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(), """
27
config as _mod_config,
34
revision as _mod_revision,
39
from breezy.bzr import (
48
from breezy.recordcounter import RecordCounter
49
from breezy.testament import Testament
50
from breezy.i18n import gettext
56
from ..decorators import (
59
from .inventory import (
65
from ..repository import (
71
from .repository import (
73
RepositoryFormatMetaDir,
76
from ..sixish import (
88
class VersionedFileRepositoryFormat(RepositoryFormat):
89
"""Base class for all repository formats that are VersionedFiles-based."""
91
supports_full_versioned_files = True
92
supports_versioned_directories = True
93
supports_unreferenced_revisions = True
95
# Should commit add an inventory, or an inventory delta to the repository.
96
_commit_inv_deltas = True
97
# What order should fetch operations request streams in?
98
# The default is unordered as that is the cheapest for an origin to
100
_fetch_order = 'unordered'
101
# Does this repository format use deltas that can be fetched as-deltas ?
102
# (E.g. knits, where the knit deltas can be transplanted intact.
103
# We default to False, which will ensure that enough data to get
104
# a full text out of any fetch stream will be grabbed.
105
_fetch_uses_deltas = False
108
class VersionedFileCommitBuilder(CommitBuilder):
109
"""Commit builder implementation for versioned files based repositories.
112
def __init__(self, repository, parents, config_stack, timestamp=None,
113
timezone=None, committer=None, revprops=None,
114
revision_id=None, lossy=False):
115
super(VersionedFileCommitBuilder, self).__init__(repository,
116
parents, config_stack, timestamp, timezone, committer, revprops,
119
basis_id = self.parents[0]
121
basis_id = _mod_revision.NULL_REVISION
122
self.basis_delta_revision = basis_id
123
self._new_inventory = None
124
self._basis_delta = []
125
self.__heads = graph.HeadsCache(repository.get_graph()).heads
126
# memo'd check for no-op commits.
127
self._any_changes = False
129
def any_changes(self):
130
"""Return True if any entries were changed.
132
This includes merge-only changes. It is the core for the --unchanged
135
:return: True if any changes have occured.
137
return self._any_changes
139
def _ensure_fallback_inventories(self):
140
"""Ensure that appropriate inventories are available.
142
This only applies to repositories that are stacked, and is about
143
enusring the stacking invariants. Namely, that for any revision that is
144
present, we either have all of the file content, or we have the parent
145
inventory and the delta file content.
147
if not self.repository._fallback_repositories:
149
if not self.repository._format.supports_chks:
150
raise errors.BzrError("Cannot commit directly to a stacked branch"
151
" in pre-2a formats. See "
152
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
153
# This is a stacked repo, we need to make sure we have the parent
154
# inventories for the parents.
155
parent_keys = [(p,) for p in self.parents]
156
parent_map = self.repository.inventories._index.get_parent_map(
158
missing_parent_keys = {pk for pk in parent_keys
159
if pk not in parent_map}
160
fallback_repos = list(reversed(self.repository._fallback_repositories))
161
missing_keys = [('inventories', pk[0])
162
for pk in missing_parent_keys]
164
while missing_keys and fallback_repos:
165
fallback_repo = fallback_repos.pop()
166
source = fallback_repo._get_source(self.repository._format)
167
sink = self.repository._get_sink()
168
missing_keys = sink.insert_missing_keys(source, missing_keys)
170
raise errors.BzrError('Unable to fill in parent inventories for a'
173
def commit(self, message):
174
"""Make the actual commit.
176
:return: The revision id of the recorded revision.
178
self._validate_unicode_text(message, 'commit message')
179
rev = _mod_revision.Revision(
180
timestamp=self._timestamp,
181
timezone=self._timezone,
182
committer=self._committer,
184
inventory_sha1=self.inv_sha1,
185
revision_id=self._new_revision_id,
186
properties=self._revprops)
187
rev.parent_ids = self.parents
188
if self._config_stack.get('create_signatures') == _mod_config.SIGN_ALWAYS:
189
testament = Testament(rev, self.revision_tree())
190
plaintext = testament.as_short_text()
191
self.repository.store_revision_signature(
192
gpg.GPGStrategy(self._config_stack), plaintext,
193
self._new_revision_id)
194
self.repository._add_revision(rev)
195
self._ensure_fallback_inventories()
196
self.repository.commit_write_group()
197
return self._new_revision_id
200
"""Abort the commit that is being built.
202
self.repository.abort_write_group()
204
def revision_tree(self):
205
"""Return the tree that was just committed.
207
After calling commit() this can be called to get a
208
RevisionTree representing the newly committed tree. This is
209
preferred to calling Repository.revision_tree() because that may
210
require deserializing the inventory, while we already have a copy in
213
if self._new_inventory is None:
214
self._new_inventory = self.repository.get_inventory(
215
self._new_revision_id)
216
return inventorytree.InventoryRevisionTree(self.repository,
217
self._new_inventory, self._new_revision_id)
219
def finish_inventory(self):
220
"""Tell the builder that the inventory is finished.
222
:return: The inventory id in the repository, which can be used with
223
repository.get_inventory.
225
# an inventory delta was accumulated without creating a new
227
basis_id = self.basis_delta_revision
228
self.inv_sha1, self._new_inventory = self.repository.add_inventory_by_delta(
229
basis_id, self._basis_delta, self._new_revision_id,
231
return self._new_revision_id
233
def _require_root_change(self, tree):
234
"""Enforce an appropriate root object change.
236
This is called once when record_iter_changes is called, if and only if
237
the root was not in the delta calculated by record_iter_changes.
239
:param tree: The tree which is being committed.
241
if self.repository.supports_rich_root():
243
if len(self.parents) == 0:
244
raise errors.RootMissing()
245
entry = entry_factory['directory'](tree.path2id(''), '',
247
entry.revision = self._new_revision_id
248
self._basis_delta.append(('', '', entry.file_id, entry))
250
def _get_delta(self, ie, basis_inv, path):
251
"""Get a delta against the basis inventory for ie."""
252
if not basis_inv.has_id(ie.file_id):
254
result = (None, path, ie.file_id, ie)
255
self._basis_delta.append(result)
257
elif ie != basis_inv.get_entry(ie.file_id):
259
# TODO: avoid tis id2path call.
260
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
261
self._basis_delta.append(result)
267
def _heads(self, file_id, revision_ids):
268
"""Calculate the graph heads for revision_ids in the graph of file_id.
270
This can use either a per-file graph or a global revision graph as we
271
have an identity relationship between the two graphs.
273
return self.__heads(revision_ids)
275
def get_basis_delta(self):
276
"""Return the complete inventory delta versus the basis inventory.
278
:return: An inventory delta, suitable for use with apply_delta, or
279
Repository.add_inventory_by_delta, etc.
281
return self._basis_delta
283
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
284
_entry_factory=entry_factory):
285
"""Record a new tree via iter_changes.
287
:param tree: The tree to obtain text contents from for changed objects.
288
:param basis_revision_id: The revision id of the tree the iter_changes
289
has been generated against. Currently assumed to be the same
290
as self.parents[0] - if it is not, errors may occur.
291
:param iter_changes: An iter_changes iterator with the changes to apply
292
to basis_revision_id. The iterator must not include any items with
293
a current kind of None - missing items must be either filtered out
294
or errored-on before record_iter_changes sees the item.
295
:param _entry_factory: Private method to bind entry_factory locally for
297
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
300
# Create an inventory delta based on deltas between all the parents and
301
# deltas between all the parent inventories. We use inventory delta's
302
# between the inventory objects because iter_changes masks
303
# last-changed-field only changes.
305
# file_id -> change map, change is fileid, paths, changed, versioneds,
306
# parents, names, kinds, executables
308
# {file_id -> revision_id -> inventory entry, for entries in parent
309
# trees that are not parents[0]
313
revtrees = list(self.repository.revision_trees(self.parents))
314
except errors.NoSuchRevision:
315
# one or more ghosts, slow path.
317
for revision_id in self.parents:
319
revtrees.append(self.repository.revision_tree(revision_id))
320
except errors.NoSuchRevision:
322
basis_revision_id = _mod_revision.NULL_REVISION
324
revtrees.append(self.repository.revision_tree(
325
_mod_revision.NULL_REVISION))
326
# The basis inventory from a repository
328
basis_tree = revtrees[0]
330
basis_tree = self.repository.revision_tree(
331
_mod_revision.NULL_REVISION)
332
basis_inv = basis_tree.root_inventory
333
if len(self.parents) > 0:
334
if basis_revision_id != self.parents[0] and not ghost_basis:
336
"arbitrary basis parents not yet supported with merges")
337
for revtree in revtrees[1:]:
338
for change in revtree.root_inventory._make_delta(basis_inv):
339
if change[1] is None:
340
# Not present in this parent.
342
if change[2] not in merged_ids:
343
if change[0] is not None:
344
basis_entry = basis_inv.get_entry(change[2])
345
merged_ids[change[2]] = [
347
basis_entry.revision,
350
parent_entries[change[2]] = {
352
basis_entry.revision: basis_entry,
354
change[3].revision: change[3],
357
merged_ids[change[2]] = [change[3].revision]
358
parent_entries[change[2]] = {
359
change[3].revision: change[3]}
361
merged_ids[change[2]].append(change[3].revision)
362
parent_entries[change[2]
363
][change[3].revision] = change[3]
366
# Setup the changes from the tree:
367
# changes maps file_id -> (change, [parent revision_ids])
369
for change in iter_changes:
370
# This probably looks up in basis_inv way to much.
371
if change[1][0] is not None:
372
head_candidate = [basis_inv.get_entry(change[0]).revision]
375
changes[change[0]] = change, merged_ids.get(change[0],
377
unchanged_merged = set(merged_ids) - set(changes)
378
# Extend the changes dict with synthetic changes to record merges of
380
for file_id in unchanged_merged:
381
# Record a merged version of these items that did not change vs the
382
# basis. This can be either identical parallel changes, or a revert
383
# of a specific file after a merge. The recorded content will be
384
# that of the current tree (which is the same as the basis), but
385
# the per-file graph will reflect a merge.
386
# NB:XXX: We are reconstructing path information we had, this
387
# should be preserved instead.
388
# inv delta change: (file_id, (path_in_source, path_in_target),
389
# changed_content, versioned, parent, name, kind,
392
basis_entry = basis_inv.get_entry(file_id)
393
except errors.NoSuchId:
394
# a change from basis->some_parents but file_id isn't in basis
395
# so was new in the merge, which means it must have changed
396
# from basis -> current, and as it hasn't the add was reverted
397
# by the user. So we discard this change.
401
(basis_inv.id2path(file_id), tree.id2path(file_id)),
403
(basis_entry.parent_id, basis_entry.parent_id),
404
(basis_entry.name, basis_entry.name),
405
(basis_entry.kind, basis_entry.kind),
406
(basis_entry.executable, basis_entry.executable))
407
changes[file_id] = (change, merged_ids[file_id])
408
# changes contains tuples with the change and a set of inventory
409
# candidates for the file.
411
# old_path, new_path, file_id, new_inventory_entry
412
seen_root = False # Is the root in the basis delta?
413
inv_delta = self._basis_delta
414
modified_rev = self._new_revision_id
415
for change, head_candidates in viewvalues(changes):
416
if change[3][1]: # versioned in target.
417
# Several things may be happening here:
418
# We may have a fork in the per-file graph
419
# - record a change with the content from tree
420
# We may have a change against < all trees
421
# - carry over the tree that hasn't changed
422
# We may have a change against all trees
423
# - record the change with the content from tree
426
entry = _entry_factory[kind](file_id, change[5][1],
428
head_set = self._heads(change[0], set(head_candidates))
431
for head_candidate in head_candidates:
432
if head_candidate in head_set:
433
heads.append(head_candidate)
434
head_set.remove(head_candidate)
437
# Could be a carry-over situation:
438
parent_entry_revs = parent_entries.get(file_id, None)
439
if parent_entry_revs:
440
parent_entry = parent_entry_revs.get(heads[0], None)
443
if parent_entry is None:
444
# The parent iter_changes was called against is the one
445
# that is the per-file head, so any change is relevant
446
# iter_changes is valid.
447
carry_over_possible = False
449
# could be a carry over situation
450
# A change against the basis may just indicate a merge,
451
# we need to check the content against the source of the
452
# merge to determine if it was changed after the merge
454
if (parent_entry.kind != entry.kind
455
or parent_entry.parent_id != entry.parent_id
456
or parent_entry.name != entry.name):
457
# Metadata common to all entries has changed
458
# against per-file parent
459
carry_over_possible = False
461
carry_over_possible = True
462
# per-type checks for changes against the parent_entry
465
# Cannot be a carry-over situation
466
carry_over_possible = False
467
# Populate the entry in the delta
469
# XXX: There is still a small race here: If someone reverts the content of a file
470
# after iter_changes examines and decides it has changed,
471
# we will unconditionally record a new version even if some
472
# other process reverts it while commit is running (with
473
# the revert happening after iter_changes did its
476
entry.executable = True
478
entry.executable = False
479
if (carry_over_possible
480
and parent_entry.executable == entry.executable):
481
# Check the file length, content hash after reading
483
nostore_sha = parent_entry.text_sha1
486
file_obj, stat_value = tree.get_file_with_stat(change[1][1])
488
entry.text_sha1, entry.text_size = self._add_file_to_weave(
489
file_id, file_obj, heads, nostore_sha)
490
yield file_id, change[1][1], (entry.text_sha1, stat_value)
491
except errors.ExistingContent:
492
# No content change against a carry_over parent
493
# Perhaps this should also yield a fs hash update?
495
entry.text_size = parent_entry.text_size
496
entry.text_sha1 = parent_entry.text_sha1
499
elif kind == 'symlink':
501
entry.symlink_target = tree.get_symlink_target(
503
if (carry_over_possible and
504
parent_entry.symlink_target ==
505
entry.symlink_target):
508
self._add_file_to_weave(
509
change[0], BytesIO(), heads, None)
510
elif kind == 'directory':
511
if carry_over_possible:
514
# Nothing to set on the entry.
515
# XXX: split into the Root and nonRoot versions.
516
if change[1][1] != '' or self.repository.supports_rich_root():
517
self._add_file_to_weave(
518
change[0], BytesIO(), heads, None)
519
elif kind == 'tree-reference':
520
if not self.repository._format.supports_tree_reference:
521
# This isn't quite sane as an error, but we shouldn't
522
# ever see this code path in practice: tree's don't
523
# permit references when the repo doesn't support tree
525
raise errors.UnsupportedOperation(
526
tree.add_reference, self.repository)
527
reference_revision = tree.get_reference_revision(
529
entry.reference_revision = reference_revision
530
if (carry_over_possible
531
and parent_entry.reference_revision ==
535
self._add_file_to_weave(
536
change[0], BytesIO(), heads, None)
538
raise AssertionError('unknown kind %r' % kind)
540
entry.revision = modified_rev
542
entry.revision = parent_entry.revision
545
new_path = change[1][1]
546
inv_delta.append((change[1][0], new_path, change[0], entry))
549
# The initial commit adds a root directory, but this in itself is not
550
# a worthwhile commit.
551
if ((len(inv_delta) > 0 and basis_revision_id != _mod_revision.NULL_REVISION)
552
or (len(inv_delta) > 1 and basis_revision_id == _mod_revision.NULL_REVISION)):
553
# This should perhaps be guarded by a check that the basis we
554
# commit against is the basis for the commit and if not do a delta
556
self._any_changes = True
558
# housekeeping root entry changes do not affect no-change commits.
559
self._require_root_change(tree)
560
self.basis_delta_revision = basis_revision_id
562
def _add_file_to_weave(self, file_id, fileobj, parents, nostore_sha):
563
parent_keys = tuple([(file_id, parent) for parent in parents])
564
return self.repository.texts.add_lines(
565
(file_id, self._new_revision_id), parent_keys, fileobj.readlines(),
566
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
569
class VersionedFileRepository(Repository):
570
"""Repository holding history for one or more branches.
572
The repository holds and retrieves historical information including
573
revisions and file history. It's normally accessed only by the Branch,
574
which views a particular line of development through that history.
576
The Repository builds on top of some byte storage facilies (the revisions,
577
signatures, inventories, texts and chk_bytes attributes) and a Transport,
578
which respectively provide byte storage and a means to access the (possibly
581
The byte storage facilities are addressed via tuples, which we refer to
582
as 'keys' throughout the code base. Revision_keys, inventory_keys and
583
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
584
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
585
byte string made up of a hash identifier and a hash value.
586
We use this interface because it allows low friction with the underlying
587
code that implements disk indices, network encoding and other parts of
590
:ivar revisions: A breezy.versionedfile.VersionedFiles instance containing
591
the serialised revisions for the repository. This can be used to obtain
592
revision graph information or to access raw serialised revisions.
593
The result of trying to insert data into the repository via this store
594
is undefined: it should be considered read-only except for implementors
596
:ivar signatures: A breezy.versionedfile.VersionedFiles instance containing
597
the serialised signatures for the repository. This can be used to
598
obtain access to raw serialised signatures. The result of trying to
599
insert data into the repository via this store is undefined: it should
600
be considered read-only except for implementors of repositories.
601
:ivar inventories: A breezy.versionedfile.VersionedFiles instance containing
602
the serialised inventories for the repository. This can be used to
603
obtain unserialised inventories. The result of trying to insert data
604
into the repository via this store is undefined: it should be
605
considered read-only except for implementors of repositories.
606
:ivar texts: A breezy.versionedfile.VersionedFiles instance containing the
607
texts of files and directories for the repository. This can be used to
608
obtain file texts or file graphs. Note that Repository.iter_file_bytes
609
is usually a better interface for accessing file texts.
610
The result of trying to insert data into the repository via this store
611
is undefined: it should be considered read-only except for implementors
613
:ivar chk_bytes: A breezy.versionedfile.VersionedFiles instance containing
614
any data the repository chooses to store or have indexed by its hash.
615
The result of trying to insert data into the repository via this store
616
is undefined: it should be considered read-only except for implementors
618
:ivar _transport: Transport for file access to repository, typically
619
pointing to .bzr/repository.
622
# What class to use for a CommitBuilder. Often it's simpler to change this
623
# in a Repository class subclass rather than to override
624
# get_commit_builder.
625
_commit_builder_class = VersionedFileCommitBuilder
627
def add_fallback_repository(self, repository):
628
"""Add a repository to use for looking up data not held locally.
630
:param repository: A repository.
632
if not self._format.supports_external_lookups:
633
raise errors.UnstackableRepositoryFormat(self._format, self.base)
634
# This can raise an exception, so should be done before we lock the
635
# fallback repository.
636
self._check_fallback_repository(repository)
638
# This repository will call fallback.unlock() when we transition to
639
# the unlocked state, so we make sure to increment the lock count
640
repository.lock_read()
641
self._fallback_repositories.append(repository)
642
self.texts.add_fallback_versioned_files(repository.texts)
643
self.inventories.add_fallback_versioned_files(repository.inventories)
644
self.revisions.add_fallback_versioned_files(repository.revisions)
645
self.signatures.add_fallback_versioned_files(repository.signatures)
646
if self.chk_bytes is not None:
647
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
649
@only_raises(errors.LockNotHeld, errors.LockBroken)
651
super(VersionedFileRepository, self).unlock()
652
if self.control_files._lock_count == 0:
653
self._inventory_entry_cache.clear()
655
def add_inventory(self, revision_id, inv, parents):
656
"""Add the inventory inv to the repository as revision_id.
658
:param parents: The revision ids of the parents that revision_id
659
is known to have and are in the repository already.
661
:returns: The validator(which is a sha1 digest, though what is sha'd is
662
repository format specific) of the serialized inventory.
664
if not self.is_in_write_group():
665
raise AssertionError("%r not in write group" % (self,))
666
_mod_revision.check_not_reserved_id(revision_id)
667
if not (inv.revision_id is None or inv.revision_id == revision_id):
668
raise AssertionError(
669
"Mismatch between inventory revision"
670
" id and insertion revid (%r, %r)"
671
% (inv.revision_id, revision_id))
673
raise errors.RootMissing()
674
return self._add_inventory_checked(revision_id, inv, parents)
676
def _add_inventory_checked(self, revision_id, inv, parents):
677
"""Add inv to the repository after checking the inputs.
679
This function can be overridden to allow different inventory styles.
681
:seealso: add_inventory, for the contract.
683
inv_lines = self._serializer.write_inventory_to_lines(inv)
684
return self._inventory_add_lines(revision_id, parents,
685
inv_lines, check_content=False)
687
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
688
parents, basis_inv=None, propagate_caches=False):
689
"""Add a new inventory expressed as a delta against another revision.
691
See the inventory developers documentation for the theory behind
694
:param basis_revision_id: The inventory id the delta was created
695
against. (This does not have to be a direct parent.)
696
:param delta: The inventory delta (see Inventory.apply_delta for
698
:param new_revision_id: The revision id that the inventory is being
700
:param parents: The revision ids of the parents that revision_id is
701
known to have and are in the repository already. These are supplied
702
for repositories that depend on the inventory graph for revision
703
graph access, as well as for those that pun ancestry with delta
705
:param basis_inv: The basis inventory if it is already known,
707
:param propagate_caches: If True, the caches for this inventory are
708
copied to and updated for the result if possible.
710
:returns: (validator, new_inv)
711
The validator(which is a sha1 digest, though what is sha'd is
712
repository format specific) of the serialized inventory, and the
715
if not self.is_in_write_group():
716
raise AssertionError("%r not in write group" % (self,))
717
_mod_revision.check_not_reserved_id(new_revision_id)
718
basis_tree = self.revision_tree(basis_revision_id)
719
with basis_tree.lock_read():
720
# Note that this mutates the inventory of basis_tree, which not all
721
# inventory implementations may support: A better idiom would be to
722
# return a new inventory, but as there is no revision tree cache in
723
# repository this is safe for now - RBC 20081013
724
if basis_inv is None:
725
basis_inv = basis_tree.root_inventory
726
basis_inv.apply_delta(delta)
727
basis_inv.revision_id = new_revision_id
728
return (self.add_inventory(new_revision_id, basis_inv, parents),
731
def _inventory_add_lines(self, revision_id, parents, lines,
733
"""Store lines in inv_vf and return the sha1 of the inventory."""
734
parents = [(parent,) for parent in parents]
735
result = self.inventories.add_lines((revision_id,), parents, lines,
736
check_content=check_content)[0]
737
self.inventories._access.flush()
740
def add_revision(self, revision_id, rev, inv=None):
741
"""Add rev to the revision store as revision_id.
743
:param revision_id: the revision id to use.
744
:param rev: The revision object.
745
:param inv: The inventory for the revision. if None, it will be looked
746
up in the inventory storer
748
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
750
_mod_revision.check_not_reserved_id(revision_id)
751
# check inventory present
752
if not self.inventories.get_parent_map([(revision_id,)]):
754
raise errors.WeaveRevisionNotPresent(revision_id,
757
# yes, this is not suitable for adding with ghosts.
758
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
762
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
763
self._add_revision(rev)
765
def _add_revision(self, revision):
766
text = self._serializer.write_revision_to_string(revision)
767
key = (revision.revision_id,)
768
parents = tuple((parent,) for parent in revision.parent_ids)
769
self.revisions.add_lines(key, parents, osutils.split_lines(text))
771
def _check_inventories(self, checker):
772
"""Check the inventories found from the revision scan.
774
This is responsible for verifying the sha1 of inventories and
775
creating a pending_keys set that covers data referenced by inventories.
777
with ui.ui_factory.nested_progress_bar() as bar:
778
self._do_check_inventories(checker, bar)
780
def _do_check_inventories(self, checker, bar):
781
"""Helper for _check_inventories."""
783
keys = {'chk_bytes': set(), 'inventories': set(), 'texts': set()}
784
kinds = ['chk_bytes', 'texts']
785
count = len(checker.pending_keys)
786
bar.update(gettext("inventories"), 0, 2)
787
current_keys = checker.pending_keys
788
checker.pending_keys = {}
789
# Accumulate current checks.
790
for key in current_keys:
791
if key[0] != 'inventories' and key[0] not in kinds:
792
checker._report_items.append('unknown key type %r' % (key,))
793
keys[key[0]].add(key[1:])
794
if keys['inventories']:
795
# NB: output order *should* be roughly sorted - topo or
796
# inverse topo depending on repository - either way decent
797
# to just delta against. However, pre-CHK formats didn't
798
# try to optimise inventory layout on disk. As such the
799
# pre-CHK code path does not use inventory deltas.
801
for record in self.inventories.check(keys=keys['inventories']):
802
if record.storage_kind == 'absent':
803
checker._report_items.append(
804
'Missing inventory {%s}' % (record.key,))
806
last_object = self._check_record('inventories', record,
807
checker, last_object,
808
current_keys[('inventories',) + record.key])
809
del keys['inventories']
812
bar.update(gettext("texts"), 1)
813
while (checker.pending_keys or keys['chk_bytes'] or
815
# Something to check.
816
current_keys = checker.pending_keys
817
checker.pending_keys = {}
818
# Accumulate current checks.
819
for key in current_keys:
820
if key[0] not in kinds:
821
checker._report_items.append(
822
'unknown key type %r' % (key,))
823
keys[key[0]].add(key[1:])
824
# Check the outermost kind only - inventories || chk_bytes || texts
828
for record in getattr(self, kind).check(keys=keys[kind]):
829
if record.storage_kind == 'absent':
830
checker._report_items.append(
831
'Missing %s {%s}' % (kind, record.key,))
833
last_object = self._check_record(kind, record,
834
checker, last_object, current_keys[(kind,) + record.key])
838
def _check_record(self, kind, record, checker, last_object, item_data):
839
"""Check a single text from this repository."""
840
if kind == 'inventories':
841
rev_id = record.key[0]
842
inv = self._deserialise_inventory(rev_id,
843
record.get_bytes_as('fulltext'))
844
if last_object is not None:
845
delta = inv._make_delta(last_object)
846
for old_path, path, file_id, ie in delta:
849
ie.check(checker, rev_id, inv)
851
for path, ie in inv.iter_entries():
852
ie.check(checker, rev_id, inv)
853
if self._format.fast_deltas:
855
elif kind == 'chk_bytes':
856
# No code written to check chk_bytes for this repo format.
857
checker._report_items.append(
858
'unsupported key type chk_bytes for %s' % (record.key,))
859
elif kind == 'texts':
860
self._check_text(record, checker, item_data)
862
checker._report_items.append(
863
'unknown key type %s for %s' % (kind, record.key))
865
def _check_text(self, record, checker, item_data):
866
"""Check a single text."""
867
# Check it is extractable.
868
# TODO: check length.
869
if record.storage_kind == 'chunked':
870
chunks = record.get_bytes_as(record.storage_kind)
871
sha1 = osutils.sha_strings(chunks)
872
length = sum(map(len, chunks))
874
content = record.get_bytes_as('fulltext')
875
sha1 = osutils.sha_string(content)
876
length = len(content)
877
if item_data and sha1 != item_data[1]:
878
checker._report_items.append(
879
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
880
(record.key, sha1, item_data[1], item_data[2]))
882
def _eliminate_revisions_not_present(self, revision_ids):
883
"""Check every revision id in revision_ids to see if we have it.
885
Returns a set of the present revisions.
887
with self.lock_read():
889
graph = self.get_graph()
890
parent_map = graph.get_parent_map(revision_ids)
891
# The old API returned a list, should this actually be a set?
892
return list(parent_map)
894
def __init__(self, _format, a_controldir, control_files):
895
"""Instantiate a VersionedFileRepository.
897
:param _format: The format of the repository on disk.
898
:param controldir: The ControlDir of the repository.
899
:param control_files: Control files to use for locking, etc.
901
# In the future we will have a single api for all stores for
902
# getting file texts, inventories and revisions, then
903
# this construct will accept instances of those things.
904
super(VersionedFileRepository, self).__init__(_format, a_controldir,
906
self._transport = control_files._transport
907
self.base = self._transport.base
909
self._reconcile_does_inventory_gc = True
910
self._reconcile_fixes_text_parents = False
911
self._reconcile_backsup_inventory = True
912
# An InventoryEntry cache, used during deserialization
913
self._inventory_entry_cache = fifo_cache.FIFOCache(10 * 1024)
914
# Is it safe to return inventory entries directly from the entry cache,
915
# rather copying them?
916
self._safe_to_return_from_cache = False
918
def fetch(self, source, revision_id=None, find_ghosts=False,
920
"""Fetch the content required to construct revision_id from source.
922
If revision_id is None and fetch_spec is None, then all content is
925
fetch() may not be used when the repository is in a write group -
926
either finish the current write group before using fetch, or use
927
fetch before starting the write group.
929
:param find_ghosts: Find and copy revisions in the source that are
930
ghosts in the target (and not reachable directly by walking out to
931
the first-present revision in target from revision_id).
932
:param revision_id: If specified, all the content needed for this
933
revision ID will be copied to the target. Fetch will determine for
934
itself which content needs to be copied.
935
:param fetch_spec: If specified, a SearchResult or
936
PendingAncestryResult that describes which revisions to copy. This
937
allows copying multiple heads at once. Mutually exclusive with
940
if fetch_spec is not None and revision_id is not None:
941
raise AssertionError(
942
"fetch_spec and revision_id are mutually exclusive.")
943
if self.is_in_write_group():
944
raise errors.InternalBzrError(
945
"May not fetch while in a write group.")
946
# fast path same-url fetch operations
947
# TODO: lift out to somewhere common with RemoteRepository
948
# <https://bugs.launchpad.net/bzr/+bug/401646>
949
if (self.has_same_location(source) and
950
fetch_spec is None and
951
self._has_same_fallbacks(source)):
952
# check that last_revision is in 'from' and then return a
954
if (revision_id is not None
955
and not _mod_revision.is_null(revision_id)):
956
self.get_revision(revision_id)
958
inter = InterRepository.get(source, self)
959
if (fetch_spec is not None
960
and not getattr(inter, "supports_fetch_spec", False)):
961
raise errors.UnsupportedOperation(
962
"fetch_spec not supported for %r" % inter)
963
return inter.fetch(revision_id=revision_id,
964
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
966
def gather_stats(self, revid=None, committers=None):
967
"""See Repository.gather_stats()."""
968
with self.lock_read():
969
result = super(VersionedFileRepository,
970
self).gather_stats(revid, committers)
971
# now gather global repository information
972
# XXX: This is available for many repos regardless of listability.
973
if self.user_transport.listable():
974
# XXX: do we want to __define len__() ?
975
# Maybe the versionedfiles object should provide a different
976
# method to get the number of keys.
977
result['revisions'] = len(self.revisions.keys())
981
def get_commit_builder(self, branch, parents, config_stack, timestamp=None,
982
timezone=None, committer=None, revprops=None,
983
revision_id=None, lossy=False):
984
"""Obtain a CommitBuilder for this repository.
986
:param branch: Branch to commit to.
987
:param parents: Revision ids of the parents of the new revision.
988
:param config_stack: Configuration stack to use.
989
:param timestamp: Optional timestamp recorded for commit.
990
:param timezone: Optional timezone for timestamp.
991
:param committer: Optional committer to set for commit.
992
:param revprops: Optional dictionary of revision properties.
993
:param revision_id: Optional revision id.
994
:param lossy: Whether to discard data that can not be natively
995
represented, when pushing to a foreign VCS
997
if self._fallback_repositories and not self._format.supports_chks:
998
raise errors.BzrError("Cannot commit directly to a stacked branch"
999
" in pre-2a formats. See "
1000
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1001
result = self._commit_builder_class(self, parents, config_stack,
1002
timestamp, timezone, committer, revprops, revision_id,
1004
self.start_write_group()
1007
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1008
"""Return the keys of missing inventory parents for revisions added in
1011
A revision is not complete if the inventory delta for that revision
1012
cannot be calculated. Therefore if the parent inventories of a
1013
revision are not present, the revision is incomplete, and e.g. cannot
1014
be streamed by a smart server. This method finds missing inventory
1015
parents for revisions added in this write group.
1017
if not self._format.supports_external_lookups:
1018
# This is only an issue for stacked repositories
1020
if not self.is_in_write_group():
1021
raise AssertionError('not in a write group')
1023
# XXX: We assume that every added revision already has its
1024
# corresponding inventory, so we only check for parent inventories that
1025
# might be missing, rather than all inventories.
1026
parents = set(self.revisions._index.get_missing_parents())
1027
parents.discard(_mod_revision.NULL_REVISION)
1028
unstacked_inventories = self.inventories._index
1029
present_inventories = unstacked_inventories.get_parent_map(
1030
key[-1:] for key in parents)
1031
parents.difference_update(present_inventories)
1032
if len(parents) == 0:
1033
# No missing parent inventories.
1035
if not check_for_missing_texts:
1036
return set(('inventories', rev_id) for (rev_id,) in parents)
1037
# Ok, now we have a list of missing inventories. But these only matter
1038
# if the inventories that reference them are missing some texts they
1039
# appear to introduce.
1040
# XXX: Texts referenced by all added inventories need to be present,
1041
# but at the moment we're only checking for texts referenced by
1042
# inventories at the graph's edge.
1043
key_deps = self.revisions._index._key_dependencies
1044
key_deps.satisfy_refs_for_keys(present_inventories)
1045
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1046
file_ids = self.fileids_altered_by_revision_ids(referrers)
1047
missing_texts = set()
1048
for file_id, version_ids in viewitems(file_ids):
1049
missing_texts.update(
1050
(file_id, version_id) for version_id in version_ids)
1051
present_texts = self.texts.get_parent_map(missing_texts)
1052
missing_texts.difference_update(present_texts)
1053
if not missing_texts:
1054
# No texts are missing, so all revisions and their deltas are
1057
# Alternatively the text versions could be returned as the missing
1058
# keys, but this is likely to be less data.
1059
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1062
def has_revisions(self, revision_ids):
1063
"""Probe to find out the presence of multiple revisions.
1065
:param revision_ids: An iterable of revision_ids.
1066
:return: A set of the revision_ids that were present.
1068
with self.lock_read():
1069
parent_map = self.revisions.get_parent_map(
1070
[(rev_id,) for rev_id in revision_ids])
1072
if _mod_revision.NULL_REVISION in revision_ids:
1073
result.add(_mod_revision.NULL_REVISION)
1074
result.update([key[0] for key in parent_map])
1077
def get_revision_reconcile(self, revision_id):
1078
"""'reconcile' helper routine that allows access to a revision always.
1080
This variant of get_revision does not cross check the weave graph
1081
against the revision one as get_revision does: but it should only
1082
be used by reconcile, or reconcile-alike commands that are correcting
1083
or testing the revision graph.
1085
with self.lock_read():
1086
return self.get_revisions([revision_id])[0]
1088
def iter_revisions(self, revision_ids):
1089
"""Iterate over revision objects.
1091
:param revision_ids: An iterable of revisions to examine. None may be
1092
passed to request all revisions known to the repository. Note that
1093
not all repositories can find unreferenced revisions; for those
1094
repositories only referenced ones will be returned.
1095
:return: An iterator of (revid, revision) tuples. Absent revisions (
1096
those asked for but not available) are returned as (revid, None).
1098
with self.lock_read():
1099
for rev_id in revision_ids:
1100
if not rev_id or not isinstance(rev_id, bytes):
1101
raise errors.InvalidRevisionId(
1102
revision_id=rev_id, branch=self)
1103
keys = [(key,) for key in revision_ids]
1104
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1105
for record in stream:
1106
revid = record.key[0]
1107
if record.storage_kind == 'absent':
1110
text = record.get_bytes_as('fulltext')
1111
rev = self._serializer.read_revision_from_string(text)
1114
def add_signature_text(self, revision_id, signature):
1115
"""Store a signature text for a revision.
1117
:param revision_id: Revision id of the revision
1118
:param signature: Signature text.
1120
with self.lock_write():
1121
self.signatures.add_lines((revision_id,), (),
1122
osutils.split_lines(signature))
1124
def find_text_key_references(self):
1125
"""Find the text key references within the repository.
1127
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1128
to whether they were referred to by the inventory of the
1129
revision_id that they contain. The inventory texts from all present
1130
revision ids are assessed to generate this report.
1132
revision_keys = self.revisions.keys()
1133
w = self.inventories
1134
with ui.ui_factory.nested_progress_bar() as pb:
1135
return self._serializer._find_text_key_references(
1136
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1138
def _inventory_xml_lines_for_keys(self, keys):
1139
"""Get a line iterator of the sort needed for findind references.
1141
Not relevant for non-xml inventory repositories.
1143
Ghosts in revision_keys are ignored.
1145
:param revision_keys: The revision keys for the inventories to inspect.
1146
:return: An iterator over (inventory line, revid) for the fulltexts of
1147
all of the xml inventories specified by revision_keys.
1149
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1150
for record in stream:
1151
if record.storage_kind != 'absent':
1152
chunks = record.get_bytes_as('chunked')
1153
revid = record.key[-1]
1154
lines = osutils.chunks_to_lines(chunks)
1158
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1160
"""Helper routine for fileids_altered_by_revision_ids.
1162
This performs the translation of xml lines to revision ids.
1164
:param line_iterator: An iterator of lines, origin_version_id
1165
:param revision_keys: The revision ids to filter for. This should be a
1166
set or other type which supports efficient __contains__ lookups, as
1167
the revision key from each parsed line will be looked up in the
1168
revision_keys filter.
1169
:return: a dictionary mapping altered file-ids to an iterable of
1170
revision_ids. Each altered file-ids has the exact revision_ids that
1171
altered it listed explicitly.
1173
seen = set(self._serializer._find_text_key_references(line_iterator))
1174
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
1175
parent_seen = set(self._serializer._find_text_key_references(
1176
self._inventory_xml_lines_for_keys(parent_keys)))
1177
new_keys = seen - parent_seen
1179
setdefault = result.setdefault
1180
for key in new_keys:
1181
setdefault(key[0], set()).add(key[-1])
1184
def _find_parent_keys_of_revisions(self, revision_keys):
1185
"""Similar to _find_parent_ids_of_revisions, but used with keys.
1187
:param revision_keys: An iterable of revision_keys.
1188
:return: The parents of all revision_keys that are not already in
1191
parent_map = self.revisions.get_parent_map(revision_keys)
1192
parent_keys = set(itertools.chain.from_iterable(
1193
viewvalues(parent_map)))
1194
parent_keys.difference_update(revision_keys)
1195
parent_keys.discard(_mod_revision.NULL_REVISION)
1198
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1199
"""Find the file ids and versions affected by revisions.
1201
:param revisions: an iterable containing revision ids.
1202
:param _inv_weave: The inventory weave from this repository or None.
1203
If None, the inventory weave will be opened automatically.
1204
:return: a dictionary mapping altered file-ids to an iterable of
1205
revision_ids. Each altered file-ids has the exact revision_ids that
1206
altered it listed explicitly.
1208
selected_keys = set((revid,) for revid in revision_ids)
1209
w = _inv_weave or self.inventories
1210
return self._find_file_ids_from_xml_inventory_lines(
1211
w.iter_lines_added_or_present_in_keys(
1212
selected_keys, pb=None),
1215
def iter_files_bytes(self, desired_files):
1216
"""Iterate through file versions.
1218
Files will not necessarily be returned in the order they occur in
1219
desired_files. No specific order is guaranteed.
1221
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1222
value supplied by the caller as part of desired_files. It should
1223
uniquely identify the file version in the caller's context. (Examples:
1224
an index number or a TreeTransform trans_id.)
1226
bytes_iterator is an iterable of bytestrings for the file. The
1227
kind of iterable and length of the bytestrings are unspecified, but for
1228
this implementation, it is a list of bytes produced by
1229
VersionedFile.get_record_stream().
1231
:param desired_files: a list of (file_id, revision_id, identifier)
1235
for file_id, revision_id, callable_data in desired_files:
1236
text_keys[(file_id, revision_id)] = callable_data
1237
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1238
if record.storage_kind == 'absent':
1239
raise errors.RevisionNotPresent(record.key[1], record.key[0])
1240
yield text_keys[record.key], record.get_bytes_as('chunked')
1242
def _generate_text_key_index(self, text_key_references=None,
1244
"""Generate a new text key index for the repository.
1246
This is an expensive function that will take considerable time to run.
1248
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1249
list of parents, also text keys. When a given key has no parents,
1250
the parents list will be [NULL_REVISION].
1252
# All revisions, to find inventory parents.
1253
if ancestors is None:
1254
graph = self.get_graph()
1255
ancestors = graph.get_parent_map(self.all_revision_ids())
1256
if text_key_references is None:
1257
text_key_references = self.find_text_key_references()
1258
with ui.ui_factory.nested_progress_bar() as pb:
1259
return self._do_generate_text_key_index(ancestors,
1260
text_key_references, pb)
1262
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1263
"""Helper for _generate_text_key_index to avoid deep nesting."""
1264
revision_order = tsort.topo_sort(ancestors)
1265
invalid_keys = set()
1267
for revision_id in revision_order:
1268
revision_keys[revision_id] = set()
1269
text_count = len(text_key_references)
1270
# a cache of the text keys to allow reuse; costs a dict of all the
1271
# keys, but saves a 2-tuple for every child of a given key.
1273
for text_key, valid in viewitems(text_key_references):
1275
invalid_keys.add(text_key)
1277
revision_keys[text_key[1]].add(text_key)
1278
text_key_cache[text_key] = text_key
1279
del text_key_references
1281
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1282
NULL_REVISION = _mod_revision.NULL_REVISION
1283
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1284
# too small for large or very branchy trees. However, for 55K path
1285
# trees, it would be easy to use too much memory trivially. Ideally we
1286
# could gauge this by looking at available real memory etc, but this is
1287
# always a tricky proposition.
1288
inventory_cache = lru_cache.LRUCache(10)
1289
batch_size = 10 # should be ~150MB on a 55K path tree
1290
batch_count = len(revision_order) // batch_size + 1
1292
pb.update(gettext("Calculating text parents"),
1293
processed_texts, text_count)
1294
for offset in range(batch_count):
1295
to_query = revision_order[offset * batch_size:(offset + 1)
1299
for revision_id in to_query:
1300
parent_ids = ancestors[revision_id]
1301
for text_key in revision_keys[revision_id]:
1302
pb.update(gettext("Calculating text parents"),
1304
processed_texts += 1
1305
candidate_parents = []
1306
for parent_id in parent_ids:
1307
parent_text_key = (text_key[0], parent_id)
1309
check_parent = parent_text_key not in \
1310
revision_keys[parent_id]
1312
# the parent parent_id is a ghost:
1313
check_parent = False
1314
# truncate the derived graph against this ghost.
1315
parent_text_key = None
1317
# look at the parent commit details inventories to
1318
# determine possible candidates in the per file graph.
1321
inv = inventory_cache[parent_id]
1323
inv = self.revision_tree(
1324
parent_id).root_inventory
1325
inventory_cache[parent_id] = inv
1327
parent_entry = inv.get_entry(text_key[0])
1328
except (KeyError, errors.NoSuchId):
1330
if parent_entry is not None:
1332
text_key[0], parent_entry.revision)
1334
parent_text_key = None
1335
if parent_text_key is not None:
1336
candidate_parents.append(
1337
text_key_cache[parent_text_key])
1338
parent_heads = text_graph.heads(candidate_parents)
1339
new_parents = list(parent_heads)
1340
new_parents.sort(key=lambda x: candidate_parents.index(x))
1341
if new_parents == []:
1342
new_parents = [NULL_REVISION]
1343
text_index[text_key] = new_parents
1345
for text_key in invalid_keys:
1346
text_index[text_key] = [NULL_REVISION]
1349
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
1350
"""Get an iterable listing the keys of all the data introduced by a set
1353
The keys will be ordered so that the corresponding items can be safely
1354
fetched and inserted in that order.
1356
:returns: An iterable producing tuples of (knit-kind, file-id,
1357
versions). knit-kind is one of 'file', 'inventory', 'signatures',
1358
'revisions'. file-id is None unless knit-kind is 'file'.
1360
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
1363
for result in self._find_non_file_keys_to_fetch(revision_ids):
1366
def _find_file_keys_to_fetch(self, revision_ids, pb):
1367
# XXX: it's a bit weird to control the inventory weave caching in this
1368
# generator. Ideally the caching would be done in fetch.py I think. Or
1369
# maybe this generator should explicitly have the contract that it
1370
# should not be iterated until the previously yielded item has been
1372
inv_w = self.inventories
1374
# file ids that changed
1375
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
1377
num_file_ids = len(file_ids)
1378
for file_id, altered_versions in viewitems(file_ids):
1380
pb.update(gettext("Fetch texts"), count, num_file_ids)
1382
yield ("file", file_id, altered_versions)
1384
def _find_non_file_keys_to_fetch(self, revision_ids):
1386
yield ("inventory", None, revision_ids)
1389
# XXX: Note ATM no callers actually pay attention to this return
1390
# instead they just use the list of revision ids and ignore
1391
# missing sigs. Consider removing this work entirely
1392
revisions_with_signatures = set(self.signatures.get_parent_map(
1393
[(r,) for r in revision_ids]))
1394
revisions_with_signatures = {r for (r,) in revisions_with_signatures}
1395
revisions_with_signatures.intersection_update(revision_ids)
1396
yield ("signatures", None, revisions_with_signatures)
1399
yield ("revisions", None, revision_ids)
1401
def get_inventory(self, revision_id):
1402
"""Get Inventory object by revision id."""
1403
with self.lock_read():
1404
return next(self.iter_inventories([revision_id]))
1406
def iter_inventories(self, revision_ids, ordering=None):
1407
"""Get many inventories by revision_ids.
1409
This will buffer some or all of the texts used in constructing the
1410
inventories in memory, but will only parse a single inventory at a
1413
:param revision_ids: The expected revision ids of the inventories.
1414
:param ordering: optional ordering, e.g. 'topological'. If not
1415
specified, the order of revision_ids will be preserved (by
1416
buffering if necessary).
1417
:return: An iterator of inventories.
1419
if ((None in revision_ids) or
1420
(_mod_revision.NULL_REVISION in revision_ids)):
1421
raise ValueError('cannot get null revision inventory')
1422
for inv, revid in self._iter_inventories(revision_ids, ordering):
1424
raise errors.NoSuchRevision(self, revid)
1427
def _iter_inventories(self, revision_ids, ordering):
1428
"""single-document based inventory iteration."""
1429
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
1430
for text, revision_id in inv_xmls:
1432
yield None, revision_id
1434
yield self._deserialise_inventory(revision_id, text), revision_id
1436
def _iter_inventory_xmls(self, revision_ids, ordering):
1437
if ordering is None:
1438
order_as_requested = True
1439
ordering = 'unordered'
1441
order_as_requested = False
1442
keys = [(revision_id,) for revision_id in revision_ids]
1445
if order_as_requested:
1446
key_iter = iter(keys)
1447
next_key = next(key_iter)
1448
stream = self.inventories.get_record_stream(keys, ordering, True)
1450
for record in stream:
1451
if record.storage_kind != 'absent':
1452
chunks = record.get_bytes_as('chunked')
1453
if order_as_requested:
1454
text_chunks[record.key] = chunks
1456
yield b''.join(chunks), record.key[-1]
1458
yield None, record.key[-1]
1459
if order_as_requested:
1460
# Yield as many results as we can while preserving order.
1461
while next_key in text_chunks:
1462
chunks = text_chunks.pop(next_key)
1463
yield b''.join(chunks), next_key[-1]
1465
next_key = next(key_iter)
1466
except StopIteration:
1467
# We still want to fully consume the get_record_stream,
1468
# just in case it is not actually finished at this point
1472
def _deserialise_inventory(self, revision_id, xml):
1473
"""Transform the xml into an inventory object.
1475
:param revision_id: The expected revision id of the inventory.
1476
:param xml: A serialised inventory.
1478
result = self._serializer.read_inventory_from_string(xml, revision_id,
1479
entry_cache=self._inventory_entry_cache,
1480
return_from_cache=self._safe_to_return_from_cache)
1481
if result.revision_id != revision_id:
1482
raise AssertionError('revision id mismatch %s != %s' % (
1483
result.revision_id, revision_id))
1486
def get_serializer_format(self):
1487
return self._serializer.format_num
1489
def _get_inventory_xml(self, revision_id):
1490
"""Get serialized inventory as a string."""
1491
with self.lock_read():
1492
texts = self._iter_inventory_xmls([revision_id], 'unordered')
1493
text, revision_id = next(texts)
1495
raise errors.NoSuchRevision(self, revision_id)
1498
def revision_tree(self, revision_id):
1499
"""Return Tree for a revision on this branch.
1501
`revision_id` may be NULL_REVISION for the empty tree revision.
1503
revision_id = _mod_revision.ensure_null(revision_id)
1504
# TODO: refactor this to use an existing revision object
1505
# so we don't need to read it in twice.
1506
if revision_id == _mod_revision.NULL_REVISION:
1507
return inventorytree.InventoryRevisionTree(self,
1508
Inventory(root_id=None), _mod_revision.NULL_REVISION)
1510
with self.lock_read():
1511
inv = self.get_inventory(revision_id)
1512
return inventorytree.InventoryRevisionTree(self, inv, revision_id)
1514
def revision_trees(self, revision_ids):
1515
"""Return Trees for revisions in this repository.
1517
:param revision_ids: a sequence of revision-ids;
1518
a revision-id may not be None or b'null:'
1520
inventories = self.iter_inventories(revision_ids)
1521
for inv in inventories:
1522
yield inventorytree.InventoryRevisionTree(self, inv, inv.revision_id)
1524
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1525
"""Produce a generator of revision deltas.
1527
Note that the input is a sequence of REVISIONS, not revision_ids.
1528
Trees will be held in memory until the generator exits.
1529
Each delta is relative to the revision's lefthand predecessor.
1531
:param specific_fileids: if not None, the result is filtered
1532
so that only those file-ids, their parents and their
1533
children are included.
1535
# Get the revision-ids of interest
1536
required_trees = set()
1537
for revision in revisions:
1538
required_trees.add(revision.revision_id)
1539
required_trees.update(revision.parent_ids[:1])
1541
# Get the matching filtered trees. Note that it's more
1542
# efficient to pass filtered trees to changes_from() rather
1543
# than doing the filtering afterwards. changes_from() could
1544
# arguably do the filtering itself but it's path-based, not
1545
# file-id based, so filtering before or afterwards is
1547
if specific_fileids is None:
1548
trees = dict((t.get_revision_id(), t) for
1549
t in self.revision_trees(required_trees))
1551
trees = dict((t.get_revision_id(), t) for
1552
t in self._filtered_revision_trees(required_trees,
1555
# Calculate the deltas
1556
for revision in revisions:
1557
if not revision.parent_ids:
1558
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1560
old_tree = trees[revision.parent_ids[0]]
1561
yield trees[revision.revision_id].changes_from(old_tree)
1563
def _filtered_revision_trees(self, revision_ids, file_ids):
1564
"""Return Tree for a revision on this branch with only some files.
1566
:param revision_ids: a sequence of revision-ids;
1567
a revision-id may not be None or b'null:'
1568
:param file_ids: if not None, the result is filtered
1569
so that only those file-ids, their parents and their
1570
children are included.
1572
inventories = self.iter_inventories(revision_ids)
1573
for inv in inventories:
1574
# Should we introduce a FilteredRevisionTree class rather
1575
# than pre-filter the inventory here?
1576
filtered_inv = inv.filter(file_ids)
1577
yield inventorytree.InventoryRevisionTree(self, filtered_inv, filtered_inv.revision_id)
1579
def get_parent_map(self, revision_ids):
1580
"""See graph.StackedParentsProvider.get_parent_map"""
1581
# revisions index works in keys; this just works in revisions
1582
# therefore wrap and unwrap
1585
for revision_id in revision_ids:
1586
if revision_id == _mod_revision.NULL_REVISION:
1587
result[revision_id] = ()
1588
elif revision_id is None:
1589
raise ValueError('get_parent_map(None) is not valid')
1591
query_keys.append((revision_id,))
1592
for (revision_id,), parent_keys in viewitems(
1593
self.revisions.get_parent_map(query_keys)):
1595
result[revision_id] = tuple([parent_revid
1596
for (parent_revid,) in parent_keys])
1598
result[revision_id] = (_mod_revision.NULL_REVISION,)
1601
def get_known_graph_ancestry(self, revision_ids):
1602
"""Return the known graph for a set of revision ids and their ancestors.
1604
st = static_tuple.StaticTuple
1605
revision_keys = [st(r_id).intern() for r_id in revision_ids]
1606
with self.lock_read():
1607
known_graph = self.revisions.get_known_graph_ancestry(
1609
return graph.GraphThunkIdsToKeys(known_graph)
1611
def get_file_graph(self):
1612
"""Return the graph walker for text revisions."""
1613
with self.lock_read():
1614
return graph.Graph(self.texts)
1616
def revision_ids_to_search_result(self, result_set):
1617
"""Convert a set of revision ids to a graph SearchResult."""
1618
result_parents = set(itertools.chain.from_iterable(viewvalues(
1619
self.get_graph().get_parent_map(result_set))))
1620
included_keys = result_set.intersection(result_parents)
1621
start_keys = result_set.difference(included_keys)
1622
exclude_keys = result_parents.difference(result_set)
1623
result = vf_search.SearchResult(start_keys, exclude_keys,
1624
len(result_set), result_set)
1627
def _get_versioned_file_checker(self, text_key_references=None,
1629
"""Return an object suitable for checking versioned files.
1631
:param text_key_references: if non-None, an already built
1632
dictionary mapping text keys ((fileid, revision_id) tuples)
1633
to whether they were referred to by the inventory of the
1634
revision_id that they contain. If None, this will be
1636
:param ancestors: Optional result from
1637
self.get_graph().get_parent_map(self.all_revision_ids()) if already
1640
return _VersionedFileChecker(self,
1641
text_key_references=text_key_references, ancestors=ancestors)
1643
def has_signature_for_revision_id(self, revision_id):
1644
"""Query for a revision signature for revision_id in the repository."""
1645
with self.lock_read():
1646
if not self.has_revision(revision_id):
1647
raise errors.NoSuchRevision(self, revision_id)
1648
sig_present = (1 == len(
1649
self.signatures.get_parent_map([(revision_id,)])))
1652
def get_signature_text(self, revision_id):
1653
"""Return the text for a signature."""
1654
with self.lock_read():
1655
stream = self.signatures.get_record_stream([(revision_id,)],
1657
record = next(stream)
1658
if record.storage_kind == 'absent':
1659
raise errors.NoSuchRevision(self, revision_id)
1660
return record.get_bytes_as('fulltext')
1662
def _check(self, revision_ids, callback_refs, check_repo):
1663
with self.lock_read():
1664
result = check.VersionedFileCheck(self, check_repo=check_repo)
1665
result.check(callback_refs)
1668
def _find_inconsistent_revision_parents(self, revisions_iterator=None):
1669
"""Find revisions with different parent lists in the revision object
1670
and in the index graph.
1672
:param revisions_iterator: None, or an iterator of (revid,
1673
Revision-or-None). This iterator controls the revisions checked.
1674
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1675
parents-in-revision).
1677
if not self.is_locked():
1678
raise AssertionError()
1680
if revisions_iterator is None:
1681
revisions_iterator = self.iter_revisions(self.all_revision_ids())
1682
for revid, revision in revisions_iterator:
1683
if revision is None:
1685
parent_map = vf.get_parent_map([(revid,)])
1686
parents_according_to_index = tuple(parent[-1] for parent in
1687
parent_map[(revid,)])
1688
parents_according_to_revision = tuple(revision.parent_ids)
1689
if parents_according_to_index != parents_according_to_revision:
1690
yield (revid, parents_according_to_index,
1691
parents_according_to_revision)
1693
def _check_for_inconsistent_revision_parents(self):
1694
inconsistencies = list(self._find_inconsistent_revision_parents())
1696
raise errors.BzrCheckError(
1697
"Revision knit has inconsistent parents.")
1699
def _get_sink(self):
1700
"""Return a sink for streaming into this repository."""
1701
return StreamSink(self)
1703
def _get_source(self, to_format):
1704
"""Return a source for streaming from this repository."""
1705
return StreamSource(self, to_format)
1708
class MetaDirVersionedFileRepository(MetaDirRepository,
1709
VersionedFileRepository):
1710
"""Repositories in a meta-dir, that work via versioned file objects."""
1712
def __init__(self, _format, a_controldir, control_files):
1713
super(MetaDirVersionedFileRepository, self).__init__(_format, a_controldir,
1717
class MetaDirVersionedFileRepositoryFormat(RepositoryFormatMetaDir,
1718
VersionedFileRepositoryFormat):
1719
"""Base class for repository formats using versioned files in metadirs."""
1722
class StreamSink(object):
1723
"""An object that can insert a stream into a repository.
1725
This interface handles the complexity of reserialising inventories and
1726
revisions from different formats, and allows unidirectional insertion into
1727
stacked repositories without looking for the missing basis parents
1731
def __init__(self, target_repo):
1732
self.target_repo = target_repo
1734
def insert_missing_keys(self, source, missing_keys):
1735
"""Insert missing keys from another source.
1737
:param source: StreamSource to stream from
1738
:param missing_keys: Keys to insert
1739
:return: keys still missing
1741
stream = source.get_stream_for_missing_keys(missing_keys)
1742
return self.insert_stream_without_locking(stream,
1743
self.target_repo._format)
1745
def insert_stream(self, stream, src_format, resume_tokens):
1746
"""Insert a stream's content into the target repository.
1748
:param src_format: a bzr repository format.
1750
:return: a list of resume tokens and an iterable of keys additional
1751
items required before the insertion can be completed.
1753
with self.target_repo.lock_write():
1755
self.target_repo.resume_write_group(resume_tokens)
1758
self.target_repo.start_write_group()
1761
# locked_insert_stream performs a commit|suspend.
1762
missing_keys = self.insert_stream_without_locking(stream,
1763
src_format, is_resume)
1765
# suspend the write group and tell the caller what we is
1766
# missing. We know we can suspend or else we would not have
1767
# entered this code path. (All repositories that can handle
1768
# missing keys can handle suspending a write group).
1769
write_group_tokens = self.target_repo.suspend_write_group()
1770
return write_group_tokens, missing_keys
1771
hint = self.target_repo.commit_write_group()
1772
to_serializer = self.target_repo._format._serializer
1773
src_serializer = src_format._serializer
1774
if (to_serializer != src_serializer
1775
and self.target_repo._format.pack_compresses):
1776
self.target_repo.pack(hint=hint)
1779
self.target_repo.abort_write_group(suppress_errors=True)
1782
def insert_stream_without_locking(self, stream, src_format,
1784
"""Insert a stream's content into the target repository.
1786
This assumes that you already have a locked repository and an active
1789
:param src_format: a bzr repository format.
1790
:param is_resume: Passed down to get_missing_parent_inventories to
1791
indicate if we should be checking for missing texts at the same
1794
:return: A set of keys that are missing.
1796
if not self.target_repo.is_write_locked():
1797
raise errors.ObjectNotLocked(self)
1798
if not self.target_repo.is_in_write_group():
1799
raise errors.BzrError('you must already be in a write group')
1800
to_serializer = self.target_repo._format._serializer
1801
src_serializer = src_format._serializer
1803
if to_serializer == src_serializer:
1804
# If serializers match and the target is a pack repository, set the
1805
# write cache size on the new pack. This avoids poor performance
1806
# on transports where append is unbuffered (such as
1807
# RemoteTransport). This is safe to do because nothing should read
1808
# back from the target repository while a stream with matching
1809
# serialization is being inserted.
1810
# The exception is that a delta record from the source that should
1811
# be a fulltext may need to be expanded by the target (see
1812
# test_fetch_revisions_with_deltas_into_pack); but we take care to
1813
# explicitly flush any buffered writes first in that rare case.
1815
new_pack = self.target_repo._pack_collection._new_pack
1816
except AttributeError:
1817
# Not a pack repository
1820
new_pack.set_write_cache_size(1024 * 1024)
1821
for substream_type, substream in stream:
1822
if 'stream' in debug.debug_flags:
1823
mutter('inserting substream: %s', substream_type)
1824
if substream_type == 'texts':
1825
self.target_repo.texts.insert_record_stream(substream)
1826
elif substream_type == 'inventories':
1827
if src_serializer == to_serializer:
1828
self.target_repo.inventories.insert_record_stream(
1831
self._extract_and_insert_inventories(
1832
substream, src_serializer)
1833
elif substream_type == 'inventory-deltas':
1834
self._extract_and_insert_inventory_deltas(
1835
substream, src_serializer)
1836
elif substream_type == 'chk_bytes':
1837
# XXX: This doesn't support conversions, as it assumes the
1838
# conversion was done in the fetch code.
1839
self.target_repo.chk_bytes.insert_record_stream(substream)
1840
elif substream_type == 'revisions':
1841
# This may fallback to extract-and-insert more often than
1842
# required if the serializers are different only in terms of
1844
if src_serializer == to_serializer:
1845
self.target_repo.revisions.insert_record_stream(substream)
1847
self._extract_and_insert_revisions(substream,
1849
elif substream_type == 'signatures':
1850
self.target_repo.signatures.insert_record_stream(substream)
1852
raise AssertionError('kaboom! %s' % (substream_type,))
1853
# Done inserting data, and the missing_keys calculations will try to
1854
# read back from the inserted data, so flush the writes to the new pack
1855
# (if this is pack format).
1856
if new_pack is not None:
1857
new_pack._write_data(b'', flush=True)
1858
# Find all the new revisions (including ones from resume_tokens)
1859
missing_keys = self.target_repo.get_missing_parent_inventories(
1860
check_for_missing_texts=is_resume)
1862
for prefix, versioned_file in (
1863
('texts', self.target_repo.texts),
1864
('inventories', self.target_repo.inventories),
1865
('revisions', self.target_repo.revisions),
1866
('signatures', self.target_repo.signatures),
1867
('chk_bytes', self.target_repo.chk_bytes),
1869
if versioned_file is None:
1871
# TODO: key is often going to be a StaticTuple object
1872
# I don't believe we can define a method by which
1873
# (prefix,) + StaticTuple will work, though we could
1874
# define a StaticTuple.sq_concat that would allow you to
1875
# pass in either a tuple or a StaticTuple as the second
1876
# object, so instead we could have:
1877
# StaticTuple(prefix) + key here...
1878
missing_keys.update((prefix,) + key for key in
1879
versioned_file.get_missing_compression_parent_keys())
1880
except NotImplementedError:
1881
# cannot even attempt suspending, and missing would have failed
1882
# during stream insertion.
1883
missing_keys = set()
1886
def _extract_and_insert_inventory_deltas(self, substream, serializer):
1887
target_rich_root = self.target_repo._format.rich_root_data
1888
target_tree_refs = self.target_repo._format.supports_tree_reference
1889
for record in substream:
1890
# Insert the delta directly
1891
inventory_delta_bytes = record.get_bytes_as('fulltext')
1892
deserialiser = inventory_delta.InventoryDeltaDeserializer()
1894
parse_result = deserialiser.parse_text_bytes(
1895
inventory_delta_bytes)
1896
except inventory_delta.IncompatibleInventoryDelta as err:
1897
mutter("Incompatible delta: %s", err.msg)
1898
raise errors.IncompatibleRevision(self.target_repo._format)
1899
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
1900
revision_id = new_id
1901
parents = [key[0] for key in record.parents]
1902
self.target_repo.add_inventory_by_delta(
1903
basis_id, inv_delta, revision_id, parents)
1905
def _extract_and_insert_inventories(self, substream, serializer,
1907
"""Generate a new inventory versionedfile in target, converting data.
1909
The inventory is retrieved from the source, (deserializing it), and
1910
stored in the target (reserializing it in a different format).
1912
target_rich_root = self.target_repo._format.rich_root_data
1913
target_tree_refs = self.target_repo._format.supports_tree_reference
1914
for record in substream:
1915
# It's not a delta, so it must be a fulltext in the source
1916
# serializer's format.
1917
bytes = record.get_bytes_as('fulltext')
1918
revision_id = record.key[0]
1919
inv = serializer.read_inventory_from_string(bytes, revision_id)
1920
parents = [key[0] for key in record.parents]
1921
self.target_repo.add_inventory(revision_id, inv, parents)
1922
# No need to keep holding this full inv in memory when the rest of
1923
# the substream is likely to be all deltas.
1926
def _extract_and_insert_revisions(self, substream, serializer):
1927
for record in substream:
1928
bytes = record.get_bytes_as('fulltext')
1929
revision_id = record.key[0]
1930
rev = serializer.read_revision_from_string(bytes)
1931
if rev.revision_id != revision_id:
1932
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
1933
self.target_repo.add_revision(revision_id, rev)
1936
if self.target_repo._format._fetch_reconcile:
1937
self.target_repo.reconcile()
1940
class StreamSource(object):
1941
"""A source of a stream for fetching between repositories."""
1943
def __init__(self, from_repository, to_format):
1944
"""Create a StreamSource streaming from from_repository."""
1945
self.from_repository = from_repository
1946
self.to_format = to_format
1947
self._record_counter = RecordCounter()
1949
def delta_on_metadata(self):
1950
"""Return True if delta's are permitted on metadata streams.
1952
That is on revisions and signatures.
1954
src_serializer = self.from_repository._format._serializer
1955
target_serializer = self.to_format._serializer
1956
return (self.to_format._fetch_uses_deltas
1957
and src_serializer == target_serializer)
1959
def _fetch_revision_texts(self, revs):
1960
# fetch signatures first and then the revision texts
1961
# may need to be a InterRevisionStore call here.
1962
from_sf = self.from_repository.signatures
1963
# A missing signature is just skipped.
1964
keys = [(rev_id,) for rev_id in revs]
1965
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
1967
self.to_format._fetch_order,
1968
not self.to_format._fetch_uses_deltas))
1969
# If a revision has a delta, this is actually expanded inside the
1970
# insert_record_stream code now, which is an alternate fix for
1972
from_rf = self.from_repository.revisions
1973
revisions = from_rf.get_record_stream(
1975
self.to_format._fetch_order,
1976
not self.delta_on_metadata())
1977
return [('signatures', signatures), ('revisions', revisions)]
1979
def _generate_root_texts(self, revs):
1980
"""This will be called by get_stream between fetching weave texts and
1981
fetching the inventory weave.
1983
if self._rich_root_upgrade():
1984
return _mod_fetch.Inter1and2Helper(
1985
self.from_repository).generate_root_texts(revs)
1989
def get_stream(self, search):
1991
revs = search.get_keys()
1992
graph = self.from_repository.get_graph()
1993
revs = tsort.topo_sort(graph.get_parent_map(revs))
1994
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
1996
for knit_kind, file_id, revisions in data_to_fetch:
1997
if knit_kind != phase:
1999
# Make a new progress bar for this phase
2000
if knit_kind == "file":
2001
# Accumulate file texts
2002
text_keys.extend([(file_id, revision) for revision in
2004
elif knit_kind == "inventory":
2005
# Now copy the file texts.
2006
from_texts = self.from_repository.texts
2007
yield ('texts', from_texts.get_record_stream(
2008
text_keys, self.to_format._fetch_order,
2009
not self.to_format._fetch_uses_deltas))
2010
# Cause an error if a text occurs after we have done the
2013
# Before we process the inventory we generate the root
2014
# texts (if necessary) so that the inventories references
2016
for _ in self._generate_root_texts(revs):
2018
# we fetch only the referenced inventories because we do not
2019
# know for unselected inventories whether all their required
2020
# texts are present in the other repository - it could be
2022
for info in self._get_inventory_stream(revs):
2024
elif knit_kind == "signatures":
2025
# Nothing to do here; this will be taken care of when
2026
# _fetch_revision_texts happens.
2028
elif knit_kind == "revisions":
2029
for record in self._fetch_revision_texts(revs):
2032
raise AssertionError("Unknown knit kind %r" % knit_kind)
2034
def get_stream_for_missing_keys(self, missing_keys):
2035
# missing keys can only occur when we are byte copying and not
2036
# translating (because translation means we don't send
2037
# unreconstructable deltas ever).
2039
keys['texts'] = set()
2040
keys['revisions'] = set()
2041
keys['inventories'] = set()
2042
keys['chk_bytes'] = set()
2043
keys['signatures'] = set()
2044
for key in missing_keys:
2045
keys[key[0]].add(key[1:])
2046
if len(keys['revisions']):
2047
# If we allowed copying revisions at this point, we could end up
2048
# copying a revision without copying its required texts: a
2049
# violation of the requirements for repository integrity.
2050
raise AssertionError(
2051
'cannot copy revisions to fill in missing deltas %s' % (
2052
keys['revisions'],))
2053
for substream_kind, keys in viewitems(keys):
2054
vf = getattr(self.from_repository, substream_kind)
2055
if vf is None and keys:
2056
raise AssertionError(
2057
"cannot fill in keys for a versioned file we don't"
2058
" have: %s needs %s" % (substream_kind, keys))
2060
# No need to stream something we don't have
2062
if substream_kind == 'inventories':
2063
# Some missing keys are genuinely ghosts, filter those out.
2064
present = self.from_repository.inventories.get_parent_map(keys)
2065
revs = [key[0] for key in present]
2066
# Get the inventory stream more-or-less as we do for the
2067
# original stream; there's no reason to assume that records
2068
# direct from the source will be suitable for the sink. (Think
2069
# e.g. 2a -> 1.9-rich-root).
2070
for info in self._get_inventory_stream(revs, missing=True):
2074
# Ask for full texts always so that we don't need more round trips
2075
# after this stream.
2076
# Some of the missing keys are genuinely ghosts, so filter absent
2077
# records. The Sink is responsible for doing another check to
2078
# ensure that ghosts don't introduce missing data for future
2080
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
2081
self.to_format._fetch_order, True))
2082
yield substream_kind, stream
2084
def inventory_fetch_order(self):
2085
if self._rich_root_upgrade():
2086
return 'topological'
2088
return self.to_format._fetch_order
2090
def _rich_root_upgrade(self):
2091
return (not self.from_repository._format.rich_root_data
2092
and self.to_format.rich_root_data)
2094
def _get_inventory_stream(self, revision_ids, missing=False):
2095
from_format = self.from_repository._format
2096
if (from_format.supports_chks and self.to_format.supports_chks
2097
and from_format.network_name() == self.to_format.network_name()):
2098
raise AssertionError(
2099
"this case should be handled by GroupCHKStreamSource")
2100
elif 'forceinvdeltas' in debug.debug_flags:
2101
return self._get_convertable_inventory_stream(revision_ids,
2102
delta_versus_null=missing)
2103
elif from_format.network_name() == self.to_format.network_name():
2105
return self._get_simple_inventory_stream(revision_ids,
2107
elif (not from_format.supports_chks and not self.to_format.supports_chks and
2108
from_format._serializer == self.to_format._serializer):
2109
# Essentially the same format.
2110
return self._get_simple_inventory_stream(revision_ids,
2113
# Any time we switch serializations, we want to use an
2114
# inventory-delta based approach.
2115
return self._get_convertable_inventory_stream(revision_ids,
2116
delta_versus_null=missing)
2118
def _get_simple_inventory_stream(self, revision_ids, missing=False):
2119
# NB: This currently reopens the inventory weave in source;
2120
# using a single stream interface instead would avoid this.
2121
from_weave = self.from_repository.inventories
2123
delta_closure = True
2125
delta_closure = not self.delta_on_metadata()
2126
yield ('inventories', from_weave.get_record_stream(
2127
[(rev_id,) for rev_id in revision_ids],
2128
self.inventory_fetch_order(), delta_closure))
2130
def _get_convertable_inventory_stream(self, revision_ids,
2131
delta_versus_null=False):
2132
# The two formats are sufficiently different that there is no fast
2133
# path, so we need to send just inventorydeltas, which any
2134
# sufficiently modern client can insert into any repository.
2135
# The StreamSink code expects to be able to
2136
# convert on the target, so we need to put bytes-on-the-wire that can
2137
# be converted. That means inventory deltas (if the remote is <1.19,
2138
# RemoteStreamSink will fallback to VFS to insert the deltas).
2139
yield ('inventory-deltas',
2140
self._stream_invs_as_deltas(revision_ids,
2141
delta_versus_null=delta_versus_null))
2143
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
2144
"""Return a stream of inventory-deltas for the given rev ids.
2146
:param revision_ids: The list of inventories to transmit
2147
:param delta_versus_null: Don't try to find a minimal delta for this
2148
entry, instead compute the delta versus the NULL_REVISION. This
2149
effectively streams a complete inventory. Used for stuff like
2150
filling in missing parents, etc.
2152
from_repo = self.from_repository
2153
revision_keys = [(rev_id,) for rev_id in revision_ids]
2154
parent_map = from_repo.inventories.get_parent_map(revision_keys)
2155
# XXX: possibly repos could implement a more efficient iter_inv_deltas
2157
inventories = self.from_repository.iter_inventories(
2158
revision_ids, 'topological')
2159
format = from_repo._format
2160
invs_sent_so_far = {_mod_revision.NULL_REVISION}
2161
inventory_cache = lru_cache.LRUCache(50)
2162
null_inventory = from_repo.revision_tree(
2163
_mod_revision.NULL_REVISION).root_inventory
2164
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
2165
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
2166
# repo back into a non-rich-root repo ought to be allowed)
2167
serializer = inventory_delta.InventoryDeltaSerializer(
2168
versioned_root=format.rich_root_data,
2169
tree_references=format.supports_tree_reference)
2170
for inv in inventories:
2171
key = (inv.revision_id,)
2172
parent_keys = parent_map.get(key, ())
2174
if not delta_versus_null and parent_keys:
2175
# The caller did not ask for complete inventories and we have
2176
# some parents that we can delta against. Make a delta against
2177
# each parent so that we can find the smallest.
2178
parent_ids = [parent_key[0] for parent_key in parent_keys]
2179
for parent_id in parent_ids:
2180
if parent_id not in invs_sent_so_far:
2181
# We don't know that the remote side has this basis, so
2184
if parent_id == _mod_revision.NULL_REVISION:
2185
parent_inv = null_inventory
2187
parent_inv = inventory_cache.get(parent_id, None)
2188
if parent_inv is None:
2189
parent_inv = from_repo.get_inventory(parent_id)
2190
candidate_delta = inv._make_delta(parent_inv)
2192
or len(delta) > len(candidate_delta)):
2193
delta = candidate_delta
2194
basis_id = parent_id
2196
# Either none of the parents ended up being suitable, or we
2197
# were asked to delta against NULL
2198
basis_id = _mod_revision.NULL_REVISION
2199
delta = inv._make_delta(null_inventory)
2200
invs_sent_so_far.add(inv.revision_id)
2201
inventory_cache[inv.revision_id] = inv
2202
delta_serialized = b''.join(
2203
serializer.delta_to_lines(basis_id, key[-1], delta))
2204
yield versionedfile.FulltextContentFactory(
2205
key, parent_keys, None, delta_serialized)
2208
class _VersionedFileChecker(object):
2210
def __init__(self, repository, text_key_references=None, ancestors=None):
2211
self.repository = repository
2212
self.text_index = self.repository._generate_text_key_index(
2213
text_key_references=text_key_references, ancestors=ancestors)
2215
def calculate_file_version_parents(self, text_key):
2216
"""Calculate the correct parents for a file version according to
2219
parent_keys = self.text_index[text_key]
2220
if parent_keys == [_mod_revision.NULL_REVISION]:
2222
return tuple(parent_keys)
2224
def check_file_version_parents(self, texts, progress_bar=None):
2225
"""Check the parents stored in a versioned file are correct.
2227
It also detects file versions that are not referenced by their
2228
corresponding revision's inventory.
2230
:returns: A tuple of (wrong_parents, dangling_file_versions).
2231
wrong_parents is a dict mapping {revision_id: (stored_parents,
2232
correct_parents)} for each revision_id where the stored parents
2233
are not correct. dangling_file_versions is a set of (file_id,
2234
revision_id) tuples for versions that are present in this versioned
2235
file, but not used by the corresponding inventory.
2237
local_progress = None
2238
if progress_bar is None:
2239
local_progress = ui.ui_factory.nested_progress_bar()
2240
progress_bar = local_progress
2242
return self._check_file_version_parents(texts, progress_bar)
2245
local_progress.finished()
2247
def _check_file_version_parents(self, texts, progress_bar):
2248
"""See check_file_version_parents."""
2250
self.file_ids = {file_id for file_id, _ in self.text_index}
2251
# text keys is now grouped by file_id
2252
n_versions = len(self.text_index)
2253
progress_bar.update(gettext('loading text store'), 0, n_versions)
2254
parent_map = self.repository.texts.get_parent_map(self.text_index)
2255
# On unlistable transports this could well be empty/error...
2256
text_keys = self.repository.texts.keys()
2257
unused_keys = frozenset(text_keys) - set(self.text_index)
2258
for num, key in enumerate(self.text_index):
2259
progress_bar.update(
2260
gettext('checking text graph'), num, n_versions)
2261
correct_parents = self.calculate_file_version_parents(key)
2263
knit_parents = parent_map[key]
2264
except errors.RevisionNotPresent:
2267
if correct_parents != knit_parents:
2268
wrong_parents[key] = (knit_parents, correct_parents)
2269
return wrong_parents, unused_keys
2272
class InterVersionedFileRepository(InterRepository):
2274
_walk_to_common_revisions_batch_size = 50
2276
supports_fetch_spec = True
2278
def fetch(self, revision_id=None, find_ghosts=False,
2280
"""Fetch the content required to construct revision_id.
2282
The content is copied from self.source to self.target.
2284
:param revision_id: if None all content is copied, if NULL_REVISION no
2288
if self.target._format.experimental:
2289
ui.ui_factory.show_user_warning(
2290
'experimental_format_fetch',
2291
from_format=self.source._format,
2292
to_format=self.target._format)
2293
from breezy.bzr.fetch import RepoFetcher
2294
# See <https://launchpad.net/bugs/456077> asking for a warning here
2295
if self.source._format.network_name() != self.target._format.network_name():
2296
ui.ui_factory.show_user_warning(
2297
'cross_format_fetch', from_format=self.source._format,
2298
to_format=self.target._format)
2299
with self.lock_write():
2300
f = RepoFetcher(to_repository=self.target,
2301
from_repository=self.source,
2302
last_revision=revision_id,
2303
fetch_spec=fetch_spec,
2304
find_ghosts=find_ghosts)
2306
def _walk_to_common_revisions(self, revision_ids, if_present_ids=None):
2307
"""Walk out from revision_ids in source to revisions target has.
2309
:param revision_ids: The start point for the search.
2310
:return: A set of revision ids.
2312
target_graph = self.target.get_graph()
2313
revision_ids = frozenset(revision_ids)
2315
all_wanted_revs = revision_ids.union(if_present_ids)
2317
all_wanted_revs = revision_ids
2318
missing_revs = set()
2319
source_graph = self.source.get_graph()
2320
# ensure we don't pay silly lookup costs.
2321
searcher = source_graph._make_breadth_first_searcher(all_wanted_revs)
2322
null_set = frozenset([_mod_revision.NULL_REVISION])
2323
searcher_exhausted = False
2327
# Iterate the searcher until we have enough next_revs
2328
while len(next_revs) < self._walk_to_common_revisions_batch_size:
2330
next_revs_part, ghosts_part = searcher.next_with_ghosts()
2331
next_revs.update(next_revs_part)
2332
ghosts.update(ghosts_part)
2333
except StopIteration:
2334
searcher_exhausted = True
2336
# If there are ghosts in the source graph, and the caller asked for
2337
# them, make sure that they are present in the target.
2338
# We don't care about other ghosts as we can't fetch them and
2339
# haven't been asked to.
2340
ghosts_to_check = set(revision_ids.intersection(ghosts))
2341
revs_to_get = set(next_revs).union(ghosts_to_check)
2343
have_revs = set(target_graph.get_parent_map(revs_to_get))
2344
# we always have NULL_REVISION present.
2345
have_revs = have_revs.union(null_set)
2346
# Check if the target is missing any ghosts we need.
2347
ghosts_to_check.difference_update(have_revs)
2349
# One of the caller's revision_ids is a ghost in both the
2350
# source and the target.
2351
raise errors.NoSuchRevision(
2352
self.source, ghosts_to_check.pop())
2353
missing_revs.update(next_revs - have_revs)
2354
# Because we may have walked past the original stop point, make
2355
# sure everything is stopped
2356
stop_revs = searcher.find_seen_ancestors(have_revs)
2357
searcher.stop_searching_any(stop_revs)
2358
if searcher_exhausted:
2360
(started_keys, excludes, included_keys) = searcher.get_state()
2361
return vf_search.SearchResult(started_keys, excludes,
2362
len(included_keys), included_keys)
2364
def search_missing_revision_ids(self,
2365
find_ghosts=True, revision_ids=None, if_present_ids=None,
2367
"""Return the revision ids that source has that target does not.
2369
:param revision_ids: return revision ids included by these
2370
revision_ids. NoSuchRevision will be raised if any of these
2371
revisions are not present.
2372
:param if_present_ids: like revision_ids, but will not cause
2373
NoSuchRevision if any of these are absent, instead they will simply
2374
not be in the result. This is useful for e.g. finding revisions
2375
to fetch for tags, which may reference absent revisions.
2376
:param find_ghosts: If True find missing revisions in deep history
2377
rather than just finding the surface difference.
2378
:return: A breezy.graph.SearchResult.
2380
with self.lock_read():
2381
# stop searching at found target revisions.
2382
if not find_ghosts and (revision_ids is not None or if_present_ids is
2384
result = self._walk_to_common_revisions(revision_ids,
2385
if_present_ids=if_present_ids)
2388
result_set = result.get_keys()
2390
# generic, possibly worst case, slow code path.
2391
target_ids = set(self.target.all_revision_ids())
2392
source_ids = self._present_source_revisions_for(
2393
revision_ids, if_present_ids)
2394
result_set = set(source_ids).difference(target_ids)
2395
if limit is not None:
2396
topo_ordered = self.source.get_graph().iter_topo_order(result_set)
2397
result_set = set(itertools.islice(topo_ordered, limit))
2398
return self.source.revision_ids_to_search_result(result_set)
2400
def _present_source_revisions_for(self, revision_ids, if_present_ids=None):
2401
"""Returns set of all revisions in ancestry of revision_ids present in
2404
:param revision_ids: if None, all revisions in source are returned.
2405
:param if_present_ids: like revision_ids, but if any/all of these are
2406
absent no error is raised.
2408
if revision_ids is not None or if_present_ids is not None:
2409
# First, ensure all specified revisions exist. Callers expect
2410
# NoSuchRevision when they pass absent revision_ids here.
2411
if revision_ids is None:
2412
revision_ids = set()
2413
if if_present_ids is None:
2414
if_present_ids = set()
2415
revision_ids = set(revision_ids)
2416
if_present_ids = set(if_present_ids)
2417
all_wanted_ids = revision_ids.union(if_present_ids)
2418
graph = self.source.get_graph()
2419
present_revs = set(graph.get_parent_map(all_wanted_ids))
2420
missing = revision_ids.difference(present_revs)
2422
raise errors.NoSuchRevision(self.source, missing.pop())
2423
found_ids = all_wanted_ids.intersection(present_revs)
2424
source_ids = [rev_id for (rev_id, parents) in
2425
graph.iter_ancestry(found_ids)
2426
if rev_id != _mod_revision.NULL_REVISION and
2427
parents is not None]
2429
source_ids = self.source.all_revision_ids()
2430
return set(source_ids)
2433
def _get_repo_format_to_test(self):
2437
def is_compatible(cls, source, target):
2438
# The default implementation is compatible with everything
2439
return (source._format.supports_full_versioned_files
2440
and target._format.supports_full_versioned_files)
2443
class InterDifferingSerializer(InterVersionedFileRepository):
2446
def _get_repo_format_to_test(self):
2450
def is_compatible(source, target):
2451
if not source._format.supports_full_versioned_files:
2453
if not target._format.supports_full_versioned_files:
2455
# This is redundant with format.check_conversion_target(), however that
2456
# raises an exception, and we just want to say "False" as in we won't
2457
# support converting between these formats.
2458
if 'IDS_never' in debug.debug_flags:
2460
if source.supports_rich_root() and not target.supports_rich_root():
2462
if (source._format.supports_tree_reference and
2463
not target._format.supports_tree_reference):
2465
if target._fallback_repositories and target._format.supports_chks:
2466
# IDS doesn't know how to copy CHKs for the parent inventories it
2467
# adds to stacked repos.
2469
if 'IDS_always' in debug.debug_flags:
2471
# Only use this code path for local source and target. IDS does far
2472
# too much IO (both bandwidth and roundtrips) over a network.
2473
if not source.controldir.transport.base.startswith('file:///'):
2475
if not target.controldir.transport.base.startswith('file:///'):
2479
def _get_trees(self, revision_ids, cache):
2481
for rev_id in revision_ids:
2483
possible_trees.append((rev_id, cache[rev_id]))
2485
# Not cached, but inventory might be present anyway.
2487
tree = self.source.revision_tree(rev_id)
2488
except errors.NoSuchRevision:
2489
# Nope, parent is ghost.
2492
cache[rev_id] = tree
2493
possible_trees.append((rev_id, tree))
2494
return possible_trees
2496
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
2497
"""Get the best delta and base for this revision.
2499
:return: (basis_id, delta)
2502
# Generate deltas against each tree, to find the shortest.
2503
# FIXME: Support nested trees
2504
texts_possibly_new_in_tree = set()
2505
for basis_id, basis_tree in possible_trees:
2506
delta = tree.root_inventory._make_delta(basis_tree.root_inventory)
2507
for old_path, new_path, file_id, new_entry in delta:
2508
if new_path is None:
2509
# This file_id isn't present in the new rev, so we don't
2513
# Rich roots are handled elsewhere...
2515
kind = new_entry.kind
2516
if kind != 'directory' and kind != 'file':
2517
# No text record associated with this inventory entry.
2519
# This is a directory or file that has changed somehow.
2520
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
2521
deltas.append((len(delta), basis_id, delta))
2523
return deltas[0][1:]
2525
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
2526
"""Find all parent revisions that are absent, but for which the
2527
inventory is present, and copy those inventories.
2529
This is necessary to preserve correctness when the source is stacked
2530
without fallbacks configured. (Note that in cases like upgrade the
2531
source may be not have _fallback_repositories even though it is
2534
parent_revs = set(itertools.chain.from_iterable(viewvalues(
2536
present_parents = self.source.get_parent_map(parent_revs)
2537
absent_parents = parent_revs.difference(present_parents)
2538
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
2539
(rev_id,) for rev_id in absent_parents)
2540
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
2541
for parent_tree in self.source.revision_trees(parent_inv_ids):
2542
current_revision_id = parent_tree.get_revision_id()
2543
parents_parents_keys = parent_invs_keys_for_stacking[
2544
(current_revision_id,)]
2545
parents_parents = [key[-1] for key in parents_parents_keys]
2546
basis_id = _mod_revision.NULL_REVISION
2547
basis_tree = self.source.revision_tree(basis_id)
2548
delta = parent_tree.root_inventory._make_delta(
2549
basis_tree.root_inventory)
2550
self.target.add_inventory_by_delta(
2551
basis_id, delta, current_revision_id, parents_parents)
2552
cache[current_revision_id] = parent_tree
2554
def _fetch_batch(self, revision_ids, basis_id, cache):
2555
"""Fetch across a few revisions.
2557
:param revision_ids: The revisions to copy
2558
:param basis_id: The revision_id of a tree that must be in cache, used
2559
as a basis for delta when no other base is available
2560
:param cache: A cache of RevisionTrees that we can use.
2561
:return: The revision_id of the last converted tree. The RevisionTree
2562
for it will be in cache
2564
# Walk though all revisions; get inventory deltas, copy referenced
2565
# texts that delta references, insert the delta, revision and
2567
root_keys_to_create = set()
2570
pending_revisions = []
2571
parent_map = self.source.get_parent_map(revision_ids)
2572
self._fetch_parent_invs_for_stacking(parent_map, cache)
2573
self.source._safe_to_return_from_cache = True
2574
for tree in self.source.revision_trees(revision_ids):
2575
# Find a inventory delta for this revision.
2576
# Find text entries that need to be copied, too.
2577
current_revision_id = tree.get_revision_id()
2578
parent_ids = parent_map.get(current_revision_id, ())
2579
parent_trees = self._get_trees(parent_ids, cache)
2580
possible_trees = list(parent_trees)
2581
if len(possible_trees) == 0:
2582
# There either aren't any parents, or the parents are ghosts,
2583
# so just use the last converted tree.
2584
possible_trees.append((basis_id, cache[basis_id]))
2585
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
2587
revision = self.source.get_revision(current_revision_id)
2588
pending_deltas.append((basis_id, delta,
2589
current_revision_id, revision.parent_ids))
2590
if self._converting_to_rich_root:
2591
self._revision_id_to_root_id[current_revision_id] = \
2593
# Determine which texts are in present in this revision but not in
2594
# any of the available parents.
2595
texts_possibly_new_in_tree = set()
2596
for old_path, new_path, file_id, entry in delta:
2597
if new_path is None:
2598
# This file_id isn't present in the new rev
2602
if not self.target.supports_rich_root():
2603
# The target doesn't support rich root, so we don't
2606
if self._converting_to_rich_root:
2607
# This can't be copied normally, we have to insert
2609
root_keys_to_create.add((file_id, entry.revision))
2612
texts_possibly_new_in_tree.add((file_id, entry.revision))
2613
for basis_id, basis_tree in possible_trees:
2614
basis_inv = basis_tree.root_inventory
2615
for file_key in list(texts_possibly_new_in_tree):
2616
file_id, file_revision = file_key
2618
entry = basis_inv.get_entry(file_id)
2619
except errors.NoSuchId:
2621
if entry.revision == file_revision:
2622
texts_possibly_new_in_tree.remove(file_key)
2623
text_keys.update(texts_possibly_new_in_tree)
2624
pending_revisions.append(revision)
2625
cache[current_revision_id] = tree
2626
basis_id = current_revision_id
2627
self.source._safe_to_return_from_cache = False
2629
from_texts = self.source.texts
2630
to_texts = self.target.texts
2631
if root_keys_to_create:
2632
root_stream = _mod_fetch._new_root_data_stream(
2633
root_keys_to_create, self._revision_id_to_root_id, parent_map,
2635
to_texts.insert_record_stream(root_stream)
2636
to_texts.insert_record_stream(from_texts.get_record_stream(
2637
text_keys, self.target._format._fetch_order,
2638
not self.target._format._fetch_uses_deltas))
2639
# insert inventory deltas
2640
for delta in pending_deltas:
2641
self.target.add_inventory_by_delta(*delta)
2642
if self.target._fallback_repositories:
2643
# Make sure this stacked repository has all the parent inventories
2644
# for the new revisions that we are about to insert. We do this
2645
# before adding the revisions so that no revision is added until
2646
# all the inventories it may depend on are added.
2647
# Note that this is overzealous, as we may have fetched these in an
2650
revision_ids = set()
2651
for revision in pending_revisions:
2652
revision_ids.add(revision.revision_id)
2653
parent_ids.update(revision.parent_ids)
2654
parent_ids.difference_update(revision_ids)
2655
parent_ids.discard(_mod_revision.NULL_REVISION)
2656
parent_map = self.source.get_parent_map(parent_ids)
2657
# we iterate over parent_map and not parent_ids because we don't
2658
# want to try copying any revision which is a ghost
2659
for parent_tree in self.source.revision_trees(parent_map):
2660
current_revision_id = parent_tree.get_revision_id()
2661
parents_parents = parent_map[current_revision_id]
2662
possible_trees = self._get_trees(parents_parents, cache)
2663
if len(possible_trees) == 0:
2664
# There either aren't any parents, or the parents are
2665
# ghosts, so just use the last converted tree.
2666
possible_trees.append((basis_id, cache[basis_id]))
2667
basis_id, delta = self._get_delta_for_revision(parent_tree,
2668
parents_parents, possible_trees)
2669
self.target.add_inventory_by_delta(
2670
basis_id, delta, current_revision_id, parents_parents)
2671
# insert signatures and revisions
2672
for revision in pending_revisions:
2674
signature = self.source.get_signature_text(
2675
revision.revision_id)
2676
self.target.add_signature_text(revision.revision_id,
2678
except errors.NoSuchRevision:
2680
self.target.add_revision(revision.revision_id, revision)
2683
def _fetch_all_revisions(self, revision_ids, pb):
2684
"""Fetch everything for the list of revisions.
2686
:param revision_ids: The list of revisions to fetch. Must be in
2688
:param pb: A ProgressTask
2691
basis_id, basis_tree = self._get_basis(revision_ids[0])
2693
cache = lru_cache.LRUCache(100)
2694
cache[basis_id] = basis_tree
2695
del basis_tree # We don't want to hang on to it here
2699
for offset in range(0, len(revision_ids), batch_size):
2700
self.target.start_write_group()
2702
pb.update(gettext('Transferring revisions'), offset,
2704
batch = revision_ids[offset:offset + batch_size]
2705
basis_id = self._fetch_batch(batch, basis_id, cache)
2707
self.source._safe_to_return_from_cache = False
2708
self.target.abort_write_group()
2711
hint = self.target.commit_write_group()
2714
if hints and self.target._format.pack_compresses:
2715
self.target.pack(hint=hints)
2716
pb.update(gettext('Transferring revisions'), len(revision_ids),
2719
def fetch(self, revision_id=None, find_ghosts=False,
2721
"""See InterRepository.fetch()."""
2722
if fetch_spec is not None:
2723
revision_ids = fetch_spec.get_keys()
2726
if self.source._format.experimental:
2727
ui.ui_factory.show_user_warning('experimental_format_fetch',
2728
from_format=self.source._format,
2729
to_format=self.target._format)
2730
if (not self.source.supports_rich_root() and
2731
self.target.supports_rich_root()):
2732
self._converting_to_rich_root = True
2733
self._revision_id_to_root_id = {}
2735
self._converting_to_rich_root = False
2736
# See <https://launchpad.net/bugs/456077> asking for a warning here
2737
if self.source._format.network_name() != self.target._format.network_name():
2738
ui.ui_factory.show_user_warning('cross_format_fetch',
2739
from_format=self.source._format,
2740
to_format=self.target._format)
2741
with self.lock_write():
2742
if revision_ids is None:
2744
search_revision_ids = [revision_id]
2746
search_revision_ids = None
2747
revision_ids = self.target.search_missing_revision_ids(self.source,
2748
revision_ids=search_revision_ids,
2749
find_ghosts=find_ghosts).get_keys()
2750
if not revision_ids:
2752
revision_ids = tsort.topo_sort(
2753
self.source.get_graph().get_parent_map(revision_ids))
2754
if not revision_ids:
2756
# Walk though all revisions; get inventory deltas, copy referenced
2757
# texts that delta references, insert the delta, revision and
2759
with ui.ui_factory.nested_progress_bar() as pb:
2760
self._fetch_all_revisions(revision_ids, pb)
2761
return len(revision_ids), 0
2763
def _get_basis(self, first_revision_id):
2764
"""Get a revision and tree which exists in the target.
2766
This assumes that first_revision_id is selected for transmission
2767
because all other ancestors are already present. If we can't find an
2768
ancestor we fall back to NULL_REVISION since we know that is safe.
2770
:return: (basis_id, basis_tree)
2772
first_rev = self.source.get_revision(first_revision_id)
2774
basis_id = first_rev.parent_ids[0]
2775
# only valid as a basis if the target has it
2776
self.target.get_revision(basis_id)
2777
# Try to get a basis tree - if it's a ghost it will hit the
2778
# NoSuchRevision case.
2779
basis_tree = self.source.revision_tree(basis_id)
2780
except (IndexError, errors.NoSuchRevision):
2781
basis_id = _mod_revision.NULL_REVISION
2782
basis_tree = self.source.revision_tree(basis_id)
2783
return basis_id, basis_tree
2786
class InterSameDataRepository(InterVersionedFileRepository):
2787
"""Code for converting between repositories that represent the same data.
2789
Data format and model must match for this to work.
2793
def _get_repo_format_to_test(self):
2794
"""Repository format for testing with.
2796
InterSameData can pull from subtree to subtree and from non-subtree to
2797
non-subtree, so we test this with the richest repository format.
2799
from breezy.bzr import knitrepo
2800
return knitrepo.RepositoryFormatKnit3()
2803
def is_compatible(source, target):
2805
InterRepository._same_model(source, target)
2806
and source._format.supports_full_versioned_files
2807
and target._format.supports_full_versioned_files)
2810
InterRepository.register_optimiser(InterVersionedFileRepository)
2811
InterRepository.register_optimiser(InterDifferingSerializer)
2812
InterRepository.register_optimiser(InterSameDataRepository)
2815
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2816
"""Install all revision data into a repository.
2818
Accepts an iterable of revision, tree, signature tuples. The signature
2821
repository.start_write_group()
2823
inventory_cache = lru_cache.LRUCache(10)
2824
for n, (revision, revision_tree, signature) in enumerate(iterable):
2825
_install_revision(repository, revision, revision_tree, signature,
2828
pb.update(gettext('Transferring revisions'),
2829
n + 1, num_revisions)
2831
repository.abort_write_group()
2834
repository.commit_write_group()
2837
def _install_revision(repository, rev, revision_tree, signature,
2839
"""Install all revision data into a repository."""
2840
present_parents = []
2842
for p_id in rev.parent_ids:
2843
if repository.has_revision(p_id):
2844
present_parents.append(p_id)
2845
parent_trees[p_id] = repository.revision_tree(p_id)
2847
parent_trees[p_id] = repository.revision_tree(
2848
_mod_revision.NULL_REVISION)
2850
# FIXME: Support nested trees
2851
inv = revision_tree.root_inventory
2852
entries = inv.iter_entries()
2853
# backwards compatibility hack: skip the root id.
2854
if not repository.supports_rich_root():
2855
path, root = next(entries)
2856
if root.revision != rev.revision_id:
2857
raise errors.IncompatibleRevision(repr(repository))
2859
for path, ie in entries:
2860
text_keys[(ie.file_id, ie.revision)] = ie
2861
text_parent_map = repository.texts.get_parent_map(text_keys)
2862
missing_texts = set(text_keys) - set(text_parent_map)
2863
# Add the texts that are not already present
2864
for text_key in missing_texts:
2865
ie = text_keys[text_key]
2867
# FIXME: TODO: The following loop overlaps/duplicates that done by
2868
# commit to determine parents. There is a latent/real bug here where
2869
# the parents inserted are not those commit would do - in particular
2870
# they are not filtered by heads(). RBC, AB
2871
for revision, tree in viewitems(parent_trees):
2872
if not tree.has_id(ie.file_id):
2874
path = tree.id2path(ie.file_id)
2875
parent_id = tree.get_file_revision(path)
2876
if parent_id in text_parents:
2878
text_parents.append((ie.file_id, parent_id))
2879
revision_tree_path = revision_tree.id2path(ie.file_id)
2880
with revision_tree.get_file(revision_tree_path) as f:
2881
lines = f.readlines()
2882
repository.texts.add_lines(text_key, text_parents, lines)
2884
# install the inventory
2885
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2886
# Cache this inventory
2887
inventory_cache[rev.revision_id] = inv
2889
basis_inv = inventory_cache[rev.parent_ids[0]]
2891
repository.add_inventory(rev.revision_id, inv, present_parents)
2893
delta = inv._make_delta(basis_inv)
2894
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2895
rev.revision_id, present_parents)
2897
repository.add_inventory(rev.revision_id, inv, present_parents)
2898
except errors.RevisionAlreadyPresent:
2900
if signature is not None:
2901
repository.add_signature_text(rev.revision_id, signature)
2902
repository.add_revision(rev.revision_id, rev, inv)
2905
def install_revision(repository, rev, revision_tree):
2906
"""Install all revision data into a repository."""
2907
install_revisions(repository, [(rev, revision_tree, None)])
added
removed
breezy/bzr/vf_repository.py
