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, NoSuchRevision,
43
from bzrlib.revision import NULL_REVISION
38
44
from bzrlib.trace import mutter
39
from bzrlib.progress import ProgressBar, ProgressPhase
40
from bzrlib.reconcile import RepoReconciler
41
from bzrlib.revision import NULL_REVISION
42
from bzrlib.symbol_versioning import *
45
# TODO: Avoid repeatedly opening weaves so many times.
47
# XXX: This doesn't handle ghost (not present in branch) revisions at
48
# all yet. I'm not sure they really should be supported.
50
# NOTE: This doesn't copy revisions which may be present but not
51
# merged into the last revision. I'm not sure we want to do that.
53
# - get a list of revisions that need to be pulled in
54
# - for each one, pull in that revision file
55
# and get the inventory, and store the inventory with right
57
# - and get the ancestry, and store that with right parents too
58
# - and keep a note of all file ids and version seen
59
# - then go through all files; for each one get the weave,
60
# and add in all file versions
63
@deprecated_function(zero_eight)
64
def greedy_fetch(to_branch, from_branch, revision=None, pb=None):
65
"""Legacy API, please see branch.fetch(from_branch, last_revision, pb)."""
66
f = Fetcher(to_branch, from_branch, revision, pb)
67
return f.count_copied, f.failed_revisions
72
47
class RepoFetcher(object):
73
48
"""Pull revisions and texts from one repository to another.
76
if set, try to limit to the data this revision references.
79
count_copied -- number of revisions copied
81
This should not be used directory, its essential a object to encapsulate
50
This should not be used directly, it's essential a object to encapsulate
82
51
the logic in InterRepository.fetch().
84
def __init__(self, to_repository, from_repository, last_revision=None, pb=None):
86
self.failed_revisions = []
88
if to_repository.control_files._transport.base == from_repository.control_files._transport.base:
89
# check that last_revision is in 'from' and then return a no-operation.
90
if last_revision not in (None, NULL_REVISION):
91
from_repository.get_revision(last_revision)
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.
93
64
self.to_repository = to_repository
94
65
self.from_repository = from_repository
66
self.sink = to_repository._get_sink()
95
67
# must not mutate self._last_revision as its potentially a shared instance
96
68
self._last_revision = last_revision
98
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
99
self.nested_pb = self.pb
102
self.nested_pb = None
69
self._fetch_spec = fetch_spec
70
self.find_ghosts = find_ghosts
103
71
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)
105
self.to_repository.lock_write()
109
if self.nested_pb is not None:
110
self.nested_pb.finished()
111
self.to_repository.unlock()
113
78
self.from_repository.unlock()
115
80
def __fetch(self):
116
81
"""Primary worker function.
118
This initialises all the needed variables, and then fetches the
83
This initialises all the needed variables, and then fetches the
119
84
requested revisions, finally clearing the progress bar.
121
self.to_weaves = self.to_repository.weave_store
122
self.to_control = self.to_repository.control_weaves
123
self.from_weaves = self.from_repository.weave_store
124
self.from_control = self.from_repository.control_weaves
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))
125
92
self.count_total = 0
126
93
self.file_ids_names = {}
127
pp = ProgressPhase('fetch phase', 4, self.pb)
129
revs = self._revids_to_fetch()
133
self._fetch_weave_texts(revs)
135
self._fetch_inventory_weave(revs)
137
self._fetch_revision_texts(revs)
138
self.count_copied += len(revs)
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")
142
153
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
143
161
mutter('fetch up to rev {%s}', self._last_revision)
144
162
if self._last_revision is NULL_REVISION:
145
163
# explicit limit of no revisions needed
147
if (self._last_revision != None and
148
self.to_repository.has_revision(self._last_revision)):
152
return self.to_repository.missing_revision_ids(self.from_repository,
154
except errors.NoSuchRevision:
155
raise InstallFailed([self._last_revision])
157
def _fetch_weave_texts(self, revs):
158
texts_pb = bzrlib.ui.ui_factory.nested_progress_bar()
160
file_ids = self.from_repository.fileid_involved_by_set(revs)
162
num_file_ids = len(file_ids)
163
for file_id in file_ids:
164
texts_pb.update("fetch texts", count, num_file_ids)
167
to_weave = self.to_weaves.get_weave(file_id,
168
self.to_repository.get_transaction())
169
except errors.NoSuchFile:
170
# destination is empty, just copy it.
171
# this copies all the texts, which is useful and
172
# on per-file basis quite cheap.
173
self.to_weaves.copy_multi(
177
self.from_repository.get_transaction(),
178
self.to_repository.get_transaction())
180
# destination has contents, must merge
181
from_weave = self.from_weaves.get_weave(file_id,
182
self.from_repository.get_transaction())
183
# we fetch all the texts, because texts do
184
# not reference anything, and its cheap enough
185
to_weave.join(from_weave)
189
def _fetch_inventory_weave(self, revs):
190
pb = bzrlib.ui.ui_factory.nested_progress_bar()
192
pb.update("fetch inventory", 0, 2)
193
to_weave = self.to_control.get_weave('inventory',
194
self.to_repository.get_transaction())
196
child_pb = bzrlib.ui.ui_factory.nested_progress_bar()
198
# just merge, this is optimisable and its means we dont
199
# copy unreferenced data such as not-needed inventories.
200
pb.update("fetch inventory", 1, 3)
201
from_weave = self.from_repository.get_inventory_weave()
202
pb.update("fetch inventory", 2, 3)
203
# we fetch only the referenced inventories because we do not
204
# know for unselected inventories whether all their required
205
# texts are present in the other repository - it could be
207
to_weave.join(from_weave, pb=child_pb, msg='merge inventory',
215
class GenericRepoFetcher(RepoFetcher):
216
"""This is a generic repo to repo fetcher.
218
This makes minimal assumptions about repo layout and contents.
219
It triggers a reconciliation after fetching to ensure integrity.
222
def _fetch_revision_texts(self, revs):
223
"""Fetch revision object texts"""
224
rev_pb = bzrlib.ui.ui_factory.nested_progress_bar()
226
to_txn = self.to_transaction = self.to_repository.get_transaction()
229
to_store = self.to_repository._revision_store
231
pb = bzrlib.ui.ui_factory.nested_progress_bar()
233
pb.update('copying revisions', count, total)
235
sig_text = self.from_repository.get_signature_text(rev)
236
to_store.add_revision_signature_text(rev, sig_text, to_txn)
237
except errors.NoSuchRevision:
240
to_store.add_revision(self.from_repository.get_revision(rev),
245
# fixup inventory if needed:
246
# this is expensive because we have no inverse index to current ghosts.
247
# but on local disk its a few seconds and sftp push is already insane.
249
# FIXME: repository should inform if this is needed.
250
self.to_repository.reconcile()
255
class KnitRepoFetcher(RepoFetcher):
256
"""This is a knit format repository specific fetcher.
258
This differs from the GenericRepoFetcher by not doing a
259
reconciliation after copying, and using knit joining to
263
def _fetch_revision_texts(self, revs):
264
# may need to be a InterRevisionStore call here.
265
from_transaction = self.from_repository.get_transaction()
266
to_transaction = self.to_repository.get_transaction()
267
to_sf = self.to_repository._revision_store.get_signature_file(
269
from_sf = self.from_repository._revision_store.get_signature_file(
271
to_sf.join(from_sf, version_ids=revs, ignore_missing=True)
272
to_rf = self.to_repository._revision_store.get_revision_file(
274
from_rf = self.from_repository._revision_store.get_revision_file(
276
to_rf.join(from_rf, version_ids=revs)
279
class Fetcher(object):
280
"""Backwards compatability glue for branch.fetch()."""
282
@deprecated_method(zero_eight)
283
def __init__(self, to_branch, from_branch, last_revision=None, pb=None):
284
"""Please see branch.fetch()."""
285
to_branch.fetch(from_branch, last_revision, pb)
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)
182
class Inter1and2Helper(object):
183
"""Helper for operations that convert data from model 1 and 2
185
This is for use by fetchers and converters.
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.
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]
added
removed
bzrlib/fetch.py
