1209
1198
return new_pack
1212
class ReconcilePacker(Packer):
1213
"""A packer which regenerates indices etc as it copies.
1215
This is used by ``bzr reconcile`` to cause parent text pointers to be
1219
def _extra_init(self):
1220
self._data_changed = False
1222
def _process_inventory_lines(self, inv_lines):
1223
"""Generate a text key reference map rather for reconciling with."""
1224
repo = self._pack_collection.repo
1225
refs = repo._find_text_key_references_from_xml_inventory_lines(
1227
self._text_refs = refs
1228
# during reconcile we:
1229
# - convert unreferenced texts to full texts
1230
# - correct texts which reference a text not copied to be full texts
1231
# - copy all others as-is but with corrected parents.
1232
# - so at this point we don't know enough to decide what becomes a full
1234
self._text_filter = None
1236
def _copy_text_texts(self):
1237
"""generate what texts we should have and then copy."""
1238
self.pb.update("Copying content texts", 3)
1239
# we have three major tasks here:
1240
# 1) generate the ideal index
1241
repo = self._pack_collection.repo
1242
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1243
_1, key, _2, refs in
1244
self.new_pack.revision_index.iter_all_entries()])
1245
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1246
# 2) generate a text_nodes list that contains all the deltas that can
1247
# be used as-is, with corrected parents.
1250
discarded_nodes = []
1251
NULL_REVISION = _mod_revision.NULL_REVISION
1252
text_index_map, text_nodes = self._get_text_nodes()
1253
for node in text_nodes:
1259
ideal_parents = tuple(ideal_index[node[1]])
1261
discarded_nodes.append(node)
1262
self._data_changed = True
1264
if ideal_parents == (NULL_REVISION,):
1266
if ideal_parents == node[3][0]:
1268
ok_nodes.append(node)
1269
elif ideal_parents[0:1] == node[3][0][0:1]:
1270
# the left most parent is the same, or there are no parents
1271
# today. Either way, we can preserve the representation as
1272
# long as we change the refs to be inserted.
1273
self._data_changed = True
1274
ok_nodes.append((node[0], node[1], node[2],
1275
(ideal_parents, node[3][1])))
1276
self._data_changed = True
1278
# Reinsert this text completely
1279
bad_texts.append((node[1], ideal_parents))
1280
self._data_changed = True
1281
# we're finished with some data.
1284
# 3) bulk copy the ok data
1285
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1286
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1287
self.new_pack.text_index, readv_group_iter, total_items))
1288
# 4) adhoc copy all the other texts.
1289
# We have to topologically insert all texts otherwise we can fail to
1290
# reconcile when parts of a single delta chain are preserved intact,
1291
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1292
# reinserted, and if d3 has incorrect parents it will also be
1293
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1294
# copied), so we will try to delta, but d2 is not currently able to be
1295
# extracted because its basis d1 is not present. Topologically sorting
1296
# addresses this. The following generates a sort for all the texts that
1297
# are being inserted without having to reference the entire text key
1298
# space (we only topo sort the revisions, which is smaller).
1299
topo_order = tsort.topo_sort(ancestors)
1300
rev_order = dict(zip(topo_order, range(len(topo_order))))
1301
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1302
transaction = repo.get_transaction()
1303
file_id_index = GraphIndexPrefixAdapter(
1304
self.new_pack.text_index,
1306
add_nodes_callback=self.new_pack.text_index.add_nodes)
1307
data_access = _DirectPackAccess(
1308
{self.new_pack.text_index:self.new_pack.access_tuple()})
1309
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1310
self.new_pack.access_tuple())
1311
output_texts = KnitVersionedFiles(
1312
_KnitGraphIndex(self.new_pack.text_index,
1313
add_callback=self.new_pack.text_index.add_nodes,
1314
deltas=True, parents=True, is_locked=repo.is_locked),
1315
data_access=data_access, max_delta_chain=200)
1316
for key, parent_keys in bad_texts:
1317
# We refer to the new pack to delta data being output.
1318
# A possible improvement would be to catch errors on short reads
1319
# and only flush then.
1320
self.new_pack.flush()
1322
for parent_key in parent_keys:
1323
if parent_key[0] != key[0]:
1324
# Graph parents must match the fileid
1325
raise errors.BzrError('Mismatched key parent %r:%r' %
1327
parents.append(parent_key[1])
1328
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1329
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1330
output_texts.add_lines(key, parent_keys, text_lines,
1331
random_id=True, check_content=False)
1332
# 5) check that nothing inserted has a reference outside the keyspace.
1333
missing_text_keys = self.new_pack.text_index._external_references()
1334
if missing_text_keys:
1335
raise errors.BzrCheckError('Reference to missing compression parents %r'
1336
% (missing_text_keys,))
1337
self._log_copied_texts()
1339
def _use_pack(self, new_pack):
1340
"""Override _use_pack to check for reconcile having changed content."""
1341
# XXX: we might be better checking this at the copy time.
1342
original_inventory_keys = set()
1343
inv_index = self._pack_collection.inventory_index.combined_index
1344
for entry in inv_index.iter_all_entries():
1345
original_inventory_keys.add(entry[1])
1346
new_inventory_keys = set()
1347
for entry in new_pack.inventory_index.iter_all_entries():
1348
new_inventory_keys.add(entry[1])
1349
if new_inventory_keys != original_inventory_keys:
1350
self._data_changed = True
1351
return new_pack.data_inserted() and self._data_changed
1354
1201
class RepositoryPackCollection(object):
1355
1202
"""Management of packs within a repository.
2092
# These attributes are inherited from the Repository base class. Setting
2093
# them to None ensures that if the constructor is changed to not initialize
2094
# them, or a subclass fails to call the constructor, that an error will
2095
# occur rather than the system working but generating incorrect data.
2096
_commit_builder_class = None
2225
2099
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2227
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2228
_commit_builder_class, _serializer)
2229
index_transport = self._transport.clone('indices')
2230
self._pack_collection = RepositoryPackCollection(self, self._transport,
2232
self._transport.clone('upload'),
2233
self._transport.clone('packs'),
2234
_format.index_builder_class,
2235
_format.index_class,
2236
use_chk_index=self._format.supports_chks,
2238
self.inventories = KnitVersionedFiles(
2239
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2240
add_callback=self._pack_collection.inventory_index.add_callback,
2241
deltas=True, parents=True, is_locked=self.is_locked),
2242
data_access=self._pack_collection.inventory_index.data_access,
2243
max_delta_chain=200)
2244
self.revisions = KnitVersionedFiles(
2245
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2246
add_callback=self._pack_collection.revision_index.add_callback,
2247
deltas=False, parents=True, is_locked=self.is_locked,
2248
track_external_parent_refs=True),
2249
data_access=self._pack_collection.revision_index.data_access,
2251
self.signatures = KnitVersionedFiles(
2252
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2253
add_callback=self._pack_collection.signature_index.add_callback,
2254
deltas=False, parents=False, is_locked=self.is_locked),
2255
data_access=self._pack_collection.signature_index.data_access,
2257
self.texts = KnitVersionedFiles(
2258
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2259
add_callback=self._pack_collection.text_index.add_callback,
2260
deltas=True, parents=True, is_locked=self.is_locked),
2261
data_access=self._pack_collection.text_index.data_access,
2262
max_delta_chain=200)
2263
if _format.supports_chks:
2264
# No graph, no compression:- references from chks are between
2265
# different objects not temporal versions of the same; and without
2266
# some sort of temporal structure knit compression will just fail.
2267
self.chk_bytes = KnitVersionedFiles(
2268
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2269
add_callback=self._pack_collection.chk_index.add_callback,
2270
deltas=False, parents=False, is_locked=self.is_locked),
2271
data_access=self._pack_collection.chk_index.data_access,
2274
self.chk_bytes = None
2275
# True when the repository object is 'write locked' (as opposed to the
2276
# physical lock only taken out around changes to the pack-names list.)
2277
# Another way to represent this would be a decorator around the control
2278
# files object that presents logical locks as physical ones - if this
2279
# gets ugly consider that alternative design. RBC 20071011
2280
self._write_lock_count = 0
2281
self._transaction = None
2283
self._reconcile_does_inventory_gc = True
2101
MetaDirRepository.__init__(self, _format, a_bzrdir, control_files)
2102
self._commit_builder_class = _commit_builder_class
2103
self._serializer = _serializer
2284
2104
self._reconcile_fixes_text_parents = True
2285
self._reconcile_backsup_inventory = False
2287
def _warn_if_deprecated(self, branch=None):
2288
# This class isn't deprecated, but one sub-format is
2289
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2290
super(KnitPackRepository, self)._warn_if_deprecated(branch)
2107
def _all_revision_ids(self):
2108
"""See Repository.all_revision_ids()."""
2109
return [key[0] for key in self.revisions.keys()]
2292
2111
def _abort_write_group(self):
2293
2112
self.revisions._index._key_dependencies.clear()
2294
2113
self._pack_collection._abort_write_group()
2296
def _get_source(self, to_format):
2297
if to_format.network_name() == self._format.network_name():
2298
return KnitPackStreamSource(self, to_format)
2299
return super(KnitPackRepository, self)._get_source(to_format)
2301
2115
def _make_parents_provider(self):
2302
2116
return graph.CachingParentsProvider(self)
2438
class KnitPackStreamSource(StreamSource):
2439
"""A StreamSource used to transfer data between same-format KnitPack repos.
2441
This source assumes:
2442
1) Same serialization format for all objects
2443
2) Same root information
2444
3) XML format inventories
2445
4) Atomic inserts (so we can stream inventory texts before text
2450
def __init__(self, from_repository, to_format):
2451
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2452
self._text_keys = None
2453
self._text_fetch_order = 'unordered'
2455
def _get_filtered_inv_stream(self, revision_ids):
2456
from_repo = self.from_repository
2457
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2458
parent_keys = [(p,) for p in parent_ids]
2459
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2460
parent_text_keys = set(find_text_keys(
2461
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2462
content_text_keys = set()
2463
knit = KnitVersionedFiles(None, None)
2464
factory = KnitPlainFactory()
2465
def find_text_keys_from_content(record):
2466
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2467
raise ValueError("Unknown content storage kind for"
2468
" inventory text: %s" % (record.storage_kind,))
2469
# It's a knit record, it has a _raw_record field (even if it was
2470
# reconstituted from a network stream).
2471
raw_data = record._raw_record
2472
# read the entire thing
2473
revision_id = record.key[-1]
2474
content, _ = knit._parse_record(revision_id, raw_data)
2475
if record.storage_kind == 'knit-delta-gz':
2476
line_iterator = factory.get_linedelta_content(content)
2477
elif record.storage_kind == 'knit-ft-gz':
2478
line_iterator = factory.get_fulltext_content(content)
2479
content_text_keys.update(find_text_keys(
2480
[(line, revision_id) for line in line_iterator]))
2481
revision_keys = [(r,) for r in revision_ids]
2482
def _filtered_inv_stream():
2483
source_vf = from_repo.inventories
2484
stream = source_vf.get_record_stream(revision_keys,
2486
for record in stream:
2487
if record.storage_kind == 'absent':
2488
raise errors.NoSuchRevision(from_repo, record.key)
2489
find_text_keys_from_content(record)
2491
self._text_keys = content_text_keys - parent_text_keys
2492
return ('inventories', _filtered_inv_stream())
2494
def _get_text_stream(self):
2495
# Note: We know we don't have to handle adding root keys, because both
2496
# the source and target are the identical network name.
2497
text_stream = self.from_repository.texts.get_record_stream(
2498
self._text_keys, self._text_fetch_order, False)
2499
return ('texts', text_stream)
2501
def get_stream(self, search):
2502
revision_ids = search.get_keys()
2503
for stream_info in self._fetch_revision_texts(revision_ids):
2505
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2506
yield self._get_filtered_inv_stream(revision_ids)
2507
yield self._get_text_stream()
2511
2251
class RepositoryFormatPack(MetaDirRepositoryFormat):
2512
2252
"""Format logic for pack structured repositories.
2586
2329
_serializer=self._serializer)
2589
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2590
"""A no-subtrees parameterized Pack repository.
2592
This format was introduced in 0.92.
2595
repository_class = KnitPackRepository
2596
_commit_builder_class = PackCommitBuilder
2598
def _serializer(self):
2599
return xml5.serializer_v5
2600
# What index classes to use
2601
index_builder_class = InMemoryGraphIndex
2602
index_class = GraphIndex
2604
def _get_matching_bzrdir(self):
2605
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2607
def _ignore_setting_bzrdir(self, format):
2610
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2612
def get_format_string(self):
2613
"""See RepositoryFormat.get_format_string()."""
2614
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2616
def get_format_description(self):
2617
"""See RepositoryFormat.get_format_description()."""
2618
return "Packs containing knits without subtree support"
2621
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2622
"""A subtrees parameterized Pack repository.
2624
This repository format uses the xml7 serializer to get:
2625
- support for recording full info about the tree root
2626
- support for recording tree-references
2628
This format was introduced in 0.92.
2631
repository_class = KnitPackRepository
2632
_commit_builder_class = PackRootCommitBuilder
2633
rich_root_data = True
2635
supports_tree_reference = True
2637
def _serializer(self):
2638
return xml7.serializer_v7
2639
# What index classes to use
2640
index_builder_class = InMemoryGraphIndex
2641
index_class = GraphIndex
2643
def _get_matching_bzrdir(self):
2644
return bzrdir.format_registry.make_bzrdir(
2645
'pack-0.92-subtree')
2647
def _ignore_setting_bzrdir(self, format):
2650
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2652
def get_format_string(self):
2653
"""See RepositoryFormat.get_format_string()."""
2654
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2656
def get_format_description(self):
2657
"""See RepositoryFormat.get_format_description()."""
2658
return "Packs containing knits with subtree support\n"
2661
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2662
"""A rich-root, no subtrees parameterized Pack repository.
2664
This repository format uses the xml6 serializer to get:
2665
- support for recording full info about the tree root
2667
This format was introduced in 1.0.
2670
repository_class = KnitPackRepository
2671
_commit_builder_class = PackRootCommitBuilder
2672
rich_root_data = True
2673
supports_tree_reference = False
2675
def _serializer(self):
2676
return xml6.serializer_v6
2677
# What index classes to use
2678
index_builder_class = InMemoryGraphIndex
2679
index_class = GraphIndex
2681
def _get_matching_bzrdir(self):
2682
return bzrdir.format_registry.make_bzrdir(
2685
def _ignore_setting_bzrdir(self, format):
2688
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2690
def get_format_string(self):
2691
"""See RepositoryFormat.get_format_string()."""
2692
return ("Bazaar pack repository format 1 with rich root"
2693
" (needs bzr 1.0)\n")
2695
def get_format_description(self):
2696
"""See RepositoryFormat.get_format_description()."""
2697
return "Packs containing knits with rich root support\n"
2700
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2701
"""Repository that supports external references to allow stacking.
2705
Supports external lookups, which results in non-truncated ghosts after
2706
reconcile compared to pack-0.92 formats.
2709
repository_class = KnitPackRepository
2710
_commit_builder_class = PackCommitBuilder
2711
supports_external_lookups = True
2712
# What index classes to use
2713
index_builder_class = InMemoryGraphIndex
2714
index_class = GraphIndex
2717
def _serializer(self):
2718
return xml5.serializer_v5
2720
def _get_matching_bzrdir(self):
2721
return bzrdir.format_registry.make_bzrdir('1.6')
2723
def _ignore_setting_bzrdir(self, format):
2726
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2728
def get_format_string(self):
2729
"""See RepositoryFormat.get_format_string()."""
2730
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2732
def get_format_description(self):
2733
"""See RepositoryFormat.get_format_description()."""
2734
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2737
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2738
"""A repository with rich roots and stacking.
2740
New in release 1.6.1.
2742
Supports stacking on other repositories, allowing data to be accessed
2743
without being stored locally.
2746
repository_class = KnitPackRepository
2747
_commit_builder_class = PackRootCommitBuilder
2748
rich_root_data = True
2749
supports_tree_reference = False # no subtrees
2750
supports_external_lookups = True
2751
# What index classes to use
2752
index_builder_class = InMemoryGraphIndex
2753
index_class = GraphIndex
2756
def _serializer(self):
2757
return xml6.serializer_v6
2759
def _get_matching_bzrdir(self):
2760
return bzrdir.format_registry.make_bzrdir(
2763
def _ignore_setting_bzrdir(self, format):
2766
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2768
def get_format_string(self):
2769
"""See RepositoryFormat.get_format_string()."""
2770
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2772
def get_format_description(self):
2773
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2776
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2777
"""A repository with rich roots and external references.
2781
Supports external lookups, which results in non-truncated ghosts after
2782
reconcile compared to pack-0.92 formats.
2784
This format was deprecated because the serializer it uses accidentally
2785
supported subtrees, when the format was not intended to. This meant that
2786
someone could accidentally fetch from an incorrect repository.
2789
repository_class = KnitPackRepository
2790
_commit_builder_class = PackRootCommitBuilder
2791
rich_root_data = True
2792
supports_tree_reference = False # no subtrees
2794
supports_external_lookups = True
2795
# What index classes to use
2796
index_builder_class = InMemoryGraphIndex
2797
index_class = GraphIndex
2800
def _serializer(self):
2801
return xml7.serializer_v7
2803
def _get_matching_bzrdir(self):
2804
matching = bzrdir.format_registry.make_bzrdir(
2806
matching.repository_format = self
2809
def _ignore_setting_bzrdir(self, format):
2812
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2814
def get_format_string(self):
2815
"""See RepositoryFormat.get_format_string()."""
2816
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2818
def get_format_description(self):
2819
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2823
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2824
"""A repository with stacking and btree indexes,
2825
without rich roots or subtrees.
2827
This is equivalent to pack-1.6 with B+Tree indices.
2830
repository_class = KnitPackRepository
2831
_commit_builder_class = PackCommitBuilder
2832
supports_external_lookups = True
2833
# What index classes to use
2834
index_builder_class = btree_index.BTreeBuilder
2835
index_class = btree_index.BTreeGraphIndex
2838
def _serializer(self):
2839
return xml5.serializer_v5
2841
def _get_matching_bzrdir(self):
2842
return bzrdir.format_registry.make_bzrdir('1.9')
2844
def _ignore_setting_bzrdir(self, format):
2847
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2849
def get_format_string(self):
2850
"""See RepositoryFormat.get_format_string()."""
2851
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2853
def get_format_description(self):
2854
"""See RepositoryFormat.get_format_description()."""
2855
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2858
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2859
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2861
1.6-rich-root with B+Tree indices.
2864
repository_class = KnitPackRepository
2865
_commit_builder_class = PackRootCommitBuilder
2866
rich_root_data = True
2867
supports_tree_reference = False # no subtrees
2868
supports_external_lookups = True
2869
# What index classes to use
2870
index_builder_class = btree_index.BTreeBuilder
2871
index_class = btree_index.BTreeGraphIndex
2874
def _serializer(self):
2875
return xml6.serializer_v6
2877
def _get_matching_bzrdir(self):
2878
return bzrdir.format_registry.make_bzrdir(
2881
def _ignore_setting_bzrdir(self, format):
2884
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2886
def get_format_string(self):
2887
"""See RepositoryFormat.get_format_string()."""
2888
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2890
def get_format_description(self):
2891
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2894
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2895
"""A subtrees development repository.
2897
This format should be retained until the second release after bzr 1.7.
2899
1.6.1-subtree[as it might have been] with B+Tree indices.
2901
This is [now] retained until we have a CHK based subtree format in
2905
repository_class = KnitPackRepository
2906
_commit_builder_class = PackRootCommitBuilder
2907
rich_root_data = True
2909
supports_tree_reference = True
2910
supports_external_lookups = True
2911
# What index classes to use
2912
index_builder_class = btree_index.BTreeBuilder
2913
index_class = btree_index.BTreeGraphIndex
2916
def _serializer(self):
2917
return xml7.serializer_v7
2919
def _get_matching_bzrdir(self):
2920
return bzrdir.format_registry.make_bzrdir(
2921
'development5-subtree')
2923
def _ignore_setting_bzrdir(self, format):
2926
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2928
def get_format_string(self):
2929
"""See RepositoryFormat.get_format_string()."""
2930
return ("Bazaar development format 2 with subtree support "
2931
"(needs bzr.dev from before 1.8)\n")
2933
def get_format_description(self):
2934
"""See RepositoryFormat.get_format_description()."""
2935
return ("Development repository format, currently the same as "
2936
"1.6.1-subtree with B+Tree indices.\n")
2332
class RetryPackOperations(errors.RetryWithNewPacks):
2333
"""Raised when we are packing and we find a missing file.
2335
Meant as a signaling exception, to tell the RepositoryPackCollection.pack
2336
code it should try again.
2339
internal_error = True
2341
_fmt = ("Pack files have changed, reload and try pack again."
2342
" context: %(context)s %(orig_error)s")
2345
class _DirectPackAccess(object):
2346
"""Access to data in one or more packs with less translation."""
2348
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
2349
"""Create a _DirectPackAccess object.
2351
:param index_to_packs: A dict mapping index objects to the transport
2352
and file names for obtaining data.
2353
:param reload_func: A function to call if we determine that the pack
2354
files have moved and we need to reload our caches. See
2355
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
2357
self._container_writer = None
2358
self._write_index = None
2359
self._indices = index_to_packs
2360
self._reload_func = reload_func
2361
self._flush_func = flush_func
2363
def add_raw_records(self, key_sizes, raw_data):
2364
"""Add raw knit bytes to a storage area.
2366
The data is spooled to the container writer in one bytes-record per
2369
:param sizes: An iterable of tuples containing the key and size of each
2371
:param raw_data: A bytestring containing the data.
2372
:return: A list of memos to retrieve the record later. Each memo is an
2373
opaque index memo. For _DirectPackAccess the memo is (index, pos,
2374
length), where the index field is the write_index object supplied
2375
to the PackAccess object.
2377
if type(raw_data) is not str:
2378
raise AssertionError(
2379
'data must be plain bytes was %s' % type(raw_data))
2382
for key, size in key_sizes:
2383
p_offset, p_length = self._container_writer.add_bytes_record(
2384
raw_data[offset:offset+size], [])
2386
result.append((self._write_index, p_offset, p_length))
2390
"""Flush pending writes on this access object.
2392
This will flush any buffered writes to a NewPack.
2394
if self._flush_func is not None:
2397
def get_raw_records(self, memos_for_retrieval):
2398
"""Get the raw bytes for a records.
2400
:param memos_for_retrieval: An iterable containing the (index, pos,
2401
length) memo for retrieving the bytes. The Pack access method
2402
looks up the pack to use for a given record in its index_to_pack
2404
:return: An iterator over the bytes of the records.
2406
# first pass, group into same-index requests
2408
current_index = None
2409
for (index, offset, length) in memos_for_retrieval:
2410
if current_index == index:
2411
current_list.append((offset, length))
2413
if current_index is not None:
2414
request_lists.append((current_index, current_list))
2415
current_index = index
2416
current_list = [(offset, length)]
2417
# handle the last entry
2418
if current_index is not None:
2419
request_lists.append((current_index, current_list))
2420
for index, offsets in request_lists:
2422
transport, path = self._indices[index]
2424
# A KeyError here indicates that someone has triggered an index
2425
# reload, and this index has gone missing, we need to start
2427
if self._reload_func is None:
2428
# If we don't have a _reload_func there is nothing that can
2431
raise errors.RetryWithNewPacks(index,
2432
reload_occurred=True,
2433
exc_info=sys.exc_info())
2435
reader = pack.make_readv_reader(transport, path, offsets)
2436
for names, read_func in reader.iter_records():
2437
yield read_func(None)
2438
except errors.NoSuchFile:
2439
# A NoSuchFile error indicates that a pack file has gone
2440
# missing on disk, we need to trigger a reload, and start over.
2441
if self._reload_func is None:
2443
raise errors.RetryWithNewPacks(transport.abspath(path),
2444
reload_occurred=False,
2445
exc_info=sys.exc_info())
2447
def set_writer(self, writer, index, transport_packname):
2448
"""Set a writer to use for adding data."""
2449
if index is not None:
2450
self._indices[index] = transport_packname
2451
self._container_writer = writer
2452
self._write_index = index
2454
def reload_or_raise(self, retry_exc):
2455
"""Try calling the reload function, or re-raise the original exception.
2457
This should be called after _DirectPackAccess raises a
2458
RetryWithNewPacks exception. This function will handle the common logic
2459
of determining when the error is fatal versus being temporary.
2460
It will also make sure that the original exception is raised, rather
2461
than the RetryWithNewPacks exception.
2463
If this function returns, then the calling function should retry
2464
whatever operation was being performed. Otherwise an exception will
2467
:param retry_exc: A RetryWithNewPacks exception.
2470
if self._reload_func is None:
2472
elif not self._reload_func():
2473
# The reload claimed that nothing changed
2474
if not retry_exc.reload_occurred:
2475
# If there wasn't an earlier reload, then we really were
2476
# expecting to find changes. We didn't find them, so this is a
2480
exc_class, exc_value, exc_traceback = retry_exc.exc_info
2481
raise exc_class, exc_value, exc_traceback
added
removed
bzrlib/repofmt/pack_repo.py
