304
293
self.bundle.add_info_record(serializer=serializer_format,
305
294
supports_rich_root=supports_rich_root)
296
def iter_file_revisions(self):
297
"""Iterate through all relevant revisions of all files.
299
This is the correct implementation, but is not compatible with bzr.dev,
300
because certain old revisions were not converted correctly, and have
301
the wrong "revision" marker in inventories.
303
transaction = self.repository.get_transaction()
304
altered = self.repository.fileids_altered_by_revision_ids(
306
for file_id, file_revision_ids in altered.iteritems():
307
vf = self.repository.weave_store.get_weave(file_id, transaction)
308
yield vf, file_id, file_revision_ids
310
def iter_file_revisions_aggressive(self):
311
"""Iterate through all relevant revisions of all files.
313
This uses the standard iter_file_revisions to determine what revisions
314
are referred to by inventories, but then uses the versionedfile to
315
determine what the build-dependencies of each required revision.
317
All build dependencies which are not ancestors of the base revision
320
for vf, file_id, file_revision_ids in self.iter_file_revisions():
321
new_revision_ids = set()
322
pending = list(file_revision_ids)
323
while len(pending) > 0:
324
revision_id = pending.pop()
325
if revision_id in new_revision_ids:
327
if revision_id in self.base_ancestry:
329
new_revision_ids.add(revision_id)
330
pending.extend(vf.get_parents(revision_id))
331
yield vf, file_id, new_revision_ids
307
333
def write_files(self):
308
334
"""Write bundle records for all revisions of all files"""
310
altered_fileids = self.repository.fileids_altered_by_revision_ids(
312
for file_id, revision_ids in altered_fileids.iteritems():
313
for revision_id in revision_ids:
314
text_keys.append((file_id, revision_id))
315
self._add_mp_records_keys('file', self.repository.texts, text_keys)
335
for vf, file_id, revision_ids in self.iter_file_revisions():
336
self.add_mp_records('file', file_id, vf, revision_ids)
317
338
def write_revisions(self):
318
339
"""Write bundle records for all revisions and signatures"""
319
inv_vf = self.repository.inventories
320
topological_order = [key[-1] for key in multiparent.topo_iter_keys(
321
inv_vf, self.revision_keys)]
322
revision_order = topological_order
340
inv_vf = self.repository.get_inventory_weave()
341
revision_order = list(multiparent.topo_iter(inv_vf, self.revision_ids))
323
342
if self.target is not None and self.target in self.revision_ids:
324
# Make sure the target revision is always the last entry
325
revision_order = list(topological_order)
326
343
revision_order.remove(self.target)
327
344
revision_order.append(self.target)
328
if self.repository._serializer.support_altered_by_hack:
329
# Repositories that support_altered_by_hack means that
330
# inventories.make_mpdiffs() contains all the data about the tree
331
# shape. Formats without support_altered_by_hack require
332
# chk_bytes/etc, so we use a different code path.
333
self._add_mp_records_keys('inventory', inv_vf,
334
[(revid,) for revid in topological_order])
336
# Inventories should always be added in pure-topological order, so
337
# that we can apply the mpdiff for the child to the parent texts.
338
self._add_inventory_mpdiffs_from_serializer(topological_order)
339
self._add_revision_texts(revision_order)
341
def _add_inventory_mpdiffs_from_serializer(self, revision_order):
342
"""Generate mpdiffs by serializing inventories.
344
The current repository only has part of the tree shape information in
345
the 'inventories' vf. So we use serializer.write_inventory_to_string to
346
get a 'full' representation of the tree shape, and then generate
347
mpdiffs on that data stream. This stream can then be reconstructed on
350
inventory_key_order = [(r,) for r in revision_order]
351
parent_map = self.repository.inventories.get_parent_map(
353
missing_keys = set(inventory_key_order).difference(parent_map)
355
raise errors.RevisionNotPresent(list(missing_keys)[0],
356
self.repository.inventories)
357
inv_to_str = self.repository._serializer.write_inventory_to_string
358
# Make sure that we grab the parent texts first
360
map(just_parents.update, parent_map.itervalues())
361
just_parents.difference_update(parent_map)
362
# Ignore ghost parents
363
present_parents = self.repository.inventories.get_parent_map(
365
ghost_keys = just_parents.difference(present_parents)
366
needed_inventories = list(present_parents) + inventory_key_order
367
needed_inventories = [k[-1] for k in needed_inventories]
369
for inv in self.repository.iter_inventories(needed_inventories):
370
revision_id = inv.revision_id
372
as_bytes = inv_to_str(inv)
373
# The sha1 is validated as the xml/textual form, not as the
374
# form-in-the-repository
375
sha1 = osutils.sha_string(as_bytes)
376
as_lines = osutils.split_lines(as_bytes)
378
all_lines[key] = as_lines
379
if key in just_parents:
380
# We don't transmit those entries
382
# Create an mpdiff for this text, and add it to the output
383
parent_keys = parent_map[key]
384
# See the comment in VF.make_mpdiffs about how this effects
385
# ordering when there are ghosts present. I think we have a latent
387
parent_lines = [all_lines[p_key] for p_key in parent_keys
388
if p_key not in ghost_keys]
389
diff = multiparent.MultiParent.from_lines(
390
as_lines, parent_lines)
391
text = ''.join(diff.to_patch())
392
parent_ids = [k[-1] for k in parent_keys]
393
self.bundle.add_multiparent_record(text, sha1, parent_ids,
394
'inventory', revision_id, None)
396
def _add_revision_texts(self, revision_order):
345
self.add_mp_records('inventory', None, inv_vf, revision_order)
397
346
parent_map = self.repository.get_parent_map(revision_order)
398
revision_to_str = self.repository._serializer.write_revision_to_string
399
revisions = self.repository.get_revisions(revision_order)
400
for revision in revisions:
401
revision_id = revision.revision_id
347
for revision_id in revision_order:
402
348
parents = parent_map.get(revision_id, None)
403
revision_text = revision_to_str(revision)
349
revision_text = self.repository.get_revision_xml(revision_id)
404
350
self.bundle.add_fulltext_record(revision_text, parents,
405
351
'revision', revision_id)
425
371
base = parents[0]
426
372
return base, target
428
def _add_mp_records_keys(self, repo_kind, vf, keys):
374
def add_mp_records(self, repo_kind, file_id, vf, revision_ids):
429
375
"""Add multi-parent diff records to a bundle"""
430
ordered_keys = list(multiparent.topo_iter_keys(vf, keys))
431
mpdiffs = vf.make_mpdiffs(ordered_keys)
432
sha1s = vf.get_sha1s(ordered_keys)
433
parent_map = vf.get_parent_map(ordered_keys)
434
for mpdiff, item_key, in zip(mpdiffs, ordered_keys):
435
sha1 = sha1s[item_key]
436
parents = [key[-1] for key in parent_map[item_key]]
376
revision_ids = list(multiparent.topo_iter(vf, revision_ids))
377
mpdiffs = vf.make_mpdiffs(revision_ids)
378
sha1s = vf.get_sha1s(revision_ids)
379
for mpdiff, revision_id, sha1, in zip(mpdiffs, revision_ids, sha1s):
380
parents = vf.get_parents(revision_id)
437
381
text = ''.join(mpdiff.to_patch())
438
# Infer file id records as appropriate.
439
if len(item_key) == 2:
440
file_id = item_key[0]
443
382
self.bundle.add_multiparent_record(text, sha1, parents, repo_kind,
444
item_key[-1], file_id)
383
revision_id, file_id)
447
386
class BundleInfoV4(object):
554
493
for bytes, metadata, repo_kind, revision_id, file_id in\
555
494
self._container.iter_records():
556
495
if repo_kind == 'info':
557
if self._info is not None:
558
raise AssertionError()
496
assert self._info is None
559
497
self._handle_info(metadata)
560
if (pending_file_records and
561
(repo_kind, file_id) != ('file', current_file)):
562
# Flush the data for a single file - prevents memory
563
# spiking due to buffering all files in memory.
564
self._install_mp_records_keys(self._repository.texts,
565
pending_file_records)
498
if (repo_kind, file_id) != ('file', current_file):
499
if len(pending_file_records) > 0:
500
self._install_mp_records(current_versionedfile,
501
pending_file_records)
566
502
current_file = None
567
del pending_file_records[:]
503
current_versionedfile = None
504
pending_file_records = []
568
505
if len(pending_inventory_records) > 0 and repo_kind != 'inventory':
569
self._install_inventory_records(pending_inventory_records)
506
self._install_inventory_records(inventory_vf,
507
pending_inventory_records)
570
508
pending_inventory_records = []
571
509
if repo_kind == 'inventory':
572
pending_inventory_records.append(((revision_id,), metadata, bytes))
510
if inventory_vf is None:
511
inventory_vf = self._repository.get_inventory_weave()
512
if revision_id not in inventory_vf:
513
pending_inventory_records.append((revision_id, metadata,
573
515
if repo_kind == 'revision':
574
516
target_revision = revision_id
575
517
self._install_revision(revision_id, metadata, bytes)
599
548
records if r not in versionedfile]
600
549
versionedfile.add_mpdiffs(vf_records)
602
def _install_mp_records_keys(self, versionedfile, records):
603
d_func = multiparent.MultiParent.from_patch
605
for key, meta, text in records:
606
# Adapt to tuple interface: A length two key is a file_id,
607
# revision_id pair, a length 1 key is a
608
# revision/signature/inventory. We need to do this because
609
# the metadata extraction from the bundle has not yet been updated
610
# to use the consistent tuple interface itself.
615
parents = [prefix + (parent,) for parent in meta['parents']]
616
vf_records.append((key, parents, meta['sha1'], d_func(text)))
617
versionedfile.add_mpdiffs(vf_records)
619
def _get_parent_inventory_texts(self, inventory_text_cache,
620
inventory_cache, parent_ids):
621
cached_parent_texts = {}
622
remaining_parent_ids = []
623
for parent_id in parent_ids:
624
p_text = inventory_text_cache.get(parent_id, None)
626
remaining_parent_ids.append(parent_id)
628
cached_parent_texts[parent_id] = p_text
630
# TODO: Use inventory_cache to grab inventories we already have in
632
if remaining_parent_ids:
633
# first determine what keys are actually present in the local
634
# inventories object (don't use revisions as they haven't been
636
parent_keys = [(r,) for r in remaining_parent_ids]
637
present_parent_map = self._repository.inventories.get_parent_map(
639
present_parent_ids = []
641
for p_id in remaining_parent_ids:
642
if (p_id,) in present_parent_map:
643
present_parent_ids.append(p_id)
646
to_string = self._source_serializer.write_inventory_to_string
647
for parent_inv in self._repository.iter_inventories(
649
p_text = to_string(parent_inv)
650
inventory_cache[parent_inv.revision_id] = parent_inv
651
cached_parent_texts[parent_inv.revision_id] = p_text
652
inventory_text_cache[parent_inv.revision_id] = p_text
654
parent_texts = [cached_parent_texts[parent_id]
655
for parent_id in parent_ids
656
if parent_id not in ghosts]
659
def _install_inventory_records(self, records):
660
if (self._info['serializer'] == self._repository._serializer.format_num
661
and self._repository._serializer.support_altered_by_hack):
662
return self._install_mp_records_keys(self._repository.inventories,
664
# Use a 10MB text cache, since these are string xml inventories. Note
665
# that 10MB is fairly small for large projects (a single inventory can
666
# be >5MB). Another possibility is to cache 10-20 inventory texts
668
inventory_text_cache = lru_cache.LRUSizeCache(10*1024*1024)
669
# Also cache the in-memory representation. This allows us to create
670
# inventory deltas to apply rather than calling add_inventory from
672
inventory_cache = lru_cache.LRUCache(10)
673
pb = ui.ui_factory.nested_progress_bar()
675
num_records = len(records)
676
for idx, (key, metadata, bytes) in enumerate(records):
677
pb.update('installing inventory', idx, num_records)
678
revision_id = key[-1]
679
parent_ids = metadata['parents']
680
# Note: This assumes the local ghosts are identical to the
681
# ghosts in the source, as the Bundle serialization
682
# format doesn't record ghosts.
683
p_texts = self._get_parent_inventory_texts(inventory_text_cache,
686
# Why does to_lines() take strings as the source, it seems that
687
# it would have to cast to a list of lines, which we get back
688
# as lines and then cast back to a string.
689
target_lines = multiparent.MultiParent.from_patch(bytes
691
inv_text = ''.join(target_lines)
693
sha1 = osutils.sha_string(inv_text)
694
if sha1 != metadata['sha1']:
695
raise errors.BadBundle("Can't convert to target format")
696
# Add this to the cache so we don't have to extract it again.
697
inventory_text_cache[revision_id] = inv_text
698
target_inv = self._source_serializer.read_inventory_from_string(
700
self._handle_root(target_inv, parent_ids)
703
parent_inv = inventory_cache.get(parent_ids[0], None)
705
if parent_inv is None:
706
self._repository.add_inventory(revision_id, target_inv,
709
delta = target_inv._make_delta(parent_inv)
710
self._repository.add_inventory_by_delta(parent_ids[0],
711
delta, revision_id, parent_ids)
712
except errors.UnsupportedInventoryKind:
713
raise errors.IncompatibleRevision(repr(self._repository))
714
inventory_cache[revision_id] = target_inv
551
def _install_inventory_records(self, vf, records):
552
if self._info['serializer'] == self._repository._serializer.format_num:
553
return self._install_mp_records(vf, records)
554
for revision_id, metadata, bytes in records:
555
parent_ids = metadata['parents']
556
parents = [self._repository.get_inventory(p)
558
p_texts = [self._source_serializer.write_inventory_to_string(p)
560
target_lines = multiparent.MultiParent.from_patch(bytes).to_lines(
562
sha1 = osutils.sha_strings(target_lines)
563
if sha1 != metadata['sha1']:
564
raise errors.BadBundle("Can't convert to target format")
565
target_inv = self._source_serializer.read_inventory_from_string(
566
''.join(target_lines))
567
self._handle_root(target_inv, parent_ids)
569
self._repository.add_inventory(revision_id, target_inv,
571
except errors.UnsupportedInventoryKind:
572
raise errors.IncompatibleRevision(repr(self._repository))
718
574
def _handle_root(self, target_inv, parent_ids):
719
575
revision_id = target_inv.revision_id
720
576
if self.update_root:
721
text_key = (target_inv.root.file_id, revision_id)
722
parent_keys = [(target_inv.root.file_id, parent) for
723
parent in parent_ids]
724
self._repository.texts.add_lines(text_key, parent_keys, [])
577
target_inv.root.revision = revision_id
578
store = self._repository.weave_store
579
transaction = self._repository.get_transaction()
580
vf = store.get_weave_or_empty(target_inv.root.file_id, transaction)
581
vf.add_lines(revision_id, parent_ids, [])
725
582
elif not self._repository.supports_rich_root():
726
583
if target_inv.root.revision != revision_id:
727
584
raise errors.IncompatibleRevision(repr(self._repository))
729
587
def _install_revision(self, revision_id, metadata, text):
730
588
if self._repository.has_revision(revision_id):
732
revision = self._source_serializer.read_revision_from_string(text)
733
self._repository.add_revision(revision.revision_id, revision)
590
self._repository._add_revision_text(revision_id, text)
735
592
def _install_signature(self, revision_id, metadata, text):
736
593
transaction = self._repository.get_transaction()
737
if self._repository.has_signature_for_revision_id(revision_id):
594
if self._repository._revision_store.has_signature(revision_id,
739
self._repository.add_signature_text(revision_id, text)
597
self._repository._revision_store.add_revision_signature_text(
598
revision_id, text, transaction)