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(), """
40
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,
41
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
44
66
class RepoFetcher(object):
45
67
"""Pull revisions and texts from one repository to another.
47
This should not be used directly, it's essential a object to encapsulate
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
48
76
the logic in InterRepository.fetch().
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.
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)
61
88
self.to_repository = to_repository
62
89
self.from_repository = from_repository
63
self.sink = to_repository._get_sink()
64
90
# must not mutate self._last_revision as its potentially a shared instance
65
91
self._last_revision = last_revision
66
self._fetch_spec = fetch_spec
67
self.find_ghosts = find_ghosts
93
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
94
self.nested_pb = self.pb
68
98
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()
75
115
self.from_repository.unlock()
77
117
def __fetch(self):
78
118
"""Primary worker function.
80
This initialises all the needed variables, and then fetches the
120
This initialises all the needed variables, and then fetches the
81
121
requested revisions, finally clearing the progress bar.
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))
123
self.to_weaves = self.to_repository.weave_store
124
self.to_control = self.to_repository.control_weaves
125
self.from_weaves = self.from_repository.weave_store
126
self.from_control = self.from_repository.control_weaves
89
127
self.count_total = 0
90
128
self.file_ids_names = {}
91
pb = ui.ui_factory.nested_progress_bar()
92
pb.show_pct = pb.show_count = False
129
pp = ProgressPhase('Fetch phase', 4, self.pb)
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)
132
revs = self._revids_to_fetch()
133
self._fetch_everything_for_revisions(revs, pp)
103
def _fetch_everything_for_search(self, search):
137
def _fetch_everything_for_revisions(self, revs, pp):
104
138
"""Fetch all data for the given set of revisions."""
105
141
# The first phase is "file". We pass the progress bar for it directly
106
142
# into item_keys_introduced_by, which has more information about how
107
143
# that phase is progressing than we do. Progress updates for the other
110
146
# item_keys_introduced_by should have a richer API than it does at the
111
147
# moment, so that it can feed the progress information back to this
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
pb = bzrlib.ui.ui_factory.nested_progress_bar()
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")
152
data_to_fetch = self.from_repository.item_keys_introduced_by(revs, pb)
153
for knit_kind, file_id, revisions in data_to_fetch:
154
if knit_kind != phase:
156
# Make a new progress bar for this phase
159
pb = bzrlib.ui.ui_factory.nested_progress_bar()
160
if knit_kind == "file":
161
self._fetch_weave_text(file_id, revisions)
162
elif knit_kind == "inventory":
164
# Once we've processed all the files, then we generate the root
165
# texts (if necessary), then we process the inventory. It's a
166
# bit distasteful to have knit_kind == "inventory" mean this,
167
# perhaps it should happen on the first non-"file" knit, in case
168
# it's not always inventory?
169
self._generate_root_texts(revs)
170
self._fetch_inventory_weave(revs, pb)
171
elif knit_kind == "signatures":
172
# Nothing to do here; this will be taken care of when
173
# _fetch_revision_texts happens.
175
elif knit_kind == "revisions":
176
self._fetch_revision_texts(revs, pb)
178
raise AssertionError("Unknown knit kind %r" % knit_kind)
182
self.count_copied += len(revs)
150
184
def _revids_to_fetch(self):
151
185
"""Determines the exact revisions needed from self.from_repository to
152
186
install self._last_revision in self.to_repository.
154
188
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
158
190
mutter('fetch up to rev {%s}', self._last_revision)
159
191
if self._last_revision is NULL_REVISION:
160
192
# explicit limit of no revisions needed
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)
194
if (self._last_revision is not None and
195
self.to_repository.has_revision(self._last_revision)):
199
return self.to_repository.missing_revision_ids(self.from_repository,
201
except errors.NoSuchRevision:
202
raise InstallFailed([self._last_revision])
204
def _fetch_weave_text(self, file_id, required_versions):
205
to_weave = self.to_weaves.get_weave_or_empty(file_id,
206
self.to_repository.get_transaction())
207
from_weave = self.from_weaves.get_weave(file_id,
208
self.from_repository.get_transaction())
209
# we fetch all the texts, because texts do
210
# not reference anything, and its cheap enough
211
to_weave.join(from_weave, version_ids=required_versions)
212
# we don't need *all* of this data anymore, but we dont know
213
# what we do. This cache clearing will result in a new read
214
# of the knit data when we do the checkout, but probably we
215
# want to emit the needed data on the fly rather than at the
217
# the from weave should know not to cache data being joined,
218
# but its ok to ask it to clear.
219
from_weave.clear_cache()
220
to_weave.clear_cache()
222
def _fetch_inventory_weave(self, revs, pb):
223
pb.update("fetch inventory", 0, 2)
224
to_weave = self.to_control.get_weave('inventory',
225
self.to_repository.get_transaction())
227
child_pb = bzrlib.ui.ui_factory.nested_progress_bar()
229
# just merge, this is optimisable and its means we don't
230
# copy unreferenced data such as not-needed inventories.
231
pb.update("fetch inventory", 1, 3)
232
from_weave = self.from_repository.get_inventory_weave()
233
pb.update("fetch inventory", 2, 3)
234
# we fetch only the referenced inventories because we do not
235
# know for unselected inventories whether all their required
236
# texts are present in the other repository - it could be
238
to_weave.join(from_weave, pb=child_pb, msg='merge inventory',
240
from_weave.clear_cache()
244
def _generate_root_texts(self, revs):
245
"""This will be called by __fetch between fetching weave texts and
246
fetching the inventory weave.
248
Subclasses should override this if they need to generate root texts
249
after fetching weave texts.
254
class GenericRepoFetcher(RepoFetcher):
255
"""This is a generic repo to repo fetcher.
257
This makes minimal assumptions about repo layout and contents.
258
It triggers a reconciliation after fetching to ensure integrity.
261
def _fetch_revision_texts(self, revs, pb):
262
"""Fetch revision object texts"""
263
to_txn = self.to_transaction = self.to_repository.get_transaction()
266
to_store = self.to_repository._revision_store
268
pb.update('copying revisions', count, total)
270
sig_text = self.from_repository.get_signature_text(rev)
271
to_store.add_revision_signature_text(rev, sig_text, to_txn)
272
except errors.NoSuchRevision:
275
to_store.add_revision(self.from_repository.get_revision(rev),
278
# fixup inventory if needed:
279
# this is expensive because we have no inverse index to current ghosts.
280
# but on local disk its a few seconds and sftp push is already insane.
282
# FIXME: repository should inform if this is needed.
283
self.to_repository.reconcile()
286
class KnitRepoFetcher(RepoFetcher):
287
"""This is a knit format repository specific fetcher.
289
This differs from the GenericRepoFetcher by not doing a
290
reconciliation after copying, and using knit joining to
294
def _fetch_revision_texts(self, revs, pb):
295
# may need to be a InterRevisionStore call here.
296
from_transaction = self.from_repository.get_transaction()
297
to_transaction = self.to_repository.get_transaction()
298
to_sf = self.to_repository._revision_store.get_signature_file(
300
from_sf = self.from_repository._revision_store.get_signature_file(
302
to_sf.join(from_sf, version_ids=revs, ignore_missing=True)
303
to_rf = self.to_repository._revision_store.get_revision_file(
305
from_rf = self.from_repository._revision_store.get_revision_file(
307
to_rf.join(from_rf, version_ids=revs)
179
310
class Inter1and2Helper(object):
180
311
"""Helper for operations that convert data from model 1 and 2
182
313
This is for use by fetchers and converters.
185
def __init__(self, source):
316
def __init__(self, source, target):
188
319
:param source: The repository data comes from
320
:param target: The repository data goes to
190
322
self.source = source
192
325
def iter_rev_trees(self, revs):
193
326
"""Iterate through RevisionTrees efficiently.
209
340
revs = revs[100:]
211
def _find_root_ids(self, revs, parent_map, graph):
213
for tree in self.iter_rev_trees(revs):
214
revision_id = tree.inventory.root.revision
215
root_id = tree.get_root_id()
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
342
def generate_root_texts(self, revs):
230
343
"""Generate VersionedFiles for all root ids.
232
345
: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.
247
# XXX: not covered by tests, should have a flag to always run
248
# this. -- mbp 20100129
249
graph = self.source_repo.get_known_graph_ancestry(revs)
250
new_roots_stream = _new_root_data_stream(
251
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
252
return [('texts', new_roots_stream)]
255
def _get_rich_root_heads_graph(source_repo, revision_ids):
256
"""Get a Graph object suitable for asking heads() for new rich roots."""
260
def _new_root_data_stream(
261
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
262
"""Generate a texts substream of synthesised root entries.
264
Used in fetches that do rich-root upgrades.
266
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
267
the root entries to create.
268
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
269
calculating the parents. If a parent rev_id is not found here then it
270
will be recalculated.
271
:param parent_map: a parent map for all the revisions in
273
:param graph: a graph to use instead of repo.get_graph().
275
for root_key in root_keys_to_create:
276
root_id, rev_id = root_key
277
parent_keys = _parent_keys_for_root_version(
278
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
279
yield versionedfile.FulltextContentFactory(
280
root_key, parent_keys, None, '')
283
def _parent_keys_for_root_version(
284
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
285
"""Get the parent keys for a given root id.
287
A helper function for _new_root_data_stream.
289
# Include direct parents of the revision, but only if they used the same
290
# root_id and are heads.
291
rev_parents = parent_map[rev_id]
293
for parent_id in rev_parents:
294
if parent_id == NULL_REVISION:
296
if parent_id not in rev_id_to_root_id_map:
297
# We probably didn't read this revision, go spend the extra effort
300
tree = repo.revision_tree(parent_id)
301
except errors.NoSuchRevision:
302
# Ghost, fill out rev_id_to_root_id in case we encounter this
304
# But set parent_root_id to None since we don't really know
305
parent_root_id = None
307
parent_root_id = tree.get_root_id()
308
rev_id_to_root_id_map[parent_id] = None
310
# rev_id_to_root_id_map[parent_id] = parent_root_id
311
# memory consumption maybe?
313
parent_root_id = rev_id_to_root_id_map[parent_id]
314
if root_id == parent_root_id:
315
# With stacking we _might_ want to refer to a non-local revision,
316
# but this code path only applies when we have the full content
317
# available, so ghosts really are ghosts, not just the edge of
319
parent_ids.append(parent_id)
321
# root_id may be in the parent anyway.
323
tree = repo.revision_tree(parent_id)
324
except errors.NoSuchRevision:
325
# ghost, can't refer to it.
329
parent_ids.append(tree.inventory[root_id].revision)
330
except errors.NoSuchId:
333
# Drop non-head parents
335
graph = repo.get_graph()
336
heads = graph.heads(parent_ids)
338
for parent_id in parent_ids:
339
if parent_id in heads and parent_id not in selected_ids:
340
selected_ids.append(parent_id)
341
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
347
inventory_weave = self.source.get_inventory_weave()
350
to_store = self.target.weave_store
351
for tree in self.iter_rev_trees(revs):
352
revision_id = tree.inventory.root.revision
353
root_id = tree.inventory.root.file_id
354
parents = inventory_weave.get_parents(revision_id)
355
if root_id not in versionedfile:
356
versionedfile[root_id] = to_store.get_weave_or_empty(root_id,
357
self.target.get_transaction())
358
parent_texts[root_id] = versionedfile[root_id].add_lines(
359
revision_id, parents, [], parent_texts)
361
def regenerate_inventory(self, revs):
362
"""Generate a new inventory versionedfile in target, convertin data.
364
The inventory is retrieved from the source, (deserializing it), and
365
stored in the target (reserializing it in a different format).
366
:param revs: The revisions to include
368
inventory_weave = self.source.get_inventory_weave()
369
for tree in self.iter_rev_trees(revs):
370
parents = inventory_weave.get_parents(tree.get_revision_id())
371
self.target.add_inventory(tree.get_revision_id(), tree.inventory,
375
class Model1toKnit2Fetcher(GenericRepoFetcher):
376
"""Fetch from a Model1 repository into a Knit2 repository
378
def __init__(self, to_repository, from_repository, last_revision=None,
380
self.helper = Inter1and2Helper(from_repository, to_repository)
381
GenericRepoFetcher.__init__(self, to_repository, from_repository,
384
def _generate_root_texts(self, revs):
385
self.helper.generate_root_texts(revs)
387
def _fetch_inventory_weave(self, revs, pb):
388
self.helper.regenerate_inventory(revs)
391
class Knit1to2Fetcher(KnitRepoFetcher):
392
"""Fetch from a Knit1 repository into a Knit2 repository"""
394
def __init__(self, to_repository, from_repository, last_revision=None,
396
self.helper = Inter1and2Helper(from_repository, to_repository)
397
KnitRepoFetcher.__init__(self, to_repository, from_repository,
400
def _generate_root_texts(self, revs):
401
self.helper.generate_root_texts(revs)
403
def _fetch_inventory_weave(self, revs, pb):
404
self.helper.regenerate_inventory(revs)