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
35
import bzrlib.errors as errors
36
from bzrlib.errors import (InstallFailed, NoSuchRevision,
33
from bzrlib.revision import NULL_REVISION
34
from bzrlib.tsort import topo_sort
38
35
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
37
from bzrlib.versionedfile import FulltextContentFactory
72
40
class RepoFetcher(object):
73
41
"""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
43
This should not be used directly, it's essential a object to encapsulate
82
44
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)
47
def __init__(self, to_repository, from_repository, last_revision=None,
48
pb=None, find_ghosts=True, fetch_spec=None):
49
"""Create a repo fetcher.
51
:param last_revision: If set, try to limit to the data this revision
53
:param find_ghosts: If True search the entire history for ghosts.
54
:param pb: ProgressBar object to use; deprecated and ignored.
55
This method will just create one on top of the stack.
58
symbol_versioning.warn(
59
symbol_versioning.deprecated_in((1, 14, 0))
60
% "pb parameter to RepoFetcher.__init__")
61
# and for simplicity it is in fact ignored
62
if to_repository.has_same_location(from_repository):
63
# repository.fetch should be taking care of this case.
64
raise errors.BzrError('RepoFetcher run '
65
'between two objects at the same location: '
66
'%r and %r' % (to_repository, from_repository))
93
67
self.to_repository = to_repository
94
68
self.from_repository = from_repository
69
self.sink = to_repository._get_sink()
95
70
# must not mutate self._last_revision as its potentially a shared instance
96
71
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
72
self._fetch_spec = fetch_spec
73
self.find_ghosts = find_ghosts
103
74
self.from_repository.lock_read()
75
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
76
self.from_repository, self.from_repository._format,
77
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
81
self.from_repository.unlock()
115
83
def __fetch(self):
116
84
"""Primary worker function.
118
This initialises all the needed variables, and then fetches the
86
This initialises all the needed variables, and then fetches the
119
87
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
89
# Roughly this is what we're aiming for fetch to become:
91
# missing = self.sink.insert_stream(self.source.get_stream(search))
93
# missing = self.sink.insert_stream(self.source.get_items(missing))
125
95
self.count_total = 0
126
96
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)
97
pb = bzrlib.ui.ui_factory.nested_progress_bar()
98
pb.show_pct = pb.show_count = False
100
pb.update("Finding revisions", 0, 2)
101
search = self._revids_to_fetch()
104
pb.update("Fetching revisions", 1, 2)
105
self._fetch_everything_for_search(search)
109
def _fetch_everything_for_search(self, search):
110
"""Fetch all data for the given set of revisions."""
111
# The first phase is "file". We pass the progress bar for it directly
112
# into item_keys_introduced_by, which has more information about how
113
# that phase is progressing than we do. Progress updates for the other
114
# phases are taken care of in this function.
115
# XXX: there should be a clear owner of the progress reporting. Perhaps
116
# item_keys_introduced_by should have a richer API than it does at the
117
# moment, so that it can feed the progress information back to this
119
if (self.from_repository._format.rich_root_data and
120
not self.to_repository._format.rich_root_data):
121
raise errors.IncompatibleRepositories(
122
self.from_repository, self.to_repository,
123
"different rich-root support")
124
pb = bzrlib.ui.ui_factory.nested_progress_bar()
126
pb.update("Get stream source")
127
source = self.from_repository._get_source(
128
self.to_repository._format)
129
stream = source.get_stream(search)
130
from_format = self.from_repository._format
131
pb.update("Inserting stream")
132
resume_tokens, missing_keys = self.sink.insert_stream(
133
stream, from_format, [])
134
if self.to_repository._fallback_repositories:
136
self._parent_inventories(search.get_keys()))
138
pb.update("Missing keys")
139
stream = source.get_stream_for_missing_keys(missing_keys)
140
pb.update("Inserting missing keys")
141
resume_tokens, missing_keys = self.sink.insert_stream(
142
stream, from_format, resume_tokens)
144
raise AssertionError(
145
"second push failed to complete a fetch %r." % (
148
raise AssertionError(
149
"second push failed to commit the fetch %r." % (
151
pb.update("Finishing stream")
142
156
def _revids_to_fetch(self):
157
"""Determines the exact revisions needed from self.from_repository to
158
install self._last_revision in self.to_repository.
160
If no revisions need to be fetched, then this just returns None.
162
if self._fetch_spec is not None:
163
return self._fetch_spec
143
164
mutter('fetch up to rev {%s}', self._last_revision)
144
165
if self._last_revision is NULL_REVISION:
145
166
# 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)
168
return self.to_repository.search_missing_revision_ids(
169
self.from_repository, self._last_revision,
170
find_ghosts=self.find_ghosts)
172
def _parent_inventories(self, revision_ids):
173
# Find all the parent revisions referenced by the stream, but
174
# not present in the stream, and make sure we send their
176
parent_maps = self.to_repository.get_parent_map(revision_ids)
178
map(parents.update, parent_maps.itervalues())
179
parents.discard(NULL_REVISION)
180
parents.difference_update(revision_ids)
181
missing_keys = set(('inventories', rev_id) for rev_id in parents)
185
class Inter1and2Helper(object):
186
"""Helper for operations that convert data from model 1 and 2
188
This is for use by fetchers and converters.
191
def __init__(self, source):
194
:param source: The repository data comes from
198
def iter_rev_trees(self, revs):
199
"""Iterate through RevisionTrees efficiently.
201
Additionally, the inventory's revision_id is set if unset.
203
Trees are retrieved in batches of 100, and then yielded in the order
206
:param revs: A list of revision ids
208
# In case that revs is not a list.
211
for tree in self.source.revision_trees(revs[:100]):
212
if tree.inventory.revision_id is None:
213
tree.inventory.revision_id = tree.get_revision_id()
217
def _find_root_ids(self, revs, parent_map, graph):
219
planned_versions = {}
220
for tree in self.iter_rev_trees(revs):
221
revision_id = tree.inventory.root.revision
222
root_id = tree.get_root_id()
223
planned_versions.setdefault(root_id, []).append(revision_id)
224
revision_root[revision_id] = root_id
225
# Find out which parents we don't already know root ids for
227
for revision_parents in parent_map.itervalues():
228
parents.update(revision_parents)
229
parents.difference_update(revision_root.keys() + [NULL_REVISION])
230
# Limit to revisions present in the versionedfile
231
parents = graph.get_parent_map(parents).keys()
232
for tree in self.iter_rev_trees(parents):
233
root_id = tree.get_root_id()
234
revision_root[tree.get_revision_id()] = root_id
235
return revision_root, planned_versions
237
def generate_root_texts(self, revs):
238
"""Generate VersionedFiles for all root ids.
240
:param revs: the revisions to include
242
graph = self.source.get_graph()
243
parent_map = graph.get_parent_map(revs)
244
rev_order = topo_sort(parent_map)
245
rev_id_to_root_id, root_id_to_rev_ids = self._find_root_ids(
246
revs, parent_map, graph)
247
root_id_order = [(rev_id_to_root_id[rev_id], rev_id) for rev_id in
249
# Guaranteed stable, this groups all the file id operations together
250
# retaining topological order within the revisions of a file id.
251
# File id splits and joins would invalidate this, but they don't exist
252
# yet, and are unlikely to in non-rich-root environments anyway.
253
root_id_order.sort(key=operator.itemgetter(0))
254
# Create a record stream containing the roots to create.
256
for key in root_id_order:
257
root_id, rev_id = key
258
rev_parents = parent_map[rev_id]
259
# We drop revision parents with different file-ids, because
260
# that represents a rename of the root to a different location
261
# - its not actually a parent for us. (We could look for that
262
# file id in the revision tree at considerably more expense,
263
# but for now this is sufficient (and reconcile will catch and
264
# correct this anyway).
265
# When a parent revision is a ghost, we guess that its root id
266
# was unchanged (rather than trimming it from the parent list).
267
parent_keys = tuple((root_id, parent) for parent in rev_parents
268
if parent != NULL_REVISION and
269
rev_id_to_root_id.get(parent, root_id) == root_id)
270
yield FulltextContentFactory(key, parent_keys, None, '')
271
return [('texts', yield_roots())]
added
removed
bzrlib/fetch.py
