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,
56
from bzrlib.static_tuple import StaticTuple
59
class GCPack(NewPack):
61
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
62
"""Create a NewPack instance.
64
:param pack_collection: A PackCollection into which this is being
66
:param upload_suffix: An optional suffix to be given to any temporary
67
files created during the pack creation. e.g '.autopack'
68
:param file_mode: An optional file mode to create the new files with.
70
# replaced from NewPack to:
71
# - change inventory reference list length to 1
72
# - change texts reference lists to 1
73
# TODO: patch this to be parameterised
75
# The relative locations of the packs are constrained, but all are
76
# passed in because the caller has them, so as to avoid object churn.
77
index_builder_class = pack_collection._index_builder_class
79
if pack_collection.chk_index is not None:
80
chk_index = index_builder_class(reference_lists=0)
84
# Revisions: parents list, no text compression.
85
index_builder_class(reference_lists=1),
86
# Inventory: We want to map compression only, but currently the
87
# knit code hasn't been updated enough to understand that, so we
88
# have a regular 2-list index giving parents and compression
90
index_builder_class(reference_lists=1),
91
# Texts: per file graph, for all fileids - so one reference list
92
# and two elements in the key tuple.
93
index_builder_class(reference_lists=1, key_elements=2),
94
# Signatures: Just blobs to store, no compression, no parents
96
index_builder_class(reference_lists=0),
97
# CHK based storage - just blobs, no compression or parents.
100
self._pack_collection = pack_collection
101
# When we make readonly indices, we need this.
102
self.index_class = pack_collection._index_class
103
# where should the new pack be opened
104
self.upload_transport = pack_collection._upload_transport
105
# where are indices written out to
106
self.index_transport = pack_collection._index_transport
107
# where is the pack renamed to when it is finished?
108
self.pack_transport = pack_collection._pack_transport
109
# What file mode to upload the pack and indices with.
110
self._file_mode = file_mode
111
# tracks the content written to the .pack file.
112
self._hash = osutils.md5()
113
# a four-tuple with the length in bytes of the indices, once the pack
114
# is finalised. (rev, inv, text, sigs)
115
self.index_sizes = None
116
# How much data to cache when writing packs. Note that this is not
117
# synchronised with reads, because it's not in the transport layer, so
118
# is not safe unless the client knows it won't be reading from the pack
120
self._cache_limit = 0
121
# the temporary pack file name.
122
self.random_name = osutils.rand_chars(20) + upload_suffix
123
# when was this pack started ?
124
self.start_time = time.time()
125
# open an output stream for the data added to the pack.
126
self.write_stream = self.upload_transport.open_write_stream(
127
self.random_name, mode=self._file_mode)
128
if 'pack' in debug.debug_flags:
129
trace.mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
130
time.ctime(), self.upload_transport.base, self.random_name,
131
time.time() - self.start_time)
132
# A list of byte sequences to be written to the new pack, and the
133
# aggregate size of them. Stored as a list rather than separate
134
# variables so that the _write_data closure below can update them.
135
self._buffer = [[], 0]
136
# create a callable for adding data
138
# robertc says- this is a closure rather than a method on the object
139
# so that the variables are locals, and faster than accessing object
141
def _write_data(bytes, flush=False, _buffer=self._buffer,
142
_write=self.write_stream.write, _update=self._hash.update):
143
_buffer[0].append(bytes)
144
_buffer[1] += len(bytes)
146
if _buffer[1] > self._cache_limit or flush:
147
bytes = ''.join(_buffer[0])
151
# expose this on self, for the occasion when clients want to add data.
152
self._write_data = _write_data
153
# a pack writer object to serialise pack records.
154
self._writer = pack.ContainerWriter(self._write_data)
156
# what state is the pack in? (open, finished, aborted)
158
# no name until we finish writing the content
161
def _check_references(self):
162
"""Make sure our external references are present.
164
Packs are allowed to have deltas whose base is not in the pack, but it
165
must be present somewhere in this collection. It is not allowed to
166
have deltas based on a fallback repository.
167
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
169
# Groupcompress packs don't have any external references, arguably CHK
170
# pages have external references, but we cannot 'cheaply' determine
171
# them without actually walking all of the chk pages.
174
class ResumedGCPack(ResumedPack):
176
def _check_references(self):
177
"""Make sure our external compression parents are present."""
178
# See GCPack._check_references for why this is empty
180
def _get_external_refs(self, index):
181
# GC repositories don't have compression parents external to a given
186
class GCCHKPacker(Packer):
187
"""This class understand what it takes to collect a GCCHK repo."""
189
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
191
super(GCCHKPacker, self).__init__(pack_collection, packs, suffix,
192
revision_ids=revision_ids,
193
reload_func=reload_func)
194
self._pack_collection = pack_collection
195
# ATM, We only support this for GCCHK repositories
196
if pack_collection.chk_index is None:
197
raise AssertionError('pack_collection.chk_index should not be None')
198
self._gather_text_refs = False
199
self._chk_id_roots = []
200
self._chk_p_id_roots = []
201
self._text_refs = None
202
# set by .pack() if self.revision_ids is not None
203
self.revision_keys = None
205
def _get_progress_stream(self, source_vf, keys, message, pb):
207
substream = source_vf.get_record_stream(keys, 'groupcompress', True)
208
for idx, record in enumerate(substream):
210
pb.update(message, idx + 1, len(keys))
214
def _get_filtered_inv_stream(self, source_vf, keys, message, pb=None):
215
"""Filter the texts of inventories, to find the chk pages."""
216
total_keys = len(keys)
217
def _filtered_inv_stream():
219
p_id_roots_set = set()
220
stream = source_vf.get_record_stream(keys, 'groupcompress', True)
221
for idx, record in enumerate(stream):
222
# Inventories should always be with revisions; assume success.
223
bytes = record.get_bytes_as('fulltext')
224
chk_inv = inventory.CHKInventory.deserialise(None, bytes,
227
pb.update('inv', idx, total_keys)
228
key = chk_inv.id_to_entry.key()
229
if key not in id_roots_set:
230
self._chk_id_roots.append(key)
231
id_roots_set.add(key)
232
p_id_map = chk_inv.parent_id_basename_to_file_id
234
raise AssertionError('Parent id -> file_id map not set')
236
if key not in p_id_roots_set:
237
p_id_roots_set.add(key)
238
self._chk_p_id_roots.append(key)
240
# We have finished processing all of the inventory records, we
241
# don't need these sets anymore
243
p_id_roots_set.clear()
244
return _filtered_inv_stream()
246
def _get_chk_streams(self, source_vf, keys, pb=None):
247
# We want to stream the keys from 'id_roots', and things they
248
# reference, and then stream things from p_id_roots and things they
249
# reference, and then any remaining keys that we didn't get to.
251
# We also group referenced texts together, so if one root references a
252
# text with prefix 'a', and another root references a node with prefix
253
# 'a', we want to yield those nodes before we yield the nodes for 'b'
254
# This keeps 'similar' nodes together.
256
# Note: We probably actually want multiple streams here, to help the
257
# client understand that the different levels won't compress well
258
# against each other.
259
# Test the difference between using one Group per level, and
260
# using 1 Group per prefix. (so '' (root) would get a group, then
261
# all the references to search-key 'a' would get a group, etc.)
262
total_keys = len(keys)
263
remaining_keys = set(keys)
265
if self._gather_text_refs:
266
bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
267
self._text_refs = set()
268
def _get_referenced_stream(root_keys, parse_leaf_nodes=False):
271
keys_by_search_prefix = {}
272
remaining_keys.difference_update(cur_keys)
274
def handle_internal_node(node):
275
for prefix, value in node._items.iteritems():
276
# We don't want to request the same key twice, and we
277
# want to order it by the first time it is seen.
278
# Even further, we don't want to request a key which is
279
# not in this group of pack files (it should be in the
280
# repo, but it doesn't have to be in the group being
282
# TODO: consider how to treat externally referenced chk
283
# pages as 'external_references' so that we
284
# always fill them in for stacked branches
285
if value not in next_keys and value in remaining_keys:
286
keys_by_search_prefix.setdefault(prefix,
289
def handle_leaf_node(node):
290
# Store is None, because we know we have a LeafNode, and we
291
# just want its entries
292
for file_id, bytes in node.iteritems(None):
293
name_utf8, file_id, revision_id = bytes_to_info(bytes)
294
self._text_refs.add((file_id, revision_id))
296
stream = source_vf.get_record_stream(cur_keys,
297
'as-requested', True)
298
for record in stream:
299
if record.storage_kind == 'absent':
300
# An absent CHK record: we assume that the missing
301
# record is in a different pack - e.g. a page not
302
# altered by the commit we're packing.
304
bytes = record.get_bytes_as('fulltext')
305
# We don't care about search_key_func for this code,
306
# because we only care about external references.
307
node = chk_map._deserialise(bytes, record.key,
308
search_key_func=None)
309
common_base = node._search_prefix
310
if isinstance(node, chk_map.InternalNode):
311
handle_internal_node(node)
312
elif parse_leaf_nodes:
313
handle_leaf_node(node)
316
pb.update('chk node', counter[0], total_keys)
319
# Double check that we won't be emitting any keys twice
320
# If we get rid of the pre-calculation of all keys, we could
321
# turn this around and do
322
# next_keys.difference_update(seen_keys)
323
# However, we also may have references to chk pages in another
324
# pack file during autopack. We filter earlier, so we should no
325
# longer need to do this
326
# next_keys = next_keys.intersection(remaining_keys)
328
for prefix in sorted(keys_by_search_prefix):
329
cur_keys.extend(keys_by_search_prefix.pop(prefix))
330
for stream in _get_referenced_stream(self._chk_id_roots,
331
self._gather_text_refs):
333
del self._chk_id_roots
334
# while it isn't really possible for chk_id_roots to not be in the
335
# local group of packs, it is possible that the tree shape has not
336
# changed recently, so we need to filter _chk_p_id_roots by the
338
chk_p_id_roots = [key for key in self._chk_p_id_roots
339
if key in remaining_keys]
340
del self._chk_p_id_roots
341
for stream in _get_referenced_stream(chk_p_id_roots, False):
344
trace.mutter('There were %d keys in the chk index, %d of which'
345
' were not referenced', total_keys,
347
if self.revision_ids is None:
348
stream = source_vf.get_record_stream(remaining_keys,
352
def _build_vf(self, index_name, parents, delta, for_write=False):
353
"""Build a VersionedFiles instance on top of this group of packs."""
354
index_name = index_name + '_index'
356
access = knit._DirectPackAccess(index_to_pack,
357
reload_func=self._reload_func)
360
if self.new_pack is None:
361
raise AssertionError('No new pack has been set')
362
index = getattr(self.new_pack, index_name)
363
index_to_pack[index] = self.new_pack.access_tuple()
364
index.set_optimize(for_size=True)
365
access.set_writer(self.new_pack._writer, index,
366
self.new_pack.access_tuple())
367
add_callback = index.add_nodes
370
for pack in self.packs:
371
sub_index = getattr(pack, index_name)
372
index_to_pack[sub_index] = pack.access_tuple()
373
indices.append(sub_index)
374
index = _mod_index.CombinedGraphIndex(indices)
376
vf = GroupCompressVersionedFiles(
378
add_callback=add_callback,
380
is_locked=self._pack_collection.repo.is_locked),
385
def _build_vfs(self, index_name, parents, delta):
386
"""Build the source and target VersionedFiles."""
387
source_vf = self._build_vf(index_name, parents,
388
delta, for_write=False)
389
target_vf = self._build_vf(index_name, parents,
390
delta, for_write=True)
391
return source_vf, target_vf
393
def _copy_stream(self, source_vf, target_vf, keys, message, vf_to_stream,
395
trace.mutter('repacking %d %s', len(keys), message)
396
self.pb.update('repacking %s' % (message,), pb_offset)
397
child_pb = ui.ui_factory.nested_progress_bar()
399
stream = vf_to_stream(source_vf, keys, message, child_pb)
400
for _ in target_vf._insert_record_stream(stream,
407
def _copy_revision_texts(self):
408
source_vf, target_vf = self._build_vfs('revision', True, False)
409
if not self.revision_keys:
410
# We are doing a full fetch, aka 'pack'
411
self.revision_keys = source_vf.keys()
412
self._copy_stream(source_vf, target_vf, self.revision_keys,
413
'revisions', self._get_progress_stream, 1)
415
def _copy_inventory_texts(self):
416
source_vf, target_vf = self._build_vfs('inventory', True, True)
417
# It is not sufficient to just use self.revision_keys, as stacked
418
# repositories can have more inventories than they have revisions.
419
# One alternative would be to do something with
420
# get_parent_map(self.revision_keys), but that shouldn't be any faster
422
inventory_keys = source_vf.keys()
423
missing_inventories = set(self.revision_keys).difference(inventory_keys)
424
if missing_inventories:
425
missing_inventories = sorted(missing_inventories)
426
raise ValueError('We are missing inventories for revisions: %s'
427
% (missing_inventories,))
428
self._copy_stream(source_vf, target_vf, inventory_keys,
429
'inventories', self._get_filtered_inv_stream, 2)
431
def _copy_chk_texts(self):
432
source_vf, target_vf = self._build_vfs('chk', False, False)
433
# TODO: This is technically spurious... if it is a performance issue,
435
total_keys = source_vf.keys()
436
trace.mutter('repacking chk: %d id_to_entry roots,'
437
' %d p_id_map roots, %d total keys',
438
len(self._chk_id_roots), len(self._chk_p_id_roots),
440
self.pb.update('repacking chk', 3)
441
child_pb = ui.ui_factory.nested_progress_bar()
443
for stream in self._get_chk_streams(source_vf, total_keys,
445
for _ in target_vf._insert_record_stream(stream,
452
def _copy_text_texts(self):
453
source_vf, target_vf = self._build_vfs('text', True, True)
454
# XXX: We don't walk the chk map to determine referenced (file_id,
455
# revision_id) keys. We don't do it yet because you really need
456
# to filter out the ones that are present in the parents of the
457
# rev just before the ones you are copying, otherwise the filter
458
# is grabbing too many keys...
459
text_keys = source_vf.keys()
460
self._copy_stream(source_vf, target_vf, text_keys,
461
'texts', self._get_progress_stream, 4)
463
def _copy_signature_texts(self):
464
source_vf, target_vf = self._build_vfs('signature', False, False)
465
signature_keys = source_vf.keys()
466
signature_keys.intersection(self.revision_keys)
467
self._copy_stream(source_vf, target_vf, signature_keys,
468
'signatures', self._get_progress_stream, 5)
470
def _create_pack_from_packs(self):
471
self.pb.update('repacking', 0, 7)
472
self.new_pack = self.open_pack()
473
# Is this necessary for GC ?
474
self.new_pack.set_write_cache_size(1024*1024)
475
self._copy_revision_texts()
476
self._copy_inventory_texts()
477
self._copy_chk_texts()
478
self._copy_text_texts()
479
self._copy_signature_texts()
480
self.new_pack._check_references()
481
if not self._use_pack(self.new_pack):
482
self.new_pack.abort()
484
self.new_pack.finish_content()
485
if len(self.packs) == 1:
486
old_pack = self.packs[0]
487
if old_pack.name == self.new_pack._hash.hexdigest():
488
# The single old pack was already optimally packed.
489
trace.mutter('single pack %s was already optimally packed',
491
self.new_pack.abort()
493
self.pb.update('finishing repack', 6, 7)
494
self.new_pack.finish()
495
self._pack_collection.allocate(self.new_pack)
499
class GCCHKReconcilePacker(GCCHKPacker):
500
"""A packer which regenerates indices etc as it copies.
502
This is used by ``bzr reconcile`` to cause parent text pointers to be
506
def __init__(self, *args, **kwargs):
507
super(GCCHKReconcilePacker, self).__init__(*args, **kwargs)
508
self._data_changed = False
509
self._gather_text_refs = True
511
def _copy_inventory_texts(self):
512
source_vf, target_vf = self._build_vfs('inventory', True, True)
513
self._copy_stream(source_vf, target_vf, self.revision_keys,
514
'inventories', self._get_filtered_inv_stream, 2)
515
if source_vf.keys() != self.revision_keys:
516
self._data_changed = True
518
def _copy_text_texts(self):
519
"""generate what texts we should have and then copy."""
520
source_vf, target_vf = self._build_vfs('text', True, True)
521
trace.mutter('repacking %d texts', len(self._text_refs))
522
self.pb.update("repacking texts", 4)
523
# we have three major tasks here:
524
# 1) generate the ideal index
525
repo = self._pack_collection.repo
526
# We want the one we just wrote, so base it on self.new_pack
527
revision_vf = self._build_vf('revision', True, False, for_write=True)
528
ancestor_keys = revision_vf.get_parent_map(revision_vf.keys())
529
# Strip keys back into revision_ids.
530
ancestors = dict((k[0], tuple([p[0] for p in parents]))
531
for k, parents in ancestor_keys.iteritems())
533
# TODO: _generate_text_key_index should be much cheaper to generate from
534
# a chk repository, rather than the current implementation
535
ideal_index = repo._generate_text_key_index(None, ancestors)
536
file_id_parent_map = source_vf.get_parent_map(self._text_refs)
537
# 2) generate a keys list that contains all the entries that can
538
# be used as-is, with corrected parents.
540
new_parent_keys = {} # (key, parent_keys)
542
NULL_REVISION = _mod_revision.NULL_REVISION
543
for key in self._text_refs:
549
ideal_parents = tuple(ideal_index[key])
551
discarded_keys.append(key)
552
self._data_changed = True
554
if ideal_parents == (NULL_REVISION,):
556
source_parents = file_id_parent_map[key]
557
if ideal_parents == source_parents:
561
# We need to change the parent graph, but we don't need to
562
# re-insert the text (since we don't pun the compression
563
# parent with the parents list)
564
self._data_changed = True
565
new_parent_keys[key] = ideal_parents
566
# we're finished with some data.
568
del file_id_parent_map
569
# 3) bulk copy the data, updating records than need it
570
def _update_parents_for_texts():
571
stream = source_vf.get_record_stream(self._text_refs,
572
'groupcompress', False)
573
for record in stream:
574
if record.key in new_parent_keys:
575
record.parents = new_parent_keys[record.key]
577
target_vf.insert_record_stream(_update_parents_for_texts())
579
def _use_pack(self, new_pack):
580
"""Override _use_pack to check for reconcile having changed content."""
581
return new_pack.data_inserted() and self._data_changed
584
class GCRepositoryPackCollection(RepositoryPackCollection):
586
pack_factory = GCPack
587
resumed_pack_factory = ResumedGCPack
589
def _check_new_inventories(self):
590
"""Detect missing inventories or chk root entries for the new revisions
593
:returns: list of strs, summarising any problems found. If the list is
594
empty no problems were found.
596
# Ensure that all revisions added in this write group have:
597
# - corresponding inventories,
598
# - chk root entries for those inventories,
599
# - and any present parent inventories have their chk root
601
# And all this should be independent of any fallback repository.
603
key_deps = self.repo.revisions._index._key_dependencies
604
new_revisions_keys = key_deps.get_new_keys()
605
no_fallback_inv_index = self.repo.inventories._index
606
no_fallback_chk_bytes_index = self.repo.chk_bytes._index
607
no_fallback_texts_index = self.repo.texts._index
608
inv_parent_map = no_fallback_inv_index.get_parent_map(
610
# Are any inventories for corresponding to the new revisions missing?
611
corresponding_invs = set(inv_parent_map)
612
missing_corresponding = set(new_revisions_keys)
613
missing_corresponding.difference_update(corresponding_invs)
614
if missing_corresponding:
615
problems.append("inventories missing for revisions %s" %
616
(sorted(missing_corresponding),))
618
# Are any chk root entries missing for any inventories? This includes
619
# any present parent inventories, which may be used when calculating
620
# deltas for streaming.
621
all_inv_keys = set(corresponding_invs)
622
for parent_inv_keys in inv_parent_map.itervalues():
623
all_inv_keys.update(parent_inv_keys)
624
# Filter out ghost parents.
625
all_inv_keys.intersection_update(
626
no_fallback_inv_index.get_parent_map(all_inv_keys))
627
parent_invs_only_keys = all_inv_keys.symmetric_difference(
630
inv_ids = [key[-1] for key in all_inv_keys]
631
parent_invs_only_ids = [key[-1] for key in parent_invs_only_keys]
632
root_key_info = _build_interesting_key_sets(
633
self.repo, inv_ids, parent_invs_only_ids)
634
expected_chk_roots = root_key_info.all_keys()
635
present_chk_roots = no_fallback_chk_bytes_index.get_parent_map(
637
missing_chk_roots = expected_chk_roots.difference(present_chk_roots)
638
if missing_chk_roots:
639
problems.append("missing referenced chk root keys: %s"
640
% (sorted(missing_chk_roots),))
641
# Don't bother checking any further.
643
# Find all interesting chk_bytes records, and make sure they are
644
# present, as well as the text keys they reference.
645
chk_bytes_no_fallbacks = self.repo.chk_bytes.without_fallbacks()
646
chk_bytes_no_fallbacks._search_key_func = \
647
self.repo.chk_bytes._search_key_func
648
chk_diff = chk_map.iter_interesting_nodes(
649
chk_bytes_no_fallbacks, root_key_info.interesting_root_keys,
650
root_key_info.uninteresting_root_keys)
651
bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
654
for record in _filter_text_keys(chk_diff, text_keys, bytes_to_info):
656
except errors.NoSuchRevision, e:
657
# XXX: It would be nice if we could give a more precise error here.
658
problems.append("missing chk node(s) for id_to_entry maps")
659
chk_diff = chk_map.iter_interesting_nodes(
660
chk_bytes_no_fallbacks, root_key_info.interesting_pid_root_keys,
661
root_key_info.uninteresting_pid_root_keys)
663
for interesting_rec, interesting_map in chk_diff:
665
except errors.NoSuchRevision, e:
667
"missing chk node(s) for parent_id_basename_to_file_id maps")
668
present_text_keys = no_fallback_texts_index.get_parent_map(text_keys)
669
missing_text_keys = text_keys.difference(present_text_keys)
670
if missing_text_keys:
671
problems.append("missing text keys: %r"
672
% (sorted(missing_text_keys),))
675
def _execute_pack_operations(self, pack_operations,
676
_packer_class=GCCHKPacker,
678
"""Execute a series of pack operations.
680
:param pack_operations: A list of [revision_count, packs_to_combine].
681
:param _packer_class: The class of packer to use (default: Packer).
684
# XXX: Copied across from RepositoryPackCollection simply because we
685
# want to override the _packer_class ... :(
686
for revision_count, packs in pack_operations:
687
# we may have no-ops from the setup logic
690
packer = GCCHKPacker(self, packs, '.autopack',
691
reload_func=reload_func)
693
result = packer.pack()
694
except errors.RetryWithNewPacks:
695
# An exception is propagating out of this context, make sure
696
# this packer has cleaned up. Packer() doesn't set its new_pack
697
# state into the RepositoryPackCollection object, so we only
698
# have access to it directly here.
699
if packer.new_pack is not None:
700
packer.new_pack.abort()
705
self._remove_pack_from_memory(pack)
706
# record the newly available packs and stop advertising the old
708
result = self._save_pack_names(clear_obsolete_packs=True)
709
# Move the old packs out of the way now they are no longer referenced.
710
for revision_count, packs in pack_operations:
711
self._obsolete_packs(packs)
715
class CHKInventoryRepository(KnitPackRepository):
716
"""subclass of KnitPackRepository that uses CHK based inventories."""
718
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
720
"""Overridden to change pack collection class."""
721
KnitPackRepository.__init__(self, _format, a_bzrdir, control_files,
722
_commit_builder_class, _serializer)
723
# and now replace everything it did :)
724
index_transport = self._transport.clone('indices')
725
self._pack_collection = GCRepositoryPackCollection(self,
726
self._transport, index_transport,
727
self._transport.clone('upload'),
728
self._transport.clone('packs'),
729
_format.index_builder_class,
731
use_chk_index=self._format.supports_chks,
733
self.inventories = GroupCompressVersionedFiles(
734
_GCGraphIndex(self._pack_collection.inventory_index.combined_index,
735
add_callback=self._pack_collection.inventory_index.add_callback,
736
parents=True, is_locked=self.is_locked,
737
inconsistency_fatal=False),
738
access=self._pack_collection.inventory_index.data_access)
739
self.revisions = GroupCompressVersionedFiles(
740
_GCGraphIndex(self._pack_collection.revision_index.combined_index,
741
add_callback=self._pack_collection.revision_index.add_callback,
742
parents=True, is_locked=self.is_locked,
743
track_external_parent_refs=True, track_new_keys=True),
744
access=self._pack_collection.revision_index.data_access,
746
self.signatures = GroupCompressVersionedFiles(
747
_GCGraphIndex(self._pack_collection.signature_index.combined_index,
748
add_callback=self._pack_collection.signature_index.add_callback,
749
parents=False, is_locked=self.is_locked,
750
inconsistency_fatal=False),
751
access=self._pack_collection.signature_index.data_access,
753
self.texts = GroupCompressVersionedFiles(
754
_GCGraphIndex(self._pack_collection.text_index.combined_index,
755
add_callback=self._pack_collection.text_index.add_callback,
756
parents=True, is_locked=self.is_locked,
757
inconsistency_fatal=False),
758
access=self._pack_collection.text_index.data_access)
759
# No parents, individual CHK pages don't have specific ancestry
760
self.chk_bytes = GroupCompressVersionedFiles(
761
_GCGraphIndex(self._pack_collection.chk_index.combined_index,
762
add_callback=self._pack_collection.chk_index.add_callback,
763
parents=False, is_locked=self.is_locked,
764
inconsistency_fatal=False),
765
access=self._pack_collection.chk_index.data_access)
766
search_key_name = self._format._serializer.search_key_name
767
search_key_func = chk_map.search_key_registry.get(search_key_name)
768
self.chk_bytes._search_key_func = search_key_func
769
# True when the repository object is 'write locked' (as opposed to the
770
# physical lock only taken out around changes to the pack-names list.)
771
# Another way to represent this would be a decorator around the control
772
# files object that presents logical locks as physical ones - if this
773
# gets ugly consider that alternative design. RBC 20071011
774
self._write_lock_count = 0
775
self._transaction = None
777
self._reconcile_does_inventory_gc = True
778
self._reconcile_fixes_text_parents = True
779
self._reconcile_backsup_inventory = False
781
def _add_inventory_checked(self, revision_id, inv, parents):
782
"""Add inv to the repository after checking the inputs.
784
This function can be overridden to allow different inventory styles.
786
:seealso: add_inventory, for the contract.
789
serializer = self._format._serializer
790
result = inventory.CHKInventory.from_inventory(self.chk_bytes, inv,
791
maximum_size=serializer.maximum_size,
792
search_key_name=serializer.search_key_name)
793
inv_lines = result.to_lines()
794
return self._inventory_add_lines(revision_id, parents,
795
inv_lines, check_content=False)
797
def _create_inv_from_null(self, delta, revision_id):
798
"""This will mutate new_inv directly.
800
This is a simplified form of create_by_apply_delta which knows that all
801
the old values must be None, so everything is a create.
803
serializer = self._format._serializer
804
new_inv = inventory.CHKInventory(serializer.search_key_name)
805
new_inv.revision_id = revision_id
806
entry_to_bytes = new_inv._entry_to_bytes
807
id_to_entry_dict = {}
808
parent_id_basename_dict = {}
809
for old_path, new_path, file_id, entry in delta:
810
if old_path is not None:
811
raise ValueError('Invalid delta, somebody tried to delete %r'
812
' from the NULL_REVISION'
813
% ((old_path, file_id),))
815
raise ValueError('Invalid delta, delta from NULL_REVISION has'
816
' no new_path %r' % (file_id,))
818
new_inv.root_id = file_id
819
parent_id_basename_key = StaticTuple('', '').intern()
821
utf8_entry_name = entry.name.encode('utf-8')
822
parent_id_basename_key = StaticTuple(entry.parent_id,
823
utf8_entry_name).intern()
824
new_value = entry_to_bytes(entry)
826
# new_inv._path_to_fileid_cache[new_path] = file_id
827
key = StaticTuple(file_id).intern()
828
id_to_entry_dict[key] = new_value
829
parent_id_basename_dict[parent_id_basename_key] = file_id
831
new_inv._populate_from_dicts(self.chk_bytes, id_to_entry_dict,
832
parent_id_basename_dict, maximum_size=serializer.maximum_size)
835
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
836
parents, basis_inv=None, propagate_caches=False):
837
"""Add a new inventory expressed as a delta against another revision.
839
:param basis_revision_id: The inventory id the delta was created
841
:param delta: The inventory delta (see Inventory.apply_delta for
843
:param new_revision_id: The revision id that the inventory is being
845
:param parents: The revision ids of the parents that revision_id is
846
known to have and are in the repository already. These are supplied
847
for repositories that depend on the inventory graph for revision
848
graph access, as well as for those that pun ancestry with delta
850
:param basis_inv: The basis inventory if it is already known,
852
:param propagate_caches: If True, the caches for this inventory are
853
copied to and updated for the result if possible.
855
:returns: (validator, new_inv)
856
The validator(which is a sha1 digest, though what is sha'd is
857
repository format specific) of the serialized inventory, and the
860
if not self.is_in_write_group():
861
raise AssertionError("%r not in write group" % (self,))
862
_mod_revision.check_not_reserved_id(new_revision_id)
864
if basis_inv is None:
865
if basis_revision_id == _mod_revision.NULL_REVISION:
866
new_inv = self._create_inv_from_null(delta, new_revision_id)
867
inv_lines = new_inv.to_lines()
868
return self._inventory_add_lines(new_revision_id, parents,
869
inv_lines, check_content=False), new_inv
871
basis_tree = self.revision_tree(basis_revision_id)
872
basis_tree.lock_read()
873
basis_inv = basis_tree.inventory
875
result = basis_inv.create_by_apply_delta(delta, new_revision_id,
876
propagate_caches=propagate_caches)
877
inv_lines = result.to_lines()
878
return self._inventory_add_lines(new_revision_id, parents,
879
inv_lines, check_content=False), result
881
if basis_tree is not None:
884
def deserialise_inventory(self, revision_id, bytes):
885
return inventory.CHKInventory.deserialise(self.chk_bytes, bytes,
888
def _iter_inventories(self, revision_ids, ordering):
889
"""Iterate over many inventory objects."""
891
ordering = 'unordered'
892
keys = [(revision_id,) for revision_id in revision_ids]
893
stream = self.inventories.get_record_stream(keys, ordering, True)
895
for record in stream:
896
if record.storage_kind != 'absent':
897
texts[record.key] = record.get_bytes_as('fulltext')
899
raise errors.NoSuchRevision(self, record.key)
901
yield inventory.CHKInventory.deserialise(self.chk_bytes, texts[key], key)
903
def _iter_inventory_xmls(self, revision_ids, ordering):
904
# Without a native 'xml' inventory, this method doesn't make sense.
905
# However older working trees, and older bundles want it - so we supply
906
# it allowing get_inventory_xml to work. Bundles currently use the
907
# serializer directly; this also isn't ideal, but there isn't an xml
908
# iteration interface offered at all for repositories. We could make
909
# _iter_inventory_xmls be part of the contract, even if kept private.
910
inv_to_str = self._serializer.write_inventory_to_string
911
for inv in self.iter_inventories(revision_ids, ordering=ordering):
912
yield inv_to_str(inv), inv.revision_id
914
def _find_present_inventory_keys(self, revision_keys):
915
parent_map = self.inventories.get_parent_map(revision_keys)
916
present_inventory_keys = set(k for k in parent_map)
917
return present_inventory_keys
919
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
920
"""Find the file ids and versions affected by revisions.
922
:param revisions: an iterable containing revision ids.
923
:param _inv_weave: The inventory weave from this repository or None.
924
If None, the inventory weave will be opened automatically.
925
:return: a dictionary mapping altered file-ids to an iterable of
926
revision_ids. Each altered file-ids has the exact revision_ids that
927
altered it listed explicitly.
929
rich_root = self.supports_rich_root()
930
bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
931
file_id_revisions = {}
932
pb = ui.ui_factory.nested_progress_bar()
934
revision_keys = [(r,) for r in revision_ids]
935
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
936
# TODO: instead of using _find_present_inventory_keys, change the
937
# code paths to allow missing inventories to be tolerated.
938
# However, we only want to tolerate missing parent
939
# inventories, not missing inventories for revision_ids
940
present_parent_inv_keys = self._find_present_inventory_keys(
942
present_parent_inv_ids = set(
943
[k[-1] for k in present_parent_inv_keys])
944
inventories_to_read = set(revision_ids)
945
inventories_to_read.update(present_parent_inv_ids)
946
root_key_info = _build_interesting_key_sets(
947
self, inventories_to_read, present_parent_inv_ids)
948
interesting_root_keys = root_key_info.interesting_root_keys
949
uninteresting_root_keys = root_key_info.uninteresting_root_keys
950
chk_bytes = self.chk_bytes
951
for record, items in chk_map.iter_interesting_nodes(chk_bytes,
952
interesting_root_keys, uninteresting_root_keys,
954
for name, bytes in items:
955
(name_utf8, file_id, revision_id) = bytes_to_info(bytes)
956
# TODO: consider interning file_id, revision_id here, or
957
# pushing that intern() into bytes_to_info()
958
# TODO: rich_root should always be True here, for all
959
# repositories that support chk_bytes
960
if not rich_root and name_utf8 == '':
963
file_id_revisions[file_id].add(revision_id)
965
file_id_revisions[file_id] = set([revision_id])
968
return file_id_revisions
970
def find_text_key_references(self):
971
"""Find the text key references within the repository.
973
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
974
to whether they were referred to by the inventory of the
975
revision_id that they contain. The inventory texts from all present
976
revision ids are assessed to generate this report.
978
# XXX: Slow version but correct: rewrite as a series of delta
979
# examinations/direct tree traversal. Note that that will require care
980
# as a common node is reachable both from the inventory that added it,
981
# and others afterwards.
982
revision_keys = self.revisions.keys()
984
rich_roots = self.supports_rich_root()
985
pb = ui.ui_factory.nested_progress_bar()
987
all_revs = self.all_revision_ids()
988
total = len(all_revs)
989
for pos, inv in enumerate(self.iter_inventories(all_revs)):
990
pb.update("Finding text references", pos, total)
991
for _, entry in inv.iter_entries():
992
if not rich_roots and entry.file_id == inv.root_id:
994
key = (entry.file_id, entry.revision)
995
result.setdefault(key, False)
996
if entry.revision == inv.revision_id:
1002
def _reconcile_pack(self, collection, packs, extension, revs, pb):
1003
packer = GCCHKReconcilePacker(collection, packs, extension)
1004
return packer.pack(pb)
1006
def _get_source(self, to_format):
1007
"""Return a source for streaming from this repository."""
1008
if self._format._serializer == to_format._serializer:
1009
# We must be exactly the same format, otherwise stuff like the chk
1010
# page layout might be different.
1011
# Actually, this test is just slightly looser than exact so that
1012
# CHK2 <-> 2a transfers will work.
1013
return GroupCHKStreamSource(self, to_format)
1014
return super(CHKInventoryRepository, self)._get_source(to_format)
1017
class GroupCHKStreamSource(KnitPackStreamSource):
1018
"""Used when both the source and target repo are GroupCHK repos."""
1020
def __init__(self, from_repository, to_format):
1021
"""Create a StreamSource streaming from from_repository."""
1022
super(GroupCHKStreamSource, self).__init__(from_repository, to_format)
1023
self._revision_keys = None
1024
self._text_keys = None
1025
self._text_fetch_order = 'groupcompress'
1026
self._chk_id_roots = None
1027
self._chk_p_id_roots = None
1029
def _get_inventory_stream(self, inventory_keys, allow_absent=False):
1030
"""Get a stream of inventory texts.
1032
When this function returns, self._chk_id_roots and self._chk_p_id_roots
1033
should be populated.
1035
self._chk_id_roots = []
1036
self._chk_p_id_roots = []
1037
def _filtered_inv_stream():
1038
id_roots_set = set()
1039
p_id_roots_set = set()
1040
source_vf = self.from_repository.inventories
1041
stream = source_vf.get_record_stream(inventory_keys,
1042
'groupcompress', True)
1043
for record in stream:
1044
if record.storage_kind == 'absent':
1048
raise errors.NoSuchRevision(self, record.key)
1049
bytes = record.get_bytes_as('fulltext')
1050
chk_inv = inventory.CHKInventory.deserialise(None, bytes,
1052
key = chk_inv.id_to_entry.key()
1053
if key not in id_roots_set:
1054
self._chk_id_roots.append(key)
1055
id_roots_set.add(key)
1056
p_id_map = chk_inv.parent_id_basename_to_file_id
1057
if p_id_map is None:
1058
raise AssertionError('Parent id -> file_id map not set')
1059
key = p_id_map.key()
1060
if key not in p_id_roots_set:
1061
p_id_roots_set.add(key)
1062
self._chk_p_id_roots.append(key)
1064
# We have finished processing all of the inventory records, we
1065
# don't need these sets anymore
1066
id_roots_set.clear()
1067
p_id_roots_set.clear()
1068
return ('inventories', _filtered_inv_stream())
1070
def _get_filtered_chk_streams(self, excluded_revision_keys):
1071
self._text_keys = set()
1072
excluded_revision_keys.discard(_mod_revision.NULL_REVISION)
1073
if not excluded_revision_keys:
1074
uninteresting_root_keys = set()
1075
uninteresting_pid_root_keys = set()
1077
# filter out any excluded revisions whose inventories are not
1079
# TODO: Update Repository.iter_inventories() to add
1080
# ignore_missing=True
1081
present_keys = self.from_repository._find_present_inventory_keys(
1082
excluded_revision_keys)
1083
present_ids = [k[-1] for k in present_keys]
1084
uninteresting_root_keys = set()
1085
uninteresting_pid_root_keys = set()
1086
for inv in self.from_repository.iter_inventories(present_ids):
1087
uninteresting_root_keys.add(inv.id_to_entry.key())
1088
uninteresting_pid_root_keys.add(
1089
inv.parent_id_basename_to_file_id.key())
1090
bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
1091
chk_bytes = self.from_repository.chk_bytes
1092
def _filter_id_to_entry():
1093
interesting_nodes = chk_map.iter_interesting_nodes(chk_bytes,
1094
self._chk_id_roots, uninteresting_root_keys)
1095
for record in _filter_text_keys(interesting_nodes, self._text_keys,
1097
if record is not None:
1100
self._chk_id_roots = None
1101
yield 'chk_bytes', _filter_id_to_entry()
1102
def _get_parent_id_basename_to_file_id_pages():
1103
for record, items in chk_map.iter_interesting_nodes(chk_bytes,
1104
self._chk_p_id_roots, uninteresting_pid_root_keys):
1105
if record is not None:
1108
self._chk_p_id_roots = None
1109
yield 'chk_bytes', _get_parent_id_basename_to_file_id_pages()
1111
def get_stream(self, search):
1112
revision_ids = search.get_keys()
1113
for stream_info in self._fetch_revision_texts(revision_ids):
1115
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
1116
self.from_repository.revisions.clear_cache()
1117
self.from_repository.signatures.clear_cache()
1118
yield self._get_inventory_stream(self._revision_keys)
1119
self.from_repository.inventories.clear_cache()
1120
# TODO: The keys to exclude might be part of the search recipe
1121
# For now, exclude all parents that are at the edge of ancestry, for
1122
# which we have inventories
1123
from_repo = self.from_repository
1124
parent_keys = from_repo._find_parent_keys_of_revisions(
1125
self._revision_keys)
1126
for stream_info in self._get_filtered_chk_streams(parent_keys):
1128
self.from_repository.chk_bytes.clear_cache()
1129
yield self._get_text_stream()
1130
self.from_repository.texts.clear_cache()
1132
def get_stream_for_missing_keys(self, missing_keys):
1133
# missing keys can only occur when we are byte copying and not
1134
# translating (because translation means we don't send
1135
# unreconstructable deltas ever).
1136
missing_inventory_keys = set()
1137
for key in missing_keys:
1138
if key[0] != 'inventories':
1139
raise AssertionError('The only missing keys we should'
1140
' be filling in are inventory keys, not %s'
1142
missing_inventory_keys.add(key[1:])
1143
if self._chk_id_roots or self._chk_p_id_roots:
1144
raise AssertionError('Cannot call get_stream_for_missing_keys'
1145
' until all of get_stream() has been consumed.')
1146
# Yield the inventory stream, so we can find the chk stream
1147
# Some of the missing_keys will be missing because they are ghosts.
1148
# As such, we can ignore them. The Sink is required to verify there are
1149
# no unavailable texts when the ghost inventories are not filled in.
1150
yield self._get_inventory_stream(missing_inventory_keys,
1152
# We use the empty set for excluded_revision_keys, to make it clear
1153
# that we want to transmit all referenced chk pages.
1154
for stream_info in self._get_filtered_chk_streams(set()):
1158
class _InterestingKeyInfo(object):
1160
self.interesting_root_keys = set()
1161
self.interesting_pid_root_keys = set()
1162
self.uninteresting_root_keys = set()
1163
self.uninteresting_pid_root_keys = set()
1165
def all_interesting(self):
1166
return self.interesting_root_keys.union(self.interesting_pid_root_keys)
1168
def all_uninteresting(self):
1169
return self.uninteresting_root_keys.union(
1170
self.uninteresting_pid_root_keys)
1173
return self.all_interesting().union(self.all_uninteresting())
1176
def _build_interesting_key_sets(repo, inventory_ids, parent_only_inv_ids):
1177
result = _InterestingKeyInfo()
1178
for inv in repo.iter_inventories(inventory_ids, 'unordered'):
1179
root_key = inv.id_to_entry.key()
1180
pid_root_key = inv.parent_id_basename_to_file_id.key()
1181
if inv.revision_id in parent_only_inv_ids:
1182
result.uninteresting_root_keys.add(root_key)
1183
result.uninteresting_pid_root_keys.add(pid_root_key)
1185
result.interesting_root_keys.add(root_key)
1186
result.interesting_pid_root_keys.add(pid_root_key)
1190
def _filter_text_keys(interesting_nodes_iterable, text_keys, bytes_to_info):
1191
"""Iterate the result of iter_interesting_nodes, yielding the records
1192
and adding to text_keys.
1194
for record, items in interesting_nodes_iterable:
1195
for name, bytes in items:
1196
# Note: we don't care about name_utf8, because groupcompress repos
1197
# are always rich-root, so there are no synthesised root records to
1199
_, file_id, revision_id = bytes_to_info(bytes)
1200
file_id = intern(file_id)
1201
revision_id = intern(revision_id)
1202
text_keys.add(StaticTuple(file_id, revision_id).intern())
1208
class RepositoryFormatCHK1(RepositoryFormatPack):
1209
"""A hashed CHK+group compress pack repository."""
1211
repository_class = CHKInventoryRepository
1212
supports_external_lookups = True
1213
supports_chks = True
1214
# For right now, setting this to True gives us InterModel1And2 rather
1215
# than InterDifferingSerializer
1216
_commit_builder_class = PackRootCommitBuilder
1217
rich_root_data = True
1218
_serializer = chk_serializer.chk_serializer_255_bigpage
1219
_commit_inv_deltas = True
1220
# What index classes to use
1221
index_builder_class = BTreeBuilder
1222
index_class = BTreeGraphIndex
1223
# Note: We cannot unpack a delta that references a text we haven't
1224
# seen yet. There are 2 options, work in fulltexts, or require
1225
# topological sorting. Using fulltexts is more optimal for local
1226
# operations, because the source can be smart about extracting
1227
# multiple in-a-row (and sharing strings). Topological is better
1228
# for remote, because we access less data.
1229
_fetch_order = 'unordered'
1230
_fetch_uses_deltas = False # essentially ignored by the groupcompress code.
1232
pack_compresses = True
1234
def _get_matching_bzrdir(self):
1235
return bzrdir.format_registry.make_bzrdir('development6-rich-root')
1237
def _ignore_setting_bzrdir(self, format):
1240
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1242
def get_format_string(self):
1243
"""See RepositoryFormat.get_format_string()."""
1244
return ('Bazaar development format - group compression and chk inventory'
1245
' (needs bzr.dev from 1.14)\n')
1247
def get_format_description(self):
1248
"""See RepositoryFormat.get_format_description()."""
1249
return ("Development repository format - rich roots, group compression"
1250
" and chk inventories")
1253
class RepositoryFormatCHK2(RepositoryFormatCHK1):
1254
"""A CHK repository that uses the bencode revision serializer."""
1256
_serializer = chk_serializer.chk_bencode_serializer
1258
def _get_matching_bzrdir(self):
1259
return bzrdir.format_registry.make_bzrdir('development7-rich-root')
1261
def _ignore_setting_bzrdir(self, format):
1264
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1266
def get_format_string(self):
1267
"""See RepositoryFormat.get_format_string()."""
1268
return ('Bazaar development format - chk repository with bencode '
1269
'revision serialization (needs bzr.dev from 1.16)\n')
1272
class RepositoryFormat2a(RepositoryFormatCHK2):
1273
"""A CHK repository that uses the bencode revision serializer.
1275
This is the same as RepositoryFormatCHK2 but with a public name.
1278
_serializer = chk_serializer.chk_bencode_serializer
1280
def _get_matching_bzrdir(self):
1281
return bzrdir.format_registry.make_bzrdir('2a')
1283
def _ignore_setting_bzrdir(self, format):
1286
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1288
def get_format_string(self):
1289
return ('Bazaar repository format 2a (needs bzr 1.16 or later)\n')
1291
def get_format_description(self):
1292
"""See RepositoryFormat.get_format_description()."""
1293
return ("Repository format 2a - rich roots, group compression"
1294
" and chk inventories")
added
removed
bzrlib/repofmt/groupcompress_repo.py
