« back to all changes in this revision
Viewing changes to bzrlib/repository.py
- browse files at revision 2382.2.5
- compare with another revision
- download diff
- download tarball
- view history from revision 2382.2.5
- Committer: Martin Pool
- Date: 2007-04-04 06:17:31 UTC
- mto: This revision was merged to the branch mainline in revision 2397.
- Revision ID: mbp@sourcefrog.net-20070404061731-tt2xrzllqhbodn83
Contents of TODO file moved into bug tracker
- files added:
- build-api
- bzrlib/bundle/old
- bzrlib/tests/interversionedfile_implementations
- bzrlib/tests/revisionstore_implementations
- bzrlib/util/configobj/docs
- files removed:
- bzrlib/_btree_serializer_c.pyx
- bzrlib/filters
- bzrlib/help_topics
- bzrlib/help_topics/en
- bzrlib/plugins/netrc_credential_store
- bzrlib/plugins/netrc_credential_store/tests
- bzrlib/smart
- bzrlib/tests/commands
- bzrlib/tests/ftp_server
- bzrlib/tests/inventory_implementations
- bzrlib/tests/per_interbranch
- bzrlib/tests/per_repository_chk
- bzrlib/tests/per_repository_reference
- bzrlib/tests/ssl_certs
- bzrlib/transport/ftp
- bzrlib/util/tests
- doc/developers
- doc/en
- doc/en/admin-guide
- doc/en/developer-guide
- doc/en/mini-tutorial
- doc/en/quick-reference
- doc/en/release-notes
- doc/en/tutorials
- doc/en/user-guide
- doc/en/user-guide/images
- doc/en/user-reference
- doc/es
- doc/es/guia-desarrollador
- doc/es/guia-usuario
- doc/es/mini-tutorial
- doc/es/notas-version
- doc/es/referencia
- doc/es/referencia-rapida
- man1
- tools/packaging
- files renamed:
- doc/developers/HACKING.txt => HACKING
- bzrlib/deprecated_graph.py => bzrlib/graph.py
- bzrlib/help_topics/__init__.py => bzrlib/help_topics.py
- bzrlib/_patiencediff_py.py => bzrlib/patiencediff.py
- bzrlib/plugins/launchpad/lp_directory.py => bzrlib/plugins/launchpad/lp_indirect.py
- bzrlib/plugins/launchpad/test_lp_directory.py => bzrlib/plugins/launchpad/test_lp_indirect.py
- bzrlib/tests/http_utils.py => bzrlib/tests/HTTPTestUtil.py
- bzrlib/tests/http_server.py => bzrlib/tests/HttpServer.py
- bzrlib/tests/blackbox/test_send.py => bzrlib/tests/blackbox/test_bundle.py
- bzrlib/tests/per_repository/ => bzrlib/tests/repository_implementations/
- bzrlib/tests/test_deprecated_graph.py => bzrlib/tests/test_graph.py
- bzrlib/tests/test_revisionspec.py => bzrlib/tests/test_revisionnamespaces.py
- bzrlib/transport/ftp/__init__.py => bzrlib/transport/ftp.py
- bzrlib/transport/remote.py => bzrlib/transport/smart.py
- bzrlib/xml8.py => bzrlib/xml5.py
- doc/en/tutorials/centralized_workflow.txt => doc/centralized_workflow.txt
- bzrlib/help_topics/en/configuration.txt => doc/configuration.txt
- doc/en/user-guide/http_smart_server.txt => doc/http_smart_server.txt
- doc/en/user-guide/index.txt => doc/index.txt
- doc/en/user-guide/plugins.txt => doc/plugins.txt
- doc/en/user-guide/server.txt => doc/server.txt
- doc/en/user-guide/setting_up_email.txt => doc/setting_up_email.txt
- doc/en/user-guide/specifying_revisions.txt => doc/specifying_revisions.txt
- doc/en/tutorials/tutorial.txt => doc/tutorial.txt
- doc/en/user-guide/using_aliases.txt => doc/using_aliases.txt
- doc/en/user-guide/version_info.txt => doc/version_info.txt
- files modified:
- .bzrignore
- setup.py *
added
removed
bzrlib/repository.py
1
# Copyright (C) 2005, 2006, 2007, 2008, 2009 Canonical Ltd
1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
2
2
#
3
3
# This program is free software; you can redistribute it and/or modify
4
4
# it under the terms of the GNU General Public License as published by
12
12
#
13
13
# You should have received a copy of the GNU General Public License
14
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
16
17
from cStringIO import StringIO
16
18
17
19
from bzrlib.lazy_import import lazy_import
18
20
lazy_import(globals(), """
19
import cStringIO
20
21
import re
21
22
import time
23
import unittest
22
24
23
25
from bzrlib import (
24
26
bzrdir,
25
27
check,
26
chk_map,
27
debug,
28
28
errors,
29
fifo_cache,
30
29
generate_ids,
31
30
gpg,
32
31
graph,
33
inventory,
34
32
lazy_regex,
35
33
lockable_files,
36
34
lockdir,
37
lru_cache,
38
35
osutils,
36
registry,
39
37
revision as _mod_revision,
40
38
symbol_versioning,
41
tsort,
39
transactions,
42
40
ui,
43
versionedfile,
44
41
)
45
from bzrlib.bundle import serializer
46
42
from bzrlib.revisiontree import RevisionTree
47
43
from bzrlib.store.versioned import VersionedFileStore
44
from bzrlib.store.text import TextStore
48
45
from bzrlib.testament import Testament
46
49
47
""")
50
48
51
49
from bzrlib.decorators import needs_read_lock, needs_write_lock
52
50
from bzrlib.inter import InterObject
53
from bzrlib.inventory import (
54
Inventory,
55
InventoryDirectory,
56
ROOT_ID,
57
entry_factory,
58
)
59
from bzrlib import registry
51
from bzrlib.inventory import Inventory, InventoryDirectory, ROOT_ID
60
52
from bzrlib.symbol_versioning import (
61
53
deprecated_method,
54
zero_nine,
62
55
)
63
from bzrlib.trace import (
64
log_exception_quietly, note, mutter, mutter_callsite, warning)
56
from bzrlib.trace import mutter, note, warning
65
57
66
58
67
59
# Old formats display a warning, but only once
68
60
_deprecation_warning_done = False
69
61
70
62
71
class CommitBuilder(object):
72
"""Provides an interface to build up a commit.
73
74
This allows describing a tree to be committed without needing to
75
know the internals of the format of the repository.
76
"""
77
78
# all clients should supply tree roots.
79
record_root_entry = True
80
# the default CommitBuilder does not manage trees whose root is versioned.
81
_versioned_root = False
82
83
def __init__(self, repository, parents, config, timestamp=None,
84
timezone=None, committer=None, revprops=None,
85
revision_id=None):
86
"""Initiate a CommitBuilder.
87
88
:param repository: Repository to commit to.
89
:param parents: Revision ids of the parents of the new revision.
90
:param config: Configuration to use.
91
:param timestamp: Optional timestamp recorded for commit.
92
:param timezone: Optional timezone for timestamp.
93
:param committer: Optional committer to set for commit.
94
:param revprops: Optional dictionary of revision properties.
95
:param revision_id: Optional revision id.
96
"""
97
self._config = config
98
99
if committer is None:
100
self._committer = self._config.username()
101
else:
102
self._committer = committer
103
104
self.new_inventory = Inventory(None)
105
self._new_revision_id = revision_id
106
self.parents = parents
107
self.repository = repository
108
109
self._revprops = {}
110
if revprops is not None:
111
self._validate_revprops(revprops)
112
self._revprops.update(revprops)
113
114
if timestamp is None:
115
timestamp = time.time()
116
# Restrict resolution to 1ms
117
self._timestamp = round(timestamp, 3)
118
119
if timezone is None:
120
self._timezone = osutils.local_time_offset()
121
else:
122
self._timezone = int(timezone)
123
124
self._generate_revision_if_needed()
125
self.__heads = graph.HeadsCache(repository.get_graph()).heads
126
self._basis_delta = []
127
# API compatibility, older code that used CommitBuilder did not call
128
# .record_delete(), which means the delta that is computed would not be
129
# valid. Callers that will call record_delete() should call
130
# .will_record_deletes() to indicate that.
131
self._recording_deletes = False
132
# memo'd check for no-op commits.
133
self._any_changes = False
134
135
def any_changes(self):
136
"""Return True if any entries were changed.
137
138
This includes merge-only changes. It is the core for the --unchanged
139
detection in commit.
140
141
:return: True if any changes have occured.
142
"""
143
return self._any_changes
144
145
def _validate_unicode_text(self, text, context):
146
"""Verify things like commit messages don't have bogus characters."""
147
if '\r' in text:
148
raise ValueError('Invalid value for %s: %r' % (context, text))
149
150
def _validate_revprops(self, revprops):
151
for key, value in revprops.iteritems():
152
# We know that the XML serializers do not round trip '\r'
153
# correctly, so refuse to accept them
154
if not isinstance(value, basestring):
155
raise ValueError('revision property (%s) is not a valid'
156
' (unicode) string: %r' % (key, value))
157
self._validate_unicode_text(value,
158
'revision property (%s)' % (key,))
159
160
def commit(self, message):
161
"""Make the actual commit.
162
163
:return: The revision id of the recorded revision.
164
"""
165
self._validate_unicode_text(message, 'commit message')
166
rev = _mod_revision.Revision(
167
timestamp=self._timestamp,
168
timezone=self._timezone,
169
committer=self._committer,
170
message=message,
171
inventory_sha1=self.inv_sha1,
172
revision_id=self._new_revision_id,
173
properties=self._revprops)
174
rev.parent_ids = self.parents
175
self.repository.add_revision(self._new_revision_id, rev,
176
self.new_inventory, self._config)
177
self.repository.commit_write_group()
178
return self._new_revision_id
179
180
def abort(self):
181
"""Abort the commit that is being built.
182
"""
183
self.repository.abort_write_group()
184
185
def revision_tree(self):
186
"""Return the tree that was just committed.
187
188
After calling commit() this can be called to get a RevisionTree
189
representing the newly committed tree. This is preferred to
190
calling Repository.revision_tree() because that may require
191
deserializing the inventory, while we already have a copy in
192
memory.
193
"""
194
if self.new_inventory is None:
195
self.new_inventory = self.repository.get_inventory(
196
self._new_revision_id)
197
return RevisionTree(self.repository, self.new_inventory,
198
self._new_revision_id)
199
200
def finish_inventory(self):
201
"""Tell the builder that the inventory is finished.
202
203
:return: The inventory id in the repository, which can be used with
204
repository.get_inventory.
205
"""
206
if self.new_inventory is None:
207
# an inventory delta was accumulated without creating a new
208
# inventory.
209
basis_id = self.basis_delta_revision
210
self.inv_sha1 = self.repository.add_inventory_by_delta(
211
basis_id, self._basis_delta, self._new_revision_id,
212
self.parents)
213
else:
214
if self.new_inventory.root is None:
215
raise AssertionError('Root entry should be supplied to'
216
' record_entry_contents, as of bzr 0.10.')
217
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
218
self.new_inventory.revision_id = self._new_revision_id
219
self.inv_sha1 = self.repository.add_inventory(
220
self._new_revision_id,
221
self.new_inventory,
222
self.parents
223
)
224
return self._new_revision_id
225
226
def _gen_revision_id(self):
227
"""Return new revision-id."""
228
return generate_ids.gen_revision_id(self._config.username(),
229
self._timestamp)
230
231
def _generate_revision_if_needed(self):
232
"""Create a revision id if None was supplied.
233
234
If the repository can not support user-specified revision ids
235
they should override this function and raise CannotSetRevisionId
236
if _new_revision_id is not None.
237
238
:raises: CannotSetRevisionId
239
"""
240
if self._new_revision_id is None:
241
self._new_revision_id = self._gen_revision_id()
242
self.random_revid = True
243
else:
244
self.random_revid = False
245
246
def _heads(self, file_id, revision_ids):
247
"""Calculate the graph heads for revision_ids in the graph of file_id.
248
249
This can use either a per-file graph or a global revision graph as we
250
have an identity relationship between the two graphs.
251
"""
252
return self.__heads(revision_ids)
253
254
def _check_root(self, ie, parent_invs, tree):
255
"""Helper for record_entry_contents.
256
257
:param ie: An entry being added.
258
:param parent_invs: The inventories of the parent revisions of the
259
commit.
260
:param tree: The tree that is being committed.
261
"""
262
# In this revision format, root entries have no knit or weave When
263
# serializing out to disk and back in root.revision is always
264
# _new_revision_id
265
ie.revision = self._new_revision_id
266
267
def _require_root_change(self, tree):
268
"""Enforce an appropriate root object change.
269
270
This is called once when record_iter_changes is called, if and only if
271
the root was not in the delta calculated by record_iter_changes.
272
273
:param tree: The tree which is being committed.
274
"""
275
# NB: if there are no parents then this method is not called, so no
276
# need to guard on parents having length.
277
entry = entry_factory['directory'](tree.path2id(''), '',
278
None)
279
entry.revision = self._new_revision_id
280
self._basis_delta.append(('', '', entry.file_id, entry))
281
282
def _get_delta(self, ie, basis_inv, path):
283
"""Get a delta against the basis inventory for ie."""
284
if ie.file_id not in basis_inv:
285
# add
286
result = (None, path, ie.file_id, ie)
287
self._basis_delta.append(result)
288
return result
289
elif ie != basis_inv[ie.file_id]:
290
# common but altered
291
# TODO: avoid tis id2path call.
292
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
293
self._basis_delta.append(result)
294
return result
295
else:
296
# common, unaltered
297
return None
298
299
def get_basis_delta(self):
300
"""Return the complete inventory delta versus the basis inventory.
301
302
This has been built up with the calls to record_delete and
303
record_entry_contents. The client must have already called
304
will_record_deletes() to indicate that they will be generating a
305
complete delta.
306
307
:return: An inventory delta, suitable for use with apply_delta, or
308
Repository.add_inventory_by_delta, etc.
309
"""
310
if not self._recording_deletes:
311
raise AssertionError("recording deletes not activated.")
312
return self._basis_delta
313
314
def record_delete(self, path, file_id):
315
"""Record that a delete occured against a basis tree.
316
317
This is an optional API - when used it adds items to the basis_delta
318
being accumulated by the commit builder. It cannot be called unless the
319
method will_record_deletes() has been called to inform the builder that
320
a delta is being supplied.
321
322
:param path: The path of the thing deleted.
323
:param file_id: The file id that was deleted.
324
"""
325
if not self._recording_deletes:
326
raise AssertionError("recording deletes not activated.")
327
delta = (path, None, file_id, None)
328
self._basis_delta.append(delta)
329
self._any_changes = True
330
return delta
331
332
def will_record_deletes(self):
333
"""Tell the commit builder that deletes are being notified.
334
335
This enables the accumulation of an inventory delta; for the resulting
336
commit to be valid, deletes against the basis MUST be recorded via
337
builder.record_delete().
338
"""
339
self._recording_deletes = True
340
try:
341
basis_id = self.parents[0]
342
except IndexError:
343
basis_id = _mod_revision.NULL_REVISION
344
self.basis_delta_revision = basis_id
345
346
def record_entry_contents(self, ie, parent_invs, path, tree,
347
content_summary):
348
"""Record the content of ie from tree into the commit if needed.
349
350
Side effect: sets ie.revision when unchanged
351
352
:param ie: An inventory entry present in the commit.
353
:param parent_invs: The inventories of the parent revisions of the
354
commit.
355
:param path: The path the entry is at in the tree.
356
:param tree: The tree which contains this entry and should be used to
357
obtain content.
358
:param content_summary: Summary data from the tree about the paths
359
content - stat, length, exec, sha/link target. This is only
360
accessed when the entry has a revision of None - that is when it is
361
a candidate to commit.
362
:return: A tuple (change_delta, version_recorded, fs_hash).
363
change_delta is an inventory_delta change for this entry against
364
the basis tree of the commit, or None if no change occured against
365
the basis tree.
366
version_recorded is True if a new version of the entry has been
367
recorded. For instance, committing a merge where a file was only
368
changed on the other side will return (delta, False).
369
fs_hash is either None, or the hash details for the path (currently
370
a tuple of the contents sha1 and the statvalue returned by
371
tree.get_file_with_stat()).
372
"""
373
if self.new_inventory.root is None:
374
if ie.parent_id is not None:
375
raise errors.RootMissing()
376
self._check_root(ie, parent_invs, tree)
377
if ie.revision is None:
378
kind = content_summary[0]
379
else:
380
# ie is carried over from a prior commit
381
kind = ie.kind
382
# XXX: repository specific check for nested tree support goes here - if
383
# the repo doesn't want nested trees we skip it ?
384
if (kind == 'tree-reference' and
385
not self.repository._format.supports_tree_reference):
386
# mismatch between commit builder logic and repository:
387
# this needs the entry creation pushed down into the builder.
388
raise NotImplementedError('Missing repository subtree support.')
389
self.new_inventory.add(ie)
390
391
# TODO: slow, take it out of the inner loop.
392
try:
393
basis_inv = parent_invs[0]
394
except IndexError:
395
basis_inv = Inventory(root_id=None)
396
397
# ie.revision is always None if the InventoryEntry is considered
398
# for committing. We may record the previous parents revision if the
399
# content is actually unchanged against a sole head.
400
if ie.revision is not None:
401
if not self._versioned_root and path == '':
402
# repositories that do not version the root set the root's
403
# revision to the new commit even when no change occurs (more
404
# specifically, they do not record a revision on the root; and
405
# the rev id is assigned to the root during deserialisation -
406
# this masks when a change may have occurred against the basis.
407
# To match this we always issue a delta, because the revision
408
# of the root will always be changing.
409
if ie.file_id in basis_inv:
410
delta = (basis_inv.id2path(ie.file_id), path,
411
ie.file_id, ie)
412
else:
413
# add
414
delta = (None, path, ie.file_id, ie)
415
self._basis_delta.append(delta)
416
return delta, False, None
417
else:
418
# we don't need to commit this, because the caller already
419
# determined that an existing revision of this file is
420
# appropriate. If its not being considered for committing then
421
# it and all its parents to the root must be unaltered so
422
# no-change against the basis.
423
if ie.revision == self._new_revision_id:
424
raise AssertionError("Impossible situation, a skipped "
425
"inventory entry (%r) claims to be modified in this "
426
"commit (%r).", (ie, self._new_revision_id))
427
return None, False, None
428
# XXX: Friction: parent_candidates should return a list not a dict
429
# so that we don't have to walk the inventories again.
430
parent_candiate_entries = ie.parent_candidates(parent_invs)
431
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
432
heads = []
433
for inv in parent_invs:
434
if ie.file_id in inv:
435
old_rev = inv[ie.file_id].revision
436
if old_rev in head_set:
437
heads.append(inv[ie.file_id].revision)
438
head_set.remove(inv[ie.file_id].revision)
439
440
store = False
441
# now we check to see if we need to write a new record to the
442
# file-graph.
443
# We write a new entry unless there is one head to the ancestors, and
444
# the kind-derived content is unchanged.
445
446
# Cheapest check first: no ancestors, or more the one head in the
447
# ancestors, we write a new node.
448
if len(heads) != 1:
449
store = True
450
if not store:
451
# There is a single head, look it up for comparison
452
parent_entry = parent_candiate_entries[heads[0]]
453
# if the non-content specific data has changed, we'll be writing a
454
# node:
455
if (parent_entry.parent_id != ie.parent_id or
456
parent_entry.name != ie.name):
457
store = True
458
# now we need to do content specific checks:
459
if not store:
460
# if the kind changed the content obviously has
461
if kind != parent_entry.kind:
462
store = True
463
# Stat cache fingerprint feedback for the caller - None as we usually
464
# don't generate one.
465
fingerprint = None
466
if kind == 'file':
467
if content_summary[2] is None:
468
raise ValueError("Files must not have executable = None")
469
if not store:
470
if (# if the file length changed we have to store:
471
parent_entry.text_size != content_summary[1] or
472
# if the exec bit has changed we have to store:
473
parent_entry.executable != content_summary[2]):
474
store = True
475
elif parent_entry.text_sha1 == content_summary[3]:
476
# all meta and content is unchanged (using a hash cache
477
# hit to check the sha)
478
ie.revision = parent_entry.revision
479
ie.text_size = parent_entry.text_size
480
ie.text_sha1 = parent_entry.text_sha1
481
ie.executable = parent_entry.executable
482
return self._get_delta(ie, basis_inv, path), False, None
483
else:
484
# Either there is only a hash change(no hash cache entry,
485
# or same size content change), or there is no change on
486
# this file at all.
487
# Provide the parent's hash to the store layer, so that the
488
# content is unchanged we will not store a new node.
489
nostore_sha = parent_entry.text_sha1
490
if store:
491
# We want to record a new node regardless of the presence or
492
# absence of a content change in the file.
493
nostore_sha = None
494
ie.executable = content_summary[2]
495
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
496
try:
497
lines = file_obj.readlines()
498
finally:
499
file_obj.close()
500
try:
501
ie.text_sha1, ie.text_size = self._add_text_to_weave(
502
ie.file_id, lines, heads, nostore_sha)
503
# Let the caller know we generated a stat fingerprint.
504
fingerprint = (ie.text_sha1, stat_value)
505
except errors.ExistingContent:
506
# Turns out that the file content was unchanged, and we were
507
# only going to store a new node if it was changed. Carry over
508
# the entry.
509
ie.revision = parent_entry.revision
510
ie.text_size = parent_entry.text_size
511
ie.text_sha1 = parent_entry.text_sha1
512
ie.executable = parent_entry.executable
513
return self._get_delta(ie, basis_inv, path), False, None
514
elif kind == 'directory':
515
if not store:
516
# all data is meta here, nothing specific to directory, so
517
# carry over:
518
ie.revision = parent_entry.revision
519
return self._get_delta(ie, basis_inv, path), False, None
520
lines = []
521
self._add_text_to_weave(ie.file_id, lines, heads, None)
522
elif kind == 'symlink':
523
current_link_target = content_summary[3]
524
if not store:
525
# symlink target is not generic metadata, check if it has
526
# changed.
527
if current_link_target != parent_entry.symlink_target:
528
store = True
529
if not store:
530
# unchanged, carry over.
531
ie.revision = parent_entry.revision
532
ie.symlink_target = parent_entry.symlink_target
533
return self._get_delta(ie, basis_inv, path), False, None
534
ie.symlink_target = current_link_target
535
lines = []
536
self._add_text_to_weave(ie.file_id, lines, heads, None)
537
elif kind == 'tree-reference':
538
if not store:
539
if content_summary[3] != parent_entry.reference_revision:
540
store = True
541
if not store:
542
# unchanged, carry over.
543
ie.reference_revision = parent_entry.reference_revision
544
ie.revision = parent_entry.revision
545
return self._get_delta(ie, basis_inv, path), False, None
546
ie.reference_revision = content_summary[3]
547
lines = []
548
self._add_text_to_weave(ie.file_id, lines, heads, None)
549
else:
550
raise NotImplementedError('unknown kind')
551
ie.revision = self._new_revision_id
552
self._any_changes = True
553
return self._get_delta(ie, basis_inv, path), True, fingerprint
554
555
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
556
_entry_factory=entry_factory):
557
"""Record a new tree via iter_changes.
558
559
:param tree: The tree to obtain text contents from for changed objects.
560
:param basis_revision_id: The revision id of the tree the iter_changes
561
has been generated against. Currently assumed to be the same
562
as self.parents[0] - if it is not, errors may occur.
563
:param iter_changes: An iter_changes iterator with the changes to apply
564
to basis_revision_id. The iterator must not include any items with
565
a current kind of None - missing items must be either filtered out
566
or errored-on beefore record_iter_changes sees the item.
567
:param _entry_factory: Private method to bind entry_factory locally for
568
performance.
569
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
570
tree._observed_sha1.
571
"""
572
# Create an inventory delta based on deltas between all the parents and
573
# deltas between all the parent inventories. We use inventory delta's
574
# between the inventory objects because iter_changes masks
575
# last-changed-field only changes.
576
# Working data:
577
# file_id -> change map, change is fileid, paths, changed, versioneds,
578
# parents, names, kinds, executables
579
merged_ids = {}
580
# {file_id -> revision_id -> inventory entry, for entries in parent
581
# trees that are not parents[0]
582
parent_entries = {}
583
ghost_basis = False
584
try:
585
revtrees = list(self.repository.revision_trees(self.parents))
586
except errors.NoSuchRevision:
587
# one or more ghosts, slow path.
588
revtrees = []
589
for revision_id in self.parents:
590
try:
591
revtrees.append(self.repository.revision_tree(revision_id))
592
except errors.NoSuchRevision:
593
if not revtrees:
594
basis_revision_id = _mod_revision.NULL_REVISION
595
ghost_basis = True
596
revtrees.append(self.repository.revision_tree(
597
_mod_revision.NULL_REVISION))
598
# The basis inventory from a repository
599
if revtrees:
600
basis_inv = revtrees[0].inventory
601
else:
602
basis_inv = self.repository.revision_tree(
603
_mod_revision.NULL_REVISION).inventory
604
if len(self.parents) > 0:
605
if basis_revision_id != self.parents[0] and not ghost_basis:
606
raise Exception(
607
"arbitrary basis parents not yet supported with merges")
608
for revtree in revtrees[1:]:
609
for change in revtree.inventory._make_delta(basis_inv):
610
if change[1] is None:
611
# Not present in this parent.
612
continue
613
if change[2] not in merged_ids:
614
if change[0] is not None:
615
basis_entry = basis_inv[change[2]]
616
merged_ids[change[2]] = [
617
# basis revid
618
basis_entry.revision,
619
# new tree revid
620
change[3].revision]
621
parent_entries[change[2]] = {
622
# basis parent
623
basis_entry.revision:basis_entry,
624
# this parent
625
change[3].revision:change[3],
626
}
627
else:
628
merged_ids[change[2]] = [change[3].revision]
629
parent_entries[change[2]] = {change[3].revision:change[3]}
630
else:
631
merged_ids[change[2]].append(change[3].revision)
632
parent_entries[change[2]][change[3].revision] = change[3]
633
else:
634
merged_ids = {}
635
# Setup the changes from the tree:
636
# changes maps file_id -> (change, [parent revision_ids])
637
changes= {}
638
for change in iter_changes:
639
# This probably looks up in basis_inv way to much.
640
if change[1][0] is not None:
641
head_candidate = [basis_inv[change[0]].revision]
642
else:
643
head_candidate = []
644
changes[change[0]] = change, merged_ids.get(change[0],
645
head_candidate)
646
unchanged_merged = set(merged_ids) - set(changes)
647
# Extend the changes dict with synthetic changes to record merges of
648
# texts.
649
for file_id in unchanged_merged:
650
# Record a merged version of these items that did not change vs the
651
# basis. This can be either identical parallel changes, or a revert
652
# of a specific file after a merge. The recorded content will be
653
# that of the current tree (which is the same as the basis), but
654
# the per-file graph will reflect a merge.
655
# NB:XXX: We are reconstructing path information we had, this
656
# should be preserved instead.
657
# inv delta change: (file_id, (path_in_source, path_in_target),
658
# changed_content, versioned, parent, name, kind,
659
# executable)
660
try:
661
basis_entry = basis_inv[file_id]
662
except errors.NoSuchId:
663
# a change from basis->some_parents but file_id isn't in basis
664
# so was new in the merge, which means it must have changed
665
# from basis -> current, and as it hasn't the add was reverted
666
# by the user. So we discard this change.
667
pass
668
else:
669
change = (file_id,
670
(basis_inv.id2path(file_id), tree.id2path(file_id)),
671
False, (True, True),
672
(basis_entry.parent_id, basis_entry.parent_id),
673
(basis_entry.name, basis_entry.name),
674
(basis_entry.kind, basis_entry.kind),
675
(basis_entry.executable, basis_entry.executable))
676
changes[file_id] = (change, merged_ids[file_id])
677
# changes contains tuples with the change and a set of inventory
678
# candidates for the file.
679
# inv delta is:
680
# old_path, new_path, file_id, new_inventory_entry
681
seen_root = False # Is the root in the basis delta?
682
inv_delta = self._basis_delta
683
modified_rev = self._new_revision_id
684
for change, head_candidates in changes.values():
685
if change[3][1]: # versioned in target.
686
# Several things may be happening here:
687
# We may have a fork in the per-file graph
688
# - record a change with the content from tree
689
# We may have a change against < all trees
690
# - carry over the tree that hasn't changed
691
# We may have a change against all trees
692
# - record the change with the content from tree
693
kind = change[6][1]
694
file_id = change[0]
695
entry = _entry_factory[kind](file_id, change[5][1],
696
change[4][1])
697
head_set = self._heads(change[0], set(head_candidates))
698
heads = []
699
# Preserve ordering.
700
for head_candidate in head_candidates:
701
if head_candidate in head_set:
702
heads.append(head_candidate)
703
head_set.remove(head_candidate)
704
carried_over = False
705
if len(heads) == 1:
706
# Could be a carry-over situation:
707
parent_entry_revs = parent_entries.get(file_id, None)
708
if parent_entry_revs:
709
parent_entry = parent_entry_revs.get(heads[0], None)
710
else:
711
parent_entry = None
712
if parent_entry is None:
713
# The parent iter_changes was called against is the one
714
# that is the per-file head, so any change is relevant
715
# iter_changes is valid.
716
carry_over_possible = False
717
else:
718
# could be a carry over situation
719
# A change against the basis may just indicate a merge,
720
# we need to check the content against the source of the
721
# merge to determine if it was changed after the merge
722
# or carried over.
723
if (parent_entry.kind != entry.kind or
724
parent_entry.parent_id != entry.parent_id or
725
parent_entry.name != entry.name):
726
# Metadata common to all entries has changed
727
# against per-file parent
728
carry_over_possible = False
729
else:
730
carry_over_possible = True
731
# per-type checks for changes against the parent_entry
732
# are done below.
733
else:
734
# Cannot be a carry-over situation
735
carry_over_possible = False
736
# Populate the entry in the delta
737
if kind == 'file':
738
# XXX: There is still a small race here: If someone reverts the content of a file
739
# after iter_changes examines and decides it has changed,
740
# we will unconditionally record a new version even if some
741
# other process reverts it while commit is running (with
742
# the revert happening after iter_changes did it's
743
# examination).
744
if change[7][1]:
745
entry.executable = True
746
else:
747
entry.executable = False
748
if (carry_over_possible and
749
parent_entry.executable == entry.executable):
750
# Check the file length, content hash after reading
751
# the file.
752
nostore_sha = parent_entry.text_sha1
753
else:
754
nostore_sha = None
755
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
756
try:
757
lines = file_obj.readlines()
758
finally:
759
file_obj.close()
760
try:
761
entry.text_sha1, entry.text_size = self._add_text_to_weave(
762
file_id, lines, heads, nostore_sha)
763
yield file_id, change[1][1], (entry.text_sha1, stat_value)
764
except errors.ExistingContent:
765
# No content change against a carry_over parent
766
# Perhaps this should also yield a fs hash update?
767
carried_over = True
768
entry.text_size = parent_entry.text_size
769
entry.text_sha1 = parent_entry.text_sha1
770
elif kind == 'symlink':
771
# Wants a path hint?
772
entry.symlink_target = tree.get_symlink_target(file_id)
773
if (carry_over_possible and
774
parent_entry.symlink_target == entry.symlink_target):
775
carried_over = True
776
else:
777
self._add_text_to_weave(change[0], [], heads, None)
778
elif kind == 'directory':
779
if carry_over_possible:
780
carried_over = True
781
else:
782
# Nothing to set on the entry.
783
# XXX: split into the Root and nonRoot versions.
784
if change[1][1] != '' or self.repository.supports_rich_root():
785
self._add_text_to_weave(change[0], [], heads, None)
786
elif kind == 'tree-reference':
787
if not self.repository._format.supports_tree_reference:
788
# This isn't quite sane as an error, but we shouldn't
789
# ever see this code path in practice: tree's don't
790
# permit references when the repo doesn't support tree
791
# references.
792
raise errors.UnsupportedOperation(tree.add_reference,
793
self.repository)
794
entry.reference_revision = \
795
tree.get_reference_revision(change[0])
796
if (carry_over_possible and
797
parent_entry.reference_revision == reference_revision):
798
carried_over = True
799
else:
800
self._add_text_to_weave(change[0], [], heads, None)
801
else:
802
raise AssertionError('unknown kind %r' % kind)
803
if not carried_over:
804
entry.revision = modified_rev
805
else:
806
entry.revision = parent_entry.revision
807
else:
808
entry = None
809
new_path = change[1][1]
810
inv_delta.append((change[1][0], new_path, change[0], entry))
811
if new_path == '':
812
seen_root = True
813
self.new_inventory = None
814
if len(inv_delta):
815
self._any_changes = True
816
if not seen_root:
817
# housekeeping root entry changes do not affect no-change commits.
818
self._require_root_change(tree)
819
self.basis_delta_revision = basis_revision_id
820
821
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
822
# Note: as we read the content directly from the tree, we know its not
823
# been turned into unicode or badly split - but a broken tree
824
# implementation could give us bad output from readlines() so this is
825
# not a guarantee of safety. What would be better is always checking
826
# the content during test suite execution. RBC 20070912
827
parent_keys = tuple((file_id, parent) for parent in parents)
828
return self.repository.texts.add_lines(
829
(file_id, self._new_revision_id), parent_keys, new_lines,
830
nostore_sha=nostore_sha, random_id=self.random_revid,
831
check_content=False)[0:2]
832
833
834
class RootCommitBuilder(CommitBuilder):
835
"""This commitbuilder actually records the root id"""
836
837
# the root entry gets versioned properly by this builder.
838
_versioned_root = True
839
840
def _check_root(self, ie, parent_invs, tree):
841
"""Helper for record_entry_contents.
842
843
:param ie: An entry being added.
844
:param parent_invs: The inventories of the parent revisions of the
845
commit.
846
:param tree: The tree that is being committed.
847
"""
848
849
def _require_root_change(self, tree):
850
"""Enforce an appropriate root object change.
851
852
This is called once when record_iter_changes is called, if and only if
853
the root was not in the delta calculated by record_iter_changes.
854
855
:param tree: The tree which is being committed.
856
"""
857
# versioned roots do not change unless the tree found a change.
858
859
860
63
######################################################################
861
64
# Repositories
862
65
867
70
revisions and file history. It's normally accessed only by the Branch,
868
71
which views a particular line of development through that history.
869
72
870
The Repository builds on top of some byte storage facilies (the revisions,
871
signatures, inventories, texts and chk_bytes attributes) and a Transport,
872
which respectively provide byte storage and a means to access the (possibly
73
The Repository builds on top of Stores and a Transport, which respectively
74
describe the disk data format and the way of accessing the (possibly
873
75
remote) disk.
874
875
The byte storage facilities are addressed via tuples, which we refer to
876
as 'keys' throughout the code base. Revision_keys, inventory_keys and
877
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
878
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
879
byte string made up of a hash identifier and a hash value.
880
We use this interface because it allows low friction with the underlying
881
code that implements disk indices, network encoding and other parts of
882
bzrlib.
883
884
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
885
the serialised revisions for the repository. This can be used to obtain
886
revision graph information or to access raw serialised revisions.
887
The result of trying to insert data into the repository via this store
888
is undefined: it should be considered read-only except for implementors
889
of repositories.
890
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
891
the serialised signatures for the repository. This can be used to
892
obtain access to raw serialised signatures. The result of trying to
893
insert data into the repository via this store is undefined: it should
894
be considered read-only except for implementors of repositories.
895
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
896
the serialised inventories for the repository. This can be used to
897
obtain unserialised inventories. The result of trying to insert data
898
into the repository via this store is undefined: it should be
899
considered read-only except for implementors of repositories.
900
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
901
texts of files and directories for the repository. This can be used to
902
obtain file texts or file graphs. Note that Repository.iter_file_bytes
903
is usually a better interface for accessing file texts.
904
The result of trying to insert data into the repository via this store
905
is undefined: it should be considered read-only except for implementors
906
of repositories.
907
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
908
any data the repository chooses to store or have indexed by its hash.
909
The result of trying to insert data into the repository via this store
910
is undefined: it should be considered read-only except for implementors
911
of repositories.
912
:ivar _transport: Transport for file access to repository, typically
913
pointing to .bzr/repository.
914
76
"""
915
77
916
# What class to use for a CommitBuilder. Often its simpler to change this
917
# in a Repository class subclass rather than to override
918
# get_commit_builder.
919
_commit_builder_class = CommitBuilder
920
# The search regex used by xml based repositories to determine what things
921
# where changed in a single commit.
922
78
_file_ids_altered_regex = lazy_regex.lazy_compile(
923
79
r'file_id="(?P<file_id>[^"]+)"'
924
r'.* revision="(?P<revision_id>[^"]+)"'
80
r'.*revision="(?P<revision_id>[^"]+)"'
925
81
)
926
82
927
def abort_write_group(self, suppress_errors=False):
928
"""Commit the contents accrued within the current write group.
929
930
:param suppress_errors: if true, abort_write_group will catch and log
931
unexpected errors that happen during the abort, rather than
932
allowing them to propagate. Defaults to False.
933
934
:seealso: start_write_group.
935
"""
936
if self._write_group is not self.get_transaction():
937
# has an unlock or relock occured ?
938
raise errors.BzrError(
939
'mismatched lock context and write group. %r, %r' %
940
(self._write_group, self.get_transaction()))
941
try:
942
self._abort_write_group()
943
except Exception, exc:
944
self._write_group = None
945
if not suppress_errors:
946
raise
947
mutter('abort_write_group failed')
948
log_exception_quietly()
949
note('bzr: ERROR (ignored): %s', exc)
950
self._write_group = None
951
952
def _abort_write_group(self):
953
"""Template method for per-repository write group cleanup.
954
955
This is called during abort before the write group is considered to be
956
finished and should cleanup any internal state accrued during the write
957
group. There is no requirement that data handed to the repository be
958
*not* made available - this is not a rollback - but neither should any
959
attempt be made to ensure that data added is fully commited. Abort is
960
invoked when an error has occured so futher disk or network operations
961
may not be possible or may error and if possible should not be
962
attempted.
963
"""
964
965
def add_fallback_repository(self, repository):
966
"""Add a repository to use for looking up data not held locally.
967
968
:param repository: A repository.
969
"""
970
if not self._format.supports_external_lookups:
971
raise errors.UnstackableRepositoryFormat(self._format, self.base)
972
self._check_fallback_repository(repository)
973
self._fallback_repositories.append(repository)
974
self.texts.add_fallback_versioned_files(repository.texts)
975
self.inventories.add_fallback_versioned_files(repository.inventories)
976
self.revisions.add_fallback_versioned_files(repository.revisions)
977
self.signatures.add_fallback_versioned_files(repository.signatures)
978
if self.chk_bytes is not None:
979
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
980
981
def _check_fallback_repository(self, repository):
982
"""Check that this repository can fallback to repository safely.
983
984
Raise an error if not.
985
986
:param repository: A repository to fallback to.
987
"""
988
return InterRepository._assert_same_model(self, repository)
989
83
@needs_write_lock
990
84
def add_inventory(self, revision_id, inv, parents):
991
85
"""Add the inventory inv to the repository as revision_id.
992
86
993
87
:param parents: The revision ids of the parents that revision_id
994
88
is known to have and are in the repository already.
995
89
996
:returns: The validator(which is a sha1 digest, though what is sha'd is
997
repository format specific) of the serialized inventory.
90
returns the sha1 of the serialized inventory.
998
91
"""
999
if not self.is_in_write_group():
1000
raise AssertionError("%r not in write group" % (self,))
92
revision_id = osutils.safe_revision_id(revision_id)
1001
93
_mod_revision.check_not_reserved_id(revision_id)
1002
if not (inv.revision_id is None or inv.revision_id == revision_id):
1003
raise AssertionError(
1004
"Mismatch between inventory revision"
1005
" id and insertion revid (%r, %r)"
1006
% (inv.revision_id, revision_id))
1007
if inv.root is None:
1008
raise AssertionError()
1009
return self._add_inventory_checked(revision_id, inv, parents)
1010
1011
def _add_inventory_checked(self, revision_id, inv, parents):
1012
"""Add inv to the repository after checking the inputs.
1013
1014
This function can be overridden to allow different inventory styles.
1015
1016
:seealso: add_inventory, for the contract.
1017
"""
1018
inv_lines = self._serialise_inventory_to_lines(inv)
1019
return self._inventory_add_lines(revision_id, parents,
1020
inv_lines, check_content=False)
1021
1022
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1023
parents, basis_inv=None, propagate_caches=False):
1024
"""Add a new inventory expressed as a delta against another revision.
1025
1026
:param basis_revision_id: The inventory id the delta was created
1027
against. (This does not have to be a direct parent.)
1028
:param delta: The inventory delta (see Inventory.apply_delta for
1029
details).
1030
:param new_revision_id: The revision id that the inventory is being
1031
added for.
1032
:param parents: The revision ids of the parents that revision_id is
1033
known to have and are in the repository already. These are supplied
1034
for repositories that depend on the inventory graph for revision
1035
graph access, as well as for those that pun ancestry with delta
1036
compression.
1037
:param basis_inv: The basis inventory if it is already known,
1038
otherwise None.
1039
:param propagate_caches: If True, the caches for this inventory are
1040
copied to and updated for the result if possible.
1041
1042
:returns: (validator, new_inv)
1043
The validator(which is a sha1 digest, though what is sha'd is
1044
repository format specific) of the serialized inventory, and the
1045
resulting inventory.
1046
"""
1047
if not self.is_in_write_group():
1048
raise AssertionError("%r not in write group" % (self,))
1049
_mod_revision.check_not_reserved_id(new_revision_id)
1050
basis_tree = self.revision_tree(basis_revision_id)
1051
basis_tree.lock_read()
1052
try:
1053
# Note that this mutates the inventory of basis_tree, which not all
1054
# inventory implementations may support: A better idiom would be to
1055
# return a new inventory, but as there is no revision tree cache in
1056
# repository this is safe for now - RBC 20081013
1057
if basis_inv is None:
1058
basis_inv = basis_tree.inventory
1059
basis_inv.apply_delta(delta)
1060
basis_inv.revision_id = new_revision_id
1061
return (self.add_inventory(new_revision_id, basis_inv, parents),
1062
basis_inv)
1063
finally:
1064
basis_tree.unlock()
1065
1066
def _inventory_add_lines(self, revision_id, parents, lines,
1067
check_content=True):
1068
"""Store lines in inv_vf and return the sha1 of the inventory."""
1069
parents = [(parent,) for parent in parents]
1070
return self.inventories.add_lines((revision_id,), parents, lines,
1071
check_content=check_content)[0]
1072
94
assert inv.revision_id is None or inv.revision_id == revision_id, \
95
"Mismatch between inventory revision" \
96
" id and insertion revid (%r, %r)" % (inv.revision_id, revision_id)
97
assert inv.root is not None
98
inv_text = self.serialise_inventory(inv)
99
inv_sha1 = osutils.sha_string(inv_text)
100
inv_vf = self.control_weaves.get_weave('inventory',
101
self.get_transaction())
102
self._inventory_add_lines(inv_vf, revision_id, parents,
103
osutils.split_lines(inv_text))
104
return inv_sha1
105
106
def _inventory_add_lines(self, inv_vf, revision_id, parents, lines):
107
final_parents = []
108
for parent in parents:
109
if parent in inv_vf:
110
final_parents.append(parent)
111
112
inv_vf.add_lines(revision_id, final_parents, lines)
113
114
@needs_write_lock
1073
115
def add_revision(self, revision_id, rev, inv=None, config=None):
1074
116
"""Add rev to the revision store as revision_id.
1075
117
1081
123
If supplied its signature_needed method will be used
1082
124
to determine if a signature should be made.
1083
125
"""
126
revision_id = osutils.safe_revision_id(revision_id)
1084
127
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1085
128
# rev.parent_ids?
1086
129
_mod_revision.check_not_reserved_id(revision_id)
1090
133
plaintext = Testament(rev, inv).as_short_text()
1091
134
self.store_revision_signature(
1092
135
gpg.GPGStrategy(config), plaintext, revision_id)
1093
# check inventory present
1094
if not self.inventories.get_parent_map([(revision_id,)]):
136
if not revision_id in self.get_inventory_weave():
1095
137
if inv is None:
1096
138
raise errors.WeaveRevisionNotPresent(revision_id,
1097
self.inventories)
139
self.get_inventory_weave())
1098
140
else:
1099
141
# yes, this is not suitable for adding with ghosts.
1100
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1101
rev.parent_ids)
1102
else:
1103
key = (revision_id,)
1104
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1105
self._add_revision(rev)
142
self.add_inventory(revision_id, inv, rev.parent_ids)
143
self._revision_store.add_revision(rev, self.get_transaction())
1106
144
1107
def _add_revision(self, revision):
1108
text = self._serializer.write_revision_to_string(revision)
1109
key = (revision.revision_id,)
1110
parents = tuple((parent,) for parent in revision.parent_ids)
1111
self.revisions.add_lines(key, parents, osutils.split_lines(text))
145
@needs_read_lock
146
def _all_possible_ids(self):
147
"""Return all the possible revisions that we could find."""
148
return self.get_inventory_weave().versions()
1112
149
1113
150
def all_revision_ids(self):
1114
"""Returns a list of all the revision ids in the repository.
151
"""Returns a list of all the revision ids in the repository.
1115
152
1116
This is conceptually deprecated because code should generally work on
1117
the graph reachable from a particular revision, and ignore any other
1118
revisions that might be present. There is no direct replacement
1119
method.
153
This is deprecated because code should generally work on the graph
154
reachable from a particular revision, and ignore any other revisions
155
that might be present. There is no direct replacement method.
1120
156
"""
1121
if 'evil' in debug.debug_flags:
1122
mutter_callsite(2, "all_revision_ids is linear with history.")
1123
157
return self._all_revision_ids()
1124
158
159
@needs_read_lock
1125
160
def _all_revision_ids(self):
1126
"""Returns a list of all the revision ids in the repository.
161
"""Returns a list of all the revision ids in the repository.
1127
162
1128
These are in as much topological order as the underlying store can
1129
present.
163
These are in as much topological order as the underlying store can
164
present: for weaves ghosts may lead to a lack of correctness until
165
the reweave updates the parents list.
1130
166
"""
1131
raise NotImplementedError(self._all_revision_ids)
167
if self._revision_store.text_store.listable():
168
return self._revision_store.all_revision_ids(self.get_transaction())
169
result = self._all_possible_ids()
170
# TODO: jam 20070210 Ensure that _all_possible_ids returns non-unicode
171
# ids. (It should, since _revision_store's API should change to
172
# return utf8 revision_ids)
173
return self._eliminate_revisions_not_present(result)
1132
174
1133
175
def break_lock(self):
1134
176
"""Break a lock if one is present from another instance.
1145
187
Returns a set of the present revisions.
1146
188
"""
1147
189
result = []
1148
graph = self.get_graph()
1149
parent_map = graph.get_parent_map(revision_ids)
1150
# The old API returned a list, should this actually be a set?
1151
return parent_map.keys()
190
for id in revision_ids:
191
if self.has_revision(id):
192
result.append(id)
193
return result
1152
194
1153
195
@staticmethod
1154
196
def create(a_bzrdir):
1155
197
"""Construct the current default format repository in a_bzrdir."""
1156
198
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1157
199
1158
def __init__(self, _format, a_bzrdir, control_files):
200
def __init__(self, _format, a_bzrdir, control_files, _revision_store, control_store, text_store):
1159
201
"""instantiate a Repository.
1160
202
1161
203
:param _format: The format of the repository on disk.
1170
212
# the following are part of the public API for Repository:
1171
213
self.bzrdir = a_bzrdir
1172
214
self.control_files = control_files
1173
self._transport = control_files._transport
1174
self.base = self._transport.base
1175
# for tests
1176
self._reconcile_does_inventory_gc = True
1177
self._reconcile_fixes_text_parents = False
1178
self._reconcile_backsup_inventory = True
1179
# not right yet - should be more semantically clear ?
1180
#
215
self._revision_store = _revision_store
216
self.text_store = text_store
217
# backwards compatibility
218
self.weave_store = text_store
219
# not right yet - should be more semantically clear ?
220
#
221
self.control_store = control_store
222
self.control_weaves = control_store
1181
223
# TODO: make sure to construct the right store classes, etc, depending
1182
224
# on whether escaping is required.
1183
225
self._warn_if_deprecated()
1184
self._write_group = None
1185
# Additional places to query for data.
1186
self._fallback_repositories = []
1187
# An InventoryEntry cache, used during deserialization
1188
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1189
226
1190
227
def __repr__(self):
1191
return '%s(%r)' % (self.__class__.__name__,
1192
self.base)
1193
1194
def has_same_location(self, other):
1195
"""Returns a boolean indicating if this repository is at the same
1196
location as another repository.
1197
1198
This might return False even when two repository objects are accessing
1199
the same physical repository via different URLs.
1200
"""
1201
if self.__class__ is not other.__class__:
1202
return False
1203
return (self._transport.base == other._transport.base)
1204
1205
def is_in_write_group(self):
1206
"""Return True if there is an open write group.
1207
1208
:seealso: start_write_group.
1209
"""
1210
return self._write_group is not None
228
return '%s(%r)' % (self.__class__.__name__,
229
self.bzrdir.transport.base)
1211
230
1212
231
def is_locked(self):
1213
232
return self.control_files.is_locked()
1214
233
1215
def is_write_locked(self):
1216
"""Return True if this object is write locked."""
1217
return self.is_locked() and self.control_files._lock_mode == 'w'
1218
1219
def lock_write(self, token=None):
1220
"""Lock this repository for writing.
1221
1222
This causes caching within the repository obejct to start accumlating
1223
data during reads, and allows a 'write_group' to be obtained. Write
1224
groups must be used for actual data insertion.
1225
1226
:param token: if this is already locked, then lock_write will fail
1227
unless the token matches the existing lock.
1228
:returns: a token if this instance supports tokens, otherwise None.
1229
:raises TokenLockingNotSupported: when a token is given but this
1230
instance doesn't support using token locks.
1231
:raises MismatchedToken: if the specified token doesn't match the token
1232
of the existing lock.
1233
:seealso: start_write_group.
1234
1235
A token should be passed in if you know that you have locked the object
1236
some other way, and need to synchronise this object's state with that
1237
fact.
1238
1239
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1240
"""
1241
locked = self.is_locked()
1242
result = self.control_files.lock_write(token=token)
1243
for repo in self._fallback_repositories:
1244
# Writes don't affect fallback repos
1245
repo.lock_read()
1246
if not locked:
1247
self._refresh_data()
1248
return result
234
def lock_write(self):
235
self.control_files.lock_write()
1249
236
1250
237
def lock_read(self):
1251
locked = self.is_locked()
1252
238
self.control_files.lock_read()
1253
for repo in self._fallback_repositories:
1254
repo.lock_read()
1255
if not locked:
1256
self._refresh_data()
1257
239
1258
240
def get_physical_lock_status(self):
1259
241
return self.control_files.get_physical_lock_status()
1260
242
1261
def leave_lock_in_place(self):
1262
"""Tell this repository not to release the physical lock when this
1263
object is unlocked.
1264
1265
If lock_write doesn't return a token, then this method is not supported.
1266
"""
1267
self.control_files.leave_in_place()
1268
1269
def dont_leave_lock_in_place(self):
1270
"""Tell this repository to release the physical lock when this
1271
object is unlocked, even if it didn't originally acquire it.
1272
1273
If lock_write doesn't return a token, then this method is not supported.
1274
"""
1275
self.control_files.dont_leave_in_place()
1276
1277
243
@needs_read_lock
1278
244
def gather_stats(self, revid=None, committers=None):
1279
245
"""Gather statistics from a revision id.
1318
284
last_revision.timezone)
1319
285
1320
286
# now gather global repository information
1321
# XXX: This is available for many repos regardless of listability.
1322
287
if self.bzrdir.root_transport.listable():
1323
# XXX: do we want to __define len__() ?
1324
# Maybe the versionedfiles object should provide a different
1325
# method to get the number of keys.
1326
result['revisions'] = len(self.revisions.keys())
1327
# result['size'] = t
288
c, t = self._revision_store.total_size(self.get_transaction())
289
result['revisions'] = c
290
result['size'] = t
1328
291
return result
1329
292
1330
def find_branches(self, using=False):
1331
"""Find branches underneath this repository.
1332
1333
This will include branches inside other branches.
1334
1335
:param using: If True, list only branches using this repository.
1336
"""
1337
if using and not self.is_shared():
1338
try:
1339
return [self.bzrdir.open_branch()]
1340
except errors.NotBranchError:
1341
return []
1342
class Evaluator(object):
1343
1344
def __init__(self):
1345
self.first_call = True
1346
1347
def __call__(self, bzrdir):
1348
# On the first call, the parameter is always the bzrdir
1349
# containing the current repo.
1350
if not self.first_call:
1351
try:
1352
repository = bzrdir.open_repository()
1353
except errors.NoRepositoryPresent:
1354
pass
1355
else:
1356
return False, (None, repository)
1357
self.first_call = False
1358
try:
1359
value = (bzrdir.open_branch(), None)
1360
except errors.NotBranchError:
1361
value = (None, None)
1362
return True, value
1363
1364
branches = []
1365
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1366
self.bzrdir.root_transport, evaluate=Evaluator()):
1367
if branch is not None:
1368
branches.append(branch)
1369
if not using and repository is not None:
1370
branches.extend(repository.find_branches())
1371
return branches
1372
1373
293
@needs_read_lock
1374
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
294
def missing_revision_ids(self, other, revision_id=None):
1375
295
"""Return the revision ids that other has that this does not.
1376
296
1377
297
These are returned in topological order.
1378
298
1379
299
revision_id: only return revision ids included by revision_id.
1380
300
"""
1381
return InterRepository.get(other, self).search_missing_revision_ids(
1382
revision_id, find_ghosts)
301
revision_id = osutils.safe_revision_id(revision_id)
302
return InterRepository.get(other, self).missing_revision_ids(revision_id)
1383
303
1384
304
@staticmethod
1385
305
def open(base):
1393
313
1394
314
def copy_content_into(self, destination, revision_id=None):
1395
315
"""Make a complete copy of the content in self into destination.
1396
1397
This is a destructive operation! Do not use it on existing
316
317
This is a destructive operation! Do not use it on existing
1398
318
repositories.
1399
319
"""
320
revision_id = osutils.safe_revision_id(revision_id)
1400
321
return InterRepository.get(self, destination).copy_content(revision_id)
1401
322
1402
def commit_write_group(self):
1403
"""Commit the contents accrued within the current write group.
1404
1405
:seealso: start_write_group.
1406
"""
1407
if self._write_group is not self.get_transaction():
1408
# has an unlock or relock occured ?
1409
raise errors.BzrError('mismatched lock context %r and '
1410
'write group %r.' %
1411
(self.get_transaction(), self._write_group))
1412
self._commit_write_group()
1413
self._write_group = None
1414
1415
def _commit_write_group(self):
1416
"""Template method for per-repository write group cleanup.
1417
1418
This is called before the write group is considered to be
1419
finished and should ensure that all data handed to the repository
1420
for writing during the write group is safely committed (to the
1421
extent possible considering file system caching etc).
1422
"""
1423
1424
def suspend_write_group(self):
1425
raise errors.UnsuspendableWriteGroup(self)
1426
1427
def get_missing_parent_inventories(self):
1428
"""Return the keys of missing inventory parents for revisions added in
1429
this write group.
1430
1431
A revision is not complete if the inventory delta for that revision
1432
cannot be calculated. Therefore if the parent inventories of a
1433
revision are not present, the revision is incomplete, and e.g. cannot
1434
be streamed by a smart server. This method finds missing inventory
1435
parents for revisions added in this write group.
1436
"""
1437
if not self._format.supports_external_lookups:
1438
# This is only an issue for stacked repositories
1439
return set()
1440
if not self.is_in_write_group():
1441
raise AssertionError('not in a write group')
1442
1443
# XXX: We assume that every added revision already has its
1444
# corresponding inventory, so we only check for parent inventories that
1445
# might be missing, rather than all inventories.
1446
parents = set(self.revisions._index.get_missing_parents())
1447
parents.discard(_mod_revision.NULL_REVISION)
1448
unstacked_inventories = self.inventories._index
1449
present_inventories = unstacked_inventories.get_parent_map(
1450
key[-1:] for key in parents)
1451
parents.difference_update(present_inventories)
1452
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1453
return missing_keys
1454
1455
def refresh_data(self):
1456
"""Re-read any data needed to to synchronise with disk.
1457
1458
This method is intended to be called after another repository instance
1459
(such as one used by a smart server) has inserted data into the
1460
repository. It may not be called during a write group, but may be
1461
called at any other time.
1462
"""
1463
if self.is_in_write_group():
1464
raise errors.InternalBzrError(
1465
"May not refresh_data while in a write group.")
1466
self._refresh_data()
1467
1468
def resume_write_group(self, tokens):
1469
if not self.is_write_locked():
1470
raise errors.NotWriteLocked(self)
1471
if self._write_group:
1472
raise errors.BzrError('already in a write group')
1473
self._resume_write_group(tokens)
1474
# so we can detect unlock/relock - the write group is now entered.
1475
self._write_group = self.get_transaction()
1476
1477
def _resume_write_group(self, tokens):
1478
raise errors.UnsuspendableWriteGroup(self)
1479
1480
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1481
fetch_spec=None):
323
def fetch(self, source, revision_id=None, pb=None):
1482
324
"""Fetch the content required to construct revision_id from source.
1483
325
1484
If revision_id is None and fetch_spec is None, then all content is
1485
copied.
1486
1487
fetch() may not be used when the repository is in a write group -
1488
either finish the current write group before using fetch, or use
1489
fetch before starting the write group.
1490
1491
:param find_ghosts: Find and copy revisions in the source that are
1492
ghosts in the target (and not reachable directly by walking out to
1493
the first-present revision in target from revision_id).
1494
:param revision_id: If specified, all the content needed for this
1495
revision ID will be copied to the target. Fetch will determine for
1496
itself which content needs to be copied.
1497
:param fetch_spec: If specified, a SearchResult or
1498
PendingAncestryResult that describes which revisions to copy. This
1499
allows copying multiple heads at once. Mutually exclusive with
1500
revision_id.
326
If revision_id is None all content is copied.
1501
327
"""
1502
if fetch_spec is not None and revision_id is not None:
1503
raise AssertionError(
1504
"fetch_spec and revision_id are mutually exclusive.")
1505
if self.is_in_write_group():
1506
raise errors.InternalBzrError(
1507
"May not fetch while in a write group.")
1508
# fast path same-url fetch operations
1509
if self.has_same_location(source) and fetch_spec is None:
1510
# check that last_revision is in 'from' and then return a
1511
# no-operation.
1512
if (revision_id is not None and
1513
not _mod_revision.is_null(revision_id)):
1514
self.get_revision(revision_id)
1515
return 0, []
1516
# if there is no specific appropriate InterRepository, this will get
1517
# the InterRepository base class, which raises an
1518
# IncompatibleRepositories when asked to fetch.
328
revision_id = osutils.safe_revision_id(revision_id)
1519
329
inter = InterRepository.get(source, self)
1520
return inter.fetch(revision_id=revision_id, pb=pb,
1521
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1522
1523
def create_bundle(self, target, base, fileobj, format=None):
1524
return serializer.write_bundle(self, target, base, fileobj, format)
1525
1526
def get_commit_builder(self, branch, parents, config, timestamp=None,
1527
timezone=None, committer=None, revprops=None,
330
try:
331
return inter.fetch(revision_id=revision_id, pb=pb)
332
except NotImplementedError:
333
raise errors.IncompatibleRepositories(source, self)
334
335
def get_commit_builder(self, branch, parents, config, timestamp=None,
336
timezone=None, committer=None, revprops=None,
1528
337
revision_id=None):
1529
338
"""Obtain a CommitBuilder for this repository.
1530
339
1531
340
:param branch: Branch to commit to.
1532
341
:param parents: Revision ids of the parents of the new revision.
1533
342
:param config: Configuration to use.
1537
346
:param revprops: Optional dictionary of revision properties.
1538
347
:param revision_id: Optional revision id.
1539
348
"""
1540
result = self._commit_builder_class(self, parents, config,
1541
timestamp, timezone, committer, revprops, revision_id)
1542
self.start_write_group()
1543
return result
349
revision_id = osutils.safe_revision_id(revision_id)
350
return _CommitBuilder(self, parents, config, timestamp, timezone,
351
committer, revprops, revision_id)
1544
352
1545
353
def unlock(self):
1546
if (self.control_files._lock_count == 1 and
1547
self.control_files._lock_mode == 'w'):
1548
if self._write_group is not None:
1549
self.abort_write_group()
1550
self.control_files.unlock()
1551
raise errors.BzrError(
1552
'Must end write groups before releasing write locks.')
1553
354
self.control_files.unlock()
1554
if self.control_files._lock_count == 0:
1555
self._inventory_entry_cache.clear()
1556
for repo in self._fallback_repositories:
1557
repo.unlock()
1558
355
1559
356
@needs_read_lock
1560
357
def clone(self, a_bzrdir, revision_id=None):
1565
362
1566
363
:return: The newly created destination repository.
1567
364
"""
1568
# TODO: deprecate after 0.16; cloning this with all its settings is
1569
# probably not very useful -- mbp 20070423
1570
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1571
self.copy_content_into(dest_repo, revision_id)
1572
return dest_repo
1573
1574
def start_write_group(self):
1575
"""Start a write group in the repository.
1576
1577
Write groups are used by repositories which do not have a 1:1 mapping
1578
between file ids and backend store to manage the insertion of data from
1579
both fetch and commit operations.
1580
1581
A write lock is required around the start_write_group/commit_write_group
1582
for the support of lock-requiring repository formats.
1583
1584
One can only insert data into a repository inside a write group.
1585
1586
:return: None.
1587
"""
1588
if not self.is_write_locked():
1589
raise errors.NotWriteLocked(self)
1590
if self._write_group:
1591
raise errors.BzrError('already in a write group')
1592
self._start_write_group()
1593
# so we can detect unlock/relock - the write group is now entered.
1594
self._write_group = self.get_transaction()
1595
1596
def _start_write_group(self):
1597
"""Template method for per-repository write group startup.
1598
1599
This is called before the write group is considered to be
1600
entered.
1601
"""
1602
1603
@needs_read_lock
1604
def sprout(self, to_bzrdir, revision_id=None):
1605
"""Create a descendent repository for new development.
1606
1607
Unlike clone, this does not copy the settings of the repository.
1608
"""
1609
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1610
dest_repo.fetch(self, revision_id=revision_id)
1611
return dest_repo
1612
1613
def _create_sprouting_repo(self, a_bzrdir, shared):
1614
365
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1615
366
# use target default format.
1616
367
dest_repo = a_bzrdir.create_repository()
1618
369
# Most control formats need the repository to be specifically
1619
370
# created, but on some old all-in-one formats it's not needed
1620
371
try:
1621
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
372
dest_repo = self._format.initialize(a_bzrdir, shared=self.is_shared())
1622
373
except errors.UninitializableFormat:
1623
374
dest_repo = a_bzrdir.open_repository()
375
self.copy_content_into(dest_repo, revision_id)
1624
376
return dest_repo
1625
377
1626
def _get_sink(self):
1627
"""Return a sink for streaming into this repository."""
1628
return StreamSink(self)
1629
1630
def _get_source(self, to_format):
1631
"""Return a source for streaming from this repository."""
1632
return StreamSource(self, to_format)
1633
1634
378
@needs_read_lock
1635
379
def has_revision(self, revision_id):
1636
380
"""True if this repository has a copy of the revision."""
1637
return revision_id in self.has_revisions((revision_id,))
1638
1639
@needs_read_lock
1640
def has_revisions(self, revision_ids):
1641
"""Probe to find out the presence of multiple revisions.
1642
1643
:param revision_ids: An iterable of revision_ids.
1644
:return: A set of the revision_ids that were present.
1645
"""
1646
parent_map = self.revisions.get_parent_map(
1647
[(rev_id,) for rev_id in revision_ids])
1648
result = set()
1649
if _mod_revision.NULL_REVISION in revision_ids:
1650
result.add(_mod_revision.NULL_REVISION)
1651
result.update([key[0] for key in parent_map])
1652
return result
1653
1654
@needs_read_lock
1655
def get_revision(self, revision_id):
1656
"""Return the Revision object for a named revision."""
1657
return self.get_revisions([revision_id])[0]
381
revision_id = osutils.safe_revision_id(revision_id)
382
return self._revision_store.has_revision_id(revision_id,
383
self.get_transaction())
1658
384
1659
385
@needs_read_lock
1660
386
def get_revision_reconcile(self, revision_id):
1661
387
"""'reconcile' helper routine that allows access to a revision always.
1662
388
1663
389
This variant of get_revision does not cross check the weave graph
1664
390
against the revision one as get_revision does: but it should only
1665
391
be used by reconcile, or reconcile-alike commands that are correcting
1666
392
or testing the revision graph.
1667
393
"""
1668
return self._get_revisions([revision_id])[0]
394
if not revision_id or not isinstance(revision_id, basestring):
395
raise errors.InvalidRevisionId(revision_id=revision_id,
396
branch=self)
397
return self.get_revisions([revision_id])[0]
1669
398
1670
399
@needs_read_lock
1671
400
def get_revisions(self, revision_ids):
1672
"""Get many revisions at once."""
1673
return self._get_revisions(revision_ids)
1674
1675
@needs_read_lock
1676
def _get_revisions(self, revision_ids):
1677
"""Core work logic to get many revisions without sanity checks."""
1678
for rev_id in revision_ids:
1679
if not rev_id or not isinstance(rev_id, basestring):
1680
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1681
keys = [(key,) for key in revision_ids]
1682
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1683
revs = {}
1684
for record in stream:
1685
if record.storage_kind == 'absent':
1686
raise errors.NoSuchRevision(self, record.key[0])
1687
text = record.get_bytes_as('fulltext')
1688
rev = self._serializer.read_revision_from_string(text)
1689
revs[record.key[0]] = rev
1690
return [revs[revid] for revid in revision_ids]
401
revision_ids = [osutils.safe_revision_id(r) for r in revision_ids]
402
revs = self._revision_store.get_revisions(revision_ids,
403
self.get_transaction())
404
for rev in revs:
405
assert not isinstance(rev.revision_id, unicode)
406
for parent_id in rev.parent_ids:
407
assert not isinstance(parent_id, unicode)
408
return revs
1691
409
1692
410
@needs_read_lock
1693
411
def get_revision_xml(self, revision_id):
1694
412
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1695
413
# would have already do it.
1696
414
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1697
# TODO: this can't just be replaced by:
1698
# return self._serializer.write_revision_to_string(
1699
# self.get_revision(revision_id))
1700
# as cStringIO preservers the encoding unlike write_revision_to_string
1701
# or some other call down the path.
415
revision_id = osutils.safe_revision_id(revision_id)
1702
416
rev = self.get_revision(revision_id)
1703
rev_tmp = cStringIO.StringIO()
417
rev_tmp = StringIO()
1704
418
# the current serializer..
1705
self._serializer.write_revision(rev, rev_tmp)
419
self._revision_store._serializer.write_revision(rev, rev_tmp)
1706
420
rev_tmp.seek(0)
1707
421
return rev_tmp.getvalue()
1708
422
1709
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
423
@needs_read_lock
424
def get_revision(self, revision_id):
425
"""Return the Revision object for a named revision"""
426
# TODO: jam 20070210 get_revision_reconcile should do this for us
427
revision_id = osutils.safe_revision_id(revision_id)
428
r = self.get_revision_reconcile(revision_id)
429
# weave corruption can lead to absent revision markers that should be
430
# present.
431
# the following test is reasonably cheap (it needs a single weave read)
432
# and the weave is cached in read transactions. In write transactions
433
# it is not cached but typically we only read a small number of
434
# revisions. For knits when they are introduced we will probably want
435
# to ensure that caching write transactions are in use.
436
inv = self.get_inventory_weave()
437
self._check_revision_parents(r, inv)
438
return r
439
440
@needs_read_lock
441
def get_deltas_for_revisions(self, revisions):
1710
442
"""Produce a generator of revision deltas.
1711
443
1712
444
Note that the input is a sequence of REVISIONS, not revision_ids.
1713
445
Trees will be held in memory until the generator exits.
1714
446
Each delta is relative to the revision's lefthand predecessor.
1715
1716
:param specific_fileids: if not None, the result is filtered
1717
so that only those file-ids, their parents and their
1718
children are included.
1719
447
"""
1720
# Get the revision-ids of interest
1721
448
required_trees = set()
1722
449
for revision in revisions:
1723
450
required_trees.add(revision.revision_id)
1724
451
required_trees.update(revision.parent_ids[:1])
1725
1726
# Get the matching filtered trees. Note that it's more
1727
# efficient to pass filtered trees to changes_from() rather
1728
# than doing the filtering afterwards. changes_from() could
1729
# arguably do the filtering itself but it's path-based, not
1730
# file-id based, so filtering before or afterwards is
1731
# currently easier.
1732
if specific_fileids is None:
1733
trees = dict((t.get_revision_id(), t) for
1734
t in self.revision_trees(required_trees))
1735
else:
1736
trees = dict((t.get_revision_id(), t) for
1737
t in self._filtered_revision_trees(required_trees,
1738
specific_fileids))
1739
1740
# Calculate the deltas
452
trees = dict((t.get_revision_id(), t) for
453
t in self.revision_trees(required_trees))
1741
454
for revision in revisions:
1742
455
if not revision.parent_ids:
1743
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
456
old_tree = self.revision_tree(None)
1744
457
else:
1745
458
old_tree = trees[revision.parent_ids[0]]
1746
459
yield trees[revision.revision_id].changes_from(old_tree)
1747
460
1748
461
@needs_read_lock
1749
def get_revision_delta(self, revision_id, specific_fileids=None):
462
def get_revision_delta(self, revision_id):
1750
463
"""Return the delta for one revision.
1751
464
1752
465
The delta is relative to the left-hand predecessor of the
1753
466
revision.
1754
1755
:param specific_fileids: if not None, the result is filtered
1756
so that only those file-ids, their parents and their
1757
children are included.
1758
467
"""
1759
468
r = self.get_revision(revision_id)
1760
return list(self.get_deltas_for_revisions([r],
1761
specific_fileids=specific_fileids))[0]
469
return list(self.get_deltas_for_revisions([r]))[0]
470
471
def _check_revision_parents(self, revision, inventory):
472
"""Private to Repository and Fetch.
473
474
This checks the parentage of revision in an inventory weave for
475
consistency and is only applicable to inventory-weave-for-ancestry
476
using repository formats & fetchers.
477
"""
478
weave_parents = inventory.get_parents(revision.revision_id)
479
weave_names = inventory.versions()
480
for parent_id in revision.parent_ids:
481
if parent_id in weave_names:
482
# this parent must not be a ghost.
483
if not parent_id in weave_parents:
484
# but it is a ghost
485
raise errors.CorruptRepository(self)
1762
486
1763
487
@needs_write_lock
1764
488
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
489
revision_id = osutils.safe_revision_id(revision_id)
1765
490
signature = gpg_strategy.sign(plaintext)
1766
self.add_signature_text(revision_id, signature)
1767
1768
@needs_write_lock
1769
def add_signature_text(self, revision_id, signature):
1770
self.signatures.add_lines((revision_id,), (),
1771
osutils.split_lines(signature))
1772
1773
def find_text_key_references(self):
1774
"""Find the text key references within the repository.
1775
1776
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1777
to whether they were referred to by the inventory of the
1778
revision_id that they contain. The inventory texts from all present
1779
revision ids are assessed to generate this report.
1780
"""
1781
revision_keys = self.revisions.keys()
1782
w = self.inventories
1783
pb = ui.ui_factory.nested_progress_bar()
1784
try:
1785
return self._find_text_key_references_from_xml_inventory_lines(
1786
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1787
finally:
1788
pb.finished()
1789
1790
def _find_text_key_references_from_xml_inventory_lines(self,
1791
line_iterator):
1792
"""Core routine for extracting references to texts from inventories.
1793
1794
This performs the translation of xml lines to revision ids.
1795
1796
:param line_iterator: An iterator of lines, origin_version_id
1797
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1798
to whether they were referred to by the inventory of the
1799
revision_id that they contain. Note that if that revision_id was
1800
not part of the line_iterator's output then False will be given -
1801
even though it may actually refer to that key.
1802
"""
1803
if not self._serializer.support_altered_by_hack:
1804
raise AssertionError(
1805
"_find_text_key_references_from_xml_inventory_lines only "
1806
"supported for branches which store inventory as unnested xml"
1807
", not on %r" % self)
491
self._revision_store.add_revision_signature_text(revision_id,
492
signature,
493
self.get_transaction())
494
495
def fileids_altered_by_revision_ids(self, revision_ids):
496
"""Find the file ids and versions affected by revisions.
497
498
:param revisions: an iterable containing revision ids.
499
:return: a dictionary mapping altered file-ids to an iterable of
500
revision_ids. Each altered file-ids has the exact revision_ids that
501
altered it listed explicitly.
502
"""
503
assert self._serializer.support_altered_by_hack, \
504
("fileids_altered_by_revision_ids only supported for branches "
505
"which store inventory as unnested xml, not on %r" % self)
506
selected_revision_ids = set(osutils.safe_revision_id(r)
507
for r in revision_ids)
508
w = self.get_inventory_weave()
1808
509
result = {}
1809
510
1810
511
# this code needs to read every new line in every inventory for the
1811
512
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1812
# not present in one of those inventories is unnecessary but not
513
# not present in one of those inventories is unnecessary but not
1813
514
# harmful because we are filtering by the revision id marker in the
1814
# inventory lines : we only select file ids altered in one of those
515
# inventory lines : we only select file ids altered in one of those
1815
516
# revisions. We don't need to see all lines in the inventory because
1816
517
# only those added in an inventory in rev X can contain a revision=X
1817
518
# line.
1826
527
search = self._file_ids_altered_regex.search
1827
528
unescape = _unescape_xml
1828
529
setdefault = result.setdefault
1829
for line, line_key in line_iterator:
1830
match = search(line)
1831
if match is None:
1832
continue
1833
# One call to match.group() returning multiple items is quite a
1834
# bit faster than 2 calls to match.group() each returning 1
1835
file_id, revision_id = match.group('file_id', 'revision_id')
1836
1837
# Inlining the cache lookups helps a lot when you make 170,000
1838
# lines and 350k ids, versus 8.4 unique ids.
1839
# Using a cache helps in 2 ways:
1840
# 1) Avoids unnecessary decoding calls
1841
# 2) Re-uses cached strings, which helps in future set and
1842
# equality checks.
1843
# (2) is enough that removing encoding entirely along with
1844
# the cache (so we are using plain strings) results in no
1845
# performance improvement.
1846
try:
1847
revision_id = unescape_revid_cache[revision_id]
1848
except KeyError:
1849
unescaped = unescape(revision_id)
1850
unescape_revid_cache[revision_id] = unescaped
1851
revision_id = unescaped
1852
1853
# Note that unconditionally unescaping means that we deserialise
1854
# every fileid, which for general 'pull' is not great, but we don't
1855
# really want to have some many fulltexts that this matters anyway.
1856
# RBC 20071114.
1857
try:
1858
file_id = unescape_fileid_cache[file_id]
1859
except KeyError:
1860
unescaped = unescape(file_id)
1861
unescape_fileid_cache[file_id] = unescaped
1862
file_id = unescaped
1863
1864
key = (file_id, revision_id)
1865
setdefault(key, False)
1866
if revision_id == line_key[-1]:
1867
result[key] = True
1868
return result
1869
1870
def _inventory_xml_lines_for_keys(self, keys):
1871
"""Get a line iterator of the sort needed for findind references.
1872
1873
Not relevant for non-xml inventory repositories.
1874
1875
Ghosts in revision_keys are ignored.
1876
1877
:param revision_keys: The revision keys for the inventories to inspect.
1878
:return: An iterator over (inventory line, revid) for the fulltexts of
1879
all of the xml inventories specified by revision_keys.
1880
"""
1881
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1882
for record in stream:
1883
if record.storage_kind != 'absent':
1884
chunks = record.get_bytes_as('chunked')
1885
revid = record.key[-1]
1886
lines = osutils.chunks_to_lines(chunks)
1887
for line in lines:
1888
yield line, revid
1889
1890
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1891
revision_ids):
1892
"""Helper routine for fileids_altered_by_revision_ids.
1893
1894
This performs the translation of xml lines to revision ids.
1895
1896
:param line_iterator: An iterator of lines, origin_version_id
1897
:param revision_ids: The revision ids to filter for. This should be a
1898
set or other type which supports efficient __contains__ lookups, as
1899
the revision id from each parsed line will be looked up in the
1900
revision_ids filter.
1901
:return: a dictionary mapping altered file-ids to an iterable of
1902
revision_ids. Each altered file-ids has the exact revision_ids that
1903
altered it listed explicitly.
1904
"""
1905
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1906
line_iterator).iterkeys())
1907
# Note that revision_ids are revision keys.
1908
parent_maps = self.revisions.get_parent_map(revision_ids)
1909
parents = set()
1910
map(parents.update, parent_maps.itervalues())
1911
parents.difference_update(revision_ids)
1912
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1913
self._inventory_xml_lines_for_keys(parents)))
1914
new_keys = seen - parent_seen
1915
result = {}
1916
setdefault = result.setdefault
1917
for key in new_keys:
1918
setdefault(key[0], set()).add(key[-1])
1919
return result
1920
1921
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1922
"""Find the file ids and versions affected by revisions.
1923
1924
:param revisions: an iterable containing revision ids.
1925
:param _inv_weave: The inventory weave from this repository or None.
1926
If None, the inventory weave will be opened automatically.
1927
:return: a dictionary mapping altered file-ids to an iterable of
1928
revision_ids. Each altered file-ids has the exact revision_ids that
1929
altered it listed explicitly.
1930
"""
1931
selected_keys = set((revid,) for revid in revision_ids)
1932
w = _inv_weave or self.inventories
1933
pb = ui.ui_factory.nested_progress_bar()
1934
try:
1935
return self._find_file_ids_from_xml_inventory_lines(
1936
w.iter_lines_added_or_present_in_keys(
1937
selected_keys, pb=pb),
1938
selected_keys)
1939
finally:
1940
pb.finished()
1941
1942
def iter_files_bytes(self, desired_files):
1943
"""Iterate through file versions.
1944
1945
Files will not necessarily be returned in the order they occur in
1946
desired_files. No specific order is guaranteed.
1947
1948
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1949
value supplied by the caller as part of desired_files. It should
1950
uniquely identify the file version in the caller's context. (Examples:
1951
an index number or a TreeTransform trans_id.)
1952
1953
bytes_iterator is an iterable of bytestrings for the file. The
1954
kind of iterable and length of the bytestrings are unspecified, but for
1955
this implementation, it is a list of bytes produced by
1956
VersionedFile.get_record_stream().
1957
1958
:param desired_files: a list of (file_id, revision_id, identifier)
1959
triples
1960
"""
1961
text_keys = {}
1962
for file_id, revision_id, callable_data in desired_files:
1963
text_keys[(file_id, revision_id)] = callable_data
1964
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1965
if record.storage_kind == 'absent':
1966
raise errors.RevisionNotPresent(record.key, self)
1967
yield text_keys[record.key], record.get_bytes_as('chunked')
1968
1969
def _generate_text_key_index(self, text_key_references=None,
1970
ancestors=None):
1971
"""Generate a new text key index for the repository.
1972
1973
This is an expensive function that will take considerable time to run.
1974
1975
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1976
list of parents, also text keys. When a given key has no parents,
1977
the parents list will be [NULL_REVISION].
1978
"""
1979
# All revisions, to find inventory parents.
1980
if ancestors is None:
1981
graph = self.get_graph()
1982
ancestors = graph.get_parent_map(self.all_revision_ids())
1983
if text_key_references is None:
1984
text_key_references = self.find_text_key_references()
1985
pb = ui.ui_factory.nested_progress_bar()
1986
try:
1987
return self._do_generate_text_key_index(ancestors,
1988
text_key_references, pb)
1989
finally:
1990
pb.finished()
1991
1992
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1993
"""Helper for _generate_text_key_index to avoid deep nesting."""
1994
revision_order = tsort.topo_sort(ancestors)
1995
invalid_keys = set()
1996
revision_keys = {}
1997
for revision_id in revision_order:
1998
revision_keys[revision_id] = set()
1999
text_count = len(text_key_references)
2000
# a cache of the text keys to allow reuse; costs a dict of all the
2001
# keys, but saves a 2-tuple for every child of a given key.
2002
text_key_cache = {}
2003
for text_key, valid in text_key_references.iteritems():
2004
if not valid:
2005
invalid_keys.add(text_key)
2006
else:
2007
revision_keys[text_key[1]].add(text_key)
2008
text_key_cache[text_key] = text_key
2009
del text_key_references
2010
text_index = {}
2011
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2012
NULL_REVISION = _mod_revision.NULL_REVISION
2013
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2014
# too small for large or very branchy trees. However, for 55K path
2015
# trees, it would be easy to use too much memory trivially. Ideally we
2016
# could gauge this by looking at available real memory etc, but this is
2017
# always a tricky proposition.
2018
inventory_cache = lru_cache.LRUCache(10)
2019
batch_size = 10 # should be ~150MB on a 55K path tree
2020
batch_count = len(revision_order) / batch_size + 1
2021
processed_texts = 0
2022
pb.update("Calculating text parents", processed_texts, text_count)
2023
for offset in xrange(batch_count):
2024
to_query = revision_order[offset * batch_size:(offset + 1) *
2025
batch_size]
2026
if not to_query:
2027
break
2028
for rev_tree in self.revision_trees(to_query):
2029
revision_id = rev_tree.get_revision_id()
2030
parent_ids = ancestors[revision_id]
2031
for text_key in revision_keys[revision_id]:
2032
pb.update("Calculating text parents", processed_texts)
2033
processed_texts += 1
2034
candidate_parents = []
2035
for parent_id in parent_ids:
2036
parent_text_key = (text_key[0], parent_id)
2037
try:
2038
check_parent = parent_text_key not in \
2039
revision_keys[parent_id]
2040
except KeyError:
2041
# the parent parent_id is a ghost:
2042
check_parent = False
2043
# truncate the derived graph against this ghost.
2044
parent_text_key = None
2045
if check_parent:
2046
# look at the parent commit details inventories to
2047
# determine possible candidates in the per file graph.
2048
# TODO: cache here.
2049
try:
2050
inv = inventory_cache[parent_id]
2051
except KeyError:
2052
inv = self.revision_tree(parent_id).inventory
2053
inventory_cache[parent_id] = inv
2054
try:
2055
parent_entry = inv[text_key[0]]
2056
except (KeyError, errors.NoSuchId):
2057
parent_entry = None
2058
if parent_entry is not None:
2059
parent_text_key = (
2060
text_key[0], parent_entry.revision)
2061
else:
2062
parent_text_key = None
2063
if parent_text_key is not None:
2064
candidate_parents.append(
2065
text_key_cache[parent_text_key])
2066
parent_heads = text_graph.heads(candidate_parents)
2067
new_parents = list(parent_heads)
2068
new_parents.sort(key=lambda x:candidate_parents.index(x))
2069
if new_parents == []:
2070
new_parents = [NULL_REVISION]
2071
text_index[text_key] = new_parents
2072
2073
for text_key in invalid_keys:
2074
text_index[text_key] = [NULL_REVISION]
2075
return text_index
2076
2077
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2078
"""Get an iterable listing the keys of all the data introduced by a set
2079
of revision IDs.
2080
2081
The keys will be ordered so that the corresponding items can be safely
2082
fetched and inserted in that order.
2083
2084
:returns: An iterable producing tuples of (knit-kind, file-id,
2085
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2086
'revisions'. file-id is None unless knit-kind is 'file'.
2087
"""
2088
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2089
yield result
2090
del _files_pb
2091
for result in self._find_non_file_keys_to_fetch(revision_ids):
2092
yield result
2093
2094
def _find_file_keys_to_fetch(self, revision_ids, pb):
2095
# XXX: it's a bit weird to control the inventory weave caching in this
2096
# generator. Ideally the caching would be done in fetch.py I think. Or
2097
# maybe this generator should explicitly have the contract that it
2098
# should not be iterated until the previously yielded item has been
2099
# processed?
2100
inv_w = self.inventories
2101
2102
# file ids that changed
2103
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2104
count = 0
2105
num_file_ids = len(file_ids)
2106
for file_id, altered_versions in file_ids.iteritems():
2107
if pb is not None:
2108
pb.update("fetch texts", count, num_file_ids)
2109
count += 1
2110
yield ("file", file_id, altered_versions)
2111
2112
def _find_non_file_keys_to_fetch(self, revision_ids):
2113
# inventory
2114
yield ("inventory", None, revision_ids)
2115
2116
# signatures
2117
# XXX: Note ATM no callers actually pay attention to this return
2118
# instead they just use the list of revision ids and ignore
2119
# missing sigs. Consider removing this work entirely
2120
revisions_with_signatures = set(self.signatures.get_parent_map(
2121
[(r,) for r in revision_ids]))
2122
revisions_with_signatures = set(
2123
[r for (r,) in revisions_with_signatures])
2124
revisions_with_signatures.intersection_update(revision_ids)
2125
yield ("signatures", None, revisions_with_signatures)
2126
2127
# revisions
2128
yield ("revisions", None, revision_ids)
530
pb = ui.ui_factory.nested_progress_bar()
531
try:
532
for line in w.iter_lines_added_or_present_in_versions(
533
selected_revision_ids, pb=pb):
534
match = search(line)
535
if match is None:
536
continue
537
# One call to match.group() returning multiple items is quite a
538
# bit faster than 2 calls to match.group() each returning 1
539
file_id, revision_id = match.group('file_id', 'revision_id')
540
541
# Inlining the cache lookups helps a lot when you make 170,000
542
# lines and 350k ids, versus 8.4 unique ids.
543
# Using a cache helps in 2 ways:
544
# 1) Avoids unnecessary decoding calls
545
# 2) Re-uses cached strings, which helps in future set and
546
# equality checks.
547
# (2) is enough that removing encoding entirely along with
548
# the cache (so we are using plain strings) results in no
549
# performance improvement.
550
try:
551
revision_id = unescape_revid_cache[revision_id]
552
except KeyError:
553
unescaped = unescape(revision_id)
554
unescape_revid_cache[revision_id] = unescaped
555
revision_id = unescaped
556
557
if revision_id in selected_revision_ids:
558
try:
559
file_id = unescape_fileid_cache[file_id]
560
except KeyError:
561
unescaped = unescape(file_id)
562
unescape_fileid_cache[file_id] = unescaped
563
file_id = unescaped
564
setdefault(file_id, set()).add(revision_id)
565
finally:
566
pb.finished()
567
return result
568
569
@needs_read_lock
570
def get_inventory_weave(self):
571
return self.control_weaves.get_weave('inventory',
572
self.get_transaction())
2129
573
2130
574
@needs_read_lock
2131
575
def get_inventory(self, revision_id):
2132
"""Get Inventory object by revision id."""
2133
return self.iter_inventories([revision_id]).next()
2134
2135
def iter_inventories(self, revision_ids):
2136
"""Get many inventories by revision_ids.
2137
2138
This will buffer some or all of the texts used in constructing the
2139
inventories in memory, but will only parse a single inventory at a
2140
time.
2141
2142
:param revision_ids: The expected revision ids of the inventories.
2143
:return: An iterator of inventories.
2144
"""
2145
if ((None in revision_ids)
2146
or (_mod_revision.NULL_REVISION in revision_ids)):
2147
raise ValueError('cannot get null revision inventory')
2148
return self._iter_inventories(revision_ids)
2149
2150
def _iter_inventories(self, revision_ids):
2151
"""single-document based inventory iteration."""
2152
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2153
yield self.deserialise_inventory(revision_id, text)
2154
2155
def _iter_inventory_xmls(self, revision_ids):
2156
keys = [(revision_id,) for revision_id in revision_ids]
2157
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2158
text_chunks = {}
2159
for record in stream:
2160
if record.storage_kind != 'absent':
2161
text_chunks[record.key] = record.get_bytes_as('chunked')
2162
else:
2163
raise errors.NoSuchRevision(self, record.key)
2164
for key in keys:
2165
chunks = text_chunks.pop(key)
2166
yield ''.join(chunks), key[-1]
576
"""Get Inventory object by hash."""
577
# TODO: jam 20070210 Technically we don't need to sanitize, since all
578
# called functions must sanitize.
579
revision_id = osutils.safe_revision_id(revision_id)
580
return self.deserialise_inventory(
581
revision_id, self.get_inventory_xml(revision_id))
2167
582
2168
583
def deserialise_inventory(self, revision_id, xml):
2169
"""Transform the xml into an inventory object.
584
"""Transform the xml into an inventory object.
2170
585
2171
586
:param revision_id: The expected revision id of the inventory.
2172
587
:param xml: A serialised inventory.
2173
588
"""
2174
result = self._serializer.read_inventory_from_string(xml, revision_id,
2175
entry_cache=self._inventory_entry_cache)
2176
if result.revision_id != revision_id:
2177
raise AssertionError('revision id mismatch %s != %s' % (
2178
result.revision_id, revision_id))
589
revision_id = osutils.safe_revision_id(revision_id)
590
result = self._serializer.read_inventory_from_string(xml)
591
result.root.revision = revision_id
2179
592
return result
2180
593
2181
594
def serialise_inventory(self, inv):
2182
595
return self._serializer.write_inventory_to_string(inv)
2183
596
2184
def _serialise_inventory_to_lines(self, inv):
2185
return self._serializer.write_inventory_to_lines(inv)
2186
2187
def get_serializer_format(self):
2188
return self._serializer.format_num
2189
2190
597
@needs_read_lock
2191
598
def get_inventory_xml(self, revision_id):
2192
599
"""Get inventory XML as a file object."""
2193
texts = self._iter_inventory_xmls([revision_id])
600
revision_id = osutils.safe_revision_id(revision_id)
2194
601
try:
2195
text, revision_id = texts.next()
2196
except StopIteration:
602
assert isinstance(revision_id, str), type(revision_id)
603
iw = self.get_inventory_weave()
604
return iw.get_text(revision_id)
605
except IndexError:
2197
606
raise errors.HistoryMissing(self, 'inventory', revision_id)
2198
return text
2199
607
2200
608
@needs_read_lock
2201
609
def get_inventory_sha1(self, revision_id):
2202
610
"""Return the sha1 hash of the inventory entry
2203
611
"""
612
# TODO: jam 20070210 Shouldn't this be deprecated / removed?
613
revision_id = osutils.safe_revision_id(revision_id)
2204
614
return self.get_revision(revision_id).inventory_sha1
2205
615
616
@needs_read_lock
617
def get_revision_graph(self, revision_id=None):
618
"""Return a dictionary containing the revision graph.
619
620
:param revision_id: The revision_id to get a graph from. If None, then
621
the entire revision graph is returned. This is a deprecated mode of
622
operation and will be removed in the future.
623
:return: a dictionary of revision_id->revision_parents_list.
624
"""
625
# special case NULL_REVISION
626
if revision_id == _mod_revision.NULL_REVISION:
627
return {}
628
revision_id = osutils.safe_revision_id(revision_id)
629
a_weave = self.get_inventory_weave()
630
all_revisions = self._eliminate_revisions_not_present(
631
a_weave.versions())
632
entire_graph = dict([(node, a_weave.get_parents(node)) for
633
node in all_revisions])
634
if revision_id is None:
635
return entire_graph
636
elif revision_id not in entire_graph:
637
raise errors.NoSuchRevision(self, revision_id)
638
else:
639
# add what can be reached from revision_id
640
result = {}
641
pending = set([revision_id])
642
while len(pending) > 0:
643
node = pending.pop()
644
result[node] = entire_graph[node]
645
for revision_id in result[node]:
646
if revision_id not in result:
647
pending.add(revision_id)
648
return result
649
650
@needs_read_lock
651
def get_revision_graph_with_ghosts(self, revision_ids=None):
652
"""Return a graph of the revisions with ghosts marked as applicable.
653
654
:param revision_ids: an iterable of revisions to graph or None for all.
655
:return: a Graph object with the graph reachable from revision_ids.
656
"""
657
result = graph.Graph()
658
if not revision_ids:
659
pending = set(self.all_revision_ids())
660
required = set([])
661
else:
662
pending = set(osutils.safe_revision_id(r) for r in revision_ids)
663
# special case NULL_REVISION
664
if _mod_revision.NULL_REVISION in pending:
665
pending.remove(_mod_revision.NULL_REVISION)
666
required = set(pending)
667
done = set([])
668
while len(pending):
669
revision_id = pending.pop()
670
try:
671
rev = self.get_revision(revision_id)
672
except errors.NoSuchRevision:
673
if revision_id in required:
674
raise
675
# a ghost
676
result.add_ghost(revision_id)
677
continue
678
for parent_id in rev.parent_ids:
679
# is this queued or done ?
680
if (parent_id not in pending and
681
parent_id not in done):
682
# no, queue it.
683
pending.add(parent_id)
684
result.add_node(revision_id, rev.parent_ids)
685
done.add(revision_id)
686
return result
687
688
def _get_history_vf(self):
689
"""Get a versionedfile whose history graph reflects all revisions.
690
691
For weave repositories, this is the inventory weave.
692
"""
693
return self.get_inventory_weave()
694
2206
695
def iter_reverse_revision_history(self, revision_id):
2207
696
"""Iterate backwards through revision ids in the lefthand history
2208
697
2209
698
:param revision_id: The revision id to start with. All its lefthand
2210
699
ancestors will be traversed.
2211
700
"""
2212
graph = self.get_graph()
701
revision_id = osutils.safe_revision_id(revision_id)
702
if revision_id in (None, _mod_revision.NULL_REVISION):
703
return
2213
704
next_id = revision_id
705
versionedfile = self._get_history_vf()
2214
706
while True:
2215
if next_id in (None, _mod_revision.NULL_REVISION):
2216
return
2217
707
yield next_id
2218
# Note: The following line may raise KeyError in the event of
2219
# truncated history. We decided not to have a try:except:raise
2220
# RevisionNotPresent here until we see a use for it, because of the
2221
# cost in an inner loop that is by its very nature O(history).
2222
# Robert Collins 20080326
2223
parents = graph.get_parent_map([next_id])[next_id]
708
parents = versionedfile.get_parents(next_id)
2224
709
if len(parents) == 0:
2225
710
return
2226
711
else:
2242
727
else:
2243
728
return self.get_inventory(revision_id)
2244
729
730
@needs_read_lock
2245
731
def is_shared(self):
2246
732
"""Return True if this repository is flagged as a shared repository."""
2247
733
raise NotImplementedError(self.is_shared)
2253
739
reconciler = RepoReconciler(self, thorough=thorough)
2254
740
reconciler.reconcile()
2255
741
return reconciler
2256
2257
def _refresh_data(self):
2258
"""Helper called from lock_* to ensure coherency with disk.
2259
2260
The default implementation does nothing; it is however possible
2261
for repositories to maintain loaded indices across multiple locks
2262
by checking inside their implementation of this method to see
2263
whether their indices are still valid. This depends of course on
2264
the disk format being validatable in this manner. This method is
2265
also called by the refresh_data() public interface to cause a refresh
2266
to occur while in a write lock so that data inserted by a smart server
2267
push operation is visible on the client's instance of the physical
2268
repository.
2269
"""
2270
742
2271
743
@needs_read_lock
2272
744
def revision_tree(self, revision_id):
2273
745
"""Return Tree for a revision on this branch.
2274
746
2275
`revision_id` may be NULL_REVISION for the empty tree revision.
747
`revision_id` may be None for the empty tree revision.
2276
748
"""
2277
revision_id = _mod_revision.ensure_null(revision_id)
2278
749
# TODO: refactor this to use an existing revision object
2279
750
# so we don't need to read it in twice.
2280
if revision_id == _mod_revision.NULL_REVISION:
2281
return RevisionTree(self, Inventory(root_id=None),
751
if revision_id is None or revision_id == _mod_revision.NULL_REVISION:
752
return RevisionTree(self, Inventory(root_id=None),
2282
753
_mod_revision.NULL_REVISION)
2283
754
else:
755
revision_id = osutils.safe_revision_id(revision_id)
2284
756
inv = self.get_revision_inventory(revision_id)
2285
757
return RevisionTree(self, inv, revision_id)
2286
758
759
@needs_read_lock
2287
760
def revision_trees(self, revision_ids):
2288
"""Return Trees for revisions in this repository.
2289
2290
:param revision_ids: a sequence of revision-ids;
2291
a revision-id may not be None or 'null:'
2292
"""
2293
inventories = self.iter_inventories(revision_ids)
2294
for inv in inventories:
2295
yield RevisionTree(self, inv, inv.revision_id)
2296
2297
def _filtered_revision_trees(self, revision_ids, file_ids):
2298
"""Return Tree for a revision on this branch with only some files.
2299
2300
:param revision_ids: a sequence of revision-ids;
2301
a revision-id may not be None or 'null:'
2302
:param file_ids: if not None, the result is filtered
2303
so that only those file-ids, their parents and their
2304
children are included.
2305
"""
2306
inventories = self.iter_inventories(revision_ids)
2307
for inv in inventories:
2308
# Should we introduce a FilteredRevisionTree class rather
2309
# than pre-filter the inventory here?
2310
filtered_inv = inv.filter(file_ids)
2311
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
761
"""Return Tree for a revision on this branch.
762
763
`revision_id` may not be None or 'null:'"""
764
assert None not in revision_ids
765
assert _mod_revision.NULL_REVISION not in revision_ids
766
texts = self.get_inventory_weave().get_texts(revision_ids)
767
for text, revision_id in zip(texts, revision_ids):
768
inv = self.deserialise_inventory(revision_id, text)
769
yield RevisionTree(self, inv, revision_id)
2312
770
2313
771
@needs_read_lock
2314
def get_ancestry(self, revision_id, topo_sorted=True):
772
def get_ancestry(self, revision_id):
2315
773
"""Return a list of revision-ids integrated by a revision.
2316
774
2317
The first element of the list is always None, indicating the origin
2318
revision. This might change when we have history horizons, or
775
The first element of the list is always None, indicating the origin
776
revision. This might change when we have history horizons, or
2319
777
perhaps we should have a new API.
2320
778
2321
779
This is topologically sorted.
2322
780
"""
2323
if _mod_revision.is_null(revision_id):
781
if revision_id is None:
2324
782
return [None]
783
revision_id = osutils.safe_revision_id(revision_id)
2325
784
if not self.has_revision(revision_id):
2326
785
raise errors.NoSuchRevision(self, revision_id)
2327
graph = self.get_graph()
2328
keys = set()
2329
search = graph._make_breadth_first_searcher([revision_id])
2330
while True:
2331
try:
2332
found, ghosts = search.next_with_ghosts()
2333
except StopIteration:
2334
break
2335
keys.update(found)
2336
if _mod_revision.NULL_REVISION in keys:
2337
keys.remove(_mod_revision.NULL_REVISION)
2338
if topo_sorted:
2339
parent_map = graph.get_parent_map(keys)
2340
keys = tsort.topo_sort(parent_map)
2341
return [None] + list(keys)
2342
2343
def pack(self):
2344
"""Compress the data within the repository.
2345
2346
This operation only makes sense for some repository types. For other
2347
types it should be a no-op that just returns.
2348
2349
This stub method does not require a lock, but subclasses should use
2350
@needs_write_lock as this is a long running call its reasonable to
2351
implicitly lock for the user.
786
w = self.get_inventory_weave()
787
candidates = w.get_ancestry(revision_id)
788
return [None] + candidates # self._eliminate_revisions_not_present(candidates)
789
790
@needs_read_lock
791
def print_file(self, file, revision_id):
792
"""Print `file` to stdout.
793
794
FIXME RBC 20060125 as John Meinel points out this is a bad api
795
- it writes to stdout, it assumes that that is valid etc. Fix
796
by creating a new more flexible convenience function.
2352
797
"""
798
revision_id = osutils.safe_revision_id(revision_id)
799
tree = self.revision_tree(revision_id)
800
# use inventory as it was in that revision
801
file_id = tree.inventory.path2id(file)
802
if not file_id:
803
# TODO: jam 20060427 Write a test for this code path
804
# it had a bug in it, and was raising the wrong
805
# exception.
806
raise errors.BzrError("%r is not present in revision %s" % (file, revision_id))
807
tree.print_file(file_id)
2353
808
2354
809
def get_transaction(self):
2355
810
return self.control_files.get_transaction()
2356
811
2357
def get_parent_map(self, revision_ids):
2358
"""See graph._StackedParentsProvider.get_parent_map"""
2359
# revisions index works in keys; this just works in revisions
2360
# therefore wrap and unwrap
2361
query_keys = []
2362
result = {}
2363
for revision_id in revision_ids:
2364
if revision_id == _mod_revision.NULL_REVISION:
2365
result[revision_id] = ()
2366
elif revision_id is None:
2367
raise ValueError('get_parent_map(None) is not valid')
2368
else:
2369
query_keys.append((revision_id ,))
2370
for ((revision_id,), parent_keys) in \
2371
self.revisions.get_parent_map(query_keys).iteritems():
2372
if parent_keys:
2373
result[revision_id] = tuple(parent_revid
2374
for (parent_revid,) in parent_keys)
2375
else:
2376
result[revision_id] = (_mod_revision.NULL_REVISION,)
2377
return result
2378
2379
def _make_parents_provider(self):
2380
return self
2381
2382
def get_graph(self, other_repository=None):
2383
"""Return the graph walker for this repository format"""
2384
parents_provider = self._make_parents_provider()
2385
if (other_repository is not None and
2386
not self.has_same_location(other_repository)):
2387
parents_provider = graph._StackedParentsProvider(
2388
[parents_provider, other_repository._make_parents_provider()])
2389
return graph.Graph(parents_provider)
2390
2391
def _get_versioned_file_checker(self, text_key_references=None):
2392
"""Return an object suitable for checking versioned files.
2393
2394
:param text_key_references: if non-None, an already built
2395
dictionary mapping text keys ((fileid, revision_id) tuples)
2396
to whether they were referred to by the inventory of the
2397
revision_id that they contain. If None, this will be
2398
calculated.
2399
"""
2400
return _VersionedFileChecker(self,
2401
text_key_references=text_key_references)
2402
2403
def revision_ids_to_search_result(self, result_set):
2404
"""Convert a set of revision ids to a graph SearchResult."""
2405
result_parents = set()
2406
for parents in self.get_graph().get_parent_map(
2407
result_set).itervalues():
2408
result_parents.update(parents)
2409
included_keys = result_set.intersection(result_parents)
2410
start_keys = result_set.difference(included_keys)
2411
exclude_keys = result_parents.difference(result_set)
2412
result = graph.SearchResult(start_keys, exclude_keys,
2413
len(result_set), result_set)
2414
return result
812
def revision_parents(self, revision_id):
813
revision_id = osutils.safe_revision_id(revision_id)
814
return self.get_inventory_weave().parent_names(revision_id)
2415
815
2416
816
@needs_write_lock
2417
817
def set_make_working_trees(self, new_value):
2424
824
working trees.
2425
825
"""
2426
826
raise NotImplementedError(self.set_make_working_trees)
2427
827
2428
828
def make_working_trees(self):
2429
829
"""Returns the policy for making working trees on new branches."""
2430
830
raise NotImplementedError(self.make_working_trees)
2431
831
2432
832
@needs_write_lock
2433
833
def sign_revision(self, revision_id, gpg_strategy):
834
revision_id = osutils.safe_revision_id(revision_id)
2434
835
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2435
836
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2436
837
2437
838
@needs_read_lock
2438
839
def has_signature_for_revision_id(self, revision_id):
2439
840
"""Query for a revision signature for revision_id in the repository."""
2440
if not self.has_revision(revision_id):
2441
raise errors.NoSuchRevision(self, revision_id)
2442
sig_present = (1 == len(
2443
self.signatures.get_parent_map([(revision_id,)])))
2444
return sig_present
841
revision_id = osutils.safe_revision_id(revision_id)
842
return self._revision_store.has_signature(revision_id,
843
self.get_transaction())
2445
844
2446
845
@needs_read_lock
2447
846
def get_signature_text(self, revision_id):
2448
847
"""Return the text for a signature."""
2449
stream = self.signatures.get_record_stream([(revision_id,)],
2450
'unordered', True)
2451
record = stream.next()
2452
if record.storage_kind == 'absent':
2453
raise errors.NoSuchRevision(self, revision_id)
2454
return record.get_bytes_as('fulltext')
848
revision_id = osutils.safe_revision_id(revision_id)
849
return self._revision_store.get_signature_text(revision_id,
850
self.get_transaction())
2455
851
2456
852
@needs_read_lock
2457
def check(self, revision_ids=None):
853
def check(self, revision_ids):
2458
854
"""Check consistency of all history of given revision_ids.
2459
855
2460
856
Different repository implementations should override _check().
2462
858
:param revision_ids: A non-empty list of revision_ids whose ancestry
2463
859
will be checked. Typically the last revision_id of a branch.
2464
860
"""
861
if not revision_ids:
862
raise ValueError("revision_ids must be non-empty in %s.check"
863
% (self,))
864
revision_ids = [osutils.safe_revision_id(r) for r in revision_ids]
2465
865
return self._check(revision_ids)
2466
866
2467
867
def _check(self, revision_ids):
2496
896
except UnicodeDecodeError:
2497
897
raise errors.NonAsciiRevisionId(method, self)
2498
898
2499
def revision_graph_can_have_wrong_parents(self):
2500
"""Is it possible for this repository to have a revision graph with
2501
incorrect parents?
2502
2503
If True, then this repository must also implement
2504
_find_inconsistent_revision_parents so that check and reconcile can
2505
check for inconsistencies before proceeding with other checks that may
2506
depend on the revision index being consistent.
2507
"""
2508
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2509
899
2510
900
2511
901
# remove these delegates a while after bzr 0.15
2544
934
2545
935
def install_revision(repository, rev, revision_tree):
2546
936
"""Install all revision data into a repository."""
2547
install_revisions(repository, [(rev, revision_tree, None)])
2548
2549
2550
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2551
"""Install all revision data into a repository.
2552
2553
Accepts an iterable of revision, tree, signature tuples. The signature
2554
may be None.
2555
"""
2556
repository.start_write_group()
2557
try:
2558
inventory_cache = lru_cache.LRUCache(10)
2559
for n, (revision, revision_tree, signature) in enumerate(iterable):
2560
_install_revision(repository, revision, revision_tree, signature,
2561
inventory_cache)
2562
if pb is not None:
2563
pb.update('Transferring revisions', n + 1, num_revisions)
2564
except:
2565
repository.abort_write_group()
2566
raise
2567
else:
2568
repository.commit_write_group()
2569
2570
2571
def _install_revision(repository, rev, revision_tree, signature,
2572
inventory_cache):
2573
"""Install all revision data into a repository."""
2574
937
present_parents = []
2575
938
parent_trees = {}
2576
939
for p_id in rev.parent_ids:
2578
941
present_parents.append(p_id)
2579
942
parent_trees[p_id] = repository.revision_tree(p_id)
2580
943
else:
2581
parent_trees[p_id] = repository.revision_tree(
2582
_mod_revision.NULL_REVISION)
944
parent_trees[p_id] = repository.revision_tree(None)
2583
945
2584
946
inv = revision_tree.inventory
2585
947
entries = inv.iter_entries()
2586
# backwards compatibility hack: skip the root id.
948
# backwards compatability hack: skip the root id.
2587
949
if not repository.supports_rich_root():
2588
950
path, root = entries.next()
2589
951
if root.revision != rev.revision_id:
2590
952
raise errors.IncompatibleRevision(repr(repository))
2591
text_keys = {}
953
# Add the texts that are not already present
2592
954
for path, ie in entries:
2593
text_keys[(ie.file_id, ie.revision)] = ie
2594
text_parent_map = repository.texts.get_parent_map(text_keys)
2595
missing_texts = set(text_keys) - set(text_parent_map)
2596
# Add the texts that are not already present
2597
for text_key in missing_texts:
2598
ie = text_keys[text_key]
2599
text_parents = []
2600
# FIXME: TODO: The following loop overlaps/duplicates that done by
2601
# commit to determine parents. There is a latent/real bug here where
2602
# the parents inserted are not those commit would do - in particular
2603
# they are not filtered by heads(). RBC, AB
2604
for revision, tree in parent_trees.iteritems():
2605
if ie.file_id not in tree:
2606
continue
2607
parent_id = tree.inventory[ie.file_id].revision
2608
if parent_id in text_parents:
2609
continue
2610
text_parents.append((ie.file_id, parent_id))
2611
lines = revision_tree.get_file(ie.file_id).readlines()
2612
repository.texts.add_lines(text_key, text_parents, lines)
955
w = repository.weave_store.get_weave_or_empty(ie.file_id,
956
repository.get_transaction())
957
if ie.revision not in w:
958
text_parents = []
959
# FIXME: TODO: The following loop *may* be overlapping/duplicate
960
# with InventoryEntry.find_previous_heads(). if it is, then there
961
# is a latent bug here where the parents may have ancestors of each
962
# other. RBC, AB
963
for revision, tree in parent_trees.iteritems():
964
if ie.file_id not in tree:
965
continue
966
parent_id = tree.inventory[ie.file_id].revision
967
if parent_id in text_parents:
968
continue
969
text_parents.append(parent_id)
970
971
vfile = repository.weave_store.get_weave_or_empty(ie.file_id,
972
repository.get_transaction())
973
lines = revision_tree.get_file(ie.file_id).readlines()
974
vfile.add_lines(rev.revision_id, text_parents, lines)
2613
975
try:
2614
976
# install the inventory
2615
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2616
# Cache this inventory
2617
inventory_cache[rev.revision_id] = inv
2618
try:
2619
basis_inv = inventory_cache[rev.parent_ids[0]]
2620
except KeyError:
2621
repository.add_inventory(rev.revision_id, inv, present_parents)
2622
else:
2623
delta = inv._make_delta(basis_inv)
2624
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2625
rev.revision_id, present_parents)
2626
else:
2627
repository.add_inventory(rev.revision_id, inv, present_parents)
977
repository.add_inventory(rev.revision_id, inv, present_parents)
2628
978
except errors.RevisionAlreadyPresent:
2629
979
pass
2630
if signature is not None:
2631
repository.add_signature_text(rev.revision_id, signature)
2632
980
repository.add_revision(rev.revision_id, rev, inv)
2633
981
2634
982
2635
983
class MetaDirRepository(Repository):
2636
"""Repositories in the new meta-dir layout.
2637
2638
:ivar _transport: Transport for access to repository control files,
2639
typically pointing to .bzr/repository.
2640
"""
2641
2642
def __init__(self, _format, a_bzrdir, control_files):
2643
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2644
self._transport = control_files._transport
2645
984
"""Repositories in the new meta-dir layout."""
985
986
def __init__(self, _format, a_bzrdir, control_files, _revision_store, control_store, text_store):
987
super(MetaDirRepository, self).__init__(_format,
988
a_bzrdir,
989
control_files,
990
_revision_store,
991
control_store,
992
text_store)
993
dir_mode = self.control_files._dir_mode
994
file_mode = self.control_files._file_mode
995
996
@needs_read_lock
2646
997
def is_shared(self):
2647
998
"""Return True if this repository is flagged as a shared repository."""
2648
return self._transport.has('shared-storage')
999
return self.control_files._transport.has('shared-storage')
2649
1000
2650
1001
@needs_write_lock
2651
1002
def set_make_working_trees(self, new_value):
2659
1010
"""
2660
1011
if new_value:
2661
1012
try:
2662
self._transport.delete('no-working-trees')
1013
self.control_files._transport.delete('no-working-trees')
2663
1014
except errors.NoSuchFile:
2664
1015
pass
2665
1016
else:
2666
self._transport.put_bytes('no-working-trees', '',
2667
mode=self.bzrdir._get_file_mode())
2668
1017
self.control_files.put_utf8('no-working-trees', '')
1018
2669
1019
def make_working_trees(self):
2670
1020
"""Returns the policy for making working trees on new branches."""
2671
return not self._transport.has('no-working-trees')
2672
2673
2674
class MetaDirVersionedFileRepository(MetaDirRepository):
2675
"""Repositories in a meta-dir, that work via versioned file objects."""
2676
2677
def __init__(self, _format, a_bzrdir, control_files):
2678
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2679
control_files)
2680
2681
2682
network_format_registry = registry.FormatRegistry()
2683
"""Registry of formats indexed by their network name.
2684
2685
The network name for a repository format is an identifier that can be used when
2686
referring to formats with smart server operations. See
2687
RepositoryFormat.network_name() for more detail.
2688
"""
2689
2690
2691
format_registry = registry.FormatRegistry(network_format_registry)
2692
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
1021
return not self.control_files._transport.has('no-working-trees')
1022
1023
1024
class RepositoryFormatRegistry(registry.Registry):
1025
"""Registry of RepositoryFormats.
1026
"""
1027
1028
def get(self, format_string):
1029
r = registry.Registry.get(self, format_string)
1030
if callable(r):
1031
r = r()
1032
return r
1033
1034
1035
format_registry = RepositoryFormatRegistry()
1036
"""Registry of formats, indexed by their identifying format string.
2693
1037
2694
1038
This can contain either format instances themselves, or classes/factories that
2695
1039
can be called to obtain one.
2702
1046
class RepositoryFormat(object):
2703
1047
"""A repository format.
2704
1048
2705
Formats provide four things:
1049
Formats provide three things:
2706
1050
* An initialization routine to construct repository data on disk.
2707
* a optional format string which is used when the BzrDir supports
2708
versioned children.
1051
* a format string which is used when the BzrDir supports versioned
1052
children.
2709
1053
* an open routine which returns a Repository instance.
2710
* A network name for referring to the format in smart server RPC
2711
methods.
2712
2713
There is one and only one Format subclass for each on-disk format. But
2714
there can be one Repository subclass that is used for several different
2715
formats. The _format attribute on a Repository instance can be used to
2716
determine the disk format.
2717
2718
Formats are placed in a registry by their format string for reference
2719
during opening. These should be subclasses of RepositoryFormat for
2720
consistency.
1054
1055
Formats are placed in an dict by their format string for reference
1056
during opening. These should be subclasses of RepositoryFormat
1057
for consistency.
2721
1058
2722
1059
Once a format is deprecated, just deprecate the initialize and open
2723
methods on the format class. Do not deprecate the object, as the
2724
object may be created even when a repository instance hasn't been
2725
created.
1060
methods on the format class. Do not deprecate the object, as the
1061
object will be created every system load.
2726
1062
2727
1063
Common instance attributes:
2728
1064
_matchingbzrdir - the bzrdir format that the repository format was
2729
1065
originally written to work with. This can be used if manually
2730
1066
constructing a bzrdir and repository, or more commonly for test suite
2731
parameterization.
1067
parameterisation.
2732
1068
"""
2733
1069
2734
# Set to True or False in derived classes. True indicates that the format
2735
# supports ghosts gracefully.
2736
supports_ghosts = None
2737
# Can this repository be given external locations to lookup additional
2738
# data. Set to True or False in derived classes.
2739
supports_external_lookups = None
2740
# Does this format support CHK bytestring lookups. Set to True or False in
2741
# derived classes.
2742
supports_chks = None
2743
# Should commit add an inventory, or an inventory delta to the repository.
2744
_commit_inv_deltas = True
2745
# What order should fetch operations request streams in?
2746
# The default is unordered as that is the cheapest for an origin to
2747
# provide.
2748
_fetch_order = 'unordered'
2749
# Does this repository format use deltas that can be fetched as-deltas ?
2750
# (E.g. knits, where the knit deltas can be transplanted intact.
2751
# We default to False, which will ensure that enough data to get
2752
# a full text out of any fetch stream will be grabbed.
2753
_fetch_uses_deltas = False
2754
# Should fetch trigger a reconcile after the fetch? Only needed for
2755
# some repository formats that can suffer internal inconsistencies.
2756
_fetch_reconcile = False
2757
# Does this format have < O(tree_size) delta generation. Used to hint what
2758
# code path for commit, amongst other things.
2759
fast_deltas = None
2760
2761
1070
def __str__(self):
2762
1071
return "<%s>" % self.__class__.__name__
2763
1072
2771
1080
@classmethod
2772
1081
def find_format(klass, a_bzrdir):
2773
1082
"""Return the format for the repository object in a_bzrdir.
2774
1083
2775
1084
This is used by bzr native formats that have a "format" file in
2776
the repository. Other methods may be used by different types of
1085
the repository. Other methods may be used by different types of
2777
1086
control directory.
2778
1087
"""
2779
1088
try:
2783
1092
except errors.NoSuchFile:
2784
1093
raise errors.NoRepositoryPresent(a_bzrdir)
2785
1094
except KeyError:
2786
raise errors.UnknownFormatError(format=format_string,
2787
kind='repository')
1095
raise errors.UnknownFormatError(format=format_string)
2788
1096
2789
1097
@classmethod
2790
1098
def register_format(klass, format):
2793
1101
@classmethod
2794
1102
def unregister_format(klass, format):
2795
1103
format_registry.remove(format.get_format_string())
2796
1104
2797
1105
@classmethod
2798
1106
def get_default_format(klass):
2799
1107
"""Return the current default format."""
2800
1108
from bzrlib import bzrdir
2801
1109
return bzrdir.format_registry.make_bzrdir('default').repository_format
2802
1110
1111
def _get_control_store(self, repo_transport, control_files):
1112
"""Return the control store for this repository."""
1113
raise NotImplementedError(self._get_control_store)
1114
2803
1115
def get_format_string(self):
2804
1116
"""Return the ASCII format string that identifies this format.
2805
2806
Note that in pre format ?? repositories the format string is
1117
1118
Note that in pre format ?? repositories the format string is
2807
1119
not permitted nor written to disk.
2808
1120
"""
2809
1121
raise NotImplementedError(self.get_format_string)
2812
1124
"""Return the short description for this format."""
2813
1125
raise NotImplementedError(self.get_format_description)
2814
1126
1127
def _get_revision_store(self, repo_transport, control_files):
1128
"""Return the revision store object for this a_bzrdir."""
1129
raise NotImplementedError(self._get_revision_store)
1130
1131
def _get_text_rev_store(self,
1132
transport,
1133
control_files,
1134
name,
1135
compressed=True,
1136
prefixed=False,
1137
serializer=None):
1138
"""Common logic for getting a revision store for a repository.
1139
1140
see self._get_revision_store for the subclass-overridable method to
1141
get the store for a repository.
1142
"""
1143
from bzrlib.store.revision.text import TextRevisionStore
1144
dir_mode = control_files._dir_mode
1145
file_mode = control_files._file_mode
1146
text_store = TextStore(transport.clone(name),
1147
prefixed=prefixed,
1148
compressed=compressed,
1149
dir_mode=dir_mode,
1150
file_mode=file_mode)
1151
_revision_store = TextRevisionStore(text_store, serializer)
1152
return _revision_store
1153
2815
1154
# TODO: this shouldn't be in the base class, it's specific to things that
2816
1155
# use weaves or knits -- mbp 20070207
2817
1156
def _get_versioned_file_store(self,
2839
1178
2840
1179
:param a_bzrdir: The bzrdir to put the new repository in it.
2841
1180
:param shared: The repository should be initialized as a sharable one.
2842
:returns: The new repository object.
2843
1181
2844
1182
This may raise UninitializableFormat if shared repository are not
2845
1183
compatible the a_bzrdir.
2846
1184
"""
2847
raise NotImplementedError(self.initialize)
2848
1185
2849
1186
def is_supported(self):
2850
1187
"""Is this format supported?
2851
1188
2852
1189
Supported formats must be initializable and openable.
2853
Unsupported formats may not support initialization or committing or
1190
Unsupported formats may not support initialization or committing or
2854
1191
some other features depending on the reason for not being supported.
2855
1192
"""
2856
1193
return True
2857
1194
2858
def network_name(self):
2859
"""A simple byte string uniquely identifying this format for RPC calls.
2860
2861
MetaDir repository formats use their disk format string to identify the
2862
repository over the wire. All in one formats such as bzr < 0.8, and
2863
foreign formats like svn/git and hg should use some marker which is
2864
unique and immutable.
2865
"""
2866
raise NotImplementedError(self.network_name)
2867
2868
1195
def check_conversion_target(self, target_format):
2869
1196
raise NotImplementedError(self.check_conversion_target)
2870
1197
2871
1198
def open(self, a_bzrdir, _found=False):
2872
1199
"""Return an instance of this format for the bzrdir a_bzrdir.
2873
1200
2874
1201
_found is a private parameter, do not use it.
2875
1202
"""
2876
1203
raise NotImplementedError(self.open)
2881
1208
2882
1209
rich_root_data = False
2883
1210
supports_tree_reference = False
2884
supports_external_lookups = False
2885
2886
@property
2887
def _matchingbzrdir(self):
2888
matching = bzrdir.BzrDirMetaFormat1()
2889
matching.repository_format = self
2890
return matching
1211
_matchingbzrdir = bzrdir.BzrDirMetaFormat1()
2891
1212
2892
1213
def __init__(self):
2893
1214
super(MetaDirRepositoryFormat, self).__init__()
2906
1227
"""Upload the initial blank content."""
2907
1228
control_files = self._create_control_files(a_bzrdir)
2908
1229
control_files.lock_write()
2909
transport = control_files._transport
2910
if shared == True:
2911
utf8_files += [('shared-storage', '')]
2912
1230
try:
2913
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
2914
for (filename, content_stream) in files:
2915
transport.put_file(filename, content_stream,
2916
mode=a_bzrdir._get_file_mode())
2917
for (filename, content_bytes) in utf8_files:
2918
transport.put_bytes_non_atomic(filename, content_bytes,
2919
mode=a_bzrdir._get_file_mode())
1231
control_files._transport.mkdir_multi(dirs,
1232
mode=control_files._dir_mode)
1233
for file, content in files:
1234
control_files.put(file, content)
1235
for file, content in utf8_files:
1236
control_files.put_utf8(file, content)
1237
if shared == True:
1238
control_files.put_utf8('shared-storage', '')
2920
1239
finally:
2921
1240
control_files.unlock()
2922
1241
2923
def network_name(self):
2924
"""Metadir formats have matching disk and network format strings."""
2925
return self.get_format_string()
2926
2927
2928
# Pre-0.8 formats that don't have a disk format string (because they are
2929
# versioned by the matching control directory). We use the control directories
2930
# disk format string as a key for the network_name because they meet the
2931
# constraints (simple string, unique, immutable).
2932
network_format_registry.register_lazy(
2933
"Bazaar-NG branch, format 5\n",
2934
'bzrlib.repofmt.weaverepo',
2935
'RepositoryFormat5',
2936
)
2937
network_format_registry.register_lazy(
2938
"Bazaar-NG branch, format 6\n",
2939
'bzrlib.repofmt.weaverepo',
2940
'RepositoryFormat6',
2941
)
2942
2943
# formats which have no format string are not discoverable or independently
2944
# creatable on disk, so are not registered in format_registry. They're
1242
1243
# formats which have no format string are not discoverable
1244
# and not independently creatable, so are not registered. They're
2945
1245
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
2946
1246
# needed, it's constructed directly by the BzrDir. Non-native formats where
2947
1247
# the repository is not separately opened are similar.
2951
1251
'bzrlib.repofmt.weaverepo',
2952
1252
'RepositoryFormat7'
2953
1253
)
1254
# KEEP in sync with bzrdir.format_registry default, which controls the overall
1255
# default control directory format
2954
1256
2955
1257
format_registry.register_lazy(
2956
1258
'Bazaar-NG Knit Repository Format 1',
2957
1259
'bzrlib.repofmt.knitrepo',
2958
1260
'RepositoryFormatKnit1',
2959
1261
)
1262
format_registry.default_key = 'Bazaar-NG Knit Repository Format 1'
2960
1263
2961
1264
format_registry.register_lazy(
2962
1265
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
2964
1267
'RepositoryFormatKnit3',
2965
1268
)
2966
1269
2967
format_registry.register_lazy(
2968
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
2969
'bzrlib.repofmt.knitrepo',
2970
'RepositoryFormatKnit4',
2971
)
2972
2973
# Pack-based formats. There is one format for pre-subtrees, and one for
2974
# post-subtrees to allow ease of testing.
2975
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
2976
format_registry.register_lazy(
2977
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
2978
'bzrlib.repofmt.pack_repo',
2979
'RepositoryFormatKnitPack1',
2980
)
2981
format_registry.register_lazy(
2982
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
2983
'bzrlib.repofmt.pack_repo',
2984
'RepositoryFormatKnitPack3',
2985
)
2986
format_registry.register_lazy(
2987
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
2988
'bzrlib.repofmt.pack_repo',
2989
'RepositoryFormatKnitPack4',
2990
)
2991
format_registry.register_lazy(
2992
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
2993
'bzrlib.repofmt.pack_repo',
2994
'RepositoryFormatKnitPack5',
2995
)
2996
format_registry.register_lazy(
2997
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
2998
'bzrlib.repofmt.pack_repo',
2999
'RepositoryFormatKnitPack5RichRoot',
3000
)
3001
format_registry.register_lazy(
3002
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3003
'bzrlib.repofmt.pack_repo',
3004
'RepositoryFormatKnitPack5RichRootBroken',
3005
)
3006
format_registry.register_lazy(
3007
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3008
'bzrlib.repofmt.pack_repo',
3009
'RepositoryFormatKnitPack6',
3010
)
3011
format_registry.register_lazy(
3012
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3013
'bzrlib.repofmt.pack_repo',
3014
'RepositoryFormatKnitPack6RichRoot',
3015
)
3016
3017
# Development formats.
3018
# Obsolete but kept pending a CHK based subtree format.
3019
format_registry.register_lazy(
3020
("Bazaar development format 2 with subtree support "
3021
"(needs bzr.dev from before 1.8)\n"),
3022
'bzrlib.repofmt.pack_repo',
3023
'RepositoryFormatPackDevelopment2Subtree',
3024
)
3025
3026
# 1.14->1.16 go below here
3027
format_registry.register_lazy(
3028
'Bazaar development format - group compression and chk inventory'
3029
' (needs bzr.dev from 1.14)\n',
3030
'bzrlib.repofmt.groupcompress_repo',
3031
'RepositoryFormatCHK1',
3032
)
3033
3034
1270
3035
1271
class InterRepository(InterObject):
3036
1272
"""This class represents operations taking place between two repositories.
3037
1273
3038
1274
Its instances have methods like copy_content and fetch, and contain
3039
references to the source and target repositories these operations can be
1275
references to the source and target repositories these operations can be
3040
1276
carried out on.
3041
1277
3042
1278
Often we will provide convenience methods on 'repository' which carry out
3044
1280
InterRepository.get(other).method_name(parameters).
3045
1281
"""
3046
1282
3047
_walk_to_common_revisions_batch_size = 50
3048
1283
_optimisers = []
3049
1284
"""The available optimised InterRepository types."""
3050
1285
3051
@needs_write_lock
3052
1286
def copy_content(self, revision_id=None):
3053
"""Make a complete copy of the content in self into destination.
3054
3055
This is a destructive operation! Do not use it on existing
3056
repositories.
3057
3058
:param revision_id: Only copy the content needed to construct
3059
revision_id and its parents.
3060
"""
3061
try:
3062
self.target.set_make_working_trees(self.source.make_working_trees())
3063
except NotImplementedError:
3064
pass
3065
self.target.fetch(self.source, revision_id=revision_id)
3066
3067
@needs_write_lock
3068
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3069
fetch_spec=None):
1287
raise NotImplementedError(self.copy_content)
1288
1289
def fetch(self, revision_id=None, pb=None):
3070
1290
"""Fetch the content required to construct revision_id.
3071
1291
3072
1292
The content is copied from self.source to self.target.
3075
1295
content is copied.
3076
1296
:param pb: optional progress bar to use for progress reports. If not
3077
1297
provided a default one will be created.
3078
:return: None.
3079
"""
3080
from bzrlib.fetch import RepoFetcher
3081
f = RepoFetcher(to_repository=self.target,
3082
from_repository=self.source,
3083
last_revision=revision_id,
3084
fetch_spec=fetch_spec,
3085
pb=pb, find_ghosts=find_ghosts)
3086
3087
def _walk_to_common_revisions(self, revision_ids):
3088
"""Walk out from revision_ids in source to revisions target has.
3089
3090
:param revision_ids: The start point for the search.
3091
:return: A set of revision ids.
3092
"""
3093
target_graph = self.target.get_graph()
3094
revision_ids = frozenset(revision_ids)
3095
missing_revs = set()
3096
source_graph = self.source.get_graph()
3097
# ensure we don't pay silly lookup costs.
3098
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3099
null_set = frozenset([_mod_revision.NULL_REVISION])
3100
searcher_exhausted = False
3101
while True:
3102
next_revs = set()
3103
ghosts = set()
3104
# Iterate the searcher until we have enough next_revs
3105
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3106
try:
3107
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3108
next_revs.update(next_revs_part)
3109
ghosts.update(ghosts_part)
3110
except StopIteration:
3111
searcher_exhausted = True
3112
break
3113
# If there are ghosts in the source graph, and the caller asked for
3114
# them, make sure that they are present in the target.
3115
# We don't care about other ghosts as we can't fetch them and
3116
# haven't been asked to.
3117
ghosts_to_check = set(revision_ids.intersection(ghosts))
3118
revs_to_get = set(next_revs).union(ghosts_to_check)
3119
if revs_to_get:
3120
have_revs = set(target_graph.get_parent_map(revs_to_get))
3121
# we always have NULL_REVISION present.
3122
have_revs = have_revs.union(null_set)
3123
# Check if the target is missing any ghosts we need.
3124
ghosts_to_check.difference_update(have_revs)
3125
if ghosts_to_check:
3126
# One of the caller's revision_ids is a ghost in both the
3127
# source and the target.
3128
raise errors.NoSuchRevision(
3129
self.source, ghosts_to_check.pop())
3130
missing_revs.update(next_revs - have_revs)
3131
# Because we may have walked past the original stop point, make
3132
# sure everything is stopped
3133
stop_revs = searcher.find_seen_ancestors(have_revs)
3134
searcher.stop_searching_any(stop_revs)
3135
if searcher_exhausted:
3136
break
3137
return searcher.get_result()
3138
1298
1299
Returns the copied revision count and the failed revisions in a tuple:
1300
(copied, failures).
1301
"""
1302
raise NotImplementedError(self.fetch)
1303
3139
1304
@needs_read_lock
3140
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1305
def missing_revision_ids(self, revision_id=None):
3141
1306
"""Return the revision ids that source has that target does not.
1307
1308
These are returned in topological order.
3142
1309
3143
1310
:param revision_id: only return revision ids included by this
3144
1311
revision_id.
3145
:param find_ghosts: If True find missing revisions in deep history
3146
rather than just finding the surface difference.
3147
:return: A bzrlib.graph.SearchResult.
3148
1312
"""
3149
# stop searching at found target revisions.
3150
if not find_ghosts and revision_id is not None:
3151
return self._walk_to_common_revisions([revision_id])
3152
1313
# generic, possibly worst case, slow code path.
3153
1314
target_ids = set(self.target.all_revision_ids())
3154
1315
if revision_id is not None:
1316
# TODO: jam 20070210 InterRepository is internal enough that it
1317
# should assume revision_ids are already utf-8
1318
revision_id = osutils.safe_revision_id(revision_id)
3155
1319
source_ids = self.source.get_ancestry(revision_id)
3156
if source_ids[0] is not None:
3157
raise AssertionError()
1320
assert source_ids[0] is None
3158
1321
source_ids.pop(0)
3159
1322
else:
3160
1323
source_ids = self.source.all_revision_ids()
3161
1324
result_set = set(source_ids).difference(target_ids)
3162
return self.source.revision_ids_to_search_result(result_set)
3163
3164
@staticmethod
3165
def _same_model(source, target):
3166
"""True if source and target have the same data representation.
3167
3168
Note: this is always called on the base class; overriding it in a
3169
subclass will have no effect.
3170
"""
3171
try:
3172
InterRepository._assert_same_model(source, target)
3173
return True
3174
except errors.IncompatibleRepositories, e:
3175
return False
3176
3177
@staticmethod
3178
def _assert_same_model(source, target):
3179
"""Raise an exception if two repositories do not use the same model.
3180
"""
3181
if source.supports_rich_root() != target.supports_rich_root():
3182
raise errors.IncompatibleRepositories(source, target,
3183
"different rich-root support")
3184
if source._serializer != target._serializer:
3185
raise errors.IncompatibleRepositories(source, target,
3186
"different serializers")
1325
# this may look like a no-op: its not. It preserves the ordering
1326
# other_ids had while only returning the members from other_ids
1327
# that we've decided we need.
1328
return [rev_id for rev_id in source_ids if rev_id in result_set]
3187
1329
3188
1330
3189
1331
class InterSameDataRepository(InterRepository):
3190
1332
"""Code for converting between repositories that represent the same data.
3191
1333
3192
1334
Data format and model must match for this to work.
3193
1335
"""
3194
1336
3195
1337
@classmethod
3196
1338
def _get_repo_format_to_test(self):
3197
"""Repository format for testing with.
3198
3199
InterSameData can pull from subtree to subtree and from non-subtree to
3200
non-subtree, so we test this with the richest repository format.
3201
"""
3202
from bzrlib.repofmt import knitrepo
3203
return knitrepo.RepositoryFormatKnit3()
1339
"""Repository format for testing with."""
1340
return RepositoryFormat.get_default_format()
3204
1341
3205
1342
@staticmethod
3206
1343
def is_compatible(source, target):
3207
return InterRepository._same_model(source, target)
1344
if not isinstance(source, Repository):
1345
return False
1346
if not isinstance(target, Repository):
1347
return False
1348
if source._format.rich_root_data != target._format.rich_root_data:
1349
return False
1350
if source._serializer != target._serializer:
1351
return False
1352
else:
1353
return True
1354
1355
@needs_write_lock
1356
def copy_content(self, revision_id=None):
1357
"""Make a complete copy of the content in self into destination.
1358
1359
This is a destructive operation! Do not use it on existing
1360
repositories.
1361
1362
:param revision_id: Only copy the content needed to construct
1363
revision_id and its parents.
1364
"""
1365
try:
1366
self.target.set_make_working_trees(self.source.make_working_trees())
1367
except NotImplementedError:
1368
pass
1369
# TODO: jam 20070210 This is fairly internal, so we should probably
1370
# just assert that revision_id is not unicode.
1371
revision_id = osutils.safe_revision_id(revision_id)
1372
# but don't bother fetching if we have the needed data now.
1373
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1374
self.target.has_revision(revision_id)):
1375
return
1376
self.target.fetch(self.source, revision_id=revision_id)
1377
1378
@needs_write_lock
1379
def fetch(self, revision_id=None, pb=None):
1380
"""See InterRepository.fetch()."""
1381
from bzrlib.fetch import GenericRepoFetcher
1382
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1383
self.source, self.source._format, self.target,
1384
self.target._format)
1385
# TODO: jam 20070210 This should be an assert, not a translate
1386
revision_id = osutils.safe_revision_id(revision_id)
1387
f = GenericRepoFetcher(to_repository=self.target,
1388
from_repository=self.source,
1389
last_revision=revision_id,
1390
pb=pb)
1391
return f.count_copied, f.failed_revisions
3208
1392
3209
1393
3210
1394
class InterWeaveRepo(InterSameDataRepository):
3211
"""Optimised code paths between Weave based repositories.
3212
3213
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3214
implemented lazy inter-object optimisation.
3215
"""
1395
"""Optimised code paths between Weave based repositories."""
3216
1396
3217
1397
@classmethod
3218
1398
def _get_repo_format_to_test(self):
3222
1402
@staticmethod
3223
1403
def is_compatible(source, target):
3224
1404
"""Be compatible with known Weave formats.
3225
1405
3226
1406
We don't test for the stores being of specific types because that
3227
could lead to confusing results, and there is no need to be
1407
could lead to confusing results, and there is no need to be
3228
1408
overly general.
3229
1409
"""
3230
1410
from bzrlib.repofmt.weaverepo import (
3241
1421
RepositoryFormat7)))
3242
1422
except AttributeError:
3243
1423
return False
3244
1424
3245
1425
@needs_write_lock
3246
1426
def copy_content(self, revision_id=None):
3247
1427
"""See InterRepository.copy_content()."""
3248
1428
# weave specific optimised path:
1429
# TODO: jam 20070210 Internal, should be an assert, not translate
1430
revision_id = osutils.safe_revision_id(revision_id)
3249
1431
try:
3250
1432
self.target.set_make_working_trees(self.source.make_working_trees())
3251
except (errors.RepositoryUpgradeRequired, NotImplemented):
1433
except NotImplementedError:
3252
1434
pass
3253
1435
# FIXME do not peek!
3254
if self.source._transport.listable():
1436
if self.source.control_files._transport.listable():
3255
1437
pb = ui.ui_factory.nested_progress_bar()
3256
1438
try:
3257
self.target.texts.insert_record_stream(
3258
self.source.texts.get_record_stream(
3259
self.source.texts.keys(), 'topological', False))
1439
self.target.weave_store.copy_all_ids(
1440
self.source.weave_store,
1441
pb=pb,
1442
from_transaction=self.source.get_transaction(),
1443
to_transaction=self.target.get_transaction())
3260
1444
pb.update('copying inventory', 0, 1)
3261
self.target.inventories.insert_record_stream(
3262
self.source.inventories.get_record_stream(
3263
self.source.inventories.keys(), 'topological', False))
3264
self.target.signatures.insert_record_stream(
3265
self.source.signatures.get_record_stream(
3266
self.source.signatures.keys(),
3267
'unordered', True))
3268
self.target.revisions.insert_record_stream(
3269
self.source.revisions.get_record_stream(
3270
self.source.revisions.keys(),
3271
'topological', True))
1445
self.target.control_weaves.copy_multi(
1446
self.source.control_weaves, ['inventory'],
1447
from_transaction=self.source.get_transaction(),
1448
to_transaction=self.target.get_transaction())
1449
self.target._revision_store.text_store.copy_all_ids(
1450
self.source._revision_store.text_store,
1451
pb=pb)
3272
1452
finally:
3273
1453
pb.finished()
3274
1454
else:
3275
1455
self.target.fetch(self.source, revision_id=revision_id)
3276
1456
1457
@needs_write_lock
1458
def fetch(self, revision_id=None, pb=None):
1459
"""See InterRepository.fetch()."""
1460
from bzrlib.fetch import GenericRepoFetcher
1461
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1462
self.source, self.source._format, self.target, self.target._format)
1463
# TODO: jam 20070210 This should be an assert, not a translate
1464
revision_id = osutils.safe_revision_id(revision_id)
1465
f = GenericRepoFetcher(to_repository=self.target,
1466
from_repository=self.source,
1467
last_revision=revision_id,
1468
pb=pb)
1469
return f.count_copied, f.failed_revisions
1470
3277
1471
@needs_read_lock
3278
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1472
def missing_revision_ids(self, revision_id=None):
3279
1473
"""See InterRepository.missing_revision_ids()."""
3280
1474
# we want all revisions to satisfy revision_id in source.
3281
1475
# but we don't want to stat every file here and there.
3282
# we want then, all revisions other needs to satisfy revision_id
1476
# we want then, all revisions other needs to satisfy revision_id
3283
1477
# checked, but not those that we have locally.
3284
# so the first thing is to get a subset of the revisions to
1478
# so the first thing is to get a subset of the revisions to
3285
1479
# satisfy revision_id in source, and then eliminate those that
3286
# we do already have.
3287
# this is slow on high latency connection to self, but as this
3288
# disk format scales terribly for push anyway due to rewriting
1480
# we do already have.
1481
# this is slow on high latency connection to self, but as as this
1482
# disk format scales terribly for push anyway due to rewriting
3289
1483
# inventory.weave, this is considered acceptable.
3290
1484
# - RBC 20060209
3291
1485
if revision_id is not None:
3292
1486
source_ids = self.source.get_ancestry(revision_id)
3293
if source_ids[0] is not None:
3294
raise AssertionError()
1487
assert source_ids[0] is None
3295
1488
source_ids.pop(0)
3296
1489
else:
3297
1490
source_ids = self.source._all_possible_ids()
3302
1495
# we do not have a revision as that would be pointless.
3303
1496
target_ids = set(self.target._all_possible_ids())
3304
1497
possibly_present_revisions = target_ids.intersection(source_ids_set)
3305
actually_present_revisions = set(
3306
self.target._eliminate_revisions_not_present(possibly_present_revisions))
1498
actually_present_revisions = set(self.target._eliminate_revisions_not_present(possibly_present_revisions))
3307
1499
required_revisions = source_ids_set.difference(actually_present_revisions)
1500
required_topo_revisions = [rev_id for rev_id in source_ids if rev_id in required_revisions]
3308
1501
if revision_id is not None:
3309
1502
# we used get_ancestry to determine source_ids then we are assured all
3310
1503
# revisions referenced are present as they are installed in topological order.
3311
1504
# and the tip revision was validated by get_ancestry.
3312
result_set = required_revisions
1505
return required_topo_revisions
3313
1506
else:
3314
# if we just grabbed the possibly available ids, then
1507
# if we just grabbed the possibly available ids, then
3315
1508
# we only have an estimate of whats available and need to validate
3316
1509
# that against the revision records.
3317
result_set = set(
3318
self.source._eliminate_revisions_not_present(required_revisions))
3319
return self.source.revision_ids_to_search_result(result_set)
1510
return self.source._eliminate_revisions_not_present(required_topo_revisions)
3320
1511
3321
1512
3322
1513
class InterKnitRepo(InterSameDataRepository):
3330
1521
@staticmethod
3331
1522
def is_compatible(source, target):
3332
1523
"""Be compatible with known Knit formats.
3333
1524
3334
1525
We don't test for the stores being of specific types because that
3335
could lead to confusing results, and there is no need to be
1526
could lead to confusing results, and there is no need to be
3336
1527
overly general.
3337
1528
"""
3338
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
1529
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1
3339
1530
try:
3340
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3341
isinstance(target._format, RepositoryFormatKnit))
1531
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1532
isinstance(target._format, (RepositoryFormatKnit1)))
3342
1533
except AttributeError:
3343
1534
return False
3344
return are_knits and InterRepository._same_model(source, target)
1535
1536
@needs_write_lock
1537
def fetch(self, revision_id=None, pb=None):
1538
"""See InterRepository.fetch()."""
1539
from bzrlib.fetch import KnitRepoFetcher
1540
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1541
self.source, self.source._format, self.target, self.target._format)
1542
# TODO: jam 20070210 This should be an assert, not a translate
1543
revision_id = osutils.safe_revision_id(revision_id)
1544
f = KnitRepoFetcher(to_repository=self.target,
1545
from_repository=self.source,
1546
last_revision=revision_id,
1547
pb=pb)
1548
return f.count_copied, f.failed_revisions
3345
1549
3346
1550
@needs_read_lock
3347
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1551
def missing_revision_ids(self, revision_id=None):
3348
1552
"""See InterRepository.missing_revision_ids()."""
3349
1553
if revision_id is not None:
3350
1554
source_ids = self.source.get_ancestry(revision_id)
3351
if source_ids[0] is not None:
3352
raise AssertionError()
1555
assert source_ids[0] is None
3353
1556
source_ids.pop(0)
3354
1557
else:
3355
source_ids = self.source.all_revision_ids()
1558
source_ids = self.source._all_possible_ids()
3356
1559
source_ids_set = set(source_ids)
3357
1560
# source_ids is the worst possible case we may need to pull.
3358
1561
# now we want to filter source_ids against what we actually
3359
1562
# have in target, but don't try to check for existence where we know
3360
1563
# we do not have a revision as that would be pointless.
3361
target_ids = set(self.target.all_revision_ids())
1564
target_ids = set(self.target._all_possible_ids())
3362
1565
possibly_present_revisions = target_ids.intersection(source_ids_set)
3363
actually_present_revisions = set(
3364
self.target._eliminate_revisions_not_present(possibly_present_revisions))
1566
actually_present_revisions = set(self.target._eliminate_revisions_not_present(possibly_present_revisions))
3365
1567
required_revisions = source_ids_set.difference(actually_present_revisions)
1568
required_topo_revisions = [rev_id for rev_id in source_ids if rev_id in required_revisions]
3366
1569
if revision_id is not None:
3367
1570
# we used get_ancestry to determine source_ids then we are assured all
3368
1571
# revisions referenced are present as they are installed in topological order.
3369
1572
# and the tip revision was validated by get_ancestry.
3370
result_set = required_revisions
1573
return required_topo_revisions
3371
1574
else:
3372
# if we just grabbed the possibly available ids, then
1575
# if we just grabbed the possibly available ids, then
3373
1576
# we only have an estimate of whats available and need to validate
3374
1577
# that against the revision records.
3375
result_set = set(
3376
self.source._eliminate_revisions_not_present(required_revisions))
3377
return self.source.revision_ids_to_search_result(result_set)
3378
3379
3380
class InterPackRepo(InterSameDataRepository):
3381
"""Optimised code paths between Pack based repositories."""
1578
return self.source._eliminate_revisions_not_present(required_topo_revisions)
1579
1580
1581
class InterModel1and2(InterRepository):
3382
1582
3383
1583
@classmethod
3384
1584
def _get_repo_format_to_test(self):
3385
from bzrlib.repofmt import pack_repo
3386
return pack_repo.RepositoryFormatKnitPack6RichRoot()
1585
return None
3387
1586
3388
1587
@staticmethod
3389
1588
def is_compatible(source, target):
3390
"""Be compatible with known Pack formats.
3391
3392
We don't test for the stores being of specific types because that
3393
could lead to confusing results, and there is no need to be
3394
overly general.
3395
3396
InterPackRepo does not support CHK based repositories.
1589
if not isinstance(source, Repository):
1590
return False
1591
if not isinstance(target, Repository):
1592
return False
1593
if not source._format.rich_root_data and target._format.rich_root_data:
1594
return True
1595
else:
1596
return False
1597
1598
@needs_write_lock
1599
def fetch(self, revision_id=None, pb=None):
1600
"""See InterRepository.fetch()."""
1601
from bzrlib.fetch import Model1toKnit2Fetcher
1602
# TODO: jam 20070210 This should be an assert, not a translate
1603
revision_id = osutils.safe_revision_id(revision_id)
1604
f = Model1toKnit2Fetcher(to_repository=self.target,
1605
from_repository=self.source,
1606
last_revision=revision_id,
1607
pb=pb)
1608
return f.count_copied, f.failed_revisions
1609
1610
@needs_write_lock
1611
def copy_content(self, revision_id=None):
1612
"""Make a complete copy of the content in self into destination.
1613
1614
This is a destructive operation! Do not use it on existing
1615
repositories.
1616
1617
:param revision_id: Only copy the content needed to construct
1618
revision_id and its parents.
3397
1619
"""
3398
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3399
from bzrlib.repofmt.groupcompress_repo import RepositoryFormatCHK1
3400
try:
3401
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3402
isinstance(target._format, RepositoryFormatPack))
3403
not_packs = (isinstance(source._format, RepositoryFormatCHK1) or
3404
isinstance(target._format, RepositoryFormatCHK1))
1620
try:
1621
self.target.set_make_working_trees(self.source.make_working_trees())
1622
except NotImplementedError:
1623
pass
1624
# TODO: jam 20070210 Internal, assert, don't translate
1625
revision_id = osutils.safe_revision_id(revision_id)
1626
# but don't bother fetching if we have the needed data now.
1627
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1628
self.target.has_revision(revision_id)):
1629
return
1630
self.target.fetch(self.source, revision_id=revision_id)
1631
1632
1633
class InterKnit1and2(InterKnitRepo):
1634
1635
@classmethod
1636
def _get_repo_format_to_test(self):
1637
return None
1638
1639
@staticmethod
1640
def is_compatible(source, target):
1641
"""Be compatible with Knit1 source and Knit3 target"""
1642
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit3
1643
try:
1644
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1, \
1645
RepositoryFormatKnit3
1646
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1647
isinstance(target._format, (RepositoryFormatKnit3)))
3405
1648
except AttributeError:
3406
1649
return False
3407
if not_packs or not are_packs:
3408
return False
3409
return InterRepository._same_model(source, target)
3410
3411
@needs_write_lock
3412
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3413
fetch_spec=None):
3414
"""See InterRepository.fetch()."""
3415
if (len(self.source._fallback_repositories) > 0 or
3416
len(self.target._fallback_repositories) > 0):
3417
# The pack layer is not aware of fallback repositories, so when
3418
# fetching from a stacked repository or into a stacked repository
3419
# we use the generic fetch logic which uses the VersionedFiles
3420
# attributes on repository.
3421
from bzrlib.fetch import RepoFetcher
3422
fetcher = RepoFetcher(self.target, self.source, revision_id,
3423
pb, find_ghosts, fetch_spec=fetch_spec)
3424
if fetch_spec is not None:
3425
if len(list(fetch_spec.heads)) != 1:
3426
raise AssertionError(
3427
"InterPackRepo.fetch doesn't support "
3428
"fetching multiple heads yet.")
3429
revision_id = list(fetch_spec.heads)[0]
3430
fetch_spec = None
3431
if revision_id is None:
3432
# TODO:
3433
# everything to do - use pack logic
3434
# to fetch from all packs to one without
3435
# inventory parsing etc, IFF nothing to be copied is in the target.
3436
# till then:
3437
source_revision_ids = frozenset(self.source.all_revision_ids())
3438
revision_ids = source_revision_ids - \
3439
frozenset(self.target.get_parent_map(source_revision_ids))
3440
revision_keys = [(revid,) for revid in revision_ids]
3441
index = self.target._pack_collection.revision_index.combined_index
3442
present_revision_ids = set(item[1][0] for item in
3443
index.iter_entries(revision_keys))
3444
revision_ids = set(revision_ids) - present_revision_ids
3445
# implementing the TODO will involve:
3446
# - detecting when all of a pack is selected
3447
# - avoiding as much as possible pre-selection, so the
3448
# more-core routines such as create_pack_from_packs can filter in
3449
# a just-in-time fashion. (though having a HEADS list on a
3450
# repository might make this a lot easier, because we could
3451
# sensibly detect 'new revisions' without doing a full index scan.
3452
elif _mod_revision.is_null(revision_id):
3453
# nothing to do:
3454
return (0, [])
3455
else:
3456
revision_ids = self.search_missing_revision_ids(revision_id,
3457
find_ghosts=find_ghosts).get_keys()
3458
if len(revision_ids) == 0:
3459
return (0, [])
3460
return self._pack(self.source, self.target, revision_ids)
3461
3462
def _pack(self, source, target, revision_ids):
3463
from bzrlib.repofmt.pack_repo import Packer
3464
packs = source._pack_collection.all_packs()
3465
pack = Packer(self.target._pack_collection, packs, '.fetch',
3466
revision_ids).pack()
3467
if pack is not None:
3468
self.target._pack_collection._save_pack_names()
3469
copied_revs = pack.get_revision_count()
3470
# Trigger an autopack. This may duplicate effort as we've just done
3471
# a pack creation, but for now it is simpler to think about as
3472
# 'upload data, then repack if needed'.
3473
self.target._pack_collection.autopack()
3474
return (copied_revs, [])
3475
else:
3476
return (0, [])
3477
3478
@needs_read_lock
3479
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3480
"""See InterRepository.missing_revision_ids().
3481
3482
:param find_ghosts: Find ghosts throughout the ancestry of
3483
revision_id.
3484
"""
3485
if not find_ghosts and revision_id is not None:
3486
return self._walk_to_common_revisions([revision_id])
3487
elif revision_id is not None:
3488
# Find ghosts: search for revisions pointing from one repository to
3489
# the other, and vice versa, anywhere in the history of revision_id.
3490
graph = self.target.get_graph(other_repository=self.source)
3491
searcher = graph._make_breadth_first_searcher([revision_id])
3492
found_ids = set()
3493
while True:
3494
try:
3495
next_revs, ghosts = searcher.next_with_ghosts()
3496
except StopIteration:
3497
break
3498
if revision_id in ghosts:
3499
raise errors.NoSuchRevision(self.source, revision_id)
3500
found_ids.update(next_revs)
3501
found_ids.update(ghosts)
3502
found_ids = frozenset(found_ids)
3503
# Double query here: should be able to avoid this by changing the
3504
# graph api further.
3505
result_set = found_ids - frozenset(
3506
self.target.get_parent_map(found_ids))
3507
else:
3508
source_ids = self.source.all_revision_ids()
3509
# source_ids is the worst possible case we may need to pull.
3510
# now we want to filter source_ids against what we actually
3511
# have in target, but don't try to check for existence where we know
3512
# we do not have a revision as that would be pointless.
3513
target_ids = set(self.target.all_revision_ids())
3514
result_set = set(source_ids).difference(target_ids)
3515
return self.source.revision_ids_to_search_result(result_set)
3516
3517
3518
class InterDifferingSerializer(InterRepository):
3519
3520
@classmethod
3521
def _get_repo_format_to_test(self):
3522
return None
3523
3524
@staticmethod
3525
def is_compatible(source, target):
3526
"""Be compatible with Knit2 source and Knit3 target"""
3527
# This is redundant with format.check_conversion_target(), however that
3528
# raises an exception, and we just want to say "False" as in we won't
3529
# support converting between these formats.
3530
if source.supports_rich_root() and not target.supports_rich_root():
3531
return False
3532
if (source._format.supports_tree_reference
3533
and not target._format.supports_tree_reference):
3534
return False
3535
return True
3536
3537
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3538
"""Get the best delta and base for this revision.
3539
3540
:return: (basis_id, delta)
3541
"""
3542
possible_trees = [(parent_id, cache[parent_id])
3543
for parent_id in parent_ids
3544
if parent_id in cache]
3545
if len(possible_trees) == 0:
3546
# There either aren't any parents, or the parents aren't in the
3547
# cache, so just use the last converted tree
3548
possible_trees.append((basis_id, cache[basis_id]))
3549
deltas = []
3550
for basis_id, basis_tree in possible_trees:
3551
delta = tree.inventory._make_delta(basis_tree.inventory)
3552
deltas.append((len(delta), basis_id, delta))
3553
deltas.sort()
3554
return deltas[0][1:]
3555
3556
def _get_parent_keys(self, root_key, parent_map):
3557
"""Get the parent keys for a given root id."""
3558
root_id, rev_id = root_key
3559
# Include direct parents of the revision, but only if they used
3560
# the same root_id.
3561
parent_keys = []
3562
for parent_id in parent_map[rev_id]:
3563
if parent_id == _mod_revision.NULL_REVISION:
3564
continue
3565
if parent_id not in self._revision_id_to_root_id:
3566
# We probably didn't read this revision, go spend the
3567
# extra effort to actually check
3568
try:
3569
tree = self.source.revision_tree(parent_id)
3570
except errors.NoSuchRevision:
3571
# Ghost, fill out _revision_id_to_root_id in case we
3572
# encounter this again.
3573
# But set parent_root_id to None since we don't really know
3574
parent_root_id = None
3575
else:
3576
parent_root_id = tree.get_root_id()
3577
self._revision_id_to_root_id[parent_id] = None
3578
else:
3579
parent_root_id = self._revision_id_to_root_id[parent_id]
3580
if root_id == parent_root_id or parent_root_id is None:
3581
parent_keys.append((root_id, parent_id))
3582
return tuple(parent_keys)
3583
3584
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3585
for root_key in root_keys_to_create:
3586
parent_keys = self._get_parent_keys(root_key, parent_map)
3587
yield versionedfile.FulltextContentFactory(root_key,
3588
parent_keys, None, '')
3589
3590
def _fetch_batch(self, revision_ids, basis_id, cache):
3591
"""Fetch across a few revisions.
3592
3593
:param revision_ids: The revisions to copy
3594
:param basis_id: The revision_id of a tree that must be in cache, used
3595
as a basis for delta when no other base is available
3596
:param cache: A cache of RevisionTrees that we can use.
3597
:return: The revision_id of the last converted tree. The RevisionTree
3598
for it will be in cache
3599
"""
3600
# Walk though all revisions; get inventory deltas, copy referenced
3601
# texts that delta references, insert the delta, revision and
3602
# signature.
3603
root_keys_to_create = set()
3604
text_keys = set()
3605
pending_deltas = []
3606
pending_revisions = []
3607
parent_map = self.source.get_parent_map(revision_ids)
3608
for tree in self.source.revision_trees(revision_ids):
3609
current_revision_id = tree.get_revision_id()
3610
parent_ids = parent_map.get(current_revision_id, ())
3611
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3612
basis_id, cache)
3613
if self._converting_to_rich_root:
3614
self._revision_id_to_root_id[current_revision_id] = \
3615
tree.get_root_id()
3616
# Find text entries that need to be copied
3617
for old_path, new_path, file_id, entry in delta:
3618
if new_path is not None:
3619
if not new_path:
3620
# This is the root
3621
if not self.target.supports_rich_root():
3622
# The target doesn't support rich root, so we don't
3623
# copy
3624
continue
3625
if self._converting_to_rich_root:
3626
# This can't be copied normally, we have to insert
3627
# it specially
3628
root_keys_to_create.add((file_id, entry.revision))
3629
continue
3630
text_keys.add((file_id, entry.revision))
3631
revision = self.source.get_revision(current_revision_id)
3632
pending_deltas.append((basis_id, delta,
3633
current_revision_id, revision.parent_ids))
3634
pending_revisions.append(revision)
3635
cache[current_revision_id] = tree
3636
basis_id = current_revision_id
3637
# Copy file texts
3638
from_texts = self.source.texts
3639
to_texts = self.target.texts
3640
if root_keys_to_create:
3641
root_stream = self._new_root_data_stream(root_keys_to_create,
3642
parent_map)
3643
to_texts.insert_record_stream(root_stream)
3644
to_texts.insert_record_stream(from_texts.get_record_stream(
3645
text_keys, self.target._format._fetch_order,
3646
not self.target._format._fetch_uses_deltas))
3647
# insert inventory deltas
3648
for delta in pending_deltas:
3649
self.target.add_inventory_by_delta(*delta)
3650
if self.target._fallback_repositories:
3651
# Make sure this stacked repository has all the parent inventories
3652
# for the new revisions that we are about to insert. We do this
3653
# before adding the revisions so that no revision is added until
3654
# all the inventories it may depend on are added.
3655
parent_ids = set()
3656
revision_ids = set()
3657
for revision in pending_revisions:
3658
revision_ids.add(revision.revision_id)
3659
parent_ids.update(revision.parent_ids)
3660
parent_ids.difference_update(revision_ids)
3661
parent_ids.discard(_mod_revision.NULL_REVISION)
3662
parent_map = self.source.get_parent_map(parent_ids)
3663
for parent_tree in self.source.revision_trees(parent_ids):
3664
basis_id, delta = self._get_delta_for_revision(tree, parent_ids, basis_id, cache)
3665
current_revision_id = parent_tree.get_revision_id()
3666
parents_parents = parent_map[current_revision_id]
3667
self.target.add_inventory_by_delta(
3668
basis_id, delta, current_revision_id, parents_parents)
3669
# insert signatures and revisions
3670
for revision in pending_revisions:
3671
try:
3672
signature = self.source.get_signature_text(
3673
revision.revision_id)
3674
self.target.add_signature_text(revision.revision_id,
3675
signature)
3676
except errors.NoSuchRevision:
3677
pass
3678
self.target.add_revision(revision.revision_id, revision)
3679
return basis_id
3680
3681
def _fetch_all_revisions(self, revision_ids, pb):
3682
"""Fetch everything for the list of revisions.
3683
3684
:param revision_ids: The list of revisions to fetch. Must be in
3685
topological order.
3686
:param pb: A ProgressBar
3687
:return: None
3688
"""
3689
basis_id, basis_tree = self._get_basis(revision_ids[0])
3690
batch_size = 100
3691
cache = lru_cache.LRUCache(100)
3692
cache[basis_id] = basis_tree
3693
del basis_tree # We don't want to hang on to it here
3694
for offset in range(0, len(revision_ids), batch_size):
3695
self.target.start_write_group()
3696
try:
3697
pb.update('Transferring revisions', offset,
3698
len(revision_ids))
3699
batch = revision_ids[offset:offset+batch_size]
3700
basis_id = self._fetch_batch(batch, basis_id, cache)
3701
except:
3702
self.target.abort_write_group()
3703
raise
3704
else:
3705
self.target.commit_write_group()
3706
pb.update('Transferring revisions', len(revision_ids),
3707
len(revision_ids))
3708
3709
@needs_write_lock
3710
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3711
fetch_spec=None):
3712
"""See InterRepository.fetch()."""
3713
if fetch_spec is not None:
3714
raise AssertionError("Not implemented yet...")
3715
if (not self.source.supports_rich_root()
3716
and self.target.supports_rich_root()):
3717
self._converting_to_rich_root = True
3718
self._revision_id_to_root_id = {}
3719
else:
3720
self._converting_to_rich_root = False
3721
revision_ids = self.target.search_missing_revision_ids(self.source,
3722
revision_id, find_ghosts=find_ghosts).get_keys()
3723
if not revision_ids:
3724
return 0, 0
3725
revision_ids = tsort.topo_sort(
3726
self.source.get_graph().get_parent_map(revision_ids))
3727
if not revision_ids:
3728
return 0, 0
3729
# Walk though all revisions; get inventory deltas, copy referenced
3730
# texts that delta references, insert the delta, revision and
3731
# signature.
3732
first_rev = self.source.get_revision(revision_ids[0])
3733
if pb is None:
3734
my_pb = ui.ui_factory.nested_progress_bar()
3735
pb = my_pb
3736
else:
3737
symbol_versioning.warn(
3738
symbol_versioning.deprecated_in((1, 14, 0))
3739
% "pb parameter to fetch()")
3740
my_pb = None
3741
try:
3742
self._fetch_all_revisions(revision_ids, pb)
3743
finally:
3744
if my_pb is not None:
3745
my_pb.finished()
3746
return len(revision_ids), 0
3747
3748
def _get_basis(self, first_revision_id):
3749
"""Get a revision and tree which exists in the target.
3750
3751
This assumes that first_revision_id is selected for transmission
3752
because all other ancestors are already present. If we can't find an
3753
ancestor we fall back to NULL_REVISION since we know that is safe.
3754
3755
:return: (basis_id, basis_tree)
3756
"""
3757
first_rev = self.source.get_revision(first_revision_id)
3758
try:
3759
basis_id = first_rev.parent_ids[0]
3760
# only valid as a basis if the target has it
3761
self.target.get_revision(basis_id)
3762
# Try to get a basis tree - if its a ghost it will hit the
3763
# NoSuchRevision case.
3764
basis_tree = self.source.revision_tree(basis_id)
3765
except (IndexError, errors.NoSuchRevision):
3766
basis_id = _mod_revision.NULL_REVISION
3767
basis_tree = self.source.revision_tree(basis_id)
3768
return basis_id, basis_tree
3769
3770
3771
InterRepository.register_optimiser(InterDifferingSerializer)
1650
1651
@needs_write_lock
1652
def fetch(self, revision_id=None, pb=None):
1653
"""See InterRepository.fetch()."""
1654
from bzrlib.fetch import Knit1to2Fetcher
1655
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1656
self.source, self.source._format, self.target,
1657
self.target._format)
1658
# TODO: jam 20070210 This should be an assert, not a translate
1659
revision_id = osutils.safe_revision_id(revision_id)
1660
f = Knit1to2Fetcher(to_repository=self.target,
1661
from_repository=self.source,
1662
last_revision=revision_id,
1663
pb=pb)
1664
return f.count_copied, f.failed_revisions
1665
1666
3772
1667
InterRepository.register_optimiser(InterSameDataRepository)
3773
1668
InterRepository.register_optimiser(InterWeaveRepo)
3774
1669
InterRepository.register_optimiser(InterKnitRepo)
3775
InterRepository.register_optimiser(InterPackRepo)
1670
InterRepository.register_optimiser(InterModel1and2)
1671
InterRepository.register_optimiser(InterKnit1and2)
1672
1673
1674
class RepositoryTestProviderAdapter(object):
1675
"""A tool to generate a suite testing multiple repository formats at once.
1676
1677
This is done by copying the test once for each transport and injecting
1678
the transport_server, transport_readonly_server, and bzrdir_format and
1679
repository_format classes into each copy. Each copy is also given a new id()
1680
to make it easy to identify.
1681
"""
1682
1683
def __init__(self, transport_server, transport_readonly_server, formats):
1684
self._transport_server = transport_server
1685
self._transport_readonly_server = transport_readonly_server
1686
self._formats = formats
1687
1688
def adapt(self, test):
1689
result = unittest.TestSuite()
1690
for repository_format, bzrdir_format in self._formats:
1691
from copy import deepcopy
1692
new_test = deepcopy(test)
1693
new_test.transport_server = self._transport_server
1694
new_test.transport_readonly_server = self._transport_readonly_server
1695
new_test.bzrdir_format = bzrdir_format
1696
new_test.repository_format = repository_format
1697
def make_new_test_id():
1698
new_id = "%s(%s)" % (new_test.id(), repository_format.__class__.__name__)
1699
return lambda: new_id
1700
new_test.id = make_new_test_id()
1701
result.addTest(new_test)
1702
return result
1703
1704
1705
class InterRepositoryTestProviderAdapter(object):
1706
"""A tool to generate a suite testing multiple inter repository formats.
1707
1708
This is done by copying the test once for each interrepo provider and injecting
1709
the transport_server, transport_readonly_server, repository_format and
1710
repository_to_format classes into each copy.
1711
Each copy is also given a new id() to make it easy to identify.
1712
"""
1713
1714
def __init__(self, transport_server, transport_readonly_server, formats):
1715
self._transport_server = transport_server
1716
self._transport_readonly_server = transport_readonly_server
1717
self._formats = formats
1718
1719
def adapt(self, test):
1720
result = unittest.TestSuite()
1721
for interrepo_class, repository_format, repository_format_to in self._formats:
1722
from copy import deepcopy
1723
new_test = deepcopy(test)
1724
new_test.transport_server = self._transport_server
1725
new_test.transport_readonly_server = self._transport_readonly_server
1726
new_test.interrepo_class = interrepo_class
1727
new_test.repository_format = repository_format
1728
new_test.repository_format_to = repository_format_to
1729
def make_new_test_id():
1730
new_id = "%s(%s)" % (new_test.id(), interrepo_class.__name__)
1731
return lambda: new_id
1732
new_test.id = make_new_test_id()
1733
result.addTest(new_test)
1734
return result
1735
1736
@staticmethod
1737
def default_test_list():
1738
"""Generate the default list of interrepo permutations to test."""
1739
from bzrlib.repofmt import knitrepo, weaverepo
1740
result = []
1741
# test the default InterRepository between format 6 and the current
1742
# default format.
1743
# XXX: robertc 20060220 reinstate this when there are two supported
1744
# formats which do not have an optimal code path between them.
1745
#result.append((InterRepository,
1746
# RepositoryFormat6(),
1747
# RepositoryFormatKnit1()))
1748
for optimiser_class in InterRepository._optimisers:
1749
format_to_test = optimiser_class._get_repo_format_to_test()
1750
if format_to_test is not None:
1751
result.append((optimiser_class,
1752
format_to_test, format_to_test))
1753
# if there are specific combinations we want to use, we can add them
1754
# here.
1755
result.append((InterModel1and2,
1756
weaverepo.RepositoryFormat5(),
1757
knitrepo.RepositoryFormatKnit3()))
1758
result.append((InterKnit1and2,
1759
knitrepo.RepositoryFormatKnit1(),
1760
knitrepo.RepositoryFormatKnit3()))
1761
return result
3776
1762
3777
1763
3778
1764
class CopyConverter(object):
3779
1765
"""A repository conversion tool which just performs a copy of the content.
3780
1766
3781
1767
This is slow but quite reliable.
3782
1768
"""
3783
1769
3787
1773
:param target_format: The format the resulting repository should be.
3788
1774
"""
3789
1775
self.target_format = target_format
3790
1776
3791
1777
def convert(self, repo, pb):
3792
1778
"""Perform the conversion of to_convert, giving feedback via pb.
3793
1779
3827
1813
self.pb.update(message, self.count, self.total)
3828
1814
3829
1815
1816
class CommitBuilder(object):
1817
"""Provides an interface to build up a commit.
1818
1819
This allows describing a tree to be committed without needing to
1820
know the internals of the format of the repository.
1821
"""
1822
1823
record_root_entry = False
1824
def __init__(self, repository, parents, config, timestamp=None,
1825
timezone=None, committer=None, revprops=None,
1826
revision_id=None):
1827
"""Initiate a CommitBuilder.
1828
1829
:param repository: Repository to commit to.
1830
:param parents: Revision ids of the parents of the new revision.
1831
:param config: Configuration to use.
1832
:param timestamp: Optional timestamp recorded for commit.
1833
:param timezone: Optional timezone for timestamp.
1834
:param committer: Optional committer to set for commit.
1835
:param revprops: Optional dictionary of revision properties.
1836
:param revision_id: Optional revision id.
1837
"""
1838
self._config = config
1839
1840
if committer is None:
1841
self._committer = self._config.username()
1842
else:
1843
assert isinstance(committer, basestring), type(committer)
1844
self._committer = committer
1845
1846
self.new_inventory = Inventory(None)
1847
self._new_revision_id = osutils.safe_revision_id(revision_id)
1848
self.parents = parents
1849
self.repository = repository
1850
1851
self._revprops = {}
1852
if revprops is not None:
1853
self._revprops.update(revprops)
1854
1855
if timestamp is None:
1856
timestamp = time.time()
1857
# Restrict resolution to 1ms
1858
self._timestamp = round(timestamp, 3)
1859
1860
if timezone is None:
1861
self._timezone = osutils.local_time_offset()
1862
else:
1863
self._timezone = int(timezone)
1864
1865
self._generate_revision_if_needed()
1866
1867
def commit(self, message):
1868
"""Make the actual commit.
1869
1870
:return: The revision id of the recorded revision.
1871
"""
1872
rev = _mod_revision.Revision(
1873
timestamp=self._timestamp,
1874
timezone=self._timezone,
1875
committer=self._committer,
1876
message=message,
1877
inventory_sha1=self.inv_sha1,
1878
revision_id=self._new_revision_id,
1879
properties=self._revprops)
1880
rev.parent_ids = self.parents
1881
self.repository.add_revision(self._new_revision_id, rev,
1882
self.new_inventory, self._config)
1883
return self._new_revision_id
1884
1885
def revision_tree(self):
1886
"""Return the tree that was just committed.
1887
1888
After calling commit() this can be called to get a RevisionTree
1889
representing the newly committed tree. This is preferred to
1890
calling Repository.revision_tree() because that may require
1891
deserializing the inventory, while we already have a copy in
1892
memory.
1893
"""
1894
return RevisionTree(self.repository, self.new_inventory,
1895
self._new_revision_id)
1896
1897
def finish_inventory(self):
1898
"""Tell the builder that the inventory is finished."""
1899
if self.new_inventory.root is None:
1900
symbol_versioning.warn('Root entry should be supplied to'
1901
' record_entry_contents, as of bzr 0.10.',
1902
DeprecationWarning, stacklevel=2)
1903
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
1904
self.new_inventory.revision_id = self._new_revision_id
1905
self.inv_sha1 = self.repository.add_inventory(
1906
self._new_revision_id,
1907
self.new_inventory,
1908
self.parents
1909
)
1910
1911
def _gen_revision_id(self):
1912
"""Return new revision-id."""
1913
return generate_ids.gen_revision_id(self._config.username(),
1914
self._timestamp)
1915
1916
def _generate_revision_if_needed(self):
1917
"""Create a revision id if None was supplied.
1918
1919
If the repository can not support user-specified revision ids
1920
they should override this function and raise CannotSetRevisionId
1921
if _new_revision_id is not None.
1922
1923
:raises: CannotSetRevisionId
1924
"""
1925
if self._new_revision_id is None:
1926
self._new_revision_id = self._gen_revision_id()
1927
1928
def record_entry_contents(self, ie, parent_invs, path, tree):
1929
"""Record the content of ie from tree into the commit if needed.
1930
1931
Side effect: sets ie.revision when unchanged
1932
1933
:param ie: An inventory entry present in the commit.
1934
:param parent_invs: The inventories of the parent revisions of the
1935
commit.
1936
:param path: The path the entry is at in the tree.
1937
:param tree: The tree which contains this entry and should be used to
1938
obtain content.
1939
"""
1940
if self.new_inventory.root is None and ie.parent_id is not None:
1941
symbol_versioning.warn('Root entry should be supplied to'
1942
' record_entry_contents, as of bzr 0.10.',
1943
DeprecationWarning, stacklevel=2)
1944
self.record_entry_contents(tree.inventory.root.copy(), parent_invs,
1945
'', tree)
1946
self.new_inventory.add(ie)
1947
1948
# ie.revision is always None if the InventoryEntry is considered
1949
# for committing. ie.snapshot will record the correct revision
1950
# which may be the sole parent if it is untouched.
1951
if ie.revision is not None:
1952
return
1953
1954
# In this revision format, root entries have no knit or weave
1955
if ie is self.new_inventory.root:
1956
# When serializing out to disk and back in
1957
# root.revision is always _new_revision_id
1958
ie.revision = self._new_revision_id
1959
return
1960
previous_entries = ie.find_previous_heads(
1961
parent_invs,
1962
self.repository.weave_store,
1963
self.repository.get_transaction())
1964
# we are creating a new revision for ie in the history store
1965
# and inventory.
1966
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
1967
1968
def modified_directory(self, file_id, file_parents):
1969
"""Record the presence of a symbolic link.
1970
1971
:param file_id: The file_id of the link to record.
1972
:param file_parents: The per-file parent revision ids.
1973
"""
1974
self._add_text_to_weave(file_id, [], file_parents.keys())
1975
1976
def modified_reference(self, file_id, file_parents):
1977
"""Record the modification of a reference.
1978
1979
:param file_id: The file_id of the link to record.
1980
:param file_parents: The per-file parent revision ids.
1981
"""
1982
self._add_text_to_weave(file_id, [], file_parents.keys())
1983
1984
def modified_file_text(self, file_id, file_parents,
1985
get_content_byte_lines, text_sha1=None,
1986
text_size=None):
1987
"""Record the text of file file_id
1988
1989
:param file_id: The file_id of the file to record the text of.
1990
:param file_parents: The per-file parent revision ids.
1991
:param get_content_byte_lines: A callable which will return the byte
1992
lines for the file.
1993
:param text_sha1: Optional SHA1 of the file contents.
1994
:param text_size: Optional size of the file contents.
1995
"""
1996
# mutter('storing text of file {%s} in revision {%s} into %r',
1997
# file_id, self._new_revision_id, self.repository.weave_store)
1998
# special case to avoid diffing on renames or
1999
# reparenting
2000
if (len(file_parents) == 1
2001
and text_sha1 == file_parents.values()[0].text_sha1
2002
and text_size == file_parents.values()[0].text_size):
2003
previous_ie = file_parents.values()[0]
2004
versionedfile = self.repository.weave_store.get_weave(file_id,
2005
self.repository.get_transaction())
2006
versionedfile.clone_text(self._new_revision_id,
2007
previous_ie.revision, file_parents.keys())
2008
return text_sha1, text_size
2009
else:
2010
new_lines = get_content_byte_lines()
2011
# TODO: Rather than invoking sha_strings here, _add_text_to_weave
2012
# should return the SHA1 and size
2013
self._add_text_to_weave(file_id, new_lines, file_parents.keys())
2014
return osutils.sha_strings(new_lines), \
2015
sum(map(len, new_lines))
2016
2017
def modified_link(self, file_id, file_parents, link_target):
2018
"""Record the presence of a symbolic link.
2019
2020
:param file_id: The file_id of the link to record.
2021
:param file_parents: The per-file parent revision ids.
2022
:param link_target: Target location of this link.
2023
"""
2024
self._add_text_to_weave(file_id, [], file_parents.keys())
2025
2026
def _add_text_to_weave(self, file_id, new_lines, parents):
2027
versionedfile = self.repository.weave_store.get_weave_or_empty(
2028
file_id, self.repository.get_transaction())
2029
versionedfile.add_lines(self._new_revision_id, parents, new_lines)
2030
versionedfile.clear_cache()
2031
2032
2033
class _CommitBuilder(CommitBuilder):
2034
"""Temporary class so old CommitBuilders are detected properly
2035
2036
Note: CommitBuilder works whether or not root entry is recorded.
2037
"""
2038
2039
record_root_entry = True
2040
2041
2042
class RootCommitBuilder(CommitBuilder):
2043
"""This commitbuilder actually records the root id"""
2044
2045
record_root_entry = True
2046
2047
def record_entry_contents(self, ie, parent_invs, path, tree):
2048
"""Record the content of ie from tree into the commit if needed.
2049
2050
Side effect: sets ie.revision when unchanged
2051
2052
:param ie: An inventory entry present in the commit.
2053
:param parent_invs: The inventories of the parent revisions of the
2054
commit.
2055
:param path: The path the entry is at in the tree.
2056
:param tree: The tree which contains this entry and should be used to
2057
obtain content.
2058
"""
2059
assert self.new_inventory.root is not None or ie.parent_id is None
2060
self.new_inventory.add(ie)
2061
2062
# ie.revision is always None if the InventoryEntry is considered
2063
# for committing. ie.snapshot will record the correct revision
2064
# which may be the sole parent if it is untouched.
2065
if ie.revision is not None:
2066
return
2067
2068
previous_entries = ie.find_previous_heads(
2069
parent_invs,
2070
self.repository.weave_store,
2071
self.repository.get_transaction())
2072
# we are creating a new revision for ie in the history store
2073
# and inventory.
2074
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
2075
2076
3830
2077
_unescape_map = {
3831
2078
'apos':"'",
3832
2079
'quot':'"',
3855
2102
if _unescape_re is None:
3856
2103
_unescape_re = re.compile('\&([^;]*);')
3857
2104
return _unescape_re.sub(_unescaper, data)
3858
3859
3860
class _VersionedFileChecker(object):
3861
3862
def __init__(self, repository, text_key_references=None):
3863
self.repository = repository
3864
self.text_index = self.repository._generate_text_key_index(
3865
text_key_references=text_key_references)
3866
3867
def calculate_file_version_parents(self, text_key):
3868
"""Calculate the correct parents for a file version according to
3869
the inventories.
3870
"""
3871
parent_keys = self.text_index[text_key]
3872
if parent_keys == [_mod_revision.NULL_REVISION]:
3873
return ()
3874
return tuple(parent_keys)
3875
3876
def check_file_version_parents(self, texts, progress_bar=None):
3877
"""Check the parents stored in a versioned file are correct.
3878
3879
It also detects file versions that are not referenced by their
3880
corresponding revision's inventory.
3881
3882
:returns: A tuple of (wrong_parents, dangling_file_versions).
3883
wrong_parents is a dict mapping {revision_id: (stored_parents,
3884
correct_parents)} for each revision_id where the stored parents
3885
are not correct. dangling_file_versions is a set of (file_id,
3886
revision_id) tuples for versions that are present in this versioned
3887
file, but not used by the corresponding inventory.
3888
"""
3889
wrong_parents = {}
3890
self.file_ids = set([file_id for file_id, _ in
3891
self.text_index.iterkeys()])
3892
# text keys is now grouped by file_id
3893
n_weaves = len(self.file_ids)
3894
files_in_revisions = {}
3895
revisions_of_files = {}
3896
n_versions = len(self.text_index)
3897
progress_bar.update('loading text store', 0, n_versions)
3898
parent_map = self.repository.texts.get_parent_map(self.text_index)
3899
# On unlistable transports this could well be empty/error...
3900
text_keys = self.repository.texts.keys()
3901
unused_keys = frozenset(text_keys) - set(self.text_index)
3902
for num, key in enumerate(self.text_index.iterkeys()):
3903
if progress_bar is not None:
3904
progress_bar.update('checking text graph', num, n_versions)
3905
correct_parents = self.calculate_file_version_parents(key)
3906
try:
3907
knit_parents = parent_map[key]
3908
except errors.RevisionNotPresent:
3909
# Missing text!
3910
knit_parents = None
3911
if correct_parents != knit_parents:
3912
wrong_parents[key] = (knit_parents, correct_parents)
3913
return wrong_parents, unused_keys
3914
3915
3916
def _old_get_graph(repository, revision_id):
3917
"""DO NOT USE. That is all. I'm serious."""
3918
graph = repository.get_graph()
3919
revision_graph = dict(((key, value) for key, value in
3920
graph.iter_ancestry([revision_id]) if value is not None))
3921
return _strip_NULL_ghosts(revision_graph)
3922
3923
3924
def _strip_NULL_ghosts(revision_graph):
3925
"""Also don't use this. more compatibility code for unmigrated clients."""
3926
# Filter ghosts, and null:
3927
if _mod_revision.NULL_REVISION in revision_graph:
3928
del revision_graph[_mod_revision.NULL_REVISION]
3929
for key, parents in revision_graph.items():
3930
revision_graph[key] = tuple(parent for parent in parents if parent
3931
in revision_graph)
3932
return revision_graph
3933
3934
3935
class StreamSink(object):
3936
"""An object that can insert a stream into a repository.
3937
3938
This interface handles the complexity of reserialising inventories and
3939
revisions from different formats, and allows unidirectional insertion into
3940
stacked repositories without looking for the missing basis parents
3941
beforehand.
3942
"""
3943
3944
def __init__(self, target_repo):
3945
self.target_repo = target_repo
3946
3947
def insert_stream(self, stream, src_format, resume_tokens):
3948
"""Insert a stream's content into the target repository.
3949
3950
:param src_format: a bzr repository format.
3951
3952
:return: a list of resume tokens and an iterable of keys additional
3953
items required before the insertion can be completed.
3954
"""
3955
self.target_repo.lock_write()
3956
try:
3957
if resume_tokens:
3958
self.target_repo.resume_write_group(resume_tokens)
3959
else:
3960
self.target_repo.start_write_group()
3961
try:
3962
# locked_insert_stream performs a commit|suspend.
3963
return self._locked_insert_stream(stream, src_format)
3964
except:
3965
self.target_repo.abort_write_group(suppress_errors=True)
3966
raise
3967
finally:
3968
self.target_repo.unlock()
3969
3970
def _locked_insert_stream(self, stream, src_format):
3971
to_serializer = self.target_repo._format._serializer
3972
src_serializer = src_format._serializer
3973
if to_serializer == src_serializer:
3974
# If serializers match and the target is a pack repository, set the
3975
# write cache size on the new pack. This avoids poor performance
3976
# on transports where append is unbuffered (such as
3977
# RemoteTransport). This is safe to do because nothing should read
3978
# back from the target repository while a stream with matching
3979
# serialization is being inserted.
3980
# The exception is that a delta record from the source that should
3981
# be a fulltext may need to be expanded by the target (see
3982
# test_fetch_revisions_with_deltas_into_pack); but we take care to
3983
# explicitly flush any buffered writes first in that rare case.
3984
try:
3985
new_pack = self.target_repo._pack_collection._new_pack
3986
except AttributeError:
3987
# Not a pack repository
3988
pass
3989
else:
3990
new_pack.set_write_cache_size(1024*1024)
3991
for substream_type, substream in stream:
3992
if substream_type == 'texts':
3993
self.target_repo.texts.insert_record_stream(substream)
3994
elif substream_type == 'inventories':
3995
if src_serializer == to_serializer:
3996
self.target_repo.inventories.insert_record_stream(
3997
substream)
3998
else:
3999
self._extract_and_insert_inventories(
4000
substream, src_serializer)
4001
elif substream_type == 'chk_bytes':
4002
# XXX: This doesn't support conversions, as it assumes the
4003
# conversion was done in the fetch code.
4004
self.target_repo.chk_bytes.insert_record_stream(substream)
4005
elif substream_type == 'revisions':
4006
# This may fallback to extract-and-insert more often than
4007
# required if the serializers are different only in terms of
4008
# the inventory.
4009
if src_serializer == to_serializer:
4010
self.target_repo.revisions.insert_record_stream(
4011
substream)
4012
else:
4013
self._extract_and_insert_revisions(substream,
4014
src_serializer)
4015
elif substream_type == 'signatures':
4016
self.target_repo.signatures.insert_record_stream(substream)
4017
else:
4018
raise AssertionError('kaboom! %s' % (substream_type,))
4019
# Find all the new revisions (including ones from resume_tokens)
4020
missing_keys = self.target_repo.get_missing_parent_inventories()
4021
try:
4022
for prefix, versioned_file in (
4023
('texts', self.target_repo.texts),
4024
('inventories', self.target_repo.inventories),
4025
('revisions', self.target_repo.revisions),
4026
('signatures', self.target_repo.signatures),
4027
):
4028
missing_keys.update((prefix,) + key for key in
4029
versioned_file.get_missing_compression_parent_keys())
4030
except NotImplementedError:
4031
# cannot even attempt suspending, and missing would have failed
4032
# during stream insertion.
4033
missing_keys = set()
4034
else:
4035
if missing_keys:
4036
# suspend the write group and tell the caller what we is
4037
# missing. We know we can suspend or else we would not have
4038
# entered this code path. (All repositories that can handle
4039
# missing keys can handle suspending a write group).
4040
write_group_tokens = self.target_repo.suspend_write_group()
4041
return write_group_tokens, missing_keys
4042
self.target_repo.commit_write_group()
4043
return [], set()
4044
4045
def _extract_and_insert_inventories(self, substream, serializer):
4046
"""Generate a new inventory versionedfile in target, converting data.
4047
4048
The inventory is retrieved from the source, (deserializing it), and
4049
stored in the target (reserializing it in a different format).
4050
"""
4051
for record in substream:
4052
bytes = record.get_bytes_as('fulltext')
4053
revision_id = record.key[0]
4054
inv = serializer.read_inventory_from_string(bytes, revision_id)
4055
parents = [key[0] for key in record.parents]
4056
self.target_repo.add_inventory(revision_id, inv, parents)
4057
4058
def _extract_and_insert_revisions(self, substream, serializer):
4059
for record in substream:
4060
bytes = record.get_bytes_as('fulltext')
4061
revision_id = record.key[0]
4062
rev = serializer.read_revision_from_string(bytes)
4063
if rev.revision_id != revision_id:
4064
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4065
self.target_repo.add_revision(revision_id, rev)
4066
4067
def finished(self):
4068
if self.target_repo._format._fetch_reconcile:
4069
self.target_repo.reconcile()
4070
4071
4072
class StreamSource(object):
4073
"""A source of a stream for fetching between repositories."""
4074
4075
def __init__(self, from_repository, to_format):
4076
"""Create a StreamSource streaming from from_repository."""
4077
self.from_repository = from_repository
4078
self.to_format = to_format
4079
4080
def delta_on_metadata(self):
4081
"""Return True if delta's are permitted on metadata streams.
4082
4083
That is on revisions and signatures.
4084
"""
4085
src_serializer = self.from_repository._format._serializer
4086
target_serializer = self.to_format._serializer
4087
return (self.to_format._fetch_uses_deltas and
4088
src_serializer == target_serializer)
4089
4090
def _fetch_revision_texts(self, revs):
4091
# fetch signatures first and then the revision texts
4092
# may need to be a InterRevisionStore call here.
4093
from_sf = self.from_repository.signatures
4094
# A missing signature is just skipped.
4095
keys = [(rev_id,) for rev_id in revs]
4096
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4097
keys,
4098
self.to_format._fetch_order,
4099
not self.to_format._fetch_uses_deltas))
4100
# If a revision has a delta, this is actually expanded inside the
4101
# insert_record_stream code now, which is an alternate fix for
4102
# bug #261339
4103
from_rf = self.from_repository.revisions
4104
revisions = from_rf.get_record_stream(
4105
keys,
4106
self.to_format._fetch_order,
4107
not self.delta_on_metadata())
4108
return [('signatures', signatures), ('revisions', revisions)]
4109
4110
def _generate_root_texts(self, revs):
4111
"""This will be called by __fetch between fetching weave texts and
4112
fetching the inventory weave.
4113
4114
Subclasses should override this if they need to generate root texts
4115
after fetching weave texts.
4116
"""
4117
if self._rich_root_upgrade():
4118
import bzrlib.fetch
4119
return bzrlib.fetch.Inter1and2Helper(
4120
self.from_repository).generate_root_texts(revs)
4121
else:
4122
return []
4123
4124
def get_stream(self, search):
4125
phase = 'file'
4126
revs = search.get_keys()
4127
graph = self.from_repository.get_graph()
4128
revs = list(graph.iter_topo_order(revs))
4129
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4130
text_keys = []
4131
for knit_kind, file_id, revisions in data_to_fetch:
4132
if knit_kind != phase:
4133
phase = knit_kind
4134
# Make a new progress bar for this phase
4135
if knit_kind == "file":
4136
# Accumulate file texts
4137
text_keys.extend([(file_id, revision) for revision in
4138
revisions])
4139
elif knit_kind == "inventory":
4140
# Now copy the file texts.
4141
from_texts = self.from_repository.texts
4142
yield ('texts', from_texts.get_record_stream(
4143
text_keys, self.to_format._fetch_order,
4144
not self.to_format._fetch_uses_deltas))
4145
# Cause an error if a text occurs after we have done the
4146
# copy.
4147
text_keys = None
4148
# Before we process the inventory we generate the root
4149
# texts (if necessary) so that the inventories references
4150
# will be valid.
4151
for _ in self._generate_root_texts(revs):
4152
yield _
4153
# NB: This currently reopens the inventory weave in source;
4154
# using a single stream interface instead would avoid this.
4155
from_weave = self.from_repository.inventories
4156
# we fetch only the referenced inventories because we do not
4157
# know for unselected inventories whether all their required
4158
# texts are present in the other repository - it could be
4159
# corrupt.
4160
for info in self._get_inventory_stream(revs):
4161
yield info
4162
elif knit_kind == "signatures":
4163
# Nothing to do here; this will be taken care of when
4164
# _fetch_revision_texts happens.
4165
pass
4166
elif knit_kind == "revisions":
4167
for record in self._fetch_revision_texts(revs):
4168
yield record
4169
else:
4170
raise AssertionError("Unknown knit kind %r" % knit_kind)
4171
4172
def get_stream_for_missing_keys(self, missing_keys):
4173
# missing keys can only occur when we are byte copying and not
4174
# translating (because translation means we don't send
4175
# unreconstructable deltas ever).
4176
keys = {}
4177
keys['texts'] = set()
4178
keys['revisions'] = set()
4179
keys['inventories'] = set()
4180
keys['signatures'] = set()
4181
for key in missing_keys:
4182
keys[key[0]].add(key[1:])
4183
if len(keys['revisions']):
4184
# If we allowed copying revisions at this point, we could end up
4185
# copying a revision without copying its required texts: a
4186
# violation of the requirements for repository integrity.
4187
raise AssertionError(
4188
'cannot copy revisions to fill in missing deltas %s' % (
4189
keys['revisions'],))
4190
for substream_kind, keys in keys.iteritems():
4191
vf = getattr(self.from_repository, substream_kind)
4192
# Ask for full texts always so that we don't need more round trips
4193
# after this stream.
4194
stream = vf.get_record_stream(keys,
4195
self.to_format._fetch_order, True)
4196
yield substream_kind, stream
4197
4198
def inventory_fetch_order(self):
4199
if self._rich_root_upgrade():
4200
return 'topological'
4201
else:
4202
return self.to_format._fetch_order
4203
4204
def _rich_root_upgrade(self):
4205
return (not self.from_repository._format.rich_root_data and
4206
self.to_format.rich_root_data)
4207
4208
def _get_inventory_stream(self, revision_ids):
4209
from_format = self.from_repository._format
4210
if (from_format.supports_chks and self.to_format.supports_chks
4211
and (from_format._serializer == self.to_format._serializer)):
4212
# Both sides support chks, and they use the same serializer, so it
4213
# is safe to transmit the chk pages and inventory pages across
4214
# as-is.
4215
return self._get_chk_inventory_stream(revision_ids)
4216
elif (not from_format.supports_chks):
4217
# Source repository doesn't support chks. So we can transmit the
4218
# inventories 'as-is' and either they are just accepted on the
4219
# target, or the Sink will properly convert it.
4220
return self._get_simple_inventory_stream(revision_ids)
4221
else:
4222
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4223
# inventories. Note that this should probably be done somehow
4224
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4225
# figure out how a non-chk repository could possibly handle
4226
# deserializing an inventory stream from a chk repo, as it
4227
# doesn't have a way to understand individual pages.
4228
return self._get_convertable_inventory_stream(revision_ids)
4229
4230
def _get_simple_inventory_stream(self, revision_ids):
4231
from_weave = self.from_repository.inventories
4232
yield ('inventories', from_weave.get_record_stream(
4233
[(rev_id,) for rev_id in revision_ids],
4234
self.inventory_fetch_order(),
4235
not self.delta_on_metadata()))
4236
4237
def _get_chk_inventory_stream(self, revision_ids):
4238
"""Fetch the inventory texts, along with the associated chk maps."""
4239
# We want an inventory outside of the search set, so that we can filter
4240
# out uninteresting chk pages. For now we use
4241
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4242
# could use that instead.
4243
start_rev_id = self.from_repository._find_revision_outside_set(
4244
revision_ids)
4245
start_rev_key = (start_rev_id,)
4246
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4247
if start_rev_id != _mod_revision.NULL_REVISION:
4248
inv_keys_to_fetch.append((start_rev_id,))
4249
# Any repo that supports chk_bytes must also support out-of-order
4250
# insertion. At least, that is how we expect it to work
4251
# We use get_record_stream instead of iter_inventories because we want
4252
# to be able to insert the stream as well. We could instead fetch
4253
# allowing deltas, and then iter_inventories, but we don't know whether
4254
# source or target is more 'local' anway.
4255
inv_stream = self.from_repository.inventories.get_record_stream(
4256
inv_keys_to_fetch, 'unordered',
4257
True) # We need them as full-texts so we can find their references
4258
uninteresting_chk_roots = set()
4259
interesting_chk_roots = set()
4260
def filter_inv_stream(inv_stream):
4261
for idx, record in enumerate(inv_stream):
4262
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4263
bytes = record.get_bytes_as('fulltext')
4264
chk_inv = inventory.CHKInventory.deserialise(
4265
self.from_repository.chk_bytes, bytes, record.key)
4266
if record.key == start_rev_key:
4267
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4268
p_id_map = chk_inv.parent_id_basename_to_file_id
4269
if p_id_map is not None:
4270
uninteresting_chk_roots.add(p_id_map.key())
4271
else:
4272
yield record
4273
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4274
p_id_map = chk_inv.parent_id_basename_to_file_id
4275
if p_id_map is not None:
4276
interesting_chk_roots.add(p_id_map.key())
4277
### pb.update('fetch inventory', 0, 2)
4278
yield ('inventories', filter_inv_stream(inv_stream))
4279
# Now that we have worked out all of the interesting root nodes, grab
4280
# all of the interesting pages and insert them
4281
### pb.update('fetch inventory', 1, 2)
4282
interesting = chk_map.iter_interesting_nodes(
4283
self.from_repository.chk_bytes, interesting_chk_roots,
4284
uninteresting_chk_roots)
4285
def to_stream_adapter():
4286
"""Adapt the iter_interesting_nodes result to a single stream.
4287
4288
iter_interesting_nodes returns records as it processes them, along
4289
with keys. However, we only want to return the records themselves.
4290
"""
4291
for record, items in interesting:
4292
if record is not None:
4293
yield record
4294
# XXX: We could instead call get_record_stream(records.keys())
4295
# ATM, this will always insert the records as fulltexts, and
4296
# requires that you can hang on to records once you have gone
4297
# on to the next one. Further, it causes the target to
4298
# recompress the data. Testing shows it to be faster than
4299
# requesting the records again, though.
4300
yield ('chk_bytes', to_stream_adapter())
4301
### pb.update('fetch inventory', 2, 2)
4302
4303
def _get_convertable_inventory_stream(self, revision_ids):
4304
# XXX: One of source or target is using chks, and they don't have
4305
# compatible serializations. The StreamSink code expects to be
4306
# able to convert on the target, so we need to put
4307
# bytes-on-the-wire that can be converted
4308
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4309
4310
def _stream_invs_as_fulltexts(self, revision_ids):
4311
from_repo = self.from_repository
4312
from_serializer = from_repo._format._serializer
4313
revision_keys = [(rev_id,) for rev_id in revision_ids]
4314
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4315
for inv in self.from_repository.iter_inventories(revision_ids):
4316
# XXX: This is a bit hackish, but it works. Basically,
4317
# CHKSerializer 'accidentally' supports
4318
# read/write_inventory_to_string, even though that is never
4319
# the format that is stored on disk. It *does* give us a
4320
# single string representation for an inventory, so live with
4321
# it for now.
4322
# This would be far better if we had a 'serialized inventory
4323
# delta' form. Then we could use 'inventory._make_delta', and
4324
# transmit that. This would both be faster to generate, and
4325
# result in fewer bytes-on-the-wire.
4326
as_bytes = from_serializer.write_inventory_to_string(inv)
4327
key = (inv.revision_id,)
4328
parent_keys = parent_map.get(key, ())
4329
yield versionedfile.FulltextContentFactory(
4330
key, parent_keys, None, as_bytes)
4331
Loggerhead is a web-based interface for Breezy
Version: 2.0.1