1
# Copyright (C) 2005-2011 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
"""Repository formats built around versioned files."""
19
from __future__ import absolute_import
22
from ..lazy_import import lazy_import
23
lazy_import(globals(), """
28
config as _mod_config,
36
revision as _mod_revision,
37
serializer as _mod_serializer,
42
from breezy.bzr import (
49
from breezy.recordcounter import RecordCounter
50
from breezy.testament import Testament
51
from breezy.i18n import gettext
57
from ..decorators import (
62
from .inventory import (
69
from ..repository import (
75
from .repository import (
77
RepositoryFormatMetaDir,
80
from ..sixish import (
91
class VersionedFileRepositoryFormat(RepositoryFormat):
92
"""Base class for all repository formats that are VersionedFiles-based."""
94
supports_full_versioned_files = True
95
supports_versioned_directories = True
96
supports_unreferenced_revisions = True
98
# Should commit add an inventory, or an inventory delta to the repository.
99
_commit_inv_deltas = True
100
# What order should fetch operations request streams in?
101
# The default is unordered as that is the cheapest for an origin to
103
_fetch_order = 'unordered'
104
# Does this repository format use deltas that can be fetched as-deltas ?
105
# (E.g. knits, where the knit deltas can be transplanted intact.
106
# We default to False, which will ensure that enough data to get
107
# a full text out of any fetch stream will be grabbed.
108
_fetch_uses_deltas = False
111
class VersionedFileCommitBuilder(CommitBuilder):
112
"""Commit builder implementation for versioned files based repositories.
115
# the default CommitBuilder does not manage trees whose root is versioned.
116
_versioned_root = False
118
def __init__(self, repository, parents, config_stack, timestamp=None,
119
timezone=None, committer=None, revprops=None,
120
revision_id=None, lossy=False):
121
super(VersionedFileCommitBuilder, self).__init__(repository,
122
parents, config_stack, timestamp, timezone, committer, revprops,
125
basis_id = self.parents[0]
127
basis_id = _mod_revision.NULL_REVISION
128
self.basis_delta_revision = basis_id
129
self._new_inventory = None
130
self._basis_delta = []
131
self.__heads = graph.HeadsCache(repository.get_graph()).heads
132
# memo'd check for no-op commits.
133
self._any_changes = False
135
def any_changes(self):
136
"""Return True if any entries were changed.
138
This includes merge-only changes. It is the core for the --unchanged
141
:return: True if any changes have occured.
143
return self._any_changes
145
def _ensure_fallback_inventories(self):
146
"""Ensure that appropriate inventories are available.
148
This only applies to repositories that are stacked, and is about
149
enusring the stacking invariants. Namely, that for any revision that is
150
present, we either have all of the file content, or we have the parent
151
inventory and the delta file content.
153
if not self.repository._fallback_repositories:
155
if not self.repository._format.supports_chks:
156
raise errors.BzrError("Cannot commit directly to a stacked branch"
157
" in pre-2a formats. See "
158
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
159
# This is a stacked repo, we need to make sure we have the parent
160
# inventories for the parents.
161
parent_keys = [(p,) for p in self.parents]
162
parent_map = self.repository.inventories._index.get_parent_map(parent_keys)
163
missing_parent_keys = {pk for pk in parent_keys
164
if pk not in parent_map}
165
fallback_repos = list(reversed(self.repository._fallback_repositories))
166
missing_keys = [('inventories', pk[0])
167
for pk in missing_parent_keys]
169
while missing_keys and fallback_repos:
170
fallback_repo = fallback_repos.pop()
171
source = fallback_repo._get_source(self.repository._format)
172
sink = self.repository._get_sink()
173
stream = source.get_stream_for_missing_keys(missing_keys)
174
missing_keys = sink.insert_stream_without_locking(stream,
175
self.repository._format)
177
raise errors.BzrError('Unable to fill in parent inventories for a'
180
def commit(self, message):
181
"""Make the actual commit.
183
:return: The revision id of the recorded revision.
185
self._validate_unicode_text(message, 'commit message')
186
rev = _mod_revision.Revision(
187
timestamp=self._timestamp,
188
timezone=self._timezone,
189
committer=self._committer,
191
inventory_sha1=self.inv_sha1,
192
revision_id=self._new_revision_id,
193
properties=self._revprops)
194
rev.parent_ids = self.parents
195
if self._config_stack.get('create_signatures') == _mod_config.SIGN_ALWAYS:
196
testament = Testament(rev, self.revision_tree())
197
plaintext = testament.as_short_text()
198
self.repository.store_revision_signature(
199
gpg.GPGStrategy(self._config_stack), plaintext,
200
self._new_revision_id)
201
self.repository._add_revision(rev)
202
self._ensure_fallback_inventories()
203
self.repository.commit_write_group()
204
return self._new_revision_id
207
"""Abort the commit that is being built.
209
self.repository.abort_write_group()
211
def revision_tree(self):
212
"""Return the tree that was just committed.
214
After calling commit() this can be called to get a
215
RevisionTree representing the newly committed tree. This is
216
preferred to calling Repository.revision_tree() because that may
217
require deserializing the inventory, while we already have a copy in
220
if self._new_inventory is None:
221
self._new_inventory = self.repository.get_inventory(
222
self._new_revision_id)
223
return inventorytree.InventoryRevisionTree(self.repository,
224
self._new_inventory, self._new_revision_id)
226
def finish_inventory(self):
227
"""Tell the builder that the inventory is finished.
229
:return: The inventory id in the repository, which can be used with
230
repository.get_inventory.
232
# an inventory delta was accumulated without creating a new
234
basis_id = self.basis_delta_revision
235
self.inv_sha1, self._new_inventory = self.repository.add_inventory_by_delta(
236
basis_id, self._basis_delta, self._new_revision_id,
238
return self._new_revision_id
240
def _require_root_change(self, tree):
241
"""Enforce an appropriate root object change.
243
This is called once when record_iter_changes is called, if and only if
244
the root was not in the delta calculated by record_iter_changes.
246
:param tree: The tree which is being committed.
248
if len(self.parents) == 0:
249
raise errors.RootMissing()
250
entry = entry_factory['directory'](tree.path2id(''), '',
252
entry.revision = self._new_revision_id
253
self._basis_delta.append(('', '', entry.file_id, entry))
255
def _get_delta(self, ie, basis_inv, path):
256
"""Get a delta against the basis inventory for ie."""
257
if not basis_inv.has_id(ie.file_id):
259
result = (None, path, ie.file_id, ie)
260
self._basis_delta.append(result)
262
elif ie != basis_inv[ie.file_id]:
264
# TODO: avoid tis id2path call.
265
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
266
self._basis_delta.append(result)
272
def _heads(self, file_id, revision_ids):
273
"""Calculate the graph heads for revision_ids in the graph of file_id.
275
This can use either a per-file graph or a global revision graph as we
276
have an identity relationship between the two graphs.
278
return self.__heads(revision_ids)
280
def get_basis_delta(self):
281
"""Return the complete inventory delta versus the basis inventory.
283
:return: An inventory delta, suitable for use with apply_delta, or
284
Repository.add_inventory_by_delta, etc.
286
return self._basis_delta
288
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
289
_entry_factory=entry_factory):
290
"""Record a new tree via iter_changes.
292
:param tree: The tree to obtain text contents from for changed objects.
293
:param basis_revision_id: The revision id of the tree the iter_changes
294
has been generated against. Currently assumed to be the same
295
as self.parents[0] - if it is not, errors may occur.
296
:param iter_changes: An iter_changes iterator with the changes to apply
297
to basis_revision_id. The iterator must not include any items with
298
a current kind of None - missing items must be either filtered out
299
or errored-on before record_iter_changes sees the item.
300
:param _entry_factory: Private method to bind entry_factory locally for
302
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
305
# Create an inventory delta based on deltas between all the parents and
306
# deltas between all the parent inventories. We use inventory delta's
307
# between the inventory objects because iter_changes masks
308
# last-changed-field only changes.
310
# file_id -> change map, change is fileid, paths, changed, versioneds,
311
# parents, names, kinds, executables
313
# {file_id -> revision_id -> inventory entry, for entries in parent
314
# trees that are not parents[0]
318
revtrees = list(self.repository.revision_trees(self.parents))
319
except errors.NoSuchRevision:
320
# one or more ghosts, slow path.
322
for revision_id in self.parents:
324
revtrees.append(self.repository.revision_tree(revision_id))
325
except errors.NoSuchRevision:
327
basis_revision_id = _mod_revision.NULL_REVISION
329
revtrees.append(self.repository.revision_tree(
330
_mod_revision.NULL_REVISION))
331
# The basis inventory from a repository
333
basis_tree = revtrees[0]
335
basis_tree = self.repository.revision_tree(
336
_mod_revision.NULL_REVISION)
337
basis_inv = basis_tree.root_inventory
338
if len(self.parents) > 0:
339
if basis_revision_id != self.parents[0] and not ghost_basis:
341
"arbitrary basis parents not yet supported with merges")
342
for revtree in revtrees[1:]:
343
for change in revtree.root_inventory._make_delta(basis_inv):
344
if change[1] is None:
345
# Not present in this parent.
347
if change[2] not in merged_ids:
348
if change[0] is not None:
349
basis_entry = basis_inv[change[2]]
350
merged_ids[change[2]] = [
352
basis_entry.revision,
355
parent_entries[change[2]] = {
357
basis_entry.revision:basis_entry,
359
change[3].revision:change[3],
362
merged_ids[change[2]] = [change[3].revision]
363
parent_entries[change[2]] = {change[3].revision:change[3]}
365
merged_ids[change[2]].append(change[3].revision)
366
parent_entries[change[2]][change[3].revision] = change[3]
369
# Setup the changes from the tree:
370
# changes maps file_id -> (change, [parent revision_ids])
372
for change in iter_changes:
373
# This probably looks up in basis_inv way to much.
374
if change[1][0] is not None:
375
head_candidate = [basis_inv[change[0]].revision]
378
changes[change[0]] = change, merged_ids.get(change[0],
380
unchanged_merged = set(merged_ids) - set(changes)
381
# Extend the changes dict with synthetic changes to record merges of
383
for file_id in unchanged_merged:
384
# Record a merged version of these items that did not change vs the
385
# basis. This can be either identical parallel changes, or a revert
386
# of a specific file after a merge. The recorded content will be
387
# that of the current tree (which is the same as the basis), but
388
# the per-file graph will reflect a merge.
389
# NB:XXX: We are reconstructing path information we had, this
390
# should be preserved instead.
391
# inv delta change: (file_id, (path_in_source, path_in_target),
392
# changed_content, versioned, parent, name, kind,
395
basis_entry = basis_inv[file_id]
396
except errors.NoSuchId:
397
# a change from basis->some_parents but file_id isn't in basis
398
# so was new in the merge, which means it must have changed
399
# from basis -> current, and as it hasn't the add was reverted
400
# by the user. So we discard this change.
404
(basis_inv.id2path(file_id), tree.id2path(file_id)),
406
(basis_entry.parent_id, basis_entry.parent_id),
407
(basis_entry.name, basis_entry.name),
408
(basis_entry.kind, basis_entry.kind),
409
(basis_entry.executable, basis_entry.executable))
410
changes[file_id] = (change, merged_ids[file_id])
411
# changes contains tuples with the change and a set of inventory
412
# candidates for the file.
414
# old_path, new_path, file_id, new_inventory_entry
415
seen_root = False # Is the root in the basis delta?
416
inv_delta = self._basis_delta
417
modified_rev = self._new_revision_id
418
for change, head_candidates in viewvalues(changes):
419
if change[3][1]: # versioned in target.
420
# Several things may be happening here:
421
# We may have a fork in the per-file graph
422
# - record a change with the content from tree
423
# We may have a change against < all trees
424
# - carry over the tree that hasn't changed
425
# We may have a change against all trees
426
# - record the change with the content from tree
429
entry = _entry_factory[kind](file_id, change[5][1],
431
head_set = self._heads(change[0], set(head_candidates))
434
for head_candidate in head_candidates:
435
if head_candidate in head_set:
436
heads.append(head_candidate)
437
head_set.remove(head_candidate)
440
# Could be a carry-over situation:
441
parent_entry_revs = parent_entries.get(file_id, None)
442
if parent_entry_revs:
443
parent_entry = parent_entry_revs.get(heads[0], None)
446
if parent_entry is None:
447
# The parent iter_changes was called against is the one
448
# that is the per-file head, so any change is relevant
449
# iter_changes is valid.
450
carry_over_possible = False
452
# could be a carry over situation
453
# A change against the basis may just indicate a merge,
454
# we need to check the content against the source of the
455
# merge to determine if it was changed after the merge
457
if (parent_entry.kind != entry.kind or
458
parent_entry.parent_id != entry.parent_id or
459
parent_entry.name != entry.name):
460
# Metadata common to all entries has changed
461
# against per-file parent
462
carry_over_possible = False
464
carry_over_possible = True
465
# per-type checks for changes against the parent_entry
468
# Cannot be a carry-over situation
469
carry_over_possible = False
470
# Populate the entry in the delta
472
# XXX: There is still a small race here: If someone reverts the content of a file
473
# after iter_changes examines and decides it has changed,
474
# we will unconditionally record a new version even if some
475
# other process reverts it while commit is running (with
476
# the revert happening after iter_changes did its
479
entry.executable = True
481
entry.executable = False
482
if (carry_over_possible and
483
parent_entry.executable == entry.executable):
484
# Check the file length, content hash after reading
486
nostore_sha = parent_entry.text_sha1
489
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
491
text = file_obj.read()
495
entry.text_sha1, entry.text_size = self._add_text_to_weave(
496
file_id, text, heads, nostore_sha)
497
yield file_id, change[1][1], (entry.text_sha1, stat_value)
498
except errors.ExistingContent:
499
# No content change against a carry_over parent
500
# Perhaps this should also yield a fs hash update?
502
entry.text_size = parent_entry.text_size
503
entry.text_sha1 = parent_entry.text_sha1
504
elif kind == 'symlink':
506
entry.symlink_target = tree.get_symlink_target(file_id)
507
if (carry_over_possible and
508
parent_entry.symlink_target == entry.symlink_target):
511
self._add_text_to_weave(change[0], '', heads, None)
512
elif kind == 'directory':
513
if carry_over_possible:
516
# Nothing to set on the entry.
517
# XXX: split into the Root and nonRoot versions.
518
if change[1][1] != '' or self.repository.supports_rich_root():
519
self._add_text_to_weave(change[0], '', heads, None)
520
elif kind == 'tree-reference':
521
if not self.repository._format.supports_tree_reference:
522
# This isn't quite sane as an error, but we shouldn't
523
# ever see this code path in practice: tree's don't
524
# permit references when the repo doesn't support tree
526
raise errors.UnsupportedOperation(tree.add_reference,
528
reference_revision = tree.get_reference_revision(change[0])
529
entry.reference_revision = reference_revision
530
if (carry_over_possible and
531
parent_entry.reference_revision == reference_revision):
534
self._add_text_to_weave(change[0], '', heads, None)
536
raise AssertionError('unknown kind %r' % kind)
538
entry.revision = modified_rev
540
entry.revision = parent_entry.revision
543
new_path = change[1][1]
544
inv_delta.append((change[1][0], new_path, change[0], entry))
547
# The initial commit adds a root directory, but this in itself is not
548
# a worthwhile commit.
549
if ((len(inv_delta) > 0 and basis_revision_id != _mod_revision.NULL_REVISION) or
550
(len(inv_delta) > 1 and basis_revision_id == _mod_revision.NULL_REVISION)):
551
# This should perhaps be guarded by a check that the basis we
552
# commit against is the basis for the commit and if not do a delta
554
self._any_changes = True
556
# housekeeping root entry changes do not affect no-change commits.
557
self._require_root_change(tree)
558
self.basis_delta_revision = basis_revision_id
560
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
561
parent_keys = tuple([(file_id, parent) for parent in parents])
562
return self.repository.texts._add_text(
563
(file_id, self._new_revision_id), parent_keys, new_text,
564
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
567
class VersionedFileRootCommitBuilder(VersionedFileCommitBuilder):
568
"""This commitbuilder actually records the root id"""
570
# the root entry gets versioned properly by this builder.
571
_versioned_root = True
573
def _require_root_change(self, tree):
574
"""Enforce an appropriate root object change.
576
This is called once when record_iter_changes is called, if and only if
577
the root was not in the delta calculated by record_iter_changes.
579
:param tree: The tree which is being committed.
581
# versioned roots do not change unless the tree found a change.
584
class VersionedFileRepository(Repository):
585
"""Repository holding history for one or more branches.
587
The repository holds and retrieves historical information including
588
revisions and file history. It's normally accessed only by the Branch,
589
which views a particular line of development through that history.
591
The Repository builds on top of some byte storage facilies (the revisions,
592
signatures, inventories, texts and chk_bytes attributes) and a Transport,
593
which respectively provide byte storage and a means to access the (possibly
596
The byte storage facilities are addressed via tuples, which we refer to
597
as 'keys' throughout the code base. Revision_keys, inventory_keys and
598
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
599
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
600
byte string made up of a hash identifier and a hash value.
601
We use this interface because it allows low friction with the underlying
602
code that implements disk indices, network encoding and other parts of
605
:ivar revisions: A breezy.versionedfile.VersionedFiles instance containing
606
the serialised revisions for the repository. This can be used to obtain
607
revision graph information or to access raw serialised revisions.
608
The result of trying to insert data into the repository via this store
609
is undefined: it should be considered read-only except for implementors
611
:ivar signatures: A breezy.versionedfile.VersionedFiles instance containing
612
the serialised signatures for the repository. This can be used to
613
obtain access to raw serialised signatures. The result of trying to
614
insert data into the repository via this store is undefined: it should
615
be considered read-only except for implementors of repositories.
616
:ivar inventories: A breezy.versionedfile.VersionedFiles instance containing
617
the serialised inventories for the repository. This can be used to
618
obtain unserialised inventories. The result of trying to insert data
619
into the repository via this store is undefined: it should be
620
considered read-only except for implementors of repositories.
621
:ivar texts: A breezy.versionedfile.VersionedFiles instance containing the
622
texts of files and directories for the repository. This can be used to
623
obtain file texts or file graphs. Note that Repository.iter_file_bytes
624
is usually a better interface for accessing file texts.
625
The result of trying to insert data into the repository via this store
626
is undefined: it should be considered read-only except for implementors
628
:ivar chk_bytes: A breezy.versionedfile.VersionedFiles instance containing
629
any data the repository chooses to store or have indexed by its hash.
630
The result of trying to insert data into the repository via this store
631
is undefined: it should be considered read-only except for implementors
633
:ivar _transport: Transport for file access to repository, typically
634
pointing to .bzr/repository.
637
# What class to use for a CommitBuilder. Often it's simpler to change this
638
# in a Repository class subclass rather than to override
639
# get_commit_builder.
640
_commit_builder_class = VersionedFileCommitBuilder
642
def add_fallback_repository(self, repository):
643
"""Add a repository to use for looking up data not held locally.
645
:param repository: A repository.
647
if not self._format.supports_external_lookups:
648
raise errors.UnstackableRepositoryFormat(self._format, self.base)
649
# This can raise an exception, so should be done before we lock the
650
# fallback repository.
651
self._check_fallback_repository(repository)
653
# This repository will call fallback.unlock() when we transition to
654
# the unlocked state, so we make sure to increment the lock count
655
repository.lock_read()
656
self._fallback_repositories.append(repository)
657
self.texts.add_fallback_versioned_files(repository.texts)
658
self.inventories.add_fallback_versioned_files(repository.inventories)
659
self.revisions.add_fallback_versioned_files(repository.revisions)
660
self.signatures.add_fallback_versioned_files(repository.signatures)
661
if self.chk_bytes is not None:
662
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
664
@only_raises(errors.LockNotHeld, errors.LockBroken)
666
super(VersionedFileRepository, self).unlock()
667
if self.control_files._lock_count == 0:
668
self._inventory_entry_cache.clear()
670
def add_inventory(self, revision_id, inv, parents):
671
"""Add the inventory inv to the repository as revision_id.
673
:param parents: The revision ids of the parents that revision_id
674
is known to have and are in the repository already.
676
:returns: The validator(which is a sha1 digest, though what is sha'd is
677
repository format specific) of the serialized inventory.
679
if not self.is_in_write_group():
680
raise AssertionError("%r not in write group" % (self,))
681
_mod_revision.check_not_reserved_id(revision_id)
682
if not (inv.revision_id is None or inv.revision_id == revision_id):
683
raise AssertionError(
684
"Mismatch between inventory revision"
685
" id and insertion revid (%r, %r)"
686
% (inv.revision_id, revision_id))
688
raise errors.RootMissing()
689
return self._add_inventory_checked(revision_id, inv, parents)
691
def _add_inventory_checked(self, revision_id, inv, parents):
692
"""Add inv to the repository after checking the inputs.
694
This function can be overridden to allow different inventory styles.
696
:seealso: add_inventory, for the contract.
698
inv_lines = self._serializer.write_inventory_to_lines(inv)
699
return self._inventory_add_lines(revision_id, parents,
700
inv_lines, check_content=False)
702
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
703
parents, basis_inv=None, propagate_caches=False):
704
"""Add a new inventory expressed as a delta against another revision.
706
See the inventory developers documentation for the theory behind
709
:param basis_revision_id: The inventory id the delta was created
710
against. (This does not have to be a direct parent.)
711
:param delta: The inventory delta (see Inventory.apply_delta for
713
:param new_revision_id: The revision id that the inventory is being
715
:param parents: The revision ids of the parents that revision_id is
716
known to have and are in the repository already. These are supplied
717
for repositories that depend on the inventory graph for revision
718
graph access, as well as for those that pun ancestry with delta
720
:param basis_inv: The basis inventory if it is already known,
722
:param propagate_caches: If True, the caches for this inventory are
723
copied to and updated for the result if possible.
725
:returns: (validator, new_inv)
726
The validator(which is a sha1 digest, though what is sha'd is
727
repository format specific) of the serialized inventory, and the
730
if not self.is_in_write_group():
731
raise AssertionError("%r not in write group" % (self,))
732
_mod_revision.check_not_reserved_id(new_revision_id)
733
basis_tree = self.revision_tree(basis_revision_id)
734
basis_tree.lock_read()
736
# Note that this mutates the inventory of basis_tree, which not all
737
# inventory implementations may support: A better idiom would be to
738
# return a new inventory, but as there is no revision tree cache in
739
# repository this is safe for now - RBC 20081013
740
if basis_inv is None:
741
basis_inv = basis_tree.root_inventory
742
basis_inv.apply_delta(delta)
743
basis_inv.revision_id = new_revision_id
744
return (self.add_inventory(new_revision_id, basis_inv, parents),
749
def _inventory_add_lines(self, revision_id, parents, lines,
751
"""Store lines in inv_vf and return the sha1 of the inventory."""
752
parents = [(parent,) for parent in parents]
753
result = self.inventories.add_lines((revision_id,), parents, lines,
754
check_content=check_content)[0]
755
self.inventories._access.flush()
758
def add_revision(self, revision_id, rev, inv=None):
759
"""Add rev to the revision store as revision_id.
761
:param revision_id: the revision id to use.
762
:param rev: The revision object.
763
:param inv: The inventory for the revision. if None, it will be looked
764
up in the inventory storer
766
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
768
_mod_revision.check_not_reserved_id(revision_id)
769
# check inventory present
770
if not self.inventories.get_parent_map([(revision_id,)]):
772
raise errors.WeaveRevisionNotPresent(revision_id,
775
# yes, this is not suitable for adding with ghosts.
776
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
780
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
781
self._add_revision(rev)
783
def _add_revision(self, revision):
784
text = self._serializer.write_revision_to_string(revision)
785
key = (revision.revision_id,)
786
parents = tuple((parent,) for parent in revision.parent_ids)
787
self.revisions.add_lines(key, parents, osutils.split_lines(text))
789
def _check_inventories(self, checker):
790
"""Check the inventories found from the revision scan.
792
This is responsible for verifying the sha1 of inventories and
793
creating a pending_keys set that covers data referenced by inventories.
795
bar = ui.ui_factory.nested_progress_bar()
797
self._do_check_inventories(checker, bar)
801
def _do_check_inventories(self, checker, bar):
802
"""Helper for _check_inventories."""
804
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
805
kinds = ['chk_bytes', 'texts']
806
count = len(checker.pending_keys)
807
bar.update(gettext("inventories"), 0, 2)
808
current_keys = checker.pending_keys
809
checker.pending_keys = {}
810
# Accumulate current checks.
811
for key in current_keys:
812
if key[0] != 'inventories' and key[0] not in kinds:
813
checker._report_items.append('unknown key type %r' % (key,))
814
keys[key[0]].add(key[1:])
815
if keys['inventories']:
816
# NB: output order *should* be roughly sorted - topo or
817
# inverse topo depending on repository - either way decent
818
# to just delta against. However, pre-CHK formats didn't
819
# try to optimise inventory layout on disk. As such the
820
# pre-CHK code path does not use inventory deltas.
822
for record in self.inventories.check(keys=keys['inventories']):
823
if record.storage_kind == 'absent':
824
checker._report_items.append(
825
'Missing inventory {%s}' % (record.key,))
827
last_object = self._check_record('inventories', record,
828
checker, last_object,
829
current_keys[('inventories',) + record.key])
830
del keys['inventories']
833
bar.update(gettext("texts"), 1)
834
while (checker.pending_keys or keys['chk_bytes']
836
# Something to check.
837
current_keys = checker.pending_keys
838
checker.pending_keys = {}
839
# Accumulate current checks.
840
for key in current_keys:
841
if key[0] not in kinds:
842
checker._report_items.append('unknown key type %r' % (key,))
843
keys[key[0]].add(key[1:])
844
# Check the outermost kind only - inventories || chk_bytes || texts
848
for record in getattr(self, kind).check(keys=keys[kind]):
849
if record.storage_kind == 'absent':
850
checker._report_items.append(
851
'Missing %s {%s}' % (kind, record.key,))
853
last_object = self._check_record(kind, record,
854
checker, last_object, current_keys[(kind,) + record.key])
858
def _check_record(self, kind, record, checker, last_object, item_data):
859
"""Check a single text from this repository."""
860
if kind == 'inventories':
861
rev_id = record.key[0]
862
inv = self._deserialise_inventory(rev_id,
863
record.get_bytes_as('fulltext'))
864
if last_object is not None:
865
delta = inv._make_delta(last_object)
866
for old_path, path, file_id, ie in delta:
869
ie.check(checker, rev_id, inv)
871
for path, ie in inv.iter_entries():
872
ie.check(checker, rev_id, inv)
873
if self._format.fast_deltas:
875
elif kind == 'chk_bytes':
876
# No code written to check chk_bytes for this repo format.
877
checker._report_items.append(
878
'unsupported key type chk_bytes for %s' % (record.key,))
879
elif kind == 'texts':
880
self._check_text(record, checker, item_data)
882
checker._report_items.append(
883
'unknown key type %s for %s' % (kind, record.key))
885
def _check_text(self, record, checker, item_data):
886
"""Check a single text."""
887
# Check it is extractable.
888
# TODO: check length.
889
if record.storage_kind == 'chunked':
890
chunks = record.get_bytes_as(record.storage_kind)
891
sha1 = osutils.sha_strings(chunks)
892
length = sum(map(len, chunks))
894
content = record.get_bytes_as('fulltext')
895
sha1 = osutils.sha_string(content)
896
length = len(content)
897
if item_data and sha1 != item_data[1]:
898
checker._report_items.append(
899
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
900
(record.key, sha1, item_data[1], item_data[2]))
903
def _eliminate_revisions_not_present(self, revision_ids):
904
"""Check every revision id in revision_ids to see if we have it.
906
Returns a set of the present revisions.
909
graph = self.get_graph()
910
parent_map = graph.get_parent_map(revision_ids)
911
# The old API returned a list, should this actually be a set?
912
return list(parent_map)
914
def __init__(self, _format, a_controldir, control_files):
915
"""Instantiate a VersionedFileRepository.
917
:param _format: The format of the repository on disk.
918
:param controldir: The ControlDir of the repository.
919
:param control_files: Control files to use for locking, etc.
921
# In the future we will have a single api for all stores for
922
# getting file texts, inventories and revisions, then
923
# this construct will accept instances of those things.
924
super(VersionedFileRepository, self).__init__(_format, a_controldir,
926
self._transport = control_files._transport
927
self.base = self._transport.base
929
self._reconcile_does_inventory_gc = True
930
self._reconcile_fixes_text_parents = False
931
self._reconcile_backsup_inventory = True
932
# An InventoryEntry cache, used during deserialization
933
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
934
# Is it safe to return inventory entries directly from the entry cache,
935
# rather copying them?
936
self._safe_to_return_from_cache = False
938
def fetch(self, source, revision_id=None, find_ghosts=False,
940
"""Fetch the content required to construct revision_id from source.
942
If revision_id is None and fetch_spec is None, then all content is
945
fetch() may not be used when the repository is in a write group -
946
either finish the current write group before using fetch, or use
947
fetch before starting the write group.
949
:param find_ghosts: Find and copy revisions in the source that are
950
ghosts in the target (and not reachable directly by walking out to
951
the first-present revision in target from revision_id).
952
:param revision_id: If specified, all the content needed for this
953
revision ID will be copied to the target. Fetch will determine for
954
itself which content needs to be copied.
955
:param fetch_spec: If specified, a SearchResult or
956
PendingAncestryResult that describes which revisions to copy. This
957
allows copying multiple heads at once. Mutually exclusive with
960
if fetch_spec is not None and revision_id is not None:
961
raise AssertionError(
962
"fetch_spec and revision_id are mutually exclusive.")
963
if self.is_in_write_group():
964
raise errors.InternalBzrError(
965
"May not fetch while in a write group.")
966
# fast path same-url fetch operations
967
# TODO: lift out to somewhere common with RemoteRepository
968
# <https://bugs.launchpad.net/bzr/+bug/401646>
969
if (self.has_same_location(source)
970
and fetch_spec is None
971
and self._has_same_fallbacks(source)):
972
# check that last_revision is in 'from' and then return a
974
if (revision_id is not None and
975
not _mod_revision.is_null(revision_id)):
976
self.get_revision(revision_id)
978
inter = InterRepository.get(source, self)
979
if (fetch_spec is not None and
980
not getattr(inter, "supports_fetch_spec", False)):
981
raise errors.UnsupportedOperation(
982
"fetch_spec not supported for %r" % inter)
983
return inter.fetch(revision_id=revision_id,
984
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
987
def gather_stats(self, revid=None, committers=None):
988
"""See Repository.gather_stats()."""
989
result = super(VersionedFileRepository, self).gather_stats(revid, committers)
990
# now gather global repository information
991
# XXX: This is available for many repos regardless of listability.
992
if self.user_transport.listable():
993
# XXX: do we want to __define len__() ?
994
# Maybe the versionedfiles object should provide a different
995
# method to get the number of keys.
996
result['revisions'] = len(self.revisions.keys())
1000
def get_commit_builder(self, branch, parents, config_stack, timestamp=None,
1001
timezone=None, committer=None, revprops=None,
1002
revision_id=None, lossy=False):
1003
"""Obtain a CommitBuilder for this repository.
1005
:param branch: Branch to commit to.
1006
:param parents: Revision ids of the parents of the new revision.
1007
:param config_stack: Configuration stack to use.
1008
:param timestamp: Optional timestamp recorded for commit.
1009
:param timezone: Optional timezone for timestamp.
1010
:param committer: Optional committer to set for commit.
1011
:param revprops: Optional dictionary of revision properties.
1012
:param revision_id: Optional revision id.
1013
:param lossy: Whether to discard data that can not be natively
1014
represented, when pushing to a foreign VCS
1016
if self._fallback_repositories and not self._format.supports_chks:
1017
raise errors.BzrError("Cannot commit directly to a stacked branch"
1018
" in pre-2a formats. See "
1019
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1020
result = self._commit_builder_class(self, parents, config_stack,
1021
timestamp, timezone, committer, revprops, revision_id,
1023
self.start_write_group()
1026
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1027
"""Return the keys of missing inventory parents for revisions added in
1030
A revision is not complete if the inventory delta for that revision
1031
cannot be calculated. Therefore if the parent inventories of a
1032
revision are not present, the revision is incomplete, and e.g. cannot
1033
be streamed by a smart server. This method finds missing inventory
1034
parents for revisions added in this write group.
1036
if not self._format.supports_external_lookups:
1037
# This is only an issue for stacked repositories
1039
if not self.is_in_write_group():
1040
raise AssertionError('not in a write group')
1042
# XXX: We assume that every added revision already has its
1043
# corresponding inventory, so we only check for parent inventories that
1044
# might be missing, rather than all inventories.
1045
parents = set(self.revisions._index.get_missing_parents())
1046
parents.discard(_mod_revision.NULL_REVISION)
1047
unstacked_inventories = self.inventories._index
1048
present_inventories = unstacked_inventories.get_parent_map(
1049
key[-1:] for key in parents)
1050
parents.difference_update(present_inventories)
1051
if len(parents) == 0:
1052
# No missing parent inventories.
1054
if not check_for_missing_texts:
1055
return set(('inventories', rev_id) for (rev_id,) in parents)
1056
# Ok, now we have a list of missing inventories. But these only matter
1057
# if the inventories that reference them are missing some texts they
1058
# appear to introduce.
1059
# XXX: Texts referenced by all added inventories need to be present,
1060
# but at the moment we're only checking for texts referenced by
1061
# inventories at the graph's edge.
1062
key_deps = self.revisions._index._key_dependencies
1063
key_deps.satisfy_refs_for_keys(present_inventories)
1064
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1065
file_ids = self.fileids_altered_by_revision_ids(referrers)
1066
missing_texts = set()
1067
for file_id, version_ids in viewitems(file_ids):
1068
missing_texts.update(
1069
(file_id, version_id) for version_id in version_ids)
1070
present_texts = self.texts.get_parent_map(missing_texts)
1071
missing_texts.difference_update(present_texts)
1072
if not missing_texts:
1073
# No texts are missing, so all revisions and their deltas are
1076
# Alternatively the text versions could be returned as the missing
1077
# keys, but this is likely to be less data.
1078
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1082
def has_revisions(self, revision_ids):
1083
"""Probe to find out the presence of multiple revisions.
1085
:param revision_ids: An iterable of revision_ids.
1086
:return: A set of the revision_ids that were present.
1088
parent_map = self.revisions.get_parent_map(
1089
[(rev_id,) for rev_id in revision_ids])
1091
if _mod_revision.NULL_REVISION in revision_ids:
1092
result.add(_mod_revision.NULL_REVISION)
1093
result.update([key[0] for key in parent_map])
1097
def get_revision_reconcile(self, revision_id):
1098
"""'reconcile' helper routine that allows access to a revision always.
1100
This variant of get_revision does not cross check the weave graph
1101
against the revision one as get_revision does: but it should only
1102
be used by reconcile, or reconcile-alike commands that are correcting
1103
or testing the revision graph.
1105
return self._get_revisions([revision_id])[0]
1108
def get_revisions(self, revision_ids):
1109
"""Get many revisions at once.
1111
Repositories that need to check data on every revision read should
1112
subclass this method.
1114
return self._get_revisions(revision_ids)
1117
def _get_revisions(self, revision_ids):
1118
"""Core work logic to get many revisions without sanity checks."""
1120
for revid, rev in self._iter_revisions(revision_ids):
1122
raise errors.NoSuchRevision(self, revid)
1124
return [revs[revid] for revid in revision_ids]
1126
def _iter_revisions(self, revision_ids):
1127
"""Iterate over revision objects.
1129
:param revision_ids: An iterable of revisions to examine. None may be
1130
passed to request all revisions known to the repository. Note that
1131
not all repositories can find unreferenced revisions; for those
1132
repositories only referenced ones will be returned.
1133
:return: An iterator of (revid, revision) tuples. Absent revisions (
1134
those asked for but not available) are returned as (revid, None).
1136
if revision_ids is None:
1137
revision_ids = self.all_revision_ids()
1139
for rev_id in revision_ids:
1140
if not rev_id or not isinstance(rev_id, bytes):
1141
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1142
keys = [(key,) for key in revision_ids]
1143
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1144
for record in stream:
1145
revid = record.key[0]
1146
if record.storage_kind == 'absent':
1149
text = record.get_bytes_as('fulltext')
1150
rev = self._serializer.read_revision_from_string(text)
1154
def add_signature_text(self, revision_id, signature):
1155
"""Store a signature text for a revision.
1157
:param revision_id: Revision id of the revision
1158
:param signature: Signature text.
1160
self.signatures.add_lines((revision_id,), (),
1161
osutils.split_lines(signature))
1163
def find_text_key_references(self):
1164
"""Find the text key references within the repository.
1166
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1167
to whether they were referred to by the inventory of the
1168
revision_id that they contain. The inventory texts from all present
1169
revision ids are assessed to generate this report.
1171
revision_keys = self.revisions.keys()
1172
w = self.inventories
1173
pb = ui.ui_factory.nested_progress_bar()
1175
return self._serializer._find_text_key_references(
1176
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1180
def _inventory_xml_lines_for_keys(self, keys):
1181
"""Get a line iterator of the sort needed for findind references.
1183
Not relevant for non-xml inventory repositories.
1185
Ghosts in revision_keys are ignored.
1187
:param revision_keys: The revision keys for the inventories to inspect.
1188
:return: An iterator over (inventory line, revid) for the fulltexts of
1189
all of the xml inventories specified by revision_keys.
1191
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1192
for record in stream:
1193
if record.storage_kind != 'absent':
1194
chunks = record.get_bytes_as('chunked')
1195
revid = record.key[-1]
1196
lines = osutils.chunks_to_lines(chunks)
1200
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1202
"""Helper routine for fileids_altered_by_revision_ids.
1204
This performs the translation of xml lines to revision ids.
1206
:param line_iterator: An iterator of lines, origin_version_id
1207
:param revision_keys: The revision ids to filter for. This should be a
1208
set or other type which supports efficient __contains__ lookups, as
1209
the revision key from each parsed line will be looked up in the
1210
revision_keys filter.
1211
:return: a dictionary mapping altered file-ids to an iterable of
1212
revision_ids. Each altered file-ids has the exact revision_ids that
1213
altered it listed explicitly.
1215
seen = set(self._serializer._find_text_key_references(line_iterator))
1216
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
1217
parent_seen = set(self._serializer._find_text_key_references(
1218
self._inventory_xml_lines_for_keys(parent_keys)))
1219
new_keys = seen - parent_seen
1221
setdefault = result.setdefault
1222
for key in new_keys:
1223
setdefault(key[0], set()).add(key[-1])
1226
def _find_parent_keys_of_revisions(self, revision_keys):
1227
"""Similar to _find_parent_ids_of_revisions, but used with keys.
1229
:param revision_keys: An iterable of revision_keys.
1230
:return: The parents of all revision_keys that are not already in
1233
parent_map = self.revisions.get_parent_map(revision_keys)
1234
parent_keys = set(itertools.chain.from_iterable(
1235
viewvalues(parent_map)))
1236
parent_keys.difference_update(revision_keys)
1237
parent_keys.discard(_mod_revision.NULL_REVISION)
1240
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1241
"""Find the file ids and versions affected by revisions.
1243
:param revisions: an iterable containing revision ids.
1244
:param _inv_weave: The inventory weave from this repository or None.
1245
If None, the inventory weave will be opened automatically.
1246
:return: a dictionary mapping altered file-ids to an iterable of
1247
revision_ids. Each altered file-ids has the exact revision_ids that
1248
altered it listed explicitly.
1250
selected_keys = set((revid,) for revid in revision_ids)
1251
w = _inv_weave or self.inventories
1252
return self._find_file_ids_from_xml_inventory_lines(
1253
w.iter_lines_added_or_present_in_keys(
1254
selected_keys, pb=None),
1257
def iter_files_bytes(self, desired_files):
1258
"""Iterate through file versions.
1260
Files will not necessarily be returned in the order they occur in
1261
desired_files. No specific order is guaranteed.
1263
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1264
value supplied by the caller as part of desired_files. It should
1265
uniquely identify the file version in the caller's context. (Examples:
1266
an index number or a TreeTransform trans_id.)
1268
bytes_iterator is an iterable of bytestrings for the file. The
1269
kind of iterable and length of the bytestrings are unspecified, but for
1270
this implementation, it is a list of bytes produced by
1271
VersionedFile.get_record_stream().
1273
:param desired_files: a list of (file_id, revision_id, identifier)
1277
for file_id, revision_id, callable_data in desired_files:
1278
text_keys[(file_id, revision_id)] = callable_data
1279
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1280
if record.storage_kind == 'absent':
1281
raise errors.RevisionNotPresent(record.key[1], record.key[0])
1282
yield text_keys[record.key], record.get_bytes_as('chunked')
1284
def _generate_text_key_index(self, text_key_references=None,
1286
"""Generate a new text key index for the repository.
1288
This is an expensive function that will take considerable time to run.
1290
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1291
list of parents, also text keys. When a given key has no parents,
1292
the parents list will be [NULL_REVISION].
1294
# All revisions, to find inventory parents.
1295
if ancestors is None:
1296
graph = self.get_graph()
1297
ancestors = graph.get_parent_map(self.all_revision_ids())
1298
if text_key_references is None:
1299
text_key_references = self.find_text_key_references()
1300
pb = ui.ui_factory.nested_progress_bar()
1302
return self._do_generate_text_key_index(ancestors,
1303
text_key_references, pb)
1307
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1308
"""Helper for _generate_text_key_index to avoid deep nesting."""
1309
revision_order = tsort.topo_sort(ancestors)
1310
invalid_keys = set()
1312
for revision_id in revision_order:
1313
revision_keys[revision_id] = set()
1314
text_count = len(text_key_references)
1315
# a cache of the text keys to allow reuse; costs a dict of all the
1316
# keys, but saves a 2-tuple for every child of a given key.
1318
for text_key, valid in viewitems(text_key_references):
1320
invalid_keys.add(text_key)
1322
revision_keys[text_key[1]].add(text_key)
1323
text_key_cache[text_key] = text_key
1324
del text_key_references
1326
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1327
NULL_REVISION = _mod_revision.NULL_REVISION
1328
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1329
# too small for large or very branchy trees. However, for 55K path
1330
# trees, it would be easy to use too much memory trivially. Ideally we
1331
# could gauge this by looking at available real memory etc, but this is
1332
# always a tricky proposition.
1333
inventory_cache = lru_cache.LRUCache(10)
1334
batch_size = 10 # should be ~150MB on a 55K path tree
1335
batch_count = len(revision_order) / batch_size + 1
1337
pb.update(gettext("Calculating text parents"), processed_texts, text_count)
1338
for offset in range(batch_count):
1339
to_query = revision_order[offset * batch_size:(offset + 1) *
1343
for revision_id in to_query:
1344
parent_ids = ancestors[revision_id]
1345
for text_key in revision_keys[revision_id]:
1346
pb.update(gettext("Calculating text parents"), processed_texts)
1347
processed_texts += 1
1348
candidate_parents = []
1349
for parent_id in parent_ids:
1350
parent_text_key = (text_key[0], parent_id)
1352
check_parent = parent_text_key not in \
1353
revision_keys[parent_id]
1355
# the parent parent_id is a ghost:
1356
check_parent = False
1357
# truncate the derived graph against this ghost.
1358
parent_text_key = None
1360
# look at the parent commit details inventories to
1361
# determine possible candidates in the per file graph.
1364
inv = inventory_cache[parent_id]
1366
inv = self.revision_tree(parent_id).root_inventory
1367
inventory_cache[parent_id] = inv
1369
parent_entry = inv[text_key[0]]
1370
except (KeyError, errors.NoSuchId):
1372
if parent_entry is not None:
1374
text_key[0], parent_entry.revision)
1376
parent_text_key = None
1377
if parent_text_key is not None:
1378
candidate_parents.append(
1379
text_key_cache[parent_text_key])
1380
parent_heads = text_graph.heads(candidate_parents)
1381
new_parents = list(parent_heads)
1382
new_parents.sort(key=lambda x:candidate_parents.index(x))
1383
if new_parents == []:
1384
new_parents = [NULL_REVISION]
1385
text_index[text_key] = new_parents
1387
for text_key in invalid_keys:
1388
text_index[text_key] = [NULL_REVISION]
1391
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
1392
"""Get an iterable listing the keys of all the data introduced by a set
1395
The keys will be ordered so that the corresponding items can be safely
1396
fetched and inserted in that order.
1398
:returns: An iterable producing tuples of (knit-kind, file-id,
1399
versions). knit-kind is one of 'file', 'inventory', 'signatures',
1400
'revisions'. file-id is None unless knit-kind is 'file'.
1402
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
1405
for result in self._find_non_file_keys_to_fetch(revision_ids):
1408
def _find_file_keys_to_fetch(self, revision_ids, pb):
1409
# XXX: it's a bit weird to control the inventory weave caching in this
1410
# generator. Ideally the caching would be done in fetch.py I think. Or
1411
# maybe this generator should explicitly have the contract that it
1412
# should not be iterated until the previously yielded item has been
1414
inv_w = self.inventories
1416
# file ids that changed
1417
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
1419
num_file_ids = len(file_ids)
1420
for file_id, altered_versions in viewitems(file_ids):
1422
pb.update(gettext("Fetch texts"), count, num_file_ids)
1424
yield ("file", file_id, altered_versions)
1426
def _find_non_file_keys_to_fetch(self, revision_ids):
1428
yield ("inventory", None, revision_ids)
1431
# XXX: Note ATM no callers actually pay attention to this return
1432
# instead they just use the list of revision ids and ignore
1433
# missing sigs. Consider removing this work entirely
1434
revisions_with_signatures = set(self.signatures.get_parent_map(
1435
[(r,) for r in revision_ids]))
1436
revisions_with_signatures = {r for (r,) in revisions_with_signatures}
1437
revisions_with_signatures.intersection_update(revision_ids)
1438
yield ("signatures", None, revisions_with_signatures)
1441
yield ("revisions", None, revision_ids)
1444
def get_inventory(self, revision_id):
1445
"""Get Inventory object by revision id."""
1446
return next(self.iter_inventories([revision_id]))
1448
def iter_inventories(self, revision_ids, ordering=None):
1449
"""Get many inventories by revision_ids.
1451
This will buffer some or all of the texts used in constructing the
1452
inventories in memory, but will only parse a single inventory at a
1455
:param revision_ids: The expected revision ids of the inventories.
1456
:param ordering: optional ordering, e.g. 'topological'. If not
1457
specified, the order of revision_ids will be preserved (by
1458
buffering if necessary).
1459
:return: An iterator of inventories.
1461
if ((None in revision_ids)
1462
or (_mod_revision.NULL_REVISION in revision_ids)):
1463
raise ValueError('cannot get null revision inventory')
1464
for inv, revid in self._iter_inventories(revision_ids, ordering):
1466
raise errors.NoSuchRevision(self, revid)
1469
def _iter_inventories(self, revision_ids, ordering):
1470
"""single-document based inventory iteration."""
1471
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
1472
for text, revision_id in inv_xmls:
1474
yield None, revision_id
1476
yield self._deserialise_inventory(revision_id, text), revision_id
1478
def _iter_inventory_xmls(self, revision_ids, ordering):
1479
if ordering is None:
1480
order_as_requested = True
1481
ordering = 'unordered'
1483
order_as_requested = False
1484
keys = [(revision_id,) for revision_id in revision_ids]
1487
if order_as_requested:
1488
key_iter = iter(keys)
1489
next_key = next(key_iter)
1490
stream = self.inventories.get_record_stream(keys, ordering, True)
1492
for record in stream:
1493
if record.storage_kind != 'absent':
1494
chunks = record.get_bytes_as('chunked')
1495
if order_as_requested:
1496
text_chunks[record.key] = chunks
1498
yield ''.join(chunks), record.key[-1]
1500
yield None, record.key[-1]
1501
if order_as_requested:
1502
# Yield as many results as we can while preserving order.
1503
while next_key in text_chunks:
1504
chunks = text_chunks.pop(next_key)
1505
yield ''.join(chunks), next_key[-1]
1507
next_key = next(key_iter)
1508
except StopIteration:
1509
# We still want to fully consume the get_record_stream,
1510
# just in case it is not actually finished at this point
1514
def _deserialise_inventory(self, revision_id, xml):
1515
"""Transform the xml into an inventory object.
1517
:param revision_id: The expected revision id of the inventory.
1518
:param xml: A serialised inventory.
1520
result = self._serializer.read_inventory_from_string(xml, revision_id,
1521
entry_cache=self._inventory_entry_cache,
1522
return_from_cache=self._safe_to_return_from_cache)
1523
if result.revision_id != revision_id:
1524
raise AssertionError('revision id mismatch %s != %s' % (
1525
result.revision_id, revision_id))
1528
def get_serializer_format(self):
1529
return self._serializer.format_num
1532
def _get_inventory_xml(self, revision_id):
1533
"""Get serialized inventory as a string."""
1534
texts = self._iter_inventory_xmls([revision_id], 'unordered')
1535
text, revision_id = next(texts)
1537
raise errors.NoSuchRevision(self, revision_id)
1541
def revision_tree(self, revision_id):
1542
"""Return Tree for a revision on this branch.
1544
`revision_id` may be NULL_REVISION for the empty tree revision.
1546
revision_id = _mod_revision.ensure_null(revision_id)
1547
# TODO: refactor this to use an existing revision object
1548
# so we don't need to read it in twice.
1549
if revision_id == _mod_revision.NULL_REVISION:
1550
return inventorytree.InventoryRevisionTree(self,
1551
Inventory(root_id=None), _mod_revision.NULL_REVISION)
1553
inv = self.get_inventory(revision_id)
1554
return inventorytree.InventoryRevisionTree(self, inv, revision_id)
1556
def revision_trees(self, revision_ids):
1557
"""Return Trees for revisions in this repository.
1559
:param revision_ids: a sequence of revision-ids;
1560
a revision-id may not be None or 'null:'
1562
inventories = self.iter_inventories(revision_ids)
1563
for inv in inventories:
1564
yield inventorytree.InventoryRevisionTree(self, inv, inv.revision_id)
1566
def _filtered_revision_trees(self, revision_ids, file_ids):
1567
"""Return Tree for a revision on this branch with only some files.
1569
:param revision_ids: a sequence of revision-ids;
1570
a revision-id may not be None or 'null:'
1571
:param file_ids: if not None, the result is filtered
1572
so that only those file-ids, their parents and their
1573
children are included.
1575
inventories = self.iter_inventories(revision_ids)
1576
for inv in inventories:
1577
# Should we introduce a FilteredRevisionTree class rather
1578
# than pre-filter the inventory here?
1579
filtered_inv = inv.filter(file_ids)
1580
yield inventorytree.InventoryRevisionTree(self, filtered_inv, filtered_inv.revision_id)
1582
def get_parent_map(self, revision_ids):
1583
"""See graph.StackedParentsProvider.get_parent_map"""
1584
# revisions index works in keys; this just works in revisions
1585
# therefore wrap and unwrap
1588
for revision_id in revision_ids:
1589
if revision_id == _mod_revision.NULL_REVISION:
1590
result[revision_id] = ()
1591
elif revision_id is None:
1592
raise ValueError('get_parent_map(None) is not valid')
1594
query_keys.append((revision_id ,))
1595
for (revision_id,), parent_keys in viewitems(
1596
self.revisions.get_parent_map(query_keys)):
1598
result[revision_id] = tuple([parent_revid
1599
for (parent_revid,) in parent_keys])
1601
result[revision_id] = (_mod_revision.NULL_REVISION,)
1605
def get_known_graph_ancestry(self, revision_ids):
1606
"""Return the known graph for a set of revision ids and their ancestors.
1608
st = static_tuple.StaticTuple
1609
revision_keys = [st(r_id).intern() for r_id in revision_ids]
1610
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
1611
return graph.GraphThunkIdsToKeys(known_graph)
1614
def get_file_graph(self):
1615
"""Return the graph walker for text revisions."""
1616
return graph.Graph(self.texts)
1618
def revision_ids_to_search_result(self, result_set):
1619
"""Convert a set of revision ids to a graph SearchResult."""
1620
result_parents = set(itertools.chain.from_iterable(viewvalues(
1621
self.get_graph().get_parent_map(result_set))))
1622
included_keys = result_set.intersection(result_parents)
1623
start_keys = result_set.difference(included_keys)
1624
exclude_keys = result_parents.difference(result_set)
1625
result = vf_search.SearchResult(start_keys, exclude_keys,
1626
len(result_set), result_set)
1629
def _get_versioned_file_checker(self, text_key_references=None,
1631
"""Return an object suitable for checking versioned files.
1633
:param text_key_references: if non-None, an already built
1634
dictionary mapping text keys ((fileid, revision_id) tuples)
1635
to whether they were referred to by the inventory of the
1636
revision_id that they contain. If None, this will be
1638
:param ancestors: Optional result from
1639
self.get_graph().get_parent_map(self.all_revision_ids()) if already
1642
return _VersionedFileChecker(self,
1643
text_key_references=text_key_references, ancestors=ancestors)
1646
def has_signature_for_revision_id(self, revision_id):
1647
"""Query for a revision signature for revision_id in the repository."""
1648
if not self.has_revision(revision_id):
1649
raise errors.NoSuchRevision(self, revision_id)
1650
sig_present = (1 == len(
1651
self.signatures.get_parent_map([(revision_id,)])))
1655
def get_signature_text(self, revision_id):
1656
"""Return the text for a signature."""
1657
stream = self.signatures.get_record_stream([(revision_id,)],
1659
record = next(stream)
1660
if record.storage_kind == 'absent':
1661
raise errors.NoSuchRevision(self, revision_id)
1662
return record.get_bytes_as('fulltext')
1665
def _check(self, revision_ids, callback_refs, check_repo):
1666
result = check.VersionedFileCheck(self, check_repo=check_repo)
1667
result.check(callback_refs)
1670
def _find_inconsistent_revision_parents(self, revisions_iterator=None):
1671
"""Find revisions with different parent lists in the revision object
1672
and in the index graph.
1674
:param revisions_iterator: None, or an iterator of (revid,
1675
Revision-or-None). This iterator controls the revisions checked.
1676
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1677
parents-in-revision).
1679
if not self.is_locked():
1680
raise AssertionError()
1682
if revisions_iterator is None:
1683
revisions_iterator = self._iter_revisions(None)
1684
for revid, revision in revisions_iterator:
1685
if revision is None:
1687
parent_map = vf.get_parent_map([(revid,)])
1688
parents_according_to_index = tuple(parent[-1] for parent in
1689
parent_map[(revid,)])
1690
parents_according_to_revision = tuple(revision.parent_ids)
1691
if parents_according_to_index != parents_according_to_revision:
1692
yield (revid, parents_according_to_index,
1693
parents_according_to_revision)
1695
def _check_for_inconsistent_revision_parents(self):
1696
inconsistencies = list(self._find_inconsistent_revision_parents())
1698
raise errors.BzrCheckError(
1699
"Revision knit has inconsistent parents.")
1701
def _get_sink(self):
1702
"""Return a sink for streaming into this repository."""
1703
return StreamSink(self)
1705
def _get_source(self, to_format):
1706
"""Return a source for streaming from this repository."""
1707
return StreamSource(self, to_format)
1710
class MetaDirVersionedFileRepository(MetaDirRepository,
1711
VersionedFileRepository):
1712
"""Repositories in a meta-dir, that work via versioned file objects."""
1714
def __init__(self, _format, a_controldir, control_files):
1715
super(MetaDirVersionedFileRepository, self).__init__(_format, a_controldir,
1719
class MetaDirVersionedFileRepositoryFormat(RepositoryFormatMetaDir,
1720
VersionedFileRepositoryFormat):
1721
"""Base class for repository formats using versioned files in metadirs."""
1724
class StreamSink(object):
1725
"""An object that can insert a stream into a repository.
1727
This interface handles the complexity of reserialising inventories and
1728
revisions from different formats, and allows unidirectional insertion into
1729
stacked repositories without looking for the missing basis parents
1733
def __init__(self, target_repo):
1734
self.target_repo = target_repo
1736
def insert_stream(self, stream, src_format, resume_tokens):
1737
"""Insert a stream's content into the target repository.
1739
:param src_format: a bzr repository format.
1741
:return: a list of resume tokens and an iterable of keys additional
1742
items required before the insertion can be completed.
1744
self.target_repo.lock_write()
1747
self.target_repo.resume_write_group(resume_tokens)
1750
self.target_repo.start_write_group()
1753
# locked_insert_stream performs a commit|suspend.
1754
missing_keys = self.insert_stream_without_locking(stream,
1755
src_format, is_resume)
1757
# suspend the write group and tell the caller what we is
1758
# missing. We know we can suspend or else we would not have
1759
# entered this code path. (All repositories that can handle
1760
# missing keys can handle suspending a write group).
1761
write_group_tokens = self.target_repo.suspend_write_group()
1762
return write_group_tokens, missing_keys
1763
hint = self.target_repo.commit_write_group()
1764
to_serializer = self.target_repo._format._serializer
1765
src_serializer = src_format._serializer
1766
if (to_serializer != src_serializer and
1767
self.target_repo._format.pack_compresses):
1768
self.target_repo.pack(hint=hint)
1771
self.target_repo.abort_write_group(suppress_errors=True)
1774
self.target_repo.unlock()
1776
def insert_stream_without_locking(self, stream, src_format,
1778
"""Insert a stream's content into the target repository.
1780
This assumes that you already have a locked repository and an active
1783
:param src_format: a bzr repository format.
1784
:param is_resume: Passed down to get_missing_parent_inventories to
1785
indicate if we should be checking for missing texts at the same
1788
:return: A set of keys that are missing.
1790
if not self.target_repo.is_write_locked():
1791
raise errors.ObjectNotLocked(self)
1792
if not self.target_repo.is_in_write_group():
1793
raise errors.BzrError('you must already be in a write group')
1794
to_serializer = self.target_repo._format._serializer
1795
src_serializer = src_format._serializer
1797
if to_serializer == src_serializer:
1798
# If serializers match and the target is a pack repository, set the
1799
# write cache size on the new pack. This avoids poor performance
1800
# on transports where append is unbuffered (such as
1801
# RemoteTransport). This is safe to do because nothing should read
1802
# back from the target repository while a stream with matching
1803
# serialization is being inserted.
1804
# The exception is that a delta record from the source that should
1805
# be a fulltext may need to be expanded by the target (see
1806
# test_fetch_revisions_with_deltas_into_pack); but we take care to
1807
# explicitly flush any buffered writes first in that rare case.
1809
new_pack = self.target_repo._pack_collection._new_pack
1810
except AttributeError:
1811
# Not a pack repository
1814
new_pack.set_write_cache_size(1024*1024)
1815
for substream_type, substream in stream:
1816
if 'stream' in debug.debug_flags:
1817
mutter('inserting substream: %s', substream_type)
1818
if substream_type == 'texts':
1819
self.target_repo.texts.insert_record_stream(substream)
1820
elif substream_type == 'inventories':
1821
if src_serializer == to_serializer:
1822
self.target_repo.inventories.insert_record_stream(
1825
self._extract_and_insert_inventories(
1826
substream, src_serializer)
1827
elif substream_type == 'inventory-deltas':
1828
self._extract_and_insert_inventory_deltas(
1829
substream, src_serializer)
1830
elif substream_type == 'chk_bytes':
1831
# XXX: This doesn't support conversions, as it assumes the
1832
# conversion was done in the fetch code.
1833
self.target_repo.chk_bytes.insert_record_stream(substream)
1834
elif substream_type == 'revisions':
1835
# This may fallback to extract-and-insert more often than
1836
# required if the serializers are different only in terms of
1838
if src_serializer == to_serializer:
1839
self.target_repo.revisions.insert_record_stream(substream)
1841
self._extract_and_insert_revisions(substream,
1843
elif substream_type == 'signatures':
1844
self.target_repo.signatures.insert_record_stream(substream)
1846
raise AssertionError('kaboom! %s' % (substream_type,))
1847
# Done inserting data, and the missing_keys calculations will try to
1848
# read back from the inserted data, so flush the writes to the new pack
1849
# (if this is pack format).
1850
if new_pack is not None:
1851
new_pack._write_data('', flush=True)
1852
# Find all the new revisions (including ones from resume_tokens)
1853
missing_keys = self.target_repo.get_missing_parent_inventories(
1854
check_for_missing_texts=is_resume)
1856
for prefix, versioned_file in (
1857
('texts', self.target_repo.texts),
1858
('inventories', self.target_repo.inventories),
1859
('revisions', self.target_repo.revisions),
1860
('signatures', self.target_repo.signatures),
1861
('chk_bytes', self.target_repo.chk_bytes),
1863
if versioned_file is None:
1865
# TODO: key is often going to be a StaticTuple object
1866
# I don't believe we can define a method by which
1867
# (prefix,) + StaticTuple will work, though we could
1868
# define a StaticTuple.sq_concat that would allow you to
1869
# pass in either a tuple or a StaticTuple as the second
1870
# object, so instead we could have:
1871
# StaticTuple(prefix) + key here...
1872
missing_keys.update((prefix,) + key for key in
1873
versioned_file.get_missing_compression_parent_keys())
1874
except NotImplementedError:
1875
# cannot even attempt suspending, and missing would have failed
1876
# during stream insertion.
1877
missing_keys = set()
1880
def _extract_and_insert_inventory_deltas(self, substream, serializer):
1881
target_rich_root = self.target_repo._format.rich_root_data
1882
target_tree_refs = self.target_repo._format.supports_tree_reference
1883
for record in substream:
1884
# Insert the delta directly
1885
inventory_delta_bytes = record.get_bytes_as('fulltext')
1886
deserialiser = inventory_delta.InventoryDeltaDeserializer()
1888
parse_result = deserialiser.parse_text_bytes(
1889
inventory_delta_bytes)
1890
except inventory_delta.IncompatibleInventoryDelta as err:
1891
mutter("Incompatible delta: %s", err.msg)
1892
raise errors.IncompatibleRevision(self.target_repo._format)
1893
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
1894
revision_id = new_id
1895
parents = [key[0] for key in record.parents]
1896
self.target_repo.add_inventory_by_delta(
1897
basis_id, inv_delta, revision_id, parents)
1899
def _extract_and_insert_inventories(self, substream, serializer,
1901
"""Generate a new inventory versionedfile in target, converting data.
1903
The inventory is retrieved from the source, (deserializing it), and
1904
stored in the target (reserializing it in a different format).
1906
target_rich_root = self.target_repo._format.rich_root_data
1907
target_tree_refs = self.target_repo._format.supports_tree_reference
1908
for record in substream:
1909
# It's not a delta, so it must be a fulltext in the source
1910
# serializer's format.
1911
bytes = record.get_bytes_as('fulltext')
1912
revision_id = record.key[0]
1913
inv = serializer.read_inventory_from_string(bytes, revision_id)
1914
parents = [key[0] for key in record.parents]
1915
self.target_repo.add_inventory(revision_id, inv, parents)
1916
# No need to keep holding this full inv in memory when the rest of
1917
# the substream is likely to be all deltas.
1920
def _extract_and_insert_revisions(self, substream, serializer):
1921
for record in substream:
1922
bytes = record.get_bytes_as('fulltext')
1923
revision_id = record.key[0]
1924
rev = serializer.read_revision_from_string(bytes)
1925
if rev.revision_id != revision_id:
1926
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
1927
self.target_repo.add_revision(revision_id, rev)
1930
if self.target_repo._format._fetch_reconcile:
1931
self.target_repo.reconcile()
1934
class StreamSource(object):
1935
"""A source of a stream for fetching between repositories."""
1937
def __init__(self, from_repository, to_format):
1938
"""Create a StreamSource streaming from from_repository."""
1939
self.from_repository = from_repository
1940
self.to_format = to_format
1941
self._record_counter = RecordCounter()
1943
def delta_on_metadata(self):
1944
"""Return True if delta's are permitted on metadata streams.
1946
That is on revisions and signatures.
1948
src_serializer = self.from_repository._format._serializer
1949
target_serializer = self.to_format._serializer
1950
return (self.to_format._fetch_uses_deltas and
1951
src_serializer == target_serializer)
1953
def _fetch_revision_texts(self, revs):
1954
# fetch signatures first and then the revision texts
1955
# may need to be a InterRevisionStore call here.
1956
from_sf = self.from_repository.signatures
1957
# A missing signature is just skipped.
1958
keys = [(rev_id,) for rev_id in revs]
1959
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
1961
self.to_format._fetch_order,
1962
not self.to_format._fetch_uses_deltas))
1963
# If a revision has a delta, this is actually expanded inside the
1964
# insert_record_stream code now, which is an alternate fix for
1966
from_rf = self.from_repository.revisions
1967
revisions = from_rf.get_record_stream(
1969
self.to_format._fetch_order,
1970
not self.delta_on_metadata())
1971
return [('signatures', signatures), ('revisions', revisions)]
1973
def _generate_root_texts(self, revs):
1974
"""This will be called by get_stream between fetching weave texts and
1975
fetching the inventory weave.
1977
if self._rich_root_upgrade():
1978
return _mod_fetch.Inter1and2Helper(
1979
self.from_repository).generate_root_texts(revs)
1983
def get_stream(self, search):
1985
revs = search.get_keys()
1986
graph = self.from_repository.get_graph()
1987
revs = tsort.topo_sort(graph.get_parent_map(revs))
1988
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
1990
for knit_kind, file_id, revisions in data_to_fetch:
1991
if knit_kind != phase:
1993
# Make a new progress bar for this phase
1994
if knit_kind == "file":
1995
# Accumulate file texts
1996
text_keys.extend([(file_id, revision) for revision in
1998
elif knit_kind == "inventory":
1999
# Now copy the file texts.
2000
from_texts = self.from_repository.texts
2001
yield ('texts', from_texts.get_record_stream(
2002
text_keys, self.to_format._fetch_order,
2003
not self.to_format._fetch_uses_deltas))
2004
# Cause an error if a text occurs after we have done the
2007
# Before we process the inventory we generate the root
2008
# texts (if necessary) so that the inventories references
2010
for _ in self._generate_root_texts(revs):
2012
# we fetch only the referenced inventories because we do not
2013
# know for unselected inventories whether all their required
2014
# texts are present in the other repository - it could be
2016
for info in self._get_inventory_stream(revs):
2018
elif knit_kind == "signatures":
2019
# Nothing to do here; this will be taken care of when
2020
# _fetch_revision_texts happens.
2022
elif knit_kind == "revisions":
2023
for record in self._fetch_revision_texts(revs):
2026
raise AssertionError("Unknown knit kind %r" % knit_kind)
2028
def get_stream_for_missing_keys(self, missing_keys):
2029
# missing keys can only occur when we are byte copying and not
2030
# translating (because translation means we don't send
2031
# unreconstructable deltas ever).
2033
keys['texts'] = set()
2034
keys['revisions'] = set()
2035
keys['inventories'] = set()
2036
keys['chk_bytes'] = set()
2037
keys['signatures'] = set()
2038
for key in missing_keys:
2039
keys[key[0]].add(key[1:])
2040
if len(keys['revisions']):
2041
# If we allowed copying revisions at this point, we could end up
2042
# copying a revision without copying its required texts: a
2043
# violation of the requirements for repository integrity.
2044
raise AssertionError(
2045
'cannot copy revisions to fill in missing deltas %s' % (
2046
keys['revisions'],))
2047
for substream_kind, keys in viewitems(keys):
2048
vf = getattr(self.from_repository, substream_kind)
2049
if vf is None and keys:
2050
raise AssertionError(
2051
"cannot fill in keys for a versioned file we don't"
2052
" have: %s needs %s" % (substream_kind, keys))
2054
# No need to stream something we don't have
2056
if substream_kind == 'inventories':
2057
# Some missing keys are genuinely ghosts, filter those out.
2058
present = self.from_repository.inventories.get_parent_map(keys)
2059
revs = [key[0] for key in present]
2060
# Get the inventory stream more-or-less as we do for the
2061
# original stream; there's no reason to assume that records
2062
# direct from the source will be suitable for the sink. (Think
2063
# e.g. 2a -> 1.9-rich-root).
2064
for info in self._get_inventory_stream(revs, missing=True):
2068
# Ask for full texts always so that we don't need more round trips
2069
# after this stream.
2070
# Some of the missing keys are genuinely ghosts, so filter absent
2071
# records. The Sink is responsible for doing another check to
2072
# ensure that ghosts don't introduce missing data for future
2074
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
2075
self.to_format._fetch_order, True))
2076
yield substream_kind, stream
2078
def inventory_fetch_order(self):
2079
if self._rich_root_upgrade():
2080
return 'topological'
2082
return self.to_format._fetch_order
2084
def _rich_root_upgrade(self):
2085
return (not self.from_repository._format.rich_root_data and
2086
self.to_format.rich_root_data)
2088
def _get_inventory_stream(self, revision_ids, missing=False):
2089
from_format = self.from_repository._format
2090
if (from_format.supports_chks and self.to_format.supports_chks and
2091
from_format.network_name() == self.to_format.network_name()):
2092
raise AssertionError(
2093
"this case should be handled by GroupCHKStreamSource")
2094
elif 'forceinvdeltas' in debug.debug_flags:
2095
return self._get_convertable_inventory_stream(revision_ids,
2096
delta_versus_null=missing)
2097
elif from_format.network_name() == self.to_format.network_name():
2099
return self._get_simple_inventory_stream(revision_ids,
2101
elif (not from_format.supports_chks and not self.to_format.supports_chks
2102
and from_format._serializer == self.to_format._serializer):
2103
# Essentially the same format.
2104
return self._get_simple_inventory_stream(revision_ids,
2107
# Any time we switch serializations, we want to use an
2108
# inventory-delta based approach.
2109
return self._get_convertable_inventory_stream(revision_ids,
2110
delta_versus_null=missing)
2112
def _get_simple_inventory_stream(self, revision_ids, missing=False):
2113
# NB: This currently reopens the inventory weave in source;
2114
# using a single stream interface instead would avoid this.
2115
from_weave = self.from_repository.inventories
2117
delta_closure = True
2119
delta_closure = not self.delta_on_metadata()
2120
yield ('inventories', from_weave.get_record_stream(
2121
[(rev_id,) for rev_id in revision_ids],
2122
self.inventory_fetch_order(), delta_closure))
2124
def _get_convertable_inventory_stream(self, revision_ids,
2125
delta_versus_null=False):
2126
# The two formats are sufficiently different that there is no fast
2127
# path, so we need to send just inventorydeltas, which any
2128
# sufficiently modern client can insert into any repository.
2129
# The StreamSink code expects to be able to
2130
# convert on the target, so we need to put bytes-on-the-wire that can
2131
# be converted. That means inventory deltas (if the remote is <1.19,
2132
# RemoteStreamSink will fallback to VFS to insert the deltas).
2133
yield ('inventory-deltas',
2134
self._stream_invs_as_deltas(revision_ids,
2135
delta_versus_null=delta_versus_null))
2137
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
2138
"""Return a stream of inventory-deltas for the given rev ids.
2140
:param revision_ids: The list of inventories to transmit
2141
:param delta_versus_null: Don't try to find a minimal delta for this
2142
entry, instead compute the delta versus the NULL_REVISION. This
2143
effectively streams a complete inventory. Used for stuff like
2144
filling in missing parents, etc.
2146
from_repo = self.from_repository
2147
revision_keys = [(rev_id,) for rev_id in revision_ids]
2148
parent_map = from_repo.inventories.get_parent_map(revision_keys)
2149
# XXX: possibly repos could implement a more efficient iter_inv_deltas
2151
inventories = self.from_repository.iter_inventories(
2152
revision_ids, 'topological')
2153
format = from_repo._format
2154
invs_sent_so_far = {_mod_revision.NULL_REVISION}
2155
inventory_cache = lru_cache.LRUCache(50)
2156
null_inventory = from_repo.revision_tree(
2157
_mod_revision.NULL_REVISION).root_inventory
2158
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
2159
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
2160
# repo back into a non-rich-root repo ought to be allowed)
2161
serializer = inventory_delta.InventoryDeltaSerializer(
2162
versioned_root=format.rich_root_data,
2163
tree_references=format.supports_tree_reference)
2164
for inv in inventories:
2165
key = (inv.revision_id,)
2166
parent_keys = parent_map.get(key, ())
2168
if not delta_versus_null and parent_keys:
2169
# The caller did not ask for complete inventories and we have
2170
# some parents that we can delta against. Make a delta against
2171
# each parent so that we can find the smallest.
2172
parent_ids = [parent_key[0] for parent_key in parent_keys]
2173
for parent_id in parent_ids:
2174
if parent_id not in invs_sent_so_far:
2175
# We don't know that the remote side has this basis, so
2178
if parent_id == _mod_revision.NULL_REVISION:
2179
parent_inv = null_inventory
2181
parent_inv = inventory_cache.get(parent_id, None)
2182
if parent_inv is None:
2183
parent_inv = from_repo.get_inventory(parent_id)
2184
candidate_delta = inv._make_delta(parent_inv)
2185
if (delta is None or
2186
len(delta) > len(candidate_delta)):
2187
delta = candidate_delta
2188
basis_id = parent_id
2190
# Either none of the parents ended up being suitable, or we
2191
# were asked to delta against NULL
2192
basis_id = _mod_revision.NULL_REVISION
2193
delta = inv._make_delta(null_inventory)
2194
invs_sent_so_far.add(inv.revision_id)
2195
inventory_cache[inv.revision_id] = inv
2196
delta_serialized = ''.join(
2197
serializer.delta_to_lines(basis_id, key[-1], delta))
2198
yield versionedfile.FulltextContentFactory(
2199
key, parent_keys, None, delta_serialized)
2202
class _VersionedFileChecker(object):
2204
def __init__(self, repository, text_key_references=None, ancestors=None):
2205
self.repository = repository
2206
self.text_index = self.repository._generate_text_key_index(
2207
text_key_references=text_key_references, ancestors=ancestors)
2209
def calculate_file_version_parents(self, text_key):
2210
"""Calculate the correct parents for a file version according to
2213
parent_keys = self.text_index[text_key]
2214
if parent_keys == [_mod_revision.NULL_REVISION]:
2216
return tuple(parent_keys)
2218
def check_file_version_parents(self, texts, progress_bar=None):
2219
"""Check the parents stored in a versioned file are correct.
2221
It also detects file versions that are not referenced by their
2222
corresponding revision's inventory.
2224
:returns: A tuple of (wrong_parents, dangling_file_versions).
2225
wrong_parents is a dict mapping {revision_id: (stored_parents,
2226
correct_parents)} for each revision_id where the stored parents
2227
are not correct. dangling_file_versions is a set of (file_id,
2228
revision_id) tuples for versions that are present in this versioned
2229
file, but not used by the corresponding inventory.
2231
local_progress = None
2232
if progress_bar is None:
2233
local_progress = ui.ui_factory.nested_progress_bar()
2234
progress_bar = local_progress
2236
return self._check_file_version_parents(texts, progress_bar)
2239
local_progress.finished()
2241
def _check_file_version_parents(self, texts, progress_bar):
2242
"""See check_file_version_parents."""
2244
self.file_ids = {file_id for file_id, _ in self.text_index}
2245
# text keys is now grouped by file_id
2246
n_versions = len(self.text_index)
2247
progress_bar.update(gettext('loading text store'), 0, n_versions)
2248
parent_map = self.repository.texts.get_parent_map(self.text_index)
2249
# On unlistable transports this could well be empty/error...
2250
text_keys = self.repository.texts.keys()
2251
unused_keys = frozenset(text_keys) - set(self.text_index)
2252
for num, key in enumerate(self.text_index):
2253
progress_bar.update(gettext('checking text graph'), num, n_versions)
2254
correct_parents = self.calculate_file_version_parents(key)
2256
knit_parents = parent_map[key]
2257
except errors.RevisionNotPresent:
2260
if correct_parents != knit_parents:
2261
wrong_parents[key] = (knit_parents, correct_parents)
2262
return wrong_parents, unused_keys
2265
class InterVersionedFileRepository(InterRepository):
2267
_walk_to_common_revisions_batch_size = 50
2269
supports_fetch_spec = True
2272
def fetch(self, revision_id=None, find_ghosts=False,
2274
"""Fetch the content required to construct revision_id.
2276
The content is copied from self.source to self.target.
2278
:param revision_id: if None all content is copied, if NULL_REVISION no
2282
if self.target._format.experimental:
2283
ui.ui_factory.show_user_warning('experimental_format_fetch',
2284
from_format=self.source._format,
2285
to_format=self.target._format)
2286
from breezy.fetch import RepoFetcher
2287
# See <https://launchpad.net/bugs/456077> asking for a warning here
2288
if self.source._format.network_name() != self.target._format.network_name():
2289
ui.ui_factory.show_user_warning('cross_format_fetch',
2290
from_format=self.source._format,
2291
to_format=self.target._format)
2292
f = RepoFetcher(to_repository=self.target,
2293
from_repository=self.source,
2294
last_revision=revision_id,
2295
fetch_spec=fetch_spec,
2296
find_ghosts=find_ghosts)
2298
def _walk_to_common_revisions(self, revision_ids, if_present_ids=None):
2299
"""Walk out from revision_ids in source to revisions target has.
2301
:param revision_ids: The start point for the search.
2302
:return: A set of revision ids.
2304
target_graph = self.target.get_graph()
2305
revision_ids = frozenset(revision_ids)
2307
all_wanted_revs = revision_ids.union(if_present_ids)
2309
all_wanted_revs = revision_ids
2310
missing_revs = set()
2311
source_graph = self.source.get_graph()
2312
# ensure we don't pay silly lookup costs.
2313
searcher = source_graph._make_breadth_first_searcher(all_wanted_revs)
2314
null_set = frozenset([_mod_revision.NULL_REVISION])
2315
searcher_exhausted = False
2319
# Iterate the searcher until we have enough next_revs
2320
while len(next_revs) < self._walk_to_common_revisions_batch_size:
2322
next_revs_part, ghosts_part = searcher.next_with_ghosts()
2323
next_revs.update(next_revs_part)
2324
ghosts.update(ghosts_part)
2325
except StopIteration:
2326
searcher_exhausted = True
2328
# If there are ghosts in the source graph, and the caller asked for
2329
# them, make sure that they are present in the target.
2330
# We don't care about other ghosts as we can't fetch them and
2331
# haven't been asked to.
2332
ghosts_to_check = set(revision_ids.intersection(ghosts))
2333
revs_to_get = set(next_revs).union(ghosts_to_check)
2335
have_revs = set(target_graph.get_parent_map(revs_to_get))
2336
# we always have NULL_REVISION present.
2337
have_revs = have_revs.union(null_set)
2338
# Check if the target is missing any ghosts we need.
2339
ghosts_to_check.difference_update(have_revs)
2341
# One of the caller's revision_ids is a ghost in both the
2342
# source and the target.
2343
raise errors.NoSuchRevision(
2344
self.source, ghosts_to_check.pop())
2345
missing_revs.update(next_revs - have_revs)
2346
# Because we may have walked past the original stop point, make
2347
# sure everything is stopped
2348
stop_revs = searcher.find_seen_ancestors(have_revs)
2349
searcher.stop_searching_any(stop_revs)
2350
if searcher_exhausted:
2352
(started_keys, excludes, included_keys) = searcher.get_state()
2353
return vf_search.SearchResult(started_keys, excludes,
2354
len(included_keys), included_keys)
2357
def search_missing_revision_ids(self,
2358
find_ghosts=True, revision_ids=None, if_present_ids=None,
2360
"""Return the revision ids that source has that target does not.
2362
:param revision_ids: return revision ids included by these
2363
revision_ids. NoSuchRevision will be raised if any of these
2364
revisions are not present.
2365
:param if_present_ids: like revision_ids, but will not cause
2366
NoSuchRevision if any of these are absent, instead they will simply
2367
not be in the result. This is useful for e.g. finding revisions
2368
to fetch for tags, which may reference absent revisions.
2369
:param find_ghosts: If True find missing revisions in deep history
2370
rather than just finding the surface difference.
2371
:return: A breezy.graph.SearchResult.
2373
# stop searching at found target revisions.
2374
if not find_ghosts and (revision_ids is not None or if_present_ids is
2376
result = self._walk_to_common_revisions(revision_ids,
2377
if_present_ids=if_present_ids)
2380
result_set = result.get_keys()
2382
# generic, possibly worst case, slow code path.
2383
target_ids = set(self.target.all_revision_ids())
2384
source_ids = self._present_source_revisions_for(
2385
revision_ids, if_present_ids)
2386
result_set = set(source_ids).difference(target_ids)
2387
if limit is not None:
2388
topo_ordered = self.source.get_graph().iter_topo_order(result_set)
2389
result_set = set(itertools.islice(topo_ordered, limit))
2390
return self.source.revision_ids_to_search_result(result_set)
2392
def _present_source_revisions_for(self, revision_ids, if_present_ids=None):
2393
"""Returns set of all revisions in ancestry of revision_ids present in
2396
:param revision_ids: if None, all revisions in source are returned.
2397
:param if_present_ids: like revision_ids, but if any/all of these are
2398
absent no error is raised.
2400
if revision_ids is not None or if_present_ids is not None:
2401
# First, ensure all specified revisions exist. Callers expect
2402
# NoSuchRevision when they pass absent revision_ids here.
2403
if revision_ids is None:
2404
revision_ids = set()
2405
if if_present_ids is None:
2406
if_present_ids = set()
2407
revision_ids = set(revision_ids)
2408
if_present_ids = set(if_present_ids)
2409
all_wanted_ids = revision_ids.union(if_present_ids)
2410
graph = self.source.get_graph()
2411
present_revs = set(graph.get_parent_map(all_wanted_ids))
2412
missing = revision_ids.difference(present_revs)
2414
raise errors.NoSuchRevision(self.source, missing.pop())
2415
found_ids = all_wanted_ids.intersection(present_revs)
2416
source_ids = [rev_id for (rev_id, parents) in
2417
graph.iter_ancestry(found_ids)
2418
if rev_id != _mod_revision.NULL_REVISION
2419
and parents is not None]
2421
source_ids = self.source.all_revision_ids()
2422
return set(source_ids)
2425
def _get_repo_format_to_test(self):
2429
def is_compatible(cls, source, target):
2430
# The default implementation is compatible with everything
2431
return (source._format.supports_full_versioned_files and
2432
target._format.supports_full_versioned_files)
2435
class InterDifferingSerializer(InterVersionedFileRepository):
2438
def _get_repo_format_to_test(self):
2442
def is_compatible(source, target):
2443
if not source._format.supports_full_versioned_files:
2445
if not target._format.supports_full_versioned_files:
2447
# This is redundant with format.check_conversion_target(), however that
2448
# raises an exception, and we just want to say "False" as in we won't
2449
# support converting between these formats.
2450
if 'IDS_never' in debug.debug_flags:
2452
if source.supports_rich_root() and not target.supports_rich_root():
2454
if (source._format.supports_tree_reference
2455
and not target._format.supports_tree_reference):
2457
if target._fallback_repositories and target._format.supports_chks:
2458
# IDS doesn't know how to copy CHKs for the parent inventories it
2459
# adds to stacked repos.
2461
if 'IDS_always' in debug.debug_flags:
2463
# Only use this code path for local source and target. IDS does far
2464
# too much IO (both bandwidth and roundtrips) over a network.
2465
if not source.controldir.transport.base.startswith('file:///'):
2467
if not target.controldir.transport.base.startswith('file:///'):
2471
def _get_trees(self, revision_ids, cache):
2473
for rev_id in revision_ids:
2475
possible_trees.append((rev_id, cache[rev_id]))
2477
# Not cached, but inventory might be present anyway.
2479
tree = self.source.revision_tree(rev_id)
2480
except errors.NoSuchRevision:
2481
# Nope, parent is ghost.
2484
cache[rev_id] = tree
2485
possible_trees.append((rev_id, tree))
2486
return possible_trees
2488
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
2489
"""Get the best delta and base for this revision.
2491
:return: (basis_id, delta)
2494
# Generate deltas against each tree, to find the shortest.
2495
# FIXME: Support nested trees
2496
texts_possibly_new_in_tree = set()
2497
for basis_id, basis_tree in possible_trees:
2498
delta = tree.root_inventory._make_delta(basis_tree.root_inventory)
2499
for old_path, new_path, file_id, new_entry in delta:
2500
if new_path is None:
2501
# This file_id isn't present in the new rev, so we don't
2505
# Rich roots are handled elsewhere...
2507
kind = new_entry.kind
2508
if kind != 'directory' and kind != 'file':
2509
# No text record associated with this inventory entry.
2511
# This is a directory or file that has changed somehow.
2512
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
2513
deltas.append((len(delta), basis_id, delta))
2515
return deltas[0][1:]
2517
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
2518
"""Find all parent revisions that are absent, but for which the
2519
inventory is present, and copy those inventories.
2521
This is necessary to preserve correctness when the source is stacked
2522
without fallbacks configured. (Note that in cases like upgrade the
2523
source may be not have _fallback_repositories even though it is
2526
parent_revs = set(itertools.chain.from_iterable(viewvalues(
2528
present_parents = self.source.get_parent_map(parent_revs)
2529
absent_parents = parent_revs.difference(present_parents)
2530
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
2531
(rev_id,) for rev_id in absent_parents)
2532
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
2533
for parent_tree in self.source.revision_trees(parent_inv_ids):
2534
current_revision_id = parent_tree.get_revision_id()
2535
parents_parents_keys = parent_invs_keys_for_stacking[
2536
(current_revision_id,)]
2537
parents_parents = [key[-1] for key in parents_parents_keys]
2538
basis_id = _mod_revision.NULL_REVISION
2539
basis_tree = self.source.revision_tree(basis_id)
2540
delta = parent_tree.root_inventory._make_delta(
2541
basis_tree.root_inventory)
2542
self.target.add_inventory_by_delta(
2543
basis_id, delta, current_revision_id, parents_parents)
2544
cache[current_revision_id] = parent_tree
2546
def _fetch_batch(self, revision_ids, basis_id, cache):
2547
"""Fetch across a few revisions.
2549
:param revision_ids: The revisions to copy
2550
:param basis_id: The revision_id of a tree that must be in cache, used
2551
as a basis for delta when no other base is available
2552
:param cache: A cache of RevisionTrees that we can use.
2553
:return: The revision_id of the last converted tree. The RevisionTree
2554
for it will be in cache
2556
# Walk though all revisions; get inventory deltas, copy referenced
2557
# texts that delta references, insert the delta, revision and
2559
root_keys_to_create = set()
2562
pending_revisions = []
2563
parent_map = self.source.get_parent_map(revision_ids)
2564
self._fetch_parent_invs_for_stacking(parent_map, cache)
2565
self.source._safe_to_return_from_cache = True
2566
for tree in self.source.revision_trees(revision_ids):
2567
# Find a inventory delta for this revision.
2568
# Find text entries that need to be copied, too.
2569
current_revision_id = tree.get_revision_id()
2570
parent_ids = parent_map.get(current_revision_id, ())
2571
parent_trees = self._get_trees(parent_ids, cache)
2572
possible_trees = list(parent_trees)
2573
if len(possible_trees) == 0:
2574
# There either aren't any parents, or the parents are ghosts,
2575
# so just use the last converted tree.
2576
possible_trees.append((basis_id, cache[basis_id]))
2577
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
2579
revision = self.source.get_revision(current_revision_id)
2580
pending_deltas.append((basis_id, delta,
2581
current_revision_id, revision.parent_ids))
2582
if self._converting_to_rich_root:
2583
self._revision_id_to_root_id[current_revision_id] = \
2585
# Determine which texts are in present in this revision but not in
2586
# any of the available parents.
2587
texts_possibly_new_in_tree = set()
2588
for old_path, new_path, file_id, entry in delta:
2589
if new_path is None:
2590
# This file_id isn't present in the new rev
2594
if not self.target.supports_rich_root():
2595
# The target doesn't support rich root, so we don't
2598
if self._converting_to_rich_root:
2599
# This can't be copied normally, we have to insert
2601
root_keys_to_create.add((file_id, entry.revision))
2604
texts_possibly_new_in_tree.add((file_id, entry.revision))
2605
for basis_id, basis_tree in possible_trees:
2606
basis_inv = basis_tree.root_inventory
2607
for file_key in list(texts_possibly_new_in_tree):
2608
file_id, file_revision = file_key
2610
entry = basis_inv[file_id]
2611
except errors.NoSuchId:
2613
if entry.revision == file_revision:
2614
texts_possibly_new_in_tree.remove(file_key)
2615
text_keys.update(texts_possibly_new_in_tree)
2616
pending_revisions.append(revision)
2617
cache[current_revision_id] = tree
2618
basis_id = current_revision_id
2619
self.source._safe_to_return_from_cache = False
2621
from_texts = self.source.texts
2622
to_texts = self.target.texts
2623
if root_keys_to_create:
2624
root_stream = _mod_fetch._new_root_data_stream(
2625
root_keys_to_create, self._revision_id_to_root_id, parent_map,
2627
to_texts.insert_record_stream(root_stream)
2628
to_texts.insert_record_stream(from_texts.get_record_stream(
2629
text_keys, self.target._format._fetch_order,
2630
not self.target._format._fetch_uses_deltas))
2631
# insert inventory deltas
2632
for delta in pending_deltas:
2633
self.target.add_inventory_by_delta(*delta)
2634
if self.target._fallback_repositories:
2635
# Make sure this stacked repository has all the parent inventories
2636
# for the new revisions that we are about to insert. We do this
2637
# before adding the revisions so that no revision is added until
2638
# all the inventories it may depend on are added.
2639
# Note that this is overzealous, as we may have fetched these in an
2642
revision_ids = set()
2643
for revision in pending_revisions:
2644
revision_ids.add(revision.revision_id)
2645
parent_ids.update(revision.parent_ids)
2646
parent_ids.difference_update(revision_ids)
2647
parent_ids.discard(_mod_revision.NULL_REVISION)
2648
parent_map = self.source.get_parent_map(parent_ids)
2649
# we iterate over parent_map and not parent_ids because we don't
2650
# want to try copying any revision which is a ghost
2651
for parent_tree in self.source.revision_trees(parent_map):
2652
current_revision_id = parent_tree.get_revision_id()
2653
parents_parents = parent_map[current_revision_id]
2654
possible_trees = self._get_trees(parents_parents, cache)
2655
if len(possible_trees) == 0:
2656
# There either aren't any parents, or the parents are
2657
# ghosts, so just use the last converted tree.
2658
possible_trees.append((basis_id, cache[basis_id]))
2659
basis_id, delta = self._get_delta_for_revision(parent_tree,
2660
parents_parents, possible_trees)
2661
self.target.add_inventory_by_delta(
2662
basis_id, delta, current_revision_id, parents_parents)
2663
# insert signatures and revisions
2664
for revision in pending_revisions:
2666
signature = self.source.get_signature_text(
2667
revision.revision_id)
2668
self.target.add_signature_text(revision.revision_id,
2670
except errors.NoSuchRevision:
2672
self.target.add_revision(revision.revision_id, revision)
2675
def _fetch_all_revisions(self, revision_ids, pb):
2676
"""Fetch everything for the list of revisions.
2678
:param revision_ids: The list of revisions to fetch. Must be in
2680
:param pb: A ProgressTask
2683
basis_id, basis_tree = self._get_basis(revision_ids[0])
2685
cache = lru_cache.LRUCache(100)
2686
cache[basis_id] = basis_tree
2687
del basis_tree # We don't want to hang on to it here
2691
for offset in range(0, len(revision_ids), batch_size):
2692
self.target.start_write_group()
2694
pb.update(gettext('Transferring revisions'), offset,
2696
batch = revision_ids[offset:offset+batch_size]
2697
basis_id = self._fetch_batch(batch, basis_id, cache)
2699
self.source._safe_to_return_from_cache = False
2700
self.target.abort_write_group()
2703
hint = self.target.commit_write_group()
2706
if hints and self.target._format.pack_compresses:
2707
self.target.pack(hint=hints)
2708
pb.update(gettext('Transferring revisions'), len(revision_ids),
2712
def fetch(self, revision_id=None, find_ghosts=False,
2714
"""See InterRepository.fetch()."""
2715
if fetch_spec is not None:
2716
revision_ids = fetch_spec.get_keys()
2719
if self.source._format.experimental:
2720
ui.ui_factory.show_user_warning('experimental_format_fetch',
2721
from_format=self.source._format,
2722
to_format=self.target._format)
2723
if (not self.source.supports_rich_root()
2724
and self.target.supports_rich_root()):
2725
self._converting_to_rich_root = True
2726
self._revision_id_to_root_id = {}
2728
self._converting_to_rich_root = False
2729
# See <https://launchpad.net/bugs/456077> asking for a warning here
2730
if self.source._format.network_name() != self.target._format.network_name():
2731
ui.ui_factory.show_user_warning('cross_format_fetch',
2732
from_format=self.source._format,
2733
to_format=self.target._format)
2734
if revision_ids is None:
2736
search_revision_ids = [revision_id]
2738
search_revision_ids = None
2739
revision_ids = self.target.search_missing_revision_ids(self.source,
2740
revision_ids=search_revision_ids,
2741
find_ghosts=find_ghosts).get_keys()
2742
if not revision_ids:
2744
revision_ids = tsort.topo_sort(
2745
self.source.get_graph().get_parent_map(revision_ids))
2746
if not revision_ids:
2748
# Walk though all revisions; get inventory deltas, copy referenced
2749
# texts that delta references, insert the delta, revision and
2751
pb = ui.ui_factory.nested_progress_bar()
2753
self._fetch_all_revisions(revision_ids, pb)
2756
return len(revision_ids), 0
2758
def _get_basis(self, first_revision_id):
2759
"""Get a revision and tree which exists in the target.
2761
This assumes that first_revision_id is selected for transmission
2762
because all other ancestors are already present. If we can't find an
2763
ancestor we fall back to NULL_REVISION since we know that is safe.
2765
:return: (basis_id, basis_tree)
2767
first_rev = self.source.get_revision(first_revision_id)
2769
basis_id = first_rev.parent_ids[0]
2770
# only valid as a basis if the target has it
2771
self.target.get_revision(basis_id)
2772
# Try to get a basis tree - if it's a ghost it will hit the
2773
# NoSuchRevision case.
2774
basis_tree = self.source.revision_tree(basis_id)
2775
except (IndexError, errors.NoSuchRevision):
2776
basis_id = _mod_revision.NULL_REVISION
2777
basis_tree = self.source.revision_tree(basis_id)
2778
return basis_id, basis_tree
2781
class InterSameDataRepository(InterVersionedFileRepository):
2782
"""Code for converting between repositories that represent the same data.
2784
Data format and model must match for this to work.
2788
def _get_repo_format_to_test(self):
2789
"""Repository format for testing with.
2791
InterSameData can pull from subtree to subtree and from non-subtree to
2792
non-subtree, so we test this with the richest repository format.
2794
from breezy.bzr import knitrepo
2795
return knitrepo.RepositoryFormatKnit3()
2798
def is_compatible(source, target):
2800
InterRepository._same_model(source, target) and
2801
source._format.supports_full_versioned_files and
2802
target._format.supports_full_versioned_files)
2805
InterRepository.register_optimiser(InterVersionedFileRepository)
2806
InterRepository.register_optimiser(InterDifferingSerializer)
2807
InterRepository.register_optimiser(InterSameDataRepository)
2810
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2811
"""Install all revision data into a repository.
2813
Accepts an iterable of revision, tree, signature tuples. The signature
2816
repository.start_write_group()
2818
inventory_cache = lru_cache.LRUCache(10)
2819
for n, (revision, revision_tree, signature) in enumerate(iterable):
2820
_install_revision(repository, revision, revision_tree, signature,
2823
pb.update(gettext('Transferring revisions'), n + 1, num_revisions)
2825
repository.abort_write_group()
2828
repository.commit_write_group()
2831
def _install_revision(repository, rev, revision_tree, signature,
2833
"""Install all revision data into a repository."""
2834
present_parents = []
2836
for p_id in rev.parent_ids:
2837
if repository.has_revision(p_id):
2838
present_parents.append(p_id)
2839
parent_trees[p_id] = repository.revision_tree(p_id)
2841
parent_trees[p_id] = repository.revision_tree(
2842
_mod_revision.NULL_REVISION)
2844
# FIXME: Support nested trees
2845
inv = revision_tree.root_inventory
2846
entries = inv.iter_entries()
2847
# backwards compatibility hack: skip the root id.
2848
if not repository.supports_rich_root():
2849
path, root = next(entries)
2850
if root.revision != rev.revision_id:
2851
raise errors.IncompatibleRevision(repr(repository))
2853
for path, ie in entries:
2854
text_keys[(ie.file_id, ie.revision)] = ie
2855
text_parent_map = repository.texts.get_parent_map(text_keys)
2856
missing_texts = set(text_keys) - set(text_parent_map)
2857
# Add the texts that are not already present
2858
for text_key in missing_texts:
2859
ie = text_keys[text_key]
2861
# FIXME: TODO: The following loop overlaps/duplicates that done by
2862
# commit to determine parents. There is a latent/real bug here where
2863
# the parents inserted are not those commit would do - in particular
2864
# they are not filtered by heads(). RBC, AB
2865
for revision, tree in viewitems(parent_trees):
2866
if not tree.has_id(ie.file_id):
2868
parent_id = tree.get_file_revision(ie.file_id)
2869
if parent_id in text_parents:
2871
text_parents.append((ie.file_id, parent_id))
2872
lines = revision_tree.get_file(ie.file_id).readlines()
2873
repository.texts.add_lines(text_key, text_parents, lines)
2875
# install the inventory
2876
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2877
# Cache this inventory
2878
inventory_cache[rev.revision_id] = inv
2880
basis_inv = inventory_cache[rev.parent_ids[0]]
2882
repository.add_inventory(rev.revision_id, inv, present_parents)
2884
delta = inv._make_delta(basis_inv)
2885
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2886
rev.revision_id, present_parents)
2888
repository.add_inventory(rev.revision_id, inv, present_parents)
2889
except errors.RevisionAlreadyPresent:
2891
if signature is not None:
2892
repository.add_signature_text(rev.revision_id, signature)
2893
repository.add_revision(rev.revision_id, rev, inv)
2896
def install_revision(repository, rev, revision_tree):
2897
"""Install all revision data into a repository."""
2898
install_revisions(repository, [(rev, revision_tree, None)])
added
removed
breezy/bzr/vf_repository.py
