14
14
# along with this program; if not, write to the Free Software
15
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from cStringIO import StringIO
17
from __future__ import absolute_import
29
repository as _mod_repository,
29
30
revision as _mod_revision,
34
from bzrlib.bundle import bundle_data, serializer as bundle_serializer
35
from bzrlib import bencode
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
38
84
class BundleWriter(object):
78
123
:revision_id: The revision id of the mpdiff being added.
79
124
:file_id: The file-id of the file, or None for inventories.
81
metadata = {'parents': parents,
82
'storage_kind': 'mpdiff',
126
metadata = {b'parents': parents,
127
b'storage_kind': b'mpdiff',
84
129
self._add_record(mp_bytes, metadata, repo_kind, revision_id, file_id)
86
131
def add_fulltext_record(self, bytes, parents, repo_kind, revision_id):
93
138
:revision_id: The revision id of the fulltext being added.
95
metadata = {'parents': parents,
96
'storage_kind': 'mpdiff'}
97
self._add_record(bytes, {'parents': parents,
98
'storage_kind': 'fulltext'}, repo_kind, revision_id, None)
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)
100
def add_info_record(self, **kwargs):
145
def add_info_record(self, kwargs):
101
146
"""Add an info record to the bundle
103
148
Any parameters may be supplied, except 'self' and 'storage_kind'.
104
149
Values must be lists, strings, integers, dicts, or a combination.
106
kwargs['storage_kind'] = 'header'
151
kwargs[b'storage_kind'] = b'header'
107
152
self._add_record(None, kwargs, 'info', None, None)
110
155
def encode_name(content_kind, revision_id, file_id=None):
111
156
"""Encode semantic ids as a container name"""
112
157
if content_kind not in ('revision', 'file', 'inventory', 'signature',
114
159
raise ValueError(content_kind)
115
160
if content_kind == 'file':
116
161
if file_id is None:
123
168
raise AssertionError()
124
169
elif revision_id is None:
125
170
raise AssertionError()
126
names = [n.replace('/', '//') for n in
127
(content_kind, revision_id, file_id) if n is not None]
128
return '/'.join(names)
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)
130
175
def _add_record(self, bytes, metadata, repo_kind, revision_id, file_id):
131
176
"""Add a bundle record to the container.
213
258
raise errors.BadBundle('Record has %d names instead of 1'
215
260
metadata = bencode.bdecode(bytes)
216
if metadata['storage_kind'] == 'header':
261
if metadata[b'storage_kind'] == b'header':
219
_unused, bytes = iterator.next()
264
_unused, bytes = next(iterator)
220
265
yield (bytes, metadata) + self.decode_name(names[0][0])
223
268
class BundleSerializerV4(bundle_serializer.BundleSerializer):
224
269
"""Implement the high-level bundle interface"""
226
def write(self, repository, revision_ids, forced_bases, fileobj):
227
"""Write a bundle to a file-like object
229
For backwards-compatibility only
231
write_op = BundleWriteOperation.from_old_args(repository, revision_ids,
232
forced_bases, fileobj)
233
return write_op.do_write()
235
271
def write_bundle(self, repository, target, base, fileobj):
236
272
"""Write a bundle to a file object
253
289
def get_source_serializer(info):
254
290
"""Retrieve the serializer for a given info object"""
255
return serializer.format_registry.get(info['serializer'])
291
return serializer.format_registry.get(info[b'serializer'].decode('ascii'))
258
294
class BundleWriteOperation(object):
259
295
"""Perform the operation of writing revisions to a bundle"""
262
def from_old_args(cls, repository, revision_ids, forced_bases, fileobj):
263
"""Create a BundleWriteOperation from old-style arguments"""
264
base, target = cls.get_base_target(revision_ids, forced_bases,
266
return BundleWriteOperation(base, target, repository, fileobj,
269
297
def __init__(self, base, target, repository, fileobj, revision_ids=None):
271
299
self.target = target
281
309
parents = graph.get_parent_map(revision_ids)
282
310
self.revision_ids = [r for r in revision_ids if r in parents]
283
self.revision_keys = set([(revid,) for revid in self.revision_ids])
311
self.revision_keys = {(revid,) for revid in self.revision_ids}
285
313
def do_write(self):
286
314
"""Write all data to the bundle"""
287
trace.note('Bundling %d revision(s).', len(self.revision_ids))
288
self.repository.lock_read()
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():
290
318
self.bundle.begin()
291
319
self.write_info()
292
320
self.write_files()
293
321
self.write_revisions()
294
322
self.bundle.end()
296
self.repository.unlock()
297
323
return self.revision_ids
299
325
def write_info(self):
301
327
serializer_format = self.repository.get_serializer_format()
302
328
supports_rich_root = {True: 1, False: 0}[
303
329
self.repository.supports_rich_root()]
304
self.bundle.add_info_record(serializer=serializer_format,
305
supports_rich_root=supports_rich_root)
330
self.bundle.add_info_record({b'serializer': serializer_format,
331
b'supports_rich_root': supports_rich_root})
307
333
def write_files(self):
308
334
"""Write bundle records for all revisions of all files"""
310
336
altered_fileids = self.repository.fileids_altered_by_revision_ids(
312
for file_id, revision_ids in altered_fileids.iteritems():
338
for file_id, revision_ids in viewitems(altered_fileids):
313
339
for revision_id in revision_ids:
314
340
text_keys.append((file_id, revision_id))
315
341
self._add_mp_records_keys('file', self.repository.texts, text_keys)
350
376
inventory_key_order = [(r,) for r in revision_order]
351
parent_map = self.repository.inventories.get_parent_map(
353
missing_keys = set(inventory_key_order).difference(parent_map)
355
raise errors.RevisionNotPresent(list(missing_keys)[0],
356
self.repository.inventories)
357
inv_to_str = self.repository._serializer.write_inventory_to_string
358
# Make sure that we grab the parent texts first
360
map(just_parents.update, parent_map.itervalues())
361
just_parents.difference_update(parent_map)
362
# Ignore ghost parents
363
present_parents = self.repository.inventories.get_parent_map(
365
ghost_keys = just_parents.difference(present_parents)
366
needed_inventories = list(present_parents) + inventory_key_order
367
needed_inventories = [k[-1] for k in needed_inventories]
369
for inv in self.repository.iter_inventories(needed_inventories):
370
revision_id = inv.revision_id
372
as_bytes = inv_to_str(inv)
373
# The sha1 is validated as the xml/textual form, not as the
374
# form-in-the-repository
375
sha1 = osutils.sha_string(as_bytes)
376
as_lines = osutils.split_lines(as_bytes)
378
all_lines[key] = as_lines
379
if key in just_parents:
380
# We don't transmit those entries
382
# Create an mpdiff for this text, and add it to the output
383
parent_keys = parent_map[key]
384
# See the comment in VF.make_mpdiffs about how this effects
385
# ordering when there are ghosts present. I think we have a latent
387
parent_lines = [all_lines[p_key] for p_key in parent_keys
388
if p_key not in ghost_keys]
389
diff = multiparent.MultiParent.from_lines(
390
as_lines, parent_lines)
391
text = ''.join(diff.to_patch())
392
parent_ids = [k[-1] for k in parent_keys]
377
generator = _MPDiffInventoryGenerator(self.repository,
379
for revision_id, parent_ids, sha1, diff in generator.iter_diffs():
380
text = b''.join(diff.to_patch())
393
381
self.bundle.add_multiparent_record(text, sha1, parent_ids,
394
382
'inventory', revision_id, None)
396
384
def _add_revision_texts(self, revision_order):
397
385
parent_map = self.repository.get_parent_map(revision_order)
398
revision_to_str = self.repository._serializer.write_revision_to_string
386
revision_to_bytes = self.repository._serializer.write_revision_to_string
399
387
revisions = self.repository.get_revisions(revision_order)
400
388
for revision in revisions:
401
389
revision_id = revision.revision_id
402
390
parents = parent_map.get(revision_id, None)
403
revision_text = revision_to_str(revision)
391
revision_text = revision_to_bytes(revision)
404
392
self.bundle.add_fulltext_record(revision_text, parents,
405
'revision', revision_id)
393
'revision', revision_id)
407
395
self.bundle.add_fulltext_record(
408
396
self.repository.get_signature_text(
409
revision_id), parents, 'signature', revision_id)
397
revision_id), parents, 'signature', revision_id)
410
398
except errors.NoSuchRevision:
552
532
added_inv = set()
553
533
target_revision = None
554
534
for bytes, metadata, repo_kind, revision_id, file_id in\
555
self._container.iter_records():
535
self._container.iter_records():
556
536
if repo_kind == 'info':
557
537
if self._info is not None:
558
538
raise AssertionError()
559
539
self._handle_info(metadata)
560
540
if (pending_file_records and
561
(repo_kind, file_id) != ('file', current_file)):
541
(repo_kind, file_id) != ('file', current_file)):
562
542
# Flush the data for a single file - prevents memory
563
543
# spiking due to buffering all files in memory.
564
544
self._install_mp_records_keys(self._repository.texts,
565
pending_file_records)
545
pending_file_records)
566
546
current_file = None
567
547
del pending_file_records[:]
568
548
if len(pending_inventory_records) > 0 and repo_kind != 'inventory':
569
549
self._install_inventory_records(pending_inventory_records)
570
550
pending_inventory_records = []
571
551
if repo_kind == 'inventory':
572
pending_inventory_records.append(((revision_id,), metadata, bytes))
552
pending_inventory_records.append(
553
((revision_id,), metadata, bytes))
573
554
if repo_kind == 'revision':
574
555
target_revision = revision_id
575
556
self._install_revision(revision_id, metadata, bytes)
577
558
self._install_signature(revision_id, metadata, bytes)
578
559
if repo_kind == 'file':
579
560
current_file = file_id
580
pending_file_records.append(((file_id, revision_id), metadata, bytes))
581
self._install_mp_records_keys(self._repository.texts, pending_file_records)
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)
582
565
return target_revision
584
567
def _handle_info(self, info):
585
568
"""Extract data from an info record"""
586
569
self._info = info
587
570
self._source_serializer = self._serializer.get_source_serializer(info)
588
if (info['supports_rich_root'] == 0 and
589
self._repository.supports_rich_root()):
571
if (info[b'supports_rich_root'] == 0 and
572
self._repository.supports_rich_root()):
590
573
self.update_root = True
592
575
self.update_root = False
615
parents = [prefix + (parent,) for parent in meta['parents']]
616
vf_records.append((key, parents, meta['sha1'], d_func(text)))
598
parents = [prefix + (parent,) for parent in meta[b'parents']]
599
vf_records.append((key, parents, meta[b'sha1'], d_func(text)))
617
600
versionedfile.add_mpdiffs(vf_records)
619
602
def _get_parent_inventory_texts(self, inventory_text_cache,
654
637
parent_texts = [cached_parent_texts[parent_id]
655
638
for parent_id in parent_ids
656
if parent_id not in ghosts]
639
if parent_id not in ghosts]
657
640
return parent_texts
659
642
def _install_inventory_records(self, records):
660
if (self._info['serializer'] == self._repository._serializer.format_num
661
and self._repository._serializer.support_altered_by_hack):
643
if (self._info[b'serializer'] == self._repository._serializer.format_num
644
and self._repository._serializer.support_altered_by_hack):
662
645
return self._install_mp_records_keys(self._repository.inventories,
664
647
# Use a 10MB text cache, since these are string xml inventories. Note
665
648
# that 10MB is fairly small for large projects (a single inventory can
666
649
# be >5MB). Another possibility is to cache 10-20 inventory texts
668
inventory_text_cache = lru_cache.LRUSizeCache(10*1024*1024)
651
inventory_text_cache = lru_cache.LRUSizeCache(10 * 1024 * 1024)
669
652
# Also cache the in-memory representation. This allows us to create
670
653
# inventory deltas to apply rather than calling add_inventory from
671
654
# scratch each time.
672
655
inventory_cache = lru_cache.LRUCache(10)
673
pb = ui.ui_factory.nested_progress_bar()
656
with ui.ui_factory.nested_progress_bar() as pb:
675
657
num_records = len(records)
676
658
for idx, (key, metadata, bytes) in enumerate(records):
677
659
pb.update('installing inventory', idx, num_records)
678
660
revision_id = key[-1]
679
parent_ids = metadata['parents']
661
parent_ids = metadata[b'parents']
680
662
# Note: This assumes the local ghosts are identical to the
681
663
# ghosts in the source, as the Bundle serialization
682
664
# format doesn't record ghosts.
687
669
# it would have to cast to a list of lines, which we get back
688
670
# as lines and then cast back to a string.
689
671
target_lines = multiparent.MultiParent.from_patch(bytes
691
inv_text = ''.join(target_lines)
673
inv_text = b''.join(target_lines)
693
675
sha1 = osutils.sha_string(inv_text)
694
if sha1 != metadata['sha1']:
676
if sha1 != metadata[b'sha1']:
695
677
raise errors.BadBundle("Can't convert to target format")
696
678
# Add this to the cache so we don't have to extract it again.
697
679
inventory_text_cache[revision_id] = inv_text
709
691
delta = target_inv._make_delta(parent_inv)
710
692
self._repository.add_inventory_by_delta(parent_ids[0],
711
delta, revision_id, parent_ids)
693
delta, revision_id, parent_ids)
712
694
except errors.UnsupportedInventoryKind:
713
695
raise errors.IncompatibleRevision(repr(self._repository))
714
696
inventory_cache[revision_id] = target_inv
718
698
def _handle_root(self, target_inv, parent_ids):
719
699
revision_id = target_inv.revision_id
720
700
if self.update_root:
721
701
text_key = (target_inv.root.file_id, revision_id)
722
702
parent_keys = [(target_inv.root.file_id, parent) for
723
parent in parent_ids]
703
parent in parent_ids]
724
704
self._repository.texts.add_lines(text_key, parent_keys, [])
725
705
elif not self._repository.supports_rich_root():
726
706
if target_inv.root.revision != revision_id: