/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: Martin
  • Date: 2018-07-01 10:53:23 UTC
  • mto: This revision was merged to the branch mainline in revision 7016.
  • Revision ID: gzlist@googlemail.com-20180701105323-dawmz8ngtj3qzdo7
Tweak copyright headers

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):
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,
 
151
        def _write_data(data, flush=False, _buffer=self._buffer,
141
152
            _write=self.write_stream.write, _update=self._hash.update):
142
 
            _buffer[0].append(bytes)
143
 
            _buffer[1] += len(bytes)
 
153
            _buffer[0].append(data)
 
154
            _buffer[1] += len(data)
144
155
            # buffer cap
145
156
            if _buffer[1] > self._cache_limit or flush:
146
 
                bytes = ''.join(_buffer[0])
147
 
                _write(bytes)
148
 
                _update(bytes)
 
157
                data = b''.join(_buffer[0])
 
158
                _write(data)
 
159
                _update(data)
149
160
                _buffer[:] = [[], 0]
150
161
        # expose this on self, for the occasion when clients want to add data.
151
162
        self._write_data = _write_data
262
273
        remaining_keys = set(keys)
263
274
        counter = [0]
264
275
        if self._gather_text_refs:
265
 
            bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
266
276
            self._text_refs = set()
267
277
        def _get_referenced_stream(root_keys, parse_leaf_nodes=False):
268
278
            cur_keys = root_keys
271
281
                remaining_keys.difference_update(cur_keys)
272
282
                next_keys = set()
273
283
                def handle_internal_node(node):
274
 
                    for prefix, value in node._items.iteritems():
 
284
                    for prefix, value in viewitems(node._items):
275
285
                        # We don't want to request the same key twice, and we
276
286
                        # want to order it by the first time it is seen.
277
287
                        # Even further, we don't want to request a key which is
289
299
                    # Store is None, because we know we have a LeafNode, and we
290
300
                    # just want its entries
291
301
                    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))
 
302
                        self._text_refs.add(chk_map._bytes_to_text_key(bytes))
294
303
                def next_stream():
295
304
                    stream = source_vf.get_record_stream(cur_keys,
296
305
                                                         'as-requested', True)
352
361
        """Build a VersionedFiles instance on top of this group of packs."""
353
362
        index_name = index_name + '_index'
354
363
        index_to_pack = {}
355
 
        access = knit._DirectPackAccess(index_to_pack,
356
 
                                        reload_func=self._reload_func)
 
364
        access = _DirectPackAccess(index_to_pack,
 
365
                                   reload_func=self._reload_func)
357
366
        if for_write:
358
367
            # Use new_pack
359
368
            if self.new_pack is None:
393
402
                     pb_offset):
394
403
        trace.mutter('repacking %d %s', len(keys), message)
395
404
        self.pb.update('repacking %s' % (message,), pb_offset)
396
 
        child_pb = ui.ui_factory.nested_progress_bar()
397
 
        try:
 
405
        with ui.ui_factory.nested_progress_bar() as child_pb:
398
406
            stream = vf_to_stream(source_vf, keys, message, child_pb)
399
407
            for _ in target_vf._insert_record_stream(stream,
400
408
                                                     random_id=True,
401
409
                                                     reuse_blocks=False):
402
410
                pass
403
 
        finally:
404
 
            child_pb.finished()
405
411
 
406
412
    def _copy_revision_texts(self):
407
413
        source_vf, target_vf = self._build_vfs('revision', True, False)
421
427
        inventory_keys = source_vf.keys()
422
428
        missing_inventories = set(self.revision_keys).difference(inventory_keys)
423
429
        if missing_inventories:
424
 
            missing_inventories = sorted(missing_inventories)
425
 
            raise ValueError('We are missing inventories for revisions: %s'
426
 
                % (missing_inventories,))
 
430
            # Go back to the original repo, to see if these are really missing
 
431
            # https://bugs.launchpad.net/bzr/+bug/437003
 
432
            # If we are packing a subset of the repo, it is fine to just have
 
433
            # the data in another Pack file, which is not included in this pack
 
434
            # operation.
 
435
            inv_index = self._pack_collection.repo.inventories._index
 
436
            pmap = inv_index.get_parent_map(missing_inventories)
 
437
            really_missing = missing_inventories.difference(pmap)
 
438
            if really_missing:
 
439
                missing_inventories = sorted(really_missing)
 
440
                raise ValueError('We are missing inventories for revisions: %s'
 
441
                    % (missing_inventories,))
427
442
        self._copy_stream(source_vf, target_vf, inventory_keys,
428
443
                          'inventories', self._get_filtered_inv_stream, 2)
429
444
 
 
445
    def _get_chk_vfs_for_copy(self):
 
446
        return self._build_vfs('chk', False, False)
 
447
 
430
448
    def _copy_chk_texts(self):
431
 
        source_vf, target_vf = self._build_vfs('chk', False, False)
 
449
        source_vf, target_vf = self._get_chk_vfs_for_copy()
432
450
        # TODO: This is technically spurious... if it is a performance issue,
433
451
        #       remove it
434
452
        total_keys = source_vf.keys()
437
455
                     len(self._chk_id_roots), len(self._chk_p_id_roots),
438
456
                     len(total_keys))
439
457
        self.pb.update('repacking chk', 3)
440
 
        child_pb = ui.ui_factory.nested_progress_bar()
441
 
        try:
 
458
        with ui.ui_factory.nested_progress_bar() as child_pb:
442
459
            for stream in self._get_chk_streams(source_vf, total_keys,
443
460
                                                pb=child_pb):
444
461
                for _ in target_vf._insert_record_stream(stream,
445
462
                                                         random_id=True,
446
463
                                                         reuse_blocks=False):
447
464
                    pass
448
 
        finally:
449
 
            child_pb.finished()
450
465
 
451
466
    def _copy_text_texts(self):
452
467
        source_vf, target_vf = self._build_vfs('text', True, True)
498
513
class GCCHKReconcilePacker(GCCHKPacker):
499
514
    """A packer which regenerates indices etc as it copies.
500
515
 
501
 
    This is used by ``bzr reconcile`` to cause parent text pointers to be
 
516
    This is used by ``brz reconcile`` to cause parent text pointers to be
502
517
    regenerated.
503
518
    """
504
519
 
527
542
        ancestor_keys = revision_vf.get_parent_map(revision_vf.keys())
528
543
        # Strip keys back into revision_ids.
529
544
        ancestors = dict((k[0], tuple([p[0] for p in parents]))
530
 
                         for k, parents in ancestor_keys.iteritems())
 
545
                         for k, parents in viewitems(ancestor_keys))
531
546
        del ancestor_keys
532
547
        # TODO: _generate_text_key_index should be much cheaper to generate from
533
548
        #       a chk repository, rather than the current implementation
580
595
        return new_pack.data_inserted() and self._data_changed
581
596
 
582
597
 
 
598
class GCCHKCanonicalizingPacker(GCCHKPacker):
 
599
    """A packer that ensures inventories have canonical-form CHK maps.
 
600
    
 
601
    Ideally this would be part of reconcile, but it's very slow and rarely
 
602
    needed.  (It repairs repositories affected by
 
603
    https://bugs.launchpad.net/bzr/+bug/522637).
 
604
    """
 
605
 
 
606
    def __init__(self, *args, **kwargs):
 
607
        super(GCCHKCanonicalizingPacker, self).__init__(*args, **kwargs)
 
608
        self._data_changed = False
 
609
 
 
610
    def _exhaust_stream(self, source_vf, keys, message, vf_to_stream, pb_offset):
 
611
        """Create and exhaust a stream, but don't insert it.
 
612
 
 
613
        This is useful to get the side-effects of generating a stream.
 
614
        """
 
615
        self.pb.update('scanning %s' % (message,), pb_offset)
 
616
        with ui.ui_factory.nested_progress_bar() as child_pb:
 
617
            list(vf_to_stream(source_vf, keys, message, child_pb))
 
618
 
 
619
    def _copy_inventory_texts(self):
 
620
        source_vf, target_vf = self._build_vfs('inventory', True, True)
 
621
        source_chk_vf, target_chk_vf = self._get_chk_vfs_for_copy()
 
622
        inventory_keys = source_vf.keys()
 
623
        # First, copy the existing CHKs on the assumption that most of them
 
624
        # will be correct.  This will save us from having to reinsert (and
 
625
        # recompress) these records later at the cost of perhaps preserving a
 
626
        # few unused CHKs. 
 
627
        # (Iterate but don't insert _get_filtered_inv_stream to populate the
 
628
        # variables needed by GCCHKPacker._copy_chk_texts.)
 
629
        self._exhaust_stream(source_vf, inventory_keys, 'inventories',
 
630
                self._get_filtered_inv_stream, 2)
 
631
        GCCHKPacker._copy_chk_texts(self)
 
632
        # Now copy and fix the inventories, and any regenerated CHKs.
 
633
        def chk_canonicalizing_inv_stream(source_vf, keys, message, pb=None):
 
634
            return self._get_filtered_canonicalizing_inv_stream(
 
635
                source_vf, keys, message, pb, source_chk_vf, target_chk_vf)
 
636
        self._copy_stream(source_vf, target_vf, inventory_keys,
 
637
                          'inventories', chk_canonicalizing_inv_stream, 4)
 
638
 
 
639
    def _copy_chk_texts(self):
 
640
        # No-op; in this class this happens during _copy_inventory_texts.
 
641
        pass
 
642
 
 
643
    def _get_filtered_canonicalizing_inv_stream(self, source_vf, keys, message,
 
644
            pb=None, source_chk_vf=None, target_chk_vf=None):
 
645
        """Filter the texts of inventories, regenerating CHKs to make sure they
 
646
        are canonical.
 
647
        """
 
648
        total_keys = len(keys)
 
649
        target_chk_vf = versionedfile.NoDupeAddLinesDecorator(target_chk_vf)
 
650
        def _filtered_inv_stream():
 
651
            stream = source_vf.get_record_stream(keys, 'groupcompress', True)
 
652
            search_key_name = None
 
653
            for idx, record in enumerate(stream):
 
654
                # Inventories should always be with revisions; assume success.
 
655
                bytes = record.get_bytes_as('fulltext')
 
656
                chk_inv = inventory.CHKInventory.deserialise(
 
657
                    source_chk_vf, bytes, record.key)
 
658
                if pb is not None:
 
659
                    pb.update('inv', idx, total_keys)
 
660
                chk_inv.id_to_entry._ensure_root()
 
661
                if search_key_name is None:
 
662
                    # Find the name corresponding to the search_key_func
 
663
                    search_key_reg = chk_map.search_key_registry
 
664
                    for search_key_name, func in viewitems(search_key_reg):
 
665
                        if func == chk_inv.id_to_entry._search_key_func:
 
666
                            break
 
667
                canonical_inv = inventory.CHKInventory.from_inventory(
 
668
                    target_chk_vf, chk_inv,
 
669
                    maximum_size=chk_inv.id_to_entry._root_node._maximum_size,
 
670
                    search_key_name=search_key_name)
 
671
                if chk_inv.id_to_entry.key() != canonical_inv.id_to_entry.key():
 
672
                    trace.mutter(
 
673
                        'Non-canonical CHK map for id_to_entry of inv: %s '
 
674
                        '(root is %s, should be %s)' % (chk_inv.revision_id,
 
675
                        chk_inv.id_to_entry.key()[0],
 
676
                        canonical_inv.id_to_entry.key()[0]))
 
677
                    self._data_changed = True
 
678
                p_id_map = chk_inv.parent_id_basename_to_file_id
 
679
                p_id_map._ensure_root()
 
680
                canon_p_id_map = canonical_inv.parent_id_basename_to_file_id
 
681
                if p_id_map.key() != canon_p_id_map.key():
 
682
                    trace.mutter(
 
683
                        'Non-canonical CHK map for parent_id_to_basename of '
 
684
                        'inv: %s (root is %s, should be %s)'
 
685
                        % (chk_inv.revision_id, p_id_map.key()[0],
 
686
                           canon_p_id_map.key()[0]))
 
687
                    self._data_changed = True
 
688
                yield versionedfile.ChunkedContentFactory(record.key,
 
689
                        record.parents, record.sha1,
 
690
                        canonical_inv.to_lines())
 
691
            # We have finished processing all of the inventory records, we
 
692
            # don't need these sets anymore
 
693
        return _filtered_inv_stream()
 
694
 
 
695
    def _use_pack(self, new_pack):
 
696
        """Override _use_pack to check for reconcile having changed content."""
 
697
        return new_pack.data_inserted() and self._data_changed
 
698
 
 
699
 
583
700
class GCRepositoryPackCollection(RepositoryPackCollection):
584
701
 
585
702
    pack_factory = GCPack
586
703
    resumed_pack_factory = ResumedGCPack
 
704
    normal_packer_class = GCCHKPacker
 
705
    optimising_packer_class = GCCHKPacker
587
706
 
588
707
    def _check_new_inventories(self):
589
708
        """Detect missing inventories or chk root entries for the new revisions
618
737
        # any present parent inventories, which may be used when calculating
619
738
        # deltas for streaming.
620
739
        all_inv_keys = set(corresponding_invs)
621
 
        for parent_inv_keys in inv_parent_map.itervalues():
 
740
        for parent_inv_keys in viewvalues(inv_parent_map):
622
741
            all_inv_keys.update(parent_inv_keys)
623
742
        # Filter out ghost parents.
624
743
        all_inv_keys.intersection_update(
635
754
            expected_chk_roots)
636
755
        missing_chk_roots = expected_chk_roots.difference(present_chk_roots)
637
756
        if missing_chk_roots:
638
 
            problems.append("missing referenced chk root keys: %s"
 
757
            problems.append(
 
758
                "missing referenced chk root keys: %s."
 
759
                "Run 'brz reconcile --canonicalize-chks' on the affected "
 
760
                "repository."
639
761
                % (sorted(missing_chk_roots),))
640
762
            # Don't bother checking any further.
641
763
            return problems
647
769
        chk_diff = chk_map.iter_interesting_nodes(
648
770
            chk_bytes_no_fallbacks, root_key_info.interesting_root_keys,
649
771
            root_key_info.uninteresting_root_keys)
650
 
        bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
651
772
        text_keys = set()
652
773
        try:
653
 
            for record in _filter_text_keys(chk_diff, text_keys, bytes_to_info):
 
774
            for record in _filter_text_keys(chk_diff, text_keys,
 
775
                                            chk_map._bytes_to_text_key):
654
776
                pass
655
 
        except errors.NoSuchRevision, e:
 
777
        except errors.NoSuchRevision as e:
656
778
            # XXX: It would be nice if we could give a more precise error here.
657
779
            problems.append("missing chk node(s) for id_to_entry maps")
658
780
        chk_diff = chk_map.iter_interesting_nodes(
661
783
        try:
662
784
            for interesting_rec, interesting_map in chk_diff:
663
785
                pass
664
 
        except errors.NoSuchRevision, e:
 
786
        except errors.NoSuchRevision as e:
665
787
            problems.append(
666
788
                "missing chk node(s) for parent_id_basename_to_file_id maps")
667
789
        present_text_keys = no_fallback_texts_index.get_parent_map(text_keys)
671
793
                % (sorted(missing_text_keys),))
672
794
        return problems
673
795
 
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,
 
796
 
 
797
class CHKInventoryRepository(PackRepository):
 
798
    """subclass of PackRepository that uses CHK based inventories."""
 
799
 
 
800
    def __init__(self, _format, a_controldir, control_files, _commit_builder_class,
719
801
        _serializer):
720
802
        """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 :)
 
803
        super(CHKInventoryRepository, self).__init__(_format, a_controldir,
 
804
            control_files, _commit_builder_class, _serializer)
724
805
        index_transport = self._transport.clone('indices')
725
806
        self._pack_collection = GCRepositoryPackCollection(self,
726
807
            self._transport, index_transport,
816
897
                                 ' no new_path %r' % (file_id,))
817
898
            if new_path == '':
818
899
                new_inv.root_id = file_id
819
 
                parent_id_basename_key = StaticTuple('', '').intern()
 
900
                parent_id_basename_key = StaticTuple(b'', b'').intern()
820
901
            else:
821
902
                utf8_entry_name = entry.name.encode('utf-8')
822
903
                parent_id_basename_key = StaticTuple(entry.parent_id,
861
942
            raise AssertionError("%r not in write group" % (self,))
862
943
        _mod_revision.check_not_reserved_id(new_revision_id)
863
944
        basis_tree = None
864
 
        if basis_inv is None:
 
945
        if basis_inv is None or not isinstance(basis_inv, inventory.CHKInventory):
865
946
            if basis_revision_id == _mod_revision.NULL_REVISION:
866
947
                new_inv = self._create_inv_from_null(delta, new_revision_id)
 
948
                if new_inv.root_id is None:
 
949
                    raise errors.RootMissing()
867
950
                inv_lines = new_inv.to_lines()
868
951
                return self._inventory_add_lines(new_revision_id, parents,
869
952
                    inv_lines, check_content=False), new_inv
870
953
            else:
871
954
                basis_tree = self.revision_tree(basis_revision_id)
872
955
                basis_tree.lock_read()
873
 
                basis_inv = basis_tree.inventory
 
956
                basis_inv = basis_tree.root_inventory
874
957
        try:
875
958
            result = basis_inv.create_by_apply_delta(delta, new_revision_id,
876
959
                propagate_caches=propagate_caches)
896
979
            if record.storage_kind != 'absent':
897
980
                texts[record.key] = record.get_bytes_as('fulltext')
898
981
            else:
899
 
                raise errors.NoSuchRevision(self, record.key)
 
982
                texts[record.key] = None
900
983
        for key in keys:
901
 
            yield inventory.CHKInventory.deserialise(self.chk_bytes, texts[key], key)
 
984
            bytes = texts[key]
 
985
            if bytes is None:
 
986
                yield (None, key[-1])
 
987
            else:
 
988
                yield (inventory.CHKInventory.deserialise(
 
989
                    self.chk_bytes, bytes, key), key[-1])
902
990
 
903
 
    def _iter_inventory_xmls(self, revision_ids, ordering):
 
991
    def _get_inventory_xml(self, revision_id):
 
992
        """Get serialized inventory as a string."""
904
993
        # Without a native 'xml' inventory, this method doesn't make sense.
905
994
        # However older working trees, and older bundles want it - so we supply
906
995
        # it allowing _get_inventory_xml to work. Bundles currently use the
907
996
        # 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
 
997
        # iteration interface offered at all for repositories.
 
998
        return self._serializer.write_inventory_to_string(
 
999
            self.get_inventory(revision_id))
913
1000
 
914
1001
    def _find_present_inventory_keys(self, revision_keys):
915
1002
        parent_map = self.inventories.get_parent_map(revision_keys)
929
1016
        rich_root = self.supports_rich_root()
930
1017
        bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
931
1018
        file_id_revisions = {}
932
 
        pb = ui.ui_factory.nested_progress_bar()
933
 
        try:
 
1019
        with ui.ui_factory.nested_progress_bar() as pb:
934
1020
            revision_keys = [(r,) for r in revision_ids]
935
1021
            parent_keys = self._find_parent_keys_of_revisions(revision_keys)
936
1022
            # TODO: instead of using _find_present_inventory_keys, change the
939
1025
            #       inventories, not missing inventories for revision_ids
940
1026
            present_parent_inv_keys = self._find_present_inventory_keys(
941
1027
                                        parent_keys)
942
 
            present_parent_inv_ids = set(
943
 
                [k[-1] for k in present_parent_inv_keys])
 
1028
            present_parent_inv_ids = {k[-1] for k in present_parent_inv_keys}
944
1029
            inventories_to_read = set(revision_ids)
945
1030
            inventories_to_read.update(present_parent_inv_ids)
946
1031
            root_key_info = _build_interesting_key_sets(
962
1047
                    try:
963
1048
                        file_id_revisions[file_id].add(revision_id)
964
1049
                    except KeyError:
965
 
                        file_id_revisions[file_id] = set([revision_id])
966
 
        finally:
967
 
            pb.finished()
 
1050
                        file_id_revisions[file_id] = {revision_id}
968
1051
        return file_id_revisions
969
1052
 
970
1053
    def find_text_key_references(self):
982
1065
        revision_keys = self.revisions.keys()
983
1066
        result = {}
984
1067
        rich_roots = self.supports_rich_root()
985
 
        pb = ui.ui_factory.nested_progress_bar()
986
 
        try:
 
1068
        with ui.ui_factory.nested_progress_bar() as pb:
987
1069
            all_revs = self.all_revision_ids()
988
1070
            total = len(all_revs)
989
1071
            for pos, inv in enumerate(self.iter_inventories(all_revs)):
996
1078
                    if entry.revision == inv.revision_id:
997
1079
                        result[key] = True
998
1080
            return result
999
 
        finally:
1000
 
            pb.finished()
 
1081
 
 
1082
    def reconcile_canonicalize_chks(self):
 
1083
        """Reconcile this repository to make sure all CHKs are in canonical
 
1084
        form.
 
1085
        """
 
1086
        from breezy.reconcile import PackReconciler
 
1087
        with self.lock_write():
 
1088
            reconciler = PackReconciler(self, thorough=True, canonicalize_chks=True)
 
1089
            reconciler.reconcile()
 
1090
            return reconciler
1001
1091
 
1002
1092
    def _reconcile_pack(self, collection, packs, extension, revs, pb):
1003
1093
        packer = GCCHKReconcilePacker(collection, packs, extension)
1004
1094
        return packer.pack(pb)
1005
1095
 
 
1096
    def _canonicalize_chks_pack(self, collection, packs, extension, revs, pb):
 
1097
        packer = GCCHKCanonicalizingPacker(collection, packs, extension, revs)
 
1098
        return packer.pack(pb)
 
1099
 
1006
1100
    def _get_source(self, to_format):
1007
1101
        """Return a source for streaming from this repository."""
1008
1102
        if self._format._serializer == to_format._serializer:
1013
1107
            return GroupCHKStreamSource(self, to_format)
1014
1108
        return super(CHKInventoryRepository, self)._get_source(to_format)
1015
1109
 
1016
 
 
1017
 
class GroupCHKStreamSource(KnitPackStreamSource):
 
1110
    def _find_inconsistent_revision_parents(self, revisions_iterator=None):
 
1111
        """Find revisions with different parent lists in the revision object
 
1112
        and in the index graph.
 
1113
 
 
1114
        :param revisions_iterator: None, or an iterator of (revid,
 
1115
            Revision-or-None). This iterator controls the revisions checked.
 
1116
        :returns: an iterator yielding tuples of (revison-id, parents-in-index,
 
1117
            parents-in-revision).
 
1118
        """
 
1119
        if not self.is_locked():
 
1120
            raise AssertionError()
 
1121
        vf = self.revisions
 
1122
        if revisions_iterator is None:
 
1123
            revisions_iterator = self.iter_revisions(self.all_revision_ids())
 
1124
        for revid, revision in revisions_iterator:
 
1125
            if revision is None:
 
1126
                pass
 
1127
            parent_map = vf.get_parent_map([(revid,)])
 
1128
            parents_according_to_index = tuple(parent[-1] for parent in
 
1129
                parent_map[(revid,)])
 
1130
            parents_according_to_revision = tuple(revision.parent_ids)
 
1131
            if parents_according_to_index != parents_according_to_revision:
 
1132
                yield (revid, parents_according_to_index,
 
1133
                    parents_according_to_revision)
 
1134
 
 
1135
    def _check_for_inconsistent_revision_parents(self):
 
1136
        inconsistencies = list(self._find_inconsistent_revision_parents())
 
1137
        if inconsistencies:
 
1138
            raise errors.BzrCheckError(
 
1139
                "Revision index has inconsistent parents.")
 
1140
 
 
1141
 
 
1142
class GroupCHKStreamSource(StreamSource):
1018
1143
    """Used when both the source and target repo are GroupCHK repos."""
1019
1144
 
1020
1145
    def __init__(self, from_repository, to_format):
1087
1212
                uninteresting_root_keys.add(inv.id_to_entry.key())
1088
1213
                uninteresting_pid_root_keys.add(
1089
1214
                    inv.parent_id_basename_to_file_id.key())
1090
 
        bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
1091
1215
        chk_bytes = self.from_repository.chk_bytes
1092
1216
        def _filter_id_to_entry():
1093
1217
            interesting_nodes = chk_map.iter_interesting_nodes(chk_bytes,
1094
1218
                        self._chk_id_roots, uninteresting_root_keys)
1095
1219
            for record in _filter_text_keys(interesting_nodes, self._text_keys,
1096
 
                    bytes_to_info):
 
1220
                    chk_map._bytes_to_text_key):
1097
1221
                if record is not None:
1098
1222
                    yield record
1099
1223
            # Consumed
1108
1232
            self._chk_p_id_roots = None
1109
1233
        yield 'chk_bytes', _get_parent_id_basename_to_file_id_pages()
1110
1234
 
 
1235
    def _get_text_stream(self):
 
1236
        # Note: We know we don't have to handle adding root keys, because both
 
1237
        # the source and target are the identical network name.
 
1238
        text_stream = self.from_repository.texts.get_record_stream(
 
1239
                        self._text_keys, self._text_fetch_order, False)
 
1240
        return ('texts', text_stream)
 
1241
 
1111
1242
    def get_stream(self, search):
 
1243
        def wrap_and_count(pb, rc, stream):
 
1244
            """Yield records from stream while showing progress."""
 
1245
            count = 0
 
1246
            for record in stream:
 
1247
                if count == rc.STEP:
 
1248
                    rc.increment(count)
 
1249
                    pb.update('Estimate', rc.current, rc.max)
 
1250
                    count = 0
 
1251
                count += 1
 
1252
                yield record
 
1253
 
1112
1254
        revision_ids = search.get_keys()
 
1255
        pb = ui.ui_factory.nested_progress_bar()
 
1256
        rc = self._record_counter
 
1257
        self._record_counter.setup(len(revision_ids))
1113
1258
        for stream_info in self._fetch_revision_texts(revision_ids):
1114
 
            yield stream_info
 
1259
            yield (stream_info[0],
 
1260
                wrap_and_count(pb, rc, stream_info[1]))
1115
1261
        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
1262
        # TODO: The keys to exclude might be part of the search recipe
1121
1263
        # For now, exclude all parents that are at the edge of ancestry, for
1122
1264
        # which we have inventories
1123
1265
        from_repo = self.from_repository
1124
1266
        parent_keys = from_repo._find_parent_keys_of_revisions(
1125
1267
                        self._revision_keys)
 
1268
        self.from_repository.revisions.clear_cache()
 
1269
        self.from_repository.signatures.clear_cache()
 
1270
        # Clear the repo's get_parent_map cache too.
 
1271
        self.from_repository._unstacked_provider.disable_cache()
 
1272
        self.from_repository._unstacked_provider.enable_cache()
 
1273
        s = self._get_inventory_stream(self._revision_keys)
 
1274
        yield (s[0], wrap_and_count(pb, rc, s[1]))
 
1275
        self.from_repository.inventories.clear_cache()
1126
1276
        for stream_info in self._get_filtered_chk_streams(parent_keys):
1127
 
            yield stream_info
 
1277
            yield (stream_info[0], wrap_and_count(pb, rc, stream_info[1]))
1128
1278
        self.from_repository.chk_bytes.clear_cache()
1129
 
        yield self._get_text_stream()
 
1279
        s = self._get_text_stream()
 
1280
        yield (s[0], wrap_and_count(pb, rc, s[1]))
1130
1281
        self.from_repository.texts.clear_cache()
 
1282
        pb.update('Done', rc.max, rc.max)
 
1283
        pb.finished()
1131
1284
 
1132
1285
    def get_stream_for_missing_keys(self, missing_keys):
1133
1286
        # missing keys can only occur when we are byte copying and not
1187
1340
    return result
1188
1341
 
1189
1342
 
1190
 
def _filter_text_keys(interesting_nodes_iterable, text_keys, bytes_to_info):
 
1343
def _filter_text_keys(interesting_nodes_iterable, text_keys, bytes_to_text_key):
1191
1344
    """Iterate the result of iter_interesting_nodes, yielding the records
1192
1345
    and adding to text_keys.
1193
1346
    """
 
1347
    text_keys_update = text_keys.update
1194
1348
    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())
 
1349
        text_keys_update([bytes_to_text_key(b) for n, b in items])
1203
1350
        yield record
1204
1351
 
1205
1352
 
1206
 
 
1207
 
 
1208
 
class RepositoryFormatCHK1(RepositoryFormatPack):
1209
 
    """A hashed CHK+group compress pack repository."""
 
1353
class RepositoryFormat2a(RepositoryFormatPack):
 
1354
    """A CHK repository that uses the bencode revision serializer."""
1210
1355
 
1211
1356
    repository_class = CHKInventoryRepository
1212
1357
    supports_external_lookups = True
1213
1358
    supports_chks = True
1214
 
    # For right now, setting this to True gives us InterModel1And2 rather
1215
 
    # than InterDifferingSerializer
1216
 
    _commit_builder_class = PackRootCommitBuilder
 
1359
    _commit_builder_class = PackCommitBuilder
1217
1360
    rich_root_data = True
1218
 
    _serializer = chk_serializer.chk_serializer_255_bigpage
 
1361
    _serializer = chk_serializer.chk_bencode_serializer
1219
1362
    _commit_inv_deltas = True
1220
1363
    # What index classes to use
1221
1364
    index_builder_class = BTreeBuilder
1232
1375
    pack_compresses = True
1233
1376
 
1234
1377
    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')
 
1378
        return controldir.format_registry.make_controldir('2a')
 
1379
 
 
1380
    def _ignore_setting_bzrdir(self, format):
 
1381
        pass
 
1382
 
 
1383
    _matchingcontroldir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
 
1384
 
 
1385
    @classmethod
 
1386
    def get_format_string(cls):
 
1387
        return b'Bazaar repository format 2a (needs bzr 1.16 or later)\n'
1290
1388
 
1291
1389
    def get_format_description(self):
1292
1390
        """See RepositoryFormat.get_format_description()."""
1293
1391
        return ("Repository format 2a - rich roots, group compression"
1294
1392
            " and chk inventories")
 
1393
 
 
1394
 
 
1395
class RepositoryFormat2aSubtree(RepositoryFormat2a):
 
1396
    """A 2a repository format that supports nested trees.
 
1397
 
 
1398
    """
 
1399
 
 
1400
    def _get_matching_bzrdir(self):
 
1401
        return controldir.format_registry.make_controldir('development-subtree')
 
1402
 
 
1403
    def _ignore_setting_bzrdir(self, format):
 
1404
        pass
 
1405
 
 
1406
    _matchingcontroldir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
 
1407
 
 
1408
    @classmethod
 
1409
    def get_format_string(cls):
 
1410
        return b'Bazaar development format 8\n'
 
1411
 
 
1412
    def get_format_description(self):
 
1413
        """See RepositoryFormat.get_format_description()."""
 
1414
        return ("Development repository format 8 - nested trees, "
 
1415
                "group compression and chk inventories")
 
1416
 
 
1417
    experimental = True
 
1418
    supports_tree_reference = True