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,
28
from bzrlib.lazy_import import lazy_import
29
lazy_import(globals(), """
40
from bzrlib.revision import NULL_REVISION
38
41
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
44
class RepoFetcher(object):
73
45
"""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
47
This should not be used directly, it's essential a object to encapsulate
82
48
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)
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 fetch_spec: A SearchResult specifying which revisions to fetch.
58
If set, this overrides last_revision.
59
:param find_ghosts: If True search the entire history for ghosts.
61
# repository.fetch has the responsibility for short-circuiting
62
# attempts to copy between a repository and itself.
93
63
self.to_repository = to_repository
94
64
self.from_repository = from_repository
65
self.sink = to_repository._get_sink()
95
66
# must not mutate self._last_revision as its potentially a shared instance
96
67
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
68
self._fetch_spec = fetch_spec
69
self.find_ghosts = find_ghosts
103
70
self.from_repository.lock_read()
71
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
72
self.from_repository, self.from_repository._format,
73
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
77
self.from_repository.unlock()
115
79
def __fetch(self):
116
80
"""Primary worker function.
118
This initialises all the needed variables, and then fetches the
82
This initialises all the needed variables, and then fetches the
119
83
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
85
# Roughly this is what we're aiming for fetch to become:
87
# missing = self.sink.insert_stream(self.source.get_stream(search))
89
# missing = self.sink.insert_stream(self.source.get_items(missing))
125
91
self.count_total = 0
126
92
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)
93
pb = ui.ui_factory.nested_progress_bar()
94
pb.show_pct = pb.show_count = False
96
pb.update("Finding revisions", 0, 2)
97
search_result = self._revids_to_fetch()
98
mutter('fetching: %s', search_result)
99
if search_result.is_empty():
101
pb.update("Fetching revisions", 1, 2)
102
self._fetch_everything_for_search(search_result)
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, [])
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")
142
150
def _revids_to_fetch(self):
143
mutter('fetch up to rev {%s}', self._last_revision)
144
if self._last_revision is NULL_REVISION:
151
"""Determines the exact revisions needed from self.from_repository to
152
install self._last_revision in self.to_repository.
154
:returns: A SearchResult of some sort. (Possibly a
155
PendingAncestryResult, EmptySearchResult, etc.)
157
if self._fetch_spec is not None:
158
# The fetch spec is already a concrete search result.
159
return self._fetch_spec
160
elif self._last_revision == NULL_REVISION:
161
# fetch_spec is None + last_revision is null => empty fetch.
145
162
# 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)
163
return graph.EmptySearchResult()
164
elif self._last_revision is not None:
165
return graph.NotInOtherForRevs(self.to_repository,
166
self.from_repository, [self._last_revision],
167
find_ghosts=self.find_ghosts).execute()
168
else: # self._last_revision is None:
169
return graph.EverythingNotInOther(self.to_repository,
170
self.from_repository,
171
find_ghosts=self.find_ghosts).execute()
174
class Inter1and2Helper(object):
175
"""Helper for operations that convert data from model 1 and 2
177
This is for use by fetchers and converters.
180
# This is a class variable so that the test suite can override it.
181
known_graph_threshold = 100
183
def __init__(self, source):
186
:param source: The repository data comes from
190
def iter_rev_trees(self, revs):
191
"""Iterate through RevisionTrees efficiently.
193
Additionally, the inventory's revision_id is set if unset.
195
Trees are retrieved in batches of 100, and then yielded in the order
198
:param revs: A list of revision ids
200
# In case that revs is not a list.
203
for tree in self.source.revision_trees(revs[:100]):
204
if tree.inventory.revision_id is None:
205
tree.inventory.revision_id = tree.get_revision_id()
209
def _find_root_ids(self, revs, parent_map, graph):
211
for tree in self.iter_rev_trees(revs):
212
revision_id = tree.inventory.root.revision
213
root_id = tree.get_root_id()
214
revision_root[revision_id] = root_id
215
# Find out which parents we don't already know root ids for
217
for revision_parents in parent_map.itervalues():
218
parents.update(revision_parents)
219
parents.difference_update(revision_root.keys() + [NULL_REVISION])
220
# Limit to revisions present in the versionedfile
221
parents = graph.get_parent_map(parents).keys()
222
for tree in self.iter_rev_trees(parents):
223
root_id = tree.get_root_id()
224
revision_root[tree.get_revision_id()] = root_id
227
def generate_root_texts(self, revs):
228
"""Generate VersionedFiles for all root ids.
230
:param revs: the revisions to include
232
graph = self.source.get_graph()
233
parent_map = graph.get_parent_map(revs)
234
rev_order = tsort.topo_sort(parent_map)
235
rev_id_to_root_id = self._find_root_ids(revs, parent_map, graph)
236
root_id_order = [(rev_id_to_root_id[rev_id], rev_id) for rev_id in
238
# Guaranteed stable, this groups all the file id operations together
239
# retaining topological order within the revisions of a file id.
240
# File id splits and joins would invalidate this, but they don't exist
241
# yet, and are unlikely to in non-rich-root environments anyway.
242
root_id_order.sort(key=operator.itemgetter(0))
243
# Create a record stream containing the roots to create.
244
if len(revs) > self.known_graph_threshold:
245
graph = self.source.get_known_graph_ancestry(revs)
246
new_roots_stream = _new_root_data_stream(
247
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
248
return [('texts', new_roots_stream)]
251
def _new_root_data_stream(
252
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
253
"""Generate a texts substream of synthesised root entries.
255
Used in fetches that do rich-root upgrades.
257
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
258
the root entries to create.
259
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
260
calculating the parents. If a parent rev_id is not found here then it
261
will be recalculated.
262
:param parent_map: a parent map for all the revisions in
264
:param graph: a graph to use instead of repo.get_graph().
266
for root_key in root_keys_to_create:
267
root_id, rev_id = root_key
268
parent_keys = _parent_keys_for_root_version(
269
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
270
yield versionedfile.FulltextContentFactory(
271
root_key, parent_keys, None, '')
274
def _parent_keys_for_root_version(
275
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
276
"""Get the parent keys for a given root id.
278
A helper function for _new_root_data_stream.
280
# Include direct parents of the revision, but only if they used the same
281
# root_id and are heads.
282
rev_parents = parent_map[rev_id]
284
for parent_id in rev_parents:
285
if parent_id == NULL_REVISION:
287
if parent_id not in rev_id_to_root_id_map:
288
# We probably didn't read this revision, go spend the extra effort
291
tree = repo.revision_tree(parent_id)
292
except errors.NoSuchRevision:
293
# Ghost, fill out rev_id_to_root_id in case we encounter this
295
# But set parent_root_id to None since we don't really know
296
parent_root_id = None
298
parent_root_id = tree.get_root_id()
299
rev_id_to_root_id_map[parent_id] = None
301
# rev_id_to_root_id_map[parent_id] = parent_root_id
302
# memory consumption maybe?
304
parent_root_id = rev_id_to_root_id_map[parent_id]
305
if root_id == parent_root_id:
306
# With stacking we _might_ want to refer to a non-local revision,
307
# but this code path only applies when we have the full content
308
# available, so ghosts really are ghosts, not just the edge of
310
parent_ids.append(parent_id)
312
# root_id may be in the parent anyway.
314
tree = repo.revision_tree(parent_id)
315
except errors.NoSuchRevision:
316
# ghost, can't refer to it.
320
parent_ids.append(tree.get_file_revision(root_id))
321
except errors.NoSuchId:
324
# Drop non-head parents
326
graph = repo.get_graph()
327
heads = graph.heads(parent_ids)
329
for parent_id in parent_ids:
330
if parent_id in heads and parent_id not in selected_ids:
331
selected_ids.append(parent_id)
332
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
336
class TargetRepoKinds(object):
337
"""An enum-like set of constants.
339
They are the possible values of FetchSpecFactory.target_repo_kinds.
342
PREEXISTING = 'preexisting'
347
class FetchSpecFactory(object):
348
"""A helper for building the best fetch spec for a sprout call.
350
Factors that go into determining the sort of fetch to perform:
351
* did the caller specify any revision IDs?
352
* did the caller specify a source branch (need to fetch its
353
heads_to_fetch(), usually the tip + tags)
354
* is there an existing target repo (don't need to refetch revs it
356
* target is stacked? (similar to pre-existing target repo: even if
357
the target itself is new don't want to refetch existing revs)
359
:ivar source_branch: the source branch if one specified, else None.
360
:ivar source_branch_stop_revision_id: fetch up to this revision of
361
source_branch, rather than its tip.
362
:ivar source_repo: the source repository if one found, else None.
363
:ivar target_repo: the target repository acquired by sprout.
364
:ivar target_repo_kind: one of the TargetRepoKinds constants.
368
self._explicit_rev_ids = set()
369
self.source_branch = None
370
self.source_branch_stop_revision_id = None
371
self.source_repo = None
372
self.target_repo = None
373
self.target_repo_kind = None
375
def add_revision_ids(self, revision_ids):
376
"""Add revision_ids to the set of revision_ids to be fetched."""
377
self._explicit_rev_ids.update(revision_ids)
379
def make_fetch_spec(self):
380
"""Build a SearchResult or PendingAncestryResult or etc."""
381
if self.target_repo_kind is None or self.source_repo is None:
382
raise AssertionError(
383
'Incomplete FetchSpecFactory: %r' % (self.__dict__,))
384
if len(self._explicit_rev_ids) == 0 and self.source_branch is None:
385
# Caller hasn't specified any revisions or source branch
386
if self.target_repo_kind == TargetRepoKinds.EMPTY:
387
return graph.EverythingResult(self.source_repo)
389
# We want everything not already in the target (or target's
391
return graph.EverythingNotInOther(
392
self.target_repo, self.source_repo).execute()
393
heads_to_fetch = set(self._explicit_rev_ids)
394
if self.source_branch is not None:
395
must_fetch, if_present_fetch = self.source_branch.heads_to_fetch()
396
if self.source_branch_stop_revision_id is not None:
397
# Replace the tip rev from must_fetch with the stop revision
398
# XXX: this might be wrong if the tip rev is also in the
399
# must_fetch set for other reasons (e.g. it's the tip of
400
# multiple loom threads?), but then it's pretty unclear what it
401
# should mean to specify a stop_revision in that case anyway.
402
must_fetch.discard(self.source_branch.last_revision())
403
must_fetch.add(self.source_branch_stop_revision_id)
404
heads_to_fetch.update(must_fetch)
406
if_present_fetch = set()
407
if self.target_repo_kind == TargetRepoKinds.EMPTY:
408
# PendingAncestryResult does not raise errors if a requested head
409
# is absent. Ideally it would support the
410
# required_ids/if_present_ids distinction, but in practice
411
# heads_to_fetch will almost certainly be present so this doesn't
413
all_heads = heads_to_fetch.union(if_present_fetch)
414
return graph.PendingAncestryResult(all_heads, self.source_repo)
415
return graph.NotInOtherForRevs(self.target_repo, self.source_repo,
416
required_ids=heads_to_fetch, if_present_ids=if_present_fetch
added
removed
bzrlib/fetch.py
