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 finally actually join
28
from bzrlib.lazy_import import lazy_import
29
lazy_import(globals(), """
43
from bzrlib.revision import NULL_REVISION
35
import bzrlib.errors as errors
36
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,
44
43
from bzrlib.trace import mutter
46
from bzrlib.lazy_import import lazy_import
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
47
66
class RepoFetcher(object):
48
67
"""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
50
75
This should not be used directly, it's essential a object to encapsulate
51
76
the logic in InterRepository.fetch().
54
def __init__(self, to_repository, from_repository, last_revision=None,
55
pb=None, find_ghosts=True, fetch_spec=None):
56
"""Create a repo fetcher.
58
:param last_revision: If set, try to limit to the data this revision
60
:param find_ghosts: If True search the entire history for ghosts.
61
:param pb: ProgressBar object to use; deprecated and ignored.
62
This method will just create one on top of the stack.
65
symbol_versioning.warn(
66
symbol_versioning.deprecated_in((1, 14, 0))
67
% "pb parameter to RepoFetcher.__init__")
68
# and for simplicity it is in fact ignored
69
# repository.fetch has the responsibility for short-circuiting
70
# attempts to copy between a repository and itself.
78
def __init__(self, to_repository, from_repository, last_revision=None, pb=None):
80
self.failed_revisions = []
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))
71
87
self.to_repository = to_repository
72
88
self.from_repository = from_repository
73
self.sink = to_repository._get_sink()
74
89
# must not mutate self._last_revision as its potentially a shared instance
75
90
self._last_revision = last_revision
76
self._fetch_spec = fetch_spec
77
self.find_ghosts = find_ghosts
92
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
93
self.nested_pb = self.pb
78
97
self.from_repository.lock_read()
79
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
80
self.from_repository, self.from_repository._format,
81
self.to_repository, self.to_repository._format)
99
self.to_repository.lock_write()
101
self.to_repository.start_write_group()
105
self.to_repository.abort_write_group()
108
self.to_repository.commit_write_group()
110
if self.nested_pb is not None:
111
self.nested_pb.finished()
112
self.to_repository.unlock()
85
114
self.from_repository.unlock()
87
116
def __fetch(self):
88
117
"""Primary worker function.
90
This initialises all the needed variables, and then fetches the
119
This initialises all the needed variables, and then fetches the
91
120
requested revisions, finally clearing the progress bar.
93
# Roughly this is what we're aiming for fetch to become:
95
# missing = self.sink.insert_stream(self.source.get_stream(search))
97
# missing = self.sink.insert_stream(self.source.get_items(missing))
122
self.to_weaves = self.to_repository.weave_store
123
self.from_weaves = self.from_repository.weave_store
99
124
self.count_total = 0
100
125
self.file_ids_names = {}
101
pb = ui.ui_factory.nested_progress_bar()
102
pb.show_pct = pb.show_count = False
126
pp = ProgressPhase('Fetch phase', 4, self.pb)
104
pb.update("Finding revisions", 0, 2)
105
search = self._revids_to_fetch()
129
revs = self._revids_to_fetch()
108
pb.update("Fetching revisions", 1, 2)
109
self._fetch_everything_for_search(search)
132
self._fetch_everything_for_revisions(revs, pp)
113
def _fetch_everything_for_search(self, search):
136
def _fetch_everything_for_revisions(self, revs, pp):
114
137
"""Fetch all data for the given set of revisions."""
115
138
# The first phase is "file". We pass the progress bar for it directly
116
139
# into item_keys_introduced_by, which has more information about how
120
143
# item_keys_introduced_by should have a richer API than it does at the
121
144
# moment, so that it can feed the progress information back to this
123
if (self.from_repository._format.rich_root_data and
124
not self.to_repository._format.rich_root_data):
125
raise errors.IncompatibleRepositories(
126
self.from_repository, self.to_repository,
127
"different rich-root support")
128
pb = ui.ui_factory.nested_progress_bar()
147
pb = bzrlib.ui.ui_factory.nested_progress_bar()
130
pb.update("Get stream source")
131
source = self.from_repository._get_source(
132
self.to_repository._format)
133
stream = source.get_stream(search)
134
from_format = self.from_repository._format
135
pb.update("Inserting stream")
136
resume_tokens, missing_keys = self.sink.insert_stream(
137
stream, from_format, [])
138
if self.to_repository._fallback_repositories:
140
self._parent_inventories(search.get_keys()))
142
pb.update("Missing keys")
143
stream = source.get_stream_for_missing_keys(missing_keys)
144
pb.update("Inserting missing keys")
145
resume_tokens, missing_keys = self.sink.insert_stream(
146
stream, from_format, resume_tokens)
148
raise AssertionError(
149
"second push failed to complete a fetch %r." % (
152
raise AssertionError(
153
"second push failed to commit the fetch %r." % (
155
pb.update("Finishing stream")
149
data_to_fetch = self.from_repository.item_keys_introduced_by(revs, pb)
150
for knit_kind, file_id, revisions in data_to_fetch:
151
if knit_kind != phase:
153
# Make a new progress bar for this phase
156
pb = bzrlib.ui.ui_factory.nested_progress_bar()
157
if knit_kind == "file":
158
self._fetch_weave_text(file_id, revisions)
159
elif knit_kind == "inventory":
161
# Once we've processed all the files, then we generate the root
162
# texts (if necessary), then we process the inventory. It's a
163
# bit distasteful to have knit_kind == "inventory" mean this,
164
# perhaps it should happen on the first non-"file" knit, in case
165
# it's not always inventory?
166
self._generate_root_texts(revs)
167
self._fetch_inventory_weave(revs, pb)
168
elif knit_kind == "signatures":
169
# Nothing to do here; this will be taken care of when
170
# _fetch_revision_texts happens.
172
elif knit_kind == "revisions":
173
self._fetch_revision_texts(revs, pb)
175
raise AssertionError("Unknown knit kind %r" % knit_kind)
179
self.count_copied += len(revs)
160
181
def _revids_to_fetch(self):
161
182
"""Determines the exact revisions needed from self.from_repository to
162
183
install self._last_revision in self.to_repository.
164
185
If no revisions need to be fetched, then this just returns None.
166
if self._fetch_spec is not None:
167
return self._fetch_spec
168
187
mutter('fetch up to rev {%s}', self._last_revision)
169
188
if self._last_revision is NULL_REVISION:
170
189
# explicit limit of no revisions needed
172
return self.to_repository.search_missing_revision_ids(
173
self.from_repository, self._last_revision,
174
find_ghosts=self.find_ghosts)
176
def _parent_inventories(self, revision_ids):
177
# Find all the parent revisions referenced by the stream, but
178
# not present in the stream, and make sure we send their
180
parent_maps = self.to_repository.get_parent_map(revision_ids)
182
map(parents.update, parent_maps.itervalues())
183
parents.discard(NULL_REVISION)
184
parents.difference_update(revision_ids)
185
missing_keys = set(('inventories', rev_id) for rev_id in parents)
191
if (self._last_revision is not None and
192
self.to_repository.has_revision(self._last_revision)):
196
# XXX: this gets the full graph on both sides, and will make sure
197
# that ghosts are filled whether or not you care about them.
198
return self.to_repository.missing_revision_ids(self.from_repository,
200
except errors.NoSuchRevision:
201
raise InstallFailed([self._last_revision])
203
def _fetch_weave_text(self, file_id, required_versions):
204
to_weave = self.to_weaves.get_weave_or_empty(file_id,
205
self.to_repository.get_transaction())
206
from_weave = self.from_weaves.get_weave(file_id,
207
self.from_repository.get_transaction())
208
# we fetch all the texts, because texts do
209
# not reference anything, and its cheap enough
210
to_weave.join(from_weave, version_ids=required_versions)
211
# we don't need *all* of this data anymore, but we dont know
212
# what we do. This cache clearing will result in a new read
213
# of the knit data when we do the checkout, but probably we
214
# want to emit the needed data on the fly rather than at the
216
# the from weave should know not to cache data being joined,
217
# but its ok to ask it to clear.
218
from_weave.clear_cache()
219
to_weave.clear_cache()
221
def _fetch_inventory_weave(self, revs, pb):
222
pb.update("fetch inventory", 0, 2)
223
to_weave = self.to_repository.get_inventory_weave()
224
child_pb = bzrlib.ui.ui_factory.nested_progress_bar()
226
# just merge, this is optimisable and its means we don't
227
# copy unreferenced data such as not-needed inventories.
228
pb.update("fetch inventory", 1, 3)
229
from_weave = self.from_repository.get_inventory_weave()
230
pb.update("fetch inventory", 2, 3)
231
# we fetch only the referenced inventories because we do not
232
# know for unselected inventories whether all their required
233
# texts are present in the other repository - it could be
235
to_weave.join(from_weave, pb=child_pb, msg='merge inventory',
237
from_weave.clear_cache()
241
def _generate_root_texts(self, revs):
242
"""This will be called by __fetch between fetching weave texts and
243
fetching the inventory weave.
245
Subclasses should override this if they need to generate root texts
246
after fetching weave texts.
251
class GenericRepoFetcher(RepoFetcher):
252
"""This is a generic repo to repo fetcher.
254
This makes minimal assumptions about repo layout and contents.
255
It triggers a reconciliation after fetching to ensure integrity.
258
def _fetch_revision_texts(self, revs, pb):
259
"""Fetch revision object texts"""
260
to_txn = self.to_transaction = self.to_repository.get_transaction()
263
to_store = self.to_repository._revision_store
265
pb.update('copying revisions', count, total)
267
sig_text = self.from_repository.get_signature_text(rev)
268
to_store.add_revision_signature_text(rev, sig_text, to_txn)
269
except errors.NoSuchRevision:
272
to_store.add_revision(self.from_repository.get_revision(rev),
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()
283
class KnitRepoFetcher(RepoFetcher):
284
"""This is a knit format repository specific fetcher.
286
This differs from the GenericRepoFetcher by not doing a
287
reconciliation after copying, and using knit joining to
291
def _fetch_revision_texts(self, revs, pb):
292
# may need to be a InterRevisionStore call here.
293
from_transaction = self.from_repository.get_transaction()
294
to_transaction = self.to_repository.get_transaction()
295
to_sf = self.to_repository._revision_store.get_signature_file(
297
from_sf = self.from_repository._revision_store.get_signature_file(
299
to_sf.join(from_sf, version_ids=revs, ignore_missing=True)
300
to_rf = self.to_repository._revision_store.get_revision_file(
302
from_rf = self.from_repository._revision_store.get_revision_file(
304
to_rf.join(from_rf, version_ids=revs)
189
307
class Inter1and2Helper(object):
190
308
"""Helper for operations that convert data from model 1 and 2
192
310
This is for use by fetchers and converters.
195
def __init__(self, source):
313
def __init__(self, source, target):
198
316
:param source: The repository data comes from
317
:param target: The repository data goes to
200
319
self.source = source
202
322
def iter_rev_trees(self, revs):
203
323
"""Iterate through RevisionTrees efficiently.
219
337
revs = revs[100:]
221
def _find_root_ids(self, revs, parent_map, graph):
223
for tree in self.iter_rev_trees(revs):
224
revision_id = tree.inventory.root.revision
225
root_id = tree.get_root_id()
226
revision_root[revision_id] = root_id
227
# Find out which parents we don't already know root ids for
229
for revision_parents in parent_map.itervalues():
230
parents.update(revision_parents)
231
parents.difference_update(revision_root.keys() + [NULL_REVISION])
232
# Limit to revisions present in the versionedfile
233
parents = graph.get_parent_map(parents).keys()
234
for tree in self.iter_rev_trees(parents):
235
root_id = tree.get_root_id()
236
revision_root[tree.get_revision_id()] = root_id
239
339
def generate_root_texts(self, revs):
240
340
"""Generate VersionedFiles for all root ids.
242
342
:param revs: the revisions to include
244
graph = self.source.get_graph()
245
parent_map = graph.get_parent_map(revs)
246
rev_order = tsort.topo_sort(parent_map)
247
rev_id_to_root_id = self._find_root_ids(revs, parent_map, graph)
248
root_id_order = [(rev_id_to_root_id[rev_id], rev_id) for rev_id in
250
# Guaranteed stable, this groups all the file id operations together
251
# retaining topological order within the revisions of a file id.
252
# File id splits and joins would invalidate this, but they don't exist
253
# yet, and are unlikely to in non-rich-root environments anyway.
254
root_id_order.sort(key=operator.itemgetter(0))
255
# Create a record stream containing the roots to create.
257
graph = _get_rich_root_heads_graph(self.source_repo, revs)
258
new_roots_stream = _new_root_data_stream(
259
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
260
return [('texts', new_roots_stream)]
263
def _get_rich_root_heads_graph(source_repo, revision_ids):
264
"""Get a Graph object suitable for asking heads() for new rich roots."""
265
st = static_tuple.StaticTuple
266
revision_keys = [st(r_id).intern() for r_id in revision_ids]
267
known_graph = source_repo.revisions.get_known_graph_ancestry(
269
return _mod_graph.GraphThunkIdsToKeys(known_graph)
272
def _new_root_data_stream(
273
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
274
"""Generate a texts substream of synthesised root entries.
276
Used in fetches that do rich-root upgrades.
278
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
279
the root entries to create.
280
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
281
calculating the parents. If a parent rev_id is not found here then it
282
will be recalculated.
283
:param parent_map: a parent map for all the revisions in
285
:param graph: a graph to use instead of repo.get_graph().
287
for root_key in root_keys_to_create:
288
root_id, rev_id = root_key
289
parent_keys = _parent_keys_for_root_version(
290
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
291
yield versionedfile.FulltextContentFactory(
292
root_key, parent_keys, None, '')
295
def _parent_keys_for_root_version(
296
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
297
"""Get the parent keys for a given root id.
299
A helper function for _new_root_data_stream.
301
# Include direct parents of the revision, but only if they used the same
302
# root_id and are heads.
303
rev_parents = parent_map[rev_id]
305
for parent_id in rev_parents:
306
if parent_id == NULL_REVISION:
308
if parent_id not in rev_id_to_root_id_map:
309
# We probably didn't read this revision, go spend the extra effort
312
tree = repo.revision_tree(parent_id)
313
except errors.NoSuchRevision:
314
# Ghost, fill out rev_id_to_root_id in case we encounter this
316
# But set parent_root_id to None since we don't really know
317
parent_root_id = None
319
parent_root_id = tree.get_root_id()
320
rev_id_to_root_id_map[parent_id] = None
322
# rev_id_to_root_id_map[parent_id] = parent_root_id
323
# memory consumption maybe?
325
parent_root_id = rev_id_to_root_id_map[parent_id]
326
if root_id == parent_root_id:
327
# With stacking we _might_ want to refer to a non-local revision,
328
# but this code path only applies when we have the full content
329
# available, so ghosts really are ghosts, not just the edge of
331
parent_ids.append(parent_id)
333
# root_id may be in the parent anyway.
335
tree = repo.revision_tree(parent_id)
336
except errors.NoSuchRevision:
337
# ghost, can't refer to it.
341
parent_ids.append(tree.inventory[root_id].revision)
342
except errors.NoSuchId:
345
# Drop non-head parents
347
graph = repo.get_graph()
348
heads = graph.heads(parent_ids)
350
for parent_id in parent_ids:
351
if parent_id in heads and parent_id not in selected_ids:
352
selected_ids.append(parent_id)
353
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
344
inventory_weave = self.source.get_inventory_weave()
347
to_store = self.target.weave_store
348
for tree in self.iter_rev_trees(revs):
349
revision_id = tree.inventory.root.revision
350
root_id = tree.inventory.root.file_id
351
parents = inventory_weave.get_parents(revision_id)
352
if root_id not in versionedfile:
353
versionedfile[root_id] = to_store.get_weave_or_empty(root_id,
354
self.target.get_transaction())
355
_, _, parent_texts[root_id] = versionedfile[root_id].add_lines(
356
revision_id, parents, [], parent_texts)
358
def regenerate_inventory(self, revs):
359
"""Generate a new inventory versionedfile in target, convertin data.
361
The inventory is retrieved from the source, (deserializing it), and
362
stored in the target (reserializing it in a different format).
363
:param revs: The revisions to include
365
inventory_weave = self.source.get_inventory_weave()
366
for tree in self.iter_rev_trees(revs):
367
parents = inventory_weave.get_parents(tree.get_revision_id())
368
self.target.add_inventory(tree.get_revision_id(), tree.inventory,
372
class Model1toKnit2Fetcher(GenericRepoFetcher):
373
"""Fetch from a Model1 repository into a Knit2 repository
375
def __init__(self, to_repository, from_repository, last_revision=None,
377
self.helper = Inter1and2Helper(from_repository, to_repository)
378
GenericRepoFetcher.__init__(self, to_repository, from_repository,
381
def _generate_root_texts(self, revs):
382
self.helper.generate_root_texts(revs)
384
def _fetch_inventory_weave(self, revs, pb):
385
self.helper.regenerate_inventory(revs)
388
class Knit1to2Fetcher(KnitRepoFetcher):
389
"""Fetch from a Knit1 repository into a Knit2 repository"""
391
def __init__(self, to_repository, from_repository, last_revision=None,
393
self.helper = Inter1and2Helper(from_repository, to_repository)
394
KnitRepoFetcher.__init__(self, to_repository, from_repository,
397
def _generate_root_texts(self, revs):
398
self.helper.generate_root_texts(revs)
400
def _fetch_inventory_weave(self, revs, pb):
401
self.helper.regenerate_inventory(revs)
404
class RemoteToOtherFetcher(GenericRepoFetcher):
406
def _fetch_everything_for_revisions(self, revs, pp):
407
data_stream = self.from_repository.get_data_stream(revs)
408
self.to_repository.insert_data_stream(data_stream)