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,
38
from bzrlib.progress import ProgressPhase
28
from bzrlib.lazy_import import lazy_import
29
lazy_import(globals(), """
40
from bzrlib.i18n import gettext
39
41
from bzrlib.revision import NULL_REVISION
40
from bzrlib.symbol_versioning import (deprecated_function,
44
42
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
75
45
class RepoFetcher(object):
76
46
"""Pull revisions and texts from one repository to another.
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
48
This should not be used directly, it's essential a object to encapsulate
85
49
the logic in InterRepository.fetch().
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)
52
def __init__(self, to_repository, from_repository, last_revision=None,
53
find_ghosts=True, fetch_spec=None):
54
"""Create a repo fetcher.
56
:param last_revision: If set, try to limit to the data this revision
58
:param fetch_spec: A SearchResult specifying which revisions to fetch.
59
If set, this overrides last_revision.
60
:param find_ghosts: If True search the entire history for ghosts.
62
# repository.fetch has the responsibility for short-circuiting
63
# attempts to copy between a repository and itself.
96
64
self.to_repository = to_repository
97
65
self.from_repository = from_repository
66
self.sink = to_repository._get_sink()
98
67
# must not mutate self._last_revision as its potentially a shared instance
99
68
self._last_revision = last_revision
101
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
102
self.nested_pb = self.pb
105
self.nested_pb = None
69
self._fetch_spec = fetch_spec
70
self.find_ghosts = find_ghosts
106
71
self.from_repository.lock_read()
72
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
73
self.from_repository, self.from_repository._format,
74
self.to_repository, self.to_repository._format)
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
78
self.from_repository.unlock()
118
80
def __fetch(self):
119
81
"""Primary worker function.
121
This initialises all the needed variables, and then fetches the
83
This initialises all the needed variables, and then fetches the
122
84
requested revisions, finally clearing the progress bar.
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
86
# Roughly this is what we're aiming for fetch to become:
88
# missing = self.sink.insert_stream(self.source.get_stream(search))
90
# missing = self.sink.insert_stream(self.source.get_items(missing))
128
92
self.count_total = 0
129
93
self.file_ids_names = {}
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)
94
pb = ui.ui_factory.nested_progress_bar()
95
pb.show_pct = pb.show_count = False
97
pb.update(gettext("Finding revisions"), 0, 2)
98
search_result = self._revids_to_fetch()
99
mutter('fetching: %s', search_result)
100
if search_result.is_empty():
102
pb.update(gettext("Fetching revisions"), 1, 2)
103
self._fetch_everything_for_search(search_result)
107
def _fetch_everything_for_search(self, search):
108
"""Fetch all data for the given set of revisions."""
109
# The first phase is "file". We pass the progress bar for it directly
110
# into item_keys_introduced_by, which has more information about how
111
# that phase is progressing than we do. Progress updates for the other
112
# phases are taken care of in this function.
113
# XXX: there should be a clear owner of the progress reporting. Perhaps
114
# item_keys_introduced_by should have a richer API than it does at the
115
# moment, so that it can feed the progress information back to this
117
if (self.from_repository._format.rich_root_data and
118
not self.to_repository._format.rich_root_data):
119
raise errors.IncompatibleRepositories(
120
self.from_repository, self.to_repository,
121
"different rich-root support")
122
pb = ui.ui_factory.nested_progress_bar()
124
pb.update("Get stream source")
125
source = self.from_repository._get_source(
126
self.to_repository._format)
127
stream = source.get_stream(search)
128
from_format = self.from_repository._format
129
pb.update("Inserting stream")
130
resume_tokens, missing_keys = self.sink.insert_stream(
131
stream, from_format, [])
133
pb.update("Missing keys")
134
stream = source.get_stream_for_missing_keys(missing_keys)
135
pb.update("Inserting missing keys")
136
resume_tokens, missing_keys = self.sink.insert_stream(
137
stream, from_format, resume_tokens)
139
raise AssertionError(
140
"second push failed to complete a fetch %r." % (
143
raise AssertionError(
144
"second push failed to commit the fetch %r." % (
146
pb.update("Finishing stream")
146
151
def _revids_to_fetch(self):
147
mutter('fetch up to rev {%s}', self._last_revision)
148
if self._last_revision is NULL_REVISION:
152
"""Determines the exact revisions needed from self.from_repository to
153
install self._last_revision in self.to_repository.
155
:returns: A SearchResult of some sort. (Possibly a
156
PendingAncestryResult, EmptySearchResult, etc.)
158
if self._fetch_spec is not None:
159
# The fetch spec is already a concrete search result.
160
return self._fetch_spec
161
elif self._last_revision == NULL_REVISION:
162
# fetch_spec is None + last_revision is null => empty fetch.
149
163
# explicit limit of no revisions needed
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)
164
return _mod_graph.EmptySearchResult()
165
elif self._last_revision is not None:
166
return _mod_graph.NotInOtherForRevs(self.to_repository,
167
self.from_repository, [self._last_revision],
168
find_ghosts=self.find_ghosts).execute()
169
else: # self._last_revision is None:
170
return _mod_graph.EverythingNotInOther(self.to_repository,
171
self.from_repository,
172
find_ghosts=self.find_ghosts).execute()
285
175
class Inter1and2Helper(object):
286
176
"""Helper for operations that convert data from model 1 and 2
288
178
This is for use by fetchers and converters.
291
def __init__(self, source, target):
181
# This is a class variable so that the test suite can override it.
182
known_graph_threshold = 100
184
def __init__(self, source):
294
187
:param source: The repository data comes from
295
:param target: The repository data goes to
297
189
self.source = source
300
191
def iter_rev_trees(self, revs):
301
192
"""Iterate through RevisionTrees efficiently.
315
208
revs = revs[100:]
210
def _find_root_ids(self, revs, parent_map, graph):
212
for tree in self.iter_rev_trees(revs):
213
revision_id = tree.inventory.root.revision
214
root_id = tree.get_root_id()
215
revision_root[revision_id] = root_id
216
# Find out which parents we don't already know root ids for
218
for revision_parents in parent_map.itervalues():
219
parents.update(revision_parents)
220
parents.difference_update(revision_root.keys() + [NULL_REVISION])
221
# Limit to revisions present in the versionedfile
222
parents = graph.get_parent_map(parents).keys()
223
for tree in self.iter_rev_trees(parents):
224
root_id = tree.get_root_id()
225
revision_root[tree.get_revision_id()] = root_id
317
228
def generate_root_texts(self, revs):
318
229
"""Generate VersionedFiles for all root ids.
320
231
:param revs: the revisions to include
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)
233
graph = self.source.get_graph()
234
parent_map = graph.get_parent_map(revs)
235
rev_order = tsort.topo_sort(parent_map)
236
rev_id_to_root_id = self._find_root_ids(revs, parent_map, graph)
237
root_id_order = [(rev_id_to_root_id[rev_id], rev_id) for rev_id in
239
# Guaranteed stable, this groups all the file id operations together
240
# retaining topological order within the revisions of a file id.
241
# File id splits and joins would invalidate this, but they don't exist
242
# yet, and are unlikely to in non-rich-root environments anyway.
243
root_id_order.sort(key=operator.itemgetter(0))
244
# Create a record stream containing the roots to create.
245
if len(revs) > self.known_graph_threshold:
246
graph = self.source.get_known_graph_ancestry(revs)
247
new_roots_stream = _new_root_data_stream(
248
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
249
return [('texts', new_roots_stream)]
252
def _new_root_data_stream(
253
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
254
"""Generate a texts substream of synthesised root entries.
256
Used in fetches that do rich-root upgrades.
258
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
259
the root entries to create.
260
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
261
calculating the parents. If a parent rev_id is not found here then it
262
will be recalculated.
263
:param parent_map: a parent map for all the revisions in
265
:param graph: a graph to use instead of repo.get_graph().
267
for root_key in root_keys_to_create:
268
root_id, rev_id = root_key
269
parent_keys = _parent_keys_for_root_version(
270
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
271
yield versionedfile.FulltextContentFactory(
272
root_key, parent_keys, None, '')
275
def _parent_keys_for_root_version(
276
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
277
"""Get the parent keys for a given root id.
279
A helper function for _new_root_data_stream.
281
# Include direct parents of the revision, but only if they used the same
282
# root_id and are heads.
283
rev_parents = parent_map[rev_id]
285
for parent_id in rev_parents:
286
if parent_id == NULL_REVISION:
288
if parent_id not in rev_id_to_root_id_map:
289
# We probably didn't read this revision, go spend the extra effort
292
tree = repo.revision_tree(parent_id)
293
except errors.NoSuchRevision:
294
# Ghost, fill out rev_id_to_root_id in case we encounter this
296
# But set parent_root_id to None since we don't really know
297
parent_root_id = None
299
parent_root_id = tree.get_root_id()
300
rev_id_to_root_id_map[parent_id] = None
302
# rev_id_to_root_id_map[parent_id] = parent_root_id
303
# memory consumption maybe?
305
parent_root_id = rev_id_to_root_id_map[parent_id]
306
if root_id == parent_root_id:
307
# With stacking we _might_ want to refer to a non-local revision,
308
# but this code path only applies when we have the full content
309
# available, so ghosts really are ghosts, not just the edge of
311
parent_ids.append(parent_id)
313
# root_id may be in the parent anyway.
315
tree = repo.revision_tree(parent_id)
316
except errors.NoSuchRevision:
317
# ghost, can't refer to it.
321
parent_ids.append(tree.get_file_revision(root_id))
322
except errors.NoSuchId:
325
# Drop non-head parents
327
graph = repo.get_graph()
328
heads = graph.heads(parent_ids)
330
for parent_id in parent_ids:
331
if parent_id in heads and parent_id not in selected_ids:
332
selected_ids.append(parent_id)
333
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
337
class TargetRepoKinds(object):
338
"""An enum-like set of constants.
340
They are the possible values of FetchSpecFactory.target_repo_kinds.
343
PREEXISTING = 'preexisting'
348
class FetchSpecFactory(object):
349
"""A helper for building the best fetch spec for a sprout call.
351
Factors that go into determining the sort of fetch to perform:
352
* did the caller specify any revision IDs?
353
* did the caller specify a source branch (need to fetch its
354
heads_to_fetch(), usually the tip + tags)
355
* is there an existing target repo (don't need to refetch revs it
357
* target is stacked? (similar to pre-existing target repo: even if
358
the target itself is new don't want to refetch existing revs)
360
:ivar source_branch: the source branch if one specified, else None.
361
:ivar source_branch_stop_revision_id: fetch up to this revision of
362
source_branch, rather than its tip.
363
:ivar source_repo: the source repository if one found, else None.
364
:ivar target_repo: the target repository acquired by sprout.
365
:ivar target_repo_kind: one of the TargetRepoKinds constants.
369
self._explicit_rev_ids = set()
370
self.source_branch = None
371
self.source_branch_stop_revision_id = None
372
self.source_repo = None
373
self.target_repo = None
374
self.target_repo_kind = None
377
def add_revision_ids(self, revision_ids):
378
"""Add revision_ids to the set of revision_ids to be fetched."""
379
self._explicit_rev_ids.update(revision_ids)
381
def make_fetch_spec(self):
382
"""Build a SearchResult or PendingAncestryResult or etc."""
383
if self.target_repo_kind is None or self.source_repo is None:
384
raise AssertionError(
385
'Incomplete FetchSpecFactory: %r' % (self.__dict__,))
386
if len(self._explicit_rev_ids) == 0 and self.source_branch is None:
387
if self.limit is not None:
388
raise NotImplementedError(
389
"limit is only supported with a source branch set")
390
# Caller hasn't specified any revisions or source branch
391
if self.target_repo_kind == TargetRepoKinds.EMPTY:
392
return _mod_graph.EverythingResult(self.source_repo)
394
# We want everything not already in the target (or target's
396
return _mod_graph.EverythingNotInOther(
397
self.target_repo, self.source_repo).execute()
398
heads_to_fetch = set(self._explicit_rev_ids)
399
if self.source_branch is not None:
400
must_fetch, if_present_fetch = self.source_branch.heads_to_fetch()
401
if self.source_branch_stop_revision_id is not None:
402
# Replace the tip rev from must_fetch with the stop revision
403
# XXX: this might be wrong if the tip rev is also in the
404
# must_fetch set for other reasons (e.g. it's the tip of
405
# multiple loom threads?), but then it's pretty unclear what it
406
# should mean to specify a stop_revision in that case anyway.
407
must_fetch.discard(self.source_branch.last_revision())
408
must_fetch.add(self.source_branch_stop_revision_id)
409
heads_to_fetch.update(must_fetch)
411
if_present_fetch = set()
412
if self.target_repo_kind == TargetRepoKinds.EMPTY:
413
# PendingAncestryResult does not raise errors if a requested head
414
# is absent. Ideally it would support the
415
# required_ids/if_present_ids distinction, but in practice
416
# heads_to_fetch will almost certainly be present so this doesn't
418
all_heads = heads_to_fetch.union(if_present_fetch)
419
ret = _mod_graph.PendingAncestryResult(all_heads, self.source_repo)
420
if self.limit is not None:
421
graph = self.source_repo.get_graph()
422
topo_order = list(graph.iter_topo_order(ret.get_keys()))
423
result_set = topo_order[:self.limit]
424
ret = self.source_repo.revision_ids_to_search_result(result_set)
427
return _mod_graph.NotInOtherForRevs(self.target_repo, self.source_repo,
428
required_ids=heads_to_fetch, if_present_ids=if_present_fetch,
429
limit=self.limit).execute()
added
removed
bzrlib/fetch.py
