21
20
that has merged into it. As the first step of a merge, pull, or
22
21
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,
25
from __future__ import absolute_import
29
from bzrlib.lazy_import import lazy_import
30
lazy_import(globals(), """
41
from bzrlib.i18n import gettext
42
from bzrlib.revision import NULL_REVISION
38
43
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
46
class RepoFetcher(object):
73
47
"""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
49
This should not be used directly, it's essential a object to encapsulate
82
50
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)
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.
93
65
self.to_repository = to_repository
94
66
self.from_repository = from_repository
67
self.sink = to_repository._get_sink()
95
68
# must not mutate self._last_revision as its potentially a shared instance
96
69
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
70
self._fetch_spec = fetch_spec
71
self.find_ghosts = find_ghosts
103
72
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)
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
79
self.from_repository.unlock()
115
81
def __fetch(self):
116
82
"""Primary worker function.
118
This initialises all the needed variables, and then fetches the
84
This initialises all the needed variables, and then fetches the
119
85
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
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))
125
93
self.count_total = 0
126
94
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)
95
pb = ui.ui_factory.nested_progress_bar()
96
pb.show_pct = pb.show_count = False
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)
108
def _fetch_everything_for_search(self, search):
109
"""Fetch all data for the given set of revisions."""
110
# The first phase is "file". We pass the progress bar for it directly
111
# into item_keys_introduced_by, which has more information about how
112
# that phase is progressing than we do. Progress updates for the other
113
# phases are taken care of in this function.
114
# XXX: there should be a clear owner of the progress reporting. Perhaps
115
# item_keys_introduced_by should have a richer API than it does at the
116
# 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()
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")
142
152
def _revids_to_fetch(self):
143
mutter('fetch up to rev {%s}', self._last_revision)
144
if self._last_revision is NULL_REVISION:
153
"""Determines the exact revisions needed from self.from_repository to
154
install self._last_revision in self.to_repository.
156
:returns: A SearchResult of some sort. (Possibly a
157
PendingAncestryResult, EmptySearchResult, etc.)
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.
145
164
# 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.fileids_altered_by_revision_ids(revs)
162
num_file_ids = len(file_ids)
163
for file_id, required_versions in file_ids.items():
164
texts_pb.update("fetch texts", count, num_file_ids)
166
to_weave = self.to_weaves.get_weave_or_empty(file_id,
167
self.to_repository.get_transaction())
168
from_weave = self.from_weaves.get_weave(file_id,
169
self.from_repository.get_transaction())
170
# we fetch all the texts, because texts do
171
# not reference anything, and its cheap enough
172
to_weave.join(from_weave, version_ids=required_versions)
173
# we don't need *all* of this data anymore, but we dont know
174
# what we do. This cache clearing will result in a new read
175
# of the knit data when we do the checkout, but probably we
176
# want to emit the needed data on the fly rather than at the
178
# the from weave should know not to cache data being joined,
179
# but its ok to ask it to clear.
180
from_weave.clear_cache()
181
to_weave.clear_cache()
185
def _fetch_inventory_weave(self, revs):
186
pb = bzrlib.ui.ui_factory.nested_progress_bar()
188
pb.update("fetch inventory", 0, 2)
189
to_weave = self.to_control.get_weave('inventory',
190
self.to_repository.get_transaction())
192
child_pb = bzrlib.ui.ui_factory.nested_progress_bar()
194
# just merge, this is optimisable and its means we don't
195
# copy unreferenced data such as not-needed inventories.
196
pb.update("fetch inventory", 1, 3)
197
from_weave = self.from_repository.get_inventory_weave()
198
pb.update("fetch inventory", 2, 3)
199
# we fetch only the referenced inventories because we do not
200
# know for unselected inventories whether all their required
201
# texts are present in the other repository - it could be
203
to_weave.join(from_weave, pb=child_pb, msg='merge inventory',
211
class GenericRepoFetcher(RepoFetcher):
212
"""This is a generic repo to repo fetcher.
214
This makes minimal assumptions about repo layout and contents.
215
It triggers a reconciliation after fetching to ensure integrity.
218
def _fetch_revision_texts(self, revs):
219
"""Fetch revision object texts"""
220
rev_pb = bzrlib.ui.ui_factory.nested_progress_bar()
222
to_txn = self.to_transaction = self.to_repository.get_transaction()
225
to_store = self.to_repository._revision_store
227
pb = bzrlib.ui.ui_factory.nested_progress_bar()
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()
176
class Inter1and2Helper(object):
177
"""Helper for operations that convert data from model 1 and 2
179
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):
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.root_inventory.revision_id is None:
207
tree.root_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
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
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.
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.
229
pb.update('copying revisions', count, total)
231
sig_text = self.from_repository.get_signature_text(rev)
232
to_store.add_revision_signature_text(rev, sig_text, to_txn)
233
except errors.NoSuchRevision:
236
to_store.add_revision(self.from_repository.get_revision(rev),
241
# fixup inventory if needed:
242
# this is expensive because we have no inverse index to current ghosts.
243
# but on local disk its a few seconds and sftp push is already insane.
245
# FIXME: repository should inform if this is needed.
246
self.to_repository.reconcile()
251
class KnitRepoFetcher(RepoFetcher):
252
"""This is a knit format repository specific fetcher.
254
This differs from the GenericRepoFetcher by not doing a
255
reconciliation after copying, and using knit joining to
259
def _fetch_revision_texts(self, revs):
260
# may need to be a InterRevisionStore call here.
261
from_transaction = self.from_repository.get_transaction()
262
to_transaction = self.to_repository.get_transaction()
263
to_sf = self.to_repository._revision_store.get_signature_file(
265
from_sf = self.from_repository._revision_store.get_signature_file(
267
to_sf.join(from_sf, version_ids=revs, ignore_missing=True)
268
to_rf = self.to_repository._revision_store.get_revision_file(
270
from_rf = self.from_repository._revision_store.get_revision_file(
272
to_rf.join(from_rf, version_ids=revs)
275
class Fetcher(object):
276
"""Backwards compatibility glue for branch.fetch()."""
278
@deprecated_method(zero_eight)
279
def __init__(self, to_branch, from_branch, last_revision=None, pb=None):
280
"""Please see branch.fetch()."""
281
to_branch.fetch(from_branch, last_revision, pb)
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()
added
removed
bzrlib/fetch.py
