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
repository as _mod_repository,
30
revision as _mod_revision,
37
versionedfile as _mod_versionedfile,
39
from ...bundle import bundle_data, serializer as bundle_serializer
40
from ...i18n import ngettext
41
from ...sixish import (
47
class _MPDiffInventoryGenerator(_mod_versionedfile._MPDiffGenerator):
48
"""Generate Inventory diffs serialized inventories."""
50
def __init__(self, repo, inventory_keys):
51
super(_MPDiffInventoryGenerator, self).__init__(repo.inventories,
57
"""Compute the diffs one at a time."""
58
# This is instead of compute_diffs() since we guarantee our ordering of
59
# inventories, we don't have to do any buffering
60
self._find_needed_keys()
61
# We actually use a slightly different ordering. We grab all of the
62
# parents first, and then grab the ordered requests.
63
needed_ids = [k[-1] for k in self.present_parents]
64
needed_ids.extend([k[-1] for k in self.ordered_keys])
65
inv_to_bytes = self.repo._serializer.write_inventory_to_string
66
for inv in self.repo.iter_inventories(needed_ids):
67
revision_id = inv.revision_id
69
if key in self.present_parents:
70
# Not a key we will transmit, which is a shame, since because
71
# of that bundles don't work with stacked branches
74
parent_ids = [k[-1] for k in self.parent_map[key]]
75
as_bytes = inv_to_bytes(inv)
76
self._process_one_record(key, (as_bytes,))
77
if parent_ids is None:
79
diff = self.diffs.pop(key)
80
sha1 = osutils.sha_string(as_bytes)
81
yield revision_id, parent_ids, sha1, diff
84
class BundleWriter(object):
85
"""Writer for bundle-format files.
87
This serves roughly the same purpose as ContainerReader, but acts as a
90
Provides ways of writing the specific record types supported this bundle
94
def __init__(self, fileobj):
95
self._container = pack.ContainerWriter(self._write_encoded)
96
self._fileobj = fileobj
97
self._compressor = bz2.BZ2Compressor()
99
def _write_encoded(self, bytes):
100
"""Write bzip2-encoded bytes to the file"""
101
self._fileobj.write(self._compressor.compress(bytes))
104
"""Start writing the bundle"""
105
self._fileobj.write(bundle_serializer._get_bundle_header('4'))
106
self._fileobj.write(b'#\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 = {b'parents': parents,
127
b'storage_kind': b'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 = {b'parents': parents,
141
b'storage_kind': b'mpdiff'}
142
self._add_record(bytes, {b'parents': parents,
143
b'storage_kind': b'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[b'storage_kind'] = b'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(b'/', b'//') for n in
172
(content_kind.encode('ascii'), revision_id, file_id) if n is not None]
173
return b'/'.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[b'storage_kind'] != b'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(b'(//?)', name)
233
for segment in segments:
236
elif segment == b'/':
240
content_kind = names[0]
244
revision_id = names[1]
247
return content_kind.decode('ascii'), 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[b'storage_kind'] == b'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_bundle(self, repository, target, base, fileobj):
272
"""Write a bundle to a file object
274
:param repository: The repository to retrieve revision data from
275
:param target: The head revision to include ancestors of
276
:param base: The ancestor of the target to stop including acestors
278
:param fileobj: The file-like object to write to
280
write_op = BundleWriteOperation(base, target, repository, fileobj)
281
return write_op.do_write()
283
def read(self, file):
284
"""return a reader object for a given file"""
285
bundle = BundleInfoV4(file, self)
289
def get_source_serializer(info):
290
"""Retrieve the serializer for a given info object"""
291
return serializer.format_registry.get(info[b'serializer'].decode('ascii'))
294
class BundleWriteOperation(object):
295
"""Perform the operation of writing revisions to a bundle"""
297
def __init__(self, base, target, repository, fileobj, revision_ids=None):
300
self.repository = repository
301
bundle = BundleWriter(fileobj)
303
if revision_ids is not None:
304
self.revision_ids = revision_ids
306
graph = repository.get_graph()
307
revision_ids = graph.find_unique_ancestors(target, [base])
309
parents = graph.get_parent_map(revision_ids)
310
self.revision_ids = [r for r in revision_ids if r in parents]
311
self.revision_keys = {(revid,) for revid in self.revision_ids}
314
"""Write all data to the bundle"""
315
trace.note(ngettext('Bundling %d revision.', 'Bundling %d revisions.',
316
len(self.revision_ids)), len(self.revision_ids))
317
with self.repository.lock_read():
321
self.write_revisions()
323
return self.revision_ids
325
def write_info(self):
326
"""Write format info"""
327
serializer_format = self.repository.get_serializer_format()
328
supports_rich_root = {True: 1, False: 0}[
329
self.repository.supports_rich_root()]
330
self.bundle.add_info_record({b'serializer': serializer_format,
331
b'supports_rich_root': supports_rich_root})
333
def write_files(self):
334
"""Write bundle records for all revisions of all files"""
336
altered_fileids = self.repository.fileids_altered_by_revision_ids(
338
for file_id, revision_ids in viewitems(altered_fileids):
339
for revision_id in revision_ids:
340
text_keys.append((file_id, revision_id))
341
self._add_mp_records_keys('file', self.repository.texts, text_keys)
343
def write_revisions(self):
344
"""Write bundle records for all revisions and signatures"""
345
inv_vf = self.repository.inventories
346
topological_order = [key[-1] for key in multiparent.topo_iter_keys(
347
inv_vf, self.revision_keys)]
348
revision_order = topological_order
349
if self.target is not None and self.target in self.revision_ids:
350
# Make sure the target revision is always the last entry
351
revision_order = list(topological_order)
352
revision_order.remove(self.target)
353
revision_order.append(self.target)
354
if self.repository._serializer.support_altered_by_hack:
355
# Repositories that support_altered_by_hack means that
356
# inventories.make_mpdiffs() contains all the data about the tree
357
# shape. Formats without support_altered_by_hack require
358
# chk_bytes/etc, so we use a different code path.
359
self._add_mp_records_keys('inventory', inv_vf,
360
[(revid,) for revid in topological_order])
362
# Inventories should always be added in pure-topological order, so
363
# that we can apply the mpdiff for the child to the parent texts.
364
self._add_inventory_mpdiffs_from_serializer(topological_order)
365
self._add_revision_texts(revision_order)
367
def _add_inventory_mpdiffs_from_serializer(self, revision_order):
368
"""Generate mpdiffs by serializing inventories.
370
The current repository only has part of the tree shape information in
371
the 'inventories' vf. So we use serializer.write_inventory_to_string to
372
get a 'full' representation of the tree shape, and then generate
373
mpdiffs on that data stream. This stream can then be reconstructed on
376
inventory_key_order = [(r,) for r in revision_order]
377
generator = _MPDiffInventoryGenerator(self.repository,
379
for revision_id, parent_ids, sha1, diff in generator.iter_diffs():
380
text = b''.join(diff.to_patch())
381
self.bundle.add_multiparent_record(text, sha1, parent_ids,
382
'inventory', revision_id, None)
384
def _add_revision_texts(self, revision_order):
385
parent_map = self.repository.get_parent_map(revision_order)
386
revision_to_bytes = self.repository._serializer.write_revision_to_string
387
revisions = self.repository.get_revisions(revision_order)
388
for revision in revisions:
389
revision_id = revision.revision_id
390
parents = parent_map.get(revision_id, None)
391
revision_text = revision_to_bytes(revision)
392
self.bundle.add_fulltext_record(revision_text, parents,
393
'revision', revision_id)
395
self.bundle.add_fulltext_record(
396
self.repository.get_signature_text(
397
revision_id), parents, 'signature', revision_id)
398
except errors.NoSuchRevision:
402
def get_base_target(revision_ids, forced_bases, repository):
403
"""Determine the base and target from old-style revision ids"""
404
if len(revision_ids) == 0:
406
target = revision_ids[0]
407
base = forced_bases.get(target)
409
parents = repository.get_revision(target).parent_ids
410
if len(parents) == 0:
411
base = _mod_revision.NULL_REVISION
416
def _add_mp_records_keys(self, repo_kind, vf, keys):
417
"""Add multi-parent diff records to a bundle"""
418
ordered_keys = list(multiparent.topo_iter_keys(vf, keys))
419
mpdiffs = vf.make_mpdiffs(ordered_keys)
420
sha1s = vf.get_sha1s(ordered_keys)
421
parent_map = vf.get_parent_map(ordered_keys)
422
for mpdiff, item_key, in zip(mpdiffs, ordered_keys):
423
sha1 = sha1s[item_key]
424
parents = [key[-1] for key in parent_map[item_key]]
425
text = b''.join(mpdiff.to_patch())
426
# Infer file id records as appropriate.
427
if len(item_key) == 2:
428
file_id = item_key[0]
431
self.bundle.add_multiparent_record(text, sha1, parents, repo_kind,
432
item_key[-1], file_id)
435
class BundleInfoV4(object):
437
"""Provide (most of) the BundleInfo interface"""
439
def __init__(self, fileobj, serializer):
440
self._fileobj = fileobj
441
self._serializer = serializer
442
self.__real_revisions = None
443
self.__revisions = None
445
def install(self, repository):
446
return self.install_revisions(repository)
448
def install_revisions(self, repository, stream_input=True):
449
"""Install this bundle's revisions into the specified repository
451
:param target_repo: The repository to install into
452
:param stream_input: If True, will stream input rather than reading it
453
all into memory at once. Reading it into memory all at once is
456
with repository.lock_write():
457
ri = RevisionInstaller(self.get_bundle_reader(stream_input),
458
self._serializer, repository)
461
def get_merge_request(self, target_repo):
462
"""Provide data for performing a merge
464
Returns suggested base, suggested target, and patch verification status
466
return None, self.target, 'inapplicable'
468
def get_bundle_reader(self, stream_input=True):
469
"""Return a new BundleReader for the associated bundle
471
:param stream_input: If True, the BundleReader stream input rather than
472
reading it all into memory at once. Reading it into memory all at
473
once is (currently) faster.
475
self._fileobj.seek(0)
476
return BundleReader(self._fileobj, stream_input)
478
def _get_real_revisions(self):
479
if self.__real_revisions is None:
480
self.__real_revisions = []
481
bundle_reader = self.get_bundle_reader()
482
for bytes, metadata, repo_kind, revision_id, file_id in \
483
bundle_reader.iter_records():
484
if repo_kind == 'info':
486
self._serializer.get_source_serializer(metadata)
487
if repo_kind == 'revision':
488
rev = serializer.read_revision_from_string(bytes)
489
self.__real_revisions.append(rev)
490
return self.__real_revisions
491
real_revisions = property(_get_real_revisions)
493
def _get_revisions(self):
494
if self.__revisions is None:
495
self.__revisions = []
496
for revision in self.real_revisions:
497
self.__revisions.append(
498
bundle_data.RevisionInfo.from_revision(revision))
499
return self.__revisions
501
revisions = property(_get_revisions)
503
def _get_target(self):
504
return self.revisions[-1].revision_id
506
target = property(_get_target)
509
class RevisionInstaller(object):
510
"""Installs revisions into a repository"""
512
def __init__(self, container, serializer, repository):
513
self._container = container
514
self._serializer = serializer
515
self._repository = repository
519
"""Perform the installation.
521
Must be called with the Repository locked.
523
with _mod_repository.WriteGroup(self._repository):
524
return self._install_in_write_group()
526
def _install_in_write_group(self):
528
current_versionedfile = None
529
pending_file_records = []
531
pending_inventory_records = []
533
target_revision = None
534
for bytes, metadata, repo_kind, revision_id, file_id in\
535
self._container.iter_records():
536
if repo_kind == 'info':
537
if self._info is not None:
538
raise AssertionError()
539
self._handle_info(metadata)
540
if (pending_file_records and
541
(repo_kind, file_id) != ('file', current_file)):
542
# Flush the data for a single file - prevents memory
543
# spiking due to buffering all files in memory.
544
self._install_mp_records_keys(self._repository.texts,
545
pending_file_records)
547
del pending_file_records[:]
548
if len(pending_inventory_records) > 0 and repo_kind != 'inventory':
549
self._install_inventory_records(pending_inventory_records)
550
pending_inventory_records = []
551
if repo_kind == 'inventory':
552
pending_inventory_records.append(
553
((revision_id,), metadata, bytes))
554
if repo_kind == 'revision':
555
target_revision = revision_id
556
self._install_revision(revision_id, metadata, bytes)
557
if repo_kind == 'signature':
558
self._install_signature(revision_id, metadata, bytes)
559
if repo_kind == 'file':
560
current_file = file_id
561
pending_file_records.append(
562
((file_id, revision_id), metadata, bytes))
563
self._install_mp_records_keys(
564
self._repository.texts, pending_file_records)
565
return target_revision
567
def _handle_info(self, info):
568
"""Extract data from an info record"""
570
self._source_serializer = self._serializer.get_source_serializer(info)
571
if (info[b'supports_rich_root'] == 0 and
572
self._repository.supports_rich_root()):
573
self.update_root = True
575
self.update_root = False
577
def _install_mp_records(self, versionedfile, records):
578
if len(records) == 0:
580
d_func = multiparent.MultiParent.from_patch
581
vf_records = [(r, m['parents'], m['sha1'], d_func(t)) for r, m, t in
582
records if r not in versionedfile]
583
versionedfile.add_mpdiffs(vf_records)
585
def _install_mp_records_keys(self, versionedfile, records):
586
d_func = multiparent.MultiParent.from_patch
588
for key, meta, text in records:
589
# Adapt to tuple interface: A length two key is a file_id,
590
# revision_id pair, a length 1 key is a
591
# revision/signature/inventory. We need to do this because
592
# the metadata extraction from the bundle has not yet been updated
593
# to use the consistent tuple interface itself.
598
parents = [prefix + (parent,) for parent in meta[b'parents']]
599
vf_records.append((key, parents, meta[b'sha1'], d_func(text)))
600
versionedfile.add_mpdiffs(vf_records)
602
def _get_parent_inventory_texts(self, inventory_text_cache,
603
inventory_cache, parent_ids):
604
cached_parent_texts = {}
605
remaining_parent_ids = []
606
for parent_id in parent_ids:
607
p_text = inventory_text_cache.get(parent_id, None)
609
remaining_parent_ids.append(parent_id)
611
cached_parent_texts[parent_id] = p_text
613
# TODO: Use inventory_cache to grab inventories we already have in
615
if remaining_parent_ids:
616
# first determine what keys are actually present in the local
617
# inventories object (don't use revisions as they haven't been
619
parent_keys = [(r,) for r in remaining_parent_ids]
620
present_parent_map = self._repository.inventories.get_parent_map(
622
present_parent_ids = []
624
for p_id in remaining_parent_ids:
625
if (p_id,) in present_parent_map:
626
present_parent_ids.append(p_id)
629
to_string = self._source_serializer.write_inventory_to_string
630
for parent_inv in self._repository.iter_inventories(
632
p_text = to_string(parent_inv)
633
inventory_cache[parent_inv.revision_id] = parent_inv
634
cached_parent_texts[parent_inv.revision_id] = p_text
635
inventory_text_cache[parent_inv.revision_id] = p_text
637
parent_texts = [cached_parent_texts[parent_id]
638
for parent_id in parent_ids
639
if parent_id not in ghosts]
642
def _install_inventory_records(self, records):
643
if (self._info[b'serializer'] == self._repository._serializer.format_num
644
and self._repository._serializer.support_altered_by_hack):
645
return self._install_mp_records_keys(self._repository.inventories,
647
# Use a 10MB text cache, since these are string xml inventories. Note
648
# that 10MB is fairly small for large projects (a single inventory can
649
# be >5MB). Another possibility is to cache 10-20 inventory texts
651
inventory_text_cache = lru_cache.LRUSizeCache(10 * 1024 * 1024)
652
# Also cache the in-memory representation. This allows us to create
653
# inventory deltas to apply rather than calling add_inventory from
655
inventory_cache = lru_cache.LRUCache(10)
656
with ui.ui_factory.nested_progress_bar() as pb:
657
num_records = len(records)
658
for idx, (key, metadata, bytes) in enumerate(records):
659
pb.update('installing inventory', idx, num_records)
660
revision_id = key[-1]
661
parent_ids = metadata[b'parents']
662
# Note: This assumes the local ghosts are identical to the
663
# ghosts in the source, as the Bundle serialization
664
# format doesn't record ghosts.
665
p_texts = self._get_parent_inventory_texts(inventory_text_cache,
668
# Why does to_lines() take strings as the source, it seems that
669
# it would have to cast to a list of lines, which we get back
670
# as lines and then cast back to a string.
671
target_lines = multiparent.MultiParent.from_patch(bytes
673
inv_text = b''.join(target_lines)
675
sha1 = osutils.sha_string(inv_text)
676
if sha1 != metadata[b'sha1']:
677
raise errors.BadBundle("Can't convert to target format")
678
# Add this to the cache so we don't have to extract it again.
679
inventory_text_cache[revision_id] = inv_text
680
target_inv = self._source_serializer.read_inventory_from_string(
682
self._handle_root(target_inv, parent_ids)
685
parent_inv = inventory_cache.get(parent_ids[0], None)
687
if parent_inv is None:
688
self._repository.add_inventory(revision_id, target_inv,
691
delta = target_inv._make_delta(parent_inv)
692
self._repository.add_inventory_by_delta(parent_ids[0],
693
delta, revision_id, parent_ids)
694
except errors.UnsupportedInventoryKind:
695
raise errors.IncompatibleRevision(repr(self._repository))
696
inventory_cache[revision_id] = target_inv
698
def _handle_root(self, target_inv, parent_ids):
699
revision_id = target_inv.revision_id
701
text_key = (target_inv.root.file_id, revision_id)
702
parent_keys = [(target_inv.root.file_id, parent) for
703
parent in parent_ids]
704
self._repository.texts.add_lines(text_key, parent_keys, [])
705
elif not self._repository.supports_rich_root():
706
if target_inv.root.revision != revision_id:
707
raise errors.IncompatibleRevision(repr(self._repository))
709
def _install_revision(self, revision_id, metadata, text):
710
if self._repository.has_revision(revision_id):
712
revision = self._source_serializer.read_revision_from_string(text)
713
self._repository.add_revision(revision.revision_id, revision)
715
def _install_signature(self, revision_id, metadata, text):
716
transaction = self._repository.get_transaction()
717
if self._repository.has_signature_for_revision_id(revision_id):
719
self._repository.add_signature_text(revision_id, text)
added
removed
breezy/bundle/serializer/v4.py
