/brz/remove-bazaar

To get this branch, use:
bzr branch http://gegoxaren.bato24.eu/bzr/brz/remove-bazaar

« back to all changes in this revision

Viewing changes to breezy/bzr/groupcompress_repo.py

  • Committer: Jelmer Vernooij
  • Date: 2019-12-23 01:39:21 UTC
  • mfrom: (7424 work)
  • mto: This revision was merged to the branch mainline in revision 7425.
  • Revision ID: jelmer@jelmer.uk-20191223013921-2kzd0wlcoylgxksk
Merge trunk.

Show diffs side-by-side

added added

removed removed

Lines of Context:
breezy/bzr/groupcompress_repo.py
1
 
# Copyright (C) 2008, 2009, 2010 Canonical Ltd
 
1
# Copyright (C) 2008-2011 Canonical Ltd
2
2
#
3
3
# This program is free software; you can redistribute it and/or modify
4
4
# it under the terms of the GNU General Public License as published by
16
16
 
17
17
"""Repository formats using CHK inventories and groupcompress compression."""
18
18
 
 
19
from __future__ import absolute_import
 
20
 
19
21
import time
20
22
 
21
 
from bzrlib import (
22
 
    bzrdir,
23
 
    chk_map,
24
 
    chk_serializer,
 
23
from .. import (
 
24
    controldir,
25
25
    debug,
26
26
    errors,
27
 
    index as _mod_index,
28
 
    inventory,
29
 
    knit,
30
27
    osutils,
31
 
    pack,
32
28
    revision as _mod_revision,
33
29
    trace,
34
30
    ui,
35
31
    )
36
 
from bzrlib.btree_index import (
 
32
from ..bzr import (
 
33
    chk_map,
 
34
    chk_serializer,
 
35
    index as _mod_index,
 
36
    inventory,
 
37
    pack,
 
38
    versionedfile,
 
39
    )
 
40
from ..bzr.btree_index import (
37
41
    BTreeGraphIndex,
38
42
    BTreeBuilder,
39
43
    )
40
 
from bzrlib.groupcompress import (
 
44
from ..bzr.groupcompress import (
41
45
    _GCGraphIndex,
42
46
    GroupCompressVersionedFiles,
43
47
    )
44
 
from bzrlib.repofmt.pack_repo import (
 
48
from .pack_repo import (
 
49
    _DirectPackAccess,
45
50
    Pack,
46
51
    NewPack,
47
 
    KnitPackRepository,
48
 
    KnitPackStreamSource,
49
 
    PackRootCommitBuilder,
 
52
    PackRepository,
 
53
    PackCommitBuilder,
50
54
    RepositoryPackCollection,
51
55
    RepositoryFormatPack,
52
56
    ResumedPack,
53
57
    Packer,
54
58
    )
55
 
from bzrlib.static_tuple import StaticTuple
 
59
from ..bzr.vf_repository import (
 
60
    StreamSource,
 
61
    )
 
62
from ..sixish import (
 
63
    viewitems,
 
64
    viewvalues,
 
65
    )
 
66
from ..static_tuple import StaticTuple
56
67
 
57
68
 
58
69
class GCPack(NewPack):
80
91
        else:
81
92
            chk_index = None
82
93
        Pack.__init__(self,
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
88
 
            # source.
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
94
 
            # listing.
95
 
            index_builder_class(reference_lists=0),
96
 
            # CHK based storage - just blobs, no compression or parents.
97
 
            chk_index=chk_index
98
 
            )
 
94
                      # Revisions: parents list, no text compression.
 
95
                      index_builder_class(reference_lists=1),
 
96
                      # Inventory: We want to map compression only, but currently the
 
97
                      # knit code hasn't been updated enough to understand that, so we
 
98
                      # have a regular 2-list index giving parents and compression
 
99
                      # source.
 
100
                      index_builder_class(reference_lists=1),
 
101
                      # Texts: per file graph, for all fileids - so one reference list
 
102
                      # and two elements in the key tuple.
 
103
                      index_builder_class(reference_lists=1, key_elements=2),
 
104
                      # Signatures: Just blobs to store, no compression, no parents
 
105
                      # listing.
 
106
                      index_builder_class(reference_lists=0),
 
107
                      # CHK based storage - just blobs, no compression or parents.
 
108
                      chk_index=chk_index
 
109
                      )
99
110
        self._pack_collection = pack_collection
100
111
        # When we make readonly indices, we need this.
101
112
        self.index_class = pack_collection._index_class
126
137
            self.random_name, mode=self._file_mode)
127
138
        if 'pack' in debug.debug_flags:
128
139
            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)
 
140
                         time.ctime(), self.upload_transport.base, self.random_name,
 
141
                         time.time() - self.start_time)
131
142
        # A list of byte sequences to be written to the new pack, and the
132
143
        # aggregate size of them.  Stored as a list rather than separate
133
144
        # variables so that the _write_data closure below can update them.
137
148
        # robertc says- this is a closure rather than a method on the object
138
149
        # so that the variables are locals, and faster than accessing object
139
150
        # members.
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)
 
151
 
 
152
        def _write_data(data, flush=False, _buffer=self._buffer,
 
153
                        _write=self.write_stream.write, _update=self._hash.update):
 
154
            _buffer[0].append(data)
 
155
            _buffer[1] += len(data)
144
156
            # buffer cap
145
157
            if _buffer[1] > self._cache_limit or flush:
146
 
                bytes = ''.join(_buffer[0])
147
 
                _write(bytes)
148
 
                _update(bytes)
 
158
                data = b''.join(_buffer[0])
 
159
                _write(data)
 
160
                _update(data)
149
161
                _buffer[:] = [[], 0]
150
162
        # expose this on self, for the occasion when clients want to add data.
151
163
        self._write_data = _write_data
193
205
        self._pack_collection = pack_collection
194
206
        # ATM, We only support this for GCCHK repositories
195
207
        if pack_collection.chk_index is None:
196
 
            raise AssertionError('pack_collection.chk_index should not be None')
 
208
            raise AssertionError(
 
209
                'pack_collection.chk_index should not be None')
197
210
        self._gather_text_refs = False
198
211
        self._chk_id_roots = []
199
212
        self._chk_p_id_roots = []
203
216
 
204
217
    def _get_progress_stream(self, source_vf, keys, message, pb):
205
218
        def pb_stream():
206
 
            substream = source_vf.get_record_stream(keys, 'groupcompress', True)
 
219
            substream = source_vf.get_record_stream(
 
220
                keys, 'groupcompress', True)
207
221
            for idx, record in enumerate(substream):
208
222
                if pb is not None:
209
223
                    pb.update(message, idx + 1, len(keys))
213
227
    def _get_filtered_inv_stream(self, source_vf, keys, message, pb=None):
214
228
        """Filter the texts of inventories, to find the chk pages."""
215
229
        total_keys = len(keys)
 
230
 
216
231
        def _filtered_inv_stream():
217
232
            id_roots_set = set()
218
233
            p_id_roots_set = set()
262
277
        remaining_keys = set(keys)
263
278
        counter = [0]
264
279
        if self._gather_text_refs:
265
 
            bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
266
280
            self._text_refs = set()
 
281
 
267
282
        def _get_referenced_stream(root_keys, parse_leaf_nodes=False):
268
283
            cur_keys = root_keys
269
284
            while cur_keys:
270
285
                keys_by_search_prefix = {}
271
286
                remaining_keys.difference_update(cur_keys)
272
287
                next_keys = set()
 
288
 
273
289
                def handle_internal_node(node):
274
 
                    for prefix, value in node._items.iteritems():
 
290
                    for prefix, value in viewitems(node._items):
275
291
                        # We don't want to request the same key twice, and we
276
292
                        # want to order it by the first time it is seen.
277
293
                        # Even further, we don't want to request a key which is
283
299
                        #       always fill them in for stacked branches
284
300
                        if value not in next_keys and value in remaining_keys:
285
301
                            keys_by_search_prefix.setdefault(prefix,
286
 
                                []).append(value)
 
302
                                                             []).append(value)
287
303
                            next_keys.add(value)
 
304
 
288
305
                def handle_leaf_node(node):
289
306
                    # Store is None, because we know we have a LeafNode, and we
290
307
                    # just want its entries
291
308
                    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))
 
309
                        self._text_refs.add(chk_map._bytes_to_text_key(bytes))
 
310
 
294
311
                def next_stream():
295
312
                    stream = source_vf.get_record_stream(cur_keys,
296
313
                                                         'as-requested', True)
352
369
        """Build a VersionedFiles instance on top of this group of packs."""
353
370
        index_name = index_name + '_index'
354
371
        index_to_pack = {}
355
 
        access = knit._DirectPackAccess(index_to_pack,
356
 
                                        reload_func=self._reload_func)
 
372
        access = _DirectPackAccess(index_to_pack,
 
373
                                   reload_func=self._reload_func)
357
374
        if for_write:
358
375
            # Use new_pack
359
376
            if self.new_pack is None:
393
410
                     pb_offset):
394
411
        trace.mutter('repacking %d %s', len(keys), message)
395
412
        self.pb.update('repacking %s' % (message,), pb_offset)
396
 
        child_pb = ui.ui_factory.nested_progress_bar()
397
 
        try:
 
413
        with ui.ui_factory.nested_progress_bar() as child_pb:
398
414
            stream = vf_to_stream(source_vf, keys, message, child_pb)
399
415
            for _ in target_vf._insert_record_stream(stream,
400
416
                                                     random_id=True,
401
417
                                                     reuse_blocks=False):
402
418
                pass
403
 
        finally:
404
 
            child_pb.finished()
405
419
 
406
420
    def _copy_revision_texts(self):
407
421
        source_vf, target_vf = self._build_vfs('revision', True, False)
419
433
        # get_parent_map(self.revision_keys), but that shouldn't be any faster
420
434
        # than this.
421
435
        inventory_keys = source_vf.keys()
422
 
        missing_inventories = set(self.revision_keys).difference(inventory_keys)
 
436
        missing_inventories = set(
 
437
            self.revision_keys).difference(inventory_keys)
423
438
        if missing_inventories:
424
 
            missing_inventories = sorted(missing_inventories)
425
 
            raise ValueError('We are missing inventories for revisions: %s'
426
 
                % (missing_inventories,))
 
439
            # Go back to the original repo, to see if these are really missing
 
440
            # https://bugs.launchpad.net/bzr/+bug/437003
 
441
            # If we are packing a subset of the repo, it is fine to just have
 
442
            # the data in another Pack file, which is not included in this pack
 
443
            # operation.
 
444
            inv_index = self._pack_collection.repo.inventories._index
 
445
            pmap = inv_index.get_parent_map(missing_inventories)
 
446
            really_missing = missing_inventories.difference(pmap)
 
447
            if really_missing:
 
448
                missing_inventories = sorted(really_missing)
 
449
                raise ValueError('We are missing inventories for revisions: %s'
 
450
                                 % (missing_inventories,))
427
451
        self._copy_stream(source_vf, target_vf, inventory_keys,
428
452
                          'inventories', self._get_filtered_inv_stream, 2)
429
453
 
 
454
    def _get_chk_vfs_for_copy(self):
 
455
        return self._build_vfs('chk', False, False)
 
456
 
430
457
    def _copy_chk_texts(self):
431
 
        source_vf, target_vf = self._build_vfs('chk', False, False)
 
458
        source_vf, target_vf = self._get_chk_vfs_for_copy()
432
459
        # TODO: This is technically spurious... if it is a performance issue,
433
460
        #       remove it
434
461
        total_keys = source_vf.keys()
437
464
                     len(self._chk_id_roots), len(self._chk_p_id_roots),
438
465
                     len(total_keys))
439
466
        self.pb.update('repacking chk', 3)
440
 
        child_pb = ui.ui_factory.nested_progress_bar()
441
 
        try:
 
467
        with ui.ui_factory.nested_progress_bar() as child_pb:
442
468
            for stream in self._get_chk_streams(source_vf, total_keys,
443
469
                                                pb=child_pb):
444
470
                for _ in target_vf._insert_record_stream(stream,
445
471
                                                         random_id=True,
446
472
                                                         reuse_blocks=False):
447
473
                    pass
448
 
        finally:
449
 
            child_pb.finished()
450
474
 
451
475
    def _copy_text_texts(self):
452
476
        source_vf, target_vf = self._build_vfs('text', True, True)
470
494
        self.pb.update('repacking', 0, 7)
471
495
        self.new_pack = self.open_pack()
472
496
        # Is this necessary for GC ?
473
 
        self.new_pack.set_write_cache_size(1024*1024)
 
497
        self.new_pack.set_write_cache_size(1024 * 1024)
474
498
        self._copy_revision_texts()
475
499
        self._copy_inventory_texts()
476
500
        self._copy_chk_texts()
486
510
            if old_pack.name == self.new_pack._hash.hexdigest():
487
511
                # The single old pack was already optimally packed.
488
512
                trace.mutter('single pack %s was already optimally packed',
489
 
                    old_pack.name)
 
513
                             old_pack.name)
490
514
                self.new_pack.abort()
491
515
                return None
492
516
        self.pb.update('finishing repack', 6, 7)
498
522
class GCCHKReconcilePacker(GCCHKPacker):
499
523
    """A packer which regenerates indices etc as it copies.
500
524
 
501
 
    This is used by ``bzr reconcile`` to cause parent text pointers to be
 
525
    This is used by ``brz reconcile`` to cause parent text pointers to be
502
526
    regenerated.
503
527
    """
504
528
 
527
551
        ancestor_keys = revision_vf.get_parent_map(revision_vf.keys())
528
552
        # Strip keys back into revision_ids.
529
553
        ancestors = dict((k[0], tuple([p[0] for p in parents]))
530
 
                         for k, parents in ancestor_keys.iteritems())
 
554
                         for k, parents in viewitems(ancestor_keys))
531
555
        del ancestor_keys
532
556
        # TODO: _generate_text_key_index should be much cheaper to generate from
533
557
        #       a chk repository, rather than the current implementation
536
560
        # 2) generate a keys list that contains all the entries that can
537
561
        #    be used as-is, with corrected parents.
538
562
        ok_keys = []
539
 
        new_parent_keys = {} # (key, parent_keys)
 
563
        new_parent_keys = {}  # (key, parent_keys)
540
564
        discarded_keys = []
541
565
        NULL_REVISION = _mod_revision.NULL_REVISION
542
566
        for key in self._text_refs:
566
590
        del ideal_index
567
591
        del file_id_parent_map
568
592
        # 3) bulk copy the data, updating records than need it
 
593
 
569
594
        def _update_parents_for_texts():
570
595
            stream = source_vf.get_record_stream(self._text_refs,
571
 
                'groupcompress', False)
 
596
                                                 'groupcompress', False)
572
597
            for record in stream:
573
598
                if record.key in new_parent_keys:
574
599
                    record.parents = new_parent_keys[record.key]
580
605
        return new_pack.data_inserted() and self._data_changed
581
606
 
582
607
 
 
608
class GCCHKCanonicalizingPacker(GCCHKPacker):
 
609
    """A packer that ensures inventories have canonical-form CHK maps.
 
610
 
 
611
    Ideally this would be part of reconcile, but it's very slow and rarely
 
612
    needed.  (It repairs repositories affected by
 
613
    https://bugs.launchpad.net/bzr/+bug/522637).
 
614
    """
 
615
 
 
616
    def __init__(self, *args, **kwargs):
 
617
        super(GCCHKCanonicalizingPacker, self).__init__(*args, **kwargs)
 
618
        self._data_changed = False
 
619
 
 
620
    def _exhaust_stream(self, source_vf, keys, message, vf_to_stream, pb_offset):
 
621
        """Create and exhaust a stream, but don't insert it.
 
622
 
 
623
        This is useful to get the side-effects of generating a stream.
 
624
        """
 
625
        self.pb.update('scanning %s' % (message,), pb_offset)
 
626
        with ui.ui_factory.nested_progress_bar() as child_pb:
 
627
            list(vf_to_stream(source_vf, keys, message, child_pb))
 
628
 
 
629
    def _copy_inventory_texts(self):
 
630
        source_vf, target_vf = self._build_vfs('inventory', True, True)
 
631
        source_chk_vf, target_chk_vf = self._get_chk_vfs_for_copy()
 
632
        inventory_keys = source_vf.keys()
 
633
        # First, copy the existing CHKs on the assumption that most of them
 
634
        # will be correct.  This will save us from having to reinsert (and
 
635
        # recompress) these records later at the cost of perhaps preserving a
 
636
        # few unused CHKs.
 
637
        # (Iterate but don't insert _get_filtered_inv_stream to populate the
 
638
        # variables needed by GCCHKPacker._copy_chk_texts.)
 
639
        self._exhaust_stream(source_vf, inventory_keys, 'inventories',
 
640
                             self._get_filtered_inv_stream, 2)
 
641
        GCCHKPacker._copy_chk_texts(self)
 
642
        # Now copy and fix the inventories, and any regenerated CHKs.
 
643
 
 
644
        def chk_canonicalizing_inv_stream(source_vf, keys, message, pb=None):
 
645
            return self._get_filtered_canonicalizing_inv_stream(
 
646
                source_vf, keys, message, pb, source_chk_vf, target_chk_vf)
 
647
        self._copy_stream(source_vf, target_vf, inventory_keys,
 
648
                          'inventories', chk_canonicalizing_inv_stream, 4)
 
649
 
 
650
    def _copy_chk_texts(self):
 
651
        # No-op; in this class this happens during _copy_inventory_texts.
 
652
        pass
 
653
 
 
654
    def _get_filtered_canonicalizing_inv_stream(self, source_vf, keys, message,
 
655
                                                pb=None, source_chk_vf=None, target_chk_vf=None):
 
656
        """Filter the texts of inventories, regenerating CHKs to make sure they
 
657
        are canonical.
 
658
        """
 
659
        total_keys = len(keys)
 
660
        target_chk_vf = versionedfile.NoDupeAddLinesDecorator(target_chk_vf)
 
661
 
 
662
        def _filtered_inv_stream():
 
663
            stream = source_vf.get_record_stream(keys, 'groupcompress', True)
 
664
            search_key_name = None
 
665
            for idx, record in enumerate(stream):
 
666
                # Inventories should always be with revisions; assume success.
 
667
                bytes = record.get_bytes_as('fulltext')
 
668
                chk_inv = inventory.CHKInventory.deserialise(
 
669
                    source_chk_vf, bytes, record.key)
 
670
                if pb is not None:
 
671
                    pb.update('inv', idx, total_keys)
 
672
                chk_inv.id_to_entry._ensure_root()
 
673
                if search_key_name is None:
 
674
                    # Find the name corresponding to the search_key_func
 
675
                    search_key_reg = chk_map.search_key_registry
 
676
                    for search_key_name, func in viewitems(search_key_reg):
 
677
                        if func == chk_inv.id_to_entry._search_key_func:
 
678
                            break
 
679
                canonical_inv = inventory.CHKInventory.from_inventory(
 
680
                    target_chk_vf, chk_inv,
 
681
                    maximum_size=chk_inv.id_to_entry._root_node._maximum_size,
 
682
                    search_key_name=search_key_name)
 
683
                if chk_inv.id_to_entry.key() != canonical_inv.id_to_entry.key():
 
684
                    trace.mutter(
 
685
                        'Non-canonical CHK map for id_to_entry of inv: %s '
 
686
                        '(root is %s, should be %s)' % (chk_inv.revision_id,
 
687
                                                        chk_inv.id_to_entry.key()[
 
688
                                                            0],
 
689
                                                        canonical_inv.id_to_entry.key()[0]))
 
690
                    self._data_changed = True
 
691
                p_id_map = chk_inv.parent_id_basename_to_file_id
 
692
                p_id_map._ensure_root()
 
693
                canon_p_id_map = canonical_inv.parent_id_basename_to_file_id
 
694
                if p_id_map.key() != canon_p_id_map.key():
 
695
                    trace.mutter(
 
696
                        'Non-canonical CHK map for parent_id_to_basename of '
 
697
                        'inv: %s (root is %s, should be %s)'
 
698
                        % (chk_inv.revision_id, p_id_map.key()[0],
 
699
                           canon_p_id_map.key()[0]))
 
700
                    self._data_changed = True
 
701
                yield versionedfile.ChunkedContentFactory(record.key,
 
702
                                                          record.parents, record.sha1,
 
703
                                                          canonical_inv.to_lines())
 
704
            # We have finished processing all of the inventory records, we
 
705
            # don't need these sets anymore
 
706
        return _filtered_inv_stream()
 
707
 
 
708
    def _use_pack(self, new_pack):
 
709
        """Override _use_pack to check for reconcile having changed content."""
 
710
        return new_pack.data_inserted() and self._data_changed
 
711
 
 
712
 
583
713
class GCRepositoryPackCollection(RepositoryPackCollection):
584
714
 
585
715
    pack_factory = GCPack
586
716
    resumed_pack_factory = ResumedGCPack
 
717
    normal_packer_class = GCCHKPacker
 
718
    optimising_packer_class = GCCHKPacker
587
719
 
588
720
    def _check_new_inventories(self):
589
721
        """Detect missing inventories or chk root entries for the new revisions
612
744
        missing_corresponding.difference_update(corresponding_invs)
613
745
        if missing_corresponding:
614
746
            problems.append("inventories missing for revisions %s" %
615
 
                (sorted(missing_corresponding),))
 
747
                            (sorted(missing_corresponding),))
616
748
            return problems
617
749
        # Are any chk root entries missing for any inventories?  This includes
618
750
        # any present parent inventories, which may be used when calculating
619
751
        # deltas for streaming.
620
752
        all_inv_keys = set(corresponding_invs)
621
 
        for parent_inv_keys in inv_parent_map.itervalues():
 
753
        for parent_inv_keys in viewvalues(inv_parent_map):
622
754
            all_inv_keys.update(parent_inv_keys)
623
755
        # Filter out ghost parents.
624
756
        all_inv_keys.intersection_update(
635
767
            expected_chk_roots)
636
768
        missing_chk_roots = expected_chk_roots.difference(present_chk_roots)
637
769
        if missing_chk_roots:
638
 
            problems.append("missing referenced chk root keys: %s"
 
770
            problems.append(
 
771
                "missing referenced chk root keys: %s."
 
772
                "Run 'brz reconcile --canonicalize-chks' on the affected "
 
773
                "repository."
639
774
                % (sorted(missing_chk_roots),))
640
775
            # Don't bother checking any further.
641
776
            return problems
647
782
        chk_diff = chk_map.iter_interesting_nodes(
648
783
            chk_bytes_no_fallbacks, root_key_info.interesting_root_keys,
649
784
            root_key_info.uninteresting_root_keys)
650
 
        bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
651
785
        text_keys = set()
652
786
        try:
653
 
            for record in _filter_text_keys(chk_diff, text_keys, bytes_to_info):
 
787
            for record in _filter_text_keys(chk_diff, text_keys,
 
788
                                            chk_map._bytes_to_text_key):
654
789
                pass
655
 
        except errors.NoSuchRevision, e:
 
790
        except errors.NoSuchRevision as e:
656
791
            # XXX: It would be nice if we could give a more precise error here.
657
792
            problems.append("missing chk node(s) for id_to_entry maps")
658
793
        chk_diff = chk_map.iter_interesting_nodes(
661
796
        try:
662
797
            for interesting_rec, interesting_map in chk_diff:
663
798
                pass
664
 
        except errors.NoSuchRevision, e:
 
799
        except errors.NoSuchRevision as e:
665
800
            problems.append(
666
801
                "missing chk node(s) for parent_id_basename_to_file_id maps")
667
802
        present_text_keys = no_fallback_texts_index.get_parent_map(text_keys)
668
803
        missing_text_keys = text_keys.difference(present_text_keys)
669
804
        if missing_text_keys:
670
805
            problems.append("missing text keys: %r"
671
 
                % (sorted(missing_text_keys),))
 
806
                            % (sorted(missing_text_keys),))
672
807
        return problems
673
808
 
674
 
    def _execute_pack_operations(self, pack_operations,
675
 
                                 _packer_class=GCCHKPacker,
676
 
                                 reload_func=None):
677
 
        """Execute a series of pack operations.
678
 
 
679
 
        :param pack_operations: A list of [revision_count, packs_to_combine].
680
 
        :param _packer_class: The class of packer to use (default: Packer).
681
 
        :return: None.
682
 
        """
683
 
        # XXX: Copied across from RepositoryPackCollection simply because we
684
 
        #      want to override the _packer_class ... :(
685
 
        for revision_count, packs in pack_operations:
686
 
            # we may have no-ops from the setup logic
687
 
            if len(packs) == 0:
688
 
                continue
689
 
            packer = GCCHKPacker(self, packs, '.autopack',
690
 
                                 reload_func=reload_func)
691
 
            try:
692
 
                result = packer.pack()
693
 
            except errors.RetryWithNewPacks:
694
 
                # An exception is propagating out of this context, make sure
695
 
                # this packer has cleaned up. Packer() doesn't set its new_pack
696
 
                # state into the RepositoryPackCollection object, so we only
697
 
                # have access to it directly here.
698
 
                if packer.new_pack is not None:
699
 
                    packer.new_pack.abort()
700
 
                raise
701
 
            if result is None:
702
 
                return
703
 
            for pack in packs:
704
 
                self._remove_pack_from_memory(pack)
705
 
        # record the newly available packs and stop advertising the old
706
 
        # packs
707
 
        to_be_obsoleted = []
708
 
        for _, packs in pack_operations:
709
 
            to_be_obsoleted.extend(packs)
710
 
        result = self._save_pack_names(clear_obsolete_packs=True,
711
 
                                       obsolete_packs=to_be_obsoleted)
712
 
        return result
713
 
 
714
 
 
715
 
class CHKInventoryRepository(KnitPackRepository):
716
 
    """subclass of KnitPackRepository that uses CHK based inventories."""
717
 
 
718
 
    def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
719
 
        _serializer):
 
809
 
 
810
class CHKInventoryRepository(PackRepository):
 
811
    """subclass of PackRepository that uses CHK based inventories."""
 
812
 
 
813
    def __init__(self, _format, a_controldir, control_files, _commit_builder_class,
 
814
                 _serializer):
720
815
        """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 :)
 
816
        super(CHKInventoryRepository, self).__init__(_format, a_controldir,
 
817
                                                     control_files, _commit_builder_class, _serializer)
724
818
        index_transport = self._transport.clone('indices')
725
819
        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,
730
 
            _format.index_class,
731
 
            use_chk_index=self._format.supports_chks,
732
 
            )
 
820
                                                           self._transport, index_transport,
 
821
                                                           self._transport.clone(
 
822
                                                               'upload'),
 
823
                                                           self._transport.clone(
 
824
                                                               'packs'),
 
825
                                                           _format.index_builder_class,
 
826
                                                           _format.index_class,
 
827
                                                           use_chk_index=self._format.supports_chks,
 
828
                                                           )
733
829
        self.inventories = GroupCompressVersionedFiles(
734
830
            _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),
 
831
                          add_callback=self._pack_collection.inventory_index.add_callback,
 
832
                          parents=True, is_locked=self.is_locked,
 
833
                          inconsistency_fatal=False),
738
834
            access=self._pack_collection.inventory_index.data_access)
739
835
        self.revisions = GroupCompressVersionedFiles(
740
836
            _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),
 
837
                          add_callback=self._pack_collection.revision_index.add_callback,
 
838
                          parents=True, is_locked=self.is_locked,
 
839
                          track_external_parent_refs=True, track_new_keys=True),
744
840
            access=self._pack_collection.revision_index.data_access,
745
841
            delta=False)
746
842
        self.signatures = GroupCompressVersionedFiles(
747
843
            _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),
 
844
                          add_callback=self._pack_collection.signature_index.add_callback,
 
845
                          parents=False, is_locked=self.is_locked,
 
846
                          inconsistency_fatal=False),
751
847
            access=self._pack_collection.signature_index.data_access,
752
848
            delta=False)
753
849
        self.texts = GroupCompressVersionedFiles(
754
850
            _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),
 
851
                          add_callback=self._pack_collection.text_index.add_callback,
 
852
                          parents=True, is_locked=self.is_locked,
 
853
                          inconsistency_fatal=False),
758
854
            access=self._pack_collection.text_index.data_access)
759
855
        # No parents, individual CHK pages don't have specific ancestry
760
856
        self.chk_bytes = GroupCompressVersionedFiles(
761
857
            _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),
 
858
                          add_callback=self._pack_collection.chk_index.add_callback,
 
859
                          parents=False, is_locked=self.is_locked,
 
860
                          inconsistency_fatal=False),
765
861
            access=self._pack_collection.chk_index.data_access)
766
862
        search_key_name = self._format._serializer.search_key_name
767
863
        search_key_func = chk_map.search_key_registry.get(search_key_name)
788
884
        # make inventory
789
885
        serializer = self._format._serializer
790
886
        result = inventory.CHKInventory.from_inventory(self.chk_bytes, inv,
791
 
            maximum_size=serializer.maximum_size,
792
 
            search_key_name=serializer.search_key_name)
 
887
                                                       maximum_size=serializer.maximum_size,
 
888
                                                       search_key_name=serializer.search_key_name)
793
889
        inv_lines = result.to_lines()
794
890
        return self._inventory_add_lines(revision_id, parents,
795
 
            inv_lines, check_content=False)
 
891
                                         inv_lines, check_content=False)
796
892
 
797
893
    def _create_inv_from_null(self, delta, revision_id):
798
894
        """This will mutate new_inv directly.
816
912
                                 ' no new_path %r' % (file_id,))
817
913
            if new_path == '':
818
914
                new_inv.root_id = file_id
819
 
                parent_id_basename_key = StaticTuple('', '').intern()
 
915
                parent_id_basename_key = StaticTuple(b'', b'').intern()
820
916
            else:
821
917
                utf8_entry_name = entry.name.encode('utf-8')
822
918
                parent_id_basename_key = StaticTuple(entry.parent_id,
829
925
            parent_id_basename_dict[parent_id_basename_key] = file_id
830
926
 
831
927
        new_inv._populate_from_dicts(self.chk_bytes, id_to_entry_dict,
832
 
            parent_id_basename_dict, maximum_size=serializer.maximum_size)
 
928
                                     parent_id_basename_dict, maximum_size=serializer.maximum_size)
833
929
        return new_inv
834
930
 
835
931
    def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
861
957
            raise AssertionError("%r not in write group" % (self,))
862
958
        _mod_revision.check_not_reserved_id(new_revision_id)
863
959
        basis_tree = None
864
 
        if basis_inv is None:
 
960
        if basis_inv is None or not isinstance(basis_inv, inventory.CHKInventory):
865
961
            if basis_revision_id == _mod_revision.NULL_REVISION:
866
962
                new_inv = self._create_inv_from_null(delta, new_revision_id)
 
963
                if new_inv.root_id is None:
 
964
                    raise errors.RootMissing()
867
965
                inv_lines = new_inv.to_lines()
868
966
                return self._inventory_add_lines(new_revision_id, parents,
869
 
                    inv_lines, check_content=False), new_inv
 
967
                                                 inv_lines, check_content=False), new_inv
870
968
            else:
871
969
                basis_tree = self.revision_tree(basis_revision_id)
872
970
                basis_tree.lock_read()
873
 
                basis_inv = basis_tree.inventory
 
971
                basis_inv = basis_tree.root_inventory
874
972
        try:
875
973
            result = basis_inv.create_by_apply_delta(delta, new_revision_id,
876
 
                propagate_caches=propagate_caches)
 
974
                                                     propagate_caches=propagate_caches)
877
975
            inv_lines = result.to_lines()
878
976
            return self._inventory_add_lines(new_revision_id, parents,
879
 
                inv_lines, check_content=False), result
 
977
                                             inv_lines, check_content=False), result
880
978
        finally:
881
979
            if basis_tree is not None:
882
980
                basis_tree.unlock()
883
981
 
884
982
    def _deserialise_inventory(self, revision_id, bytes):
885
983
        return inventory.CHKInventory.deserialise(self.chk_bytes, bytes,
886
 
            (revision_id,))
 
984
                                                  (revision_id,))
887
985
 
888
986
    def _iter_inventories(self, revision_ids, ordering):
889
987
        """Iterate over many inventory objects."""
896
994
            if record.storage_kind != 'absent':
897
995
                texts[record.key] = record.get_bytes_as('fulltext')
898
996
            else:
899
 
                raise errors.NoSuchRevision(self, record.key)
 
997
                texts[record.key] = None
900
998
        for key in keys:
901
 
            yield inventory.CHKInventory.deserialise(self.chk_bytes, texts[key], key)
 
999
            bytes = texts[key]
 
1000
            if bytes is None:
 
1001
                yield (None, key[-1])
 
1002
            else:
 
1003
                yield (inventory.CHKInventory.deserialise(
 
1004
                    self.chk_bytes, bytes, key), key[-1])
902
1005
 
903
 
    def _iter_inventory_xmls(self, revision_ids, ordering):
 
1006
    def _get_inventory_xml(self, revision_id):
 
1007
        """Get serialized inventory as a string."""
904
1008
        # Without a native 'xml' inventory, this method doesn't make sense.
905
1009
        # However older working trees, and older bundles want it - so we supply
906
1010
        # it allowing _get_inventory_xml to work. Bundles currently use the
907
1011
        # 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
 
1012
        # iteration interface offered at all for repositories.
 
1013
        return self._serializer.write_inventory_to_string(
 
1014
            self.get_inventory(revision_id))
913
1015
 
914
1016
    def _find_present_inventory_keys(self, revision_keys):
915
1017
        parent_map = self.inventories.get_parent_map(revision_keys)
929
1031
        rich_root = self.supports_rich_root()
930
1032
        bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
931
1033
        file_id_revisions = {}
932
 
        pb = ui.ui_factory.nested_progress_bar()
933
 
        try:
 
1034
        with ui.ui_factory.nested_progress_bar() as pb:
934
1035
            revision_keys = [(r,) for r in revision_ids]
935
1036
            parent_keys = self._find_parent_keys_of_revisions(revision_keys)
936
1037
            # TODO: instead of using _find_present_inventory_keys, change the
938
1039
            #       However, we only want to tolerate missing parent
939
1040
            #       inventories, not missing inventories for revision_ids
940
1041
            present_parent_inv_keys = self._find_present_inventory_keys(
941
 
                                        parent_keys)
942
 
            present_parent_inv_ids = set(
943
 
                [k[-1] for k in present_parent_inv_keys])
 
1042
                parent_keys)
 
1043
            present_parent_inv_ids = {k[-1] for k in present_parent_inv_keys}
944
1044
            inventories_to_read = set(revision_ids)
945
1045
            inventories_to_read.update(present_parent_inv_ids)
946
1046
            root_key_info = _build_interesting_key_sets(
949
1049
            uninteresting_root_keys = root_key_info.uninteresting_root_keys
950
1050
            chk_bytes = self.chk_bytes
951
1051
            for record, items in chk_map.iter_interesting_nodes(chk_bytes,
952
 
                        interesting_root_keys, uninteresting_root_keys,
953
 
                        pb=pb):
 
1052
                                                                interesting_root_keys, uninteresting_root_keys,
 
1053
                                                                pb=pb):
954
1054
                for name, bytes in items:
955
1055
                    (name_utf8, file_id, revision_id) = bytes_to_info(bytes)
956
1056
                    # TODO: consider interning file_id, revision_id here, or
962
1062
                    try:
963
1063
                        file_id_revisions[file_id].add(revision_id)
964
1064
                    except KeyError:
965
 
                        file_id_revisions[file_id] = set([revision_id])
966
 
        finally:
967
 
            pb.finished()
 
1065
                        file_id_revisions[file_id] = {revision_id}
968
1066
        return file_id_revisions
969
1067
 
970
1068
    def find_text_key_references(self):
982
1080
        revision_keys = self.revisions.keys()
983
1081
        result = {}
984
1082
        rich_roots = self.supports_rich_root()
985
 
        pb = ui.ui_factory.nested_progress_bar()
986
 
        try:
 
1083
        with ui.ui_factory.nested_progress_bar() as pb:
987
1084
            all_revs = self.all_revision_ids()
988
1085
            total = len(all_revs)
989
1086
            for pos, inv in enumerate(self.iter_inventories(all_revs)):
996
1093
                    if entry.revision == inv.revision_id:
997
1094
                        result[key] = True
998
1095
            return result
999
 
        finally:
1000
 
            pb.finished()
 
1096
 
 
1097
    def reconcile_canonicalize_chks(self):
 
1098
        """Reconcile this repository to make sure all CHKs are in canonical
 
1099
        form.
 
1100
        """
 
1101
        from .reconcile import PackReconciler
 
1102
        with self.lock_write():
 
1103
            reconciler = PackReconciler(
 
1104
                self, thorough=True, canonicalize_chks=True)
 
1105
            return reconciler.reconcile()
1001
1106
 
1002
1107
    def _reconcile_pack(self, collection, packs, extension, revs, pb):
1003
1108
        packer = GCCHKReconcilePacker(collection, packs, extension)
1004
1109
        return packer.pack(pb)
1005
1110
 
 
1111
    def _canonicalize_chks_pack(self, collection, packs, extension, revs, pb):
 
1112
        packer = GCCHKCanonicalizingPacker(collection, packs, extension, revs)
 
1113
        return packer.pack(pb)
 
1114
 
1006
1115
    def _get_source(self, to_format):
1007
1116
        """Return a source for streaming from this repository."""
1008
1117
        if self._format._serializer == to_format._serializer:
1013
1122
            return GroupCHKStreamSource(self, to_format)
1014
1123
        return super(CHKInventoryRepository, self)._get_source(to_format)
1015
1124
 
1016
 
 
1017
 
class GroupCHKStreamSource(KnitPackStreamSource):
 
1125
    def _find_inconsistent_revision_parents(self, revisions_iterator=None):
 
1126
        """Find revisions with different parent lists in the revision object
 
1127
        and in the index graph.
 
1128
 
 
1129
        :param revisions_iterator: None, or an iterator of (revid,
 
1130
            Revision-or-None). This iterator controls the revisions checked.
 
1131
        :returns: an iterator yielding tuples of (revison-id, parents-in-index,
 
1132
            parents-in-revision).
 
1133
        """
 
1134
        if not self.is_locked():
 
1135
            raise AssertionError()
 
1136
        vf = self.revisions
 
1137
        if revisions_iterator is None:
 
1138
            revisions_iterator = self.iter_revisions(self.all_revision_ids())
 
1139
        for revid, revision in revisions_iterator:
 
1140
            if revision is None:
 
1141
                pass
 
1142
            parent_map = vf.get_parent_map([(revid,)])
 
1143
            parents_according_to_index = tuple(parent[-1] for parent in
 
1144
                                               parent_map[(revid,)])
 
1145
            parents_according_to_revision = tuple(revision.parent_ids)
 
1146
            if parents_according_to_index != parents_according_to_revision:
 
1147
                yield (revid, parents_according_to_index,
 
1148
                       parents_according_to_revision)
 
1149
 
 
1150
    def _check_for_inconsistent_revision_parents(self):
 
1151
        inconsistencies = list(self._find_inconsistent_revision_parents())
 
1152
        if inconsistencies:
 
1153
            raise errors.BzrCheckError(
 
1154
                "Revision index has inconsistent parents.")
 
1155
 
 
1156
 
 
1157
class GroupCHKStreamSource(StreamSource):
1018
1158
    """Used when both the source and target repo are GroupCHK repos."""
1019
1159
 
1020
1160
    def __init__(self, from_repository, to_format):
1034
1174
        """
1035
1175
        self._chk_id_roots = []
1036
1176
        self._chk_p_id_roots = []
 
1177
 
1037
1178
        def _filtered_inv_stream():
1038
1179
            id_roots_set = set()
1039
1180
            p_id_roots_set = set()
1079
1220
            # TODO: Update Repository.iter_inventories() to add
1080
1221
            #       ignore_missing=True
1081
1222
            present_keys = self.from_repository._find_present_inventory_keys(
1082
 
                            excluded_revision_keys)
 
1223
                excluded_revision_keys)
1083
1224
            present_ids = [k[-1] for k in present_keys]
1084
1225
            uninteresting_root_keys = set()
1085
1226
            uninteresting_pid_root_keys = set()
1087
1228
                uninteresting_root_keys.add(inv.id_to_entry.key())
1088
1229
                uninteresting_pid_root_keys.add(
1089
1230
                    inv.parent_id_basename_to_file_id.key())
1090
 
        bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
1091
1231
        chk_bytes = self.from_repository.chk_bytes
 
1232
 
1092
1233
        def _filter_id_to_entry():
1093
1234
            interesting_nodes = chk_map.iter_interesting_nodes(chk_bytes,
1094
 
                        self._chk_id_roots, uninteresting_root_keys)
 
1235
                                                               self._chk_id_roots, uninteresting_root_keys)
1095
1236
            for record in _filter_text_keys(interesting_nodes, self._text_keys,
1096
 
                    bytes_to_info):
 
1237
                                            chk_map._bytes_to_text_key):
1097
1238
                if record is not None:
1098
1239
                    yield record
1099
1240
            # Consumed
1100
1241
            self._chk_id_roots = None
1101
1242
        yield 'chk_bytes', _filter_id_to_entry()
 
1243
 
1102
1244
        def _get_parent_id_basename_to_file_id_pages():
1103
1245
            for record, items in chk_map.iter_interesting_nodes(chk_bytes,
1104
 
                        self._chk_p_id_roots, uninteresting_pid_root_keys):
 
1246
                                                                self._chk_p_id_roots, uninteresting_pid_root_keys):
1105
1247
                if record is not None:
1106
1248
                    yield record
1107
1249
            # Consumed
1108
1250
            self._chk_p_id_roots = None
1109
1251
        yield 'chk_bytes', _get_parent_id_basename_to_file_id_pages()
1110
1252
 
 
1253
    def _get_text_stream(self):
 
1254
        # Note: We know we don't have to handle adding root keys, because both
 
1255
        # the source and target are the identical network name.
 
1256
        text_stream = self.from_repository.texts.get_record_stream(
 
1257
            self._text_keys, self._text_fetch_order, False)
 
1258
        return ('texts', text_stream)
 
1259
 
1111
1260
    def get_stream(self, search):
 
1261
        def wrap_and_count(pb, rc, stream):
 
1262
            """Yield records from stream while showing progress."""
 
1263
            count = 0
 
1264
            for record in stream:
 
1265
                if count == rc.STEP:
 
1266
                    rc.increment(count)
 
1267
                    pb.update('Estimate', rc.current, rc.max)
 
1268
                    count = 0
 
1269
                count += 1
 
1270
                yield record
 
1271
 
1112
1272
        revision_ids = search.get_keys()
1113
 
        for stream_info in self._fetch_revision_texts(revision_ids):
1114
 
            yield stream_info
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):
1127
 
            yield stream_info
1128
 
        self.from_repository.chk_bytes.clear_cache()
1129
 
        yield self._get_text_stream()
1130
 
        self.from_repository.texts.clear_cache()
 
1273
        with ui.ui_factory.nested_progress_bar() as pb:
 
1274
            rc = self._record_counter
 
1275
            self._record_counter.setup(len(revision_ids))
 
1276
            for stream_info in self._fetch_revision_texts(revision_ids):
 
1277
                yield (stream_info[0],
 
1278
                       wrap_and_count(pb, rc, stream_info[1]))
 
1279
            self._revision_keys = [(rev_id,) for rev_id in revision_ids]
 
1280
            # TODO: The keys to exclude might be part of the search recipe
 
1281
            # For now, exclude all parents that are at the edge of ancestry, for
 
1282
            # which we have inventories
 
1283
            from_repo = self.from_repository
 
1284
            parent_keys = from_repo._find_parent_keys_of_revisions(
 
1285
                self._revision_keys)
 
1286
            self.from_repository.revisions.clear_cache()
 
1287
            self.from_repository.signatures.clear_cache()
 
1288
            # Clear the repo's get_parent_map cache too.
 
1289
            self.from_repository._unstacked_provider.disable_cache()
 
1290
            self.from_repository._unstacked_provider.enable_cache()
 
1291
            s = self._get_inventory_stream(self._revision_keys)
 
1292
            yield (s[0], wrap_and_count(pb, rc, s[1]))
 
1293
            self.from_repository.inventories.clear_cache()
 
1294
            for stream_info in self._get_filtered_chk_streams(parent_keys):
 
1295
                yield (stream_info[0], wrap_and_count(pb, rc, stream_info[1]))
 
1296
            self.from_repository.chk_bytes.clear_cache()
 
1297
            s = self._get_text_stream()
 
1298
            yield (s[0], wrap_and_count(pb, rc, s[1]))
 
1299
            self.from_repository.texts.clear_cache()
 
1300
            pb.update('Done', rc.max, rc.max)
1131
1301
 
1132
1302
    def get_stream_for_missing_keys(self, missing_keys):
1133
1303
        # missing keys can only occur when we are byte copying and not
1137
1307
        for key in missing_keys:
1138
1308
            if key[0] != 'inventories':
1139
1309
                raise AssertionError('The only missing keys we should'
1140
 
                    ' be filling in are inventory keys, not %s'
1141
 
                    % (key[0],))
 
1310
                                     ' be filling in are inventory keys, not %s'
 
1311
                                     % (key[0],))
1142
1312
            missing_inventory_keys.add(key[1:])
1143
1313
        if self._chk_id_roots or self._chk_p_id_roots:
1144
1314
            raise AssertionError('Cannot call get_stream_for_missing_keys'
1145
 
                ' until all of get_stream() has been consumed.')
 
1315
                                 ' until all of get_stream() has been consumed.')
1146
1316
        # Yield the inventory stream, so we can find the chk stream
1147
1317
        # Some of the missing_keys will be missing because they are ghosts.
1148
1318
        # As such, we can ignore them. The Sink is required to verify there are
1187
1357
    return result
1188
1358
 
1189
1359
 
1190
 
def _filter_text_keys(interesting_nodes_iterable, text_keys, bytes_to_info):
 
1360
def _filter_text_keys(interesting_nodes_iterable, text_keys, bytes_to_text_key):
1191
1361
    """Iterate the result of iter_interesting_nodes, yielding the records
1192
1362
    and adding to text_keys.
1193
1363
    """
 
1364
    text_keys_update = text_keys.update
1194
1365
    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
1198
 
            # ignore.
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())
 
1366
        text_keys_update([bytes_to_text_key(b) for n, b in items])
1203
1367
        yield record
1204
1368
 
1205
1369
 
1206
 
 
1207
 
 
1208
 
class RepositoryFormatCHK1(RepositoryFormatPack):
1209
 
    """A hashed CHK+group compress pack repository."""
 
1370
class RepositoryFormat2a(RepositoryFormatPack):
 
1371
    """A CHK repository that uses the bencode revision serializer."""
1210
1372
 
1211
1373
    repository_class = CHKInventoryRepository
1212
1374
    supports_external_lookups = True
1213
1375
    supports_chks = True
1214
 
    # For right now, setting this to True gives us InterModel1And2 rather
1215
 
    # than InterDifferingSerializer
1216
 
    _commit_builder_class = PackRootCommitBuilder
 
1376
    _commit_builder_class = PackCommitBuilder
1217
1377
    rich_root_data = True
1218
 
    _serializer = chk_serializer.chk_serializer_255_bigpage
 
1378
    _serializer = chk_serializer.chk_bencode_serializer
1219
1379
    _commit_inv_deltas = True
1220
1380
    # What index classes to use
1221
1381
    index_builder_class = BTreeBuilder
1227
1387
    # multiple in-a-row (and sharing strings). Topological is better
1228
1388
    # for remote, because we access less data.
1229
1389
    _fetch_order = 'unordered'
1230
 
    _fetch_uses_deltas = False # essentially ignored by the groupcompress code.
 
1390
    # essentially ignored by the groupcompress code.
 
1391
    _fetch_uses_deltas = False
1231
1392
    fast_deltas = True
1232
1393
    pack_compresses = True
1233
1394
 
1234
1395
    def _get_matching_bzrdir(self):
1235
 
        return bzrdir.format_registry.make_bzrdir('development6-rich-root')
1236
 
 
1237
 
    def _ignore_setting_bzrdir(self, format):
1238
 
        pass
1239
 
 
1240
 
    _matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1241
 
 
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')
1246
 
 
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")
1251
 
 
1252
 
 
1253
 
class RepositoryFormatCHK2(RepositoryFormatCHK1):
1254
 
    """A CHK repository that uses the bencode revision serializer."""
1255
 
 
1256
 
    _serializer = chk_serializer.chk_bencode_serializer
1257
 
 
1258
 
    def _get_matching_bzrdir(self):
1259
 
        return bzrdir.format_registry.make_bzrdir('development7-rich-root')
1260
 
 
1261
 
    def _ignore_setting_bzrdir(self, format):
1262
 
        pass
1263
 
 
1264
 
    _matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1265
 
 
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')
1270
 
 
1271
 
 
1272
 
class RepositoryFormat2a(RepositoryFormatCHK2):
1273
 
    """A CHK repository that uses the bencode revision serializer.
1274
 
 
1275
 
    This is the same as RepositoryFormatCHK2 but with a public name.
1276
 
    """
1277
 
 
1278
 
    _serializer = chk_serializer.chk_bencode_serializer
1279
 
 
1280
 
    def _get_matching_bzrdir(self):
1281
 
        return bzrdir.format_registry.make_bzrdir('2a')
1282
 
 
1283
 
    def _ignore_setting_bzrdir(self, format):
1284
 
        pass
1285
 
 
1286
 
    _matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1287
 
 
1288
 
    def get_format_string(self):
1289
 
        return ('Bazaar repository format 2a (needs bzr 1.16 or later)\n')
 
1396
        return controldir.format_registry.make_controldir('2a')
 
1397
 
 
1398
    def _ignore_setting_bzrdir(self, format):
 
1399
        pass
 
1400
 
 
1401
    _matchingcontroldir = property(
 
1402
        _get_matching_bzrdir, _ignore_setting_bzrdir)
 
1403
 
 
1404
    @classmethod
 
1405
    def get_format_string(cls):
 
1406
        return b'Bazaar repository format 2a (needs bzr 1.16 or later)\n'
1290
1407
 
1291
1408
    def get_format_description(self):
1292
1409
        """See RepositoryFormat.get_format_description()."""
1293
1410
        return ("Repository format 2a - rich roots, group compression"
1294
 
            " and chk inventories")
 
1411
                " and chk inventories")
 
1412
 
 
1413
 
 
1414
class RepositoryFormat2aSubtree(RepositoryFormat2a):
 
1415
    """A 2a repository format that supports nested trees.
 
1416
 
 
1417
    """
 
1418
 
 
1419
    def _get_matching_bzrdir(self):
 
1420
        return controldir.format_registry.make_controldir('development-subtree')
 
1421
 
 
1422
    def _ignore_setting_bzrdir(self, format):
 
1423
        pass
 
1424
 
 
1425
    _matchingcontroldir = property(
 
1426
        _get_matching_bzrdir, _ignore_setting_bzrdir)
 
1427
 
 
1428
    @classmethod
 
1429
    def get_format_string(cls):
 
1430
        return b'Bazaar development format 8\n'
 
1431
 
 
1432
    def get_format_description(self):
 
1433
        """See RepositoryFormat.get_format_description()."""
 
1434
        return ("Development repository format 8 - nested trees, "
 
1435
                "group compression and chk inventories")
 
1436
 
 
1437
    experimental = True
 
1438
    supports_tree_reference = True