1
# Copyright (C) 2007-2010 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from __future__ import absolute_import
19
from cStringIO import StringIO
30
revision as _mod_revision,
34
versionedfile as _mod_versionedfile,
36
from bzrlib.bundle import bundle_data, serializer as bundle_serializer
37
from bzrlib.i18n import ngettext
38
from bzrlib import bencode
41
class _MPDiffInventoryGenerator(_mod_versionedfile._MPDiffGenerator):
42
"""Generate Inventory diffs serialized inventories."""
44
def __init__(self, repo, inventory_keys):
45
super(_MPDiffInventoryGenerator, self).__init__(repo.inventories,
51
"""Compute the diffs one at a time."""
52
# This is instead of compute_diffs() since we guarantee our ordering of
53
# inventories, we don't have to do any buffering
54
self._find_needed_keys()
55
# We actually use a slightly different ordering. We grab all of the
56
# parents first, and then grab the ordered requests.
57
needed_ids = [k[-1] for k in self.present_parents]
58
needed_ids.extend([k[-1] for k in self.ordered_keys])
59
inv_to_str = self.repo._serializer.write_inventory_to_string
60
for inv in self.repo.iter_inventories(needed_ids):
61
revision_id = inv.revision_id
63
if key in self.present_parents:
64
# Not a key we will transmit, which is a shame, since because
65
# of that bundles don't work with stacked branches
68
parent_ids = [k[-1] for k in self.parent_map[key]]
69
as_bytes = inv_to_str(inv)
70
self._process_one_record(key, (as_bytes,))
71
if parent_ids is None:
73
diff = self.diffs.pop(key)
74
sha1 = osutils.sha_string(as_bytes)
75
yield revision_id, parent_ids, sha1, diff
78
class BundleWriter(object):
79
"""Writer for bundle-format files.
81
This serves roughly the same purpose as ContainerReader, but acts as a
84
Provides ways of writing the specific record types supported this bundle
88
def __init__(self, fileobj):
89
self._container = pack.ContainerWriter(self._write_encoded)
90
self._fileobj = fileobj
91
self._compressor = bz2.BZ2Compressor()
93
def _write_encoded(self, bytes):
94
"""Write bzip2-encoded bytes to the file"""
95
self._fileobj.write(self._compressor.compress(bytes))
98
"""Start writing the bundle"""
99
self._fileobj.write(bundle_serializer._get_bundle_header(
100
bundle_serializer.v4_string))
101
self._fileobj.write('#\n')
102
self._container.begin()
105
"""Finish writing the bundle"""
106
self._container.end()
107
self._fileobj.write(self._compressor.flush())
109
def add_multiparent_record(self, mp_bytes, sha1, parents, repo_kind,
110
revision_id, file_id):
111
"""Add a record for a multi-parent diff
113
:mp_bytes: A multi-parent diff, as a bytestring
114
:sha1: The sha1 hash of the fulltext
115
:parents: a list of revision-ids of the parents
116
:repo_kind: The kind of object in the repository. May be 'file' or
118
:revision_id: The revision id of the mpdiff being added.
119
:file_id: The file-id of the file, or None for inventories.
121
metadata = {'parents': parents,
122
'storage_kind': 'mpdiff',
124
self._add_record(mp_bytes, metadata, repo_kind, revision_id, file_id)
126
def add_fulltext_record(self, bytes, parents, repo_kind, revision_id):
127
"""Add a record for a fulltext
129
:bytes: The fulltext, as a bytestring
130
:parents: a list of revision-ids of the parents
131
:repo_kind: The kind of object in the repository. May be 'revision' or
133
:revision_id: The revision id of the fulltext being added.
135
metadata = {'parents': parents,
136
'storage_kind': 'mpdiff'}
137
self._add_record(bytes, {'parents': parents,
138
'storage_kind': 'fulltext'}, repo_kind, revision_id, None)
140
def add_info_record(self, **kwargs):
141
"""Add an info record to the bundle
143
Any parameters may be supplied, except 'self' and 'storage_kind'.
144
Values must be lists, strings, integers, dicts, or a combination.
146
kwargs['storage_kind'] = 'header'
147
self._add_record(None, kwargs, 'info', None, None)
150
def encode_name(content_kind, revision_id, file_id=None):
151
"""Encode semantic ids as a container name"""
152
if content_kind not in ('revision', 'file', 'inventory', 'signature',
154
raise ValueError(content_kind)
155
if content_kind == 'file':
157
raise AssertionError()
159
if file_id is not None:
160
raise AssertionError()
161
if content_kind == 'info':
162
if revision_id is not None:
163
raise AssertionError()
164
elif revision_id is None:
165
raise AssertionError()
166
names = [n.replace('/', '//') for n in
167
(content_kind, revision_id, file_id) if n is not None]
168
return '/'.join(names)
170
def _add_record(self, bytes, metadata, repo_kind, revision_id, file_id):
171
"""Add a bundle record to the container.
173
Most bundle records are recorded as header/body pairs, with the
174
body being nameless. Records with storage_kind 'header' have no
177
name = self.encode_name(repo_kind, revision_id, file_id)
178
encoded_metadata = bencode.bencode(metadata)
179
self._container.add_bytes_record(encoded_metadata, [(name, )])
180
if metadata['storage_kind'] != 'header':
181
self._container.add_bytes_record(bytes, [])
184
class BundleReader(object):
185
"""Reader for bundle-format files.
187
This serves roughly the same purpose as ContainerReader, but acts as a
188
layer on top of it, providing metadata, a semantic name, and a record
192
def __init__(self, fileobj, stream_input=True):
195
:param fileobj: a file containing a bzip-encoded container
196
:param stream_input: If True, the BundleReader stream input rather than
197
reading it all into memory at once. Reading it into memory all at
198
once is (currently) faster.
200
line = fileobj.readline()
203
self.patch_lines = []
205
source_file = iterablefile.IterableFile(self.iter_decode(fileobj))
207
source_file = StringIO(bz2.decompress(fileobj.read()))
208
self._container_file = source_file
211
def iter_decode(fileobj):
212
"""Iterate through decoded fragments of the file"""
213
decompressor = bz2.BZ2Decompressor()
216
yield decompressor.decompress(line)
221
def decode_name(name):
222
"""Decode a name from its container form into a semantic form
224
:retval: content_kind, revision_id, file_id
226
segments = re.split('(//?)', name)
228
for segment in segments:
235
content_kind = names[0]
239
revision_id = names[1]
242
return content_kind, revision_id, file_id
244
def iter_records(self):
245
"""Iterate through bundle records
247
:return: a generator of (bytes, metadata, content_kind, revision_id,
250
iterator = pack.iter_records_from_file(self._container_file)
251
for names, bytes in iterator:
253
raise errors.BadBundle('Record has %d names instead of 1'
255
metadata = bencode.bdecode(bytes)
256
if metadata['storage_kind'] == 'header':
259
_unused, bytes = iterator.next()
260
yield (bytes, metadata) + self.decode_name(names[0][0])
263
class BundleSerializerV4(bundle_serializer.BundleSerializer):
264
"""Implement the high-level bundle interface"""
266
def write(self, repository, revision_ids, forced_bases, fileobj):
267
"""Write a bundle to a file-like object
269
For backwards-compatibility only
271
write_op = BundleWriteOperation.from_old_args(repository, revision_ids,
272
forced_bases, fileobj)
273
return write_op.do_write()
275
def write_bundle(self, repository, target, base, fileobj):
276
"""Write a bundle to a file object
278
:param repository: The repository to retrieve revision data from
279
:param target: The head revision to include ancestors of
280
:param base: The ancestor of the target to stop including acestors
282
:param fileobj: The file-like object to write to
284
write_op = BundleWriteOperation(base, target, repository, fileobj)
285
return write_op.do_write()
287
def read(self, file):
288
"""return a reader object for a given file"""
289
bundle = BundleInfoV4(file, self)
293
def get_source_serializer(info):
294
"""Retrieve the serializer for a given info object"""
295
return serializer.format_registry.get(info['serializer'])
298
class BundleWriteOperation(object):
299
"""Perform the operation of writing revisions to a bundle"""
302
def from_old_args(cls, repository, revision_ids, forced_bases, fileobj):
303
"""Create a BundleWriteOperation from old-style arguments"""
304
base, target = cls.get_base_target(revision_ids, forced_bases,
306
return BundleWriteOperation(base, target, repository, fileobj,
309
def __init__(self, base, target, repository, fileobj, revision_ids=None):
312
self.repository = repository
313
bundle = BundleWriter(fileobj)
315
if revision_ids is not None:
316
self.revision_ids = revision_ids
318
graph = repository.get_graph()
319
revision_ids = graph.find_unique_ancestors(target, [base])
321
parents = graph.get_parent_map(revision_ids)
322
self.revision_ids = [r for r in revision_ids if r in parents]
323
self.revision_keys = set([(revid,) for revid in self.revision_ids])
326
"""Write all data to the bundle"""
327
trace.note(ngettext('Bundling %d revision.', 'Bundling %d revisions.',
328
len(self.revision_ids)), len(self.revision_ids))
329
self.repository.lock_read()
334
self.write_revisions()
337
self.repository.unlock()
338
return self.revision_ids
340
def write_info(self):
341
"""Write format info"""
342
serializer_format = self.repository.get_serializer_format()
343
supports_rich_root = {True: 1, False: 0}[
344
self.repository.supports_rich_root()]
345
self.bundle.add_info_record(serializer=serializer_format,
346
supports_rich_root=supports_rich_root)
348
def write_files(self):
349
"""Write bundle records for all revisions of all files"""
351
altered_fileids = self.repository.fileids_altered_by_revision_ids(
353
for file_id, revision_ids in altered_fileids.iteritems():
354
for revision_id in revision_ids:
355
text_keys.append((file_id, revision_id))
356
self._add_mp_records_keys('file', self.repository.texts, text_keys)
358
def write_revisions(self):
359
"""Write bundle records for all revisions and signatures"""
360
inv_vf = self.repository.inventories
361
topological_order = [key[-1] for key in multiparent.topo_iter_keys(
362
inv_vf, self.revision_keys)]
363
revision_order = topological_order
364
if self.target is not None and self.target in self.revision_ids:
365
# Make sure the target revision is always the last entry
366
revision_order = list(topological_order)
367
revision_order.remove(self.target)
368
revision_order.append(self.target)
369
if self.repository._serializer.support_altered_by_hack:
370
# Repositories that support_altered_by_hack means that
371
# inventories.make_mpdiffs() contains all the data about the tree
372
# shape. Formats without support_altered_by_hack require
373
# chk_bytes/etc, so we use a different code path.
374
self._add_mp_records_keys('inventory', inv_vf,
375
[(revid,) for revid in topological_order])
377
# Inventories should always be added in pure-topological order, so
378
# that we can apply the mpdiff for the child to the parent texts.
379
self._add_inventory_mpdiffs_from_serializer(topological_order)
380
self._add_revision_texts(revision_order)
382
def _add_inventory_mpdiffs_from_serializer(self, revision_order):
383
"""Generate mpdiffs by serializing inventories.
385
The current repository only has part of the tree shape information in
386
the 'inventories' vf. So we use serializer.write_inventory_to_string to
387
get a 'full' representation of the tree shape, and then generate
388
mpdiffs on that data stream. This stream can then be reconstructed on
391
inventory_key_order = [(r,) for r in revision_order]
392
generator = _MPDiffInventoryGenerator(self.repository,
394
for revision_id, parent_ids, sha1, diff in generator.iter_diffs():
395
text = ''.join(diff.to_patch())
396
self.bundle.add_multiparent_record(text, sha1, parent_ids,
397
'inventory', revision_id, None)
399
def _add_revision_texts(self, revision_order):
400
parent_map = self.repository.get_parent_map(revision_order)
401
revision_to_str = self.repository._serializer.write_revision_to_string
402
revisions = self.repository.get_revisions(revision_order)
403
for revision in revisions:
404
revision_id = revision.revision_id
405
parents = parent_map.get(revision_id, None)
406
revision_text = revision_to_str(revision)
407
self.bundle.add_fulltext_record(revision_text, parents,
408
'revision', revision_id)
410
self.bundle.add_fulltext_record(
411
self.repository.get_signature_text(
412
revision_id), parents, 'signature', revision_id)
413
except errors.NoSuchRevision:
417
def get_base_target(revision_ids, forced_bases, repository):
418
"""Determine the base and target from old-style revision ids"""
419
if len(revision_ids) == 0:
421
target = revision_ids[0]
422
base = forced_bases.get(target)
424
parents = repository.get_revision(target).parent_ids
425
if len(parents) == 0:
426
base = _mod_revision.NULL_REVISION
431
def _add_mp_records_keys(self, repo_kind, vf, keys):
432
"""Add multi-parent diff records to a bundle"""
433
ordered_keys = list(multiparent.topo_iter_keys(vf, keys))
434
mpdiffs = vf.make_mpdiffs(ordered_keys)
435
sha1s = vf.get_sha1s(ordered_keys)
436
parent_map = vf.get_parent_map(ordered_keys)
437
for mpdiff, item_key, in zip(mpdiffs, ordered_keys):
438
sha1 = sha1s[item_key]
439
parents = [key[-1] for key in parent_map[item_key]]
440
text = ''.join(mpdiff.to_patch())
441
# Infer file id records as appropriate.
442
if len(item_key) == 2:
443
file_id = item_key[0]
446
self.bundle.add_multiparent_record(text, sha1, parents, repo_kind,
447
item_key[-1], file_id)
450
class BundleInfoV4(object):
452
"""Provide (most of) the BundleInfo interface"""
453
def __init__(self, fileobj, serializer):
454
self._fileobj = fileobj
455
self._serializer = serializer
456
self.__real_revisions = None
457
self.__revisions = None
459
def install(self, repository):
460
return self.install_revisions(repository)
462
def install_revisions(self, repository, stream_input=True):
463
"""Install this bundle's revisions into the specified repository
465
:param target_repo: The repository to install into
466
:param stream_input: If True, will stream input rather than reading it
467
all into memory at once. Reading it into memory all at once is
470
repository.lock_write()
472
ri = RevisionInstaller(self.get_bundle_reader(stream_input),
473
self._serializer, repository)
478
def get_merge_request(self, target_repo):
479
"""Provide data for performing a merge
481
Returns suggested base, suggested target, and patch verification status
483
return None, self.target, 'inapplicable'
485
def get_bundle_reader(self, stream_input=True):
486
"""Return a new BundleReader for the associated bundle
488
:param stream_input: If True, the BundleReader stream input rather than
489
reading it all into memory at once. Reading it into memory all at
490
once is (currently) faster.
492
self._fileobj.seek(0)
493
return BundleReader(self._fileobj, stream_input)
495
def _get_real_revisions(self):
496
if self.__real_revisions is None:
497
self.__real_revisions = []
498
bundle_reader = self.get_bundle_reader()
499
for bytes, metadata, repo_kind, revision_id, file_id in \
500
bundle_reader.iter_records():
501
if repo_kind == 'info':
503
self._serializer.get_source_serializer(metadata)
504
if repo_kind == 'revision':
505
rev = serializer.read_revision_from_string(bytes)
506
self.__real_revisions.append(rev)
507
return self.__real_revisions
508
real_revisions = property(_get_real_revisions)
510
def _get_revisions(self):
511
if self.__revisions is None:
512
self.__revisions = []
513
for revision in self.real_revisions:
514
self.__revisions.append(
515
bundle_data.RevisionInfo.from_revision(revision))
516
return self.__revisions
518
revisions = property(_get_revisions)
520
def _get_target(self):
521
return self.revisions[-1].revision_id
523
target = property(_get_target)
526
class RevisionInstaller(object):
527
"""Installs revisions into a repository"""
529
def __init__(self, container, serializer, repository):
530
self._container = container
531
self._serializer = serializer
532
self._repository = repository
536
"""Perform the installation.
538
Must be called with the Repository locked.
540
self._repository.start_write_group()
542
result = self._install_in_write_group()
544
self._repository.abort_write_group()
546
self._repository.commit_write_group()
549
def _install_in_write_group(self):
551
current_versionedfile = None
552
pending_file_records = []
554
pending_inventory_records = []
556
target_revision = None
557
for bytes, metadata, repo_kind, revision_id, file_id in\
558
self._container.iter_records():
559
if repo_kind == 'info':
560
if self._info is not None:
561
raise AssertionError()
562
self._handle_info(metadata)
563
if (pending_file_records and
564
(repo_kind, file_id) != ('file', current_file)):
565
# Flush the data for a single file - prevents memory
566
# spiking due to buffering all files in memory.
567
self._install_mp_records_keys(self._repository.texts,
568
pending_file_records)
570
del pending_file_records[:]
571
if len(pending_inventory_records) > 0 and repo_kind != 'inventory':
572
self._install_inventory_records(pending_inventory_records)
573
pending_inventory_records = []
574
if repo_kind == 'inventory':
575
pending_inventory_records.append(((revision_id,), metadata, bytes))
576
if repo_kind == 'revision':
577
target_revision = revision_id
578
self._install_revision(revision_id, metadata, bytes)
579
if repo_kind == 'signature':
580
self._install_signature(revision_id, metadata, bytes)
581
if repo_kind == 'file':
582
current_file = file_id
583
pending_file_records.append(((file_id, revision_id), metadata, bytes))
584
self._install_mp_records_keys(self._repository.texts, pending_file_records)
585
return target_revision
587
def _handle_info(self, info):
588
"""Extract data from an info record"""
590
self._source_serializer = self._serializer.get_source_serializer(info)
591
if (info['supports_rich_root'] == 0 and
592
self._repository.supports_rich_root()):
593
self.update_root = True
595
self.update_root = False
597
def _install_mp_records(self, versionedfile, records):
598
if len(records) == 0:
600
d_func = multiparent.MultiParent.from_patch
601
vf_records = [(r, m['parents'], m['sha1'], d_func(t)) for r, m, t in
602
records if r not in versionedfile]
603
versionedfile.add_mpdiffs(vf_records)
605
def _install_mp_records_keys(self, versionedfile, records):
606
d_func = multiparent.MultiParent.from_patch
608
for key, meta, text in records:
609
# Adapt to tuple interface: A length two key is a file_id,
610
# revision_id pair, a length 1 key is a
611
# revision/signature/inventory. We need to do this because
612
# the metadata extraction from the bundle has not yet been updated
613
# to use the consistent tuple interface itself.
618
parents = [prefix + (parent,) for parent in meta['parents']]
619
vf_records.append((key, parents, meta['sha1'], d_func(text)))
620
versionedfile.add_mpdiffs(vf_records)
622
def _get_parent_inventory_texts(self, inventory_text_cache,
623
inventory_cache, parent_ids):
624
cached_parent_texts = {}
625
remaining_parent_ids = []
626
for parent_id in parent_ids:
627
p_text = inventory_text_cache.get(parent_id, None)
629
remaining_parent_ids.append(parent_id)
631
cached_parent_texts[parent_id] = p_text
633
# TODO: Use inventory_cache to grab inventories we already have in
635
if remaining_parent_ids:
636
# first determine what keys are actually present in the local
637
# inventories object (don't use revisions as they haven't been
639
parent_keys = [(r,) for r in remaining_parent_ids]
640
present_parent_map = self._repository.inventories.get_parent_map(
642
present_parent_ids = []
644
for p_id in remaining_parent_ids:
645
if (p_id,) in present_parent_map:
646
present_parent_ids.append(p_id)
649
to_string = self._source_serializer.write_inventory_to_string
650
for parent_inv in self._repository.iter_inventories(
652
p_text = to_string(parent_inv)
653
inventory_cache[parent_inv.revision_id] = parent_inv
654
cached_parent_texts[parent_inv.revision_id] = p_text
655
inventory_text_cache[parent_inv.revision_id] = p_text
657
parent_texts = [cached_parent_texts[parent_id]
658
for parent_id in parent_ids
659
if parent_id not in ghosts]
662
def _install_inventory_records(self, records):
663
if (self._info['serializer'] == self._repository._serializer.format_num
664
and self._repository._serializer.support_altered_by_hack):
665
return self._install_mp_records_keys(self._repository.inventories,
667
# Use a 10MB text cache, since these are string xml inventories. Note
668
# that 10MB is fairly small for large projects (a single inventory can
669
# be >5MB). Another possibility is to cache 10-20 inventory texts
671
inventory_text_cache = lru_cache.LRUSizeCache(10*1024*1024)
672
# Also cache the in-memory representation. This allows us to create
673
# inventory deltas to apply rather than calling add_inventory from
675
inventory_cache = lru_cache.LRUCache(10)
676
pb = ui.ui_factory.nested_progress_bar()
678
num_records = len(records)
679
for idx, (key, metadata, bytes) in enumerate(records):
680
pb.update('installing inventory', idx, num_records)
681
revision_id = key[-1]
682
parent_ids = metadata['parents']
683
# Note: This assumes the local ghosts are identical to the
684
# ghosts in the source, as the Bundle serialization
685
# format doesn't record ghosts.
686
p_texts = self._get_parent_inventory_texts(inventory_text_cache,
689
# Why does to_lines() take strings as the source, it seems that
690
# it would have to cast to a list of lines, which we get back
691
# as lines and then cast back to a string.
692
target_lines = multiparent.MultiParent.from_patch(bytes
694
inv_text = ''.join(target_lines)
696
sha1 = osutils.sha_string(inv_text)
697
if sha1 != metadata['sha1']:
698
raise errors.BadBundle("Can't convert to target format")
699
# Add this to the cache so we don't have to extract it again.
700
inventory_text_cache[revision_id] = inv_text
701
target_inv = self._source_serializer.read_inventory_from_string(
703
self._handle_root(target_inv, parent_ids)
706
parent_inv = inventory_cache.get(parent_ids[0], None)
708
if parent_inv is None:
709
self._repository.add_inventory(revision_id, target_inv,
712
delta = target_inv._make_delta(parent_inv)
713
self._repository.add_inventory_by_delta(parent_ids[0],
714
delta, revision_id, parent_ids)
715
except errors.UnsupportedInventoryKind:
716
raise errors.IncompatibleRevision(repr(self._repository))
717
inventory_cache[revision_id] = target_inv
721
def _handle_root(self, target_inv, parent_ids):
722
revision_id = target_inv.revision_id
724
text_key = (target_inv.root.file_id, revision_id)
725
parent_keys = [(target_inv.root.file_id, parent) for
726
parent in parent_ids]
727
self._repository.texts.add_lines(text_key, parent_keys, [])
728
elif not self._repository.supports_rich_root():
729
if target_inv.root.revision != revision_id:
730
raise errors.IncompatibleRevision(repr(self._repository))
732
def _install_revision(self, revision_id, metadata, text):
733
if self._repository.has_revision(revision_id):
735
revision = self._source_serializer.read_revision_from_string(text)
736
self._repository.add_revision(revision.revision_id, revision)
738
def _install_signature(self, revision_id, metadata, text):
739
transaction = self._repository.get_transaction()
740
if self._repository.has_signature_for_revision_id(revision_id):
742
self._repository.add_signature_text(revision_id, text)
added
removed
bzrlib/bundle/serializer/v4.py
