21
21
that has merged into it. As the first step of a merge, pull, or
22
22
branch operation we copy history from the source into the destination
25
The copying is done in a slightly complicated order. We don't want to
26
add a revision to the store until everything it refers to is also
27
stored, so that if a revision is present we can totally recreate it.
28
However, we can't know what files are included in a revision until we
29
read its inventory. So we query the inventory store of the source for
30
the ids we need, and then pull those ids and then return to the inventories.
28
from bzrlib.lazy_import import lazy_import
29
lazy_import(globals(), """
36
import bzrlib.errors as errors
37
from bzrlib.errors import InstallFailed
38
from bzrlib.progress import ProgressPhase
40
39
from bzrlib.revision import NULL_REVISION
40
from bzrlib.tsort import topo_sort
41
41
from bzrlib.trace import mutter
43
from bzrlib.versionedfile import filter_absent, FulltextContentFactory
45
# TODO: Avoid repeatedly opening weaves so many times.
47
# XXX: This doesn't handle ghost (not present in branch) revisions at
48
# all yet. I'm not sure they really should be supported.
50
# NOTE: This doesn't copy revisions which may be present but not
51
# merged into the last revision. I'm not sure we want to do that.
53
# - get a list of revisions that need to be pulled in
54
# - for each one, pull in that revision file
55
# and get the inventory, and store the inventory with right
57
# - and get the ancestry, and store that with right parents too
58
# - and keep a note of all file ids and version seen
59
# - then go through all files; for each one get the weave,
60
# and add in all file versions
44
63
class RepoFetcher(object):
45
64
"""Pull revisions and texts from one repository to another.
67
if set, try to limit to the data this revision references.
47
69
This should not be used directly, it's essential a object to encapsulate
48
70
the logic in InterRepository.fetch().
51
def __init__(self, to_repository, from_repository, last_revision=None,
52
find_ghosts=True, fetch_spec=None):
73
def __init__(self, to_repository, from_repository, last_revision=None, pb=None,
53
75
"""Create a repo fetcher.
55
:param last_revision: If set, try to limit to the data this revision
57
77
:param find_ghosts: If True search the entire history for ghosts.
78
:param _write_group_acquired_callable: Don't use; this parameter only
79
exists to facilitate a hack done in InterPackRepo.fetch. We would
80
like to remove this parameter.
59
# repository.fetch has the responsibility for short-circuiting
60
# attempts to copy between a repository and itself.
82
if to_repository.has_same_location(from_repository):
83
# repository.fetch should be taking care of this case.
84
raise errors.BzrError('RepoFetcher run '
85
'between two objects at the same location: '
86
'%r and %r' % (to_repository, from_repository))
61
87
self.to_repository = to_repository
62
88
self.from_repository = from_repository
63
89
self.sink = to_repository._get_sink()
64
90
# must not mutate self._last_revision as its potentially a shared instance
65
91
self._last_revision = last_revision
66
self._fetch_spec = fetch_spec
67
92
self.find_ghosts = find_ghosts
94
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
95
self.nested_pb = self.pb
68
99
self.from_repository.lock_read()
69
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
70
self.from_repository, self.from_repository._format,
71
self.to_repository, self.to_repository._format)
104
if self.nested_pb is not None:
105
self.nested_pb.finished()
75
107
self.from_repository.unlock()
110
140
# item_keys_introduced_by should have a richer API than it does at the
111
141
# moment, so that it can feed the progress information back to this
113
if (self.from_repository._format.rich_root_data and
114
not self.to_repository._format.rich_root_data):
115
raise errors.IncompatibleRepositories(
116
self.from_repository, self.to_repository,
117
"different rich-root support")
118
pb = ui.ui_factory.nested_progress_bar()
143
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
120
pb.update("Get stream source")
121
source = self.from_repository._get_source(
122
self.to_repository._format)
123
stream = source.get_stream(search)
124
145
from_format = self.from_repository._format
125
pb.update("Inserting stream")
146
stream = self.get_stream(search, pp)
126
147
resume_tokens, missing_keys = self.sink.insert_stream(
127
148
stream, from_format, [])
128
if self.to_repository._fallback_repositories:
130
self._parent_inventories(search.get_keys()))
132
pb.update("Missing keys")
133
stream = source.get_stream_for_missing_keys(missing_keys)
134
pb.update("Inserting missing keys")
150
stream = self.get_stream_for_missing_keys(missing_keys)
135
151
resume_tokens, missing_keys = self.sink.insert_stream(
136
152
stream, from_format, resume_tokens)
142
158
raise AssertionError(
143
159
"second push failed to commit the fetch %r." % (
145
pb.update("Finishing stream")
146
161
self.sink.finished()
163
if self.pb is not None:
166
def get_stream(self, search, pp):
168
revs = search.get_keys()
169
graph = self.from_repository.get_graph()
170
revs = list(graph.iter_topo_order(revs))
171
data_to_fetch = self.from_repository.item_keys_introduced_by(
174
for knit_kind, file_id, revisions in data_to_fetch:
175
if knit_kind != phase:
177
# Make a new progress bar for this phase
180
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
181
if knit_kind == "file":
182
# Accumulate file texts
183
text_keys.extend([(file_id, revision) for revision in
185
elif knit_kind == "inventory":
186
# Now copy the file texts.
187
from_texts = self.from_repository.texts
188
yield ('texts', from_texts.get_record_stream(
189
text_keys, self.to_repository._format._fetch_order,
190
not self.to_repository._format._fetch_uses_deltas))
191
# Cause an error if a text occurs after we have done the
194
# Before we process the inventory we generate the root
195
# texts (if necessary) so that the inventories references
197
for _ in self._generate_root_texts(revs):
199
# NB: This currently reopens the inventory weave in source;
200
# using a single stream interface instead would avoid this.
201
self.pb.update("fetch inventory", 0, 1)
202
from_weave = self.from_repository.inventories
203
# we fetch only the referenced inventories because we do not
204
# know for unselected inventories whether all their required
205
# texts are present in the other repository - it could be
207
yield ('inventories', from_weave.get_record_stream(
208
[(rev_id,) for rev_id in revs],
209
self.inventory_fetch_order(),
210
not self.delta_on_metadata()))
211
elif knit_kind == "signatures":
212
# Nothing to do here; this will be taken care of when
213
# _fetch_revision_texts happens.
215
elif knit_kind == "revisions":
216
for _ in self._fetch_revision_texts(revs, self.pb):
219
raise AssertionError("Unknown knit kind %r" % knit_kind)
221
def get_stream_for_missing_keys(self, missing_keys):
222
# missing keys can only occur when we are byte copying and not
223
# translating (because translation means we don't send
224
# unreconstructable deltas ever).
226
keys['texts'] = set()
227
keys['revisions'] = set()
228
keys['inventories'] = set()
229
keys['signatures'] = set()
230
for key in missing_keys:
231
keys[key[0]].add(key[1:])
232
if len(keys['revisions']):
233
# If we allowed copying revisions at this point, we could end up
234
# copying a revision without copying its required texts: a
235
# violation of the requirements for repository integrity.
236
raise AssertionError(
237
'cannot copy revisions to fill in missing deltas %s' % (
239
for substream_kind, keys in keys.iteritems():
240
vf = getattr(self.from_repository, substream_kind)
241
# Ask for full texts always so that we don't need more round trips
243
stream = vf.get_record_stream(keys,
244
self.to_repository._format._fetch_order, True)
245
yield substream_kind, stream
150
247
def _revids_to_fetch(self):
151
248
"""Determines the exact revisions needed from self.from_repository to
154
251
If no revisions need to be fetched, then this just returns None.
156
if self._fetch_spec is not None:
157
return self._fetch_spec
158
253
mutter('fetch up to rev {%s}', self._last_revision)
159
254
if self._last_revision is NULL_REVISION:
160
255
# explicit limit of no revisions needed
162
return self.to_repository.search_missing_revision_ids(
163
self.from_repository, self._last_revision,
164
find_ghosts=self.find_ghosts)
166
def _parent_inventories(self, revision_ids):
167
# Find all the parent revisions referenced by the stream, but
168
# not present in the stream, and make sure we send their
170
parent_maps = self.to_repository.get_parent_map(revision_ids)
172
map(parents.update, parent_maps.itervalues())
173
parents.discard(NULL_REVISION)
174
parents.difference_update(revision_ids)
175
missing_keys = set(('inventories', rev_id) for rev_id in parents)
257
if (self._last_revision is not None and
258
self.to_repository.has_revision(self._last_revision)):
261
return self.to_repository.search_missing_revision_ids(
262
self.from_repository, self._last_revision,
263
find_ghosts=self.find_ghosts)
264
except errors.NoSuchRevision, e:
265
raise InstallFailed([self._last_revision])
267
def _fetch_revision_texts(self, revs, pb):
268
# fetch signatures first and then the revision texts
269
# may need to be a InterRevisionStore call here.
270
from_sf = self.from_repository.signatures
271
# A missing signature is just skipped.
272
keys = [(rev_id,) for rev_id in revs]
273
signatures = filter_absent(from_sf.get_record_stream(
275
self.to_repository._format._fetch_order,
276
not self.to_repository._format._fetch_uses_deltas))
277
# If a revision has a delta, this is actually expanded inside the
278
# insert_record_stream code now, which is an alternate fix for
280
from_rf = self.from_repository.revisions
281
revisions = from_rf.get_record_stream(
283
self.to_repository._format._fetch_order,
284
not self.delta_on_metadata())
285
return [('signatures', signatures), ('revisions', revisions)]
287
def _generate_root_texts(self, revs):
288
"""This will be called by __fetch between fetching weave texts and
289
fetching the inventory weave.
291
Subclasses should override this if they need to generate root texts
292
after fetching weave texts.
296
def inventory_fetch_order(self):
297
return self.to_repository._format._fetch_order
299
def delta_on_metadata(self):
300
src_serializer = self.from_repository._format._serializer
301
target_serializer = self.to_repository._format._serializer
302
return (self.to_repository._format._fetch_uses_deltas and
303
src_serializer == target_serializer)
179
306
class Inter1and2Helper(object):
243
373
# yet, and are unlikely to in non-rich-root environments anyway.
244
374
root_id_order.sort(key=operator.itemgetter(0))
245
375
# Create a record stream containing the roots to create.
247
# XXX: not covered by tests, should have a flag to always run
248
# this. -- mbp 20100129
249
graph = self.source_repo.get_known_graph_ancestry(revs)
250
new_roots_stream = _new_root_data_stream(
251
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
252
return [('texts', new_roots_stream)]
255
def _get_rich_root_heads_graph(source_repo, revision_ids):
256
"""Get a Graph object suitable for asking heads() for new rich roots."""
260
def _new_root_data_stream(
261
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
262
"""Generate a texts substream of synthesised root entries.
264
Used in fetches that do rich-root upgrades.
266
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
267
the root entries to create.
268
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
269
calculating the parents. If a parent rev_id is not found here then it
270
will be recalculated.
271
:param parent_map: a parent map for all the revisions in
273
:param graph: a graph to use instead of repo.get_graph().
275
for root_key in root_keys_to_create:
276
root_id, rev_id = root_key
277
parent_keys = _parent_keys_for_root_version(
278
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
279
yield versionedfile.FulltextContentFactory(
280
root_key, parent_keys, None, '')
283
def _parent_keys_for_root_version(
284
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
285
"""Get the parent keys for a given root id.
287
A helper function for _new_root_data_stream.
289
# Include direct parents of the revision, but only if they used the same
290
# root_id and are heads.
291
rev_parents = parent_map[rev_id]
293
for parent_id in rev_parents:
294
if parent_id == NULL_REVISION:
296
if parent_id not in rev_id_to_root_id_map:
297
# We probably didn't read this revision, go spend the extra effort
300
tree = repo.revision_tree(parent_id)
301
except errors.NoSuchRevision:
302
# Ghost, fill out rev_id_to_root_id in case we encounter this
304
# But set parent_root_id to None since we don't really know
305
parent_root_id = None
307
parent_root_id = tree.get_root_id()
308
rev_id_to_root_id_map[parent_id] = None
310
# rev_id_to_root_id_map[parent_id] = parent_root_id
311
# memory consumption maybe?
313
parent_root_id = rev_id_to_root_id_map[parent_id]
314
if root_id == parent_root_id:
315
# With stacking we _might_ want to refer to a non-local revision,
316
# but this code path only applies when we have the full content
317
# available, so ghosts really are ghosts, not just the edge of
319
parent_ids.append(parent_id)
321
# root_id may be in the parent anyway.
323
tree = repo.revision_tree(parent_id)
324
except errors.NoSuchRevision:
325
# ghost, can't refer to it.
329
parent_ids.append(tree.inventory[root_id].revision)
330
except errors.NoSuchId:
333
# Drop non-head parents
335
graph = repo.get_graph()
336
heads = graph.heads(parent_ids)
338
for parent_id in parent_ids:
339
if parent_id in heads and parent_id not in selected_ids:
340
selected_ids.append(parent_id)
341
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
377
for key in root_id_order:
378
root_id, rev_id = key
379
rev_parents = parent_map[rev_id]
380
# We drop revision parents with different file-ids, because
381
# that represents a rename of the root to a different location
382
# - its not actually a parent for us. (We could look for that
383
# file id in the revision tree at considerably more expense,
384
# but for now this is sufficient (and reconcile will catch and
385
# correct this anyway).
386
# When a parent revision is a ghost, we guess that its root id
387
# was unchanged (rather than trimming it from the parent list).
388
parent_keys = tuple((root_id, parent) for parent in rev_parents
389
if parent != NULL_REVISION and
390
rev_id_to_root_id.get(parent, root_id) == root_id)
391
yield FulltextContentFactory(key, parent_keys, None, '')
392
return [('texts', yield_roots())]
395
class Model1toKnit2Fetcher(RepoFetcher):
396
"""Fetch from a Model1 repository into a Knit2 repository
398
def __init__(self, to_repository, from_repository, last_revision=None,
399
pb=None, find_ghosts=True):
400
self.helper = Inter1and2Helper(from_repository)
401
RepoFetcher.__init__(self, to_repository, from_repository,
402
last_revision, pb, find_ghosts)
404
def _generate_root_texts(self, revs):
405
return self.helper.generate_root_texts(revs)
407
def inventory_fetch_order(self):
410
Knit1to2Fetcher = Model1toKnit2Fetcher