/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 bzrlib/repofmt/pack_repo.py

  • Committer: Canonical.com Patch Queue Manager
  • Date: 2009-03-24 17:01:50 UTC
  • mfrom: (4178.3.7 lru_cache_linked_lst)
  • Revision ID: pqm@pqm.ubuntu.com-20090324170150-9wtdpv5w7192zdwy
(jam) Improvements to LRUCache structure, use a double-linked-list

Show diffs side-by-side

added added

removed removed

Lines of Context:
1
 
# Copyright (C) 2007-2010 Canonical Ltd
 
1
# Copyright (C) 2005, 2006, 2007, 2008 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
23
23
import time
24
24
 
25
25
from bzrlib import (
26
 
    chk_map,
27
 
    cleanup,
28
26
    debug,
29
27
    graph,
30
28
    osutils,
37
35
    )
38
36
from bzrlib.index import (
39
37
    CombinedGraphIndex,
 
38
    GraphIndex,
 
39
    GraphIndexBuilder,
40
40
    GraphIndexPrefixAdapter,
 
41
    InMemoryGraphIndex,
41
42
    )
42
43
from bzrlib.knit import (
43
44
    KnitPlainFactory,
52
53
    errors,
53
54
    lockable_files,
54
55
    lockdir,
55
 
    revision as _mod_revision,
 
56
    symbol_versioning,
56
57
    )
57
58
 
58
 
from bzrlib.decorators import needs_write_lock, only_raises
 
59
from bzrlib.decorators import needs_write_lock
59
60
from bzrlib.btree_index import (
60
61
    BTreeGraphIndex,
61
62
    BTreeBuilder,
64
65
    GraphIndex,
65
66
    InMemoryGraphIndex,
66
67
    )
67
 
from bzrlib.lock import LogicalLockResult
68
68
from bzrlib.repofmt.knitrepo import KnitRepository
69
69
from bzrlib.repository import (
70
70
    CommitBuilder,
71
71
    MetaDirRepositoryFormat,
72
72
    RepositoryFormat,
73
 
    RepositoryWriteLockResult,
74
73
    RootCommitBuilder,
75
 
    StreamSource,
76
74
    )
 
75
import bzrlib.revision as _mod_revision
77
76
from bzrlib.trace import (
78
77
    mutter,
79
 
    note,
80
78
    warning,
81
79
    )
82
80
 
133
131
    # A map of index 'type' to the file extension and position in the
134
132
    # index_sizes array.
135
133
    index_definitions = {
136
 
        'chk': ('.cix', 4),
137
134
        'revision': ('.rix', 0),
138
135
        'inventory': ('.iix', 1),
139
136
        'text': ('.tix', 2),
141
138
        }
142
139
 
143
140
    def __init__(self, revision_index, inventory_index, text_index,
144
 
        signature_index, chk_index=None):
 
141
        signature_index):
145
142
        """Create a pack instance.
146
143
 
147
144
        :param revision_index: A GraphIndex for determining what revisions are
154
151
            texts/deltas (via (fileid, revisionid) tuples).
155
152
        :param signature_index: A GraphIndex for determining what signatures are
156
153
            present in the Pack and accessing the locations of their texts.
157
 
        :param chk_index: A GraphIndex for accessing content by CHK, if the
158
 
            pack has one.
159
154
        """
160
155
        self.revision_index = revision_index
161
156
        self.inventory_index = inventory_index
162
157
        self.text_index = text_index
163
158
        self.signature_index = signature_index
164
 
        self.chk_index = chk_index
165
159
 
166
160
    def access_tuple(self):
167
161
        """Return a tuple (transport, name) for the pack content."""
228
222
        return self.index_name('text', name)
229
223
 
230
224
    def _replace_index_with_readonly(self, index_type):
231
 
        unlimited_cache = False
232
 
        if index_type == 'chk':
233
 
            unlimited_cache = True
234
225
        setattr(self, index_type + '_index',
235
226
            self.index_class(self.index_transport,
236
227
                self.index_name(index_type, self.name),
237
 
                self.index_sizes[self.index_offset(index_type)],
238
 
                unlimited_cache=unlimited_cache))
 
228
                self.index_sizes[self.index_offset(index_type)]))
239
229
 
240
230
 
241
231
class ExistingPack(Pack):
242
232
    """An in memory proxy for an existing .pack and its disk indices."""
243
233
 
244
234
    def __init__(self, pack_transport, name, revision_index, inventory_index,
245
 
        text_index, signature_index, chk_index=None):
 
235
        text_index, signature_index):
246
236
        """Create an ExistingPack object.
247
237
 
248
238
        :param pack_transport: The transport where the pack file resides.
249
239
        :param name: The name of the pack on disk in the pack_transport.
250
240
        """
251
241
        Pack.__init__(self, revision_index, inventory_index, text_index,
252
 
            signature_index, chk_index)
 
242
            signature_index)
253
243
        self.name = name
254
244
        self.pack_transport = pack_transport
255
245
        if None in (revision_index, inventory_index, text_index,
272
262
 
273
263
    def __init__(self, name, revision_index, inventory_index, text_index,
274
264
        signature_index, upload_transport, pack_transport, index_transport,
275
 
        pack_collection, chk_index=None):
 
265
        pack_collection):
276
266
        """Create a ResumedPack object."""
277
267
        ExistingPack.__init__(self, pack_transport, name, revision_index,
278
 
            inventory_index, text_index, signature_index,
279
 
            chk_index=chk_index)
 
268
            inventory_index, text_index, signature_index)
280
269
        self.upload_transport = upload_transport
281
270
        self.index_transport = index_transport
282
271
        self.index_sizes = [None, None, None, None]
286
275
            ('text', text_index),
287
276
            ('signature', signature_index),
288
277
            ]
289
 
        if chk_index is not None:
290
 
            indices.append(('chk', chk_index))
291
 
            self.index_sizes.append(None)
292
278
        for index_type, index in indices:
293
279
            offset = self.index_offset(index_type)
294
280
            self.index_sizes[offset] = index._size
309
295
        self.upload_transport.delete(self.file_name())
310
296
        indices = [self.revision_index, self.inventory_index, self.text_index,
311
297
            self.signature_index]
312
 
        if self.chk_index is not None:
313
 
            indices.append(self.chk_index)
314
298
        for index in indices:
315
299
            index._transport.delete(index._name)
316
300
 
317
301
    def finish(self):
318
302
        self._check_references()
319
 
        index_types = ['revision', 'inventory', 'text', 'signature']
320
 
        if self.chk_index is not None:
321
 
            index_types.append('chk')
322
 
        for index_type in index_types:
 
303
        new_name = '../packs/' + self.file_name()
 
304
        self.upload_transport.rename(self.file_name(), new_name)
 
305
        for index_type in ['revision', 'inventory', 'text', 'signature']:
323
306
            old_name = self.index_name(index_type, self.name)
324
307
            new_name = '../indices/' + old_name
325
308
            self.upload_transport.rename(old_name, new_name)
326
309
            self._replace_index_with_readonly(index_type)
327
 
        new_name = '../packs/' + self.file_name()
328
 
        self.upload_transport.rename(self.file_name(), new_name)
329
310
        self._state = 'finished'
330
311
 
331
312
    def _get_external_refs(self, index):
332
 
        """Return compression parents for this index that are not present.
333
 
 
334
 
        This returns any compression parents that are referenced by this index,
335
 
        which are not contained *in* this index. They may be present elsewhere.
336
 
        """
337
313
        return index.external_references(1)
338
314
 
339
315
 
351
327
        # The relative locations of the packs are constrained, but all are
352
328
        # passed in because the caller has them, so as to avoid object churn.
353
329
        index_builder_class = pack_collection._index_builder_class
354
 
        if pack_collection.chk_index is not None:
355
 
            chk_index = index_builder_class(reference_lists=0)
356
 
        else:
357
 
            chk_index = None
358
330
        Pack.__init__(self,
359
331
            # Revisions: parents list, no text compression.
360
332
            index_builder_class(reference_lists=1),
369
341
            # Signatures: Just blobs to store, no compression, no parents
370
342
            # listing.
371
343
            index_builder_class(reference_lists=0),
372
 
            # CHK based storage - just blobs, no compression or parents.
373
 
            chk_index=chk_index
374
344
            )
375
345
        self._pack_collection = pack_collection
376
346
        # When we make readonly indices, we need this.
385
355
        self._file_mode = file_mode
386
356
        # tracks the content written to the .pack file.
387
357
        self._hash = osutils.md5()
388
 
        # a tuple with the length in bytes of the indices, once the pack
389
 
        # is finalised. (rev, inv, text, sigs, chk_if_in_use)
 
358
        # a four-tuple with the length in bytes of the indices, once the pack
 
359
        # is finalised. (rev, inv, text, sigs)
390
360
        self.index_sizes = None
391
361
        # How much data to cache when writing packs. Note that this is not
392
362
        # synchronised with reads, because it's not in the transport layer, so
430
400
        self._writer.begin()
431
401
        # what state is the pack in? (open, finished, aborted)
432
402
        self._state = 'open'
433
 
        # no name until we finish writing the content
434
 
        self.name = None
435
403
 
436
404
    def abort(self):
437
405
        """Cancel creating this pack."""
455
423
        return bool(self.get_revision_count() or
456
424
            self.inventory_index.key_count() or
457
425
            self.text_index.key_count() or
458
 
            self.signature_index.key_count() or
459
 
            (self.chk_index is not None and self.chk_index.key_count()))
460
 
 
461
 
    def finish_content(self):
462
 
        if self.name is not None:
463
 
            return
464
 
        self._writer.end()
465
 
        if self._buffer[1]:
466
 
            self._write_data('', flush=True)
467
 
        self.name = self._hash.hexdigest()
 
426
            self.signature_index.key_count())
468
427
 
469
428
    def finish(self, suspend=False):
470
429
        """Finish the new pack.
477
436
         - stores the index size tuple for the pack in the index_sizes
478
437
           attribute.
479
438
        """
480
 
        self.finish_content()
 
439
        self._writer.end()
 
440
        if self._buffer[1]:
 
441
            self._write_data('', flush=True)
 
442
        self.name = self._hash.hexdigest()
481
443
        if not suspend:
482
444
            self._check_references()
483
445
        # write indices
492
454
        self._write_index('text', self.text_index, 'file texts', suspend)
493
455
        self._write_index('signature', self.signature_index,
494
456
            'revision signatures', suspend)
495
 
        if self.chk_index is not None:
496
 
            self.index_sizes.append(None)
497
 
            self._write_index('chk', self.chk_index,
498
 
                'content hash bytes', suspend)
499
457
        self.write_stream.close()
500
458
        # Note that this will clobber an existing pack with the same name,
501
459
        # without checking for hash collisions. While this is undesirable this
574
532
    # XXX: Probably 'can be written to' could/should be separated from 'acts
575
533
    # like a knit index' -- mbp 20071024
576
534
 
577
 
    def __init__(self, reload_func=None, flush_func=None):
 
535
    def __init__(self, reload_func=None):
578
536
        """Create an AggregateIndex.
579
537
 
580
538
        :param reload_func: A function to call if we find we are missing an
585
543
        self.index_to_pack = {}
586
544
        self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
587
545
        self.data_access = _DirectPackAccess(self.index_to_pack,
588
 
                                             reload_func=reload_func,
589
 
                                             flush_func=flush_func)
 
546
                                             reload_func=reload_func)
 
547
        self.add_callback = None
 
548
 
 
549
    def replace_indices(self, index_to_pack, indices):
 
550
        """Replace the current mappings with fresh ones.
 
551
 
 
552
        This should probably not be used eventually, rather incremental add and
 
553
        removal of indices. It has been added during refactoring of existing
 
554
        code.
 
555
 
 
556
        :param index_to_pack: A mapping from index objects to
 
557
            (transport, name) tuples for the pack file data.
 
558
        :param indices: A list of indices.
 
559
        """
 
560
        # refresh the revision pack map dict without replacing the instance.
 
561
        self.index_to_pack.clear()
 
562
        self.index_to_pack.update(index_to_pack)
 
563
        # XXX: API break - clearly a 'replace' method would be good?
 
564
        self.combined_index._indices[:] = indices
 
565
        # the current add nodes callback for the current writable index if
 
566
        # there is one.
590
567
        self.add_callback = None
591
568
 
592
569
    def add_index(self, index, pack):
601
578
        # expose it to the index map
602
579
        self.index_to_pack[index] = pack.access_tuple()
603
580
        # put it at the front of the linear index list
604
 
        self.combined_index.insert_index(0, index, pack.name)
 
581
        self.combined_index.insert_index(0, index)
605
582
 
606
583
    def add_writable_index(self, index, pack):
607
584
        """Add an index which is able to have data added to it.
627
604
        self.data_access.set_writer(None, None, (None, None))
628
605
        self.index_to_pack.clear()
629
606
        del self.combined_index._indices[:]
630
 
        del self.combined_index._index_names[:]
631
607
        self.add_callback = None
632
608
 
633
 
    def remove_index(self, index):
 
609
    def remove_index(self, index, pack):
634
610
        """Remove index from the indices used to answer queries.
635
611
 
636
612
        :param index: An index from the pack parameter.
 
613
        :param pack: A Pack instance.
637
614
        """
638
615
        del self.index_to_pack[index]
639
 
        pos = self.combined_index._indices.index(index)
640
 
        del self.combined_index._indices[pos]
641
 
        del self.combined_index._index_names[pos]
 
616
        self.combined_index._indices.remove(index)
642
617
        if (self.add_callback is not None and
643
618
            getattr(index, 'add_nodes', None) == self.add_callback):
644
619
            self.add_callback = None
750
725
 
751
726
    def open_pack(self):
752
727
        """Open a pack for the pack we are creating."""
753
 
        new_pack = self._pack_collection.pack_factory(self._pack_collection,
754
 
                upload_suffix=self.suffix,
 
728
        new_pack = NewPack(self._pack_collection, upload_suffix=self.suffix,
755
729
                file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
756
730
        # We know that we will process all nodes in order, and don't need to
757
731
        # query, so don't combine any indices spilled to disk until we are done
922
896
                time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
923
897
                new_pack.signature_index.key_count(),
924
898
                time.time() - new_pack.start_time)
925
 
        # copy chk contents
926
 
        # NB XXX: how to check CHK references are present? perhaps by yielding
927
 
        # the items? How should that interact with stacked repos?
928
 
        if new_pack.chk_index is not None:
929
 
            self._copy_chks()
930
 
            if 'pack' in debug.debug_flags:
931
 
                mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
932
 
                    time.ctime(), self._pack_collection._upload_transport.base,
933
 
                    new_pack.random_name,
934
 
                    new_pack.chk_index.key_count(),
935
 
                    time.time() - new_pack.start_time)
936
899
        new_pack._check_references()
937
900
        if not self._use_pack(new_pack):
938
901
            new_pack.abort()
942
905
        self._pack_collection.allocate(new_pack)
943
906
        return new_pack
944
907
 
945
 
    def _copy_chks(self, refs=None):
946
 
        # XXX: Todo, recursive follow-pointers facility when fetching some
947
 
        # revisions only.
948
 
        chk_index_map, chk_indices = self._pack_map_and_index_list(
949
 
            'chk_index')
950
 
        chk_nodes = self._index_contents(chk_indices, refs)
951
 
        new_refs = set()
952
 
        # TODO: This isn't strictly tasteful as we are accessing some private
953
 
        #       variables (_serializer). Perhaps a better way would be to have
954
 
        #       Repository._deserialise_chk_node()
955
 
        search_key_func = chk_map.search_key_registry.get(
956
 
            self._pack_collection.repo._serializer.search_key_name)
957
 
        def accumlate_refs(lines):
958
 
            # XXX: move to a generic location
959
 
            # Yay mismatch:
960
 
            bytes = ''.join(lines)
961
 
            node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
962
 
            new_refs.update(node.refs())
963
 
        self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
964
 
            self.new_pack.chk_index, output_lines=accumlate_refs)
965
 
        return new_refs
966
 
 
967
 
    def _copy_nodes(self, nodes, index_map, writer, write_index,
968
 
        output_lines=None):
969
 
        """Copy knit nodes between packs with no graph references.
970
 
 
971
 
        :param output_lines: Output full texts of copied items.
972
 
        """
 
908
    def _copy_nodes(self, nodes, index_map, writer, write_index):
 
909
        """Copy knit nodes between packs with no graph references."""
973
910
        pb = ui.ui_factory.nested_progress_bar()
974
911
        try:
975
912
            return self._do_copy_nodes(nodes, index_map, writer,
976
 
                write_index, pb, output_lines=output_lines)
 
913
                write_index, pb)
977
914
        finally:
978
915
            pb.finished()
979
916
 
980
 
    def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
981
 
        output_lines=None):
 
917
    def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
982
918
        # for record verification
983
919
        knit = KnitVersionedFiles(None, None)
984
920
        # plan a readv on each source pack:
1018
954
                izip(reader.iter_records(), pack_readv_requests):
1019
955
                raw_data = read_func(None)
1020
956
                # check the header only
1021
 
                if output_lines is not None:
1022
 
                    output_lines(knit._parse_record(key[-1], raw_data)[0])
1023
 
                else:
1024
 
                    df, _ = knit._parse_record_header(key, raw_data)
1025
 
                    df.close()
 
957
                df, _ = knit._parse_record_header(key, raw_data)
 
958
                df.close()
1026
959
                pos, size = writer.add_bytes_record(raw_data, names)
1027
960
                write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1028
961
                pb.update("Copied record", record_index)
1102
1035
            iterator is a tuple with:
1103
1036
            index, readv_vector, node_vector. readv_vector is a list ready to
1104
1037
            hand to the transport readv method, and node_vector is a list of
1105
 
            (key, eol_flag, references) for the node retrieved by the
 
1038
            (key, eol_flag, references) for the the node retrieved by the
1106
1039
            matching readv_vector.
1107
1040
        """
1108
1041
        # group by pack so we do one readv per pack
1299
1232
        # space (we only topo sort the revisions, which is smaller).
1300
1233
        topo_order = tsort.topo_sort(ancestors)
1301
1234
        rev_order = dict(zip(topo_order, range(len(topo_order))))
1302
 
        bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
 
1235
        bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1303
1236
        transaction = repo.get_transaction()
1304
1237
        file_id_index = GraphIndexPrefixAdapter(
1305
1238
            self.new_pack.text_index,
1358
1291
    :ivar _names: map of {pack_name: (index_size,)}
1359
1292
    """
1360
1293
 
1361
 
    pack_factory = NewPack
1362
 
    resumed_pack_factory = ResumedPack
1363
 
 
1364
1294
    def __init__(self, repo, transport, index_transport, upload_transport,
1365
 
                 pack_transport, index_builder_class, index_class,
1366
 
                 use_chk_index):
 
1295
                 pack_transport, index_builder_class, index_class):
1367
1296
        """Create a new RepositoryPackCollection.
1368
1297
 
1369
1298
        :param transport: Addresses the repository base directory
1374
1303
        :param pack_transport: Addresses the directory of existing complete packs.
1375
1304
        :param index_builder_class: The index builder class to use.
1376
1305
        :param index_class: The index class to use.
1377
 
        :param use_chk_index: Whether to setup and manage a CHK index.
1378
1306
        """
1379
1307
        # XXX: This should call self.reset()
1380
1308
        self.repo = repo
1384
1312
        self._pack_transport = pack_transport
1385
1313
        self._index_builder_class = index_builder_class
1386
1314
        self._index_class = index_class
1387
 
        self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
1388
 
            '.cix': 4}
 
1315
        self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1389
1316
        self.packs = []
1390
1317
        # name:Pack mapping
1391
1318
        self._names = None
1395
1322
        # when a pack is being created by this object, the state of that pack.
1396
1323
        self._new_pack = None
1397
1324
        # aggregated revision index data
1398
 
        flush = self._flush_new_pack
1399
 
        self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1400
 
        self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1401
 
        self.text_index = AggregateIndex(self.reload_pack_names, flush)
1402
 
        self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1403
 
        all_indices = [self.revision_index, self.inventory_index,
1404
 
                self.text_index, self.signature_index]
1405
 
        if use_chk_index:
1406
 
            self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1407
 
            all_indices.append(self.chk_index)
1408
 
        else:
1409
 
            # used to determine if we're using a chk_index elsewhere.
1410
 
            self.chk_index = None
1411
 
        # Tell all the CombinedGraphIndex objects about each other, so they can
1412
 
        # share hints about which pack names to search first.
1413
 
        all_combined = [agg_idx.combined_index for agg_idx in all_indices]
1414
 
        for combined_idx in all_combined:
1415
 
            combined_idx.set_sibling_indices(
1416
 
                set(all_combined).difference([combined_idx]))
 
1325
        self.revision_index = AggregateIndex(self.reload_pack_names)
 
1326
        self.inventory_index = AggregateIndex(self.reload_pack_names)
 
1327
        self.text_index = AggregateIndex(self.reload_pack_names)
 
1328
        self.signature_index = AggregateIndex(self.reload_pack_names)
1417
1329
        # resumed packs
1418
1330
        self._resumed_packs = []
1419
1331
 
1420
 
    def __repr__(self):
1421
 
        return '%s(%r)' % (self.__class__.__name__, self.repo)
1422
 
 
1423
1332
    def add_pack_to_memory(self, pack):
1424
1333
        """Make a Pack object available to the repository to satisfy queries.
1425
1334
 
1434
1343
        self.inventory_index.add_index(pack.inventory_index, pack)
1435
1344
        self.text_index.add_index(pack.text_index, pack)
1436
1345
        self.signature_index.add_index(pack.signature_index, pack)
1437
 
        if self.chk_index is not None:
1438
 
            self.chk_index.add_index(pack.chk_index, pack)
1439
1346
 
1440
1347
    def all_packs(self):
1441
1348
        """Return a list of all the Pack objects this repository has.
1463
1370
        in synchronisation with certain steps. Otherwise the names collection
1464
1371
        is not flushed.
1465
1372
 
1466
 
        :return: Something evaluating true if packing took place.
 
1373
        :return: True if packing took place.
1467
1374
        """
1468
1375
        while True:
1469
1376
            try:
1470
1377
                return self._do_autopack()
1471
 
            except errors.RetryAutopack:
 
1378
            except errors.RetryAutopack, e:
1472
1379
                # If we get a RetryAutopack exception, we should abort the
1473
1380
                # current action, and retry.
1474
1381
                pass
1478
1385
        total_revisions = self.revision_index.combined_index.key_count()
1479
1386
        total_packs = len(self._names)
1480
1387
        if self._max_pack_count(total_revisions) >= total_packs:
1481
 
            return None
 
1388
            return False
 
1389
        # XXX: the following may want to be a class, to pack with a given
 
1390
        # policy.
1482
1391
        # determine which packs need changing
1483
1392
        pack_distribution = self.pack_distribution(total_revisions)
1484
1393
        existing_packs = []
1506
1415
            'containing %d revisions. Packing %d files into %d affecting %d'
1507
1416
            ' revisions', self, total_packs, total_revisions, num_old_packs,
1508
1417
            num_new_packs, num_revs_affected)
1509
 
        result = self._execute_pack_operations(pack_operations,
 
1418
        self._execute_pack_operations(pack_operations,
1510
1419
                                      reload_func=self._restart_autopack)
1511
 
        mutter('Auto-packing repository %s completed', self)
1512
 
        return result
 
1420
        return True
1513
1421
 
1514
1422
    def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1515
1423
                                 reload_func=None):
1517
1425
 
1518
1426
        :param pack_operations: A list of [revision_count, packs_to_combine].
1519
1427
        :param _packer_class: The class of packer to use (default: Packer).
1520
 
        :return: The new pack names.
 
1428
        :return: None.
1521
1429
        """
1522
1430
        for revision_count, packs in pack_operations:
1523
1431
            # we may have no-ops from the setup logic
1539
1447
                self._remove_pack_from_memory(pack)
1540
1448
        # record the newly available packs and stop advertising the old
1541
1449
        # packs
1542
 
        to_be_obsoleted = []
1543
 
        for _, packs in pack_operations:
1544
 
            to_be_obsoleted.extend(packs)
1545
 
        result = self._save_pack_names(clear_obsolete_packs=True,
1546
 
                                       obsolete_packs=to_be_obsoleted)
1547
 
        return result
1548
 
 
1549
 
    def _flush_new_pack(self):
1550
 
        if self._new_pack is not None:
1551
 
            self._new_pack.flush()
 
1450
        self._save_pack_names(clear_obsolete_packs=True)
 
1451
        # Move the old packs out of the way now they are no longer referenced.
 
1452
        for revision_count, packs in pack_operations:
 
1453
            self._obsolete_packs(packs)
1552
1454
 
1553
1455
    def lock_names(self):
1554
1456
        """Acquire the mutex around the pack-names index.
1558
1460
        """
1559
1461
        self.repo.control_files.lock_write()
1560
1462
 
1561
 
    def _already_packed(self):
1562
 
        """Is the collection already packed?"""
1563
 
        return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1564
 
 
1565
 
    def pack(self, hint=None, clean_obsolete_packs=False):
 
1463
    def pack(self):
1566
1464
        """Pack the pack collection totally."""
1567
1465
        self.ensure_loaded()
1568
1466
        total_packs = len(self._names)
1569
 
        if self._already_packed():
 
1467
        if total_packs < 2:
 
1468
            # This is arguably wrong because we might not be optimal, but for
 
1469
            # now lets leave it in. (e.g. reconcile -> one pack. But not
 
1470
            # optimal.
1570
1471
            return
1571
1472
        total_revisions = self.revision_index.combined_index.key_count()
1572
1473
        # XXX: the following may want to be a class, to pack with a given
1573
1474
        # policy.
1574
1475
        mutter('Packing repository %s, which has %d pack files, '
1575
 
            'containing %d revisions with hint %r.', self, total_packs,
1576
 
            total_revisions, hint)
 
1476
            'containing %d revisions into 1 packs.', self, total_packs,
 
1477
            total_revisions)
1577
1478
        # determine which packs need changing
 
1479
        pack_distribution = [1]
1578
1480
        pack_operations = [[0, []]]
1579
1481
        for pack in self.all_packs():
1580
 
            if hint is None or pack.name in hint:
1581
 
                # Either no hint was provided (so we are packing everything),
1582
 
                # or this pack was included in the hint.
1583
 
                pack_operations[-1][0] += pack.get_revision_count()
1584
 
                pack_operations[-1][1].append(pack)
 
1482
            pack_operations[-1][0] += pack.get_revision_count()
 
1483
            pack_operations[-1][1].append(pack)
1585
1484
        self._execute_pack_operations(pack_operations, OptimisingPacker)
1586
1485
 
1587
 
        if clean_obsolete_packs:
1588
 
            self._clear_obsolete_packs()
1589
 
 
1590
1486
    def plan_autopack_combinations(self, existing_packs, pack_distribution):
1591
1487
        """Plan a pack operation.
1592
1488
 
1679
1575
            inv_index = self._make_index(name, '.iix')
1680
1576
            txt_index = self._make_index(name, '.tix')
1681
1577
            sig_index = self._make_index(name, '.six')
1682
 
            if self.chk_index is not None:
1683
 
                chk_index = self._make_index(name, '.cix', unlimited_cache=True)
1684
 
            else:
1685
 
                chk_index = None
1686
1578
            result = ExistingPack(self._pack_transport, name, rev_index,
1687
 
                inv_index, txt_index, sig_index, chk_index)
 
1579
                inv_index, txt_index, sig_index)
1688
1580
            self.add_pack_to_memory(result)
1689
1581
            return result
1690
1582
 
1704
1596
            inv_index = self._make_index(name, '.iix', resume=True)
1705
1597
            txt_index = self._make_index(name, '.tix', resume=True)
1706
1598
            sig_index = self._make_index(name, '.six', resume=True)
1707
 
            if self.chk_index is not None:
1708
 
                chk_index = self._make_index(name, '.cix', resume=True,
1709
 
                                             unlimited_cache=True)
1710
 
            else:
1711
 
                chk_index = None
1712
 
            result = self.resumed_pack_factory(name, rev_index, inv_index,
1713
 
                txt_index, sig_index, self._upload_transport,
1714
 
                self._pack_transport, self._index_transport, self,
1715
 
                chk_index=chk_index)
 
1599
            result = ResumedPack(name, rev_index, inv_index, txt_index,
 
1600
                sig_index, self._upload_transport, self._pack_transport,
 
1601
                self._index_transport, self)
1716
1602
        except errors.NoSuchFile, e:
1717
1603
            raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1718
1604
        self.add_pack_to_memory(result)
1742
1628
        return self._index_class(self.transport, 'pack-names', None
1743
1629
                ).iter_all_entries()
1744
1630
 
1745
 
    def _make_index(self, name, suffix, resume=False, unlimited_cache=False):
 
1631
    def _make_index(self, name, suffix, resume=False):
1746
1632
        size_offset = self._suffix_offsets[suffix]
1747
1633
        index_name = name + suffix
1748
1634
        if resume:
1751
1637
        else:
1752
1638
            transport = self._index_transport
1753
1639
            index_size = self._names[name][size_offset]
1754
 
        return self._index_class(transport, index_name, index_size,
1755
 
                                 unlimited_cache=unlimited_cache)
 
1640
        return self._index_class(transport, index_name, index_size)
1756
1641
 
1757
1642
    def _max_pack_count(self, total_revisions):
1758
1643
        """Return the maximum number of packs to use for total revisions.
1786
1671
        :param return: None.
1787
1672
        """
1788
1673
        for pack in packs:
1789
 
            try:
1790
 
                pack.pack_transport.rename(pack.file_name(),
1791
 
                    '../obsolete_packs/' + pack.file_name())
1792
 
            except (errors.PathError, errors.TransportError), e:
1793
 
                # TODO: Should these be warnings or mutters?
1794
 
                mutter("couldn't rename obsolete pack, skipping it:\n%s"
1795
 
                       % (e,))
 
1674
            pack.pack_transport.rename(pack.file_name(),
 
1675
                '../obsolete_packs/' + pack.file_name())
1796
1676
            # TODO: Probably needs to know all possible indices for this pack
1797
1677
            # - or maybe list the directory and move all indices matching this
1798
1678
            # name whether we recognize it or not?
1799
 
            suffixes = ['.iix', '.six', '.tix', '.rix']
1800
 
            if self.chk_index is not None:
1801
 
                suffixes.append('.cix')
1802
 
            for suffix in suffixes:
1803
 
                try:
1804
 
                    self._index_transport.rename(pack.name + suffix,
1805
 
                        '../obsolete_packs/' + pack.name + suffix)
1806
 
                except (errors.PathError, errors.TransportError), e:
1807
 
                    mutter("couldn't rename obsolete index, skipping it:\n%s"
1808
 
                           % (e,))
 
1679
            for suffix in ('.iix', '.six', '.tix', '.rix'):
 
1680
                self._index_transport.rename(pack.name + suffix,
 
1681
                    '../obsolete_packs/' + pack.name + suffix)
1809
1682
 
1810
1683
    def pack_distribution(self, total_revisions):
1811
1684
        """Generate a list of the number of revisions to put in each pack.
1837
1710
        self._remove_pack_indices(pack)
1838
1711
        self.packs.remove(pack)
1839
1712
 
1840
 
    def _remove_pack_indices(self, pack, ignore_missing=False):
1841
 
        """Remove the indices for pack from the aggregated indices.
1842
 
        
1843
 
        :param ignore_missing: Suppress KeyErrors from calling remove_index.
1844
 
        """
1845
 
        for index_type in Pack.index_definitions.keys():
1846
 
            attr_name = index_type + '_index'
1847
 
            aggregate_index = getattr(self, attr_name)
1848
 
            if aggregate_index is not None:
1849
 
                pack_index = getattr(pack, attr_name)
1850
 
                try:
1851
 
                    aggregate_index.remove_index(pack_index)
1852
 
                except KeyError:
1853
 
                    if ignore_missing:
1854
 
                        continue
1855
 
                    raise
 
1713
    def _remove_pack_indices(self, pack):
 
1714
        """Remove the indices for pack from the aggregated indices."""
 
1715
        self.revision_index.remove_index(pack.revision_index, pack)
 
1716
        self.inventory_index.remove_index(pack.inventory_index, pack)
 
1717
        self.text_index.remove_index(pack.text_index, pack)
 
1718
        self.signature_index.remove_index(pack.signature_index, pack)
1856
1719
 
1857
1720
    def reset(self):
1858
1721
        """Clear all cached data."""
1859
1722
        # cached revision data
 
1723
        self.repo._revision_knit = None
1860
1724
        self.revision_index.clear()
1861
1725
        # cached signature data
 
1726
        self.repo._signature_knit = None
1862
1727
        self.signature_index.clear()
1863
1728
        # cached file text data
1864
1729
        self.text_index.clear()
 
1730
        self.repo._text_knit = None
1865
1731
        # cached inventory data
1866
1732
        self.inventory_index.clear()
1867
 
        # cached chk data
1868
 
        if self.chk_index is not None:
1869
 
            self.chk_index.clear()
1870
1733
        # remove the open pack
1871
1734
        self._new_pack = None
1872
1735
        # information about packs.
1891
1754
        disk_nodes = set()
1892
1755
        for index, key, value in self._iter_disk_pack_index():
1893
1756
            disk_nodes.add((key, value))
1894
 
        orig_disk_nodes = set(disk_nodes)
1895
1757
 
1896
1758
        # do a two-way diff against our original content
1897
1759
        current_nodes = set()
1910
1772
        disk_nodes.difference_update(deleted_nodes)
1911
1773
        disk_nodes.update(new_nodes)
1912
1774
 
1913
 
        return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
 
1775
        return disk_nodes, deleted_nodes, new_nodes
1914
1776
 
1915
1777
    def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1916
1778
        """Given the correct set of pack files, update our saved info.
1956
1818
                added.append(name)
1957
1819
        return removed, added, modified
1958
1820
 
1959
 
    def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
 
1821
    def _save_pack_names(self, clear_obsolete_packs=False):
1960
1822
        """Save the list of packs.
1961
1823
 
1962
1824
        This will take out the mutex around the pack names list for the
1966
1828
 
1967
1829
        :param clear_obsolete_packs: If True, clear out the contents of the
1968
1830
            obsolete_packs directory.
1969
 
        :param obsolete_packs: Packs that are obsolete once the new pack-names
1970
 
            file has been written.
1971
 
        :return: A list of the names saved that were not previously on disk.
1972
1831
        """
1973
 
        already_obsolete = []
1974
1832
        self.lock_names()
1975
1833
        try:
1976
1834
            builder = self._index_builder_class()
1977
 
            (disk_nodes, deleted_nodes, new_nodes,
1978
 
             orig_disk_nodes) = self._diff_pack_names()
 
1835
            disk_nodes, deleted_nodes, new_nodes = self._diff_pack_names()
1979
1836
            # TODO: handle same-name, index-size-changes here -
1980
1837
            # e.g. use the value from disk, not ours, *unless* we're the one
1981
1838
            # changing it.
1983
1840
                builder.add_node(key, value)
1984
1841
            self.transport.put_file('pack-names', builder.finish(),
1985
1842
                mode=self.repo.bzrdir._get_file_mode())
 
1843
            # move the baseline forward
1986
1844
            self._packs_at_load = disk_nodes
1987
1845
            if clear_obsolete_packs:
1988
 
                to_preserve = None
1989
 
                if obsolete_packs:
1990
 
                    to_preserve = set([o.name for o in obsolete_packs])
1991
 
                already_obsolete = self._clear_obsolete_packs(to_preserve)
 
1846
                self._clear_obsolete_packs()
1992
1847
        finally:
1993
1848
            self._unlock_names()
1994
1849
        # synchronise the memory packs list with what we just wrote:
1995
1850
        self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1996
 
        if obsolete_packs:
1997
 
            # TODO: We could add one more condition here. "if o.name not in
1998
 
            #       orig_disk_nodes and o != the new_pack we haven't written to
1999
 
            #       disk yet. However, the new pack object is not easily
2000
 
            #       accessible here (it would have to be passed through the
2001
 
            #       autopacking code, etc.)
2002
 
            obsolete_packs = [o for o in obsolete_packs
2003
 
                              if o.name not in already_obsolete]
2004
 
            self._obsolete_packs(obsolete_packs)
2005
 
        return [new_node[0][0] for new_node in new_nodes]
2006
1851
 
2007
1852
    def reload_pack_names(self):
2008
1853
        """Sync our pack listing with what is present in the repository.
2022
1867
        if first_read:
2023
1868
            return True
2024
1869
        # out the new value.
2025
 
        (disk_nodes, deleted_nodes, new_nodes,
2026
 
         orig_disk_nodes) = self._diff_pack_names()
2027
 
        # _packs_at_load is meant to be the explicit list of names in
2028
 
        # 'pack-names' at then start. As such, it should not contain any
2029
 
        # pending names that haven't been written out yet.
2030
 
        self._packs_at_load = orig_disk_nodes
 
1870
        disk_nodes, _, _ = self._diff_pack_names()
 
1871
        self._packs_at_load = disk_nodes
2031
1872
        (removed, added,
2032
1873
         modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
2033
1874
        if removed or added or modified:
2042
1883
            raise
2043
1884
        raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2044
1885
 
2045
 
    def _clear_obsolete_packs(self, preserve=None):
 
1886
    def _clear_obsolete_packs(self):
2046
1887
        """Delete everything from the obsolete-packs directory.
2047
 
 
2048
 
        :return: A list of pack identifiers (the filename without '.pack') that
2049
 
            were found in obsolete_packs.
2050
1888
        """
2051
 
        found = []
2052
1889
        obsolete_pack_transport = self.transport.clone('obsolete_packs')
2053
 
        if preserve is None:
2054
 
            preserve = set()
2055
1890
        for filename in obsolete_pack_transport.list_dir('.'):
2056
 
            name, ext = osutils.splitext(filename)
2057
 
            if ext == '.pack':
2058
 
                found.append(name)
2059
 
            if name in preserve:
2060
 
                continue
2061
1891
            try:
2062
1892
                obsolete_pack_transport.delete(filename)
2063
1893
            except (errors.PathError, errors.TransportError), e:
2064
 
                warning("couldn't delete obsolete pack, skipping it:\n%s"
2065
 
                        % (e,))
2066
 
        return found
 
1894
                warning("couldn't delete obsolete pack, skipping it:\n%s" % (e,))
2067
1895
 
2068
1896
    def _start_write_group(self):
2069
1897
        # Do not permit preparation for writing if we're not in a 'write lock'.
2070
1898
        if not self.repo.is_write_locked():
2071
1899
            raise errors.NotWriteLocked(self)
2072
 
        self._new_pack = self.pack_factory(self, upload_suffix='.pack',
 
1900
        self._new_pack = NewPack(self, upload_suffix='.pack',
2073
1901
            file_mode=self.repo.bzrdir._get_file_mode())
2074
1902
        # allow writing: queue writes to a new index
2075
1903
        self.revision_index.add_writable_index(self._new_pack.revision_index,
2078
1906
            self._new_pack)
2079
1907
        self.text_index.add_writable_index(self._new_pack.text_index,
2080
1908
            self._new_pack)
2081
 
        self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2082
1909
        self.signature_index.add_writable_index(self._new_pack.signature_index,
2083
1910
            self._new_pack)
2084
 
        if self.chk_index is not None:
2085
 
            self.chk_index.add_writable_index(self._new_pack.chk_index,
2086
 
                self._new_pack)
2087
 
            self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2088
 
            self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2089
1911
 
2090
1912
        self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2091
1913
        self.repo.revisions._index._add_callback = self.revision_index.add_callback
2096
1918
        # FIXME: just drop the transient index.
2097
1919
        # forget what names there are
2098
1920
        if self._new_pack is not None:
2099
 
            operation = cleanup.OperationWithCleanups(self._new_pack.abort)
2100
 
            operation.add_cleanup(setattr, self, '_new_pack', None)
2101
 
            # If we aborted while in the middle of finishing the write
2102
 
            # group, _remove_pack_indices could fail because the indexes are
2103
 
            # already gone.  But they're not there we shouldn't fail in this
2104
 
            # case, so we pass ignore_missing=True.
2105
 
            operation.add_cleanup(self._remove_pack_indices, self._new_pack,
2106
 
                ignore_missing=True)
2107
 
            operation.run_simple()
 
1921
            try:
 
1922
                self._new_pack.abort()
 
1923
            finally:
 
1924
                # XXX: If we aborted while in the middle of finishing the write
 
1925
                # group, _remove_pack_indices can fail because the indexes are
 
1926
                # already gone.  If they're not there we shouldn't fail in this
 
1927
                # case.  -- mbp 20081113
 
1928
                self._remove_pack_indices(self._new_pack)
 
1929
                self._new_pack = None
2108
1930
        for resumed_pack in self._resumed_packs:
2109
 
            operation = cleanup.OperationWithCleanups(resumed_pack.abort)
2110
 
            # See comment in previous finally block.
2111
 
            operation.add_cleanup(self._remove_pack_indices, resumed_pack,
2112
 
                ignore_missing=True)
2113
 
            operation.run_simple()
 
1931
            try:
 
1932
                resumed_pack.abort()
 
1933
            finally:
 
1934
                # See comment in previous finally block.
 
1935
                try:
 
1936
                    self._remove_pack_indices(resumed_pack)
 
1937
                except KeyError:
 
1938
                    pass
2114
1939
        del self._resumed_packs[:]
 
1940
        self.repo._text_knit = None
2115
1941
 
2116
1942
    def _remove_resumed_pack_indices(self):
2117
1943
        for resumed_pack in self._resumed_packs:
2118
1944
            self._remove_pack_indices(resumed_pack)
2119
1945
        del self._resumed_packs[:]
2120
1946
 
2121
 
    def _check_new_inventories(self):
2122
 
        """Detect missing inventories in this write group.
2123
 
 
2124
 
        :returns: list of strs, summarising any problems found.  If the list is
2125
 
            empty no problems were found.
2126
 
        """
2127
 
        # The base implementation does no checks.  GCRepositoryPackCollection
2128
 
        # overrides this.
2129
 
        return []
2130
 
        
2131
1947
    def _commit_write_group(self):
2132
1948
        all_missing = set()
2133
1949
        for prefix, versioned_file in (
2142
1958
            raise errors.BzrCheckError(
2143
1959
                "Repository %s has missing compression parent(s) %r "
2144
1960
                 % (self.repo, sorted(all_missing)))
2145
 
        problems = self._check_new_inventories()
2146
 
        if problems:
2147
 
            problems_summary = '\n'.join(problems)
2148
 
            raise errors.BzrCheckError(
2149
 
                "Cannot add revision(s) to repository: " + problems_summary)
2150
1961
        self._remove_pack_indices(self._new_pack)
2151
 
        any_new_content = False
 
1962
        should_autopack = False
2152
1963
        if self._new_pack.data_inserted():
2153
1964
            # get all the data to disk and read to use
2154
1965
            self._new_pack.finish()
2155
1966
            self.allocate(self._new_pack)
2156
1967
            self._new_pack = None
2157
 
            any_new_content = True
 
1968
            should_autopack = True
2158
1969
        else:
2159
1970
            self._new_pack.abort()
2160
1971
            self._new_pack = None
2165
1976
            self._remove_pack_from_memory(resumed_pack)
2166
1977
            resumed_pack.finish()
2167
1978
            self.allocate(resumed_pack)
2168
 
            any_new_content = True
 
1979
            should_autopack = True
2169
1980
        del self._resumed_packs[:]
2170
 
        if any_new_content:
2171
 
            result = self.autopack()
2172
 
            if not result:
 
1981
        if should_autopack:
 
1982
            if not self.autopack():
2173
1983
                # when autopack takes no steps, the names list is still
2174
1984
                # unsaved.
2175
 
                return self._save_pack_names()
2176
 
            return result
2177
 
        return []
 
1985
                self._save_pack_names()
 
1986
        self.repo._text_knit = None
2178
1987
 
2179
1988
    def _suspend_write_group(self):
2180
1989
        tokens = [pack.name for pack in self._resumed_packs]
2188
1997
            self._new_pack.abort()
2189
1998
            self._new_pack = None
2190
1999
        self._remove_resumed_pack_indices()
 
2000
        self.repo._text_knit = None
2191
2001
        return tokens
2192
2002
 
2193
2003
    def _resume_write_group(self, tokens):
2230
2040
            self._transport.clone('upload'),
2231
2041
            self._transport.clone('packs'),
2232
2042
            _format.index_builder_class,
2233
 
            _format.index_class,
2234
 
            use_chk_index=self._format.supports_chks,
2235
 
            )
 
2043
            _format.index_class)
2236
2044
        self.inventories = KnitVersionedFiles(
2237
2045
            _KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2238
2046
                add_callback=self._pack_collection.inventory_index.add_callback,
2242
2050
        self.revisions = KnitVersionedFiles(
2243
2051
            _KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2244
2052
                add_callback=self._pack_collection.revision_index.add_callback,
2245
 
                deltas=False, parents=True, is_locked=self.is_locked,
2246
 
                track_external_parent_refs=True),
 
2053
                deltas=False, parents=True, is_locked=self.is_locked),
2247
2054
            data_access=self._pack_collection.revision_index.data_access,
2248
2055
            max_delta_chain=0)
2249
2056
        self.signatures = KnitVersionedFiles(
2258
2065
                deltas=True, parents=True, is_locked=self.is_locked),
2259
2066
            data_access=self._pack_collection.text_index.data_access,
2260
2067
            max_delta_chain=200)
2261
 
        if _format.supports_chks:
2262
 
            # No graph, no compression:- references from chks are between
2263
 
            # different objects not temporal versions of the same; and without
2264
 
            # some sort of temporal structure knit compression will just fail.
2265
 
            self.chk_bytes = KnitVersionedFiles(
2266
 
                _KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2267
 
                    add_callback=self._pack_collection.chk_index.add_callback,
2268
 
                    deltas=False, parents=False, is_locked=self.is_locked),
2269
 
                data_access=self._pack_collection.chk_index.data_access,
2270
 
                max_delta_chain=0)
2271
 
        else:
2272
 
            self.chk_bytes = None
2273
2068
        # True when the repository object is 'write locked' (as opposed to the
2274
2069
        # physical lock only taken out around changes to the pack-names list.)
2275
2070
        # Another way to represent this would be a decorator around the control
2282
2077
        self._reconcile_fixes_text_parents = True
2283
2078
        self._reconcile_backsup_inventory = False
2284
2079
 
2285
 
    def _warn_if_deprecated(self, branch=None):
 
2080
    def _warn_if_deprecated(self):
2286
2081
        # This class isn't deprecated, but one sub-format is
2287
2082
        if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2288
 
            super(KnitPackRepository, self)._warn_if_deprecated(branch)
 
2083
            from bzrlib import repository
 
2084
            if repository._deprecation_warning_done:
 
2085
                return
 
2086
            repository._deprecation_warning_done = True
 
2087
            warning("Format %s for %s is deprecated - please use"
 
2088
                    " 'bzr upgrade --1.6.1-rich-root'"
 
2089
                    % (self._format, self.bzrdir.transport.base))
2289
2090
 
2290
2091
    def _abort_write_group(self):
2291
 
        self.revisions._index._key_dependencies.clear()
2292
2092
        self._pack_collection._abort_write_group()
2293
2093
 
2294
 
    def _get_source(self, to_format):
2295
 
        if to_format.network_name() == self._format.network_name():
2296
 
            return KnitPackStreamSource(self, to_format)
2297
 
        return super(KnitPackRepository, self)._get_source(to_format)
 
2094
    def _find_inconsistent_revision_parents(self):
 
2095
        """Find revisions with incorrectly cached parents.
 
2096
 
 
2097
        :returns: an iterator yielding tuples of (revison-id, parents-in-index,
 
2098
            parents-in-revision).
 
2099
        """
 
2100
        if not self.is_locked():
 
2101
            raise errors.ObjectNotLocked(self)
 
2102
        pb = ui.ui_factory.nested_progress_bar()
 
2103
        result = []
 
2104
        try:
 
2105
            revision_nodes = self._pack_collection.revision_index \
 
2106
                .combined_index.iter_all_entries()
 
2107
            index_positions = []
 
2108
            # Get the cached index values for all revisions, and also the location
 
2109
            # in each index of the revision text so we can perform linear IO.
 
2110
            for index, key, value, refs in revision_nodes:
 
2111
                pos, length = value[1:].split(' ')
 
2112
                index_positions.append((index, int(pos), key[0],
 
2113
                    tuple(parent[0] for parent in refs[0])))
 
2114
                pb.update("Reading revision index", 0, 0)
 
2115
            index_positions.sort()
 
2116
            batch_count = len(index_positions) / 1000 + 1
 
2117
            pb.update("Checking cached revision graph", 0, batch_count)
 
2118
            for offset in xrange(batch_count):
 
2119
                pb.update("Checking cached revision graph", offset)
 
2120
                to_query = index_positions[offset * 1000:(offset + 1) * 1000]
 
2121
                if not to_query:
 
2122
                    break
 
2123
                rev_ids = [item[2] for item in to_query]
 
2124
                revs = self.get_revisions(rev_ids)
 
2125
                for revision, item in zip(revs, to_query):
 
2126
                    index_parents = item[3]
 
2127
                    rev_parents = tuple(revision.parent_ids)
 
2128
                    if index_parents != rev_parents:
 
2129
                        result.append((revision.revision_id, index_parents, rev_parents))
 
2130
        finally:
 
2131
            pb.finished()
 
2132
        return result
2298
2133
 
2299
2134
    def _make_parents_provider(self):
2300
2135
        return graph.CachingParentsProvider(self)
2308
2143
        self._pack_collection._start_write_group()
2309
2144
 
2310
2145
    def _commit_write_group(self):
2311
 
        hint = self._pack_collection._commit_write_group()
2312
 
        self.revisions._index._key_dependencies.clear()
2313
 
        return hint
 
2146
        return self._pack_collection._commit_write_group()
2314
2147
 
2315
2148
    def suspend_write_group(self):
2316
2149
        # XXX check self._write_group is self.get_transaction()?
2317
2150
        tokens = self._pack_collection._suspend_write_group()
2318
 
        self.revisions._index._key_dependencies.clear()
2319
2151
        self._write_group = None
2320
2152
        return tokens
2321
2153
 
2322
2154
    def _resume_write_group(self, tokens):
2323
2155
        self._start_write_group()
2324
 
        try:
2325
 
            self._pack_collection._resume_write_group(tokens)
2326
 
        except errors.UnresumableWriteGroup:
2327
 
            self._abort_write_group()
2328
 
            raise
2329
 
        for pack in self._pack_collection._resumed_packs:
2330
 
            self.revisions._index.scan_unvalidated_index(pack.revision_index)
 
2156
        self._pack_collection._resume_write_group(tokens)
2331
2157
 
2332
2158
    def get_transaction(self):
2333
2159
        if self._write_lock_count:
2342
2168
        return self._write_lock_count
2343
2169
 
2344
2170
    def lock_write(self, token=None):
2345
 
        """Lock the repository for writes.
2346
 
 
2347
 
        :return: A bzrlib.repository.RepositoryWriteLockResult.
2348
 
        """
2349
2171
        locked = self.is_locked()
2350
2172
        if not self._write_lock_count and locked:
2351
2173
            raise errors.ReadOnlyError(self)
2352
2174
        self._write_lock_count += 1
2353
2175
        if self._write_lock_count == 1:
2354
2176
            self._transaction = transactions.WriteTransaction()
2355
 
        if not locked:
2356
 
            if 'relock' in debug.debug_flags and self._prev_lock == 'w':
2357
 
                note('%r was write locked again', self)
2358
 
            self._prev_lock = 'w'
2359
2177
            for repo in self._fallback_repositories:
2360
2178
                # Writes don't affect fallback repos
2361
2179
                repo.lock_read()
 
2180
        if not locked:
2362
2181
            self._refresh_data()
2363
 
        return RepositoryWriteLockResult(self.unlock, None)
2364
2182
 
2365
2183
    def lock_read(self):
2366
 
        """Lock the repository for reads.
2367
 
 
2368
 
        :return: A bzrlib.lock.LogicalLockResult.
2369
 
        """
2370
2184
        locked = self.is_locked()
2371
2185
        if self._write_lock_count:
2372
2186
            self._write_lock_count += 1
2373
2187
        else:
2374
2188
            self.control_files.lock_read()
2375
 
        if not locked:
2376
 
            if 'relock' in debug.debug_flags and self._prev_lock == 'r':
2377
 
                note('%r was read locked again', self)
2378
 
            self._prev_lock = 'r'
2379
2189
            for repo in self._fallback_repositories:
 
2190
                # Writes don't affect fallback repos
2380
2191
                repo.lock_read()
 
2192
        if not locked:
2381
2193
            self._refresh_data()
2382
 
        return LogicalLockResult(self.unlock)
2383
2194
 
2384
2195
    def leave_lock_in_place(self):
2385
2196
        # not supported - raise an error
2390
2201
        raise NotImplementedError(self.dont_leave_lock_in_place)
2391
2202
 
2392
2203
    @needs_write_lock
2393
 
    def pack(self, hint=None, clean_obsolete_packs=False):
 
2204
    def pack(self):
2394
2205
        """Compress the data within the repository.
2395
2206
 
2396
2207
        This will pack all the data to a single pack. In future it may
2397
2208
        recompress deltas or do other such expensive operations.
2398
2209
        """
2399
 
        self._pack_collection.pack(hint=hint, clean_obsolete_packs=clean_obsolete_packs)
 
2210
        self._pack_collection.pack()
2400
2211
 
2401
2212
    @needs_write_lock
2402
2213
    def reconcile(self, other=None, thorough=False):
2406
2217
        reconciler.reconcile()
2407
2218
        return reconciler
2408
2219
 
2409
 
    def _reconcile_pack(self, collection, packs, extension, revs, pb):
2410
 
        packer = ReconcilePacker(collection, packs, extension, revs)
2411
 
        return packer.pack(pb)
2412
 
 
2413
 
    @only_raises(errors.LockNotHeld, errors.LockBroken)
2414
2220
    def unlock(self):
2415
2221
        if self._write_lock_count == 1 and self._write_group is not None:
2416
2222
            self.abort_write_group()
2425
2231
                transaction = self._transaction
2426
2232
                self._transaction = None
2427
2233
                transaction.finish()
 
2234
                for repo in self._fallback_repositories:
 
2235
                    repo.unlock()
2428
2236
        else:
2429
2237
            self.control_files.unlock()
2430
 
 
2431
 
        if not self.is_locked():
2432
2238
            for repo in self._fallback_repositories:
2433
2239
                repo.unlock()
2434
2240
 
2435
2241
 
2436
 
class KnitPackStreamSource(StreamSource):
2437
 
    """A StreamSource used to transfer data between same-format KnitPack repos.
2438
 
 
2439
 
    This source assumes:
2440
 
        1) Same serialization format for all objects
2441
 
        2) Same root information
2442
 
        3) XML format inventories
2443
 
        4) Atomic inserts (so we can stream inventory texts before text
2444
 
           content)
2445
 
        5) No chk_bytes
2446
 
    """
2447
 
 
2448
 
    def __init__(self, from_repository, to_format):
2449
 
        super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2450
 
        self._text_keys = None
2451
 
        self._text_fetch_order = 'unordered'
2452
 
 
2453
 
    def _get_filtered_inv_stream(self, revision_ids):
2454
 
        from_repo = self.from_repository
2455
 
        parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2456
 
        parent_keys = [(p,) for p in parent_ids]
2457
 
        find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2458
 
        parent_text_keys = set(find_text_keys(
2459
 
            from_repo._inventory_xml_lines_for_keys(parent_keys)))
2460
 
        content_text_keys = set()
2461
 
        knit = KnitVersionedFiles(None, None)
2462
 
        factory = KnitPlainFactory()
2463
 
        def find_text_keys_from_content(record):
2464
 
            if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2465
 
                raise ValueError("Unknown content storage kind for"
2466
 
                    " inventory text: %s" % (record.storage_kind,))
2467
 
            # It's a knit record, it has a _raw_record field (even if it was
2468
 
            # reconstituted from a network stream).
2469
 
            raw_data = record._raw_record
2470
 
            # read the entire thing
2471
 
            revision_id = record.key[-1]
2472
 
            content, _ = knit._parse_record(revision_id, raw_data)
2473
 
            if record.storage_kind == 'knit-delta-gz':
2474
 
                line_iterator = factory.get_linedelta_content(content)
2475
 
            elif record.storage_kind == 'knit-ft-gz':
2476
 
                line_iterator = factory.get_fulltext_content(content)
2477
 
            content_text_keys.update(find_text_keys(
2478
 
                [(line, revision_id) for line in line_iterator]))
2479
 
        revision_keys = [(r,) for r in revision_ids]
2480
 
        def _filtered_inv_stream():
2481
 
            source_vf = from_repo.inventories
2482
 
            stream = source_vf.get_record_stream(revision_keys,
2483
 
                                                 'unordered', False)
2484
 
            for record in stream:
2485
 
                if record.storage_kind == 'absent':
2486
 
                    raise errors.NoSuchRevision(from_repo, record.key)
2487
 
                find_text_keys_from_content(record)
2488
 
                yield record
2489
 
            self._text_keys = content_text_keys - parent_text_keys
2490
 
        return ('inventories', _filtered_inv_stream())
2491
 
 
2492
 
    def _get_text_stream(self):
2493
 
        # Note: We know we don't have to handle adding root keys, because both
2494
 
        # the source and target are the identical network name.
2495
 
        text_stream = self.from_repository.texts.get_record_stream(
2496
 
                        self._text_keys, self._text_fetch_order, False)
2497
 
        return ('texts', text_stream)
2498
 
 
2499
 
    def get_stream(self, search):
2500
 
        revision_ids = search.get_keys()
2501
 
        for stream_info in self._fetch_revision_texts(revision_ids):
2502
 
            yield stream_info
2503
 
        self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2504
 
        yield self._get_filtered_inv_stream(revision_ids)
2505
 
        yield self._get_text_stream()
2506
 
 
2507
 
 
2508
 
 
2509
2242
class RepositoryFormatPack(MetaDirRepositoryFormat):
2510
2243
    """Format logic for pack structured repositories.
2511
2244
 
2535
2268
    supports_ghosts = True
2536
2269
    # External references are not supported in pack repositories yet.
2537
2270
    supports_external_lookups = False
2538
 
    # Most pack formats do not use chk lookups.
2539
 
    supports_chks = False
2540
2271
    # What index classes to use
2541
2272
    index_builder_class = None
2542
2273
    index_class = None
2558
2289
        utf8_files = [('format', self.get_format_string())]
2559
2290
 
2560
2291
        self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2561
 
        repository = self.open(a_bzrdir=a_bzrdir, _found=True)
2562
 
        self._run_post_repo_init_hooks(repository, a_bzrdir, shared)
2563
 
        return repository
 
2292
        return self.open(a_bzrdir=a_bzrdir, _found=True)
2564
2293
 
2565
2294
    def open(self, a_bzrdir, _found=False, _override_transport=None):
2566
2295
        """See RepositoryFormat.open().
2615
2344
        """See RepositoryFormat.get_format_description()."""
2616
2345
        return "Packs containing knits without subtree support"
2617
2346
 
 
2347
    def check_conversion_target(self, target_format):
 
2348
        pass
 
2349
 
2618
2350
 
2619
2351
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2620
2352
    """A subtrees parameterized Pack repository.
2629
2361
    repository_class = KnitPackRepository
2630
2362
    _commit_builder_class = PackRootCommitBuilder
2631
2363
    rich_root_data = True
2632
 
    experimental = True
2633
2364
    supports_tree_reference = True
2634
2365
    @property
2635
2366
    def _serializer(self):
2647
2378
 
2648
2379
    _matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2649
2380
 
 
2381
    def check_conversion_target(self, target_format):
 
2382
        if not target_format.rich_root_data:
 
2383
            raise errors.BadConversionTarget(
 
2384
                'Does not support rich root data.', target_format)
 
2385
        if not getattr(target_format, 'supports_tree_reference', False):
 
2386
            raise errors.BadConversionTarget(
 
2387
                'Does not support nested trees', target_format)
 
2388
 
2650
2389
    def get_format_string(self):
2651
2390
        """See RepositoryFormat.get_format_string()."""
2652
2391
        return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2685
2424
 
2686
2425
    _matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2687
2426
 
 
2427
    def check_conversion_target(self, target_format):
 
2428
        if not target_format.rich_root_data:
 
2429
            raise errors.BadConversionTarget(
 
2430
                'Does not support rich root data.', target_format)
 
2431
 
2688
2432
    def get_format_string(self):
2689
2433
        """See RepositoryFormat.get_format_string()."""
2690
2434
        return ("Bazaar pack repository format 1 with rich root"
2731
2475
        """See RepositoryFormat.get_format_description()."""
2732
2476
        return "Packs 5 (adds stacking support, requires bzr 1.6)"
2733
2477
 
 
2478
    def check_conversion_target(self, target_format):
 
2479
        pass
 
2480
 
2734
2481
 
2735
2482
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2736
2483
    """A repository with rich roots and stacking.
2763
2510
 
2764
2511
    _matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2765
2512
 
 
2513
    def check_conversion_target(self, target_format):
 
2514
        if not target_format.rich_root_data:
 
2515
            raise errors.BadConversionTarget(
 
2516
                'Does not support rich root data.', target_format)
 
2517
 
2766
2518
    def get_format_string(self):
2767
2519
        """See RepositoryFormat.get_format_string()."""
2768
2520
        return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2809
2561
 
2810
2562
    _matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2811
2563
 
 
2564
    def check_conversion_target(self, target_format):
 
2565
        if not target_format.rich_root_data:
 
2566
            raise errors.BadConversionTarget(
 
2567
                'Does not support rich root data.', target_format)
 
2568
 
2812
2569
    def get_format_string(self):
2813
2570
        """See RepositoryFormat.get_format_string()."""
2814
2571
        return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2852
2609
        """See RepositoryFormat.get_format_description()."""
2853
2610
        return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2854
2611
 
 
2612
    def check_conversion_target(self, target_format):
 
2613
        pass
 
2614
 
2855
2615
 
2856
2616
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2857
2617
    """A repository with rich roots, no subtrees, stacking and btree indexes.
2881
2641
 
2882
2642
    _matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2883
2643
 
 
2644
    def check_conversion_target(self, target_format):
 
2645
        if not target_format.rich_root_data:
 
2646
            raise errors.BadConversionTarget(
 
2647
                'Does not support rich root data.', target_format)
 
2648
 
2884
2649
    def get_format_string(self):
2885
2650
        """See RepositoryFormat.get_format_string()."""
2886
2651
        return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2889
2654
        return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2890
2655
 
2891
2656
 
 
2657
class RepositoryFormatPackDevelopment2(RepositoryFormatPack):
 
2658
    """A no-subtrees development repository.
 
2659
 
 
2660
    This format should be retained until the second release after bzr 1.7.
 
2661
 
 
2662
    This is pack-1.6.1 with B+Tree indices.
 
2663
    """
 
2664
 
 
2665
    repository_class = KnitPackRepository
 
2666
    _commit_builder_class = PackCommitBuilder
 
2667
    supports_external_lookups = True
 
2668
    # What index classes to use
 
2669
    index_builder_class = BTreeBuilder
 
2670
    index_class = BTreeGraphIndex
 
2671
    # Set to true to get the fast-commit code path tested until a really fast
 
2672
    # format lands in trunk. Not actually fast in this format.
 
2673
    fast_deltas = True
 
2674
 
 
2675
    @property
 
2676
    def _serializer(self):
 
2677
        return xml5.serializer_v5
 
2678
 
 
2679
    def _get_matching_bzrdir(self):
 
2680
        return bzrdir.format_registry.make_bzrdir('development2')
 
2681
 
 
2682
    def _ignore_setting_bzrdir(self, format):
 
2683
        pass
 
2684
 
 
2685
    _matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
 
2686
 
 
2687
    def get_format_string(self):
 
2688
        """See RepositoryFormat.get_format_string()."""
 
2689
        return "Bazaar development format 2 (needs bzr.dev from before 1.8)\n"
 
2690
 
 
2691
    def get_format_description(self):
 
2692
        """See RepositoryFormat.get_format_description()."""
 
2693
        return ("Development repository format, currently the same as "
 
2694
            "1.6.1 with B+Trees.\n")
 
2695
 
 
2696
    def check_conversion_target(self, target_format):
 
2697
        pass
 
2698
 
 
2699
 
2892
2700
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2893
2701
    """A subtrees development repository.
2894
2702
 
2895
2703
    This format should be retained until the second release after bzr 1.7.
2896
2704
 
2897
2705
    1.6.1-subtree[as it might have been] with B+Tree indices.
2898
 
 
2899
 
    This is [now] retained until we have a CHK based subtree format in
2900
 
    development.
2901
2706
    """
2902
2707
 
2903
2708
    repository_class = KnitPackRepository
2904
2709
    _commit_builder_class = PackRootCommitBuilder
2905
2710
    rich_root_data = True
2906
 
    experimental = True
2907
2711
    supports_tree_reference = True
2908
2712
    supports_external_lookups = True
2909
2713
    # What index classes to use
2916
2720
 
2917
2721
    def _get_matching_bzrdir(self):
2918
2722
        return bzrdir.format_registry.make_bzrdir(
2919
 
            'development-subtree')
 
2723
            'development2-subtree')
2920
2724
 
2921
2725
    def _ignore_setting_bzrdir(self, format):
2922
2726
        pass
2923
2727
 
2924
2728
    _matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2925
2729
 
 
2730
    def check_conversion_target(self, target_format):
 
2731
        if not target_format.rich_root_data:
 
2732
            raise errors.BadConversionTarget(
 
2733
                'Does not support rich root data.', target_format)
 
2734
        if not getattr(target_format, 'supports_tree_reference', False):
 
2735
            raise errors.BadConversionTarget(
 
2736
                'Does not support nested trees', target_format)
 
2737
 
2926
2738
    def get_format_string(self):
2927
2739
        """See RepositoryFormat.get_format_string()."""
2928
2740
        return ("Bazaar development format 2 with subtree support "
2932
2744
        """See RepositoryFormat.get_format_description()."""
2933
2745
        return ("Development repository format, currently the same as "
2934
2746
            "1.6.1-subtree with B+Tree indices.\n")
2935