20
21
that has merged into it. As the first step of a merge, pull, or
21
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.
25
from __future__ import absolute_import
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
from bzrlib.trace import mutter
29
44
from bzrlib.lazy_import import lazy_import
30
lazy_import(globals(), """
41
from bzrlib.i18n import gettext
42
from bzrlib.revision import NULL_REVISION
43
from bzrlib.trace import mutter
46
# TODO: Avoid repeatedly opening weaves so many times.
48
# XXX: This doesn't handle ghost (not present in branch) revisions at
49
# all yet. I'm not sure they really should be supported.
51
# NOTE: This doesn't copy revisions which may be present but not
52
# merged into the last revision. I'm not sure we want to do that.
54
# - get a list of revisions that need to be pulled in
55
# - for each one, pull in that revision file
56
# and get the inventory, and store the inventory with right
58
# - and get the ancestry, and store that with right parents too
59
# - and keep a note of all file ids and version seen
60
# - then go through all files; for each one get the weave,
61
# and add in all file versions
46
64
class RepoFetcher(object):
47
65
"""Pull revisions and texts from one repository to another.
68
if set, try to limit to the data this revision references.
71
count_copied -- number of revisions copied
49
73
This should not be used directly, it's essential a object to encapsulate
50
74
the logic in InterRepository.fetch().
53
def __init__(self, to_repository, from_repository, last_revision=None,
54
find_ghosts=True, fetch_spec=None):
77
def __init__(self, to_repository, from_repository, last_revision=None, pb=None,
55
79
"""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.
61
81
:param find_ghosts: If True search the entire history for ghosts.
63
# repository.fetch has the responsibility for short-circuiting
64
# attempts to copy between a repository and itself.
84
self.failed_revisions = []
86
if to_repository.has_same_location(from_repository):
87
# repository.fetch should be taking care of this case.
88
raise errors.BzrError('RepoFetcher run '
89
'between two objects at the same location: '
90
'%r and %r' % (to_repository, from_repository))
65
91
self.to_repository = to_repository
66
92
self.from_repository = from_repository
67
self.sink = to_repository._get_sink()
68
93
# must not mutate self._last_revision as its potentially a shared instance
69
94
self._last_revision = last_revision
70
self._fetch_spec = fetch_spec
71
95
self.find_ghosts = find_ghosts
97
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
98
self.nested_pb = self.pb
101
self.nested_pb = None
72
102
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)
104
self.to_repository.lock_write()
106
self.to_repository.start_write_group()
110
self.to_repository.abort_write_group()
113
self.to_repository.commit_write_group()
115
if self.nested_pb is not None:
116
self.nested_pb.finished()
117
self.to_repository.unlock()
79
119
self.from_repository.unlock()
81
121
def __fetch(self):
82
122
"""Primary worker function.
84
This initialises all the needed variables, and then fetches the
124
This initialises all the needed variables, and then fetches the
85
125
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))
127
self.to_weaves = self.to_repository.weave_store
128
self.from_weaves = self.from_repository.weave_store
93
129
self.count_total = 0
94
130
self.file_ids_names = {}
95
pb = ui.ui_factory.nested_progress_bar()
96
pb.show_pct = pb.show_count = False
131
pp = ProgressPhase('Transferring', 4, self.pb)
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():
134
search = self._revids_to_fetch()
103
pb.update(gettext("Fetching revisions"), 1, 2)
104
self._fetch_everything_for_search(search_result)
137
if getattr(self, '_fetch_everything_for_search', None) is not None:
138
self._fetch_everything_for_search(search, pp)
140
# backward compatibility
141
self._fetch_everything_for_revisions(search.get_keys, pp)
108
def _fetch_everything_for_search(self, search):
145
def _fetch_everything_for_search(self, search, pp):
109
146
"""Fetch all data for the given set of revisions."""
110
147
# The first phase is "file". We pass the progress bar for it directly
111
148
# into item_keys_introduced_by, which has more information about how
115
152
# item_keys_introduced_by should have a richer API than it does at the
116
153
# moment, so that it can feed the progress information back to this
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()
156
pb = bzrlib.ui.ui_factory.nested_progress_bar()
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")
158
revs = search.get_keys()
159
data_to_fetch = self.from_repository.item_keys_introduced_by(revs, pb)
160
for knit_kind, file_id, revisions in data_to_fetch:
161
if knit_kind != phase:
163
# Make a new progress bar for this phase
166
pb = bzrlib.ui.ui_factory.nested_progress_bar()
167
if knit_kind == "file":
168
self._fetch_weave_text(file_id, revisions)
169
elif knit_kind == "inventory":
170
# Before we process the inventory we generate the root
171
# texts (if necessary) so that the inventories references
173
self._generate_root_texts(revs)
174
# NB: This currently reopens the inventory weave in source;
175
# using a full get_data_stream instead would avoid this.
176
self._fetch_inventory_weave(revs, pb)
177
elif knit_kind == "signatures":
178
# Nothing to do here; this will be taken care of when
179
# _fetch_revision_texts happens.
181
elif knit_kind == "revisions":
182
self._fetch_revision_texts(revs, pb)
184
raise AssertionError("Unknown knit kind %r" % knit_kind)
188
self.count_copied += len(revs)
152
190
def _revids_to_fetch(self):
153
191
"""Determines the exact revisions needed from self.from_repository to
154
192
install self._last_revision in self.to_repository.
156
:returns: A SearchResult of some sort. (Possibly a
157
PendingAncestryResult, EmptySearchResult, etc.)
194
If no revisions need to be fetched, then this just returns None.
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.
196
mutter('fetch up to rev {%s}', self._last_revision)
197
if self._last_revision is NULL_REVISION:
164
198
# explicit limit of no revisions needed
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()
200
if (self._last_revision is not None and
201
self.to_repository.has_revision(self._last_revision)):
204
return self.to_repository.search_missing_revision_ids(
205
self.from_repository, self._last_revision,
206
find_ghosts=self.find_ghosts)
207
except errors.NoSuchRevision:
208
raise InstallFailed([self._last_revision])
210
def _fetch_weave_text(self, file_id, required_versions):
211
to_weave = self.to_weaves.get_weave_or_empty(file_id,
212
self.to_repository.get_transaction())
213
from_weave = self.from_weaves.get_weave(file_id,
214
self.from_repository.get_transaction())
215
# we fetch all the texts, because texts do
216
# not reference anything, and its cheap enough
217
to_weave.join(from_weave, version_ids=required_versions)
219
def _fetch_inventory_weave(self, revs, pb):
220
pb.update("fetch inventory", 0, 2)
221
to_weave = self.to_repository.get_inventory_weave()
222
child_pb = bzrlib.ui.ui_factory.nested_progress_bar()
224
# just merge, this is optimisable and its means we don't
225
# copy unreferenced data such as not-needed inventories.
226
pb.update("fetch inventory", 1, 3)
227
from_weave = self.from_repository.get_inventory_weave()
228
pb.update("fetch inventory", 2, 3)
229
# we fetch only the referenced inventories because we do not
230
# know for unselected inventories whether all their required
231
# texts are present in the other repository - it could be
233
to_weave.join(from_weave, pb=child_pb, msg='merge inventory',
238
def _generate_root_texts(self, revs):
239
"""This will be called by __fetch between fetching weave texts and
240
fetching the inventory weave.
242
Subclasses should override this if they need to generate root texts
243
after fetching weave texts.
248
class GenericRepoFetcher(RepoFetcher):
249
"""This is a generic repo to repo fetcher.
251
This makes minimal assumptions about repo layout and contents.
252
It triggers a reconciliation after fetching to ensure integrity.
255
def _fetch_revision_texts(self, revs, pb):
256
"""Fetch revision object texts"""
257
to_txn = self.to_transaction = self.to_repository.get_transaction()
260
to_store = self.to_repository._revision_store
262
pb.update('copying revisions', count, total)
264
sig_text = self.from_repository.get_signature_text(rev)
265
to_store.add_revision_signature_text(rev, sig_text, to_txn)
266
except errors.NoSuchRevision:
269
to_store.add_revision(self.from_repository.get_revision(rev),
272
# fixup inventory if needed:
273
# this is expensive because we have no inverse index to current ghosts.
274
# but on local disk its a few seconds and sftp push is already insane.
276
# FIXME: repository should inform if this is needed.
277
self.to_repository.reconcile()
280
class KnitRepoFetcher(RepoFetcher):
281
"""This is a knit format repository specific fetcher.
283
This differs from the GenericRepoFetcher by not doing a
284
reconciliation after copying, and using knit joining to
288
def _fetch_revision_texts(self, revs, pb):
289
# may need to be a InterRevisionStore call here.
290
from_transaction = self.from_repository.get_transaction()
291
to_transaction = self.to_repository.get_transaction()
292
to_sf = self.to_repository._revision_store.get_signature_file(
294
from_sf = self.from_repository._revision_store.get_signature_file(
296
to_sf.join(from_sf, version_ids=revs, ignore_missing=True)
297
to_rf = self.to_repository._revision_store.get_revision_file(
299
from_rf = self.from_repository._revision_store.get_revision_file(
301
to_rf.join(from_rf, version_ids=revs)
176
304
class Inter1and2Helper(object):
177
305
"""Helper for operations that convert data from model 1 and 2
179
307
This is for use by fetchers and converters.
182
# This is a class variable so that the test suite can override it.
183
known_graph_threshold = 100
185
def __init__(self, source):
310
def __init__(self, source, target):
188
313
:param source: The repository data comes from
314
:param target: The repository data goes to
190
316
self.source = source
192
319
def iter_rev_trees(self, revs):
193
320
"""Iterate through RevisionTrees efficiently.
203
330
revs = list(revs)
205
332
for tree in self.source.revision_trees(revs[:100]):
206
if tree.root_inventory.revision_id is None:
207
tree.root_inventory.revision_id = tree.get_revision_id()
333
if tree.inventory.revision_id is None:
334
tree.inventory.revision_id = tree.get_revision_id()
209
336
revs = revs[100:]
211
def _find_root_ids(self, revs, parent_map, graph):
213
for tree in self.iter_rev_trees(revs):
214
root_id = tree.get_root_id()
215
revision_id = tree.get_file_revision(root_id, u"")
216
revision_root[revision_id] = root_id
217
# Find out which parents we don't already know root ids for
219
for revision_parents in parent_map.itervalues():
220
parents.update(revision_parents)
221
parents.difference_update(revision_root.keys() + [NULL_REVISION])
222
# Limit to revisions present in the versionedfile
223
parents = graph.get_parent_map(parents).keys()
224
for tree in self.iter_rev_trees(parents):
225
root_id = tree.get_root_id()
226
revision_root[tree.get_revision_id()] = root_id
229
338
def generate_root_texts(self, revs):
230
339
"""Generate VersionedFiles for all root ids.
232
341
:param revs: the revisions to include
234
graph = self.source.get_graph()
235
parent_map = graph.get_parent_map(revs)
236
rev_order = tsort.topo_sort(parent_map)
237
rev_id_to_root_id = self._find_root_ids(revs, parent_map, graph)
238
root_id_order = [(rev_id_to_root_id[rev_id], rev_id) for rev_id in
240
# Guaranteed stable, this groups all the file id operations together
241
# retaining topological order within the revisions of a file id.
242
# File id splits and joins would invalidate this, but they don't exist
243
# yet, and are unlikely to in non-rich-root environments anyway.
244
root_id_order.sort(key=operator.itemgetter(0))
245
# Create a record stream containing the roots to create.
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()
343
inventory_weave = self.source.get_inventory_weave()
346
to_store = self.target.weave_store
347
parent_map = self.source.get_graph().get_parent_map(revs)
348
for tree in self.iter_rev_trees(revs):
349
revision_id = tree.inventory.root.revision
350
root_id = tree.get_root_id()
351
parents = parent_map[revision_id]
352
if parents[0] == NULL_REVISION:
354
if root_id not in versionedfile:
355
versionedfile[root_id] = to_store.get_weave_or_empty(root_id,
356
self.target.get_transaction())
357
_, _, parent_texts[root_id] = versionedfile[root_id].add_lines(
358
revision_id, parents, [], parent_texts)
360
def regenerate_inventory(self, revs):
361
"""Generate a new inventory versionedfile in target, convertin data.
363
The inventory is retrieved from the source, (deserializing it), and
364
stored in the target (reserializing it in a different format).
365
:param revs: The revisions to include
367
for tree in self.iter_rev_trees(revs):
368
parents = tree.get_parent_ids()
369
self.target.add_inventory(tree.get_revision_id(), tree.inventory,
373
class Model1toKnit2Fetcher(GenericRepoFetcher):
374
"""Fetch from a Model1 repository into a Knit2 repository
376
def __init__(self, to_repository, from_repository, last_revision=None,
377
pb=None, find_ghosts=True):
378
self.helper = Inter1and2Helper(from_repository, to_repository)
379
GenericRepoFetcher.__init__(self, to_repository, from_repository,
380
last_revision, pb, find_ghosts)
382
def _generate_root_texts(self, revs):
383
self.helper.generate_root_texts(revs)
385
def _fetch_inventory_weave(self, revs, pb):
386
self.helper.regenerate_inventory(revs)
389
class Knit1to2Fetcher(KnitRepoFetcher):
390
"""Fetch from a Knit1 repository into a Knit2 repository"""
392
def __init__(self, to_repository, from_repository, last_revision=None,
393
pb=None, find_ghosts=True):
394
self.helper = Inter1and2Helper(from_repository, to_repository)
395
KnitRepoFetcher.__init__(self, to_repository, from_repository,
396
last_revision, pb, find_ghosts)
398
def _generate_root_texts(self, revs):
399
self.helper.generate_root_texts(revs)
401
def _fetch_inventory_weave(self, revs, pb):
402
self.helper.regenerate_inventory(revs)
405
class RemoteToOtherFetcher(GenericRepoFetcher):
407
def _fetch_everything_for_search(self, search, pp):
408
data_stream = self.from_repository.get_data_stream_for_search(search)
409
self.to_repository.insert_data_stream(data_stream)
added
removed
bzrlib/fetch.py
