20
32
that has merged into it. As the first step of a merge, pull, or
21
33
branch operation we copy history from the source into the destination
36
The copying is done in a slightly complicated order. We don't want to
37
add a revision to the store until everything it refers to is also
38
stored, so that if a revision is present we can totally recreate it.
39
However, we can't know what files are included in a revision until we
40
read its inventory. Therefore, we first pull the XML and hold it in
41
memory until we've updated all of the files referenced.
25
from __future__ import absolute_import
29
from ..lazy_import import lazy_import
30
lazy_import(globals(), """
34
from breezy.bzr import (
43
from ..i18n import gettext
44
from ..revision import NULL_REVISION
45
from ..sixish import (
48
from ..trace import mutter
51
class RepoFetcher(object):
52
"""Pull revisions and texts from one repository to another.
54
This should not be used directly, it's essential a object to encapsulate
55
the logic in InterRepository.fetch().
44
# TODO: Avoid repeatedly opening weaves so many times.
46
# XXX: This doesn't handle ghost (not present in branch) revisions at
47
# all yet. I'm not sure they really should be supported.
49
# NOTE: This doesn't copy revisions which may be present but not
50
# merged into the last revision. I'm not sure we want to do that.
52
# - get a list of revisions that need to be pulled in
53
# - for each one, pull in that revision file
54
# and get the inventory, and store the inventory with right
56
# - and get the ancestry, and store that with right parents too
57
# - and keep a note of all file ids and version seen
58
# - then go through all files; for each one get the weave,
59
# and add in all file versions
63
def greedy_fetch(to_branch, from_branch, revision=None, pb=None):
64
f = Fetcher(to_branch, from_branch, revision, pb)
65
return f.count_copied, f.failed_revisions
69
class Fetcher(object):
70
"""Pull revisions and texts from one branch to another.
72
This doesn't update the destination's history; that can be done
73
separately if desired.
76
If set, pull only up to this revision_id.
80
last_revision -- if last_revision
81
is given it will be that, otherwise the last revision of
84
count_copied -- number of revisions copied
86
count_weaves -- number of file weaves copied
58
def __init__(self, to_repository, from_repository, last_revision=None,
59
find_ghosts=True, fetch_spec=None):
60
"""Create a repo fetcher.
62
:param last_revision: If set, try to limit to the data this revision
64
:param fetch_spec: A SearchResult specifying which revisions to fetch.
65
If set, this overrides last_revision.
66
:param find_ghosts: If True search the entire history for ghosts.
68
# repository.fetch has the responsibility for short-circuiting
69
# attempts to copy between a repository and itself.
70
self.to_repository = to_repository
71
self.from_repository = from_repository
72
self.sink = to_repository._get_sink()
73
# must not mutate self._last_revision as its potentially a shared instance
74
self._last_revision = last_revision
75
self._fetch_spec = fetch_spec
76
self.find_ghosts = find_ghosts
77
with self.from_repository.lock_read():
78
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
79
str(self.from_repository), str(self.from_repository._format),
80
str(self.to_repository), str(self.to_repository._format))
84
"""Primary worker function.
86
This initialises all the needed variables, and then fetches the
87
requested revisions, finally clearing the progress bar.
89
# Roughly this is what we're aiming for fetch to become:
91
# missing = self.sink.insert_stream(self.source.get_stream(search))
93
# missing = self.sink.insert_stream(self.source.get_items(missing))
88
def __init__(self, to_branch, from_branch, last_revision=None, pb=None):
89
if to_branch == from_branch:
90
raise Exception("can't fetch from a branch to itself")
91
self.to_branch = to_branch
92
self.to_weaves = to_branch.weave_store
93
self.to_control = to_branch.control_weaves
94
self.from_branch = from_branch
95
self.from_weaves = from_branch.weave_store
96
self.from_control = from_branch.control_weaves
97
self.failed_revisions = []
95
99
self.count_total = 0
96
self.file_ids_names = {}
97
with ui.ui_factory.nested_progress_bar() as pb:
98
pb.show_pct = pb.show_count = False
99
pb.update(gettext("Finding revisions"), 0, 2)
100
search_result = self._revids_to_fetch()
101
mutter('fetching: %s', str(search_result))
102
if search_result.is_empty():
104
pb.update(gettext("Fetching revisions"), 1, 2)
105
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
with ui.ui_factory.nested_progress_bar() as pb:
123
pb.update("Get stream source")
124
source = self.from_repository._get_source(
125
self.to_repository._format)
126
stream = source.get_stream(search)
127
from_format = self.from_repository._format
128
pb.update("Inserting stream")
129
resume_tokens, missing_keys = self.sink.insert_stream(
130
stream, from_format, [])
132
pb.update("Missing keys")
133
stream = source.get_stream_for_missing_keys(missing_keys)
134
pb.update("Inserting missing keys")
135
resume_tokens, missing_keys = self.sink.insert_stream(
136
stream, from_format, resume_tokens)
138
raise AssertionError(
139
"second push failed to complete a fetch %r." % (
142
raise AssertionError(
143
"second push failed to commit the fetch %r." % (
145
pb.update("Finishing stream")
148
def _revids_to_fetch(self):
149
"""Determines the exact revisions needed from self.from_repository to
150
install self._last_revision in self.to_repository.
152
:returns: A SearchResult of some sort. (Possibly a
153
PendingAncestryResult, EmptySearchResult, etc.)
155
if self._fetch_spec is not None:
156
# The fetch spec is already a concrete search result.
157
return self._fetch_spec
158
elif self._last_revision == NULL_REVISION:
159
# fetch_spec is None + last_revision is null => empty fetch.
160
# explicit limit of no revisions needed
161
return vf_search.EmptySearchResult()
162
elif self._last_revision is not None:
163
return vf_search.NotInOtherForRevs(self.to_repository,
164
self.from_repository, [
165
self._last_revision],
166
find_ghosts=self.find_ghosts).execute()
167
else: # self._last_revision is None:
168
return vf_search.EverythingNotInOther(self.to_repository,
169
self.from_repository,
170
find_ghosts=self.find_ghosts).execute()
173
class Inter1and2Helper(object):
174
"""Helper for operations that convert data from model 1 and 2
176
This is for use by fetchers and converters.
179
# This is a class variable so that the test suite can override it.
180
known_graph_threshold = 100
182
def __init__(self, source):
185
:param source: The repository data comes from
189
def iter_rev_trees(self, revs):
190
"""Iterate through RevisionTrees efficiently.
192
Additionally, the inventory's revision_id is set if unset.
194
Trees are retrieved in batches of 100, and then yielded in the order
197
:param revs: A list of revision ids
199
# In case that revs is not a list.
202
for tree in self.source.revision_trees(revs[:100]):
203
if tree.root_inventory.revision_id is None:
204
tree.root_inventory.revision_id = tree.get_revision_id()
208
def _find_root_ids(self, revs, parent_map, graph):
210
for tree in self.iter_rev_trees(revs):
211
root_id = tree.path2id('')
212
revision_id = tree.get_file_revision(u'')
213
revision_root[revision_id] = root_id
214
# Find out which parents we don't already know root ids for
215
parents = set(viewvalues(parent_map))
216
parents.difference_update(revision_root)
217
parents.discard(NULL_REVISION)
218
# Limit to revisions present in the versionedfile
219
parents = graph.get_parent_map(parents)
220
for tree in self.iter_rev_trees(parents):
221
root_id = tree.path2id('')
222
revision_root[tree.get_revision_id()] = root_id
225
def generate_root_texts(self, revs):
226
"""Generate VersionedFiles for all root ids.
228
:param revs: the revisions to include
230
graph = self.source.get_graph()
231
parent_map = graph.get_parent_map(revs)
232
rev_order = tsort.topo_sort(parent_map)
233
rev_id_to_root_id = self._find_root_ids(revs, parent_map, graph)
234
root_id_order = [(rev_id_to_root_id[rev_id], rev_id) for rev_id in
236
# Guaranteed stable, this groups all the file id operations together
237
# retaining topological order within the revisions of a file id.
238
# File id splits and joins would invalidate this, but they don't exist
239
# yet, and are unlikely to in non-rich-root environments anyway.
240
root_id_order.sort(key=operator.itemgetter(0))
241
# Create a record stream containing the roots to create.
242
if len(revs) > self.known_graph_threshold:
243
graph = self.source.get_known_graph_ancestry(revs)
244
new_roots_stream = _new_root_data_stream(
245
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
246
return [('texts', new_roots_stream)]
249
def _new_root_data_stream(
250
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
251
"""Generate a texts substream of synthesised root entries.
253
Used in fetches that do rich-root upgrades.
255
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
256
the root entries to create.
257
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
258
calculating the parents. If a parent rev_id is not found here then it
259
will be recalculated.
260
:param parent_map: a parent map for all the revisions in
262
:param graph: a graph to use instead of repo.get_graph().
264
for root_key in root_keys_to_create:
265
root_id, rev_id = root_key
266
parent_keys = _parent_keys_for_root_version(
267
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
268
yield versionedfile.ChunkedContentFactory(
269
root_key, parent_keys, None, [])
272
def _parent_keys_for_root_version(
273
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
274
"""Get the parent keys for a given root id.
276
A helper function for _new_root_data_stream.
278
# Include direct parents of the revision, but only if they used the same
279
# root_id and are heads.
280
rev_parents = parent_map[rev_id]
282
for parent_id in rev_parents:
283
if parent_id == NULL_REVISION:
285
if parent_id not in rev_id_to_root_id_map:
286
# We probably didn't read this revision, go spend the extra effort
289
tree = repo.revision_tree(parent_id)
290
except errors.NoSuchRevision:
291
# Ghost, fill out rev_id_to_root_id in case we encounter this
293
# But set parent_root_id to None since we don't really know
294
parent_root_id = None
296
parent_root_id = tree.path2id('')
297
rev_id_to_root_id_map[parent_id] = None
299
# rev_id_to_root_id_map[parent_id] = parent_root_id
300
# memory consumption maybe?
302
parent_root_id = rev_id_to_root_id_map[parent_id]
303
if root_id == parent_root_id:
304
# With stacking we _might_ want to refer to a non-local revision,
305
# but this code path only applies when we have the full content
306
# available, so ghosts really are ghosts, not just the edge of
308
parent_ids.append(parent_id)
310
# root_id may be in the parent anyway.
312
tree = repo.revision_tree(parent_id)
313
except errors.NoSuchRevision:
314
# ghost, can't refer to it.
319
tree.get_file_revision(
320
tree.id2path(root_id, recurse='none')))
321
except errors.NoSuchId:
324
# Drop non-head parents
326
graph = repo.get_graph()
327
heads = graph.heads(parent_ids)
329
for parent_id in parent_ids:
330
if parent_id in heads and parent_id not in selected_ids:
331
selected_ids.append(parent_id)
332
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
336
class TargetRepoKinds(object):
337
"""An enum-like set of constants.
339
They are the possible values of FetchSpecFactory.target_repo_kinds.
342
PREEXISTING = 'preexisting'
347
class FetchSpecFactory(object):
348
"""A helper for building the best fetch spec for a sprout call.
350
Factors that go into determining the sort of fetch to perform:
351
* did the caller specify any revision IDs?
352
* did the caller specify a source branch (need to fetch its
353
heads_to_fetch(), usually the tip + tags)
354
* is there an existing target repo (don't need to refetch revs it
356
* target is stacked? (similar to pre-existing target repo: even if
357
the target itself is new don't want to refetch existing revs)
359
:ivar source_branch: the source branch if one specified, else None.
360
:ivar source_branch_stop_revision_id: fetch up to this revision of
361
source_branch, rather than its tip.
362
:ivar source_repo: the source repository if one found, else None.
363
:ivar target_repo: the target repository acquired by sprout.
364
:ivar target_repo_kind: one of the TargetRepoKinds constants.
368
self._explicit_rev_ids = set()
369
self.source_branch = None
370
self.source_branch_stop_revision_id = None
371
self.source_repo = None
372
self.target_repo = None
373
self.target_repo_kind = None
376
def add_revision_ids(self, revision_ids):
377
"""Add revision_ids to the set of revision_ids to be fetched."""
378
self._explicit_rev_ids.update(revision_ids)
380
def make_fetch_spec(self):
381
"""Build a SearchResult or PendingAncestryResult or etc."""
382
if self.target_repo_kind is None or self.source_repo is None:
383
raise AssertionError(
384
'Incomplete FetchSpecFactory: %r' % (self.__dict__,))
385
if len(self._explicit_rev_ids) == 0 and self.source_branch is None:
386
if self.limit is not None:
387
raise NotImplementedError(
388
"limit is only supported with a source branch set")
389
# Caller hasn't specified any revisions or source branch
390
if self.target_repo_kind == TargetRepoKinds.EMPTY:
391
return vf_search.EverythingResult(self.source_repo)
393
# We want everything not already in the target (or target's
395
return vf_search.EverythingNotInOther(
396
self.target_repo, self.source_repo).execute()
397
heads_to_fetch = set(self._explicit_rev_ids)
398
if self.source_branch is not None:
399
must_fetch, if_present_fetch = self.source_branch.heads_to_fetch()
400
if self.source_branch_stop_revision_id is not None:
401
# Replace the tip rev from must_fetch with the stop revision
402
# XXX: this might be wrong if the tip rev is also in the
403
# must_fetch set for other reasons (e.g. it's the tip of
404
# multiple loom threads?), but then it's pretty unclear what it
405
# should mean to specify a stop_revision in that case anyway.
406
must_fetch.discard(self.source_branch.last_revision())
407
must_fetch.add(self.source_branch_stop_revision_id)
408
heads_to_fetch.update(must_fetch)
410
if_present_fetch = set()
411
if self.target_repo_kind == TargetRepoKinds.EMPTY:
412
# PendingAncestryResult does not raise errors if a requested head
413
# is absent. Ideally it would support the
414
# required_ids/if_present_ids distinction, but in practice
415
# heads_to_fetch will almost certainly be present so this doesn't
417
all_heads = heads_to_fetch.union(if_present_fetch)
418
ret = vf_search.PendingAncestryResult(all_heads, self.source_repo)
419
if self.limit is not None:
420
graph = self.source_repo.get_graph()
421
topo_order = list(graph.iter_topo_order(ret.get_keys()))
422
result_set = topo_order[:self.limit]
423
ret = self.source_repo.revision_ids_to_search_result(
427
return vf_search.NotInOtherForRevs(self.target_repo, self.source_repo,
428
required_ids=heads_to_fetch, if_present_ids=if_present_fetch,
429
limit=self.limit).execute()
100
self.count_weaves = 0
101
self.copied_file_ids = set()
103
self.pb = bzrlib.ui.ui_factory.progress_bar()
106
self.from_branch.lock_read()
108
self._fetch_revisions(last_revision)
110
self.from_branch.unlock()
113
def _fetch_revisions(self, last_revision):
114
self.last_revision = self._find_last_revision(last_revision)
115
mutter('fetch up to rev {%s}', self.last_revision)
116
if (self.last_revision is not None and
117
self.to_branch.has_revision(self.last_revision)):
120
revs_to_fetch = self._compare_ancestries()
122
raise InstallFailed([self.last_revision])
123
self._copy_revisions(revs_to_fetch)
124
self.new_ancestry = revs_to_fetch
126
def _find_last_revision(self, last_revision):
127
"""Find the limiting source revision.
129
Every ancestor of that revision will be merged across.
131
Returns the revision_id, or returns None if there's no history
132
in the source branch."""
135
self.pb.update('get source history')
136
from_history = self.from_branch.revision_history()
137
self.pb.update('get destination history')
139
return from_history[-1]
141
return None # no history in the source branch
144
def _compare_ancestries(self):
145
"""Get a list of revisions that must be copied.
147
That is, every revision that's in the ancestry of the source
148
branch and not in the destination branch."""
149
self.pb.update('get source ancestry')
150
self.from_ancestry = self.from_branch.get_ancestry(self.last_revision)
152
dest_last_rev = self.to_branch.last_revision()
153
self.pb.update('get destination ancestry')
155
dest_ancestry = self.to_branch.get_ancestry(dest_last_rev)
158
ss = set(dest_ancestry)
160
for rev_id in self.from_ancestry:
162
to_fetch.append(rev_id)
163
mutter('need to get revision {%s}', rev_id)
164
mutter('need to get %d revisions in total', len(to_fetch))
165
self.count_total = len(to_fetch)
168
def _copy_revisions(self, revs_to_fetch):
170
for rev_id in revs_to_fetch:
174
if self.to_branch.has_revision(rev_id):
176
self.pb.update('fetch revision', i, self.count_total)
177
self._copy_one_revision(rev_id)
178
self.count_copied += 1
181
def _copy_one_revision(self, rev_id):
182
"""Copy revision and everything referenced by it."""
183
mutter('copying revision {%s}', rev_id)
184
rev_xml = self.from_branch.get_revision_xml(rev_id)
185
inv_xml = self.from_branch.get_inventory_xml(rev_id)
186
rev = serializer_v5.read_revision_from_string(rev_xml)
187
inv = serializer_v5.read_inventory_from_string(inv_xml)
188
assert rev.revision_id == rev_id
189
assert rev.inventory_sha1 == sha_string(inv_xml)
190
mutter(' commiter %s, %d parents',
193
self._copy_new_texts(rev_id, inv)
194
parents = rev.parent_ids
195
for parent in parents:
196
if not self.to_branch.has_revision(parent):
197
parents.pop(parents.index(parent))
198
self._copy_inventory(rev_id, inv_xml, parents)
199
self.to_branch.revision_store.add(StringIO(rev_xml), rev_id)
200
mutter('copied revision %s', rev_id)
203
def _copy_inventory(self, rev_id, inv_xml, parent_ids):
204
self.to_control.add_text('inventory', rev_id,
205
split_lines(inv_xml), parent_ids,
206
self.to_branch.get_transaction())
208
def _copy_new_texts(self, rev_id, inv):
209
"""Copy any new texts occuring in this revision."""
210
# TODO: Rather than writing out weaves every time, hold them
211
# in memory until everything's done? But this way is nicer
212
# if it's interrupted.
213
for path, ie in inv.iter_entries():
214
if ie.revision != rev_id:
216
mutter('%s {%s} is changed in this revision',
218
self._copy_one_weave(rev_id, ie.file_id)
221
def _copy_one_weave(self, rev_id, file_id):
222
"""Copy one file weave."""
223
mutter('copy file {%s} modified in {%s}', file_id, rev_id)
224
if file_id in self.copied_file_ids:
225
mutter('file {%s} already copied', file_id)
227
from_weave = self.from_weaves.get_weave(file_id,
228
self.from_branch.get_transaction())
229
to_weave = self.to_weaves.get_weave_or_empty(file_id,
230
self.to_branch.get_transaction())
232
to_weave.join(from_weave)
233
except errors.WeaveParentMismatch:
234
to_weave.reweave(from_weave)
235
self.to_weaves.put_weave(file_id, to_weave,
236
self.to_branch.get_transaction())
237
self.count_weaves += 1
238
self.copied_file_ids.add(file_id)
239
mutter('copied file {%s}', file_id)
added
removed
bzrlib/fetch.py
