20
21
that has merged into it. As the first step of a merge, pull, or
21
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
25
from __future__ import absolute_import
35
import bzrlib.errors as errors
36
from bzrlib.errors import (InstallFailed,
38
from bzrlib.progress import ProgressPhase
39
from bzrlib.revision import is_null, NULL_REVISION
40
from bzrlib.symbol_versioning import (deprecated_function,
43
from bzrlib.trace import mutter
29
46
from bzrlib.lazy_import import lazy_import
30
lazy_import(globals(), """
41
from bzrlib.i18n import gettext
42
from bzrlib.revision import NULL_REVISION
43
from bzrlib.trace import mutter
48
# TODO: Avoid repeatedly opening weaves so many times.
50
# XXX: This doesn't handle ghost (not present in branch) revisions at
51
# all yet. I'm not sure they really should be supported.
53
# NOTE: This doesn't copy revisions which may be present but not
54
# merged into the last revision. I'm not sure we want to do that.
56
# - get a list of revisions that need to be pulled in
57
# - for each one, pull in that revision file
58
# and get the inventory, and store the inventory with right
60
# - and get the ancestry, and store that with right parents too
61
# - and keep a note of all file ids and version seen
62
# - then go through all files; for each one get the weave,
63
# and add in all file versions
46
66
class RepoFetcher(object):
47
67
"""Pull revisions and texts from one repository to another.
49
This should not be used directly, it's essential a object to encapsulate
70
if set, try to limit to the data this revision references.
73
count_copied -- number of revisions copied
75
This should not be used directory, its essential a object to encapsulate
50
76
the logic in InterRepository.fetch().
53
def __init__(self, to_repository, from_repository, last_revision=None,
54
find_ghosts=True, fetch_spec=None):
55
"""Create a repo fetcher.
57
:param last_revision: If set, try to limit to the data this revision
59
:param fetch_spec: A SearchResult specifying which revisions to fetch.
60
If set, this overrides last_revision.
61
:param find_ghosts: If True search the entire history for ghosts.
63
# repository.fetch has the responsibility for short-circuiting
64
# attempts to copy between a repository and itself.
78
def __init__(self, to_repository, from_repository, last_revision=None, pb=None):
80
self.failed_revisions = []
82
if to_repository.has_same_location(from_repository):
83
# check that last_revision is in 'from' and then return a
85
if last_revision is not None and not is_null(last_revision):
86
to_repository.get_revision(last_revision)
65
88
self.to_repository = to_repository
66
89
self.from_repository = from_repository
67
self.sink = to_repository._get_sink()
68
90
# must not mutate self._last_revision as its potentially a shared instance
69
91
self._last_revision = last_revision
70
self._fetch_spec = fetch_spec
71
self.find_ghosts = find_ghosts
93
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
94
self.nested_pb = self.pb
72
98
self.from_repository.lock_read()
73
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
74
self.from_repository, self.from_repository._format,
75
self.to_repository, self.to_repository._format)
100
self.to_repository.lock_write()
102
self.to_repository.start_write_group()
106
self.to_repository.abort_write_group()
109
self.to_repository.commit_write_group()
111
if self.nested_pb is not None:
112
self.nested_pb.finished()
113
self.to_repository.unlock()
79
115
self.from_repository.unlock()
81
117
def __fetch(self):
82
118
"""Primary worker function.
84
This initialises all the needed variables, and then fetches the
120
This initialises all the needed variables, and then fetches the
85
121
requested revisions, finally clearing the progress bar.
87
# Roughly this is what we're aiming for fetch to become:
89
# missing = self.sink.insert_stream(self.source.get_stream(search))
91
# missing = self.sink.insert_stream(self.source.get_items(missing))
123
self.to_weaves = self.to_repository.weave_store
124
self.from_weaves = self.from_repository.weave_store
93
125
self.count_total = 0
94
126
self.file_ids_names = {}
95
pb = ui.ui_factory.nested_progress_bar()
96
pb.show_pct = pb.show_count = False
127
pp = ProgressPhase('Fetch phase', 4, self.pb)
98
pb.update(gettext("Finding revisions"), 0, 2)
99
search_result = self._revids_to_fetch()
100
mutter('fetching: %s', search_result)
101
if search_result.is_empty():
103
pb.update(gettext("Fetching revisions"), 1, 2)
104
self._fetch_everything_for_search(search_result)
130
revs = self._revids_to_fetch()
131
self._fetch_everything_for_revisions(revs, pp)
108
def _fetch_everything_for_search(self, search):
135
def _fetch_everything_for_revisions(self, revs, pp):
109
136
"""Fetch all data for the given set of revisions."""
110
139
# The first phase is "file". We pass the progress bar for it directly
111
140
# into item_keys_introduced_by, which has more information about how
112
141
# that phase is progressing than we do. Progress updates for the other
115
144
# item_keys_introduced_by should have a richer API than it does at the
116
145
# moment, so that it can feed the progress information back to this
118
if (self.from_repository._format.rich_root_data and
119
not self.to_repository._format.rich_root_data):
120
raise errors.IncompatibleRepositories(
121
self.from_repository, self.to_repository,
122
"different rich-root support")
123
pb = ui.ui_factory.nested_progress_bar()
148
pb = bzrlib.ui.ui_factory.nested_progress_bar()
125
pb.update("Get stream source")
126
source = self.from_repository._get_source(
127
self.to_repository._format)
128
stream = source.get_stream(search)
129
from_format = self.from_repository._format
130
pb.update("Inserting stream")
131
resume_tokens, missing_keys = self.sink.insert_stream(
132
stream, from_format, [])
134
pb.update("Missing keys")
135
stream = source.get_stream_for_missing_keys(missing_keys)
136
pb.update("Inserting missing keys")
137
resume_tokens, missing_keys = self.sink.insert_stream(
138
stream, from_format, resume_tokens)
140
raise AssertionError(
141
"second push failed to complete a fetch %r." % (
144
raise AssertionError(
145
"second push failed to commit the fetch %r." % (
147
pb.update("Finishing stream")
150
data_to_fetch = self.from_repository.item_keys_introduced_by(revs, pb)
151
for knit_kind, file_id, revisions in data_to_fetch:
152
if knit_kind != phase:
154
# Make a new progress bar for this phase
157
pb = bzrlib.ui.ui_factory.nested_progress_bar()
158
if knit_kind == "file":
159
self._fetch_weave_text(file_id, revisions)
160
elif knit_kind == "inventory":
162
# Once we've processed all the files, then we generate the root
163
# texts (if necessary), then we process the inventory. It's a
164
# bit distasteful to have knit_kind == "inventory" mean this,
165
# perhaps it should happen on the first non-"file" knit, in case
166
# it's not always inventory?
167
self._generate_root_texts(revs)
168
self._fetch_inventory_weave(revs, pb)
169
elif knit_kind == "signatures":
170
# Nothing to do here; this will be taken care of when
171
# _fetch_revision_texts happens.
173
elif knit_kind == "revisions":
174
self._fetch_revision_texts(revs, pb)
176
raise AssertionError("Unknown knit kind %r" % knit_kind)
180
self.count_copied += len(revs)
152
182
def _revids_to_fetch(self):
153
183
"""Determines the exact revisions needed from self.from_repository to
154
184
install self._last_revision in self.to_repository.
156
:returns: A SearchResult of some sort. (Possibly a
157
PendingAncestryResult, EmptySearchResult, etc.)
186
If no revisions need to be fetched, then this just returns None.
159
if self._fetch_spec is not None:
160
# The fetch spec is already a concrete search result.
161
return self._fetch_spec
162
elif self._last_revision == NULL_REVISION:
163
# fetch_spec is None + last_revision is null => empty fetch.
188
mutter('fetch up to rev {%s}', self._last_revision)
189
if self._last_revision is NULL_REVISION:
164
190
# explicit limit of no revisions needed
165
return vf_search.EmptySearchResult()
166
elif self._last_revision is not None:
167
return vf_search.NotInOtherForRevs(self.to_repository,
168
self.from_repository, [self._last_revision],
169
find_ghosts=self.find_ghosts).execute()
170
else: # self._last_revision is None:
171
return vf_search.EverythingNotInOther(self.to_repository,
172
self.from_repository,
173
find_ghosts=self.find_ghosts).execute()
192
if (self._last_revision is not None and
193
self.to_repository.has_revision(self._last_revision)):
197
return self.to_repository.missing_revision_ids(self.from_repository,
199
except errors.NoSuchRevision:
200
raise InstallFailed([self._last_revision])
202
def _fetch_weave_text(self, file_id, required_versions):
203
to_weave = self.to_weaves.get_weave_or_empty(file_id,
204
self.to_repository.get_transaction())
205
from_weave = self.from_weaves.get_weave(file_id,
206
self.from_repository.get_transaction())
207
# we fetch all the texts, because texts do
208
# not reference anything, and its cheap enough
209
to_weave.join(from_weave, version_ids=required_versions)
210
# we don't need *all* of this data anymore, but we dont know
211
# what we do. This cache clearing will result in a new read
212
# of the knit data when we do the checkout, but probably we
213
# want to emit the needed data on the fly rather than at the
215
# the from weave should know not to cache data being joined,
216
# but its ok to ask it to clear.
217
from_weave.clear_cache()
218
to_weave.clear_cache()
220
def _fetch_inventory_weave(self, revs, pb):
221
pb.update("fetch inventory", 0, 2)
222
to_weave = self.to_repository.get_inventory_weave()
223
child_pb = bzrlib.ui.ui_factory.nested_progress_bar()
225
# just merge, this is optimisable and its means we don't
226
# copy unreferenced data such as not-needed inventories.
227
pb.update("fetch inventory", 1, 3)
228
from_weave = self.from_repository.get_inventory_weave()
229
pb.update("fetch inventory", 2, 3)
230
# we fetch only the referenced inventories because we do not
231
# know for unselected inventories whether all their required
232
# texts are present in the other repository - it could be
234
to_weave.join(from_weave, pb=child_pb, msg='merge inventory',
236
from_weave.clear_cache()
240
def _generate_root_texts(self, revs):
241
"""This will be called by __fetch between fetching weave texts and
242
fetching the inventory weave.
244
Subclasses should override this if they need to generate root texts
245
after fetching weave texts.
250
class GenericRepoFetcher(RepoFetcher):
251
"""This is a generic repo to repo fetcher.
253
This makes minimal assumptions about repo layout and contents.
254
It triggers a reconciliation after fetching to ensure integrity.
257
def _fetch_revision_texts(self, revs, pb):
258
"""Fetch revision object texts"""
259
to_txn = self.to_transaction = self.to_repository.get_transaction()
262
to_store = self.to_repository._revision_store
264
pb.update('copying revisions', count, total)
266
sig_text = self.from_repository.get_signature_text(rev)
267
to_store.add_revision_signature_text(rev, sig_text, to_txn)
268
except errors.NoSuchRevision:
271
to_store.add_revision(self.from_repository.get_revision(rev),
274
# fixup inventory if needed:
275
# this is expensive because we have no inverse index to current ghosts.
276
# but on local disk its a few seconds and sftp push is already insane.
278
# FIXME: repository should inform if this is needed.
279
self.to_repository.reconcile()
282
class KnitRepoFetcher(RepoFetcher):
283
"""This is a knit format repository specific fetcher.
285
This differs from the GenericRepoFetcher by not doing a
286
reconciliation after copying, and using knit joining to
290
def _fetch_revision_texts(self, revs, pb):
291
# may need to be a InterRevisionStore call here.
292
from_transaction = self.from_repository.get_transaction()
293
to_transaction = self.to_repository.get_transaction()
294
to_sf = self.to_repository._revision_store.get_signature_file(
296
from_sf = self.from_repository._revision_store.get_signature_file(
298
to_sf.join(from_sf, version_ids=revs, ignore_missing=True)
299
to_rf = self.to_repository._revision_store.get_revision_file(
301
from_rf = self.from_repository._revision_store.get_revision_file(
303
to_rf.join(from_rf, version_ids=revs)
176
306
class Inter1and2Helper(object):
177
307
"""Helper for operations that convert data from model 1 and 2
179
309
This is for use by fetchers and converters.
182
# This is a class variable so that the test suite can override it.
183
known_graph_threshold = 100
185
def __init__(self, source):
312
def __init__(self, source, target):
188
315
:param source: The repository data comes from
316
:param target: The repository data goes to
190
318
self.source = source
192
321
def iter_rev_trees(self, revs):
193
322
"""Iterate through RevisionTrees efficiently.
200
329
:param revs: A list of revision ids
202
# In case that revs is not a list.
205
332
for tree in self.source.revision_trees(revs[:100]):
206
if tree.root_inventory.revision_id is None:
207
tree.root_inventory.revision_id = tree.get_revision_id()
333
if tree.inventory.revision_id is None:
334
tree.inventory.revision_id = tree.get_revision_id()
209
336
revs = revs[100:]
211
def _find_root_ids(self, revs, parent_map, graph):
213
for tree in self.iter_rev_trees(revs):
214
root_id = tree.get_root_id()
215
revision_id = tree.get_file_revision(root_id, u"")
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
338
def generate_root_texts(self, revs):
230
339
"""Generate VersionedFiles for all root ids.
232
341
: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.
246
if len(revs) > self.known_graph_threshold:
247
graph = self.source.get_known_graph_ancestry(revs)
248
new_roots_stream = _new_root_data_stream(
249
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
250
return [('texts', new_roots_stream)]
253
def _new_root_data_stream(
254
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
255
"""Generate a texts substream of synthesised root entries.
257
Used in fetches that do rich-root upgrades.
259
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
260
the root entries to create.
261
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
262
calculating the parents. If a parent rev_id is not found here then it
263
will be recalculated.
264
:param parent_map: a parent map for all the revisions in
266
:param graph: a graph to use instead of repo.get_graph().
268
for root_key in root_keys_to_create:
269
root_id, rev_id = root_key
270
parent_keys = _parent_keys_for_root_version(
271
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
272
yield versionedfile.FulltextContentFactory(
273
root_key, parent_keys, None, '')
276
def _parent_keys_for_root_version(
277
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
278
"""Get the parent keys for a given root id.
280
A helper function for _new_root_data_stream.
282
# Include direct parents of the revision, but only if they used the same
283
# root_id and are heads.
284
rev_parents = parent_map[rev_id]
286
for parent_id in rev_parents:
287
if parent_id == NULL_REVISION:
289
if parent_id not in rev_id_to_root_id_map:
290
# We probably didn't read this revision, go spend the extra effort
293
tree = repo.revision_tree(parent_id)
294
except errors.NoSuchRevision:
295
# Ghost, fill out rev_id_to_root_id in case we encounter this
297
# But set parent_root_id to None since we don't really know
298
parent_root_id = None
300
parent_root_id = tree.get_root_id()
301
rev_id_to_root_id_map[parent_id] = None
303
# rev_id_to_root_id_map[parent_id] = parent_root_id
304
# memory consumption maybe?
306
parent_root_id = rev_id_to_root_id_map[parent_id]
307
if root_id == parent_root_id:
308
# With stacking we _might_ want to refer to a non-local revision,
309
# but this code path only applies when we have the full content
310
# available, so ghosts really are ghosts, not just the edge of
312
parent_ids.append(parent_id)
314
# root_id may be in the parent anyway.
316
tree = repo.revision_tree(parent_id)
317
except errors.NoSuchRevision:
318
# ghost, can't refer to it.
322
parent_ids.append(tree.get_file_revision(root_id))
323
except errors.NoSuchId:
326
# Drop non-head parents
328
graph = repo.get_graph()
329
heads = graph.heads(parent_ids)
331
for parent_id in parent_ids:
332
if parent_id in heads and parent_id not in selected_ids:
333
selected_ids.append(parent_id)
334
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
338
class TargetRepoKinds(object):
339
"""An enum-like set of constants.
341
They are the possible values of FetchSpecFactory.target_repo_kinds.
344
PREEXISTING = 'preexisting'
349
class FetchSpecFactory(object):
350
"""A helper for building the best fetch spec for a sprout call.
352
Factors that go into determining the sort of fetch to perform:
353
* did the caller specify any revision IDs?
354
* did the caller specify a source branch (need to fetch its
355
heads_to_fetch(), usually the tip + tags)
356
* is there an existing target repo (don't need to refetch revs it
358
* target is stacked? (similar to pre-existing target repo: even if
359
the target itself is new don't want to refetch existing revs)
361
:ivar source_branch: the source branch if one specified, else None.
362
:ivar source_branch_stop_revision_id: fetch up to this revision of
363
source_branch, rather than its tip.
364
:ivar source_repo: the source repository if one found, else None.
365
:ivar target_repo: the target repository acquired by sprout.
366
:ivar target_repo_kind: one of the TargetRepoKinds constants.
370
self._explicit_rev_ids = set()
371
self.source_branch = None
372
self.source_branch_stop_revision_id = None
373
self.source_repo = None
374
self.target_repo = None
375
self.target_repo_kind = None
378
def add_revision_ids(self, revision_ids):
379
"""Add revision_ids to the set of revision_ids to be fetched."""
380
self._explicit_rev_ids.update(revision_ids)
382
def make_fetch_spec(self):
383
"""Build a SearchResult or PendingAncestryResult or etc."""
384
if self.target_repo_kind is None or self.source_repo is None:
385
raise AssertionError(
386
'Incomplete FetchSpecFactory: %r' % (self.__dict__,))
387
if len(self._explicit_rev_ids) == 0 and self.source_branch is None:
388
if self.limit is not None:
389
raise NotImplementedError(
390
"limit is only supported with a source branch set")
391
# Caller hasn't specified any revisions or source branch
392
if self.target_repo_kind == TargetRepoKinds.EMPTY:
393
return vf_search.EverythingResult(self.source_repo)
395
# We want everything not already in the target (or target's
397
return vf_search.EverythingNotInOther(
398
self.target_repo, self.source_repo).execute()
399
heads_to_fetch = set(self._explicit_rev_ids)
400
if self.source_branch is not None:
401
must_fetch, if_present_fetch = self.source_branch.heads_to_fetch()
402
if self.source_branch_stop_revision_id is not None:
403
# Replace the tip rev from must_fetch with the stop revision
404
# XXX: this might be wrong if the tip rev is also in the
405
# must_fetch set for other reasons (e.g. it's the tip of
406
# multiple loom threads?), but then it's pretty unclear what it
407
# should mean to specify a stop_revision in that case anyway.
408
must_fetch.discard(self.source_branch.last_revision())
409
must_fetch.add(self.source_branch_stop_revision_id)
410
heads_to_fetch.update(must_fetch)
412
if_present_fetch = set()
413
if self.target_repo_kind == TargetRepoKinds.EMPTY:
414
# PendingAncestryResult does not raise errors if a requested head
415
# is absent. Ideally it would support the
416
# required_ids/if_present_ids distinction, but in practice
417
# heads_to_fetch will almost certainly be present so this doesn't
419
all_heads = heads_to_fetch.union(if_present_fetch)
420
ret = vf_search.PendingAncestryResult(all_heads, self.source_repo)
421
if self.limit is not None:
422
graph = self.source_repo.get_graph()
423
topo_order = list(graph.iter_topo_order(ret.get_keys()))
424
result_set = topo_order[:self.limit]
425
ret = self.source_repo.revision_ids_to_search_result(result_set)
428
return vf_search.NotInOtherForRevs(self.target_repo, self.source_repo,
429
required_ids=heads_to_fetch, if_present_ids=if_present_fetch,
430
limit=self.limit).execute()
343
inventory_weave = self.source.get_inventory_weave()
346
to_store = self.target.weave_store
347
for tree in self.iter_rev_trees(revs):
348
revision_id = tree.inventory.root.revision
349
root_id = tree.inventory.root.file_id
350
parents = inventory_weave.get_parents(revision_id)
351
if root_id not in versionedfile:
352
versionedfile[root_id] = to_store.get_weave_or_empty(root_id,
353
self.target.get_transaction())
354
_, _, parent_texts[root_id] = versionedfile[root_id].add_lines(
355
revision_id, parents, [], parent_texts)
357
def regenerate_inventory(self, revs):
358
"""Generate a new inventory versionedfile in target, convertin data.
360
The inventory is retrieved from the source, (deserializing it), and
361
stored in the target (reserializing it in a different format).
362
:param revs: The revisions to include
364
inventory_weave = self.source.get_inventory_weave()
365
for tree in self.iter_rev_trees(revs):
366
parents = inventory_weave.get_parents(tree.get_revision_id())
367
self.target.add_inventory(tree.get_revision_id(), tree.inventory,
371
class Model1toKnit2Fetcher(GenericRepoFetcher):
372
"""Fetch from a Model1 repository into a Knit2 repository
374
def __init__(self, to_repository, from_repository, last_revision=None,
376
self.helper = Inter1and2Helper(from_repository, to_repository)
377
GenericRepoFetcher.__init__(self, to_repository, from_repository,
380
def _generate_root_texts(self, revs):
381
self.helper.generate_root_texts(revs)
383
def _fetch_inventory_weave(self, revs, pb):
384
self.helper.regenerate_inventory(revs)
387
class Knit1to2Fetcher(KnitRepoFetcher):
388
"""Fetch from a Knit1 repository into a Knit2 repository"""
390
def __init__(self, to_repository, from_repository, last_revision=None,
392
self.helper = Inter1and2Helper(from_repository, to_repository)
393
KnitRepoFetcher.__init__(self, to_repository, from_repository,
396
def _generate_root_texts(self, revs):
397
self.helper.generate_root_texts(revs)
399
def _fetch_inventory_weave(self, revs, pb):
400
self.helper.regenerate_inventory(revs)
added
removed
bzrlib/fetch.py
