1
# Copyright (C) 2008, 2009 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
"""Repository formats using CHK inventories and groupcompress compression."""
33
revision as _mod_revision,
37
from bzrlib.btree_index import (
41
from bzrlib.groupcompress import (
43
GroupCompressVersionedFiles,
45
from bzrlib.repofmt.pack_repo import (
50
PackRootCommitBuilder,
51
RepositoryPackCollection,
58
class GCPack(NewPack):
60
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
61
"""Create a NewPack instance.
63
:param pack_collection: A PackCollection into which this is being
65
:param upload_suffix: An optional suffix to be given to any temporary
66
files created during the pack creation. e.g '.autopack'
67
:param file_mode: An optional file mode to create the new files with.
69
# replaced from NewPack to:
70
# - change inventory reference list length to 1
71
# - change texts reference lists to 1
72
# TODO: patch this to be parameterised
74
# The relative locations of the packs are constrained, but all are
75
# passed in because the caller has them, so as to avoid object churn.
76
index_builder_class = pack_collection._index_builder_class
78
if pack_collection.chk_index is not None:
79
chk_index = index_builder_class(reference_lists=0)
83
# Revisions: parents list, no text compression.
84
index_builder_class(reference_lists=1),
85
# Inventory: We want to map compression only, but currently the
86
# knit code hasn't been updated enough to understand that, so we
87
# have a regular 2-list index giving parents and compression
89
index_builder_class(reference_lists=1),
90
# Texts: per file graph, for all fileids - so one reference list
91
# and two elements in the key tuple.
92
index_builder_class(reference_lists=1, key_elements=2),
93
# Signatures: Just blobs to store, no compression, no parents
95
index_builder_class(reference_lists=0),
96
# CHK based storage - just blobs, no compression or parents.
99
self._pack_collection = pack_collection
100
# When we make readonly indices, we need this.
101
self.index_class = pack_collection._index_class
102
# where should the new pack be opened
103
self.upload_transport = pack_collection._upload_transport
104
# where are indices written out to
105
self.index_transport = pack_collection._index_transport
106
# where is the pack renamed to when it is finished?
107
self.pack_transport = pack_collection._pack_transport
108
# What file mode to upload the pack and indices with.
109
self._file_mode = file_mode
110
# tracks the content written to the .pack file.
111
self._hash = osutils.md5()
112
# a four-tuple with the length in bytes of the indices, once the pack
113
# is finalised. (rev, inv, text, sigs)
114
self.index_sizes = None
115
# How much data to cache when writing packs. Note that this is not
116
# synchronised with reads, because it's not in the transport layer, so
117
# is not safe unless the client knows it won't be reading from the pack
119
self._cache_limit = 0
120
# the temporary pack file name.
121
self.random_name = osutils.rand_chars(20) + upload_suffix
122
# when was this pack started ?
123
self.start_time = time.time()
124
# open an output stream for the data added to the pack.
125
self.write_stream = self.upload_transport.open_write_stream(
126
self.random_name, mode=self._file_mode)
127
if 'pack' in debug.debug_flags:
128
trace.mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
129
time.ctime(), self.upload_transport.base, self.random_name,
130
time.time() - self.start_time)
131
# A list of byte sequences to be written to the new pack, and the
132
# aggregate size of them. Stored as a list rather than separate
133
# variables so that the _write_data closure below can update them.
134
self._buffer = [[], 0]
135
# create a callable for adding data
137
# robertc says- this is a closure rather than a method on the object
138
# so that the variables are locals, and faster than accessing object
140
def _write_data(bytes, flush=False, _buffer=self._buffer,
141
_write=self.write_stream.write, _update=self._hash.update):
142
_buffer[0].append(bytes)
143
_buffer[1] += len(bytes)
145
if _buffer[1] > self._cache_limit or flush:
146
bytes = ''.join(_buffer[0])
150
# expose this on self, for the occasion when clients want to add data.
151
self._write_data = _write_data
152
# a pack writer object to serialise pack records.
153
self._writer = pack.ContainerWriter(self._write_data)
155
# what state is the pack in? (open, finished, aborted)
158
def _check_references(self):
159
"""Make sure our external references are present.
161
Packs are allowed to have deltas whose base is not in the pack, but it
162
must be present somewhere in this collection. It is not allowed to
163
have deltas based on a fallback repository.
164
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
166
# Groupcompress packs don't have any external references, arguably CHK
167
# pages have external references, but we cannot 'cheaply' determine
168
# them without actually walking all of the chk pages.
171
class ResumedGCPack(ResumedPack):
173
def _check_references(self):
174
"""Make sure our external compression parents are present."""
175
# See GCPack._check_references for why this is empty
177
def _get_external_refs(self, index):
178
# GC repositories don't have compression parents external to a given
183
class GCCHKPacker(Packer):
184
"""This class understand what it takes to collect a GCCHK repo."""
186
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
188
super(GCCHKPacker, self).__init__(pack_collection, packs, suffix,
189
revision_ids=revision_ids,
190
reload_func=reload_func)
191
self._pack_collection = pack_collection
192
# ATM, We only support this for GCCHK repositories
193
if pack_collection.chk_index is None:
194
raise AssertionError('pack_collection.chk_index should not be None')
195
self._gather_text_refs = False
196
self._chk_id_roots = []
197
self._chk_p_id_roots = []
198
self._text_refs = None
199
# set by .pack() if self.revision_ids is not None
200
self.revision_keys = None
202
def _get_progress_stream(self, source_vf, keys, message, pb):
204
substream = source_vf.get_record_stream(keys, 'groupcompress', True)
205
for idx, record in enumerate(substream):
207
pb.update(message, idx + 1, len(keys))
211
def _get_filtered_inv_stream(self, source_vf, keys, message, pb=None):
212
"""Filter the texts of inventories, to find the chk pages."""
213
total_keys = len(keys)
214
def _filtered_inv_stream():
216
p_id_roots_set = set()
217
stream = source_vf.get_record_stream(keys, 'groupcompress', True)
218
for idx, record in enumerate(stream):
219
# Inventories should always be with revisions; assume success.
220
bytes = record.get_bytes_as('fulltext')
221
chk_inv = inventory.CHKInventory.deserialise(None, bytes,
224
pb.update('inv', idx, total_keys)
225
key = chk_inv.id_to_entry.key()
226
if key not in id_roots_set:
227
self._chk_id_roots.append(key)
228
id_roots_set.add(key)
229
p_id_map = chk_inv.parent_id_basename_to_file_id
231
raise AssertionError('Parent id -> file_id map not set')
233
if key not in p_id_roots_set:
234
p_id_roots_set.add(key)
235
self._chk_p_id_roots.append(key)
237
# We have finished processing all of the inventory records, we
238
# don't need these sets anymore
240
p_id_roots_set.clear()
241
return _filtered_inv_stream()
243
def _get_chk_streams(self, source_vf, keys, pb=None):
244
# We want to stream the keys from 'id_roots', and things they
245
# reference, and then stream things from p_id_roots and things they
246
# reference, and then any remaining keys that we didn't get to.
248
# We also group referenced texts together, so if one root references a
249
# text with prefix 'a', and another root references a node with prefix
250
# 'a', we want to yield those nodes before we yield the nodes for 'b'
251
# This keeps 'similar' nodes together.
253
# Note: We probably actually want multiple streams here, to help the
254
# client understand that the different levels won't compress well
255
# against each other.
256
# Test the difference between using one Group per level, and
257
# using 1 Group per prefix. (so '' (root) would get a group, then
258
# all the references to search-key 'a' would get a group, etc.)
259
total_keys = len(keys)
260
remaining_keys = set(keys)
262
if self._gather_text_refs:
263
bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
264
self._text_refs = set()
265
def _get_referenced_stream(root_keys, parse_leaf_nodes=False):
268
keys_by_search_prefix = {}
269
remaining_keys.difference_update(cur_keys)
271
def handle_internal_node(node):
272
for prefix, value in node._items.iteritems():
273
# We don't want to request the same key twice, and we
274
# want to order it by the first time it is seen.
275
# Even further, we don't want to request a key which is
276
# not in this group of pack files (it should be in the
277
# repo, but it doesn't have to be in the group being
279
# TODO: consider how to treat externally referenced chk
280
# pages as 'external_references' so that we
281
# always fill them in for stacked branches
282
if value not in next_keys and value in remaining_keys:
283
keys_by_search_prefix.setdefault(prefix,
286
def handle_leaf_node(node):
287
# Store is None, because we know we have a LeafNode, and we
288
# just want its entries
289
for file_id, bytes in node.iteritems(None):
290
name_utf8, file_id, revision_id = bytes_to_info(bytes)
291
self._text_refs.add((file_id, revision_id))
293
stream = source_vf.get_record_stream(cur_keys,
294
'as-requested', True)
295
for record in stream:
296
if record.storage_kind == 'absent':
297
# An absent CHK record: we assume that the missing
298
# record is in a different pack - e.g. a page not
299
# altered by the commit we're packing.
301
bytes = record.get_bytes_as('fulltext')
302
# We don't care about search_key_func for this code,
303
# because we only care about external references.
304
node = chk_map._deserialise(bytes, record.key,
305
search_key_func=None)
306
common_base = node._search_prefix
307
if isinstance(node, chk_map.InternalNode):
308
handle_internal_node(node)
309
elif parse_leaf_nodes:
310
handle_leaf_node(node)
313
pb.update('chk node', counter[0], total_keys)
316
# Double check that we won't be emitting any keys twice
317
# If we get rid of the pre-calculation of all keys, we could
318
# turn this around and do
319
# next_keys.difference_update(seen_keys)
320
# However, we also may have references to chk pages in another
321
# pack file during autopack. We filter earlier, so we should no
322
# longer need to do this
323
# next_keys = next_keys.intersection(remaining_keys)
325
for prefix in sorted(keys_by_search_prefix):
326
cur_keys.extend(keys_by_search_prefix.pop(prefix))
327
for stream in _get_referenced_stream(self._chk_id_roots,
328
self._gather_text_refs):
330
del self._chk_id_roots
331
# while it isn't really possible for chk_id_roots to not be in the
332
# local group of packs, it is possible that the tree shape has not
333
# changed recently, so we need to filter _chk_p_id_roots by the
335
chk_p_id_roots = [key for key in self._chk_p_id_roots
336
if key in remaining_keys]
337
del self._chk_p_id_roots
338
for stream in _get_referenced_stream(chk_p_id_roots, False):
341
trace.mutter('There were %d keys in the chk index, %d of which'
342
' were not referenced', total_keys,
344
if self.revision_ids is None:
345
stream = source_vf.get_record_stream(remaining_keys,
349
def _build_vf(self, index_name, parents, delta, for_write=False):
350
"""Build a VersionedFiles instance on top of this group of packs."""
351
index_name = index_name + '_index'
353
access = knit._DirectPackAccess(index_to_pack)
356
if self.new_pack is None:
357
raise AssertionError('No new pack has been set')
358
index = getattr(self.new_pack, index_name)
359
index_to_pack[index] = self.new_pack.access_tuple()
360
index.set_optimize(for_size=True)
361
access.set_writer(self.new_pack._writer, index,
362
self.new_pack.access_tuple())
363
add_callback = index.add_nodes
366
for pack in self.packs:
367
sub_index = getattr(pack, index_name)
368
index_to_pack[sub_index] = pack.access_tuple()
369
indices.append(sub_index)
370
index = _mod_index.CombinedGraphIndex(indices)
372
vf = GroupCompressVersionedFiles(
374
add_callback=add_callback,
376
is_locked=self._pack_collection.repo.is_locked),
381
def _build_vfs(self, index_name, parents, delta):
382
"""Build the source and target VersionedFiles."""
383
source_vf = self._build_vf(index_name, parents,
384
delta, for_write=False)
385
target_vf = self._build_vf(index_name, parents,
386
delta, for_write=True)
387
return source_vf, target_vf
389
def _copy_stream(self, source_vf, target_vf, keys, message, vf_to_stream,
391
trace.mutter('repacking %d %s', len(keys), message)
392
self.pb.update('repacking %s' % (message,), pb_offset)
393
child_pb = ui.ui_factory.nested_progress_bar()
395
stream = vf_to_stream(source_vf, keys, message, child_pb)
396
for _ in target_vf._insert_record_stream(stream,
403
def _copy_revision_texts(self):
404
source_vf, target_vf = self._build_vfs('revision', True, False)
405
if not self.revision_keys:
406
# We are doing a full fetch, aka 'pack'
407
self.revision_keys = source_vf.keys()
408
self._copy_stream(source_vf, target_vf, self.revision_keys,
409
'revisions', self._get_progress_stream, 1)
411
def _copy_inventory_texts(self):
412
source_vf, target_vf = self._build_vfs('inventory', True, True)
413
self._copy_stream(source_vf, target_vf, self.revision_keys,
414
'inventories', self._get_filtered_inv_stream, 2)
416
def _copy_chk_texts(self):
417
source_vf, target_vf = self._build_vfs('chk', False, False)
418
# TODO: This is technically spurious... if it is a performance issue,
420
total_keys = source_vf.keys()
421
trace.mutter('repacking chk: %d id_to_entry roots,'
422
' %d p_id_map roots, %d total keys',
423
len(self._chk_id_roots), len(self._chk_p_id_roots),
425
self.pb.update('repacking chk', 3)
426
child_pb = ui.ui_factory.nested_progress_bar()
428
for stream in self._get_chk_streams(source_vf, total_keys,
430
for _ in target_vf._insert_record_stream(stream,
437
def _copy_text_texts(self):
438
source_vf, target_vf = self._build_vfs('text', True, True)
439
# XXX: We don't walk the chk map to determine referenced (file_id,
440
# revision_id) keys. We don't do it yet because you really need
441
# to filter out the ones that are present in the parents of the
442
# rev just before the ones you are copying, otherwise the filter
443
# is grabbing too many keys...
444
text_keys = source_vf.keys()
445
self._copy_stream(source_vf, target_vf, text_keys,
446
'text', self._get_progress_stream, 4)
448
def _copy_signature_texts(self):
449
source_vf, target_vf = self._build_vfs('signature', False, False)
450
signature_keys = source_vf.keys()
451
signature_keys.intersection(self.revision_keys)
452
self._copy_stream(source_vf, target_vf, signature_keys,
453
'signatures', self._get_progress_stream, 5)
455
def _create_pack_from_packs(self):
456
self.pb.update('repacking', 0, 7)
457
self.new_pack = self.open_pack()
458
# Is this necessary for GC ?
459
self.new_pack.set_write_cache_size(1024*1024)
460
self._copy_revision_texts()
461
self._copy_inventory_texts()
462
self._copy_chk_texts()
463
self._copy_text_texts()
464
self._copy_signature_texts()
465
self.new_pack._check_references()
466
if not self._use_pack(self.new_pack):
467
self.new_pack.abort()
469
self.pb.update('finishing repack', 6, 7)
470
self.new_pack.finish()
471
self._pack_collection.allocate(self.new_pack)
475
class GCCHKReconcilePacker(GCCHKPacker):
476
"""A packer which regenerates indices etc as it copies.
478
This is used by ``bzr reconcile`` to cause parent text pointers to be
482
def __init__(self, *args, **kwargs):
483
super(GCCHKReconcilePacker, self).__init__(*args, **kwargs)
484
self._data_changed = False
485
self._gather_text_refs = True
487
def _copy_inventory_texts(self):
488
source_vf, target_vf = self._build_vfs('inventory', True, True)
489
self._copy_stream(source_vf, target_vf, self.revision_keys,
490
'inventories', self._get_filtered_inv_stream, 2)
491
if source_vf.keys() != self.revision_keys:
492
self._data_changed = True
494
def _copy_text_texts(self):
495
"""generate what texts we should have and then copy."""
496
source_vf, target_vf = self._build_vfs('text', True, True)
497
trace.mutter('repacking %d texts', len(self._text_refs))
498
self.pb.update("repacking texts", 4)
499
# we have three major tasks here:
500
# 1) generate the ideal index
501
repo = self._pack_collection.repo
502
# We want the one we just wrote, so base it on self.new_pack
503
revision_vf = self._build_vf('revision', True, False, for_write=True)
504
ancestor_keys = revision_vf.get_parent_map(revision_vf.keys())
505
# Strip keys back into revision_ids.
506
ancestors = dict((k[0], tuple([p[0] for p in parents]))
507
for k, parents in ancestor_keys.iteritems())
509
# TODO: _generate_text_key_index should be much cheaper to generate from
510
# a chk repository, rather than the current implementation
511
ideal_index = repo._generate_text_key_index(None, ancestors)
512
file_id_parent_map = source_vf.get_parent_map(self._text_refs)
513
# 2) generate a keys list that contains all the entries that can
514
# be used as-is, with corrected parents.
516
new_parent_keys = {} # (key, parent_keys)
518
NULL_REVISION = _mod_revision.NULL_REVISION
519
for key in self._text_refs:
525
ideal_parents = tuple(ideal_index[key])
527
discarded_keys.append(key)
528
self._data_changed = True
530
if ideal_parents == (NULL_REVISION,):
532
source_parents = file_id_parent_map[key]
533
if ideal_parents == source_parents:
537
# We need to change the parent graph, but we don't need to
538
# re-insert the text (since we don't pun the compression
539
# parent with the parents list)
540
self._data_changed = True
541
new_parent_keys[key] = ideal_parents
542
# we're finished with some data.
544
del file_id_parent_map
545
# 3) bulk copy the data, updating records than need it
546
def _update_parents_for_texts():
547
stream = source_vf.get_record_stream(self._text_refs,
548
'groupcompress', False)
549
for record in stream:
550
if record.key in new_parent_keys:
551
record.parents = new_parent_keys[record.key]
553
target_vf.insert_record_stream(_update_parents_for_texts())
555
def _use_pack(self, new_pack):
556
"""Override _use_pack to check for reconcile having changed content."""
557
return new_pack.data_inserted() and self._data_changed
560
class GCRepositoryPackCollection(RepositoryPackCollection):
562
pack_factory = GCPack
563
resumed_pack_factory = ResumedGCPack
565
def _execute_pack_operations(self, pack_operations,
566
_packer_class=GCCHKPacker,
568
"""Execute a series of pack operations.
570
:param pack_operations: A list of [revision_count, packs_to_combine].
571
:param _packer_class: The class of packer to use (default: Packer).
574
# XXX: Copied across from RepositoryPackCollection simply because we
575
# want to override the _packer_class ... :(
576
for revision_count, packs in pack_operations:
577
# we may have no-ops from the setup logic
580
packer = GCCHKPacker(self, packs, '.autopack',
581
reload_func=reload_func)
584
except errors.RetryWithNewPacks:
585
# An exception is propagating out of this context, make sure
586
# this packer has cleaned up. Packer() doesn't set its new_pack
587
# state into the RepositoryPackCollection object, so we only
588
# have access to it directly here.
589
if packer.new_pack is not None:
590
packer.new_pack.abort()
593
self._remove_pack_from_memory(pack)
594
# record the newly available packs and stop advertising the old
596
self._save_pack_names(clear_obsolete_packs=True)
597
# Move the old packs out of the way now they are no longer referenced.
598
for revision_count, packs in pack_operations:
599
self._obsolete_packs(packs)
602
class CHKInventoryRepository(KnitPackRepository):
603
"""subclass of KnitPackRepository that uses CHK based inventories."""
605
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
607
"""Overridden to change pack collection class."""
608
KnitPackRepository.__init__(self, _format, a_bzrdir, control_files,
609
_commit_builder_class, _serializer)
610
# and now replace everything it did :)
611
index_transport = self._transport.clone('indices')
612
self._pack_collection = GCRepositoryPackCollection(self,
613
self._transport, index_transport,
614
self._transport.clone('upload'),
615
self._transport.clone('packs'),
616
_format.index_builder_class,
618
use_chk_index=self._format.supports_chks,
620
self.inventories = GroupCompressVersionedFiles(
621
_GCGraphIndex(self._pack_collection.inventory_index.combined_index,
622
add_callback=self._pack_collection.inventory_index.add_callback,
623
parents=True, is_locked=self.is_locked,
624
inconsistency_fatal=False),
625
access=self._pack_collection.inventory_index.data_access)
626
self.revisions = GroupCompressVersionedFiles(
627
_GCGraphIndex(self._pack_collection.revision_index.combined_index,
628
add_callback=self._pack_collection.revision_index.add_callback,
629
parents=True, is_locked=self.is_locked,
630
track_external_parent_refs=True),
631
access=self._pack_collection.revision_index.data_access,
633
self.signatures = GroupCompressVersionedFiles(
634
_GCGraphIndex(self._pack_collection.signature_index.combined_index,
635
add_callback=self._pack_collection.signature_index.add_callback,
636
parents=False, is_locked=self.is_locked,
637
inconsistency_fatal=False),
638
access=self._pack_collection.signature_index.data_access,
640
self.texts = GroupCompressVersionedFiles(
641
_GCGraphIndex(self._pack_collection.text_index.combined_index,
642
add_callback=self._pack_collection.text_index.add_callback,
643
parents=True, is_locked=self.is_locked,
644
inconsistency_fatal=False),
645
access=self._pack_collection.text_index.data_access)
646
# No parents, individual CHK pages don't have specific ancestry
647
self.chk_bytes = GroupCompressVersionedFiles(
648
_GCGraphIndex(self._pack_collection.chk_index.combined_index,
649
add_callback=self._pack_collection.chk_index.add_callback,
650
parents=False, is_locked=self.is_locked,
651
inconsistency_fatal=False),
652
access=self._pack_collection.chk_index.data_access)
653
# True when the repository object is 'write locked' (as opposed to the
654
# physical lock only taken out around changes to the pack-names list.)
655
# Another way to represent this would be a decorator around the control
656
# files object that presents logical locks as physical ones - if this
657
# gets ugly consider that alternative design. RBC 20071011
658
self._write_lock_count = 0
659
self._transaction = None
661
self._reconcile_does_inventory_gc = True
662
self._reconcile_fixes_text_parents = True
663
self._reconcile_backsup_inventory = False
665
def _add_inventory_checked(self, revision_id, inv, parents):
666
"""Add inv to the repository after checking the inputs.
668
This function can be overridden to allow different inventory styles.
670
:seealso: add_inventory, for the contract.
673
serializer = self._format._serializer
674
result = inventory.CHKInventory.from_inventory(self.chk_bytes, inv,
675
maximum_size=serializer.maximum_size,
676
search_key_name=serializer.search_key_name)
677
inv_lines = result.to_lines()
678
return self._inventory_add_lines(revision_id, parents,
679
inv_lines, check_content=False)
681
def _create_inv_from_null(self, delta, revision_id):
682
"""This will mutate new_inv directly.
684
This is a simplified form of create_by_apply_delta which knows that all
685
the old values must be None, so everything is a create.
687
serializer = self._format._serializer
688
new_inv = inventory.CHKInventory(serializer.search_key_name)
689
new_inv.revision_id = revision_id
690
entry_to_bytes = new_inv._entry_to_bytes
691
id_to_entry_dict = {}
692
parent_id_basename_dict = {}
693
for old_path, new_path, file_id, entry in delta:
694
if old_path is not None:
695
raise ValueError('Invalid delta, somebody tried to delete %r'
696
' from the NULL_REVISION'
697
% ((old_path, file_id),))
699
raise ValueError('Invalid delta, delta from NULL_REVISION has'
700
' no new_path %r' % (file_id,))
702
new_inv.root_id = file_id
703
parent_id_basename_key = ('', '')
705
utf8_entry_name = entry.name.encode('utf-8')
706
parent_id_basename_key = (entry.parent_id, utf8_entry_name)
707
new_value = entry_to_bytes(entry)
709
# new_inv._path_to_fileid_cache[new_path] = file_id
710
id_to_entry_dict[(file_id,)] = new_value
711
parent_id_basename_dict[parent_id_basename_key] = file_id
713
new_inv._populate_from_dicts(self.chk_bytes, id_to_entry_dict,
714
parent_id_basename_dict, maximum_size=serializer.maximum_size)
717
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
718
parents, basis_inv=None, propagate_caches=False):
719
"""Add a new inventory expressed as a delta against another revision.
721
:param basis_revision_id: The inventory id the delta was created
723
:param delta: The inventory delta (see Inventory.apply_delta for
725
:param new_revision_id: The revision id that the inventory is being
727
:param parents: The revision ids of the parents that revision_id is
728
known to have and are in the repository already. These are supplied
729
for repositories that depend on the inventory graph for revision
730
graph access, as well as for those that pun ancestry with delta
732
:param basis_inv: The basis inventory if it is already known,
734
:param propagate_caches: If True, the caches for this inventory are
735
copied to and updated for the result if possible.
737
:returns: (validator, new_inv)
738
The validator(which is a sha1 digest, though what is sha'd is
739
repository format specific) of the serialized inventory, and the
742
if not self.is_in_write_group():
743
raise AssertionError("%r not in write group" % (self,))
744
_mod_revision.check_not_reserved_id(new_revision_id)
746
if basis_inv is None:
747
if basis_revision_id == _mod_revision.NULL_REVISION:
748
new_inv = self._create_inv_from_null(delta, new_revision_id)
749
inv_lines = new_inv.to_lines()
750
return self._inventory_add_lines(new_revision_id, parents,
751
inv_lines, check_content=False), new_inv
753
basis_tree = self.revision_tree(basis_revision_id)
754
basis_tree.lock_read()
755
basis_inv = basis_tree.inventory
757
result = basis_inv.create_by_apply_delta(delta, new_revision_id,
758
propagate_caches=propagate_caches)
759
inv_lines = result.to_lines()
760
return self._inventory_add_lines(new_revision_id, parents,
761
inv_lines, check_content=False), result
763
if basis_tree is not None:
766
def _iter_inventories(self, revision_ids):
767
"""Iterate over many inventory objects."""
768
keys = [(revision_id,) for revision_id in revision_ids]
769
stream = self.inventories.get_record_stream(keys, 'unordered', True)
771
for record in stream:
772
if record.storage_kind != 'absent':
773
texts[record.key] = record.get_bytes_as('fulltext')
775
raise errors.NoSuchRevision(self, record.key)
777
yield inventory.CHKInventory.deserialise(self.chk_bytes, texts[key], key)
779
def _iter_inventory_xmls(self, revision_ids):
780
# Without a native 'xml' inventory, this method doesn't make sense, so
781
# make it raise to trap naughty direct users.
782
raise NotImplementedError(self._iter_inventory_xmls)
784
def _find_present_inventory_keys(self, revision_keys):
785
parent_map = self.inventories.get_parent_map(revision_keys)
786
present_inventory_keys = set(k for k in parent_map)
787
return present_inventory_keys
789
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
790
"""Find the file ids and versions affected by revisions.
792
:param revisions: an iterable containing revision ids.
793
:param _inv_weave: The inventory weave from this repository or None.
794
If None, the inventory weave will be opened automatically.
795
:return: a dictionary mapping altered file-ids to an iterable of
796
revision_ids. Each altered file-ids has the exact revision_ids that
797
altered it listed explicitly.
799
rich_root = self.supports_rich_root()
800
bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
801
file_id_revisions = {}
802
pb = ui.ui_factory.nested_progress_bar()
804
revision_keys = [(r,) for r in revision_ids]
805
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
806
# TODO: instead of using _find_present_inventory_keys, change the
807
# code paths to allow missing inventories to be tolerated.
808
# However, we only want to tolerate missing parent
809
# inventories, not missing inventories for revision_ids
810
present_parent_inv_keys = self._find_present_inventory_keys(
812
present_parent_inv_ids = set(
813
[k[-1] for k in present_parent_inv_keys])
814
uninteresting_root_keys = set()
815
interesting_root_keys = set()
816
inventories_to_read = set(revision_ids)
817
inventories_to_read.update(present_parent_inv_ids)
818
for inv in self.iter_inventories(inventories_to_read):
819
entry_chk_root_key = inv.id_to_entry.key()
820
if inv.revision_id in present_parent_inv_ids:
821
uninteresting_root_keys.add(entry_chk_root_key)
823
interesting_root_keys.add(entry_chk_root_key)
825
chk_bytes = self.chk_bytes
826
for record, items in chk_map.iter_interesting_nodes(chk_bytes,
827
interesting_root_keys, uninteresting_root_keys,
829
for name, bytes in items:
830
(name_utf8, file_id, revision_id) = bytes_to_info(bytes)
831
if not rich_root and name_utf8 == '':
834
file_id_revisions[file_id].add(revision_id)
836
file_id_revisions[file_id] = set([revision_id])
839
return file_id_revisions
841
def find_text_key_references(self):
842
"""Find the text key references within the repository.
844
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
845
to whether they were referred to by the inventory of the
846
revision_id that they contain. The inventory texts from all present
847
revision ids are assessed to generate this report.
849
# XXX: Slow version but correct: rewrite as a series of delta
850
# examinations/direct tree traversal. Note that that will require care
851
# as a common node is reachable both from the inventory that added it,
852
# and others afterwards.
853
revision_keys = self.revisions.keys()
855
rich_roots = self.supports_rich_root()
856
pb = ui.ui_factory.nested_progress_bar()
858
all_revs = self.all_revision_ids()
859
total = len(all_revs)
860
for pos, inv in enumerate(self.iter_inventories(all_revs)):
861
pb.update("Finding text references", pos, total)
862
for _, entry in inv.iter_entries():
863
if not rich_roots and entry.file_id == inv.root_id:
865
key = (entry.file_id, entry.revision)
866
result.setdefault(key, False)
867
if entry.revision == inv.revision_id:
873
def _reconcile_pack(self, collection, packs, extension, revs, pb):
874
packer = GCCHKReconcilePacker(collection, packs, extension)
875
return packer.pack(pb)
877
def _get_source(self, to_format):
878
"""Return a source for streaming from this repository."""
879
if isinstance(to_format, remote.RemoteRepositoryFormat):
880
# Can't just check attributes on to_format with the current code,
882
to_format._ensure_real()
883
to_format = to_format._custom_format
884
if to_format.__class__ is self._format.__class__:
885
# We must be exactly the same format, otherwise stuff like the chk
886
# page layout might be different
887
return GroupCHKStreamSource(self, to_format)
888
return super(CHKInventoryRepository, self)._get_source(to_format)
891
class GroupCHKStreamSource(KnitPackStreamSource):
892
"""Used when both the source and target repo are GroupCHK repos."""
894
def __init__(self, from_repository, to_format):
895
"""Create a StreamSource streaming from from_repository."""
896
super(GroupCHKStreamSource, self).__init__(from_repository, to_format)
897
self._revision_keys = None
898
self._text_keys = None
899
self._text_fetch_order = 'groupcompress'
900
self._chk_id_roots = None
901
self._chk_p_id_roots = None
903
def _get_inventory_stream(self, inventory_keys, allow_absent=False):
904
"""Get a stream of inventory texts.
906
When this function returns, self._chk_id_roots and self._chk_p_id_roots
909
self._chk_id_roots = []
910
self._chk_p_id_roots = []
911
def _filtered_inv_stream():
913
p_id_roots_set = set()
914
source_vf = self.from_repository.inventories
915
stream = source_vf.get_record_stream(inventory_keys,
916
'groupcompress', True)
917
for record in stream:
918
if record.storage_kind == 'absent':
922
raise errors.NoSuchRevision(self, record.key)
923
bytes = record.get_bytes_as('fulltext')
924
chk_inv = inventory.CHKInventory.deserialise(None, bytes,
926
key = chk_inv.id_to_entry.key()
927
if key not in id_roots_set:
928
self._chk_id_roots.append(key)
929
id_roots_set.add(key)
930
p_id_map = chk_inv.parent_id_basename_to_file_id
932
raise AssertionError('Parent id -> file_id map not set')
934
if key not in p_id_roots_set:
935
p_id_roots_set.add(key)
936
self._chk_p_id_roots.append(key)
938
# We have finished processing all of the inventory records, we
939
# don't need these sets anymore
941
p_id_roots_set.clear()
942
return ('inventories', _filtered_inv_stream())
944
def _get_filtered_chk_streams(self, excluded_revision_keys):
945
self._text_keys = set()
946
excluded_revision_keys.discard(_mod_revision.NULL_REVISION)
947
if not excluded_revision_keys:
948
uninteresting_root_keys = set()
949
uninteresting_pid_root_keys = set()
951
# filter out any excluded revisions whose inventories are not
953
# TODO: Update Repository.iter_inventories() to add
954
# ignore_missing=True
955
present_keys = self.from_repository._find_present_inventory_keys(
956
excluded_revision_keys)
957
present_ids = [k[-1] for k in present_keys]
958
uninteresting_root_keys = set()
959
uninteresting_pid_root_keys = set()
960
for inv in self.from_repository.iter_inventories(present_ids):
961
uninteresting_root_keys.add(inv.id_to_entry.key())
962
uninteresting_pid_root_keys.add(
963
inv.parent_id_basename_to_file_id.key())
964
bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
965
chk_bytes = self.from_repository.chk_bytes
966
def _filter_id_to_entry():
967
for record, items in chk_map.iter_interesting_nodes(chk_bytes,
968
self._chk_id_roots, uninteresting_root_keys):
969
for name, bytes in items:
970
# Note: we don't care about name_utf8, because we are always
972
_, file_id, revision_id = bytes_to_info(bytes)
973
self._text_keys.add((file_id, revision_id))
974
if record is not None:
977
self._chk_id_roots = None
978
yield 'chk_bytes', _filter_id_to_entry()
979
def _get_parent_id_basename_to_file_id_pages():
980
for record, items in chk_map.iter_interesting_nodes(chk_bytes,
981
self._chk_p_id_roots, uninteresting_pid_root_keys):
982
if record is not None:
985
self._chk_p_id_roots = None
986
yield 'chk_bytes', _get_parent_id_basename_to_file_id_pages()
988
def get_stream(self, search):
989
revision_ids = search.get_keys()
990
for stream_info in self._fetch_revision_texts(revision_ids):
992
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
993
yield self._get_inventory_stream(self._revision_keys)
994
# TODO: The keys to exclude might be part of the search recipe
995
# For now, exclude all parents that are at the edge of ancestry, for
996
# which we have inventories
997
from_repo = self.from_repository
998
parent_keys = from_repo._find_parent_keys_of_revisions(
1000
for stream_info in self._get_filtered_chk_streams(parent_keys):
1002
yield self._get_text_stream()
1004
def get_stream_for_missing_keys(self, missing_keys):
1005
# missing keys can only occur when we are byte copying and not
1006
# translating (because translation means we don't send
1007
# unreconstructable deltas ever).
1008
missing_inventory_keys = set()
1009
for key in missing_keys:
1010
if key[0] != 'inventories':
1011
raise AssertionError('The only missing keys we should'
1012
' be filling in are inventory keys, not %s'
1014
missing_inventory_keys.add(key[1:])
1015
if self._chk_id_roots or self._chk_p_id_roots:
1016
raise AssertionError('Cannot call get_stream_for_missing_keys'
1017
' untill all of get_stream() has been consumed.')
1018
# Yield the inventory stream, so we can find the chk stream
1019
# Some of the missing_keys will be missing because they are ghosts.
1020
# As such, we can ignore them. The Sink is required to verify there are
1021
# no unavailable texts when the ghost inventories are not filled in.
1022
yield self._get_inventory_stream(missing_inventory_keys,
1024
# We use the empty set for excluded_revision_keys, to make it clear
1025
# that we want to transmit all referenced chk pages.
1026
for stream_info in self._get_filtered_chk_streams(set()):
1030
class RepositoryFormatCHK1(RepositoryFormatPack):
1031
"""A hashed CHK+group compress pack repository."""
1033
repository_class = CHKInventoryRepository
1034
supports_external_lookups = True
1035
supports_chks = True
1036
# For right now, setting this to True gives us InterModel1And2 rather
1037
# than InterDifferingSerializer
1038
_commit_builder_class = PackRootCommitBuilder
1039
rich_root_data = True
1040
_serializer = chk_serializer.chk_serializer_255_bigpage
1041
_commit_inv_deltas = True
1042
# What index classes to use
1043
index_builder_class = BTreeBuilder
1044
index_class = BTreeGraphIndex
1045
# Note: We cannot unpack a delta that references a text we haven't
1046
# seen yet. There are 2 options, work in fulltexts, or require
1047
# topological sorting. Using fulltexts is more optimal for local
1048
# operations, because the source can be smart about extracting
1049
# multiple in-a-row (and sharing strings). Topological is better
1050
# for remote, because we access less data.
1051
_fetch_order = 'unordered'
1052
_fetch_uses_deltas = False # essentially ignored by the groupcompress code.
1054
pack_compresses = True
1056
def _get_matching_bzrdir(self):
1057
return bzrdir.format_registry.make_bzrdir('development6-rich-root')
1059
def _ignore_setting_bzrdir(self, format):
1062
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1064
def get_format_string(self):
1065
"""See RepositoryFormat.get_format_string()."""
1066
return ('Bazaar development format - group compression and chk inventory'
1067
' (needs bzr.dev from 1.14)\n')
1069
def get_format_description(self):
1070
"""See RepositoryFormat.get_format_description()."""
1071
return ("Development repository format - rich roots, group compression"
1072
" and chk inventories")
1074
def check_conversion_target(self, target_format):
1075
if not target_format.rich_root_data:
1076
raise errors.BadConversionTarget(
1077
'Does not support rich root data.', target_format)
1078
if (self.supports_tree_reference and
1079
not getattr(target_format, 'supports_tree_reference', False)):
1080
raise errors.BadConversionTarget(
1081
'Does not support nested trees', target_format)
1085
class RepositoryFormatCHK2(RepositoryFormatCHK1):
1086
"""A CHK repository that uses the bencode revision serializer."""
1088
_serializer = chk_serializer.chk_bencode_serializer
1090
def _get_matching_bzrdir(self):
1091
return bzrdir.format_registry.make_bzrdir('development7-rich-root')
1093
def _ignore_setting_bzrdir(self, format):
1096
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1098
def get_format_string(self):
1099
"""See RepositoryFormat.get_format_string()."""
1100
return ('Bazaar development format - chk repository with bencode '
1101
'revision serialization (needs bzr.dev from 1.16)\n')
1104
class RepositoryFormat2a(RepositoryFormatCHK2):
1105
"""A CHK repository that uses the bencode revision serializer.
1107
This is the same as RepositoryFormatCHK2 but with a public name.
1110
_serializer = chk_serializer.chk_bencode_serializer
1112
def _get_matching_bzrdir(self):
1113
return bzrdir.format_registry.make_bzrdir('2a')
1115
def _ignore_setting_bzrdir(self, format):
1118
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1120
def get_format_string(self):
1121
return ('Bazaar repository format 2a (needs bzr 1.16 or later)\n')
added
removed
bzrlib/repofmt/groupcompress_repo.py
