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.trace import mutter
39
from bzrlib.progress import ProgressPhase
40
40
from bzrlib.revision import NULL_REVISION
41
from bzrlib.trace import mutter
41
from bzrlib.symbol_versioning import (deprecated_function,
47
# TODO: Avoid repeatedly opening weaves so many times.
49
# XXX: This doesn't handle ghost (not present in branch) revisions at
50
# all yet. I'm not sure they really should be supported.
52
# NOTE: This doesn't copy revisions which may be present but not
53
# merged into the last revision. I'm not sure we want to do that.
55
# - get a list of revisions that need to be pulled in
56
# - for each one, pull in that revision file
57
# and get the inventory, and store the inventory with right
59
# - and get the ancestry, and store that with right parents too
60
# - and keep a note of all file ids and version seen
61
# - then go through all files; for each one get the weave,
62
# and add in all file versions
65
@deprecated_function(zero_eight)
66
def greedy_fetch(to_branch, from_branch, revision=None, pb=None):
67
"""Legacy API, please see branch.fetch(from_branch, last_revision, pb)."""
68
f = Fetcher(to_branch, from_branch, revision, pb)
69
return f.count_copied, f.failed_revisions
44
74
class RepoFetcher(object):
45
75
"""Pull revisions and texts from one repository to another.
47
This should not be used directly, it's essential a object to encapsulate
78
if set, try to limit to the data this revision references.
81
count_copied -- number of revisions copied
83
This should not be used directory, its essential a object to encapsulate
48
84
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.
86
def __init__(self, to_repository, from_repository, last_revision=None, pb=None):
88
self.failed_revisions = []
90
if to_repository.control_files._transport.base == from_repository.control_files._transport.base:
91
# check that last_revision is in 'from' and then return a no-operation.
92
if last_revision not in (None, NULL_REVISION):
93
from_repository.get_revision(last_revision)
61
95
self.to_repository = to_repository
62
96
self.from_repository = from_repository
63
self.sink = to_repository._get_sink()
64
97
# must not mutate self._last_revision as its potentially a shared instance
65
98
self._last_revision = last_revision
66
self._fetch_spec = fetch_spec
67
self.find_ghosts = find_ghosts
100
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
101
self.nested_pb = self.pb
104
self.nested_pb = None
68
105
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)
107
self.to_repository.lock_write()
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
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)
103
def _fetch_everything_for_search(self, search):
104
"""Fetch all data for the given set of revisions."""
105
# The first phase is "file". We pass the progress bar for it directly
106
# into item_keys_introduced_by, which has more information about how
107
# that phase is progressing than we do. Progress updates for the other
108
# phases are taken care of in this function.
109
# XXX: there should be a clear owner of the progress reporting. Perhaps
110
# item_keys_introduced_by should have a richer API than it does at the
111
# 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()
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")
129
pp = ProgressPhase('Fetch phase', 4, self.pb)
132
revs = self._revids_to_fetch()
136
self._fetch_weave_texts(revs)
138
self._fetch_inventory_weave(revs)
140
self._fetch_revision_texts(revs)
141
self.count_copied += len(revs)
150
145
def _revids_to_fetch(self):
151
"""Determines the exact revisions needed from self.from_repository to
152
install self._last_revision in self.to_repository.
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
158
146
mutter('fetch up to rev {%s}', self._last_revision)
159
147
if self._last_revision is NULL_REVISION:
160
148
# 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)
179
class Inter1and2Helper(object):
180
"""Helper for operations that convert data from model 1 and 2
182
This is for use by fetchers and converters.
185
# This is a class variable so that the test suite can override it.
186
known_graph_threshold = 100
188
def __init__(self, source):
191
:param source: The repository data comes from
195
def iter_rev_trees(self, revs):
196
"""Iterate through RevisionTrees efficiently.
198
Additionally, the inventory's revision_id is set if unset.
200
Trees are retrieved in batches of 100, and then yielded in the order
203
:param revs: A list of revision ids
205
# In case that revs is not a list.
208
for tree in self.source.revision_trees(revs[:100]):
209
if tree.inventory.revision_id is None:
210
tree.inventory.revision_id = tree.get_revision_id()
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
def generate_root_texts(self, revs):
233
"""Generate VersionedFiles for all root ids.
235
: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.
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.
150
if (self._last_revision != None and
151
self.to_repository.has_revision(self._last_revision)):
155
return self.to_repository.missing_revision_ids(self.from_repository,
157
except errors.NoSuchRevision:
158
raise InstallFailed([self._last_revision])
160
def _fetch_weave_texts(self, revs):
161
texts_pb = bzrlib.ui.ui_factory.nested_progress_bar()
163
file_ids = self.from_repository.fileids_altered_by_revision_ids(revs)
165
num_file_ids = len(file_ids)
166
for file_id, required_versions in file_ids.items():
167
texts_pb.update("fetch texts", count, num_file_ids)
169
to_weave = self.to_weaves.get_weave_or_empty(file_id,
170
self.to_repository.get_transaction())
171
from_weave = self.from_weaves.get_weave(file_id,
172
self.from_repository.get_transaction())
173
# we fetch all the texts, because texts do
174
# not reference anything, and its cheap enough
175
to_weave.join(from_weave, version_ids=required_versions)
176
# we don't need *all* of this data anymore, but we dont know
177
# what we do. This cache clearing will result in a new read
178
# of the knit data when we do the checkout, but probably we
179
# want to emit the needed data on the fly rather than at the
181
# the from weave should know not to cache data being joined,
182
# but its ok to ask it to clear.
183
from_weave.clear_cache()
184
to_weave.clear_cache()
188
def _fetch_inventory_weave(self, revs):
189
pb = bzrlib.ui.ui_factory.nested_progress_bar()
191
pb.update("fetch inventory", 0, 2)
192
to_weave = self.to_control.get_weave('inventory',
193
self.to_repository.get_transaction())
195
child_pb = bzrlib.ui.ui_factory.nested_progress_bar()
197
# just merge, this is optimisable and its means we don't
198
# copy unreferenced data such as not-needed inventories.
199
pb.update("fetch inventory", 1, 3)
200
from_weave = self.from_repository.get_inventory_weave()
201
pb.update("fetch inventory", 2, 3)
202
# we fetch only the referenced inventories because we do not
203
# know for unselected inventories whether all their required
204
# texts are present in the other repository - it could be
206
to_weave.join(from_weave, pb=child_pb, msg='merge inventory',
214
class GenericRepoFetcher(RepoFetcher):
215
"""This is a generic repo to repo fetcher.
217
This makes minimal assumptions about repo layout and contents.
218
It triggers a reconciliation after fetching to ensure integrity.
221
def _fetch_revision_texts(self, revs):
222
"""Fetch revision object texts"""
223
rev_pb = bzrlib.ui.ui_factory.nested_progress_bar()
225
to_txn = self.to_transaction = self.to_repository.get_transaction()
228
to_store = self.to_repository._revision_store
230
pb = bzrlib.ui.ui_factory.nested_progress_bar()
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]
232
pb.update('copying revisions', count, total)
234
sig_text = self.from_repository.get_signature_text(rev)
235
to_store.add_revision_signature_text(rev, sig_text, to_txn)
236
except errors.NoSuchRevision:
239
to_store.add_revision(self.from_repository.get_revision(rev),
244
# fixup inventory if needed:
245
# this is expensive because we have no inverse index to current ghosts.
246
# but on local disk its a few seconds and sftp push is already insane.
248
# FIXME: repository should inform if this is needed.
249
self.to_repository.reconcile()
254
class KnitRepoFetcher(RepoFetcher):
255
"""This is a knit format repository specific fetcher.
257
This differs from the GenericRepoFetcher by not doing a
258
reconciliation after copying, and using knit joining to
262
def _fetch_revision_texts(self, revs):
263
# may need to be a InterRevisionStore call here.
264
from_transaction = self.from_repository.get_transaction()
265
to_transaction = self.to_repository.get_transaction()
266
to_sf = self.to_repository._revision_store.get_signature_file(
268
from_sf = self.from_repository._revision_store.get_signature_file(
270
to_sf.join(from_sf, version_ids=revs, ignore_missing=True)
271
to_rf = self.to_repository._revision_store.get_revision_file(
273
from_rf = self.from_repository._revision_store.get_revision_file(
275
to_rf.join(from_rf, version_ids=revs)
278
class Fetcher(object):
279
"""Backwards compatibility glue for branch.fetch()."""
281
@deprecated_method(zero_eight)
282
def __init__(self, to_branch, from_branch, last_revision=None, pb=None):
283
"""Please see branch.fetch()."""
284
to_branch.fetch(from_branch, last_revision, pb)
added
removed
bzrlib/fetch.py
