~bzr-pqm/bzr/bzr.dev

« back to all changes in this revision

Viewing changes to bzrlib/repofmt/groupcompress_repo.py

  • Committer: Andrew Bennetts
  • Date: 2010-08-05 04:21:03 UTC
  • mto: This revision was merged to the branch mainline in revision 5367.
  • Revision ID: andrew.bennetts@canonical.com-20100805042103-pdc01ltvca2nu7bl
Fix test failures for bundles and upgrades.

Show diffs side-by-side

added added

removed removed

Lines of Context:
32
32
    revision as _mod_revision,
33
33
    trace,
34
34
    ui,
35
 
    versionedfile,
36
35
    )
37
36
from bzrlib.btree_index import (
38
37
    BTreeGraphIndex,
39
38
    BTreeBuilder,
40
39
    )
41
 
from bzrlib.decorators import needs_write_lock
42
40
from bzrlib.groupcompress import (
43
41
    _GCGraphIndex,
44
42
    GroupCompressVersionedFiles,
427
425
        self._copy_stream(source_vf, target_vf, inventory_keys,
428
426
                          'inventories', self._get_filtered_inv_stream, 2)
429
427
 
430
 
    def _get_chk_vfs_for_copy(self):
431
 
        return self._build_vfs('chk', False, False)
432
 
 
433
428
    def _copy_chk_texts(self):
434
 
        source_vf, target_vf = self._get_chk_vfs_for_copy()
 
429
        source_vf, target_vf = self._build_vfs('chk', False, False)
435
430
        # TODO: This is technically spurious... if it is a performance issue,
436
431
        #       remove it
437
432
        total_keys = source_vf.keys()
583
578
        return new_pack.data_inserted() and self._data_changed
584
579
 
585
580
 
586
 
class GCCHKCanonicalizingPacker(GCCHKPacker):
587
 
    """A packer that ensures inventories have canonical-form CHK maps.
588
 
    
589
 
    Ideally this would be part of reconcile, but it's very slow and rarely
590
 
    needed.  (It repairs repositories affected by
591
 
    https://bugs.launchpad.net/bzr/+bug/522637).
592
 
    """
593
 
 
594
 
    def __init__(self, *args, **kwargs):
595
 
        super(GCCHKCanonicalizingPacker, self).__init__(*args, **kwargs)
596
 
        self._data_changed = False
597
 
    
598
 
    def _exhaust_stream(self, source_vf, keys, message, vf_to_stream, pb_offset):
599
 
        """Create and exhaust a stream, but don't insert it.
600
 
        
601
 
        This is useful to get the side-effects of generating a stream.
602
 
        """
603
 
        self.pb.update('scanning %s' % (message,), pb_offset)
604
 
        child_pb = ui.ui_factory.nested_progress_bar()
605
 
        try:
606
 
            list(vf_to_stream(source_vf, keys, message, child_pb))
607
 
        finally:
608
 
            child_pb.finished()
609
 
 
610
 
    def _copy_inventory_texts(self):
611
 
        source_vf, target_vf = self._build_vfs('inventory', True, True)
612
 
        source_chk_vf, target_chk_vf = self._get_chk_vfs_for_copy()
613
 
        inventory_keys = source_vf.keys()
614
 
        # First, copy the existing CHKs on the assumption that most of them
615
 
        # will be correct.  This will save us from having to reinsert (and
616
 
        # recompress) these records later at the cost of perhaps preserving a
617
 
        # few unused CHKs. 
618
 
        # (Iterate but don't insert _get_filtered_inv_stream to populate the
619
 
        # variables needed by GCCHKPacker._copy_chk_texts.)
620
 
        self._exhaust_stream(source_vf, inventory_keys, 'inventories',
621
 
                self._get_filtered_inv_stream, 2)
622
 
        GCCHKPacker._copy_chk_texts(self)
623
 
        # Now copy and fix the inventories, and any regenerated CHKs.
624
 
        def chk_canonicalizing_inv_stream(source_vf, keys, message, pb=None):
625
 
            return self._get_filtered_canonicalizing_inv_stream(
626
 
                source_vf, keys, message, pb, source_chk_vf, target_chk_vf)
627
 
        self._copy_stream(source_vf, target_vf, inventory_keys,
628
 
                          'inventories', chk_canonicalizing_inv_stream, 4)
629
 
 
630
 
    def _copy_chk_texts(self):
631
 
        # No-op; in this class this happens during _copy_inventory_texts.
632
 
        pass
633
 
 
634
 
    def _get_filtered_canonicalizing_inv_stream(self, source_vf, keys, message,
635
 
            pb=None, source_chk_vf=None, target_chk_vf=None):
636
 
        """Filter the texts of inventories, regenerating CHKs to make sure they
637
 
        are canonical.
638
 
        """
639
 
        total_keys = len(keys)
640
 
        target_chk_vf = versionedfile.NoDupeAddLinesDecorator(target_chk_vf)
641
 
        def _filtered_inv_stream():
642
 
            stream = source_vf.get_record_stream(keys, 'groupcompress', True)
643
 
            search_key_name = None
644
 
            for idx, record in enumerate(stream):
645
 
                # Inventories should always be with revisions; assume success.
646
 
                bytes = record.get_bytes_as('fulltext')
647
 
                chk_inv = inventory.CHKInventory.deserialise(
648
 
                    source_chk_vf, bytes, record.key)
649
 
                if pb is not None:
650
 
                    pb.update('inv', idx, total_keys)
651
 
                chk_inv.id_to_entry._ensure_root()
652
 
                if search_key_name is None:
653
 
                    # Find the name corresponding to the search_key_func
654
 
                    search_key_reg = chk_map.search_key_registry
655
 
                    for search_key_name, func in search_key_reg.iteritems():
656
 
                        if func == chk_inv.id_to_entry._search_key_func:
657
 
                            break
658
 
                canonical_inv = inventory.CHKInventory.from_inventory(
659
 
                    target_chk_vf, chk_inv,
660
 
                    maximum_size=chk_inv.id_to_entry._root_node._maximum_size,
661
 
                    search_key_name=search_key_name)
662
 
                if chk_inv.id_to_entry.key() != canonical_inv.id_to_entry.key():
663
 
                    trace.mutter(
664
 
                        'Non-canonical CHK map for id_to_entry of inv: %s '
665
 
                        '(root is %s, should be %s)' % (chk_inv.revision_id,
666
 
                        chk_inv.id_to_entry.key()[0],
667
 
                        canonical_inv.id_to_entry.key()[0]))
668
 
                    self._data_changed = True
669
 
                p_id_map = chk_inv.parent_id_basename_to_file_id
670
 
                p_id_map._ensure_root()
671
 
                canon_p_id_map = canonical_inv.parent_id_basename_to_file_id
672
 
                if p_id_map.key() != canon_p_id_map.key():
673
 
                    trace.mutter(
674
 
                        'Non-canonical CHK map for parent_id_to_basename of '
675
 
                        'inv: %s (root is %s, should be %s)'
676
 
                        % (chk_inv.revision_id, p_id_map.key()[0],
677
 
                           canon_p_id_map.key()[0]))
678
 
                    self._data_changed = True
679
 
                yield versionedfile.ChunkedContentFactory(record.key,
680
 
                        record.parents, record.sha1,
681
 
                        canonical_inv.to_lines())
682
 
            # We have finished processing all of the inventory records, we
683
 
            # don't need these sets anymore
684
 
        return _filtered_inv_stream()
685
 
 
686
 
    def _use_pack(self, new_pack):
687
 
        """Override _use_pack to check for reconcile having changed content."""
688
 
        return new_pack.data_inserted() and self._data_changed
689
 
 
690
 
 
691
581
class GCRepositoryPackCollection(RepositoryPackCollection):
692
582
 
693
583
    pack_factory = GCPack
1109
999
        finally:
1110
1000
            pb.finished()
1111
1001
 
1112
 
    @needs_write_lock
1113
 
    def reconcile_canonicalize_chks(self):
1114
 
        """Reconcile this repository to make sure all CHKs are in canonical
1115
 
        form.
1116
 
        """
1117
 
        from bzrlib.reconcile import PackReconciler
1118
 
        reconciler = PackReconciler(self, thorough=True, canonicalize_chks=True)
1119
 
        reconciler.reconcile()
1120
 
        return reconciler
1121
 
 
1122
1002
    def _reconcile_pack(self, collection, packs, extension, revs, pb):
1123
1003
        packer = GCCHKReconcilePacker(collection, packs, extension)
1124
1004
        return packer.pack(pb)
1125
1005
 
1126
 
    def _canonicalize_chks_pack(self, collection, packs, extension, revs, pb):
1127
 
        packer = GCCHKCanonicalizingPacker(collection, packs, extension, revs)
1128
 
        return packer.pack(pb)
1129
 
 
1130
1006
    def _get_source(self, to_format):
1131
1007
        """Return a source for streaming from this repository."""
1132
1008
        if self._format._serializer == to_format._serializer:
1428
1304
        """See RepositoryFormat.get_format_description()."""
1429
1305
        return ("Repository format 2a - rich roots, group compression"
1430
1306
            " and chk inventories")
1431
 
 
1432
 
 
1433
 
class RepositoryFormat2aSubtree(RepositoryFormat2a):
1434
 
    """A 2a repository format that supports nested trees.
1435
 
 
1436
 
    """
1437
 
 
1438
 
    def _get_matching_bzrdir(self):
1439
 
        return bzrdir.format_registry.make_bzrdir('development-subtree')
1440
 
 
1441
 
    def _ignore_setting_bzrdir(self, format):
1442
 
        pass
1443
 
 
1444
 
    _matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1445
 
 
1446
 
    def get_format_string(self):
1447
 
        return ('Bazaar development format 8\n')
1448
 
 
1449
 
    def get_format_description(self):
1450
 
        """See RepositoryFormat.get_format_description()."""
1451
 
        return ("Development repository format 8 - nested trees, "
1452
 
                "group compression and chk inventories")
1453
 
 
1454
 
    experimental = True
1455
 
    supports_tree_reference = True