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
43
39
from bzrlib.revision import NULL_REVISION
40
from bzrlib.symbol_versioning import (deprecated_function,
44
44
from bzrlib.trace import mutter
47
from bzrlib.lazy_import import lazy_import
49
# TODO: Avoid repeatedly opening weaves so many times.
51
# XXX: This doesn't handle ghost (not present in branch) revisions at
52
# all yet. I'm not sure they really should be supported.
54
# NOTE: This doesn't copy revisions which may be present but not
55
# merged into the last revision. I'm not sure we want to do that.
57
# - get a list of revisions that need to be pulled in
58
# - for each one, pull in that revision file
59
# and get the inventory, and store the inventory with right
61
# - and get the ancestry, and store that with right parents too
62
# - and keep a note of all file ids and version seen
63
# - then go through all files; for each one get the weave,
64
# and add in all file versions
67
@deprecated_function(zero_eight)
68
def greedy_fetch(to_branch, from_branch, revision=None, pb=None):
69
"""Legacy API, please see branch.fetch(from_branch, last_revision, pb)."""
70
f = Fetcher(to_branch, from_branch, revision, pb)
71
return f.count_copied, f.failed_revisions
47
76
class RepoFetcher(object):
48
77
"""Pull revisions and texts from one repository to another.
50
This should not be used directly, it's essential a object to encapsulate
80
if set, try to limit to the data this revision references.
83
count_copied -- number of revisions copied
85
This should not be used directory, its essential a object to encapsulate
51
86
the logic in InterRepository.fetch().
54
def __init__(self, to_repository, from_repository, last_revision=None,
55
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.
62
# repository.fetch has the responsibility for short-circuiting
63
# attempts to copy between a repository and itself.
88
def __init__(self, to_repository, from_repository, last_revision=None, pb=None):
90
self.failed_revisions = []
92
if to_repository.control_files._transport.base == from_repository.control_files._transport.base:
93
# check that last_revision is in 'from' and then return a no-operation.
94
if last_revision not in (None, NULL_REVISION):
95
to_repository.get_revision(last_revision)
64
97
self.to_repository = to_repository
65
98
self.from_repository = from_repository
66
self.sink = to_repository._get_sink()
67
99
# must not mutate self._last_revision as its potentially a shared instance
68
100
self._last_revision = last_revision
69
self._fetch_spec = fetch_spec
70
self.find_ghosts = find_ghosts
102
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
103
self.nested_pb = self.pb
106
self.nested_pb = None
71
107
self.from_repository.lock_read()
72
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
73
self.from_repository, self.from_repository._format,
74
self.to_repository, self.to_repository._format)
109
self.to_repository.lock_write()
111
self.to_repository.start_write_group()
115
self.to_repository.abort_write_group()
118
self.to_repository.commit_write_group()
120
if self.nested_pb is not None:
121
self.nested_pb.finished()
122
self.to_repository.unlock()
78
124
self.from_repository.unlock()
80
126
def __fetch(self):
81
127
"""Primary worker function.
83
This initialises all the needed variables, and then fetches the
129
This initialises all the needed variables, and then fetches the
84
130
requested revisions, finally clearing the progress bar.
86
# Roughly this is what we're aiming for fetch to become:
88
# missing = self.sink.insert_stream(self.source.get_stream(search))
90
# missing = self.sink.insert_stream(self.source.get_items(missing))
132
self.to_weaves = self.to_repository.weave_store
133
self.to_control = self.to_repository.control_weaves
134
self.from_weaves = self.from_repository.weave_store
135
self.from_control = self.from_repository.control_weaves
92
136
self.count_total = 0
93
137
self.file_ids_names = {}
94
pb = ui.ui_factory.nested_progress_bar()
95
pb.show_pct = pb.show_count = False
97
pb.update("Finding revisions", 0, 2)
98
search = self._revids_to_fetch()
101
pb.update("Fetching revisions", 1, 2)
102
self._fetch_everything_for_search(search)
106
def _fetch_everything_for_search(self, search):
107
"""Fetch all data for the given set of revisions."""
108
# The first phase is "file". We pass the progress bar for it directly
109
# into item_keys_introduced_by, which has more information about how
110
# that phase is progressing than we do. Progress updates for the other
111
# phases are taken care of in this function.
112
# XXX: there should be a clear owner of the progress reporting. Perhaps
113
# item_keys_introduced_by should have a richer API than it does at the
114
# moment, so that it can feed the progress information back to this
116
if (self.from_repository._format.rich_root_data and
117
not self.to_repository._format.rich_root_data):
118
raise errors.IncompatibleRepositories(
119
self.from_repository, self.to_repository,
120
"different rich-root support")
121
pb = ui.ui_factory.nested_progress_bar()
123
pb.update("Get stream source")
124
source = self.from_repository._get_source(
125
self.to_repository._format)
126
stream = source.get_stream(search)
127
from_format = self.from_repository._format
128
pb.update("Inserting stream")
129
resume_tokens, missing_keys = self.sink.insert_stream(
130
stream, from_format, [])
131
if self.to_repository._fallback_repositories:
133
self._parent_inventories(search.get_keys()))
135
pb.update("Missing keys")
136
stream = source.get_stream_for_missing_keys(missing_keys)
137
pb.update("Inserting missing keys")
138
resume_tokens, missing_keys = self.sink.insert_stream(
139
stream, from_format, resume_tokens)
141
raise AssertionError(
142
"second push failed to complete a fetch %r." % (
145
raise AssertionError(
146
"second push failed to commit the fetch %r." % (
148
pb.update("Finishing stream")
138
pp = ProgressPhase('Fetch phase', 4, self.pb)
141
revs = self._revids_to_fetch()
145
self._fetch_weave_texts(revs)
147
self._fetch_inventory_weave(revs)
149
self._fetch_revision_texts(revs)
150
self.count_copied += len(revs)
153
154
def _revids_to_fetch(self):
154
"""Determines the exact revisions needed from self.from_repository to
155
install self._last_revision in self.to_repository.
157
If no revisions need to be fetched, then this just returns None.
159
if self._fetch_spec is not None:
160
return self._fetch_spec
161
155
mutter('fetch up to rev {%s}', self._last_revision)
162
156
if self._last_revision is NULL_REVISION:
163
157
# explicit limit of no revisions needed
165
return self.to_repository.search_missing_revision_ids(
166
self.from_repository, self._last_revision,
167
find_ghosts=self.find_ghosts)
169
def _parent_inventories(self, revision_ids):
170
# Find all the parent revisions referenced by the stream, but
171
# not present in the stream, and make sure we send their
173
parent_maps = self.to_repository.get_parent_map(revision_ids)
175
map(parents.update, parent_maps.itervalues())
176
parents.discard(NULL_REVISION)
177
parents.difference_update(revision_ids)
178
missing_keys = set(('inventories', rev_id) for rev_id in parents)
159
if (self._last_revision is not None and
160
self.to_repository.has_revision(self._last_revision)):
164
return self.to_repository.missing_revision_ids(self.from_repository,
166
except errors.NoSuchRevision:
167
raise InstallFailed([self._last_revision])
169
def _fetch_weave_texts(self, revs):
170
texts_pb = bzrlib.ui.ui_factory.nested_progress_bar()
172
# fileids_altered_by_revision_ids requires reading the inventory
173
# weave, we will need to read the inventory weave again when
174
# all this is done, so enable caching for that specific weave
175
inv_w = self.from_repository.get_inventory_weave()
177
file_ids = self.from_repository.fileids_altered_by_revision_ids(revs)
179
num_file_ids = len(file_ids)
180
for file_id, required_versions in file_ids.items():
181
texts_pb.update("fetch texts", count, num_file_ids)
183
to_weave = self.to_weaves.get_weave_or_empty(file_id,
184
self.to_repository.get_transaction())
185
from_weave = self.from_weaves.get_weave(file_id,
186
self.from_repository.get_transaction())
187
# we fetch all the texts, because texts do
188
# not reference anything, and its cheap enough
189
to_weave.join(from_weave, version_ids=required_versions)
190
# we don't need *all* of this data anymore, but we dont know
191
# what we do. This cache clearing will result in a new read
192
# of the knit data when we do the checkout, but probably we
193
# want to emit the needed data on the fly rather than at the
195
# the from weave should know not to cache data being joined,
196
# but its ok to ask it to clear.
197
from_weave.clear_cache()
198
to_weave.clear_cache()
202
def _fetch_inventory_weave(self, revs):
203
pb = bzrlib.ui.ui_factory.nested_progress_bar()
205
pb.update("fetch inventory", 0, 2)
206
to_weave = self.to_control.get_weave('inventory',
207
self.to_repository.get_transaction())
209
child_pb = bzrlib.ui.ui_factory.nested_progress_bar()
211
# just merge, this is optimisable and its means we don't
212
# copy unreferenced data such as not-needed inventories.
213
pb.update("fetch inventory", 1, 3)
214
from_weave = self.from_repository.get_inventory_weave()
215
pb.update("fetch inventory", 2, 3)
216
# we fetch only the referenced inventories because we do not
217
# know for unselected inventories whether all their required
218
# texts are present in the other repository - it could be
220
to_weave.join(from_weave, pb=child_pb, msg='merge inventory',
222
from_weave.clear_cache()
229
class GenericRepoFetcher(RepoFetcher):
230
"""This is a generic repo to repo fetcher.
232
This makes minimal assumptions about repo layout and contents.
233
It triggers a reconciliation after fetching to ensure integrity.
236
def _fetch_revision_texts(self, revs):
237
"""Fetch revision object texts"""
238
rev_pb = bzrlib.ui.ui_factory.nested_progress_bar()
240
to_txn = self.to_transaction = self.to_repository.get_transaction()
243
to_store = self.to_repository._revision_store
245
pb = bzrlib.ui.ui_factory.nested_progress_bar()
247
pb.update('copying revisions', count, total)
249
sig_text = self.from_repository.get_signature_text(rev)
250
to_store.add_revision_signature_text(rev, sig_text, to_txn)
251
except errors.NoSuchRevision:
254
to_store.add_revision(self.from_repository.get_revision(rev),
259
# fixup inventory if needed:
260
# this is expensive because we have no inverse index to current ghosts.
261
# but on local disk its a few seconds and sftp push is already insane.
263
# FIXME: repository should inform if this is needed.
264
self.to_repository.reconcile()
269
class KnitRepoFetcher(RepoFetcher):
270
"""This is a knit format repository specific fetcher.
272
This differs from the GenericRepoFetcher by not doing a
273
reconciliation after copying, and using knit joining to
277
def _fetch_revision_texts(self, revs):
278
# may need to be a InterRevisionStore call here.
279
from_transaction = self.from_repository.get_transaction()
280
to_transaction = self.to_repository.get_transaction()
281
to_sf = self.to_repository._revision_store.get_signature_file(
283
from_sf = self.from_repository._revision_store.get_signature_file(
285
to_sf.join(from_sf, version_ids=revs, ignore_missing=True)
286
to_rf = self.to_repository._revision_store.get_revision_file(
288
from_rf = self.from_repository._revision_store.get_revision_file(
290
to_rf.join(from_rf, version_ids=revs)
182
293
class Inter1and2Helper(object):
183
294
"""Helper for operations that convert data from model 1 and 2
185
296
This is for use by fetchers and converters.
188
def __init__(self, source):
299
def __init__(self, source, target):
191
302
:param source: The repository data comes from
303
:param target: The repository data goes to
193
305
self.source = source
195
308
def iter_rev_trees(self, revs):
196
309
"""Iterate through RevisionTrees efficiently.
212
323
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
232
325
def generate_root_texts(self, revs):
233
326
"""Generate VersionedFiles for all root ids.
235
328
:param revs: the revisions to include
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.
250
# XXX: not covered by tests, should have a flag to always run
251
# this. -- mbp 20100129
252
graph = _get_rich_root_heads_graph(self.source, revs)
253
new_roots_stream = _new_root_data_stream(
254
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
255
return [('texts', new_roots_stream)]
258
def _get_rich_root_heads_graph(source_repo, revision_ids):
259
"""Get a Graph object suitable for asking heads() for new rich roots."""
260
st = static_tuple.StaticTuple
261
revision_keys = [st(r_id).intern() for r_id in revision_ids]
262
known_graph = source_repo.revisions.get_known_graph_ancestry(
264
return _mod_graph.GraphThunkIdsToKeys(known_graph)
267
def _new_root_data_stream(
268
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
269
"""Generate a texts substream of synthesised root entries.
271
Used in fetches that do rich-root upgrades.
273
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
274
the root entries to create.
275
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
276
calculating the parents. If a parent rev_id is not found here then it
277
will be recalculated.
278
:param parent_map: a parent map for all the revisions in
280
:param graph: a graph to use instead of repo.get_graph().
282
for root_key in root_keys_to_create:
283
root_id, rev_id = root_key
284
parent_keys = _parent_keys_for_root_version(
285
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
286
yield versionedfile.FulltextContentFactory(
287
root_key, parent_keys, None, '')
290
def _parent_keys_for_root_version(
291
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
292
"""Get the parent keys for a given root id.
294
A helper function for _new_root_data_stream.
296
# Include direct parents of the revision, but only if they used the same
297
# root_id and are heads.
298
rev_parents = parent_map[rev_id]
300
for parent_id in rev_parents:
301
if parent_id == NULL_REVISION:
303
if parent_id not in rev_id_to_root_id_map:
304
# We probably didn't read this revision, go spend the extra effort
307
tree = repo.revision_tree(parent_id)
308
except errors.NoSuchRevision:
309
# Ghost, fill out rev_id_to_root_id in case we encounter this
311
# But set parent_root_id to None since we don't really know
312
parent_root_id = None
314
parent_root_id = tree.get_root_id()
315
rev_id_to_root_id_map[parent_id] = None
317
# rev_id_to_root_id_map[parent_id] = parent_root_id
318
# memory consumption maybe?
320
parent_root_id = rev_id_to_root_id_map[parent_id]
321
if root_id == parent_root_id:
322
# With stacking we _might_ want to refer to a non-local revision,
323
# but this code path only applies when we have the full content
324
# available, so ghosts really are ghosts, not just the edge of
326
parent_ids.append(parent_id)
328
# root_id may be in the parent anyway.
330
tree = repo.revision_tree(parent_id)
331
except errors.NoSuchRevision:
332
# ghost, can't refer to it.
336
parent_ids.append(tree.inventory[root_id].revision)
337
except errors.NoSuchId:
340
# Drop non-head parents
342
graph = repo.get_graph()
343
heads = graph.heads(parent_ids)
345
for parent_id in parent_ids:
346
if parent_id in heads and parent_id not in selected_ids:
347
selected_ids.append(parent_id)
348
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
330
inventory_weave = self.source.get_inventory_weave()
333
to_store = self.target.weave_store
334
for tree in self.iter_rev_trees(revs):
335
revision_id = tree.inventory.root.revision
336
root_id = tree.inventory.root.file_id
337
parents = inventory_weave.get_parents(revision_id)
338
if root_id not in versionedfile:
339
versionedfile[root_id] = to_store.get_weave_or_empty(root_id,
340
self.target.get_transaction())
341
parent_texts[root_id] = versionedfile[root_id].add_lines(
342
revision_id, parents, [], parent_texts)
344
def regenerate_inventory(self, revs):
345
"""Generate a new inventory versionedfile in target, convertin data.
347
The inventory is retrieved from the source, (deserializing it), and
348
stored in the target (reserializing it in a different format).
349
:param revs: The revisions to include
351
inventory_weave = self.source.get_inventory_weave()
352
for tree in self.iter_rev_trees(revs):
353
parents = inventory_weave.get_parents(tree.get_revision_id())
354
self.target.add_inventory(tree.get_revision_id(), tree.inventory,
358
class Model1toKnit2Fetcher(GenericRepoFetcher):
359
"""Fetch from a Model1 repository into a Knit2 repository
361
def __init__(self, to_repository, from_repository, last_revision=None,
363
self.helper = Inter1and2Helper(from_repository, to_repository)
364
GenericRepoFetcher.__init__(self, to_repository, from_repository,
367
def _fetch_weave_texts(self, revs):
368
GenericRepoFetcher._fetch_weave_texts(self, revs)
369
# Now generate a weave for the tree root
370
self.helper.generate_root_texts(revs)
372
def _fetch_inventory_weave(self, revs):
373
self.helper.regenerate_inventory(revs)
376
class Knit1to2Fetcher(KnitRepoFetcher):
377
"""Fetch from a Knit1 repository into a Knit2 repository"""
379
def __init__(self, to_repository, from_repository, last_revision=None,
381
self.helper = Inter1and2Helper(from_repository, to_repository)
382
KnitRepoFetcher.__init__(self, to_repository, from_repository,
385
def _fetch_weave_texts(self, revs):
386
KnitRepoFetcher._fetch_weave_texts(self, revs)
387
# Now generate a weave for the tree root
388
self.helper.generate_root_texts(revs)
390
def _fetch_inventory_weave(self, revs):
391
self.helper.regenerate_inventory(revs)
394
class Fetcher(object):
395
"""Backwards compatibility glue for branch.fetch()."""
397
@deprecated_method(zero_eight)
398
def __init__(self, to_branch, from_branch, last_revision=None, pb=None):
399
"""Please see branch.fetch()."""
400
to_branch.fetch(from_branch, last_revision, pb)