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
29
revision as _mod_revision,
36
versionedfile as _mod_versionedfile,
38
from ...bundle import bundle_data, serializer as bundle_serializer
39
from ...i18n import ngettext
40
from ...sixish import (
46
class _MPDiffInventoryGenerator(_mod_versionedfile._MPDiffGenerator):
47
"""Generate Inventory diffs serialized inventories."""
49
def __init__(self, repo, inventory_keys):
50
super(_MPDiffInventoryGenerator, self).__init__(repo.inventories,
56
"""Compute the diffs one at a time."""
57
# This is instead of compute_diffs() since we guarantee our ordering of
58
# inventories, we don't have to do any buffering
59
self._find_needed_keys()
60
# We actually use a slightly different ordering. We grab all of the
61
# parents first, and then grab the ordered requests.
62
needed_ids = [k[-1] for k in self.present_parents]
63
needed_ids.extend([k[-1] for k in self.ordered_keys])
64
inv_to_str = self.repo._serializer.write_inventory_to_string
65
for inv in self.repo.iter_inventories(needed_ids):
66
revision_id = inv.revision_id
68
if key in self.present_parents:
69
# Not a key we will transmit, which is a shame, since because
70
# of that bundles don't work with stacked branches
73
parent_ids = [k[-1] for k in self.parent_map[key]]
74
as_bytes = inv_to_str(inv)
75
self._process_one_record(key, (as_bytes,))
76
if parent_ids is None:
78
diff = self.diffs.pop(key)
79
sha1 = osutils.sha_string(as_bytes)
80
yield revision_id, parent_ids, sha1, diff
83
class BundleWriter(object):
84
"""Writer for bundle-format files.
86
This serves roughly the same purpose as ContainerReader, but acts as a
89
Provides ways of writing the specific record types supported this bundle
93
def __init__(self, fileobj):
94
self._container = pack.ContainerWriter(self._write_encoded)
95
self._fileobj = fileobj
96
self._compressor = bz2.BZ2Compressor()
98
def _write_encoded(self, bytes):
99
"""Write bzip2-encoded bytes to the file"""
100
self._fileobj.write(self._compressor.compress(bytes))
103
"""Start writing the bundle"""
104
self._fileobj.write(bundle_serializer._get_bundle_header(
105
bundle_serializer.v4_string))
106
self._fileobj.write('#\n')
107
self._container.begin()
110
"""Finish writing the bundle"""
111
self._container.end()
112
self._fileobj.write(self._compressor.flush())
114
def add_multiparent_record(self, mp_bytes, sha1, parents, repo_kind,
115
revision_id, file_id):
116
"""Add a record for a multi-parent diff
118
:mp_bytes: A multi-parent diff, as a bytestring
119
:sha1: The sha1 hash of the fulltext
120
:parents: a list of revision-ids of the parents
121
:repo_kind: The kind of object in the repository. May be 'file' or
123
:revision_id: The revision id of the mpdiff being added.
124
:file_id: The file-id of the file, or None for inventories.
126
metadata = {'parents': parents,
127
'storage_kind': 'mpdiff',
129
self._add_record(mp_bytes, metadata, repo_kind, revision_id, file_id)
131
def add_fulltext_record(self, bytes, parents, repo_kind, revision_id):
132
"""Add a record for a fulltext
134
:bytes: The fulltext, as a bytestring
135
:parents: a list of revision-ids of the parents
136
:repo_kind: The kind of object in the repository. May be 'revision' or
138
:revision_id: The revision id of the fulltext being added.
140
metadata = {'parents': parents,
141
'storage_kind': 'mpdiff'}
142
self._add_record(bytes, {'parents': parents,
143
'storage_kind': 'fulltext'}, repo_kind, revision_id, None)
145
def add_info_record(self, **kwargs):
146
"""Add an info record to the bundle
148
Any parameters may be supplied, except 'self' and 'storage_kind'.
149
Values must be lists, strings, integers, dicts, or a combination.
151
kwargs['storage_kind'] = 'header'
152
self._add_record(None, kwargs, 'info', None, None)
155
def encode_name(content_kind, revision_id, file_id=None):
156
"""Encode semantic ids as a container name"""
157
if content_kind not in ('revision', 'file', 'inventory', 'signature',
159
raise ValueError(content_kind)
160
if content_kind == 'file':
162
raise AssertionError()
164
if file_id is not None:
165
raise AssertionError()
166
if content_kind == 'info':
167
if revision_id is not None:
168
raise AssertionError()
169
elif revision_id is None:
170
raise AssertionError()
171
names = [n.replace('/', '//') for n in
172
(content_kind, revision_id, file_id) if n is not None]
173
return '/'.join(names)
175
def _add_record(self, bytes, metadata, repo_kind, revision_id, file_id):
176
"""Add a bundle record to the container.
178
Most bundle records are recorded as header/body pairs, with the
179
body being nameless. Records with storage_kind 'header' have no
182
name = self.encode_name(repo_kind, revision_id, file_id)
183
encoded_metadata = bencode.bencode(metadata)
184
self._container.add_bytes_record(encoded_metadata, [(name, )])
185
if metadata['storage_kind'] != 'header':
186
self._container.add_bytes_record(bytes, [])
189
class BundleReader(object):
190
"""Reader for bundle-format files.
192
This serves roughly the same purpose as ContainerReader, but acts as a
193
layer on top of it, providing metadata, a semantic name, and a record
197
def __init__(self, fileobj, stream_input=True):
200
:param fileobj: a file containing a bzip-encoded container
201
:param stream_input: If True, the BundleReader stream input rather than
202
reading it all into memory at once. Reading it into memory all at
203
once is (currently) faster.
205
line = fileobj.readline()
208
self.patch_lines = []
210
source_file = iterablefile.IterableFile(self.iter_decode(fileobj))
212
source_file = BytesIO(bz2.decompress(fileobj.read()))
213
self._container_file = source_file
216
def iter_decode(fileobj):
217
"""Iterate through decoded fragments of the file"""
218
decompressor = bz2.BZ2Decompressor()
221
yield decompressor.decompress(line)
226
def decode_name(name):
227
"""Decode a name from its container form into a semantic form
229
:retval: content_kind, revision_id, file_id
231
segments = re.split('(//?)', name)
233
for segment in segments:
240
content_kind = names[0]
244
revision_id = names[1]
247
return content_kind, revision_id, file_id
249
def iter_records(self):
250
"""Iterate through bundle records
252
:return: a generator of (bytes, metadata, content_kind, revision_id,
255
iterator = pack.iter_records_from_file(self._container_file)
256
for names, bytes in iterator:
258
raise errors.BadBundle('Record has %d names instead of 1'
260
metadata = bencode.bdecode(bytes)
261
if metadata['storage_kind'] == 'header':
264
_unused, bytes = next(iterator)
265
yield (bytes, metadata) + self.decode_name(names[0][0])
268
class BundleSerializerV4(bundle_serializer.BundleSerializer):
269
"""Implement the high-level bundle interface"""
271
def write(self, repository, revision_ids, forced_bases, fileobj):
272
"""Write a bundle to a file-like object
274
For backwards-compatibility only
276
write_op = BundleWriteOperation.from_old_args(repository, revision_ids,
277
forced_bases, fileobj)
278
return write_op.do_write()
280
def write_bundle(self, repository, target, base, fileobj):
281
"""Write a bundle to a file object
283
:param repository: The repository to retrieve revision data from
284
:param target: The head revision to include ancestors of
285
:param base: The ancestor of the target to stop including acestors
287
:param fileobj: The file-like object to write to
289
write_op = BundleWriteOperation(base, target, repository, fileobj)
290
return write_op.do_write()
292
def read(self, file):
293
"""return a reader object for a given file"""
294
bundle = BundleInfoV4(file, self)
298
def get_source_serializer(info):
299
"""Retrieve the serializer for a given info object"""
300
return serializer.format_registry.get(info['serializer'])
303
class BundleWriteOperation(object):
304
"""Perform the operation of writing revisions to a bundle"""
307
def from_old_args(cls, repository, revision_ids, forced_bases, fileobj):
308
"""Create a BundleWriteOperation from old-style arguments"""
309
base, target = cls.get_base_target(revision_ids, forced_bases,
311
return BundleWriteOperation(base, target, repository, fileobj,
314
def __init__(self, base, target, repository, fileobj, revision_ids=None):
317
self.repository = repository
318
bundle = BundleWriter(fileobj)
320
if revision_ids is not None:
321
self.revision_ids = revision_ids
323
graph = repository.get_graph()
324
revision_ids = graph.find_unique_ancestors(target, [base])
326
parents = graph.get_parent_map(revision_ids)
327
self.revision_ids = [r for r in revision_ids if r in parents]
328
self.revision_keys = {(revid,) for revid in self.revision_ids}
331
"""Write all data to the bundle"""
332
trace.note(ngettext('Bundling %d revision.', 'Bundling %d revisions.',
333
len(self.revision_ids)), len(self.revision_ids))
334
with self.repository.lock_read():
338
self.write_revisions()
340
return self.revision_ids
342
def write_info(self):
343
"""Write format info"""
344
serializer_format = self.repository.get_serializer_format()
345
supports_rich_root = {True: 1, False: 0}[
346
self.repository.supports_rich_root()]
347
self.bundle.add_info_record(serializer=serializer_format,
348
supports_rich_root=supports_rich_root)
350
def write_files(self):
351
"""Write bundle records for all revisions of all files"""
353
altered_fileids = self.repository.fileids_altered_by_revision_ids(
355
for file_id, revision_ids in viewitems(altered_fileids):
356
for revision_id in revision_ids:
357
text_keys.append((file_id, revision_id))
358
self._add_mp_records_keys('file', self.repository.texts, text_keys)
360
def write_revisions(self):
361
"""Write bundle records for all revisions and signatures"""
362
inv_vf = self.repository.inventories
363
topological_order = [key[-1] for key in multiparent.topo_iter_keys(
364
inv_vf, self.revision_keys)]
365
revision_order = topological_order
366
if self.target is not None and self.target in self.revision_ids:
367
# Make sure the target revision is always the last entry
368
revision_order = list(topological_order)
369
revision_order.remove(self.target)
370
revision_order.append(self.target)
371
if self.repository._serializer.support_altered_by_hack:
372
# Repositories that support_altered_by_hack means that
373
# inventories.make_mpdiffs() contains all the data about the tree
374
# shape. Formats without support_altered_by_hack require
375
# chk_bytes/etc, so we use a different code path.
376
self._add_mp_records_keys('inventory', inv_vf,
377
[(revid,) for revid in topological_order])
379
# Inventories should always be added in pure-topological order, so
380
# that we can apply the mpdiff for the child to the parent texts.
381
self._add_inventory_mpdiffs_from_serializer(topological_order)
382
self._add_revision_texts(revision_order)
384
def _add_inventory_mpdiffs_from_serializer(self, revision_order):
385
"""Generate mpdiffs by serializing inventories.
387
The current repository only has part of the tree shape information in
388
the 'inventories' vf. So we use serializer.write_inventory_to_string to
389
get a 'full' representation of the tree shape, and then generate
390
mpdiffs on that data stream. This stream can then be reconstructed on
393
inventory_key_order = [(r,) for r in revision_order]
394
generator = _MPDiffInventoryGenerator(self.repository,
396
for revision_id, parent_ids, sha1, diff in generator.iter_diffs():
397
text = ''.join(diff.to_patch())
398
self.bundle.add_multiparent_record(text, sha1, parent_ids,
399
'inventory', revision_id, None)
401
def _add_revision_texts(self, revision_order):
402
parent_map = self.repository.get_parent_map(revision_order)
403
revision_to_str = self.repository._serializer.write_revision_to_string
404
revisions = self.repository.get_revisions(revision_order)
405
for revision in revisions:
406
revision_id = revision.revision_id
407
parents = parent_map.get(revision_id, None)
408
revision_text = revision_to_str(revision)
409
self.bundle.add_fulltext_record(revision_text, parents,
410
'revision', revision_id)
412
self.bundle.add_fulltext_record(
413
self.repository.get_signature_text(
414
revision_id), parents, 'signature', revision_id)
415
except errors.NoSuchRevision:
419
def get_base_target(revision_ids, forced_bases, repository):
420
"""Determine the base and target from old-style revision ids"""
421
if len(revision_ids) == 0:
423
target = revision_ids[0]
424
base = forced_bases.get(target)
426
parents = repository.get_revision(target).parent_ids
427
if len(parents) == 0:
428
base = _mod_revision.NULL_REVISION
433
def _add_mp_records_keys(self, repo_kind, vf, keys):
434
"""Add multi-parent diff records to a bundle"""
435
ordered_keys = list(multiparent.topo_iter_keys(vf, keys))
436
mpdiffs = vf.make_mpdiffs(ordered_keys)
437
sha1s = vf.get_sha1s(ordered_keys)
438
parent_map = vf.get_parent_map(ordered_keys)
439
for mpdiff, item_key, in zip(mpdiffs, ordered_keys):
440
sha1 = sha1s[item_key]
441
parents = [key[-1] for key in parent_map[item_key]]
442
text = ''.join(mpdiff.to_patch())
443
# Infer file id records as appropriate.
444
if len(item_key) == 2:
445
file_id = item_key[0]
448
self.bundle.add_multiparent_record(text, sha1, parents, repo_kind,
449
item_key[-1], file_id)
452
class BundleInfoV4(object):
454
"""Provide (most of) the BundleInfo interface"""
455
def __init__(self, fileobj, serializer):
456
self._fileobj = fileobj
457
self._serializer = serializer
458
self.__real_revisions = None
459
self.__revisions = None
461
def install(self, repository):
462
return self.install_revisions(repository)
464
def install_revisions(self, repository, stream_input=True):
465
"""Install this bundle's revisions into the specified repository
467
:param target_repo: The repository to install into
468
:param stream_input: If True, will stream input rather than reading it
469
all into memory at once. Reading it into memory all at once is
472
repository.lock_write()
474
ri = RevisionInstaller(self.get_bundle_reader(stream_input),
475
self._serializer, repository)
480
def get_merge_request(self, target_repo):
481
"""Provide data for performing a merge
483
Returns suggested base, suggested target, and patch verification status
485
return None, self.target, 'inapplicable'
487
def get_bundle_reader(self, stream_input=True):
488
"""Return a new BundleReader for the associated bundle
490
:param stream_input: If True, the BundleReader stream input rather than
491
reading it all into memory at once. Reading it into memory all at
492
once is (currently) faster.
494
self._fileobj.seek(0)
495
return BundleReader(self._fileobj, stream_input)
497
def _get_real_revisions(self):
498
if self.__real_revisions is None:
499
self.__real_revisions = []
500
bundle_reader = self.get_bundle_reader()
501
for bytes, metadata, repo_kind, revision_id, file_id in \
502
bundle_reader.iter_records():
503
if repo_kind == 'info':
505
self._serializer.get_source_serializer(metadata)
506
if repo_kind == 'revision':
507
rev = serializer.read_revision_from_string(bytes)
508
self.__real_revisions.append(rev)
509
return self.__real_revisions
510
real_revisions = property(_get_real_revisions)
512
def _get_revisions(self):
513
if self.__revisions is None:
514
self.__revisions = []
515
for revision in self.real_revisions:
516
self.__revisions.append(
517
bundle_data.RevisionInfo.from_revision(revision))
518
return self.__revisions
520
revisions = property(_get_revisions)
522
def _get_target(self):
523
return self.revisions[-1].revision_id
525
target = property(_get_target)
528
class RevisionInstaller(object):
529
"""Installs revisions into a repository"""
531
def __init__(self, container, serializer, repository):
532
self._container = container
533
self._serializer = serializer
534
self._repository = repository
538
"""Perform the installation.
540
Must be called with the Repository locked.
542
self._repository.start_write_group()
544
result = self._install_in_write_group()
546
self._repository.abort_write_group()
548
self._repository.commit_write_group()
551
def _install_in_write_group(self):
553
current_versionedfile = None
554
pending_file_records = []
556
pending_inventory_records = []
558
target_revision = None
559
for bytes, metadata, repo_kind, revision_id, file_id in\
560
self._container.iter_records():
561
if repo_kind == 'info':
562
if self._info is not None:
563
raise AssertionError()
564
self._handle_info(metadata)
565
if (pending_file_records and
566
(repo_kind, file_id) != ('file', current_file)):
567
# Flush the data for a single file - prevents memory
568
# spiking due to buffering all files in memory.
569
self._install_mp_records_keys(self._repository.texts,
570
pending_file_records)
572
del pending_file_records[:]
573
if len(pending_inventory_records) > 0 and repo_kind != 'inventory':
574
self._install_inventory_records(pending_inventory_records)
575
pending_inventory_records = []
576
if repo_kind == 'inventory':
577
pending_inventory_records.append(((revision_id,), metadata, bytes))
578
if repo_kind == 'revision':
579
target_revision = revision_id
580
self._install_revision(revision_id, metadata, bytes)
581
if repo_kind == 'signature':
582
self._install_signature(revision_id, metadata, bytes)
583
if repo_kind == 'file':
584
current_file = file_id
585
pending_file_records.append(((file_id, revision_id), metadata, bytes))
586
self._install_mp_records_keys(self._repository.texts, pending_file_records)
587
return target_revision
589
def _handle_info(self, info):
590
"""Extract data from an info record"""
592
self._source_serializer = self._serializer.get_source_serializer(info)
593
if (info['supports_rich_root'] == 0 and
594
self._repository.supports_rich_root()):
595
self.update_root = True
597
self.update_root = False
599
def _install_mp_records(self, versionedfile, records):
600
if len(records) == 0:
602
d_func = multiparent.MultiParent.from_patch
603
vf_records = [(r, m['parents'], m['sha1'], d_func(t)) for r, m, t in
604
records if r not in versionedfile]
605
versionedfile.add_mpdiffs(vf_records)
607
def _install_mp_records_keys(self, versionedfile, records):
608
d_func = multiparent.MultiParent.from_patch
610
for key, meta, text in records:
611
# Adapt to tuple interface: A length two key is a file_id,
612
# revision_id pair, a length 1 key is a
613
# revision/signature/inventory. We need to do this because
614
# the metadata extraction from the bundle has not yet been updated
615
# to use the consistent tuple interface itself.
620
parents = [prefix + (parent,) for parent in meta['parents']]
621
vf_records.append((key, parents, meta['sha1'], d_func(text)))
622
versionedfile.add_mpdiffs(vf_records)
624
def _get_parent_inventory_texts(self, inventory_text_cache,
625
inventory_cache, parent_ids):
626
cached_parent_texts = {}
627
remaining_parent_ids = []
628
for parent_id in parent_ids:
629
p_text = inventory_text_cache.get(parent_id, None)
631
remaining_parent_ids.append(parent_id)
633
cached_parent_texts[parent_id] = p_text
635
# TODO: Use inventory_cache to grab inventories we already have in
637
if remaining_parent_ids:
638
# first determine what keys are actually present in the local
639
# inventories object (don't use revisions as they haven't been
641
parent_keys = [(r,) for r in remaining_parent_ids]
642
present_parent_map = self._repository.inventories.get_parent_map(
644
present_parent_ids = []
646
for p_id in remaining_parent_ids:
647
if (p_id,) in present_parent_map:
648
present_parent_ids.append(p_id)
651
to_string = self._source_serializer.write_inventory_to_string
652
for parent_inv in self._repository.iter_inventories(
654
p_text = to_string(parent_inv)
655
inventory_cache[parent_inv.revision_id] = parent_inv
656
cached_parent_texts[parent_inv.revision_id] = p_text
657
inventory_text_cache[parent_inv.revision_id] = p_text
659
parent_texts = [cached_parent_texts[parent_id]
660
for parent_id in parent_ids
661
if parent_id not in ghosts]
664
def _install_inventory_records(self, records):
665
if (self._info['serializer'] == self._repository._serializer.format_num
666
and self._repository._serializer.support_altered_by_hack):
667
return self._install_mp_records_keys(self._repository.inventories,
669
# Use a 10MB text cache, since these are string xml inventories. Note
670
# that 10MB is fairly small for large projects (a single inventory can
671
# be >5MB). Another possibility is to cache 10-20 inventory texts
673
inventory_text_cache = lru_cache.LRUSizeCache(10*1024*1024)
674
# Also cache the in-memory representation. This allows us to create
675
# inventory deltas to apply rather than calling add_inventory from
677
inventory_cache = lru_cache.LRUCache(10)
678
pb = ui.ui_factory.nested_progress_bar()
680
num_records = len(records)
681
for idx, (key, metadata, bytes) in enumerate(records):
682
pb.update('installing inventory', idx, num_records)
683
revision_id = key[-1]
684
parent_ids = metadata['parents']
685
# Note: This assumes the local ghosts are identical to the
686
# ghosts in the source, as the Bundle serialization
687
# format doesn't record ghosts.
688
p_texts = self._get_parent_inventory_texts(inventory_text_cache,
691
# Why does to_lines() take strings as the source, it seems that
692
# it would have to cast to a list of lines, which we get back
693
# as lines and then cast back to a string.
694
target_lines = multiparent.MultiParent.from_patch(bytes
696
inv_text = ''.join(target_lines)
698
sha1 = osutils.sha_string(inv_text)
699
if sha1 != metadata['sha1']:
700
raise errors.BadBundle("Can't convert to target format")
701
# Add this to the cache so we don't have to extract it again.
702
inventory_text_cache[revision_id] = inv_text
703
target_inv = self._source_serializer.read_inventory_from_string(
705
self._handle_root(target_inv, parent_ids)
708
parent_inv = inventory_cache.get(parent_ids[0], None)
710
if parent_inv is None:
711
self._repository.add_inventory(revision_id, target_inv,
714
delta = target_inv._make_delta(parent_inv)
715
self._repository.add_inventory_by_delta(parent_ids[0],
716
delta, revision_id, parent_ids)
717
except errors.UnsupportedInventoryKind:
718
raise errors.IncompatibleRevision(repr(self._repository))
719
inventory_cache[revision_id] = target_inv
723
def _handle_root(self, target_inv, parent_ids):
724
revision_id = target_inv.revision_id
726
text_key = (target_inv.root.file_id, revision_id)
727
parent_keys = [(target_inv.root.file_id, parent) for
728
parent in parent_ids]
729
self._repository.texts.add_lines(text_key, parent_keys, [])
730
elif not self._repository.supports_rich_root():
731
if target_inv.root.revision != revision_id:
732
raise errors.IncompatibleRevision(repr(self._repository))
734
def _install_revision(self, revision_id, metadata, text):
735
if self._repository.has_revision(revision_id):
737
revision = self._source_serializer.read_revision_from_string(text)
738
self._repository.add_revision(revision.revision_id, revision)
740
def _install_signature(self, revision_id, metadata, text):
741
transaction = self._repository.get_transaction()
742
if self._repository.has_signature_for_revision_id(revision_id):
744
self._repository.add_signature_text(revision_id, text)
added
removed
breezy/bundle/serializer/v4.py
