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."""
34
revision as _mod_revision,
38
from bzrlib.btree_index import (
42
from bzrlib.index import GraphIndex, GraphIndexBuilder
43
from bzrlib.groupcompress import (
45
GroupCompressVersionedFiles,
47
from bzrlib.repofmt.pack_repo import (
52
PackRootCommitBuilder,
53
RepositoryPackCollection,
60
class GCPack(NewPack):
62
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
63
"""Create a NewPack instance.
65
:param pack_collection: A PackCollection into which this is being
67
:param upload_suffix: An optional suffix to be given to any temporary
68
files created during the pack creation. e.g '.autopack'
69
:param file_mode: An optional file mode to create the new files with.
71
# replaced from NewPack to:
72
# - change inventory reference list length to 1
73
# - change texts reference lists to 1
74
# TODO: patch this to be parameterised
76
# The relative locations of the packs are constrained, but all are
77
# passed in because the caller has them, so as to avoid object churn.
78
index_builder_class = pack_collection._index_builder_class
80
if pack_collection.chk_index is not None:
81
chk_index = index_builder_class(reference_lists=0)
85
# Revisions: parents list, no text compression.
86
index_builder_class(reference_lists=1),
87
# Inventory: We want to map compression only, but currently the
88
# knit code hasn't been updated enough to understand that, so we
89
# have a regular 2-list index giving parents and compression
91
index_builder_class(reference_lists=1),
92
# Texts: per file graph, for all fileids - so one reference list
93
# and two elements in the key tuple.
94
index_builder_class(reference_lists=1, key_elements=2),
95
# Signatures: Just blobs to store, no compression, no parents
97
index_builder_class(reference_lists=0),
98
# CHK based storage - just blobs, no compression or parents.
101
self._pack_collection = pack_collection
102
# When we make readonly indices, we need this.
103
self.index_class = pack_collection._index_class
104
# where should the new pack be opened
105
self.upload_transport = pack_collection._upload_transport
106
# where are indices written out to
107
self.index_transport = pack_collection._index_transport
108
# where is the pack renamed to when it is finished?
109
self.pack_transport = pack_collection._pack_transport
110
# What file mode to upload the pack and indices with.
111
self._file_mode = file_mode
112
# tracks the content written to the .pack file.
113
self._hash = osutils.md5()
114
# a four-tuple with the length in bytes of the indices, once the pack
115
# is finalised. (rev, inv, text, sigs)
116
self.index_sizes = None
117
# How much data to cache when writing packs. Note that this is not
118
# synchronised with reads, because it's not in the transport layer, so
119
# is not safe unless the client knows it won't be reading from the pack
121
self._cache_limit = 0
122
# the temporary pack file name.
123
self.random_name = osutils.rand_chars(20) + upload_suffix
124
# when was this pack started ?
125
self.start_time = time.time()
126
# open an output stream for the data added to the pack.
127
self.write_stream = self.upload_transport.open_write_stream(
128
self.random_name, mode=self._file_mode)
129
if 'pack' in debug.debug_flags:
130
trace.mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
131
time.ctime(), self.upload_transport.base, self.random_name,
132
time.time() - self.start_time)
133
# A list of byte sequences to be written to the new pack, and the
134
# aggregate size of them. Stored as a list rather than separate
135
# variables so that the _write_data closure below can update them.
136
self._buffer = [[], 0]
137
# create a callable for adding data
139
# robertc says- this is a closure rather than a method on the object
140
# so that the variables are locals, and faster than accessing object
142
def _write_data(bytes, flush=False, _buffer=self._buffer,
143
_write=self.write_stream.write, _update=self._hash.update):
144
_buffer[0].append(bytes)
145
_buffer[1] += len(bytes)
147
if _buffer[1] > self._cache_limit or flush:
148
bytes = ''.join(_buffer[0])
152
# expose this on self, for the occasion when clients want to add data.
153
self._write_data = _write_data
154
# a pack writer object to serialise pack records.
155
self._writer = pack.ContainerWriter(self._write_data)
157
# what state is the pack in? (open, finished, aborted)
160
def _check_references(self):
161
"""Make sure our external references are present.
163
Packs are allowed to have deltas whose base is not in the pack, but it
164
must be present somewhere in this collection. It is not allowed to
165
have deltas based on a fallback repository.
166
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
168
# Groupcompress packs don't have any external references, arguably CHK
169
# pages have external references, but we cannot 'cheaply' determine
170
# them without actually walking all of the chk pages.
173
class ResumedGCPack(ResumedPack):
175
def _check_references(self):
176
"""Make sure our external compression parents are present."""
177
# See GCPack._check_references for why this is empty
179
def _get_external_refs(self, index):
180
# GC repositories don't have compression parents external to a given
185
class GCCHKPacker(Packer):
186
"""This class understand what it takes to collect a GCCHK repo."""
188
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
190
super(GCCHKPacker, self).__init__(pack_collection, packs, suffix,
191
revision_ids=revision_ids,
192
reload_func=reload_func)
193
self._pack_collection = pack_collection
194
# ATM, We only support this for GCCHK repositories
195
if pack_collection.chk_index is None:
196
raise AssertionError('pack_collection.chk_index should not be None')
197
self._gather_text_refs = False
198
self._chk_id_roots = []
199
self._chk_p_id_roots = []
200
self._text_refs = None
201
# set by .pack() if self.revision_ids is not None
202
self.revision_keys = None
204
def _get_progress_stream(self, source_vf, keys, message, pb):
206
substream = source_vf.get_record_stream(keys, 'groupcompress', True)
207
for idx, record in enumerate(substream):
209
pb.update(message, idx + 1, len(keys))
213
def _get_filtered_inv_stream(self, source_vf, keys, message, pb=None):
214
"""Filter the texts of inventories, to find the chk pages."""
215
total_keys = len(keys)
216
def _filtered_inv_stream():
218
p_id_roots_set = set()
219
stream = source_vf.get_record_stream(keys, 'groupcompress', True)
220
for idx, record in enumerate(stream):
221
# Inventories should always be with revisions; assume success.
222
bytes = record.get_bytes_as('fulltext')
223
chk_inv = inventory.CHKInventory.deserialise(None, bytes,
226
pb.update('inv', idx, total_keys)
227
key = chk_inv.id_to_entry.key()
228
if key not in id_roots_set:
229
self._chk_id_roots.append(key)
230
id_roots_set.add(key)
231
p_id_map = chk_inv.parent_id_basename_to_file_id
233
raise AssertionError('Parent id -> file_id map not set')
235
if key not in p_id_roots_set:
236
p_id_roots_set.add(key)
237
self._chk_p_id_roots.append(key)
239
# We have finished processing all of the inventory records, we
240
# don't need these sets anymore
242
p_id_roots_set.clear()
243
return _filtered_inv_stream()
245
def _get_chk_streams(self, source_vf, keys, pb=None):
246
# We want to stream the keys from 'id_roots', and things they
247
# reference, and then stream things from p_id_roots and things they
248
# reference, and then any remaining keys that we didn't get to.
250
# We also group referenced texts together, so if one root references a
251
# text with prefix 'a', and another root references a node with prefix
252
# 'a', we want to yield those nodes before we yield the nodes for 'b'
253
# This keeps 'similar' nodes together.
255
# Note: We probably actually want multiple streams here, to help the
256
# client understand that the different levels won't compress well
257
# against each other.
258
# Test the difference between using one Group per level, and
259
# using 1 Group per prefix. (so '' (root) would get a group, then
260
# all the references to search-key 'a' would get a group, etc.)
261
total_keys = len(keys)
262
remaining_keys = set(keys)
264
if self._gather_text_refs:
265
bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
266
self._text_refs = set()
267
def _get_referenced_stream(root_keys, parse_leaf_nodes=False):
270
keys_by_search_prefix = {}
271
remaining_keys.difference_update(cur_keys)
273
def handle_internal_node(node):
274
for prefix, value in node._items.iteritems():
275
# We don't want to request the same key twice, and we
276
# want to order it by the first time it is seen.
277
# Even further, we don't want to request a key which is
278
# not in this group of pack files (it should be in the
279
# repo, but it doesn't have to be in the group being
281
# TODO: consider how to treat externally referenced chk
282
# pages as 'external_references' so that we
283
# always fill them in for stacked branches
284
if value not in next_keys and value in remaining_keys:
285
keys_by_search_prefix.setdefault(prefix,
288
def handle_leaf_node(node):
289
# Store is None, because we know we have a LeafNode, and we
290
# just want its entries
291
for file_id, bytes in node.iteritems(None):
292
name_utf8, file_id, revision_id = bytes_to_info(bytes)
293
self._text_refs.add((file_id, revision_id))
295
stream = source_vf.get_record_stream(cur_keys,
296
'as-requested', True)
297
for record in stream:
298
if record.storage_kind == 'absent':
299
# An absent CHK record: we assume that the missing
300
# record is in a different pack - e.g. a page not
301
# altered by the commit we're packing.
303
bytes = record.get_bytes_as('fulltext')
304
# We don't care about search_key_func for this code,
305
# because we only care about external references.
306
node = chk_map._deserialise(bytes, record.key,
307
search_key_func=None)
308
common_base = node._search_prefix
309
if isinstance(node, chk_map.InternalNode):
310
handle_internal_node(node)
311
elif parse_leaf_nodes:
312
handle_leaf_node(node)
315
pb.update('chk node', counter[0], total_keys)
318
# Double check that we won't be emitting any keys twice
319
# If we get rid of the pre-calculation of all keys, we could
320
# turn this around and do
321
# next_keys.difference_update(seen_keys)
322
# However, we also may have references to chk pages in another
323
# pack file during autopack. We filter earlier, so we should no
324
# longer need to do this
325
# next_keys = next_keys.intersection(remaining_keys)
327
for prefix in sorted(keys_by_search_prefix):
328
cur_keys.extend(keys_by_search_prefix.pop(prefix))
329
for stream in _get_referenced_stream(self._chk_id_roots,
330
self._gather_text_refs):
332
del self._chk_id_roots
333
# while it isn't really possible for chk_id_roots to not be in the
334
# local group of packs, it is possible that the tree shape has not
335
# changed recently, so we need to filter _chk_p_id_roots by the
337
chk_p_id_roots = [key for key in self._chk_p_id_roots
338
if key in remaining_keys]
339
del self._chk_p_id_roots
340
for stream in _get_referenced_stream(chk_p_id_roots, False):
343
trace.mutter('There were %d keys in the chk index, %d of which'
344
' were not referenced', total_keys,
346
if self.revision_ids is None:
347
stream = source_vf.get_record_stream(remaining_keys,
351
def _build_vf(self, index_name, parents, delta, for_write=False):
352
"""Build a VersionedFiles instance on top of this group of packs."""
353
index_name = index_name + '_index'
355
access = knit._DirectPackAccess(index_to_pack)
358
if self.new_pack is None:
359
raise AssertionError('No new pack has been set')
360
index = getattr(self.new_pack, index_name)
361
index_to_pack[index] = self.new_pack.access_tuple()
362
index.set_optimize(for_size=True)
363
access.set_writer(self.new_pack._writer, index,
364
self.new_pack.access_tuple())
365
add_callback = index.add_nodes
368
for pack in self.packs:
369
sub_index = getattr(pack, index_name)
370
index_to_pack[sub_index] = pack.access_tuple()
371
indices.append(sub_index)
372
index = _mod_index.CombinedGraphIndex(indices)
374
vf = GroupCompressVersionedFiles(
376
add_callback=add_callback,
378
is_locked=self._pack_collection.repo.is_locked),
383
def _build_vfs(self, index_name, parents, delta):
384
"""Build the source and target VersionedFiles."""
385
source_vf = self._build_vf(index_name, parents,
386
delta, for_write=False)
387
target_vf = self._build_vf(index_name, parents,
388
delta, for_write=True)
389
return source_vf, target_vf
391
def _copy_stream(self, source_vf, target_vf, keys, message, vf_to_stream,
393
trace.mutter('repacking %d %s', len(keys), message)
394
self.pb.update('repacking %s' % (message,), pb_offset)
395
child_pb = ui.ui_factory.nested_progress_bar()
397
stream = vf_to_stream(source_vf, keys, message, child_pb)
398
for _ in target_vf._insert_record_stream(stream,
405
def _copy_revision_texts(self):
406
source_vf, target_vf = self._build_vfs('revision', True, False)
407
if not self.revision_keys:
408
# We are doing a full fetch, aka 'pack'
409
self.revision_keys = source_vf.keys()
410
self._copy_stream(source_vf, target_vf, self.revision_keys,
411
'revisions', self._get_progress_stream, 1)
413
def _copy_inventory_texts(self):
414
source_vf, target_vf = self._build_vfs('inventory', True, True)
415
self._copy_stream(source_vf, target_vf, self.revision_keys,
416
'inventories', self._get_filtered_inv_stream, 2)
418
def _copy_chk_texts(self):
419
source_vf, target_vf = self._build_vfs('chk', False, False)
420
# TODO: This is technically spurious... if it is a performance issue,
422
total_keys = source_vf.keys()
423
trace.mutter('repacking chk: %d id_to_entry roots,'
424
' %d p_id_map roots, %d total keys',
425
len(self._chk_id_roots), len(self._chk_p_id_roots),
427
self.pb.update('repacking chk', 3)
428
child_pb = ui.ui_factory.nested_progress_bar()
430
for stream in self._get_chk_streams(source_vf, total_keys,
432
for _ in target_vf._insert_record_stream(stream,
439
def _copy_text_texts(self):
440
source_vf, target_vf = self._build_vfs('text', True, True)
441
# XXX: We don't walk the chk map to determine referenced (file_id,
442
# revision_id) keys. We don't do it yet because you really need
443
# to filter out the ones that are present in the parents of the
444
# rev just before the ones you are copying, otherwise the filter
445
# is grabbing too many keys...
446
text_keys = source_vf.keys()
447
self._copy_stream(source_vf, target_vf, text_keys,
448
'text', self._get_progress_stream, 4)
450
def _copy_signature_texts(self):
451
source_vf, target_vf = self._build_vfs('signature', False, False)
452
signature_keys = source_vf.keys()
453
signature_keys.intersection(self.revision_keys)
454
self._copy_stream(source_vf, target_vf, signature_keys,
455
'signatures', self._get_progress_stream, 5)
457
def _create_pack_from_packs(self):
458
self.pb.update('repacking', 0, 7)
459
self.new_pack = self.open_pack()
460
# Is this necessary for GC ?
461
self.new_pack.set_write_cache_size(1024*1024)
462
self._copy_revision_texts()
463
self._copy_inventory_texts()
464
self._copy_chk_texts()
465
self._copy_text_texts()
466
self._copy_signature_texts()
467
self.new_pack._check_references()
468
if not self._use_pack(self.new_pack):
469
self.new_pack.abort()
471
self.pb.update('finishing repack', 6, 7)
472
self.new_pack.finish()
473
self._pack_collection.allocate(self.new_pack)
477
class GCCHKReconcilePacker(GCCHKPacker):
478
"""A packer which regenerates indices etc as it copies.
480
This is used by ``bzr reconcile`` to cause parent text pointers to be
484
def __init__(self, *args, **kwargs):
485
super(GCCHKReconcilePacker, self).__init__(*args, **kwargs)
486
self._data_changed = False
487
self._gather_text_refs = True
489
def _copy_inventory_texts(self):
490
source_vf, target_vf = self._build_vfs('inventory', True, True)
491
self._copy_stream(source_vf, target_vf, self.revision_keys,
492
'inventories', self._get_filtered_inv_stream, 2)
493
if source_vf.keys() != self.revision_keys:
494
self._data_changed = True
496
def _copy_text_texts(self):
497
"""generate what texts we should have and then copy."""
498
source_vf, target_vf = self._build_vfs('text', True, True)
499
trace.mutter('repacking %d texts', len(self._text_refs))
500
self.pb.update("repacking texts", 4)
501
# we have three major tasks here:
502
# 1) generate the ideal index
503
repo = self._pack_collection.repo
504
# We want the one we just wrote, so base it on self.new_pack
505
revision_vf = self._build_vf('revision', True, False, for_write=True)
506
ancestor_keys = revision_vf.get_parent_map(revision_vf.keys())
507
# Strip keys back into revision_ids.
508
ancestors = dict((k[0], tuple([p[0] for p in parents]))
509
for k, parents in ancestor_keys.iteritems())
511
# TODO: _generate_text_key_index should be much cheaper to generate from
512
# a chk repository, rather than the current implementation
513
ideal_index = repo._generate_text_key_index(None, ancestors)
514
file_id_parent_map = source_vf.get_parent_map(self._text_refs)
515
# 2) generate a keys list that contains all the entries that can
516
# be used as-is, with corrected parents.
518
new_parent_keys = {} # (key, parent_keys)
520
NULL_REVISION = _mod_revision.NULL_REVISION
521
for key in self._text_refs:
527
ideal_parents = tuple(ideal_index[key])
529
discarded_keys.append(key)
530
self._data_changed = True
532
if ideal_parents == (NULL_REVISION,):
534
source_parents = file_id_parent_map[key]
535
if ideal_parents == source_parents:
539
# We need to change the parent graph, but we don't need to
540
# re-insert the text (since we don't pun the compression
541
# parent with the parents list)
542
self._data_changed = True
543
new_parent_keys[key] = ideal_parents
544
# we're finished with some data.
546
del file_id_parent_map
547
# 3) bulk copy the data, updating records than need it
548
def _update_parents_for_texts():
549
stream = source_vf.get_record_stream(self._text_refs,
550
'groupcompress', False)
551
for record in stream:
552
if record.key in new_parent_keys:
553
record.parents = new_parent_keys[record.key]
555
target_vf.insert_record_stream(_update_parents_for_texts())
557
def _use_pack(self, new_pack):
558
"""Override _use_pack to check for reconcile having changed content."""
559
return new_pack.data_inserted() and self._data_changed
562
class GCRepositoryPackCollection(RepositoryPackCollection):
564
pack_factory = GCPack
565
resumed_pack_factory = ResumedGCPack
567
def _execute_pack_operations(self, pack_operations,
568
_packer_class=GCCHKPacker,
570
"""Execute a series of pack operations.
572
:param pack_operations: A list of [revision_count, packs_to_combine].
573
:param _packer_class: The class of packer to use (default: Packer).
576
# XXX: Copied across from RepositoryPackCollection simply because we
577
# want to override the _packer_class ... :(
578
for revision_count, packs in pack_operations:
579
# we may have no-ops from the setup logic
582
packer = GCCHKPacker(self, packs, '.autopack',
583
reload_func=reload_func)
586
except errors.RetryWithNewPacks:
587
# An exception is propagating out of this context, make sure
588
# this packer has cleaned up. Packer() doesn't set its new_pack
589
# state into the RepositoryPackCollection object, so we only
590
# have access to it directly here.
591
if packer.new_pack is not None:
592
packer.new_pack.abort()
595
self._remove_pack_from_memory(pack)
596
# record the newly available packs and stop advertising the old
598
self._save_pack_names(clear_obsolete_packs=True)
599
# Move the old packs out of the way now they are no longer referenced.
600
for revision_count, packs in pack_operations:
601
self._obsolete_packs(packs)
604
class CHKInventoryRepository(KnitPackRepository):
605
"""subclass of KnitPackRepository that uses CHK based inventories."""
607
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
609
"""Overridden to change pack collection class."""
610
KnitPackRepository.__init__(self, _format, a_bzrdir, control_files,
611
_commit_builder_class, _serializer)
612
# and now replace everything it did :)
613
index_transport = self._transport.clone('indices')
614
self._pack_collection = GCRepositoryPackCollection(self,
615
self._transport, index_transport,
616
self._transport.clone('upload'),
617
self._transport.clone('packs'),
618
_format.index_builder_class,
620
use_chk_index=self._format.supports_chks,
622
self.inventories = GroupCompressVersionedFiles(
623
_GCGraphIndex(self._pack_collection.inventory_index.combined_index,
624
add_callback=self._pack_collection.inventory_index.add_callback,
625
parents=True, is_locked=self.is_locked),
626
access=self._pack_collection.inventory_index.data_access)
627
self.revisions = GroupCompressVersionedFiles(
628
_GCGraphIndex(self._pack_collection.revision_index.combined_index,
629
add_callback=self._pack_collection.revision_index.add_callback,
630
parents=True, is_locked=self.is_locked,
631
track_external_parent_refs=True),
632
access=self._pack_collection.revision_index.data_access,
634
self.signatures = GroupCompressVersionedFiles(
635
_GCGraphIndex(self._pack_collection.signature_index.combined_index,
636
add_callback=self._pack_collection.signature_index.add_callback,
637
parents=False, is_locked=self.is_locked),
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
access=self._pack_collection.text_index.data_access)
645
# No parents, individual CHK pages don't have specific ancestry
646
self.chk_bytes = GroupCompressVersionedFiles(
647
_GCGraphIndex(self._pack_collection.chk_index.combined_index,
648
add_callback=self._pack_collection.chk_index.add_callback,
649
parents=False, is_locked=self.is_locked),
650
access=self._pack_collection.chk_index.data_access)
651
# True when the repository object is 'write locked' (as opposed to the
652
# physical lock only taken out around changes to the pack-names list.)
653
# Another way to represent this would be a decorator around the control
654
# files object that presents logical locks as physical ones - if this
655
# gets ugly consider that alternative design. RBC 20071011
656
self._write_lock_count = 0
657
self._transaction = None
659
self._reconcile_does_inventory_gc = True
660
self._reconcile_fixes_text_parents = True
661
self._reconcile_backsup_inventory = False
663
def _add_inventory_checked(self, revision_id, inv, parents):
664
"""Add inv to the repository after checking the inputs.
666
This function can be overridden to allow different inventory styles.
668
:seealso: add_inventory, for the contract.
671
serializer = self._format._serializer
672
result = inventory.CHKInventory.from_inventory(self.chk_bytes, inv,
673
maximum_size=serializer.maximum_size,
674
search_key_name=serializer.search_key_name)
675
inv_lines = result.to_lines()
676
return self._inventory_add_lines(revision_id, parents,
677
inv_lines, check_content=False)
679
def _create_inv_from_null(self, delta, revision_id):
680
"""This will mutate new_inv directly.
682
This is a simplified form of create_by_apply_delta which knows that all
683
the old values must be None, so everything is a create.
685
serializer = self._format._serializer
686
new_inv = inventory.CHKInventory(serializer.search_key_name)
687
new_inv.revision_id = revision_id
688
entry_to_bytes = new_inv._entry_to_bytes
689
id_to_entry_dict = {}
690
parent_id_basename_dict = {}
691
for old_path, new_path, file_id, entry in delta:
692
if old_path is not None:
693
raise ValueError('Invalid delta, somebody tried to delete %r'
694
' from the NULL_REVISION'
695
% ((old_path, file_id),))
697
raise ValueError('Invalid delta, delta from NULL_REVISION has'
698
' no new_path %r' % (file_id,))
700
new_inv.root_id = file_id
701
parent_id_basename_key = ('', '')
703
utf8_entry_name = entry.name.encode('utf-8')
704
parent_id_basename_key = (entry.parent_id, utf8_entry_name)
705
new_value = entry_to_bytes(entry)
707
# new_inv._path_to_fileid_cache[new_path] = file_id
708
id_to_entry_dict[(file_id,)] = new_value
709
parent_id_basename_dict[parent_id_basename_key] = file_id
711
new_inv._populate_from_dicts(self.chk_bytes, id_to_entry_dict,
712
parent_id_basename_dict, maximum_size=serializer.maximum_size)
715
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
716
parents, basis_inv=None, propagate_caches=False):
717
"""Add a new inventory expressed as a delta against another revision.
719
:param basis_revision_id: The inventory id the delta was created
721
:param delta: The inventory delta (see Inventory.apply_delta for
723
:param new_revision_id: The revision id that the inventory is being
725
:param parents: The revision ids of the parents that revision_id is
726
known to have and are in the repository already. These are supplied
727
for repositories that depend on the inventory graph for revision
728
graph access, as well as for those that pun ancestry with delta
730
:param basis_inv: The basis inventory if it is already known,
732
:param propagate_caches: If True, the caches for this inventory are
733
copied to and updated for the result if possible.
735
:returns: (validator, new_inv)
736
The validator(which is a sha1 digest, though what is sha'd is
737
repository format specific) of the serialized inventory, and the
740
if not self.is_in_write_group():
741
raise AssertionError("%r not in write group" % (self,))
742
_mod_revision.check_not_reserved_id(new_revision_id)
744
if basis_inv is None:
745
if basis_revision_id == _mod_revision.NULL_REVISION:
746
new_inv = self._create_inv_from_null(delta, new_revision_id)
747
inv_lines = new_inv.to_lines()
748
return self._inventory_add_lines(new_revision_id, parents,
749
inv_lines, check_content=False), new_inv
751
basis_tree = self.revision_tree(basis_revision_id)
752
basis_tree.lock_read()
753
basis_inv = basis_tree.inventory
755
result = basis_inv.create_by_apply_delta(delta, new_revision_id,
756
propagate_caches=propagate_caches)
757
inv_lines = result.to_lines()
758
return self._inventory_add_lines(new_revision_id, parents,
759
inv_lines, check_content=False), result
761
if basis_tree is not None:
764
def _iter_inventories(self, revision_ids):
765
"""Iterate over many inventory objects."""
766
keys = [(revision_id,) for revision_id in revision_ids]
767
stream = self.inventories.get_record_stream(keys, 'unordered', True)
769
for record in stream:
770
if record.storage_kind != 'absent':
771
texts[record.key] = record.get_bytes_as('fulltext')
773
raise errors.NoSuchRevision(self, record.key)
775
yield inventory.CHKInventory.deserialise(self.chk_bytes, texts[key], key)
777
def _iter_inventory_xmls(self, revision_ids):
778
# Without a native 'xml' inventory, this method doesn't make sense, so
779
# make it raise to trap naughty direct users.
780
raise NotImplementedError(self._iter_inventory_xmls)
782
def _find_present_inventory_keys(self, revision_keys):
783
parent_map = self.inventories.get_parent_map(revision_keys)
784
present_inventory_keys = set(k for k in parent_map)
785
return present_inventory_keys
787
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
788
"""Find the file ids and versions affected by revisions.
790
:param revisions: an iterable containing revision ids.
791
:param _inv_weave: The inventory weave from this repository or None.
792
If None, the inventory weave will be opened automatically.
793
:return: a dictionary mapping altered file-ids to an iterable of
794
revision_ids. Each altered file-ids has the exact revision_ids that
795
altered it listed explicitly.
797
rich_root = self.supports_rich_root()
798
bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
799
file_id_revisions = {}
800
pb = ui.ui_factory.nested_progress_bar()
802
revision_keys = [(r,) for r in revision_ids]
803
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
804
# TODO: instead of using _find_present_inventory_keys, change the
805
# code paths to allow missing inventories to be tolerated.
806
# However, we only want to tolerate missing parent
807
# inventories, not missing inventories for revision_ids
808
present_parent_inv_keys = self._find_present_inventory_keys(
810
present_parent_inv_ids = set(
811
[k[-1] for k in present_parent_inv_keys])
812
uninteresting_root_keys = set()
813
interesting_root_keys = set()
814
inventories_to_read = set(revision_ids)
815
inventories_to_read.update(present_parent_inv_ids)
816
for inv in self.iter_inventories(inventories_to_read):
817
entry_chk_root_key = inv.id_to_entry.key()
818
if inv.revision_id in present_parent_inv_ids:
819
uninteresting_root_keys.add(entry_chk_root_key)
821
interesting_root_keys.add(entry_chk_root_key)
823
chk_bytes = self.chk_bytes
824
for record, items in chk_map.iter_interesting_nodes(chk_bytes,
825
interesting_root_keys, uninteresting_root_keys,
827
for name, bytes in items:
828
(name_utf8, file_id, revision_id) = bytes_to_info(bytes)
829
if not rich_root and name_utf8 == '':
832
file_id_revisions[file_id].add(revision_id)
834
file_id_revisions[file_id] = set([revision_id])
837
return file_id_revisions
839
def find_text_key_references(self):
840
"""Find the text key references within the repository.
842
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
843
to whether they were referred to by the inventory of the
844
revision_id that they contain. The inventory texts from all present
845
revision ids are assessed to generate this report.
847
# XXX: Slow version but correct: rewrite as a series of delta
848
# examinations/direct tree traversal. Note that that will require care
849
# as a common node is reachable both from the inventory that added it,
850
# and others afterwards.
851
revision_keys = self.revisions.keys()
853
rich_roots = self.supports_rich_root()
854
pb = ui.ui_factory.nested_progress_bar()
856
all_revs = self.all_revision_ids()
857
total = len(all_revs)
858
for pos, inv in enumerate(self.iter_inventories(all_revs)):
859
pb.update("Finding text references", pos, total)
860
for _, entry in inv.iter_entries():
861
if not rich_roots and entry.file_id == inv.root_id:
863
key = (entry.file_id, entry.revision)
864
result.setdefault(key, False)
865
if entry.revision == inv.revision_id:
871
def _reconcile_pack(self, collection, packs, extension, revs, pb):
872
packer = GCCHKReconcilePacker(collection, packs, extension)
873
return packer.pack(pb)
875
def _get_source(self, to_format):
876
"""Return a source for streaming from this repository."""
877
if isinstance(to_format, remote.RemoteRepositoryFormat):
878
# Can't just check attributes on to_format with the current code,
880
to_format._ensure_real()
881
to_format = to_format._custom_format
882
if to_format.__class__ is self._format.__class__:
883
# We must be exactly the same format, otherwise stuff like the chk
884
# page layout might be different
885
return GroupCHKStreamSource(self, to_format)
886
return super(CHKInventoryRepository, self)._get_source(to_format)
889
class GroupCHKStreamSource(KnitPackStreamSource):
890
"""Used when both the source and target repo are GroupCHK repos."""
892
def __init__(self, from_repository, to_format):
893
"""Create a StreamSource streaming from from_repository."""
894
super(GroupCHKStreamSource, self).__init__(from_repository, to_format)
895
self._revision_keys = None
896
self._text_keys = None
897
self._text_fetch_order = 'groupcompress'
898
self._chk_id_roots = None
899
self._chk_p_id_roots = None
901
def _get_inventory_stream(self, inventory_keys, allow_absent=False):
902
"""Get a stream of inventory texts.
904
When this function returns, self._chk_id_roots and self._chk_p_id_roots
907
self._chk_id_roots = []
908
self._chk_p_id_roots = []
909
def _filtered_inv_stream():
911
p_id_roots_set = set()
912
source_vf = self.from_repository.inventories
913
stream = source_vf.get_record_stream(inventory_keys,
914
'groupcompress', True)
915
for record in stream:
916
if record.storage_kind == 'absent':
920
raise errors.NoSuchRevision(self, record.key)
921
bytes = record.get_bytes_as('fulltext')
922
chk_inv = inventory.CHKInventory.deserialise(None, bytes,
924
key = chk_inv.id_to_entry.key()
925
if key not in id_roots_set:
926
self._chk_id_roots.append(key)
927
id_roots_set.add(key)
928
p_id_map = chk_inv.parent_id_basename_to_file_id
930
raise AssertionError('Parent id -> file_id map not set')
932
if key not in p_id_roots_set:
933
p_id_roots_set.add(key)
934
self._chk_p_id_roots.append(key)
936
# We have finished processing all of the inventory records, we
937
# don't need these sets anymore
939
p_id_roots_set.clear()
940
return ('inventories', _filtered_inv_stream())
942
def _get_filtered_chk_streams(self, excluded_revision_keys):
943
self._text_keys = set()
944
excluded_revision_keys.discard(_mod_revision.NULL_REVISION)
945
if not excluded_revision_keys:
946
uninteresting_root_keys = set()
947
uninteresting_pid_root_keys = set()
949
# filter out any excluded revisions whose inventories are not
951
# TODO: Update Repository.iter_inventories() to add
952
# ignore_missing=True
953
present_keys = self.from_repository._find_present_inventory_keys(
954
excluded_revision_keys)
955
present_ids = [k[-1] for k in present_keys]
956
uninteresting_root_keys = set()
957
uninteresting_pid_root_keys = set()
958
for inv in self.from_repository.iter_inventories(present_ids):
959
uninteresting_root_keys.add(inv.id_to_entry.key())
960
uninteresting_pid_root_keys.add(
961
inv.parent_id_basename_to_file_id.key())
962
bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
963
chk_bytes = self.from_repository.chk_bytes
964
def _filter_id_to_entry():
965
for record, items in chk_map.iter_interesting_nodes(chk_bytes,
966
self._chk_id_roots, uninteresting_root_keys):
967
for name, bytes in items:
968
# Note: we don't care about name_utf8, because we are always
970
_, file_id, revision_id = bytes_to_info(bytes)
971
self._text_keys.add((file_id, revision_id))
972
if record is not None:
975
self._chk_id_roots = None
976
yield 'chk_bytes', _filter_id_to_entry()
977
def _get_parent_id_basename_to_file_id_pages():
978
for record, items in chk_map.iter_interesting_nodes(chk_bytes,
979
self._chk_p_id_roots, uninteresting_pid_root_keys):
980
if record is not None:
983
self._chk_p_id_roots = None
984
yield 'chk_bytes', _get_parent_id_basename_to_file_id_pages()
986
def get_stream(self, search):
987
revision_ids = search.get_keys()
988
for stream_info in self._fetch_revision_texts(revision_ids):
990
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
991
yield self._get_inventory_stream(self._revision_keys)
992
# TODO: The keys to exclude might be part of the search recipe
993
# For now, exclude all parents that are at the edge of ancestry, for
994
# which we have inventories
995
from_repo = self.from_repository
996
parent_keys = from_repo._find_parent_keys_of_revisions(
998
for stream_info in self._get_filtered_chk_streams(parent_keys):
1000
yield self._get_text_stream()
1002
def get_stream_for_missing_keys(self, missing_keys):
1003
# missing keys can only occur when we are byte copying and not
1004
# translating (because translation means we don't send
1005
# unreconstructable deltas ever).
1006
missing_inventory_keys = set()
1007
for key in missing_keys:
1008
if key[0] != 'inventories':
1009
raise AssertionError('The only missing keys we should'
1010
' be filling in are inventory keys, not %s'
1012
missing_inventory_keys.add(key[1:])
1013
if self._chk_id_roots or self._chk_p_id_roots:
1014
raise AssertionError('Cannot call get_stream_for_missing_keys'
1015
' untill all of get_stream() has been consumed.')
1016
# Yield the inventory stream, so we can find the chk stream
1017
# Some of the missing_keys will be missing because they are ghosts.
1018
# As such, we can ignore them. The Sink is required to verify there are
1019
# no unavailable texts when the ghost inventories are not filled in.
1020
yield self._get_inventory_stream(missing_inventory_keys,
1022
# We use the empty set for excluded_revision_keys, to make it clear
1023
# that we want to transmit all referenced chk pages.
1024
for stream_info in self._get_filtered_chk_streams(set()):
1028
class RepositoryFormatCHK1(RepositoryFormatPack):
1029
"""A hashed CHK+group compress pack repository."""
1031
repository_class = CHKInventoryRepository
1032
supports_external_lookups = True
1033
supports_chks = True
1034
# For right now, setting this to True gives us InterModel1And2 rather
1035
# than InterDifferingSerializer
1036
_commit_builder_class = PackRootCommitBuilder
1037
rich_root_data = True
1038
_serializer = chk_serializer.chk_serializer_255_bigpage
1039
_commit_inv_deltas = True
1040
# What index classes to use
1041
index_builder_class = BTreeBuilder
1042
index_class = BTreeGraphIndex
1043
# Note: We cannot unpack a delta that references a text we haven't
1044
# seen yet. There are 2 options, work in fulltexts, or require
1045
# topological sorting. Using fulltexts is more optimal for local
1046
# operations, because the source can be smart about extracting
1047
# multiple in-a-row (and sharing strings). Topological is better
1048
# for remote, because we access less data.
1049
_fetch_order = 'unordered'
1050
_fetch_uses_deltas = False # essentially ignored by the groupcompress code.
1052
pack_compresses = True
1054
def _get_matching_bzrdir(self):
1055
return bzrdir.format_registry.make_bzrdir('development6-rich-root')
1057
def _ignore_setting_bzrdir(self, format):
1060
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1062
def get_format_string(self):
1063
"""See RepositoryFormat.get_format_string()."""
1064
return ('Bazaar development format - group compression and chk inventory'
1065
' (needs bzr.dev from 1.14)\n')
1067
def get_format_description(self):
1068
"""See RepositoryFormat.get_format_description()."""
1069
return ("Development repository format - rich roots, group compression"
1070
" and chk inventories")
1072
def check_conversion_target(self, target_format):
1073
if not target_format.rich_root_data:
1074
raise errors.BadConversionTarget(
1075
'Does not support rich root data.', target_format)
1076
if (self.supports_tree_reference and
1077
not getattr(target_format, 'supports_tree_reference', False)):
1078
raise errors.BadConversionTarget(
1079
'Does not support nested trees', target_format)
1083
class RepositoryFormatCHK2(RepositoryFormatCHK1):
1084
"""A CHK repository that uses the bencode revision serializer."""
1086
_serializer = chk_serializer.chk_bencode_serializer
1088
def _get_matching_bzrdir(self):
1089
return bzrdir.format_registry.make_bzrdir('development7-rich-root')
1091
def _ignore_setting_bzrdir(self, format):
1094
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1096
def get_format_string(self):
1097
"""See RepositoryFormat.get_format_string()."""
1098
return ('Bazaar development format - chk repository with bencode '
1099
'revision serialization (needs bzr.dev from 1.16)\n')
1102
class RepositoryFormat2a(RepositoryFormatCHK2):
1103
"""A CHK repository that uses the bencode revision serializer.
1105
This is the same as RepositoryFormatCHK2 but with a public name.
1108
_serializer = chk_serializer.chk_bencode_serializer
1110
def _get_matching_bzrdir(self):
1111
return bzrdir.format_registry.make_bzrdir('2a')
1113
def _ignore_setting_bzrdir(self, format):
1116
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1118
def get_format_string(self):
1119
return ('Bazaar repository format 2a (needs bzr 1.16 or later)\n')
added
removed
bzrlib/repofmt/groupcompress_repo.py
