← Back to branch summary

~bzr-pqm/bzr/bzr.dev

« back to all changes in this revision

Viewing changes to bzrlib/repofmt/groupcompress_repo.py

  • Committer: Danny van Heumen
  • Date: 2010-03-09 21:42:11 UTC
  • mto: (4634.139.5 2.0)
  • mto: This revision was merged to the branch mainline in revision 5160.
  • Revision ID: danny@dannyvanheumen.nl-20100309214211-iqh42x6qcikgd9p3
Reverted now-useless TODO list.

Show diffs side-by-side

added added

removed removed

Lines of Context:
bzrlib/repofmt/groupcompress_repo.py
1
 
# Copyright (C) 2008, 2009 Canonical Ltd
 
1
# Copyright (C) 2008, 2009, 2010 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
154
154
        self._writer.begin()
155
155
        # what state is the pack in? (open, finished, aborted)
156
156
        self._state = 'open'
 
157
        # no name until we finish writing the content
 
158
        self.name = None
157
159
 
158
160
    def _check_references(self):
159
161
        """Make sure our external references are present.
350
352
        """Build a VersionedFiles instance on top of this group of packs."""
351
353
        index_name = index_name + '_index'
352
354
        index_to_pack = {}
353
 
        access = knit._DirectPackAccess(index_to_pack)
 
355
        access = knit._DirectPackAccess(index_to_pack,
 
356
                                        reload_func=self._reload_func)
354
357
        if for_write:
355
358
            # Use new_pack
356
359
            if self.new_pack is None:
410
413
 
411
414
    def _copy_inventory_texts(self):
412
415
        source_vf, target_vf = self._build_vfs('inventory', True, True)
413
 
        self._copy_stream(source_vf, target_vf, self.revision_keys,
 
416
        # It is not sufficient to just use self.revision_keys, as stacked
 
417
        # repositories can have more inventories than they have revisions.
 
418
        # One alternative would be to do something with
 
419
        # get_parent_map(self.revision_keys), but that shouldn't be any faster
 
420
        # than this.
 
421
        inventory_keys = source_vf.keys()
 
422
        missing_inventories = set(self.revision_keys).difference(inventory_keys)
 
423
        if missing_inventories:
 
424
            missing_inventories = sorted(missing_inventories)
 
425
            raise ValueError('We are missing inventories for revisions: %s'
 
426
                % (missing_inventories,))
 
427
        self._copy_stream(source_vf, target_vf, inventory_keys,
414
428
                          'inventories', self._get_filtered_inv_stream, 2)
415
429
 
416
430
    def _copy_chk_texts(self):
466
480
        if not self._use_pack(self.new_pack):
467
481
            self.new_pack.abort()
468
482
            return None
 
483
        self.new_pack.finish_content()
 
484
        if len(self.packs) == 1:
 
485
            old_pack = self.packs[0]
 
486
            if old_pack.name == self.new_pack._hash.hexdigest():
 
487
                # The single old pack was already optimally packed.
 
488
                trace.mutter('single pack %s was already optimally packed',
 
489
                    old_pack.name)
 
490
                self.new_pack.abort()
 
491
                return None
469
492
        self.pb.update('finishing repack', 6, 7)
470
493
        self.new_pack.finish()
471
494
        self._pack_collection.allocate(self.new_pack)
562
585
    pack_factory = GCPack
563
586
    resumed_pack_factory = ResumedGCPack
564
587
 
 
588
    def _check_new_inventories(self):
 
589
        """Detect missing inventories or chk root entries for the new revisions
 
590
        in this write group.
 
591
 
 
592
        :returns: list of strs, summarising any problems found.  If the list is
 
593
            empty no problems were found.
 
594
        """
 
595
        # Ensure that all revisions added in this write group have:
 
596
        #   - corresponding inventories,
 
597
        #   - chk root entries for those inventories,
 
598
        #   - and any present parent inventories have their chk root
 
599
        #     entries too.
 
600
        # And all this should be independent of any fallback repository.
 
601
        problems = []
 
602
        key_deps = self.repo.revisions._index._key_dependencies
 
603
        new_revisions_keys = key_deps.get_new_keys()
 
604
        no_fallback_inv_index = self.repo.inventories._index
 
605
        no_fallback_chk_bytes_index = self.repo.chk_bytes._index
 
606
        no_fallback_texts_index = self.repo.texts._index
 
607
        inv_parent_map = no_fallback_inv_index.get_parent_map(
 
608
            new_revisions_keys)
 
609
        # Are any inventories for corresponding to the new revisions missing?
 
610
        corresponding_invs = set(inv_parent_map)
 
611
        missing_corresponding = set(new_revisions_keys)
 
612
        missing_corresponding.difference_update(corresponding_invs)
 
613
        if missing_corresponding:
 
614
            problems.append("inventories missing for revisions %s" %
 
615
                (sorted(missing_corresponding),))
 
616
            return problems
 
617
        # Are any chk root entries missing for any inventories?  This includes
 
618
        # any present parent inventories, which may be used when calculating
 
619
        # deltas for streaming.
 
620
        all_inv_keys = set(corresponding_invs)
 
621
        for parent_inv_keys in inv_parent_map.itervalues():
 
622
            all_inv_keys.update(parent_inv_keys)
 
623
        # Filter out ghost parents.
 
624
        all_inv_keys.intersection_update(
 
625
            no_fallback_inv_index.get_parent_map(all_inv_keys))
 
626
        parent_invs_only_keys = all_inv_keys.symmetric_difference(
 
627
            corresponding_invs)
 
628
        all_missing = set()
 
629
        inv_ids = [key[-1] for key in all_inv_keys]
 
630
        parent_invs_only_ids = [key[-1] for key in parent_invs_only_keys]
 
631
        root_key_info = _build_interesting_key_sets(
 
632
            self.repo, inv_ids, parent_invs_only_ids)
 
633
        expected_chk_roots = root_key_info.all_keys()
 
634
        present_chk_roots = no_fallback_chk_bytes_index.get_parent_map(
 
635
            expected_chk_roots)
 
636
        missing_chk_roots = expected_chk_roots.difference(present_chk_roots)
 
637
        if missing_chk_roots:
 
638
            problems.append("missing referenced chk root keys: %s"
 
639
                % (sorted(missing_chk_roots),))
 
640
            # Don't bother checking any further.
 
641
            return problems
 
642
        # Find all interesting chk_bytes records, and make sure they are
 
643
        # present, as well as the text keys they reference.
 
644
        chk_bytes_no_fallbacks = self.repo.chk_bytes.without_fallbacks()
 
645
        chk_bytes_no_fallbacks._search_key_func = \
 
646
            self.repo.chk_bytes._search_key_func
 
647
        chk_diff = chk_map.iter_interesting_nodes(
 
648
            chk_bytes_no_fallbacks, root_key_info.interesting_root_keys,
 
649
            root_key_info.uninteresting_root_keys)
 
650
        bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
 
651
        text_keys = set()
 
652
        try:
 
653
            for record in _filter_text_keys(chk_diff, text_keys, bytes_to_info):
 
654
                pass
 
655
        except errors.NoSuchRevision, e:
 
656
            # XXX: It would be nice if we could give a more precise error here.
 
657
            problems.append("missing chk node(s) for id_to_entry maps")
 
658
        chk_diff = chk_map.iter_interesting_nodes(
 
659
            chk_bytes_no_fallbacks, root_key_info.interesting_pid_root_keys,
 
660
            root_key_info.uninteresting_pid_root_keys)
 
661
        try:
 
662
            for interesting_rec, interesting_map in chk_diff:
 
663
                pass
 
664
        except errors.NoSuchRevision, e:
 
665
            problems.append(
 
666
                "missing chk node(s) for parent_id_basename_to_file_id maps")
 
667
        present_text_keys = no_fallback_texts_index.get_parent_map(text_keys)
 
668
        missing_text_keys = text_keys.difference(present_text_keys)
 
669
        if missing_text_keys:
 
670
            problems.append("missing text keys: %r"
 
671
                % (sorted(missing_text_keys),))
 
672
        return problems
 
673
 
565
674
    def _execute_pack_operations(self, pack_operations,
566
675
                                 _packer_class=GCCHKPacker,
567
676
                                 reload_func=None):
580
689
            packer = GCCHKPacker(self, packs, '.autopack',
581
690
                                 reload_func=reload_func)
582
691
            try:
583
 
                packer.pack()
 
692
                result = packer.pack()
584
693
            except errors.RetryWithNewPacks:
585
694
                # An exception is propagating out of this context, make sure
586
695
                # this packer has cleaned up. Packer() doesn't set its new_pack
589
698
                if packer.new_pack is not None:
590
699
                    packer.new_pack.abort()
591
700
                raise
 
701
            if result is None:
 
702
                return
592
703
            for pack in packs:
593
704
                self._remove_pack_from_memory(pack)
594
705
        # record the newly available packs and stop advertising the old
595
706
        # packs
596
 
        self._save_pack_names(clear_obsolete_packs=True)
597
 
        # Move the old packs out of the way now they are no longer referenced.
598
 
        for revision_count, packs in pack_operations:
599
 
            self._obsolete_packs(packs)
 
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
600
713
 
601
714
 
602
715
class CHKInventoryRepository(KnitPackRepository):
627
740
            _GCGraphIndex(self._pack_collection.revision_index.combined_index,
628
741
                add_callback=self._pack_collection.revision_index.add_callback,
629
742
                parents=True, is_locked=self.is_locked,
630
 
                track_external_parent_refs=True),
 
743
                track_external_parent_refs=True, track_new_keys=True),
631
744
            access=self._pack_collection.revision_index.data_access,
632
745
            delta=False)
633
746
        self.signatures = GroupCompressVersionedFiles(
766
879
            if basis_tree is not None:
767
880
                basis_tree.unlock()
768
881
 
769
 
    def _iter_inventories(self, revision_ids):
 
882
    def deserialise_inventory(self, revision_id, bytes):
 
883
        return inventory.CHKInventory.deserialise(self.chk_bytes, bytes,
 
884
            (revision_id,))
 
885
 
 
886
    def _iter_inventories(self, revision_ids, ordering):
770
887
        """Iterate over many inventory objects."""
 
888
        if ordering is None:
 
889
            ordering = 'unordered'
771
890
        keys = [(revision_id,) for revision_id in revision_ids]
772
 
        stream = self.inventories.get_record_stream(keys, 'unordered', True)
 
891
        stream = self.inventories.get_record_stream(keys, ordering, True)
773
892
        texts = {}
774
893
        for record in stream:
775
894
            if record.storage_kind != 'absent':
779
898
        for key in keys:
780
899
            yield inventory.CHKInventory.deserialise(self.chk_bytes, texts[key], key)
781
900
 
782
 
    def _iter_inventory_xmls(self, revision_ids):
783
 
        # Without a native 'xml' inventory, this method doesn't make sense, so
784
 
        # make it raise to trap naughty direct users.
785
 
        raise NotImplementedError(self._iter_inventory_xmls)
 
901
    def _iter_inventory_xmls(self, revision_ids, ordering):
 
902
        # Without a native 'xml' inventory, this method doesn't make sense.
 
903
        # However older working trees, and older bundles want it - so we supply
 
904
        # it allowing get_inventory_xml to work. Bundles currently use the
 
905
        # serializer directly; this also isn't ideal, but there isn't an xml
 
906
        # iteration interface offered at all for repositories. We could make
 
907
        # _iter_inventory_xmls be part of the contract, even if kept private.
 
908
        inv_to_str = self._serializer.write_inventory_to_string
 
909
        for inv in self.iter_inventories(revision_ids, ordering=ordering):
 
910
            yield inv_to_str(inv), inv.revision_id
786
911
 
787
912
    def _find_present_inventory_keys(self, revision_keys):
788
913
        parent_map = self.inventories.get_parent_map(revision_keys)
814
939
                                        parent_keys)
815
940
            present_parent_inv_ids = set(
816
941
                [k[-1] for k in present_parent_inv_keys])
817
 
            uninteresting_root_keys = set()
818
 
            interesting_root_keys = set()
819
942
            inventories_to_read = set(revision_ids)
820
943
            inventories_to_read.update(present_parent_inv_ids)
821
 
            for inv in self.iter_inventories(inventories_to_read):
822
 
                entry_chk_root_key = inv.id_to_entry.key()
823
 
                if inv.revision_id in present_parent_inv_ids:
824
 
                    uninteresting_root_keys.add(entry_chk_root_key)
825
 
                else:
826
 
                    interesting_root_keys.add(entry_chk_root_key)
827
 
 
 
944
            root_key_info = _build_interesting_key_sets(
 
945
                self, inventories_to_read, present_parent_inv_ids)
 
946
            interesting_root_keys = root_key_info.interesting_root_keys
 
947
            uninteresting_root_keys = root_key_info.uninteresting_root_keys
828
948
            chk_bytes = self.chk_bytes
829
949
            for record, items in chk_map.iter_interesting_nodes(chk_bytes,
830
950
                        interesting_root_keys, uninteresting_root_keys,
879
999
 
880
1000
    def _get_source(self, to_format):
881
1001
        """Return a source for streaming from this repository."""
882
 
        if isinstance(to_format, remote.RemoteRepositoryFormat):
883
 
            # Can't just check attributes on to_format with the current code,
884
 
            # work around this:
885
 
            to_format._ensure_real()
886
 
            to_format = to_format._custom_format
887
 
        if to_format.__class__ is self._format.__class__:
 
1002
        if self._format._serializer == to_format._serializer:
888
1003
            # We must be exactly the same format, otherwise stuff like the chk
889
 
            # page layout might be different
 
1004
            # page layout might be different.
 
1005
            # Actually, this test is just slightly looser than exact so that
 
1006
            # CHK2 <-> 2a transfers will work.
890
1007
            return GroupCHKStreamSource(self, to_format)
891
1008
        return super(CHKInventoryRepository, self)._get_source(to_format)
892
1009
 
967
1084
        bytes_to_info = inventory.CHKInventory._bytes_to_utf8name_key
968
1085
        chk_bytes = self.from_repository.chk_bytes
969
1086
        def _filter_id_to_entry():
970
 
            for record, items in chk_map.iter_interesting_nodes(chk_bytes,
971
 
                        self._chk_id_roots, uninteresting_root_keys):
972
 
                for name, bytes in items:
973
 
                    # Note: we don't care about name_utf8, because we are always
974
 
                    # rich-root = True
975
 
                    _, file_id, revision_id = bytes_to_info(bytes)
976
 
                    self._text_keys.add((file_id, revision_id))
 
1087
            interesting_nodes = chk_map.iter_interesting_nodes(chk_bytes,
 
1088
                        self._chk_id_roots, uninteresting_root_keys)
 
1089
            for record in _filter_text_keys(interesting_nodes, self._text_keys,
 
1090
                    bytes_to_info):
977
1091
                if record is not None:
978
1092
                    yield record
979
1093
            # Consumed
1017
1131
            missing_inventory_keys.add(key[1:])
1018
1132
        if self._chk_id_roots or self._chk_p_id_roots:
1019
1133
            raise AssertionError('Cannot call get_stream_for_missing_keys'
1020
 
                ' untill all of get_stream() has been consumed.')
 
1134
                ' until all of get_stream() has been consumed.')
1021
1135
        # Yield the inventory stream, so we can find the chk stream
1022
1136
        # Some of the missing_keys will be missing because they are ghosts.
1023
1137
        # As such, we can ignore them. The Sink is required to verify there are
1030
1144
            yield stream_info
1031
1145
 
1032
1146
 
 
1147
class _InterestingKeyInfo(object):
 
1148
    def __init__(self):
 
1149
        self.interesting_root_keys = set()
 
1150
        self.interesting_pid_root_keys = set()
 
1151
        self.uninteresting_root_keys = set()
 
1152
        self.uninteresting_pid_root_keys = set()
 
1153
 
 
1154
    def all_interesting(self):
 
1155
        return self.interesting_root_keys.union(self.interesting_pid_root_keys)
 
1156
 
 
1157
    def all_uninteresting(self):
 
1158
        return self.uninteresting_root_keys.union(
 
1159
            self.uninteresting_pid_root_keys)
 
1160
 
 
1161
    def all_keys(self):
 
1162
        return self.all_interesting().union(self.all_uninteresting())
 
1163
 
 
1164
 
 
1165
def _build_interesting_key_sets(repo, inventory_ids, parent_only_inv_ids):
 
1166
    result = _InterestingKeyInfo()
 
1167
    for inv in repo.iter_inventories(inventory_ids, 'unordered'):
 
1168
        root_key = inv.id_to_entry.key()
 
1169
        pid_root_key = inv.parent_id_basename_to_file_id.key()
 
1170
        if inv.revision_id in parent_only_inv_ids:
 
1171
            result.uninteresting_root_keys.add(root_key)
 
1172
            result.uninteresting_pid_root_keys.add(pid_root_key)
 
1173
        else:
 
1174
            result.interesting_root_keys.add(root_key)
 
1175
            result.interesting_pid_root_keys.add(pid_root_key)
 
1176
    return result
 
1177
 
 
1178
 
 
1179
def _filter_text_keys(interesting_nodes_iterable, text_keys, bytes_to_info):
 
1180
    """Iterate the result of iter_interesting_nodes, yielding the records
 
1181
    and adding to text_keys.
 
1182
    """
 
1183
    for record, items in interesting_nodes_iterable:
 
1184
        for name, bytes in items:
 
1185
            # Note: we don't care about name_utf8, because groupcompress repos
 
1186
            # are always rich-root, so there are no synthesised root records to
 
1187
            # ignore.
 
1188
            _, file_id, revision_id = bytes_to_info(bytes)
 
1189
            text_keys.add((file_id, revision_id))
 
1190
        yield record
 
1191
 
 
1192
 
 
1193
 
 
1194
 
1033
1195
class RepositoryFormatCHK1(RepositoryFormatPack):
1034
1196
    """A hashed CHK+group compress pack repository."""
1035
1197
 
1074
1236
        return ("Development repository format - rich roots, group compression"
1075
1237
            " and chk inventories")
1076
1238
 
1077
 
    def check_conversion_target(self, target_format):
1078
 
        if not target_format.rich_root_data:
1079
 
            raise errors.BadConversionTarget(
1080
 
                'Does not support rich root data.', target_format)
1081
 
        if (self.supports_tree_reference and 
1082
 
            not getattr(target_format, 'supports_tree_reference', False)):
1083
 
            raise errors.BadConversionTarget(
1084
 
                'Does not support nested trees', target_format)
1085
 
 
1086
 
 
1087
1239
 
1088
1240
class RepositoryFormatCHK2(RepositoryFormatCHK1):
1089
1241
    """A CHK repository that uses the bencode revision serializer."""
1106
1258
 
1107
1259
class RepositoryFormat2a(RepositoryFormatCHK2):
1108
1260
    """A CHK repository that uses the bencode revision serializer.
1109
 
    
 
1261
 
1110
1262
    This is the same as RepositoryFormatCHK2 but with a public name.
1111
1263
    """
1112
1264
 
1122
1274
 
1123
1275
    def get_format_string(self):
1124
1276
        return ('Bazaar repository format 2a (needs bzr 1.16 or later)\n')
 
1277
 
 
1278
    def get_format_description(self):
 
1279
        """See RepositoryFormat.get_format_description()."""
 
1280
        return ("Repository format 2a - rich roots, group compression"
 
1281
            " and chk inventories")