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,
40
from bzrlib.revision import NULL_REVISION
38
41
from bzrlib.trace import mutter
39
from bzrlib.progress import ProgressPhase
40
from bzrlib.revision import NULL_REVISION
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
74
44
class RepoFetcher(object):
75
45
"""Pull revisions and texts from one repository to another.
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
47
This should not be used directly, it's essential a object to encapsulate
84
48
the logic in InterRepository.fetch().
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)
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.
95
61
self.to_repository = to_repository
96
62
self.from_repository = from_repository
63
self.sink = to_repository._get_sink()
97
64
# must not mutate self._last_revision as its potentially a shared instance
98
65
self._last_revision = last_revision
100
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
101
self.nested_pb = self.pb
104
self.nested_pb = None
66
self._fetch_spec = fetch_spec
67
self.find_ghosts = find_ghosts
105
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)
107
self.to_repository.lock_write()
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.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
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))
127
89
self.count_total = 0
128
90
self.file_ids_names = {}
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)
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")
145
150
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
146
158
mutter('fetch up to rev {%s}', self._last_revision)
147
159
if self._last_revision is NULL_REVISION:
148
160
# explicit limit of no revisions needed
150
if (self._last_revision is not 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
# fileids_altered_by_revision_ids requires reading the inventory
164
# weave, we will need to read the inventory weave again when
165
# all this is done, so enable caching for that specific weave
166
inv_w = self.from_repository.get_inventory_weave()
168
file_ids = self.from_repository.fileids_altered_by_revision_ids(revs)
170
num_file_ids = len(file_ids)
171
for file_id, required_versions in file_ids.items():
172
texts_pb.update("fetch texts", count, num_file_ids)
174
to_weave = self.to_weaves.get_weave_or_empty(file_id,
175
self.to_repository.get_transaction())
176
from_weave = self.from_weaves.get_weave(file_id,
177
self.from_repository.get_transaction())
178
# we fetch all the texts, because texts do
179
# not reference anything, and its cheap enough
180
to_weave.join(from_weave, version_ids=required_versions)
181
# we don't need *all* of this data anymore, but we dont know
182
# what we do. This cache clearing will result in a new read
183
# of the knit data when we do the checkout, but probably we
184
# want to emit the needed data on the fly rather than at the
186
# the from weave should know not to cache data being joined,
187
# but its ok to ask it to clear.
188
from_weave.clear_cache()
189
to_weave.clear_cache()
193
def _fetch_inventory_weave(self, revs):
194
pb = bzrlib.ui.ui_factory.nested_progress_bar()
196
pb.update("fetch inventory", 0, 2)
197
to_weave = self.to_control.get_weave('inventory',
198
self.to_repository.get_transaction())
200
child_pb = bzrlib.ui.ui_factory.nested_progress_bar()
202
# just merge, this is optimisable and its means we don't
203
# copy unreferenced data such as not-needed inventories.
204
pb.update("fetch inventory", 1, 3)
205
from_weave = self.from_repository.get_inventory_weave()
206
pb.update("fetch inventory", 2, 3)
207
# we fetch only the referenced inventories because we do not
208
# know for unselected inventories whether all their required
209
# texts are present in the other repository - it could be
211
to_weave.join(from_weave, pb=child_pb, msg='merge inventory',
213
from_weave.clear_cache()
220
class GenericRepoFetcher(RepoFetcher):
221
"""This is a generic repo to repo fetcher.
223
This makes minimal assumptions about repo layout and contents.
224
It triggers a reconciliation after fetching to ensure integrity.
227
def _fetch_revision_texts(self, revs):
228
"""Fetch revision object texts"""
229
rev_pb = bzrlib.ui.ui_factory.nested_progress_bar()
231
to_txn = self.to_transaction = self.to_repository.get_transaction()
234
to_store = self.to_repository._revision_store
236
pb = bzrlib.ui.ui_factory.nested_progress_bar()
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
def __init__(self, source):
188
:param source: The repository data comes from
192
def iter_rev_trees(self, revs):
193
"""Iterate through RevisionTrees efficiently.
195
Additionally, the inventory's revision_id is set if unset.
197
Trees are retrieved in batches of 100, and then yielded in the order
200
:param revs: A list of revision ids
202
# In case that revs is not a list.
205
for tree in self.source.revision_trees(revs[:100]):
206
if tree.inventory.revision_id is None:
207
tree.inventory.revision_id = tree.get_revision_id()
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
def generate_root_texts(self, revs):
230
"""Generate VersionedFiles for all root ids.
232
: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.
238
pb.update('copying revisions', count, total)
240
sig_text = self.from_repository.get_signature_text(rev)
241
to_store.add_revision_signature_text(rev, sig_text, to_txn)
242
except errors.NoSuchRevision:
245
to_store.add_revision(self.from_repository.get_revision(rev),
250
# fixup inventory if needed:
251
# this is expensive because we have no inverse index to current ghosts.
252
# but on local disk its a few seconds and sftp push is already insane.
254
# FIXME: repository should inform if this is needed.
255
self.to_repository.reconcile()
260
class KnitRepoFetcher(RepoFetcher):
261
"""This is a knit format repository specific fetcher.
263
This differs from the GenericRepoFetcher by not doing a
264
reconciliation after copying, and using knit joining to
268
def _fetch_revision_texts(self, revs):
269
# may need to be a InterRevisionStore call here.
270
from_transaction = self.from_repository.get_transaction()
271
to_transaction = self.to_repository.get_transaction()
272
to_sf = self.to_repository._revision_store.get_signature_file(
274
from_sf = self.from_repository._revision_store.get_signature_file(
276
to_sf.join(from_sf, version_ids=revs, ignore_missing=True)
277
to_rf = self.to_repository._revision_store.get_revision_file(
279
from_rf = self.from_repository._revision_store.get_revision_file(
281
to_rf.join(from_rf, version_ids=revs)
284
class Fetcher(object):
285
"""Backwards compatibility glue for branch.fetch()."""
287
@deprecated_method(zero_eight)
288
def __init__(self, to_branch, from_branch, last_revision=None, pb=None):
289
"""Please see branch.fetch()."""
290
to_branch.fetch(from_branch, last_revision, pb)
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]
added
removed
bzrlib/fetch.py
