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(), """
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,
40
from bzrlib.revision import NULL_REVISION
43
41
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
66
44
class RepoFetcher(object):
67
45
"""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
This should not be used directory, its essential a object to encapsulate
47
This should not be used directly, it's essential a object to encapsulate
76
48
the logic in InterRepository.fetch().
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
# check that last_revision is in 'from' and then return a
85
if last_revision is not None and not is_null(last_revision):
86
to_repository.get_revision(last_revision)
51
def __init__(self, to_repository, from_repository, last_revision=None,
52
find_ghosts=True, fetch_spec=None):
53
"""Create a repo fetcher.
55
:param last_revision: If set, try to limit to the data this revision
57
:param find_ghosts: If True search the entire history for ghosts.
59
# repository.fetch has the responsibility for short-circuiting
60
# attempts to copy between a repository and itself.
88
61
self.to_repository = to_repository
89
62
self.from_repository = from_repository
63
self.sink = to_repository._get_sink()
90
64
# must not mutate self._last_revision as its potentially a shared instance
91
65
self._last_revision = last_revision
93
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
94
self.nested_pb = self.pb
66
self._fetch_spec = fetch_spec
67
self.find_ghosts = find_ghosts
98
68
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)
100
self.to_repository.lock_write()
102
self.to_repository.start_write_group()
106
self.to_repository.abort_write_group()
109
self.to_repository.commit_write_group()
111
if self.nested_pb is not None:
112
self.nested_pb.finished()
113
self.to_repository.unlock()
115
75
self.from_repository.unlock()
117
77
def __fetch(self):
118
78
"""Primary worker function.
120
This initialises all the needed variables, and then fetches the
80
This initialises all the needed variables, and then fetches the
121
81
requested revisions, finally clearing the progress bar.
123
self.to_weaves = self.to_repository.weave_store
124
self.from_weaves = self.from_repository.weave_store
83
# Roughly this is what we're aiming for fetch to become:
85
# missing = self.sink.insert_stream(self.source.get_stream(search))
87
# missing = self.sink.insert_stream(self.source.get_items(missing))
125
89
self.count_total = 0
126
90
self.file_ids_names = {}
127
pp = ProgressPhase('Fetch phase', 4, self.pb)
91
pb = ui.ui_factory.nested_progress_bar()
92
pb.show_pct = pb.show_count = False
130
revs = self._revids_to_fetch()
131
self._fetch_everything_for_revisions(revs, pp)
94
pb.update("Finding revisions", 0, 2)
95
search = self._revids_to_fetch()
98
pb.update("Fetching revisions", 1, 2)
99
self._fetch_everything_for_search(search)
135
def _fetch_everything_for_revisions(self, revs, pp):
103
def _fetch_everything_for_search(self, search):
136
104
"""Fetch all data for the given set of revisions."""
139
105
# The first phase is "file". We pass the progress bar for it directly
140
106
# into item_keys_introduced_by, which has more information about how
141
107
# that phase is progressing than we do. Progress updates for the other
144
110
# item_keys_introduced_by should have a richer API than it does at the
145
111
# moment, so that it can feed the progress information back to this
148
pb = bzrlib.ui.ui_factory.nested_progress_bar()
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()
150
data_to_fetch = self.from_repository.item_keys_introduced_by(revs, pb)
151
for knit_kind, file_id, revisions in data_to_fetch:
152
if knit_kind != phase:
154
# Make a new progress bar for this phase
157
pb = bzrlib.ui.ui_factory.nested_progress_bar()
158
if knit_kind == "file":
159
self._fetch_weave_text(file_id, revisions)
160
elif knit_kind == "inventory":
162
# Once we've processed all the files, then we generate the root
163
# texts (if necessary), then we process the inventory. It's a
164
# bit distasteful to have knit_kind == "inventory" mean this,
165
# perhaps it should happen on the first non-"file" knit, in case
166
# it's not always inventory?
167
self._generate_root_texts(revs)
168
self._fetch_inventory_weave(revs, pb)
169
elif knit_kind == "signatures":
170
# Nothing to do here; this will be taken care of when
171
# _fetch_revision_texts happens.
173
elif knit_kind == "revisions":
174
self._fetch_revision_texts(revs, pb)
176
raise AssertionError("Unknown knit kind %r" % knit_kind)
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
from_format = self.from_repository._format
125
pb.update("Inserting stream")
126
resume_tokens, missing_keys = self.sink.insert_stream(
127
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")
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")
180
self.count_copied += len(revs)
182
150
def _revids_to_fetch(self):
183
151
"""Determines the exact revisions needed from self.from_repository to
184
152
install self._last_revision in self.to_repository.
186
154
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
188
158
mutter('fetch up to rev {%s}', self._last_revision)
189
159
if self._last_revision is NULL_REVISION:
190
160
# explicit limit of no revisions needed
192
if (self._last_revision is not None and
193
self.to_repository.has_revision(self._last_revision)):
197
return self.to_repository.missing_revision_ids(self.from_repository,
199
except errors.NoSuchRevision:
200
raise InstallFailed([self._last_revision])
202
def _fetch_weave_text(self, file_id, required_versions):
203
to_weave = self.to_weaves.get_weave_or_empty(file_id,
204
self.to_repository.get_transaction())
205
from_weave = self.from_weaves.get_weave(file_id,
206
self.from_repository.get_transaction())
207
# we fetch all the texts, because texts do
208
# not reference anything, and its cheap enough
209
to_weave.join(from_weave, version_ids=required_versions)
210
# we don't need *all* of this data anymore, but we dont know
211
# what we do. This cache clearing will result in a new read
212
# of the knit data when we do the checkout, but probably we
213
# want to emit the needed data on the fly rather than at the
215
# the from weave should know not to cache data being joined,
216
# but its ok to ask it to clear.
217
from_weave.clear_cache()
218
to_weave.clear_cache()
220
def _fetch_inventory_weave(self, revs, pb):
221
pb.update("fetch inventory", 0, 2)
222
to_weave = self.to_repository.get_inventory_weave()
223
child_pb = bzrlib.ui.ui_factory.nested_progress_bar()
225
# just merge, this is optimisable and its means we don't
226
# copy unreferenced data such as not-needed inventories.
227
pb.update("fetch inventory", 1, 3)
228
from_weave = self.from_repository.get_inventory_weave()
229
pb.update("fetch inventory", 2, 3)
230
# we fetch only the referenced inventories because we do not
231
# know for unselected inventories whether all their required
232
# texts are present in the other repository - it could be
234
to_weave.join(from_weave, pb=child_pb, msg='merge inventory',
236
from_weave.clear_cache()
240
def _generate_root_texts(self, revs):
241
"""This will be called by __fetch between fetching weave texts and
242
fetching the inventory weave.
244
Subclasses should override this if they need to generate root texts
245
after fetching weave texts.
250
class GenericRepoFetcher(RepoFetcher):
251
"""This is a generic repo to repo fetcher.
253
This makes minimal assumptions about repo layout and contents.
254
It triggers a reconciliation after fetching to ensure integrity.
257
def _fetch_revision_texts(self, revs, pb):
258
"""Fetch revision object texts"""
259
to_txn = self.to_transaction = self.to_repository.get_transaction()
262
to_store = self.to_repository._revision_store
264
pb.update('copying revisions', count, total)
266
sig_text = self.from_repository.get_signature_text(rev)
267
to_store.add_revision_signature_text(rev, sig_text, to_txn)
268
except errors.NoSuchRevision:
271
to_store.add_revision(self.from_repository.get_revision(rev),
274
# fixup inventory if needed:
275
# this is expensive because we have no inverse index to current ghosts.
276
# but on local disk its a few seconds and sftp push is already insane.
278
# FIXME: repository should inform if this is needed.
279
self.to_repository.reconcile()
282
class KnitRepoFetcher(RepoFetcher):
283
"""This is a knit format repository specific fetcher.
285
This differs from the GenericRepoFetcher by not doing a
286
reconciliation after copying, and using knit joining to
290
def _fetch_revision_texts(self, revs, pb):
291
# may need to be a InterRevisionStore call here.
292
from_transaction = self.from_repository.get_transaction()
293
to_transaction = self.to_repository.get_transaction()
294
to_sf = self.to_repository._revision_store.get_signature_file(
296
from_sf = self.from_repository._revision_store.get_signature_file(
298
to_sf.join(from_sf, version_ids=revs, ignore_missing=True)
299
to_rf = self.to_repository._revision_store.get_revision_file(
301
from_rf = self.from_repository._revision_store.get_revision_file(
303
to_rf.join(from_rf, version_ids=revs)
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)
306
179
class Inter1and2Helper(object):
307
180
"""Helper for operations that convert data from model 1 and 2
309
182
This is for use by fetchers and converters.
312
def __init__(self, source, target):
185
# This is a class variable so that the test suite can override it.
186
known_graph_threshold = 100
188
def __init__(self, source):
315
191
:param source: The repository data comes from
316
:param target: The repository data goes to
318
193
self.source = source
321
195
def iter_rev_trees(self, revs):
322
196
"""Iterate through RevisionTrees efficiently.
336
212
revs = revs[100:]
214
def _find_root_ids(self, revs, parent_map, graph):
216
for tree in self.iter_rev_trees(revs):
217
revision_id = tree.inventory.root.revision
218
root_id = tree.get_root_id()
219
revision_root[revision_id] = root_id
220
# Find out which parents we don't already know root ids for
222
for revision_parents in parent_map.itervalues():
223
parents.update(revision_parents)
224
parents.difference_update(revision_root.keys() + [NULL_REVISION])
225
# Limit to revisions present in the versionedfile
226
parents = graph.get_parent_map(parents).keys()
227
for tree in self.iter_rev_trees(parents):
228
root_id = tree.get_root_id()
229
revision_root[tree.get_revision_id()] = root_id
338
232
def generate_root_texts(self, revs):
339
233
"""Generate VersionedFiles for all root ids.
341
235
:param revs: the revisions to include
343
inventory_weave = self.source.get_inventory_weave()
346
to_store = self.target.weave_store
347
for tree in self.iter_rev_trees(revs):
348
revision_id = tree.inventory.root.revision
349
root_id = tree.inventory.root.file_id
350
parents = inventory_weave.get_parents(revision_id)
351
if root_id not in versionedfile:
352
versionedfile[root_id] = to_store.get_weave_or_empty(root_id,
353
self.target.get_transaction())
354
parent_texts[root_id] = versionedfile[root_id].add_lines(
355
revision_id, parents, [], parent_texts)
357
def regenerate_inventory(self, revs):
358
"""Generate a new inventory versionedfile in target, convertin data.
360
The inventory is retrieved from the source, (deserializing it), and
361
stored in the target (reserializing it in a different format).
362
:param revs: The revisions to include
364
inventory_weave = self.source.get_inventory_weave()
365
for tree in self.iter_rev_trees(revs):
366
parents = inventory_weave.get_parents(tree.get_revision_id())
367
self.target.add_inventory(tree.get_revision_id(), tree.inventory,
371
class Model1toKnit2Fetcher(GenericRepoFetcher):
372
"""Fetch from a Model1 repository into a Knit2 repository
374
def __init__(self, to_repository, from_repository, last_revision=None,
376
self.helper = Inter1and2Helper(from_repository, to_repository)
377
GenericRepoFetcher.__init__(self, to_repository, from_repository,
380
def _generate_root_texts(self, revs):
381
self.helper.generate_root_texts(revs)
383
def _fetch_inventory_weave(self, revs, pb):
384
self.helper.regenerate_inventory(revs)
387
class Knit1to2Fetcher(KnitRepoFetcher):
388
"""Fetch from a Knit1 repository into a Knit2 repository"""
390
def __init__(self, to_repository, from_repository, last_revision=None,
392
self.helper = Inter1and2Helper(from_repository, to_repository)
393
KnitRepoFetcher.__init__(self, to_repository, from_repository,
396
def _generate_root_texts(self, revs):
397
self.helper.generate_root_texts(revs)
399
def _fetch_inventory_weave(self, revs, pb):
400
self.helper.regenerate_inventory(revs)
237
graph = self.source.get_graph()
238
parent_map = graph.get_parent_map(revs)
239
rev_order = tsort.topo_sort(parent_map)
240
rev_id_to_root_id = self._find_root_ids(revs, parent_map, graph)
241
root_id_order = [(rev_id_to_root_id[rev_id], rev_id) for rev_id in
243
# Guaranteed stable, this groups all the file id operations together
244
# retaining topological order within the revisions of a file id.
245
# File id splits and joins would invalidate this, but they don't exist
246
# yet, and are unlikely to in non-rich-root environments anyway.
247
root_id_order.sort(key=operator.itemgetter(0))
248
# Create a record stream containing the roots to create.
249
if len(revs) > self.known_graph_threshold:
250
graph = self.source.get_known_graph_ancestry(revs)
251
new_roots_stream = _new_root_data_stream(
252
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
253
return [('texts', new_roots_stream)]
256
def _new_root_data_stream(
257
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
258
"""Generate a texts substream of synthesised root entries.
260
Used in fetches that do rich-root upgrades.
262
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
263
the root entries to create.
264
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
265
calculating the parents. If a parent rev_id is not found here then it
266
will be recalculated.
267
:param parent_map: a parent map for all the revisions in
269
:param graph: a graph to use instead of repo.get_graph().
271
for root_key in root_keys_to_create:
272
root_id, rev_id = root_key
273
parent_keys = _parent_keys_for_root_version(
274
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
275
yield versionedfile.FulltextContentFactory(
276
root_key, parent_keys, None, '')
279
def _parent_keys_for_root_version(
280
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
281
"""Get the parent keys for a given root id.
283
A helper function for _new_root_data_stream.
285
# Include direct parents of the revision, but only if they used the same
286
# root_id and are heads.
287
rev_parents = parent_map[rev_id]
289
for parent_id in rev_parents:
290
if parent_id == NULL_REVISION:
292
if parent_id not in rev_id_to_root_id_map:
293
# We probably didn't read this revision, go spend the extra effort
296
tree = repo.revision_tree(parent_id)
297
except errors.NoSuchRevision:
298
# Ghost, fill out rev_id_to_root_id in case we encounter this
300
# But set parent_root_id to None since we don't really know
301
parent_root_id = None
303
parent_root_id = tree.get_root_id()
304
rev_id_to_root_id_map[parent_id] = None
306
# rev_id_to_root_id_map[parent_id] = parent_root_id
307
# memory consumption maybe?
309
parent_root_id = rev_id_to_root_id_map[parent_id]
310
if root_id == parent_root_id:
311
# With stacking we _might_ want to refer to a non-local revision,
312
# but this code path only applies when we have the full content
313
# available, so ghosts really are ghosts, not just the edge of
315
parent_ids.append(parent_id)
317
# root_id may be in the parent anyway.
319
tree = repo.revision_tree(parent_id)
320
except errors.NoSuchRevision:
321
# ghost, can't refer to it.
325
parent_ids.append(tree.inventory[root_id].revision)
326
except errors.NoSuchId:
329
# Drop non-head parents
331
graph = repo.get_graph()
332
heads = graph.heads(parent_ids)
334
for parent_id in parent_ids:
335
if parent_id in heads and parent_id not in selected_ids:
336
selected_ids.append(parent_id)
337
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
added
removed
bzrlib/fetch.py
