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
27
from ..lazy_import import lazy_import
28
lazy_import(globals(), """
32
from breezy.bzr import (
41
from ..i18n import gettext
42
from ..revision import NULL_REVISION
43
from ..trace import mutter
35
import bzrlib.errors as errors
36
from bzrlib.errors import (InstallFailed,
38
from bzrlib.progress import ProgressPhase
39
from bzrlib.revision import NULL_REVISION
40
from bzrlib.symbol_versioning import (deprecated_function,
44
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
66
@deprecated_function(zero_eight)
67
def greedy_fetch(to_branch, from_branch, revision=None, pb=None):
68
"""Legacy API, please see branch.fetch(from_branch, last_revision, pb)."""
69
f = Fetcher(to_branch, from_branch, revision, pb)
70
return f.count_copied, f.failed_revisions
46
75
class RepoFetcher(object):
47
76
"""Pull revisions and texts from one repository to another.
49
This should not be used directly, it's essential a object to encapsulate
79
if set, try to limit to the data this revision references.
82
count_copied -- number of revisions copied
84
This should not be used directory, its essential a object to encapsulate
50
85
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.
87
def __init__(self, to_repository, from_repository, last_revision=None, pb=None):
89
self.failed_revisions = []
91
if to_repository.control_files._transport.base == from_repository.control_files._transport.base:
92
# check that last_revision is in 'from' and then return a no-operation.
93
if last_revision not in (None, NULL_REVISION):
94
from_repository.get_revision(last_revision)
65
96
self.to_repository = to_repository
66
97
self.from_repository = from_repository
67
self.sink = to_repository._get_sink()
68
98
# must not mutate self._last_revision as its potentially a shared instance
69
99
self._last_revision = last_revision
70
self._fetch_spec = fetch_spec
71
self.find_ghosts = find_ghosts
72
with self.from_repository.lock_read():
73
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
74
str(self.from_repository), str(self.from_repository._format),
75
str(self.to_repository), str(self.to_repository._format))
101
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
102
self.nested_pb = self.pb
105
self.nested_pb = None
106
self.from_repository.lock_read()
108
self.to_repository.lock_write()
112
if self.nested_pb is not None:
113
self.nested_pb.finished()
114
self.to_repository.unlock()
116
self.from_repository.unlock()
78
118
def __fetch(self):
79
119
"""Primary worker function.
81
This initialises all the needed variables, and then fetches the
121
This initialises all the needed variables, and then fetches the
82
122
requested revisions, finally clearing the progress bar.
84
# Roughly this is what we're aiming for fetch to become:
86
# missing = self.sink.insert_stream(self.source.get_stream(search))
88
# missing = self.sink.insert_stream(self.source.get_items(missing))
124
self.to_weaves = self.to_repository.weave_store
125
self.to_control = self.to_repository.control_weaves
126
self.from_weaves = self.from_repository.weave_store
127
self.from_control = self.from_repository.control_weaves
90
128
self.count_total = 0
91
129
self.file_ids_names = {}
92
with ui.ui_factory.nested_progress_bar() as pb:
93
pb.show_pct = pb.show_count = False
94
pb.update(gettext("Finding revisions"), 0, 2)
95
search_result = self._revids_to_fetch()
96
mutter('fetching: %s', str(search_result))
97
if search_result.is_empty():
99
pb.update(gettext("Fetching revisions"), 1, 2)
100
self._fetch_everything_for_search(search_result)
102
def _fetch_everything_for_search(self, search):
103
"""Fetch all data for the given set of revisions."""
104
# The first phase is "file". We pass the progress bar for it directly
105
# into item_keys_introduced_by, which has more information about how
106
# that phase is progressing than we do. Progress updates for the other
107
# phases are taken care of in this function.
108
# XXX: there should be a clear owner of the progress reporting. Perhaps
109
# item_keys_introduced_by should have a richer API than it does at the
110
# moment, so that it can feed the progress information back to this
112
if (self.from_repository._format.rich_root_data and
113
not self.to_repository._format.rich_root_data):
114
raise errors.IncompatibleRepositories(
115
self.from_repository, self.to_repository,
116
"different rich-root support")
117
with ui.ui_factory.nested_progress_bar() as pb:
118
pb.update("Get stream source")
119
source = self.from_repository._get_source(
120
self.to_repository._format)
121
stream = source.get_stream(search)
122
from_format = self.from_repository._format
123
pb.update("Inserting stream")
124
resume_tokens, missing_keys = self.sink.insert_stream(
125
stream, from_format, [])
127
pb.update("Missing keys")
128
stream = source.get_stream_for_missing_keys(missing_keys)
129
pb.update("Inserting missing keys")
130
resume_tokens, missing_keys = self.sink.insert_stream(
131
stream, from_format, resume_tokens)
133
raise AssertionError(
134
"second push failed to complete a fetch %r." % (
137
raise AssertionError(
138
"second push failed to commit the fetch %r." % (
140
pb.update("Finishing stream")
130
pp = ProgressPhase('Fetch phase', 4, self.pb)
133
revs = self._revids_to_fetch()
137
self._fetch_weave_texts(revs)
139
self._fetch_inventory_weave(revs)
141
self._fetch_revision_texts(revs)
142
self.count_copied += len(revs)
143
146
def _revids_to_fetch(self):
144
"""Determines the exact revisions needed from self.from_repository to
145
install self._last_revision in self.to_repository.
147
:returns: A SearchResult of some sort. (Possibly a
148
PendingAncestryResult, EmptySearchResult, etc.)
150
if self._fetch_spec is not None:
151
# The fetch spec is already a concrete search result.
152
return self._fetch_spec
153
elif self._last_revision == NULL_REVISION:
154
# fetch_spec is None + last_revision is null => empty fetch.
147
mutter('fetch up to rev {%s}', self._last_revision)
148
if self._last_revision is NULL_REVISION:
155
149
# explicit limit of no revisions needed
156
return vf_search.EmptySearchResult()
157
elif self._last_revision is not None:
158
return vf_search.NotInOtherForRevs(self.to_repository,
159
self.from_repository, [
160
self._last_revision],
161
find_ghosts=self.find_ghosts).execute()
162
else: # self._last_revision is None:
163
return vf_search.EverythingNotInOther(self.to_repository,
164
self.from_repository,
165
find_ghosts=self.find_ghosts).execute()
151
if (self._last_revision is not None and
152
self.to_repository.has_revision(self._last_revision)):
156
return self.to_repository.missing_revision_ids(self.from_repository,
158
except errors.NoSuchRevision:
159
raise InstallFailed([self._last_revision])
161
def _fetch_weave_texts(self, revs):
162
texts_pb = bzrlib.ui.ui_factory.nested_progress_bar()
164
# fileids_altered_by_revision_ids requires reading the inventory
165
# weave, we will need to read the inventory weave again when
166
# all this is done, so enable caching for that specific weave
167
inv_w = self.from_repository.get_inventory_weave()
169
file_ids = self.from_repository.fileids_altered_by_revision_ids(revs)
171
num_file_ids = len(file_ids)
172
for file_id, required_versions in file_ids.items():
173
texts_pb.update("fetch texts", count, num_file_ids)
175
to_weave = self.to_weaves.get_weave_or_empty(file_id,
176
self.to_repository.get_transaction())
177
from_weave = self.from_weaves.get_weave(file_id,
178
self.from_repository.get_transaction())
179
# we fetch all the texts, because texts do
180
# not reference anything, and its cheap enough
181
to_weave.join(from_weave, version_ids=required_versions)
182
# we don't need *all* of this data anymore, but we dont know
183
# what we do. This cache clearing will result in a new read
184
# of the knit data when we do the checkout, but probably we
185
# want to emit the needed data on the fly rather than at the
187
# the from weave should know not to cache data being joined,
188
# but its ok to ask it to clear.
189
from_weave.clear_cache()
190
to_weave.clear_cache()
194
def _fetch_inventory_weave(self, revs):
195
pb = bzrlib.ui.ui_factory.nested_progress_bar()
197
pb.update("fetch inventory", 0, 2)
198
to_weave = self.to_control.get_weave('inventory',
199
self.to_repository.get_transaction())
201
child_pb = bzrlib.ui.ui_factory.nested_progress_bar()
203
# just merge, this is optimisable and its means we don't
204
# copy unreferenced data such as not-needed inventories.
205
pb.update("fetch inventory", 1, 3)
206
from_weave = self.from_repository.get_inventory_weave()
207
pb.update("fetch inventory", 2, 3)
208
# we fetch only the referenced inventories because we do not
209
# know for unselected inventories whether all their required
210
# texts are present in the other repository - it could be
212
to_weave.join(from_weave, pb=child_pb, msg='merge inventory',
214
from_weave.clear_cache()
221
class GenericRepoFetcher(RepoFetcher):
222
"""This is a generic repo to repo fetcher.
224
This makes minimal assumptions about repo layout and contents.
225
It triggers a reconciliation after fetching to ensure integrity.
228
def _fetch_revision_texts(self, revs):
229
"""Fetch revision object texts"""
230
rev_pb = bzrlib.ui.ui_factory.nested_progress_bar()
232
to_txn = self.to_transaction = self.to_repository.get_transaction()
235
to_store = self.to_repository._revision_store
237
pb = bzrlib.ui.ui_factory.nested_progress_bar()
239
pb.update('copying revisions', count, total)
241
sig_text = self.from_repository.get_signature_text(rev)
242
to_store.add_revision_signature_text(rev, sig_text, to_txn)
243
except errors.NoSuchRevision:
246
to_store.add_revision(self.from_repository.get_revision(rev),
251
# fixup inventory if needed:
252
# this is expensive because we have no inverse index to current ghosts.
253
# but on local disk its a few seconds and sftp push is already insane.
255
# FIXME: repository should inform if this is needed.
256
self.to_repository.reconcile()
261
class KnitRepoFetcher(RepoFetcher):
262
"""This is a knit format repository specific fetcher.
264
This differs from the GenericRepoFetcher by not doing a
265
reconciliation after copying, and using knit joining to
269
def _fetch_revision_texts(self, revs):
270
# may need to be a InterRevisionStore call here.
271
from_transaction = self.from_repository.get_transaction()
272
to_transaction = self.to_repository.get_transaction()
273
to_sf = self.to_repository._revision_store.get_signature_file(
275
from_sf = self.from_repository._revision_store.get_signature_file(
277
to_sf.join(from_sf, version_ids=revs, ignore_missing=True)
278
to_rf = self.to_repository._revision_store.get_revision_file(
280
from_rf = self.from_repository._revision_store.get_revision_file(
282
to_rf.join(from_rf, version_ids=revs)
168
285
class Inter1and2Helper(object):
169
286
"""Helper for operations that convert data from model 1 and 2
171
288
This is for use by fetchers and converters.
174
# This is a class variable so that the test suite can override it.
175
known_graph_threshold = 100
177
def __init__(self, source):
291
def __init__(self, source, target):
180
294
:param source: The repository data comes from
295
:param target: The repository data goes to
182
297
self.source = source
184
300
def iter_rev_trees(self, revs):
185
301
"""Iterate through RevisionTrees efficiently.
192
308
:param revs: A list of revision ids
194
# In case that revs is not a list.
197
311
for tree in self.source.revision_trees(revs[:100]):
198
if tree.root_inventory.revision_id is None:
199
tree.root_inventory.revision_id = tree.get_revision_id()
312
if tree.inventory.revision_id is None:
313
tree.inventory.revision_id = tree.get_revision_id()
201
315
revs = revs[100:]
203
def _find_root_ids(self, revs, parent_map, graph):
205
for tree in self.iter_rev_trees(revs):
206
root_id = tree.path2id('')
207
revision_id = tree.get_file_revision(u'')
208
revision_root[revision_id] = root_id
209
# Find out which parents we don't already know root ids for
210
parents = set(parent_map.values())
211
parents.difference_update(revision_root)
212
parents.discard(NULL_REVISION)
213
# Limit to revisions present in the versionedfile
214
parents = graph.get_parent_map(parents)
215
for tree in self.iter_rev_trees(parents):
216
root_id = tree.path2id('')
217
revision_root[tree.get_revision_id()] = root_id
220
317
def generate_root_texts(self, revs):
221
318
"""Generate VersionedFiles for all root ids.
223
320
:param revs: the revisions to include
225
graph = self.source.get_graph()
226
parent_map = graph.get_parent_map(revs)
227
rev_order = tsort.topo_sort(parent_map)
228
rev_id_to_root_id = self._find_root_ids(revs, parent_map, graph)
229
root_id_order = [(rev_id_to_root_id[rev_id], rev_id) for rev_id in
231
# Guaranteed stable, this groups all the file id operations together
232
# retaining topological order within the revisions of a file id.
233
# File id splits and joins would invalidate this, but they don't exist
234
# yet, and are unlikely to in non-rich-root environments anyway.
235
root_id_order.sort(key=operator.itemgetter(0))
236
# Create a record stream containing the roots to create.
237
if len(revs) > self.known_graph_threshold:
238
graph = self.source.get_known_graph_ancestry(revs)
239
new_roots_stream = _new_root_data_stream(
240
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
241
return [('texts', new_roots_stream)]
244
def _new_root_data_stream(
245
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
246
"""Generate a texts substream of synthesised root entries.
248
Used in fetches that do rich-root upgrades.
250
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
251
the root entries to create.
252
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
253
calculating the parents. If a parent rev_id is not found here then it
254
will be recalculated.
255
:param parent_map: a parent map for all the revisions in
257
:param graph: a graph to use instead of repo.get_graph().
259
for root_key in root_keys_to_create:
260
root_id, rev_id = root_key
261
parent_keys = _parent_keys_for_root_version(
262
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
263
yield versionedfile.ChunkedContentFactory(
264
root_key, parent_keys, None, [])
267
def _parent_keys_for_root_version(
268
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
269
"""Get the parent keys for a given root id.
271
A helper function for _new_root_data_stream.
273
# Include direct parents of the revision, but only if they used the same
274
# root_id and are heads.
275
rev_parents = parent_map[rev_id]
277
for parent_id in rev_parents:
278
if parent_id == NULL_REVISION:
280
if parent_id not in rev_id_to_root_id_map:
281
# We probably didn't read this revision, go spend the extra effort
284
tree = repo.revision_tree(parent_id)
285
except errors.NoSuchRevision:
286
# Ghost, fill out rev_id_to_root_id in case we encounter this
288
# But set parent_root_id to None since we don't really know
289
parent_root_id = None
291
parent_root_id = tree.path2id('')
292
rev_id_to_root_id_map[parent_id] = None
294
# rev_id_to_root_id_map[parent_id] = parent_root_id
295
# memory consumption maybe?
297
parent_root_id = rev_id_to_root_id_map[parent_id]
298
if root_id == parent_root_id:
299
# With stacking we _might_ want to refer to a non-local revision,
300
# but this code path only applies when we have the full content
301
# available, so ghosts really are ghosts, not just the edge of
303
parent_ids.append(parent_id)
305
# root_id may be in the parent anyway.
307
tree = repo.revision_tree(parent_id)
308
except errors.NoSuchRevision:
309
# ghost, can't refer to it.
314
tree.get_file_revision(
315
tree.id2path(root_id, recurse='none')))
316
except errors.NoSuchId:
319
# Drop non-head parents
321
graph = repo.get_graph()
322
heads = graph.heads(parent_ids)
324
for parent_id in parent_ids:
325
if parent_id in heads and parent_id not in selected_ids:
326
selected_ids.append(parent_id)
327
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
331
class TargetRepoKinds(object):
332
"""An enum-like set of constants.
334
They are the possible values of FetchSpecFactory.target_repo_kinds.
337
PREEXISTING = 'preexisting'
342
class FetchSpecFactory(object):
343
"""A helper for building the best fetch spec for a sprout call.
345
Factors that go into determining the sort of fetch to perform:
346
* did the caller specify any revision IDs?
347
* did the caller specify a source branch (need to fetch its
348
heads_to_fetch(), usually the tip + tags)
349
* is there an existing target repo (don't need to refetch revs it
351
* target is stacked? (similar to pre-existing target repo: even if
352
the target itself is new don't want to refetch existing revs)
354
:ivar source_branch: the source branch if one specified, else None.
355
:ivar source_branch_stop_revision_id: fetch up to this revision of
356
source_branch, rather than its tip.
357
:ivar source_repo: the source repository if one found, else None.
358
:ivar target_repo: the target repository acquired by sprout.
359
:ivar target_repo_kind: one of the TargetRepoKinds constants.
363
self._explicit_rev_ids = set()
364
self.source_branch = None
365
self.source_branch_stop_revision_id = None
366
self.source_repo = None
367
self.target_repo = None
368
self.target_repo_kind = None
371
def add_revision_ids(self, revision_ids):
372
"""Add revision_ids to the set of revision_ids to be fetched."""
373
self._explicit_rev_ids.update(revision_ids)
375
def make_fetch_spec(self):
376
"""Build a SearchResult or PendingAncestryResult or etc."""
377
if self.target_repo_kind is None or self.source_repo is None:
378
raise AssertionError(
379
'Incomplete FetchSpecFactory: %r' % (self.__dict__,))
380
if len(self._explicit_rev_ids) == 0 and self.source_branch is None:
381
if self.limit is not None:
382
raise NotImplementedError(
383
"limit is only supported with a source branch set")
384
# Caller hasn't specified any revisions or source branch
385
if self.target_repo_kind == TargetRepoKinds.EMPTY:
386
return vf_search.EverythingResult(self.source_repo)
388
# We want everything not already in the target (or target's
390
return vf_search.EverythingNotInOther(
391
self.target_repo, self.source_repo).execute()
392
heads_to_fetch = set(self._explicit_rev_ids)
393
if self.source_branch is not None:
394
must_fetch, if_present_fetch = self.source_branch.heads_to_fetch()
395
if self.source_branch_stop_revision_id is not None:
396
# Replace the tip rev from must_fetch with the stop revision
397
# XXX: this might be wrong if the tip rev is also in the
398
# must_fetch set for other reasons (e.g. it's the tip of
399
# multiple loom threads?), but then it's pretty unclear what it
400
# should mean to specify a stop_revision in that case anyway.
401
must_fetch.discard(self.source_branch.last_revision())
402
must_fetch.add(self.source_branch_stop_revision_id)
403
heads_to_fetch.update(must_fetch)
405
if_present_fetch = set()
406
if self.target_repo_kind == TargetRepoKinds.EMPTY:
407
# PendingAncestryResult does not raise errors if a requested head
408
# is absent. Ideally it would support the
409
# required_ids/if_present_ids distinction, but in practice
410
# heads_to_fetch will almost certainly be present so this doesn't
412
all_heads = heads_to_fetch.union(if_present_fetch)
413
ret = vf_search.PendingAncestryResult(all_heads, self.source_repo)
414
if self.limit is not None:
415
graph = self.source_repo.get_graph()
416
topo_order = list(graph.iter_topo_order(ret.get_keys()))
417
result_set = topo_order[:self.limit]
418
ret = self.source_repo.revision_ids_to_search_result(
422
return vf_search.NotInOtherForRevs(self.target_repo, self.source_repo,
423
required_ids=heads_to_fetch, if_present_ids=if_present_fetch,
424
limit=self.limit).execute()
322
inventory_weave = self.source.get_inventory_weave()
325
to_store = self.target.weave_store
326
for tree in self.iter_rev_trees(revs):
327
revision_id = tree.inventory.root.revision
328
root_id = tree.inventory.root.file_id
329
parents = inventory_weave.get_parents(revision_id)
330
if root_id not in versionedfile:
331
versionedfile[root_id] = to_store.get_weave_or_empty(root_id,
332
self.target.get_transaction())
333
parent_texts[root_id] = versionedfile[root_id].add_lines(
334
revision_id, parents, [], parent_texts)
336
def regenerate_inventory(self, revs):
337
"""Generate a new inventory versionedfile in target, convertin data.
339
The inventory is retrieved from the source, (deserializing it), and
340
stored in the target (reserializing it in a different format).
341
:param revs: The revisions to include
343
inventory_weave = self.source.get_inventory_weave()
344
for tree in self.iter_rev_trees(revs):
345
parents = inventory_weave.get_parents(tree.get_revision_id())
346
self.target.add_inventory(tree.get_revision_id(), tree.inventory,
350
class Model1toKnit2Fetcher(GenericRepoFetcher):
351
"""Fetch from a Model1 repository into a Knit2 repository
353
def __init__(self, to_repository, from_repository, last_revision=None,
355
self.helper = Inter1and2Helper(from_repository, to_repository)
356
GenericRepoFetcher.__init__(self, to_repository, from_repository,
359
def _fetch_weave_texts(self, revs):
360
GenericRepoFetcher._fetch_weave_texts(self, revs)
361
# Now generate a weave for the tree root
362
self.helper.generate_root_texts(revs)
364
def _fetch_inventory_weave(self, revs):
365
self.helper.regenerate_inventory(revs)
368
class Knit1to2Fetcher(KnitRepoFetcher):
369
"""Fetch from a Knit1 repository into a Knit2 repository"""
371
def __init__(self, to_repository, from_repository, last_revision=None,
373
self.helper = Inter1and2Helper(from_repository, to_repository)
374
KnitRepoFetcher.__init__(self, to_repository, from_repository,
377
def _fetch_weave_texts(self, revs):
378
KnitRepoFetcher._fetch_weave_texts(self, revs)
379
# Now generate a weave for the tree root
380
self.helper.generate_root_texts(revs)
382
def _fetch_inventory_weave(self, revs):
383
self.helper.regenerate_inventory(revs)
386
class Fetcher(object):
387
"""Backwards compatibility glue for branch.fetch()."""
389
@deprecated_method(zero_eight)
390
def __init__(self, to_branch, from_branch, last_revision=None, pb=None):
391
"""Please see branch.fetch()."""
392
to_branch.fetch(from_branch, last_revision, pb)
added
removed
bzrlib/fetch.py
