush any buffered writes first in that rare case.
4291
new_pack = self.target_repo._pack_collection._new_pack
4292
except AttributeError:
4293
# Not a pack repository
4296
new_pack.set_write_cache_size(1024*1024)
4297
for substream_type, substream in stream:
4298
if 'stream' in debug.debug_flags:
4299
mutter('inserting substream: %s', substream_type)
4300
if substream_type == 'texts':
4301
self.target_repo.texts.insert_record_stream(substream)
4302
elif substream_type == 'inventories':
4303
if src_serializer == to_serializer:
4304
self.target_repo.inventories.insert_record_stream(
4307
self._extract_and_insert_inventories(
4308
substream, src_serializer)
4309
elif substream_type == 'inventory-deltas':
4310
ui.ui_factory.warn_cross_format_fetch(src_format,
4311
self.target_repo._format)
4312
self._extract_and_insert_inventory_deltas(
4313
substream, src_serializer)
4314
elif substream_type == 'chk_bytes':
4315
# XXX: This doesn't support conversions, as it assumes the
4316
# conversion was done in the fetch code.
4317
self.target_repo.chk_bytes.insert_record_stream(substream)
4318
elif substream_type == 'revisions':
4319
# This may fallback to extract-and-insert more often than
4320
# required if the serializers are different only in terms of
4322
if src_serializer == to_serializer:
4323
self.target_repo.revisions.insert_record_stream(
4326
self._extract_and_insert_revisions(substream,
4328
elif substream_type == 'signatures':
4329
self.target_repo.signatures.insert_record_stream(substream)
4331
raise AssertionError('kaboom! %s' % (substream_type,))
4332
# Done inserting data, and the missing_keys calculations will try to
4333
# read back from the inserted data, so flush the writes to the new pack
4334
# (if this is pack format).
4335
if new_pack is not None:
4336
new_pack._write_data('', flush=True)
4337
# Find all the new revisions (including ones from resume_tokens)
4338
missing_keys = self.target_repo.get_missing_parent_inventories(
4339
check_for_missing_texts=is_resume)
4341
for prefix, versioned_file in (
4342
('texts', self.target_repo.texts),
4343
('inventories', self.target_repo.inventories),
4344
('revisions', self.target_repo.revisions),
4345
('signatures', self.target_repo.signatures),
4346
('chk_bytes', self.target_repo.chk_bytes),
4348
if versioned_file is None:
4350
# TODO: key is often going to be a StaticTuple object
4351
# I don't believe we can define a method by which
4352
# (prefix,) + StaticTuple will work, though we could
4353
# define a StaticTuple.sq_concat that would allow you to
4354
# pass in either a tuple or a StaticTuple as the second
4355
# object, so instead we could have:
4356
# StaticTuple(prefix) + key here...
4357
missing_keys.update((prefix,) + key for key in
4358
versioned_file.get_missing_compression_parent_keys())
4359
except NotImplementedError:
4360
# cannot even attempt suspending, and missing would have failed
4361
# during stream insertion.
4362
missing_keys = set()
4365
# suspend the write group and tell the caller what we is
4366
# missing. We know we can suspend or else we would not have
4367
# entered this code path. (All repositories that can handle
4368
# missing keys can handle suspending a write group).
4369
write_group_tokens = self.target_repo.suspend_write_group()
4370
return write_group_tokens, missing_keys
4371
hint = self.target_repo.commit_write_group()
4372
if (to_serializer != src_serializer and
4373
self.target_repo._format.pack_compresses):
4374
self.target_repo.pack(hint=hint)
4377
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4378
target_rich_root = self.target_repo._format.rich_root_data
4379
target_tree_refs = self.target_repo._format.supports_tree_reference
4380
for record in substream:
4381
# Insert the delta directly
4382
inventory_delta_bytes = record.get_bytes_as('fulltext')
4383
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4385
parse_result = deserialiser.parse_text_bytes(
4386
inventory_delta_bytes)
4387
except inventory_delta.IncompatibleInventoryDelta, err:
4388
trace.mutter("Incompatible delta: %s", err.msg)
4389
raise errors.IncompatibleRevision(self.target_repo._format)
4390
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4391
revision_id = new_id
4392
parents = [key[0] for key in record.parents]
4393
self.target_repo.add_inventory_by_delta(
4394
basis_id, inv_delta, revision_id, parents)
4396
def _extract_and_insert_inventories(self, substream, serializer,
4398
"""Generate a new inventory versionedfile in target, converting data.
4400
The inventory is retrieved from the source, (deserializing it), and
4401
stored in the target (reserializing it in a different format).
4403
target_rich_root = self.target_repo._format.rich_root_data
4404
target_tree_refs = self.target_repo._format.supports_tree_reference
4405
for record in substream:
4406
# It's not a delta, so it must be a fulltext in the source
4407
# serializer's format.
4408
bytes = record.get_bytes_as('fulltext')
4409
revision_id = record.key[0]
4410
inv = serializer.read_inventory_from_string(bytes, revision_id)
4411
parents = [key[0] for key in record.parents]
4412
self.target_repo.add_inventory(revision_id, inv, parents)
4413
# No need to keep holding this full inv in memory when the rest of
4414
# the substream is likely to be all deltas.
4417
def _extract_and_insert_revisions(self, substream, serializer):
4418
for record in substream:
4419
bytes = record.get_bytes_as('fulltext')
4420
revision_id = record.key[0]
4421
rev = serializer.read_revision_from_string(bytes)
4422
if rev.revision_id != revision_id:
4423
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4424
self.target_repo.add_revision(revision_id, rev)
4427
if self.target_repo._format._fetch_reconcile:
4428
self.target_repo.reconcile()
4431
class StreamSource(object):
4432
"""A source of a stream for fetching between repositories."""
4434
def __init__(self, from_repository, to_format):
4435
"""Create a StreamSource streaming from from_repository."""
4436
self.from_repository = from_repository
4437
self.to_format = to_format
4439
def delta_on_metadata(self):
4440
"""Return True if delta's are permitted on metadata streams.
4442
That is on revisions and signatures.
4444
src_serializer = self.from_repository._format._serializer
4445
target_serializer = self.to_format._serializer
4446
return (self.to_format._fetch_uses_deltas and
4447
src_serializer == target_serializer)
4449
def _fetch_revision_texts(self, revs):
4450
# fetch signatures first and then the revision texts
4451
# may need to be a InterRevisionStore call here.
4452
from_sf = self.from_repository.signatures
4453
# A missing signature is just skipped.
4454
keys = [(rev_id,) for rev_id in revs]
4455
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4457
self.to_format._fetch_order,
4458
not self.to_format._fetch_uses_deltas))
4459
# If a revision has a delta, this is actually expanded inside the
4460
# insert_record_stream code now, which is an alternate fix for
4462
from_rf = self.from_repository.revisions
4463
revisions = from_rf.get_record_stream(
4465
self.to_format._fetch_order,
4466
not self.delta_on_metadata())
4467
return [('signatures', signatures), ('revisions', revisions)]
4469
def _generate_root_texts(self, revs):
4470
"""This will be called by get_stream between fetching weave texts and
4471
fetching the inventory weave.
4473
if self._rich_root_upgrade():
4474
return _mod_fetch.Inter1and2Helper(
4475
self.from_repository).generate_root_texts(revs)
4479
def get_stream(self, search):
4481
revs = search.get_keys()
4482
graph = self.from_repository.get_graph()
4483
revs = tsort.topo_sort(graph.get_parent_map(revs))
4484
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4486
for knit_kind, file_id, revisions in data_to_fetch:
4487
if knit_kind != phase:
4489
# Make a new progress bar for this phase
4490
if knit_kind == "file":
4491
# Accumulate file texts
4492
text_keys.extend([(file_id, revision) for revision in
4494
elif knit_kind == "inventory":
4495
# Now copy the file texts.
4496
from_texts = self.from_repository.texts
4497
yield ('texts', from_texts.get_record_stream(
4498
text_keys, self.to_format._fetch_order,
4499
not self.to_format._fetch_uses_deltas))
4500
# Cause an error if a text occurs after we have done the
4503
# Before we process the inventory we generate the root
4504
# texts (if necessary) so that the inventories references
4506
for _ in self._generate_root_texts(revs):
4508
# we fetch only the referenced inventories because we do not
4509
# know for unselected inventories whether all their required
4510
# texts are present in the other repository - it could be
4512
for info in self._get_inventory_stream(revs):
4514
elif knit_kind == "signatures":
4515
# Nothing to do here; this will be taken care of when
4516
# _fetch_revision_texts happens.
4518
elif knit_kind == "revisions":
4519
for record in self._fetch_revision_texts(revs):
4522
raise AssertionError("Unknown knit kind %r" % knit_kind)
4524
def get_stream_for_missing_keys(self, missing_keys):
4525
# missing keys can only occur when we are byte copying and not
4526
# translating (because translation means we don't send
4527
# unreconstructable deltas ever).
4529
keys['texts'] = set()
4530
keys['revisions'] = set()
4531
keys['inventories'] = set()
4532
keys['chk_bytes'] = set()
4533
keys['signatures'] = set()
4534
for key in missing_keys:
4535
keys[key[0]].add(key[1:])
4536
if len(keys['revisions']):
4537
# If we allowed copying revisions at this point, we could end up
4538
# copying a revision without copying its required texts: a
4539
# violation of the requirements for repository integrity.
4540
raise AssertionError(
4541
'cannot copy revisions to fill in missing deltas %s' % (
4542
keys['revisions'],))
4543
for substream_kind, keys in keys.iteritems():
4544
vf = getattr(self.from_repository, substream_kind)
4545
if vf is None and keys:
4546
raise AssertionError(
4547
"cannot fill in keys for a versioned file we don't"
4548
" have: %s needs %s" % (substream_kind, keys))
4550
# No need to stream something we don't have
4552
if substream_kind == 'inventories':
4553
# Some missing keys are genuinely ghosts, filter those out.
4554
present = self.from_repository.inventories.get_parent_map(keys)
4555
revs = [key[0] for key in present]
4556
# Get the inventory stream more-or-less as we do for the
4557
# original stream; there's no reason to assume that records
4558
# direct from the source will be suitable for the sink. (Think
4559
# e.g. 2a -> 1.9-rich-root).
4560
for info in self._get_inventory_stream(revs, missing=True):
4564
# Ask for full texts always so that we don't need more round trips
4565
# after this stream.
4566
# Some of the missing keys are genuinely ghosts, so filter absent
4567
# records. The Sink is responsible for doing another check to
4568
# ensure that ghosts don't introduce missing data for future
4570
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4571
self.to_format._fetch_order, True))
4572
yield substream_kind, stream
4574
def inventory_fetch_order(self):
4575
if self._rich_root_upgrade():
4576
return 'topological'
4578
return self.to_format._fetch_order
4580
def _rich_root_upgrade(self):
4581
return (not self.from_repository._format.rich_root_data and
4582
self.to_format.rich_root_data)
4584
def _get_inventory_stream(self, revision_ids, missing=False):
4585
from_format = self.from_repository._format
4586
if (from_format.supports_chks and self.to_format.supports_chks and
4587
from_format.network_name() == self.to_format.network_name()):
4588
raise AssertionError(
4589
"this case should be handled by GroupCHKStreamSource")
4590
elif 'forceinvdeltas' in debug.debug_flags:
4591
return self._get_convertable_inventory_stream(revision_ids,
4592
delta_versus_null=missing)
4593
elif from_format.network_name() == self.to_format.network_name():
4595
return self._get_simple_inventory_stream(revision_ids,
4597
elif (not from_format.supports_chks and not self.to_format.supports_chks
4598
and from_format._serializer == self.to_format._serializer):
4599
# Essentially the same format.
4600
return self._get_simple_inventory_stream(revision_ids,
4603
# Any time we switch serializations, we want to use an
4604
# inventory-delta based approach.
4605
return self._get_convertable_inventory_stream(revision_ids,
4606
delta_versus_null=missing)
4608
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4609
# NB: This currently reopens the inventory weave in source;
4610
# using a single stream interface instead would avoid this.
4611
from_weave = self.from_repository.inventories
4613
delta_closure = True
4615
delta_closure = not self.delta_on_metadata()
4616
yield ('inventories', from_weave.get_record_stream(
4617
[(rev_id,) for rev_id in revision_ids],
4618
self.inventory_fetch_order(), delta_closure))
4620
def _get_convertable_inventory_stream(self, revision_ids,
4621
delta_versus_null=False):
4622
# The two formats are sufficiently different that there is no fast
4623
# path, so we need to send just inventorydeltas, which any
4624
# sufficiently modern client can insert into any repository.
4625
# The StreamSink code expects to be able to
4626
# convert on the target, so we need to put bytes-on-the-wire that can
4627
# be converted. That means inventory deltas (if the remote is <1.19,
4628
# RemoteStreamSink will fallback to VFS to insert the deltas).
4629
yield ('inventory-deltas',
4630
self._stream_invs_as_deltas(revision_ids,
4631
delta_versus_null=delta_versus_null))
4633
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4634
"""Return a stream of inventory-deltas for the given rev ids.
4636
:param revision_ids: The list of inventories to transmit
4637
:param delta_versus_null: Don't try to find a minimal delta for this
4638
entry, instead compute the delta versus the NULL_REVISION. This
4639
effectively streams a complete inventory. Used for stuff like
4640
filling in missing parents, etc.
4642
from_repo = self.from_repository
4643
revision_keys = [(rev_id,) for rev_id in revision_ids]
4644
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4645
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4647
inventories = self.from_repository.iter_inventories(
4648
revision_ids, 'topological')
4649
format = from_repo._format
4650
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4651
inventory_cache = lru_cache.LRUCache(50)
4652
null_inventory = from_repo.revision_tree(
4653
_mod_revision.NULL_REVISION).inventory
4654
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4655
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4656
# repo back into a non-rich-root repo ought to be allowed)
4657
serializer = inventory_delta.InventoryDeltaSerializer(
4658
versioned_root=format.rich_root_data,
4659
tree_references=format.supports_tree_reference)
4660
for inv in inventories:
4661
key = (inv.revision_id,)
4662
parent_keys = parent_map.get(key, ())
4664
if not delta_versus_null and parent_keys:
4665
# The caller did not ask for complete inventories and we have
4666
# some parents that we can delta against. Make a delta against
4667
# each parent so that we can find the smallest.
4668
parent_ids = [parent_key[0] for parent_key in parent_keys]
4669
for parent_id in parent_ids:
4670
if parent_id not in invs_sent_so_far:
4671
# We don't know that the remote side has this basis, so
4674
if parent_id == _mod_revision.NULL_REVISION:
4675
parent_inv = null_inventory
4677
parent_inv = inventory_cache.get(parent_id, None)
4678
if parent_inv is None:
4679
parent_inv = from_repo.get_inventory(parent_id)
4680
candidate_delta = inv._make_delta(parent_inv)
4681
if (delta is None or
4682
len(delta) > len(candidate_delta)):
4683
delta = candidate_delta
4684
basis_id = parent_id
4686
# Either none of the parents ended up being suitable, or we
4687
# were asked to delta against NULL
4688
basis_id = _mod_revision.NULL_REVISION
4689
delta = inv._make_delta(null_inventory)
4690
invs_sent_so_far.add(inv.revision_id)
4691
inventory_cache[inv.revision_id] = inv
4692
delta_serialized = ''.join(
4693
serializer.delta_to_lines(basis_id, key[-1], delta))
4694
yield versionedfile.FulltextContentFactory(
4695
key, parent_keys, None, delta_serialized)
4698
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4699
stop_revision=None):
4700
"""Extend the partial history to include a given index
4702
If a stop_index is supplied, stop when that index has been reached.
4703
If a stop_revision is supplied, stop when that revision is
4704
encountered. Otherwise, stop when the beginning of history is
4707
:param stop_index: The index which should be present. When it is
4708
present, history extension will stop.
4709
:param stop_revision: The revision id which should be present. When
4710
it is encountered, history extension will stop.
4712
start_revision = partial_history_cache[-1]
4713
iterator = repo.iter_reverse_revision_history(start_revision)
4715
#skip the last revision in the list
4718
if (stop_index is not None and
4719
len(partial_history_cache) > stop_index):
4721
if partial_history_cache[-1] == stop_revision:
4723
revision_id = iterator.next()
4724
partial_history_cache.append(revision_id)
4725
except StopIteration: