21
20
that has merged into it. As the first step of a merge, pull, or
22
21
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.
25
from __future__ import absolute_import
36
import bzrlib.errors as errors
37
from bzrlib.errors import InstallFailed
38
from bzrlib.progress import ProgressPhase
39
from bzrlib.revision import is_null, NULL_REVISION
40
from bzrlib.symbol_versioning import (deprecated_function,
43
from bzrlib.tsort import topo_sort
29
from bzrlib.lazy_import import lazy_import
30
lazy_import(globals(), """
41
from bzrlib.i18n import gettext
42
from bzrlib.revision import NULL_REVISION
44
43
from bzrlib.trace import mutter
46
from bzrlib.versionedfile import filter_absent, FulltextContentFactory
48
# TODO: Avoid repeatedly opening weaves so many times.
50
# XXX: This doesn't handle ghost (not present in branch) revisions at
51
# all yet. I'm not sure they really should be supported.
53
# NOTE: This doesn't copy revisions which may be present but not
54
# merged into the last revision. I'm not sure we want to do that.
56
# - get a list of revisions that need to be pulled in
57
# - for each one, pull in that revision file
58
# and get the inventory, and store the inventory with right
60
# - and get the ancestry, and store that with right parents too
61
# - and keep a note of all file ids and version seen
62
# - then go through all files; for each one get the weave,
63
# and add in all file versions
66
46
class RepoFetcher(object):
67
47
"""Pull revisions and texts from one repository to another.
70
if set, try to limit to the data this revision references.
73
count_copied -- number of revisions copied
75
49
This should not be used directly, it's essential a object to encapsulate
76
50
the logic in InterRepository.fetch().
79
def __init__(self, to_repository, from_repository, last_revision=None, pb=None,
53
def __init__(self, to_repository, from_repository, last_revision=None,
54
find_ghosts=True, fetch_spec=None):
81
55
"""Create a repo fetcher.
57
:param last_revision: If set, try to limit to the data this revision
59
:param fetch_spec: A SearchResult specifying which revisions to fetch.
60
If set, this overrides last_revision.
83
61
:param find_ghosts: If True search the entire history for ghosts.
86
self.failed_revisions = []
88
if to_repository.has_same_location(from_repository):
89
# repository.fetch should be taking care of this case.
90
raise errors.BzrError('RepoFetcher run '
91
'between two objects at the same location: '
92
'%r and %r' % (to_repository, from_repository))
63
# repository.fetch has the responsibility for short-circuiting
64
# attempts to copy between a repository and itself.
93
65
self.to_repository = to_repository
94
66
self.from_repository = from_repository
67
self.sink = to_repository._get_sink()
95
68
# must not mutate self._last_revision as its potentially a shared instance
96
69
self._last_revision = last_revision
70
self._fetch_spec = fetch_spec
97
71
self.find_ghosts = find_ghosts
99
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
100
self.nested_pb = self.pb
103
self.nested_pb = None
104
72
self.from_repository.lock_read()
73
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
74
self.from_repository, self.from_repository._format,
75
self.to_repository, self.to_repository._format)
106
self.to_repository.lock_write()
108
self.to_repository.start_write_group()
112
self.to_repository.abort_write_group()
115
self.to_repository.commit_write_group()
118
if self.nested_pb is not None:
119
self.nested_pb.finished()
121
self.to_repository.unlock()
123
79
self.from_repository.unlock()
125
81
def __fetch(self):
126
82
"""Primary worker function.
128
This initialises all the needed variables, and then fetches the
84
This initialises all the needed variables, and then fetches the
129
85
requested revisions, finally clearing the progress bar.
87
# Roughly this is what we're aiming for fetch to become:
89
# missing = self.sink.insert_stream(self.source.get_stream(search))
91
# missing = self.sink.insert_stream(self.source.get_items(missing))
131
93
self.count_total = 0
132
94
self.file_ids_names = {}
133
pp = ProgressPhase('Transferring', 4, self.pb)
95
pb = ui.ui_factory.nested_progress_bar()
96
pb.show_pct = pb.show_count = False
136
search = self._revids_to_fetch()
98
pb.update(gettext("Finding revisions"), 0, 2)
99
search_result = self._revids_to_fetch()
100
mutter('fetching: %s', search_result)
101
if search_result.is_empty():
139
if getattr(self, '_fetch_everything_for_search', None) is not None:
140
self._fetch_everything_for_search(search, pp)
142
# backward compatibility
143
self._fetch_everything_for_revisions(search.get_keys, pp)
103
pb.update(gettext("Fetching revisions"), 1, 2)
104
self._fetch_everything_for_search(search_result)
147
def _fetch_everything_for_search(self, search, pp):
108
def _fetch_everything_for_search(self, search):
148
109
"""Fetch all data for the given set of revisions."""
149
110
# The first phase is "file". We pass the progress bar for it directly
150
111
# into item_keys_introduced_by, which has more information about how
154
115
# item_keys_introduced_by should have a richer API than it does at the
155
116
# moment, so that it can feed the progress information back to this
158
pb = bzrlib.ui.ui_factory.nested_progress_bar()
118
if (self.from_repository._format.rich_root_data and
119
not self.to_repository._format.rich_root_data):
120
raise errors.IncompatibleRepositories(
121
self.from_repository, self.to_repository,
122
"different rich-root support")
123
pb = ui.ui_factory.nested_progress_bar()
160
revs = search.get_keys()
161
graph = self.from_repository.get_graph()
162
revs = list(graph.iter_topo_order(revs))
163
data_to_fetch = self.from_repository.item_keys_introduced_by(revs,
166
for knit_kind, file_id, revisions in data_to_fetch:
167
if knit_kind != phase:
169
# Make a new progress bar for this phase
172
pb = bzrlib.ui.ui_factory.nested_progress_bar()
173
if knit_kind == "file":
174
# Accumulate file texts
175
text_keys.extend([(file_id, revision) for revision in
177
elif knit_kind == "inventory":
178
# Now copy the file texts.
179
to_texts = self.to_repository.texts
180
from_texts = self.from_repository.texts
181
to_texts.insert_record_stream(from_texts.get_record_stream(
182
text_keys, 'topological', False))
183
# Cause an error if a text occurs after we have done the
186
# Before we process the inventory we generate the root
187
# texts (if necessary) so that the inventories references
189
self._generate_root_texts(revs)
190
# NB: This currently reopens the inventory weave in source;
191
# using a single stream interface instead would avoid this.
192
self._fetch_inventory_weave(revs, pb)
193
elif knit_kind == "signatures":
194
# Nothing to do here; this will be taken care of when
195
# _fetch_revision_texts happens.
197
elif knit_kind == "revisions":
198
self._fetch_revision_texts(revs, pb)
200
raise AssertionError("Unknown knit kind %r" % knit_kind)
125
pb.update("Get stream source")
126
source = self.from_repository._get_source(
127
self.to_repository._format)
128
stream = source.get_stream(search)
129
from_format = self.from_repository._format
130
pb.update("Inserting stream")
131
resume_tokens, missing_keys = self.sink.insert_stream(
132
stream, from_format, [])
134
pb.update("Missing keys")
135
stream = source.get_stream_for_missing_keys(missing_keys)
136
pb.update("Inserting missing keys")
137
resume_tokens, missing_keys = self.sink.insert_stream(
138
stream, from_format, resume_tokens)
140
raise AssertionError(
141
"second push failed to complete a fetch %r." % (
144
raise AssertionError(
145
"second push failed to commit the fetch %r." % (
147
pb.update("Finishing stream")
204
self.count_copied += len(revs)
206
152
def _revids_to_fetch(self):
207
153
"""Determines the exact revisions needed from self.from_repository to
208
154
install self._last_revision in self.to_repository.
210
If no revisions need to be fetched, then this just returns None.
156
:returns: A SearchResult of some sort. (Possibly a
157
PendingAncestryResult, EmptySearchResult, etc.)
212
mutter('fetch up to rev {%s}', self._last_revision)
213
if self._last_revision is NULL_REVISION:
159
if self._fetch_spec is not None:
160
# The fetch spec is already a concrete search result.
161
return self._fetch_spec
162
elif self._last_revision == NULL_REVISION:
163
# fetch_spec is None + last_revision is null => empty fetch.
214
164
# explicit limit of no revisions needed
216
if (self._last_revision is not None and
217
self.to_repository.has_revision(self._last_revision)):
220
return self.to_repository.search_missing_revision_ids(
221
self.from_repository, self._last_revision,
222
find_ghosts=self.find_ghosts)
223
except errors.NoSuchRevision, e:
224
raise InstallFailed([self._last_revision])
226
def _fetch_inventory_weave(self, revs, pb):
227
pb.update("fetch inventory", 0, 2)
228
to_weave = self.to_repository.inventories
229
child_pb = bzrlib.ui.ui_factory.nested_progress_bar()
231
# just merge, this is optimisable and its means we don't
232
# copy unreferenced data such as not-needed inventories.
233
pb.update("fetch inventory", 1, 3)
234
from_weave = self.from_repository.inventories
235
pb.update("fetch inventory", 2, 3)
236
# we fetch only the referenced inventories because we do not
237
# know for unselected inventories whether all their required
238
# texts are present in the other repository - it could be
240
to_weave.insert_record_stream(from_weave.get_record_stream(
241
[(rev_id,) for rev_id in revs],
242
'topological', False))
246
def _generate_root_texts(self, revs):
247
"""This will be called by __fetch between fetching weave texts and
248
fetching the inventory weave.
250
Subclasses should override this if they need to generate root texts
251
after fetching weave texts.
256
class GenericRepoFetcher(RepoFetcher):
257
"""This is a generic repo to repo fetcher.
259
This makes minimal assumptions about repo layout and contents.
260
It triggers a reconciliation after fetching to ensure integrity.
263
def _fetch_revision_texts(self, revs, pb):
264
"""Fetch revision object texts"""
268
pb.update('copying revisions', count, total)
270
sig_text = self.from_repository.get_signature_text(rev)
271
self.to_repository.add_signature_text(rev, sig_text)
272
except errors.NoSuchRevision:
275
self._copy_revision(rev)
277
# fixup inventory if needed:
278
# this is expensive because we have no inverse index to current ghosts.
279
# but on local disk its a few seconds and sftp push is already insane.
281
# FIXME: repository should inform if this is needed.
282
self.to_repository.reconcile()
284
def _copy_revision(self, rev_id):
285
rev = self.from_repository.get_revision(rev_id)
286
self.to_repository.add_revision(rev_id, rev)
289
class KnitRepoFetcher(RepoFetcher):
290
"""This is a knit format repository specific fetcher.
292
This differs from the GenericRepoFetcher by not doing a
293
reconciliation after copying, and using knit joining to
297
def _fetch_revision_texts(self, revs, pb):
298
# may need to be a InterRevisionStore call here.
299
to_sf = self.to_repository.signatures
300
from_sf = self.from_repository.signatures
301
# A missing signature is just skipped.
302
to_sf.insert_record_stream(filter_absent(from_sf.get_record_stream(
303
[(rev_id,) for rev_id in revs],
304
'unordered', False)))
305
self._fetch_just_revision_texts(revs)
307
def _fetch_just_revision_texts(self, version_ids):
308
to_rf = self.to_repository.revisions
309
from_rf = self.from_repository.revisions
310
to_rf.insert_record_stream(from_rf.get_record_stream(
311
[(rev_id,) for rev_id in version_ids],
312
'topological', False))
165
return vf_search.EmptySearchResult()
166
elif self._last_revision is not None:
167
return vf_search.NotInOtherForRevs(self.to_repository,
168
self.from_repository, [self._last_revision],
169
find_ghosts=self.find_ghosts).execute()
170
else: # self._last_revision is None:
171
return vf_search.EverythingNotInOther(self.to_repository,
172
self.from_repository,
173
find_ghosts=self.find_ghosts).execute()
315
176
class Inter1and2Helper(object):
316
177
"""Helper for operations that convert data from model 1 and 2
318
179
This is for use by fetchers and converters.
321
def __init__(self, source, target):
182
# This is a class variable so that the test suite can override it.
183
known_graph_threshold = 100
185
def __init__(self, source):
324
188
:param source: The repository data comes from
325
:param target: The repository data goes to
327
190
self.source = source
330
192
def iter_rev_trees(self, revs):
331
193
"""Iterate through RevisionTrees efficiently.
385
243
# yet, and are unlikely to in non-rich-root environments anyway.
386
244
root_id_order.sort(key=operator.itemgetter(0))
387
245
# Create a record stream containing the roots to create.
389
for key in root_id_order:
390
root_id, rev_id = key
391
rev_parents = parent_map[rev_id]
392
# We drop revision parents with different file-ids, because
393
# that represents a rename of the root to a different location
394
# - its not actually a parent for us. (We could look for that
395
# file id in the revision tree at considerably more expense,
396
# but for now this is sufficient (and reconcile will catch and
397
# correct this anyway).
398
# When a parent revision is a ghost, we guess that its root id
399
# was unchanged (rather than trimming it from the parent list).
400
parent_keys = tuple((root_id, parent) for parent in rev_parents
401
if parent != NULL_REVISION and
402
rev_id_to_root_id.get(parent, root_id) == root_id)
403
yield FulltextContentFactory(key, parent_keys, None, '')
404
to_texts.insert_record_stream(yield_roots())
406
def regenerate_inventory(self, revs):
407
"""Generate a new inventory versionedfile in target, convertin data.
409
The inventory is retrieved from the source, (deserializing it), and
410
stored in the target (reserializing it in a different format).
411
:param revs: The revisions to include
413
for tree in self.iter_rev_trees(revs):
414
parents = tree.get_parent_ids()
415
self.target.add_inventory(tree.get_revision_id(), tree.inventory,
418
def fetch_revisions(self, revision_ids):
419
for revision in self.source.get_revisions(revision_ids):
420
self.target.add_revision(revision.revision_id, revision)
423
class Model1toKnit2Fetcher(GenericRepoFetcher):
424
"""Fetch from a Model1 repository into a Knit2 repository
426
def __init__(self, to_repository, from_repository, last_revision=None,
427
pb=None, find_ghosts=True):
428
self.helper = Inter1and2Helper(from_repository, to_repository)
429
GenericRepoFetcher.__init__(self, to_repository, from_repository,
430
last_revision, pb, find_ghosts)
432
def _generate_root_texts(self, revs):
433
self.helper.generate_root_texts(revs)
435
def _fetch_inventory_weave(self, revs, pb):
436
self.helper.regenerate_inventory(revs)
438
def _copy_revision(self, rev):
439
self.helper.fetch_revisions([rev])
442
class Knit1to2Fetcher(KnitRepoFetcher):
443
"""Fetch from a Knit1 repository into a Knit2 repository"""
445
def __init__(self, to_repository, from_repository, last_revision=None,
446
pb=None, find_ghosts=True):
447
self.helper = Inter1and2Helper(from_repository, to_repository)
448
KnitRepoFetcher.__init__(self, to_repository, from_repository,
449
last_revision, pb, find_ghosts)
451
def _generate_root_texts(self, revs):
452
self.helper.generate_root_texts(revs)
454
def _fetch_inventory_weave(self, revs, pb):
455
self.helper.regenerate_inventory(revs)
457
def _fetch_just_revision_texts(self, version_ids):
458
self.helper.fetch_revisions(version_ids)
246
if len(revs) > self.known_graph_threshold:
247
graph = self.source.get_known_graph_ancestry(revs)
248
new_roots_stream = _new_root_data_stream(
249
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
250
return [('texts', new_roots_stream)]
253
def _new_root_data_stream(
254
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
255
"""Generate a texts substream of synthesised root entries.
257
Used in fetches that do rich-root upgrades.
259
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
260
the root entries to create.
261
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
262
calculating the parents. If a parent rev_id is not found here then it
263
will be recalculated.
264
:param parent_map: a parent map for all the revisions in
266
:param graph: a graph to use instead of repo.get_graph().
268
for root_key in root_keys_to_create:
269
root_id, rev_id = root_key
270
parent_keys = _parent_keys_for_root_version(
271
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
272
yield versionedfile.FulltextContentFactory(
273
root_key, parent_keys, None, '')
276
def _parent_keys_for_root_version(
277
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
278
"""Get the parent keys for a given root id.
280
A helper function for _new_root_data_stream.
282
# Include direct parents of the revision, but only if they used the same
283
# root_id and are heads.
284
rev_parents = parent_map[rev_id]
286
for parent_id in rev_parents:
287
if parent_id == NULL_REVISION:
289
if parent_id not in rev_id_to_root_id_map:
290
# We probably didn't read this revision, go spend the extra effort
293
tree = repo.revision_tree(parent_id)
294
except errors.NoSuchRevision:
295
# Ghost, fill out rev_id_to_root_id in case we encounter this
297
# But set parent_root_id to None since we don't really know
298
parent_root_id = None
300
parent_root_id = tree.get_root_id()
301
rev_id_to_root_id_map[parent_id] = None
303
# rev_id_to_root_id_map[parent_id] = parent_root_id
304
# memory consumption maybe?
306
parent_root_id = rev_id_to_root_id_map[parent_id]
307
if root_id == parent_root_id:
308
# With stacking we _might_ want to refer to a non-local revision,
309
# but this code path only applies when we have the full content
310
# available, so ghosts really are ghosts, not just the edge of
312
parent_ids.append(parent_id)
314
# root_id may be in the parent anyway.
316
tree = repo.revision_tree(parent_id)
317
except errors.NoSuchRevision:
318
# ghost, can't refer to it.
322
parent_ids.append(tree.get_file_revision(root_id))
323
except errors.NoSuchId:
326
# Drop non-head parents
328
graph = repo.get_graph()
329
heads = graph.heads(parent_ids)
331
for parent_id in parent_ids:
332
if parent_id in heads and parent_id not in selected_ids:
333
selected_ids.append(parent_id)
334
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
338
class TargetRepoKinds(object):
339
"""An enum-like set of constants.
341
They are the possible values of FetchSpecFactory.target_repo_kinds.
344
PREEXISTING = 'preexisting'
349
class FetchSpecFactory(object):
350
"""A helper for building the best fetch spec for a sprout call.
352
Factors that go into determining the sort of fetch to perform:
353
* did the caller specify any revision IDs?
354
* did the caller specify a source branch (need to fetch its
355
heads_to_fetch(), usually the tip + tags)
356
* is there an existing target repo (don't need to refetch revs it
358
* target is stacked? (similar to pre-existing target repo: even if
359
the target itself is new don't want to refetch existing revs)
361
:ivar source_branch: the source branch if one specified, else None.
362
:ivar source_branch_stop_revision_id: fetch up to this revision of
363
source_branch, rather than its tip.
364
:ivar source_repo: the source repository if one found, else None.
365
:ivar target_repo: the target repository acquired by sprout.
366
:ivar target_repo_kind: one of the TargetRepoKinds constants.
370
self._explicit_rev_ids = set()
371
self.source_branch = None
372
self.source_branch_stop_revision_id = None
373
self.source_repo = None
374
self.target_repo = None
375
self.target_repo_kind = None
378
def add_revision_ids(self, revision_ids):
379
"""Add revision_ids to the set of revision_ids to be fetched."""
380
self._explicit_rev_ids.update(revision_ids)
382
def make_fetch_spec(self):
383
"""Build a SearchResult or PendingAncestryResult or etc."""
384
if self.target_repo_kind is None or self.source_repo is None:
385
raise AssertionError(
386
'Incomplete FetchSpecFactory: %r' % (self.__dict__,))
387
if len(self._explicit_rev_ids) == 0 and self.source_branch is None:
388
if self.limit is not None:
389
raise NotImplementedError(
390
"limit is only supported with a source branch set")
391
# Caller hasn't specified any revisions or source branch
392
if self.target_repo_kind == TargetRepoKinds.EMPTY:
393
return vf_search.EverythingResult(self.source_repo)
395
# We want everything not already in the target (or target's
397
return vf_search.EverythingNotInOther(
398
self.target_repo, self.source_repo).execute()
399
heads_to_fetch = set(self._explicit_rev_ids)
400
if self.source_branch is not None:
401
must_fetch, if_present_fetch = self.source_branch.heads_to_fetch()
402
if self.source_branch_stop_revision_id is not None:
403
# Replace the tip rev from must_fetch with the stop revision
404
# XXX: this might be wrong if the tip rev is also in the
405
# must_fetch set for other reasons (e.g. it's the tip of
406
# multiple loom threads?), but then it's pretty unclear what it
407
# should mean to specify a stop_revision in that case anyway.
408
must_fetch.discard(self.source_branch.last_revision())
409
must_fetch.add(self.source_branch_stop_revision_id)
410
heads_to_fetch.update(must_fetch)
412
if_present_fetch = set()
413
if self.target_repo_kind == TargetRepoKinds.EMPTY:
414
# PendingAncestryResult does not raise errors if a requested head
415
# is absent. Ideally it would support the
416
# required_ids/if_present_ids distinction, but in practice
417
# heads_to_fetch will almost certainly be present so this doesn't
419
all_heads = heads_to_fetch.union(if_present_fetch)
420
ret = vf_search.PendingAncestryResult(all_heads, self.source_repo)
421
if self.limit is not None:
422
graph = self.source_repo.get_graph()
423
topo_order = list(graph.iter_topo_order(ret.get_keys()))
424
result_set = topo_order[:self.limit]
425
ret = self.source_repo.revision_ids_to_search_result(result_set)
428
return vf_search.NotInOtherForRevs(self.target_repo, self.source_repo,
429
required_ids=heads_to_fetch, if_present_ids=if_present_fetch,
430
limit=self.limit).execute()
added
removed
bzrlib/fetch.py
