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(), """
40
from bzrlib.revision import NULL_REVISION
34
import bzrlib.errors as errors
35
from bzrlib.errors import InstallFailed
36
from bzrlib.progress import ProgressPhase
37
from bzrlib.revision import is_null, NULL_REVISION
38
from bzrlib.symbol_versioning import (deprecated_function,
41
41
from bzrlib.trace import mutter
43
from bzrlib.versionedfile import filter_absent
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.
70
count_copied -- number of revisions copied
47
72
This should not be used directly, it's essential a object to encapsulate
48
73
the logic in InterRepository.fetch().
51
def __init__(self, to_repository, from_repository, last_revision=None,
52
find_ghosts=True, fetch_spec=None):
76
def __init__(self, to_repository, from_repository, last_revision=None, pb=None,
53
78
"""Create a repo fetcher.
55
:param last_revision: If set, try to limit to the data this revision
57
:param fetch_spec: A SearchResult specifying which revisions to fetch.
58
If set, this overrides last_revision.
59
80
:param find_ghosts: If True search the entire history for ghosts.
61
# repository.fetch has the responsibility for short-circuiting
62
# attempts to copy between a repository and itself.
83
self.failed_revisions = []
85
if to_repository.has_same_location(from_repository):
86
# repository.fetch should be taking care of this case.
87
raise errors.BzrError('RepoFetcher run '
88
'between two objects at the same location: '
89
'%r and %r' % (to_repository, from_repository))
63
90
self.to_repository = to_repository
64
91
self.from_repository = from_repository
65
self.sink = to_repository._get_sink()
66
92
# must not mutate self._last_revision as its potentially a shared instance
67
93
self._last_revision = last_revision
68
self._fetch_spec = fetch_spec
69
94
self.find_ghosts = find_ghosts
96
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
97
self.nested_pb = self.pb
100
self.nested_pb = None
70
101
self.from_repository.lock_read()
71
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
72
self.from_repository, self.from_repository._format,
73
self.to_repository, self.to_repository._format)
103
self.to_repository.lock_write()
105
self.to_repository.start_write_group()
109
self.to_repository.abort_write_group()
112
self.to_repository.commit_write_group()
115
if self.nested_pb is not None:
116
self.nested_pb.finished()
118
self.to_repository.unlock()
77
120
self.from_repository.unlock()
79
122
def __fetch(self):
80
123
"""Primary worker function.
82
This initialises all the needed variables, and then fetches the
125
This initialises all the needed variables, and then fetches the
83
126
requested revisions, finally clearing the progress bar.
85
# Roughly this is what we're aiming for fetch to become:
87
# missing = self.sink.insert_stream(self.source.get_stream(search))
89
# missing = self.sink.insert_stream(self.source.get_items(missing))
128
self.to_weaves = self.to_repository.weave_store
129
self.from_weaves = self.from_repository.weave_store
91
130
self.count_total = 0
92
131
self.file_ids_names = {}
93
pb = ui.ui_factory.nested_progress_bar()
94
pb.show_pct = pb.show_count = False
132
pp = ProgressPhase('Transferring', 4, self.pb)
96
pb.update("Finding revisions", 0, 2)
97
search_result = self._revids_to_fetch()
98
mutter('fetching: %s', search_result)
99
if search_result.is_empty():
135
search = self._revids_to_fetch()
101
pb.update("Fetching revisions", 1, 2)
102
self._fetch_everything_for_search(search_result)
138
if getattr(self, '_fetch_everything_for_search', None) is not None:
139
self._fetch_everything_for_search(search, pp)
141
# backward compatibility
142
self._fetch_everything_for_revisions(search.get_keys, pp)
106
def _fetch_everything_for_search(self, search):
146
def _fetch_everything_for_search(self, search, pp):
107
147
"""Fetch all data for the given set of revisions."""
108
148
# The first phase is "file". We pass the progress bar for it directly
109
149
# into item_keys_introduced_by, which has more information about how
113
153
# item_keys_introduced_by should have a richer API than it does at the
114
154
# moment, so that it can feed the progress information back to this
116
if (self.from_repository._format.rich_root_data and
117
not self.to_repository._format.rich_root_data):
118
raise errors.IncompatibleRepositories(
119
self.from_repository, self.to_repository,
120
"different rich-root support")
121
pb = ui.ui_factory.nested_progress_bar()
157
pb = bzrlib.ui.ui_factory.nested_progress_bar()
123
pb.update("Get stream source")
124
source = self.from_repository._get_source(
125
self.to_repository._format)
126
stream = source.get_stream(search)
127
from_format = self.from_repository._format
128
pb.update("Inserting stream")
129
resume_tokens, missing_keys = self.sink.insert_stream(
130
stream, from_format, [])
132
pb.update("Missing keys")
133
stream = source.get_stream_for_missing_keys(missing_keys)
134
pb.update("Inserting missing keys")
135
resume_tokens, missing_keys = self.sink.insert_stream(
136
stream, from_format, resume_tokens)
138
raise AssertionError(
139
"second push failed to complete a fetch %r." % (
142
raise AssertionError(
143
"second push failed to commit the fetch %r." % (
145
pb.update("Finishing stream")
159
revs = search.get_keys()
160
graph = self.from_repository.get_graph()
161
revs = list(graph.iter_topo_order(revs))
162
data_to_fetch = self.from_repository.item_keys_introduced_by(revs,
164
for knit_kind, file_id, revisions in data_to_fetch:
165
if knit_kind != phase:
167
# Make a new progress bar for this phase
170
pb = bzrlib.ui.ui_factory.nested_progress_bar()
171
if knit_kind == "file":
172
self._fetch_weave_text(file_id, revisions)
173
elif knit_kind == "inventory":
174
# Before we process the inventory we generate the root
175
# texts (if necessary) so that the inventories references
177
self._generate_root_texts(revs)
178
# NB: This currently reopens the inventory weave in source;
179
# using a full get_data_stream instead would avoid this.
180
self._fetch_inventory_weave(revs, pb)
181
elif knit_kind == "signatures":
182
# Nothing to do here; this will be taken care of when
183
# _fetch_revision_texts happens.
185
elif knit_kind == "revisions":
186
self._fetch_revision_texts(revs, pb)
188
raise AssertionError("Unknown knit kind %r" % knit_kind)
192
self.count_copied += len(revs)
150
194
def _revids_to_fetch(self):
151
195
"""Determines the exact revisions needed from self.from_repository to
152
196
install self._last_revision in self.to_repository.
154
:returns: A SearchResult of some sort. (Possibly a
155
PendingAncestryResult, EmptySearchResult, etc.)
198
If no revisions need to be fetched, then this just returns None.
157
if self._fetch_spec is not None:
158
# The fetch spec is already a concrete search result.
159
return self._fetch_spec
160
elif self._last_revision == NULL_REVISION:
161
# fetch_spec is None + last_revision is null => empty fetch.
200
mutter('fetch up to rev {%s}', self._last_revision)
201
if self._last_revision is NULL_REVISION:
162
202
# explicit limit of no revisions needed
163
return _mod_graph.EmptySearchResult()
164
elif self._last_revision is not None:
165
return _mod_graph.NotInOtherForRevs(self.to_repository,
166
self.from_repository, [self._last_revision],
167
find_ghosts=self.find_ghosts).execute()
168
else: # self._last_revision is None:
169
return _mod_graph.EverythingNotInOther(self.to_repository,
170
self.from_repository,
171
find_ghosts=self.find_ghosts).execute()
204
if (self._last_revision is not None and
205
self.to_repository.has_revision(self._last_revision)):
208
return self.to_repository.search_missing_revision_ids(
209
self.from_repository, self._last_revision,
210
find_ghosts=self.find_ghosts)
211
except errors.NoSuchRevision:
212
raise InstallFailed([self._last_revision])
214
def _fetch_weave_text(self, file_id, required_versions):
215
to_weave = self.to_weaves.get_weave_or_empty(file_id,
216
self.to_repository.get_transaction())
217
from_weave = self.from_weaves.get_weave(file_id,
218
self.from_repository.get_transaction())
219
# Fetch all the texts.
220
to_weave.insert_record_stream(from_weave.get_record_stream(
221
required_versions, 'topological', False))
223
def _fetch_inventory_weave(self, revs, pb):
224
pb.update("fetch inventory", 0, 2)
225
to_weave = self.to_repository.get_inventory_weave()
226
child_pb = bzrlib.ui.ui_factory.nested_progress_bar()
228
# just merge, this is optimisable and its means we don't
229
# copy unreferenced data such as not-needed inventories.
230
pb.update("fetch inventory", 1, 3)
231
from_weave = self.from_repository.get_inventory_weave()
232
pb.update("fetch inventory", 2, 3)
233
# we fetch only the referenced inventories because we do not
234
# know for unselected inventories whether all their required
235
# texts are present in the other repository - it could be
237
to_weave.insert_record_stream(from_weave.get_record_stream(revs,
238
'topological', False))
242
def _generate_root_texts(self, revs):
243
"""This will be called by __fetch between fetching weave texts and
244
fetching the inventory weave.
246
Subclasses should override this if they need to generate root texts
247
after fetching weave texts.
252
class GenericRepoFetcher(RepoFetcher):
253
"""This is a generic repo to repo fetcher.
255
This makes minimal assumptions about repo layout and contents.
256
It triggers a reconciliation after fetching to ensure integrity.
259
def _fetch_revision_texts(self, revs, pb):
260
"""Fetch revision object texts"""
261
to_txn = self.to_transaction = self.to_repository.get_transaction()
264
to_store = self.to_repository._revision_store
266
pb.update('copying revisions', count, total)
268
sig_text = self.from_repository.get_signature_text(rev)
269
to_store.add_revision_signature_text(rev, sig_text, to_txn)
270
except errors.NoSuchRevision:
273
self._copy_revision(rev, to_txn)
275
# fixup inventory if needed:
276
# this is expensive because we have no inverse index to current ghosts.
277
# but on local disk its a few seconds and sftp push is already insane.
279
# FIXME: repository should inform if this is needed.
280
self.to_repository.reconcile()
282
def _copy_revision(self, rev, to_txn):
283
to_store = self.to_repository._revision_store
284
to_store.add_revision(self.from_repository.get_revision(rev), to_txn)
287
class KnitRepoFetcher(RepoFetcher):
288
"""This is a knit format repository specific fetcher.
290
This differs from the GenericRepoFetcher by not doing a
291
reconciliation after copying, and using knit joining to
295
def _fetch_revision_texts(self, revs, pb):
296
# may need to be a InterRevisionStore call here.
297
from_transaction = self.from_repository.get_transaction()
298
to_transaction = self.to_repository.get_transaction()
299
to_sf = self.to_repository._revision_store.get_signature_file(
301
from_sf = self.from_repository._revision_store.get_signature_file(
303
# A missing signature is just skipped.
304
to_sf.insert_record_stream(filter_absent(from_sf.get_record_stream(revs,
305
'unordered', False)))
306
self._fetch_just_revision_texts(revs, from_transaction, to_transaction)
308
def _fetch_just_revision_texts(self, version_ids, from_transaction,
310
to_rf = self.to_repository._revision_store.get_revision_file(
312
from_rf = self.from_repository._revision_store.get_revision_file(
314
to_rf.insert_record_stream(from_rf.get_record_stream(version_ids,
315
'topological', False))
174
318
class Inter1and2Helper(object):
175
319
"""Helper for operations that convert data from model 1 and 2
177
321
This is for use by fetchers and converters.
180
# This is a class variable so that the test suite can override it.
181
known_graph_threshold = 100
183
def __init__(self, source):
324
def __init__(self, source, target):
186
327
:param source: The repository data comes from
328
:param target: The repository data goes to
188
330
self.source = source
190
333
def iter_rev_trees(self, revs):
191
334
"""Iterate through RevisionTrees efficiently.
222
367
for tree in self.iter_rev_trees(parents):
223
368
root_id = tree.get_root_id()
224
369
revision_root[tree.get_revision_id()] = root_id
370
return revision_root, planned_versions
227
372
def generate_root_texts(self, revs):
228
373
"""Generate VersionedFiles for all root ids.
230
375
:param revs: the revisions to include
377
to_store = self.target.weave_store
232
378
graph = self.source.get_graph()
233
379
parent_map = graph.get_parent_map(revs)
234
rev_order = tsort.topo_sort(parent_map)
235
rev_id_to_root_id = self._find_root_ids(revs, parent_map, graph)
236
root_id_order = [(rev_id_to_root_id[rev_id], rev_id) for rev_id in
238
# Guaranteed stable, this groups all the file id operations together
239
# retaining topological order within the revisions of a file id.
240
# File id splits and joins would invalidate this, but they don't exist
241
# yet, and are unlikely to in non-rich-root environments anyway.
242
root_id_order.sort(key=operator.itemgetter(0))
243
# Create a record stream containing the roots to create.
244
if len(revs) > self.known_graph_threshold:
245
graph = self.source.get_known_graph_ancestry(revs)
246
new_roots_stream = _new_root_data_stream(
247
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
248
return [('texts', new_roots_stream)]
251
def _new_root_data_stream(
252
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
253
"""Generate a texts substream of synthesised root entries.
255
Used in fetches that do rich-root upgrades.
257
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
258
the root entries to create.
259
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
260
calculating the parents. If a parent rev_id is not found here then it
261
will be recalculated.
262
:param parent_map: a parent map for all the revisions in
264
:param graph: a graph to use instead of repo.get_graph().
266
for root_key in root_keys_to_create:
267
root_id, rev_id = root_key
268
parent_keys = _parent_keys_for_root_version(
269
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
270
yield versionedfile.FulltextContentFactory(
271
root_key, parent_keys, None, '')
274
def _parent_keys_for_root_version(
275
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
276
"""Get the parent keys for a given root id.
278
A helper function for _new_root_data_stream.
280
# Include direct parents of the revision, but only if they used the same
281
# root_id and are heads.
282
rev_parents = parent_map[rev_id]
284
for parent_id in rev_parents:
285
if parent_id == NULL_REVISION:
287
if parent_id not in rev_id_to_root_id_map:
288
# We probably didn't read this revision, go spend the extra effort
291
tree = repo.revision_tree(parent_id)
292
except errors.NoSuchRevision:
293
# Ghost, fill out rev_id_to_root_id in case we encounter this
295
# But set parent_root_id to None since we don't really know
296
parent_root_id = None
298
parent_root_id = tree.get_root_id()
299
rev_id_to_root_id_map[parent_id] = None
301
# rev_id_to_root_id_map[parent_id] = parent_root_id
302
# memory consumption maybe?
304
parent_root_id = rev_id_to_root_id_map[parent_id]
305
if root_id == parent_root_id:
306
# With stacking we _might_ want to refer to a non-local revision,
307
# but this code path only applies when we have the full content
308
# available, so ghosts really are ghosts, not just the edge of
310
parent_ids.append(parent_id)
312
# root_id may be in the parent anyway.
314
tree = repo.revision_tree(parent_id)
315
except errors.NoSuchRevision:
316
# ghost, can't refer to it.
320
parent_ids.append(tree.get_file_revision(root_id))
321
except errors.NoSuchId:
324
# Drop non-head parents
326
graph = repo.get_graph()
327
heads = graph.heads(parent_ids)
329
for parent_id in parent_ids:
330
if parent_id in heads and parent_id not in selected_ids:
331
selected_ids.append(parent_id)
332
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
336
class TargetRepoKinds(object):
337
"""An enum-like set of constants.
339
They are the possible values of FetchSpecFactory.target_repo_kinds.
342
PREEXISTING = 'preexisting'
347
class FetchSpecFactory(object):
348
"""A helper for building the best fetch spec for a sprout call.
350
Factors that go into determining the sort of fetch to perform:
351
* did the caller specify any revision IDs?
352
* did the caller specify a source branch (need to fetch its
353
heads_to_fetch(), usually the tip + tags)
354
* is there an existing target repo (don't need to refetch revs it
356
* target is stacked? (similar to pre-existing target repo: even if
357
the target itself is new don't want to refetch existing revs)
359
:ivar source_branch: the source branch if one specified, else None.
360
:ivar source_branch_stop_revision_id: fetch up to this revision of
361
source_branch, rather than its tip.
362
:ivar source_repo: the source repository if one found, else None.
363
:ivar target_repo: the target repository acquired by sprout.
364
:ivar target_repo_kind: one of the TargetRepoKinds constants.
368
self._explicit_rev_ids = set()
369
self.source_branch = None
370
self.source_branch_stop_revision_id = None
371
self.source_repo = None
372
self.target_repo = None
373
self.target_repo_kind = None
376
def add_revision_ids(self, revision_ids):
377
"""Add revision_ids to the set of revision_ids to be fetched."""
378
self._explicit_rev_ids.update(revision_ids)
380
def make_fetch_spec(self):
381
"""Build a SearchResult or PendingAncestryResult or etc."""
382
if self.target_repo_kind is None or self.source_repo is None:
383
raise AssertionError(
384
'Incomplete FetchSpecFactory: %r' % (self.__dict__,))
385
if len(self._explicit_rev_ids) == 0 and self.source_branch is None:
386
if self.limit is not None:
387
raise NotImplementedError(
388
"limit is only supported with a source branch set")
389
# Caller hasn't specified any revisions or source branch
390
if self.target_repo_kind == TargetRepoKinds.EMPTY:
391
return _mod_graph.EverythingResult(self.source_repo)
393
# We want everything not already in the target (or target's
395
return _mod_graph.EverythingNotInOther(
396
self.target_repo, self.source_repo).execute()
397
heads_to_fetch = set(self._explicit_rev_ids)
398
if self.source_branch is not None:
399
must_fetch, if_present_fetch = self.source_branch.heads_to_fetch()
400
if self.source_branch_stop_revision_id is not None:
401
# Replace the tip rev from must_fetch with the stop revision
402
# XXX: this might be wrong if the tip rev is also in the
403
# must_fetch set for other reasons (e.g. it's the tip of
404
# multiple loom threads?), but then it's pretty unclear what it
405
# should mean to specify a stop_revision in that case anyway.
406
must_fetch.discard(self.source_branch.last_revision())
407
must_fetch.add(self.source_branch_stop_revision_id)
408
heads_to_fetch.update(must_fetch)
410
if_present_fetch = set()
411
if self.target_repo_kind == TargetRepoKinds.EMPTY:
412
# PendingAncestryResult does not raise errors if a requested head
413
# is absent. Ideally it would support the
414
# required_ids/if_present_ids distinction, but in practice
415
# heads_to_fetch will almost certainly be present so this doesn't
417
all_heads = heads_to_fetch.union(if_present_fetch)
418
ret = _mod_graph.PendingAncestryResult(all_heads, self.source_repo)
419
if self.limit is not None:
420
graph = self.source_repo.get_graph()
421
topo_order = list(graph.iter_topo_order(ret.get_keys()))
422
result_set = topo_order[:self.limit]
423
ret = self.source_repo.revision_ids_to_search_result(result_set)
426
return _mod_graph.NotInOtherForRevs(self.target_repo, self.source_repo,
427
required_ids=heads_to_fetch, if_present_ids=if_present_fetch,
428
limit=self.limit).execute()
380
revision_root, planned_versions = self._find_root_ids(
381
revs, parent_map, graph)
382
for root_id, versions in planned_versions.iteritems():
383
versionedfile = to_store.get_weave_or_empty(root_id,
384
self.target.get_transaction())
386
for revision_id in versions:
387
if revision_id in versionedfile:
389
parents = parent_map[revision_id]
390
# We drop revision parents with different file-ids, because
391
# a version cannot have a version with another file-id as its
393
# When a parent revision is a ghost, we guess that its root id
395
parents = tuple(p for p in parents if p != NULL_REVISION
396
and revision_root.get(p, root_id) == root_id)
397
result = versionedfile.add_lines_with_ghosts(
398
revision_id, parents, [], parent_texts)
399
parent_texts[revision_id] = result[2]
401
def regenerate_inventory(self, revs):
402
"""Generate a new inventory versionedfile in target, convertin data.
404
The inventory is retrieved from the source, (deserializing it), and
405
stored in the target (reserializing it in a different format).
406
:param revs: The revisions to include
408
for tree in self.iter_rev_trees(revs):
409
parents = tree.get_parent_ids()
410
self.target.add_inventory(tree.get_revision_id(), tree.inventory,
413
def fetch_revisions(self, revision_ids):
414
for revision in self.source.get_revisions(revision_ids):
415
self.target.add_revision(revision.revision_id, revision)
418
class Model1toKnit2Fetcher(GenericRepoFetcher):
419
"""Fetch from a Model1 repository into a Knit2 repository
421
def __init__(self, to_repository, from_repository, last_revision=None,
422
pb=None, find_ghosts=True):
423
self.helper = Inter1and2Helper(from_repository, to_repository)
424
GenericRepoFetcher.__init__(self, to_repository, from_repository,
425
last_revision, pb, find_ghosts)
427
def _generate_root_texts(self, revs):
428
self.helper.generate_root_texts(revs)
430
def _fetch_inventory_weave(self, revs, pb):
431
self.helper.regenerate_inventory(revs)
433
def _copy_revision(self, rev, to_txn):
434
self.helper.fetch_revisions([rev])
437
class Knit1to2Fetcher(KnitRepoFetcher):
438
"""Fetch from a Knit1 repository into a Knit2 repository"""
440
def __init__(self, to_repository, from_repository, last_revision=None,
441
pb=None, find_ghosts=True):
442
self.helper = Inter1and2Helper(from_repository, to_repository)
443
KnitRepoFetcher.__init__(self, to_repository, from_repository,
444
last_revision, pb, find_ghosts)
446
def _generate_root_texts(self, revs):
447
self.helper.generate_root_texts(revs)
449
def _fetch_inventory_weave(self, revs, pb):
450
self.helper.regenerate_inventory(revs)
452
def _fetch_just_revision_texts(self, version_ids, from_transaction,
454
self.helper.fetch_revisions(version_ids)
457
class RemoteToOtherFetcher(GenericRepoFetcher):
459
def _fetch_everything_for_search(self, search, pp):
460
data_stream = self.from_repository.get_data_stream_for_search(search)
461
self.to_repository.insert_data_stream(data_stream)
added
removed
bzrlib/fetch.py
