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 cStringIO import StringIO
29
revision as _mod_revision,
33
versionedfile as _mod_versionedfile,
35
from bzrlib.bundle import bundle_data, serializer as bundle_serializer
36
from bzrlib.i18n import gettext, ngettext
37
from bzrlib import bencode
40
class _MPDiffInventoryGenerator(_mod_versionedfile._MPDiffGenerator):
41
"""Generate Inventory diffs serialized inventories."""
43
def __init__(self, repo, inventory_keys):
44
super(_MPDiffInventoryGenerator, self).__init__(repo.inventories,
50
"""Compute the diffs one at a time."""
51
# This is instead of compute_diffs() since we guarantee our ordering of
52
# inventories, we don't have to do any buffering
53
self._find_needed_keys()
54
# We actually use a slightly different ordering. We grab all of the
55
# parents first, and then grab the ordered requests.
56
needed_ids = [k[-1] for k in self.present_parents]
57
needed_ids.extend([k[-1] for k in self.ordered_keys])
58
inv_to_str = self.repo._serializer.write_inventory_to_string
59
for inv in self.repo.iter_inventories(needed_ids):
60
revision_id = inv.revision_id
62
if key in self.present_parents:
63
# Not a key we will transmit, which is a shame, since because
64
# of that bundles don't work with stacked branches
67
parent_ids = [k[-1] for k in self.parent_map[key]]
68
as_bytes = inv_to_str(inv)
69
self._process_one_record(key, (as_bytes,))
70
if parent_ids is None:
72
diff = self.diffs.pop(key)
73
sha1 = osutils.sha_string(as_bytes)
74
yield revision_id, parent_ids, sha1, diff
77
class BundleWriter(object):
78
"""Writer for bundle-format files.
80
This serves roughly the same purpose as ContainerReader, but acts as a
83
Provides ways of writing the specific record types supported this bundle
87
def __init__(self, fileobj):
88
self._container = pack.ContainerWriter(self._write_encoded)
89
self._fileobj = fileobj
90
self._compressor = bz2.BZ2Compressor()
92
def _write_encoded(self, bytes):
93
"""Write bzip2-encoded bytes to the file"""
94
self._fileobj.write(self._compressor.compress(bytes))
97
"""Start writing the bundle"""
98
self._fileobj.write(bundle_serializer._get_bundle_header(
99
bundle_serializer.v4_string))
100
self._fileobj.write('#\n')
101
self._container.begin()
104
"""Finish writing the bundle"""
105
self._container.end()
106
self._fileobj.write(self._compressor.flush())
108
def add_multiparent_record(self, mp_bytes, sha1, parents, repo_kind,
109
revision_id, file_id):
110
"""Add a record for a multi-parent diff
112
:mp_bytes: A multi-parent diff, as a bytestring
113
:sha1: The sha1 hash of the fulltext
114
:parents: a list of revision-ids of the parents
115
:repo_kind: The kind of object in the repository. May be 'file' or
117
:revision_id: The revision id of the mpdiff being added.
118
:file_id: The file-id of the file, or None for inventories.
120
metadata = {'parents': parents,
121
'storage_kind': 'mpdiff',
123
self._add_record(mp_bytes, metadata, repo_kind, revision_id, file_id)
125
def add_fulltext_record(self, bytes, parents, repo_kind, revision_id):
126
"""Add a record for a fulltext
128
:bytes: The fulltext, as a bytestring
129
:parents: a list of revision-ids of the parents
130
:repo_kind: The kind of object in the repository. May be 'revision' or
132
:revision_id: The revision id of the fulltext being added.
134
metadata = {'parents': parents,
135
'storage_kind': 'mpdiff'}
136
self._add_record(bytes, {'parents': parents,
137
'storage_kind': 'fulltext'}, repo_kind, revision_id, None)
139
def add_info_record(self, **kwargs):
140
"""Add an info record to the bundle
142
Any parameters may be supplied, except 'self' and 'storage_kind'.
143
Values must be lists, strings, integers, dicts, or a combination.
145
kwargs['storage_kind'] = 'header'
146
self._add_record(None, kwargs, 'info', None, None)
149
def encode_name(content_kind, revision_id, file_id=None):
150
"""Encode semantic ids as a container name"""
151
if content_kind not in ('revision', 'file', 'inventory', 'signature',
153
raise ValueError(content_kind)
154
if content_kind == 'file':
156
raise AssertionError()
158
if file_id is not None:
159
raise AssertionError()
160
if content_kind == 'info':
161
if revision_id is not None:
162
raise AssertionError()
163
elif revision_id is None:
164
raise AssertionError()
165
names = [n.replace('/', '//') for n in
166
(content_kind, revision_id, file_id) if n is not None]
167
return '/'.join(names)
169
def _add_record(self, bytes, metadata, repo_kind, revision_id, file_id):
170
"""Add a bundle record to the container.
172
Most bundle records are recorded as header/body pairs, with the
173
body being nameless. Records with storage_kind 'header' have no
176
name = self.encode_name(repo_kind, revision_id, file_id)
177
encoded_metadata = bencode.bencode(metadata)
178
self._container.add_bytes_record(encoded_metadata, [(name, )])
179
if metadata['storage_kind'] != 'header':
180
self._container.add_bytes_record(bytes, [])
183
class BundleReader(object):
184
"""Reader for bundle-format files.
186
This serves roughly the same purpose as ContainerReader, but acts as a
187
layer on top of it, providing metadata, a semantic name, and a record
191
def __init__(self, fileobj, stream_input=True):
194
:param fileobj: a file containing a bzip-encoded container
195
:param stream_input: If True, the BundleReader stream input rather than
196
reading it all into memory at once. Reading it into memory all at
197
once is (currently) faster.
199
line = fileobj.readline()
202
self.patch_lines = []
204
source_file = iterablefile.IterableFile(self.iter_decode(fileobj))
206
source_file = StringIO(bz2.decompress(fileobj.read()))
207
self._container_file = source_file
210
def iter_decode(fileobj):
211
"""Iterate through decoded fragments of the file"""
212
decompressor = bz2.BZ2Decompressor()
215
yield decompressor.decompress(line)
220
def decode_name(name):
221
"""Decode a name from its container form into a semantic form
223
:retval: content_kind, revision_id, file_id
225
segments = re.split('(//?)', name)
227
for segment in segments:
234
content_kind = names[0]
238
revision_id = names[1]
241
return content_kind, revision_id, file_id
243
def iter_records(self):
244
"""Iterate through bundle records
246
:return: a generator of (bytes, metadata, content_kind, revision_id,
249
iterator = pack.iter_records_from_file(self._container_file)
250
for names, bytes in iterator:
252
raise errors.BadBundle('Record has %d names instead of 1'
254
metadata = bencode.bdecode(bytes)
255
if metadata['storage_kind'] == 'header':
258
_unused, bytes = iterator.next()
259
yield (bytes, metadata) + self.decode_name(names[0][0])
262
class BundleSerializerV4(bundle_serializer.BundleSerializer):
263
"""Implement the high-level bundle interface"""
265
def write(self, repository, revision_ids, forced_bases, fileobj):
266
"""Write a bundle to a file-like object
268
For backwards-compatibility only
270
write_op = BundleWriteOperation.from_old_args(repository, revision_ids,
271
forced_bases, fileobj)
272
return write_op.do_write()
274
def write_bundle(self, repository, target, base, fileobj):
275
"""Write a bundle to a file object
277
:param repository: The repository to retrieve revision data from
278
:param target: The head revision to include ancestors of
279
:param base: The ancestor of the target to stop including acestors
281
:param fileobj: The file-like object to write to
283
write_op = BundleWriteOperation(base, target, repository, fileobj)
284
return write_op.do_write()
286
def read(self, file):
287
"""return a reader object for a given file"""
288
bundle = BundleInfoV4(file, self)
292
def get_source_serializer(info):
293
"""Retrieve the serializer for a given info object"""
294
return serializer.format_registry.get(info['serializer'])
297
class BundleWriteOperation(object):
298
"""Perform the operation of writing revisions to a bundle"""
301
def from_old_args(cls, repository, revision_ids, forced_bases, fileobj):
302
"""Create a BundleWriteOperation from old-style arguments"""
303
base, target = cls.get_base_target(revision_ids, forced_bases,
305
return BundleWriteOperation(base, target, repository, fileobj,
308
def __init__(self, base, target, repository, fileobj, revision_ids=None):
311
self.repository = repository
312
bundle = BundleWriter(fileobj)
314
if revision_ids is not None:
315
self.revision_ids = revision_ids
317
graph = repository.get_graph()
318
revision_ids = graph.find_unique_ancestors(target, [base])
320
parents = graph.get_parent_map(revision_ids)
321
self.revision_ids = [r for r in revision_ids if r in parents]
322
self.revision_keys = set([(revid,) for revid in self.revision_ids])
325
"""Write all data to the bundle"""
326
trace.note(ngettext('Bundling %d revision.', 'Bundling %d revisions.',
327
len(self.revision_ids)), len(self.revision_ids))
328
self.repository.lock_read()
333
self.write_revisions()
336
self.repository.unlock()
337
return self.revision_ids
339
def write_info(self):
340
"""Write format info"""
341
serializer_format = self.repository.get_serializer_format()
342
supports_rich_root = {True: 1, False: 0}[
343
self.repository.supports_rich_root()]
344
self.bundle.add_info_record(serializer=serializer_format,
345
supports_rich_root=supports_rich_root)
347
def write_files(self):
348
"""Write bundle records for all revisions of all files"""
350
altered_fileids = self.repository.fileids_altered_by_revision_ids(
352
for file_id, revision_ids in altered_fileids.iteritems():
353
for revision_id in revision_ids:
354
text_keys.append((file_id, revision_id))
355
self._add_mp_records_keys('file', self.repository.texts, text_keys)
357
def write_revisions(self):
358
"""Write bundle records for all revisions and signatures"""
359
inv_vf = self.repository.inventories
360
topological_order = [key[-1] for key in multiparent.topo_iter_keys(
361
inv_vf, self.revision_keys)]
362
revision_order = topological_order
363
if self.target is not None and self.target in self.revision_ids:
364
# Make sure the target revision is always the last entry
365
revision_order = list(topological_order)
366
revision_order.remove(self.target)
367
revision_order.append(self.target)
368
if self.repository._serializer.support_altered_by_hack:
369
# Repositories that support_altered_by_hack means that
370
# inventories.make_mpdiffs() contains all the data about the tree
371
# shape. Formats without support_altered_by_hack require
372
# chk_bytes/etc, so we use a different code path.
373
self._add_mp_records_keys('inventory', inv_vf,
374
[(revid,) for revid in topological_order])
376
# Inventories should always be added in pure-topological order, so
377
# that we can apply the mpdiff for the child to the parent texts.
378
self._add_inventory_mpdiffs_from_serializer(topological_order)
379
self._add_revision_texts(revision_order)
381
def _add_inventory_mpdiffs_from_serializer(self, revision_order):
382
"""Generate mpdiffs by serializing inventories.
384
The current repository only has part of the tree shape information in
385
the 'inventories' vf. So we use serializer.write_inventory_to_string to
386
get a 'full' representation of the tree shape, and then generate
387
mpdiffs on that data stream. This stream can then be reconstructed on
390
inventory_key_order = [(r,) for r in revision_order]
391
generator = _MPDiffInventoryGenerator(self.repository,
393
for revision_id, parent_ids, sha1, diff in generator.iter_diffs():
394
text = ''.join(diff.to_patch())
395
self.bundle.add_multiparent_record(text, sha1, parent_ids,
396
'inventory', revision_id, None)
398
def _add_revision_texts(self, revision_order):
399
parent_map = self.repository.get_parent_map(revision_order)
400
revision_to_str = self.repository._serializer.write_revision_to_string
401
revisions = self.repository.get_revisions(revision_order)
402
for revision in revisions:
403
revision_id = revision.revision_id
404
parents = parent_map.get(revision_id, None)
405
revision_text = revision_to_str(revision)
406
self.bundle.add_fulltext_record(revision_text, parents,
407
'revision', revision_id)
409
self.bundle.add_fulltext_record(
410
self.repository.get_signature_text(
411
revision_id), parents, 'signature', revision_id)
412
except errors.NoSuchRevision:
416
def get_base_target(revision_ids, forced_bases, repository):
417
"""Determine the base and target from old-style revision ids"""
418
if len(revision_ids) == 0:
420
target = revision_ids[0]
421
base = forced_bases.get(target)
423
parents = repository.get_revision(target).parent_ids
424
if len(parents) == 0:
425
base = _mod_revision.NULL_REVISION
430
def _add_mp_records_keys(self, repo_kind, vf, keys):
431
"""Add multi-parent diff records to a bundle"""
432
ordered_keys = list(multiparent.topo_iter_keys(vf, keys))
433
mpdiffs = vf.make_mpdiffs(ordered_keys)
434
sha1s = vf.get_sha1s(ordered_keys)
435
parent_map = vf.get_parent_map(ordered_keys)
436
for mpdiff, item_key, in zip(mpdiffs, ordered_keys):
437
sha1 = sha1s[item_key]
438
parents = [key[-1] for key in parent_map[item_key]]
439
text = ''.join(mpdiff.to_patch())
440
# Infer file id records as appropriate.
441
if len(item_key) == 2:
442
file_id = item_key[0]
445
self.bundle.add_multiparent_record(text, sha1, parents, repo_kind,
446
item_key[-1], file_id)
449
class BundleInfoV4(object):
451
"""Provide (most of) the BundleInfo interface"""
452
def __init__(self, fileobj, serializer):
453
self._fileobj = fileobj
454
self._serializer = serializer
455
self.__real_revisions = None
456
self.__revisions = None
458
def install(self, repository):
459
return self.install_revisions(repository)
461
def install_revisions(self, repository, stream_input=True):
462
"""Install this bundle's revisions into the specified repository
464
:param target_repo: The repository to install into
465
:param stream_input: If True, will stream input rather than reading it
466
all into memory at once. Reading it into memory all at once is
469
repository.lock_write()
471
ri = RevisionInstaller(self.get_bundle_reader(stream_input),
472
self._serializer, repository)
477
def get_merge_request(self, target_repo):
478
"""Provide data for performing a merge
480
Returns suggested base, suggested target, and patch verification status
482
return None, self.target, 'inapplicable'
484
def get_bundle_reader(self, stream_input=True):
485
"""Return a new BundleReader for the associated bundle
487
:param stream_input: If True, the BundleReader stream input rather than
488
reading it all into memory at once. Reading it into memory all at
489
once is (currently) faster.
491
self._fileobj.seek(0)
492
return BundleReader(self._fileobj, stream_input)
494
def _get_real_revisions(self):
495
if self.__real_revisions is None:
496
self.__real_revisions = []
497
bundle_reader = self.get_bundle_reader()
498
for bytes, metadata, repo_kind, revision_id, file_id in \
499
bundle_reader.iter_records():
500
if repo_kind == 'info':
502
self._serializer.get_source_serializer(metadata)
503
if repo_kind == 'revision':
504
rev = serializer.read_revision_from_string(bytes)
505
self.__real_revisions.append(rev)
506
return self.__real_revisions
507
real_revisions = property(_get_real_revisions)
509
def _get_revisions(self):
510
if self.__revisions is None:
511
self.__revisions = []
512
for revision in self.real_revisions:
513
self.__revisions.append(
514
bundle_data.RevisionInfo.from_revision(revision))
515
return self.__revisions
517
revisions = property(_get_revisions)
519
def _get_target(self):
520
return self.revisions[-1].revision_id
522
target = property(_get_target)
525
class RevisionInstaller(object):
526
"""Installs revisions into a repository"""
528
def __init__(self, container, serializer, repository):
529
self._container = container
530
self._serializer = serializer
531
self._repository = repository
535
"""Perform the installation.
537
Must be called with the Repository locked.
539
self._repository.start_write_group()
541
result = self._install_in_write_group()
543
self._repository.abort_write_group()
545
self._repository.commit_write_group()
548
def _install_in_write_group(self):
550
current_versionedfile = None
551
pending_file_records = []
553
pending_inventory_records = []
555
target_revision = None
556
for bytes, metadata, repo_kind, revision_id, file_id in\
557
self._container.iter_records():
558
if repo_kind == 'info':
559
if self._info is not None:
560
raise AssertionError()
561
self._handle_info(metadata)
562
if (pending_file_records and
563
(repo_kind, file_id) != ('file', current_file)):
564
# Flush the data for a single file - prevents memory
565
# spiking due to buffering all files in memory.
566
self._install_mp_records_keys(self._repository.texts,
567
pending_file_records)
569
del pending_file_records[:]
570
if len(pending_inventory_records) > 0 and repo_kind != 'inventory':
571
self._install_inventory_records(pending_inventory_records)
572
pending_inventory_records = []
573
if repo_kind == 'inventory':
574
pending_inventory_records.append(((revision_id,), metadata, bytes))
575
if repo_kind == 'revision':
576
target_revision = revision_id
577
self._install_revision(revision_id, metadata, bytes)
578
if repo_kind == 'signature':
579
self._install_signature(revision_id, metadata, bytes)
580
if repo_kind == 'file':
581
current_file = file_id
582
pending_file_records.append(((file_id, revision_id), metadata, bytes))
583
self._install_mp_records_keys(self._repository.texts, pending_file_records)
584
return target_revision
586
def _handle_info(self, info):
587
"""Extract data from an info record"""
589
self._source_serializer = self._serializer.get_source_serializer(info)
590
if (info['supports_rich_root'] == 0 and
591
self._repository.supports_rich_root()):
592
self.update_root = True
594
self.update_root = False
596
def _install_mp_records(self, versionedfile, records):
597
if len(records) == 0:
599
d_func = multiparent.MultiParent.from_patch
600
vf_records = [(r, m['parents'], m['sha1'], d_func(t)) for r, m, t in
601
records if r not in versionedfile]
602
versionedfile.add_mpdiffs(vf_records)
604
def _install_mp_records_keys(self, versionedfile, records):
605
d_func = multiparent.MultiParent.from_patch
607
for key, meta, text in records:
608
# Adapt to tuple interface: A length two key is a file_id,
609
# revision_id pair, a length 1 key is a
610
# revision/signature/inventory. We need to do this because
611
# the metadata extraction from the bundle has not yet been updated
612
# to use the consistent tuple interface itself.
617
parents = [prefix + (parent,) for parent in meta['parents']]
618
vf_records.append((key, parents, meta['sha1'], d_func(text)))
619
versionedfile.add_mpdiffs(vf_records)
621
def _get_parent_inventory_texts(self, inventory_text_cache,
622
inventory_cache, parent_ids):
623
cached_parent_texts = {}
624
remaining_parent_ids = []
625
for parent_id in parent_ids:
626
p_text = inventory_text_cache.get(parent_id, None)
628
remaining_parent_ids.append(parent_id)
630
cached_parent_texts[parent_id] = p_text
632
# TODO: Use inventory_cache to grab inventories we already have in
634
if remaining_parent_ids:
635
# first determine what keys are actually present in the local
636
# inventories object (don't use revisions as they haven't been
638
parent_keys = [(r,) for r in remaining_parent_ids]
639
present_parent_map = self._repository.inventories.get_parent_map(
641
present_parent_ids = []
643
for p_id in remaining_parent_ids:
644
if (p_id,) in present_parent_map:
645
present_parent_ids.append(p_id)
648
to_string = self._source_serializer.write_inventory_to_string
649
for parent_inv in self._repository.iter_inventories(
651
p_text = to_string(parent_inv)
652
inventory_cache[parent_inv.revision_id] = parent_inv
653
cached_parent_texts[parent_inv.revision_id] = p_text
654
inventory_text_cache[parent_inv.revision_id] = p_text
656
parent_texts = [cached_parent_texts[parent_id]
657
for parent_id in parent_ids
658
if parent_id not in ghosts]
661
def _install_inventory_records(self, records):
662
if (self._info['serializer'] == self._repository._serializer.format_num
663
and self._repository._serializer.support_altered_by_hack):
664
return self._install_mp_records_keys(self._repository.inventories,
666
# Use a 10MB text cache, since these are string xml inventories. Note
667
# that 10MB is fairly small for large projects (a single inventory can
668
# be >5MB). Another possibility is to cache 10-20 inventory texts
670
inventory_text_cache = lru_cache.LRUSizeCache(10*1024*1024)
671
# Also cache the in-memory representation. This allows us to create
672
# inventory deltas to apply rather than calling add_inventory from
674
inventory_cache = lru_cache.LRUCache(10)
675
pb = ui.ui_factory.nested_progress_bar()
677
num_records = len(records)
678
for idx, (key, metadata, bytes) in enumerate(records):
679
pb.update('installing inventory', idx, num_records)
680
revision_id = key[-1]
681
parent_ids = metadata['parents']
682
# Note: This assumes the local ghosts are identical to the
683
# ghosts in the source, as the Bundle serialization
684
# format doesn't record ghosts.
685
p_texts = self._get_parent_inventory_texts(inventory_text_cache,
688
# Why does to_lines() take strings as the source, it seems that
689
# it would have to cast to a list of lines, which we get back
690
# as lines and then cast back to a string.
691
target_lines = multiparent.MultiParent.from_patch(bytes
693
inv_text = ''.join(target_lines)
695
sha1 = osutils.sha_string(inv_text)
696
if sha1 != metadata['sha1']:
697
raise errors.BadBundle("Can't convert to target format")
698
# Add this to the cache so we don't have to extract it again.
699
inventory_text_cache[revision_id] = inv_text
700
target_inv = self._source_serializer.read_inventory_from_string(
702
self._handle_root(target_inv, parent_ids)
705
parent_inv = inventory_cache.get(parent_ids[0], None)
707
if parent_inv is None:
708
self._repository.add_inventory(revision_id, target_inv,
711
delta = target_inv._make_delta(parent_inv)
712
self._repository.add_inventory_by_delta(parent_ids[0],
713
delta, revision_id, parent_ids)
714
except errors.UnsupportedInventoryKind:
715
raise errors.IncompatibleRevision(repr(self._repository))
716
inventory_cache[revision_id] = target_inv
720
def _handle_root(self, target_inv, parent_ids):
721
revision_id = target_inv.revision_id
723
text_key = (target_inv.root.file_id, revision_id)
724
parent_keys = [(target_inv.root.file_id, parent) for
725
parent in parent_ids]
726
self._repository.texts.add_lines(text_key, parent_keys, [])
727
elif not self._repository.supports_rich_root():
728
if target_inv.root.revision != revision_id:
729
raise errors.IncompatibleRevision(repr(self._repository))
731
def _install_revision(self, revision_id, metadata, text):
732
if self._repository.has_revision(revision_id):
734
revision = self._source_serializer.read_revision_from_string(text)
735
self._repository.add_revision(revision.revision_id, revision)
737
def _install_signature(self, revision_id, metadata, text):
738
transaction = self._repository.get_transaction()
739
if self._repository.has_signature_for_revision_id(revision_id):
741
self._repository.add_signature_text(revision_id, text)
added
removed
bzrlib/bundle/serializer/v4.py
