21
34
that has merged into it. As the first step of a merge, pull, or
22
35
branch operation we copy history from the source into the destination
38
The copying is done in a slightly complicated order. We don't want to
39
add a revision to the store until everything it refers to is also
40
stored, so that if a revision is present we can totally recreate it.
41
However, we can't know what files are included in a revision until we
42
read its inventory. Therefore, we first pull the XML and hold it in
43
memory until we've updated all of the files referenced.
28
from bzrlib.lazy_import import lazy_import
29
lazy_import(globals(), """
40
from bzrlib.revision import NULL_REVISION
41
from bzrlib.trace import mutter
44
class RepoFetcher(object):
45
"""Pull revisions and texts from one repository to another.
47
This should not be used directly, it's essential a object to encapsulate
48
the logic in InterRepository.fetch().
46
# TODO: Avoid repeatedly opening weaves so many times.
48
# XXX: This doesn't handle ghost (not present in branch) revisions at
49
# all yet. I'm not sure they really should be supported.
51
# NOTE: This doesn't copy revisions which may be present but not
52
# merged into the last revision. I'm not sure we want to do that.
54
# - get a list of revisions that need to be pulled in
55
# - for each one, pull in that revision file
56
# and get the inventory, and store the inventory with right
58
# - and get the ancestry, and store that with right parents too
59
# - and keep a note of all file ids and version seen
60
# - then go through all files; for each one get the weave,
61
# and add in all file versions
65
def greedy_fetch(to_branch, from_branch, revision=None, pb=None):
66
f = Fetcher(to_branch, from_branch, revision, pb)
67
return f.count_copied, f.failed_revisions
71
class Fetcher(object):
72
"""Pull revisions and texts from one branch to another.
74
This doesn't update the destination's history; that can be done
75
separately if desired.
78
If set, pull only up to this revision_id.
82
last_revision -- if last_revision
83
is given it will be that, otherwise the last revision of
86
count_copied -- number of revisions copied
88
count_weaves -- number of file weaves copied
51
def __init__(self, to_repository, from_repository, last_revision=None,
52
find_ghosts=True, fetch_spec=None):
53
"""Create a repo fetcher.
55
:param last_revision: If set, try to limit to the data this revision
57
:param find_ghosts: If True search the entire history for ghosts.
59
# repository.fetch has the responsibility for short-circuiting
60
# attempts to copy between a repository and itself.
61
self.to_repository = to_repository
62
self.from_repository = from_repository
63
self.sink = to_repository._get_sink()
64
# must not mutate self._last_revision as its potentially a shared instance
65
self._last_revision = last_revision
66
self._fetch_spec = fetch_spec
67
self.find_ghosts = find_ghosts
68
self.from_repository.lock_read()
69
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
70
self.from_repository, self.from_repository._format,
71
self.to_repository, self.to_repository._format)
75
self.from_repository.unlock()
78
"""Primary worker function.
80
This initialises all the needed variables, and then fetches the
81
requested revisions, finally clearing the progress bar.
83
# Roughly this is what we're aiming for fetch to become:
85
# missing = self.sink.insert_stream(self.source.get_stream(search))
87
# missing = self.sink.insert_stream(self.source.get_items(missing))
90
def __init__(self, to_branch, from_branch, last_revision=None, pb=None):
91
if to_branch == from_branch:
92
raise Exception("can't fetch from a branch to itself")
93
self.to_branch = to_branch
94
self.to_weaves = to_branch.weave_store
95
self.to_control = to_branch.control_weaves
96
self.from_branch = from_branch
97
self.from_weaves = from_branch.weave_store
98
self.from_control = from_branch.control_weaves
99
self.failed_revisions = []
100
self.count_copied = 0
89
101
self.count_total = 0
102
self.count_weaves = 0
103
self.copied_file_ids = set()
90
104
self.file_ids_names = {}
91
pb = ui.ui_factory.nested_progress_bar()
92
pb.show_pct = pb.show_count = False
94
pb.update("Finding revisions", 0, 2)
95
search = self._revids_to_fetch()
98
pb.update("Fetching revisions", 1, 2)
99
self._fetch_everything_for_search(search)
103
def _fetch_everything_for_search(self, search):
104
"""Fetch all data for the given set of revisions."""
105
# The first phase is "file". We pass the progress bar for it directly
106
# into item_keys_introduced_by, which has more information about how
107
# that phase is progressing than we do. Progress updates for the other
108
# phases are taken care of in this function.
109
# XXX: there should be a clear owner of the progress reporting. Perhaps
110
# item_keys_introduced_by should have a richer API than it does at the
111
# moment, so that it can feed the progress information back to this
113
if (self.from_repository._format.rich_root_data and
114
not self.to_repository._format.rich_root_data):
115
raise errors.IncompatibleRepositories(
116
self.from_repository, self.to_repository,
117
"different rich-root support")
118
pb = ui.ui_factory.nested_progress_bar()
120
pb.update("Get stream source")
121
source = self.from_repository._get_source(
122
self.to_repository._format)
123
stream = source.get_stream(search)
124
from_format = self.from_repository._format
125
pb.update("Inserting stream")
126
resume_tokens, missing_keys = self.sink.insert_stream(
127
stream, from_format, [])
128
if self.to_repository._fallback_repositories:
130
self._parent_inventories(search.get_keys()))
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")
150
def _revids_to_fetch(self):
151
"""Determines the exact revisions needed from self.from_repository to
152
install self._last_revision in self.to_repository.
154
If no revisions need to be fetched, then this just returns None.
156
if self._fetch_spec is not None:
157
return self._fetch_spec
158
mutter('fetch up to rev {%s}', self._last_revision)
159
if self._last_revision is NULL_REVISION:
160
# explicit limit of no revisions needed
162
return self.to_repository.search_missing_revision_ids(
163
self.from_repository, self._last_revision,
164
find_ghosts=self.find_ghosts)
166
def _parent_inventories(self, revision_ids):
167
# Find all the parent revisions referenced by the stream, but
168
# not present in the stream, and make sure we send their
170
parent_maps = self.to_repository.get_parent_map(revision_ids)
172
map(parents.update, parent_maps.itervalues())
173
parents.discard(NULL_REVISION)
174
parents.difference_update(revision_ids)
175
missing_keys = set(('inventories', rev_id) for rev_id in parents)
179
class Inter1and2Helper(object):
180
"""Helper for operations that convert data from model 1 and 2
182
This is for use by fetchers and converters.
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.inventory.revision_id is None:
207
tree.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
revision_id = tree.inventory.root.revision
215
root_id = tree.get_root_id()
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.
247
# XXX: not covered by tests, should have a flag to always run
248
# this. -- mbp 20100129
249
graph = self.source_repo.get_known_graph_ancestry(revs)
250
new_roots_stream = _new_root_data_stream(
251
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
252
return [('texts', new_roots_stream)]
255
def _get_rich_root_heads_graph(source_repo, revision_ids):
256
"""Get a Graph object suitable for asking heads() for new rich roots."""
260
def _new_root_data_stream(
261
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
262
"""Generate a texts substream of synthesised root entries.
264
Used in fetches that do rich-root upgrades.
266
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
267
the root entries to create.
268
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
269
calculating the parents. If a parent rev_id is not found here then it
270
will be recalculated.
271
:param parent_map: a parent map for all the revisions in
273
:param graph: a graph to use instead of repo.get_graph().
275
for root_key in root_keys_to_create:
276
root_id, rev_id = root_key
277
parent_keys = _parent_keys_for_root_version(
278
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
279
yield versionedfile.FulltextContentFactory(
280
root_key, parent_keys, None, '')
283
def _parent_keys_for_root_version(
284
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
285
"""Get the parent keys for a given root id.
287
A helper function for _new_root_data_stream.
289
# Include direct parents of the revision, but only if they used the same
290
# root_id and are heads.
291
rev_parents = parent_map[rev_id]
293
for parent_id in rev_parents:
294
if parent_id == NULL_REVISION:
296
if parent_id not in rev_id_to_root_id_map:
297
# We probably didn't read this revision, go spend the extra effort
300
tree = repo.revision_tree(parent_id)
301
except errors.NoSuchRevision:
302
# Ghost, fill out rev_id_to_root_id in case we encounter this
304
# But set parent_root_id to None since we don't really know
305
parent_root_id = None
307
parent_root_id = tree.get_root_id()
308
rev_id_to_root_id_map[parent_id] = None
310
# rev_id_to_root_id_map[parent_id] = parent_root_id
311
# memory consumption maybe?
313
parent_root_id = rev_id_to_root_id_map[parent_id]
314
if root_id == parent_root_id:
315
# With stacking we _might_ want to refer to a non-local revision,
316
# but this code path only applies when we have the full content
317
# available, so ghosts really are ghosts, not just the edge of
319
parent_ids.append(parent_id)
321
# root_id may be in the parent anyway.
323
tree = repo.revision_tree(parent_id)
324
except errors.NoSuchRevision:
325
# ghost, can't refer to it.
329
parent_ids.append(tree.inventory[root_id].revision)
330
except errors.NoSuchId:
333
# Drop non-head parents
335
graph = repo.get_graph()
336
heads = graph.heads(parent_ids)
338
for parent_id in parent_ids:
339
if parent_id in heads and parent_id not in selected_ids:
340
selected_ids.append(parent_id)
341
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
106
self.pb = bzrlib.ui.ui_factory.progress_bar()
109
self.from_branch.lock_read()
111
self._fetch_revisions(last_revision)
113
self.from_branch.unlock()
116
def _fetch_revisions(self, last_revision):
117
self.last_revision = self._find_last_revision(last_revision)
118
mutter('fetch up to rev {%s}', self.last_revision)
119
if (self.last_revision is not None and
120
self.to_branch.has_revision(self.last_revision)):
123
revs_to_fetch = self._compare_ancestries()
125
raise InstallFailed([self.last_revision])
126
self._copy_revisions(revs_to_fetch)
127
self.new_ancestry = revs_to_fetch
129
def _find_last_revision(self, last_revision):
130
"""Find the limiting source revision.
132
Every ancestor of that revision will be merged across.
134
Returns the revision_id, or returns None if there's no history
135
in the source branch."""
138
self.pb.update('get source history')
139
from_history = self.from_branch.revision_history()
140
self.pb.update('get destination history')
142
return from_history[-1]
144
return None # no history in the source branch
147
def _compare_ancestries(self):
148
"""Get a list of revisions that must be copied.
150
That is, every revision that's in the ancestry of the source
151
branch and not in the destination branch."""
152
self.pb.update('get source ancestry')
153
self.from_ancestry = self.from_branch.get_ancestry(self.last_revision)
155
dest_last_rev = self.to_branch.last_revision()
156
self.pb.update('get destination ancestry')
158
dest_ancestry = self.to_branch.get_ancestry(dest_last_rev)
161
ss = set(dest_ancestry)
163
for rev_id in self.from_ancestry:
165
to_fetch.append(rev_id)
166
mutter('need to get revision {%s}', rev_id)
167
mutter('need to get %d revisions in total', len(to_fetch))
168
self.count_total = len(to_fetch)
171
def _copy_revisions(self, revs_to_fetch):
173
for rev_id in revs_to_fetch:
177
if self.to_branch.has_revision(rev_id):
179
self.pb.update('copy revision', i, self.count_total)
180
self._copy_one_revision(rev_id)
181
self.count_copied += 1
184
def _copy_one_revision(self, rev_id):
185
"""Copy revision and everything referenced by it."""
186
mutter('copying revision {%s}', rev_id)
187
rev_xml = self.from_branch.get_revision_xml(rev_id)
188
inv_xml = self.from_branch.get_inventory_xml(rev_id)
189
rev = serializer_v5.read_revision_from_string(rev_xml)
190
inv = serializer_v5.read_inventory_from_string(inv_xml)
191
assert rev.revision_id == rev_id
192
assert rev.inventory_sha1 == sha_string(inv_xml)
193
mutter(' commiter %s, %d parents',
196
self._copy_new_texts(rev_id, inv)
197
parents = rev.parent_ids
198
new_parents = copy(parents)
199
for parent in parents:
200
if not self.to_branch.has_revision(parent):
201
new_parents.pop(new_parents.index(parent))
202
self._copy_inventory(rev_id, inv_xml, new_parents)
203
self.to_branch.revision_store.add(StringIO(rev_xml), rev_id)
204
mutter('copied revision %s', rev_id)
206
def _copy_inventory(self, rev_id, inv_xml, parent_ids):
207
self.to_control.add_text('inventory', rev_id,
208
split_lines(inv_xml), parent_ids,
209
self.to_branch.get_transaction())
211
def _copy_new_texts(self, rev_id, inv):
212
"""Copy any new texts occuring in this revision."""
213
# TODO: Rather than writing out weaves every time, hold them
214
# in memory until everything's done? But this way is nicer
215
# if it's interrupted.
216
for path, ie in inv.iter_entries():
217
self._copy_one_weave(rev_id, ie.file_id, ie.revision)
219
def _copy_one_weave(self, rev_id, file_id, text_revision):
220
"""Copy one file weave, esuring the result contains text_revision."""
221
# check if the revision is already there
222
if file_id in self.file_ids_names.keys( ) and \
223
text_revision in self.file_ids_names[file_id]:
225
to_weave = self.to_weaves.get_weave_or_empty(file_id,
226
self.to_branch.get_transaction())
227
if not file_id in self.file_ids_names.keys( ):
228
self.file_ids_names[file_id] = to_weave.names( )
229
if text_revision in to_weave:
231
from_weave = self.from_weaves.get_weave(file_id,
232
self.from_branch.get_transaction())
233
if text_revision not in from_weave:
234
raise MissingText(self.from_branch, text_revision, file_id)
235
mutter('copy file {%s} modified in {%s}', file_id, rev_id)
237
if to_weave.numversions() > 0:
238
# destination has contents, must merge
240
to_weave.join(from_weave)
241
except errors.WeaveParentMismatch:
242
to_weave.reweave(from_weave)
244
# destination is empty, just replace it
245
to_weave = from_weave.copy( )
246
self.to_weaves.put_weave(file_id, to_weave,
247
self.to_branch.get_transaction())
248
self.count_weaves += 1
249
self.copied_file_ids.add(file_id)
250
self.file_ids_names[file_id] = to_weave.names()
251
mutter('copied file {%s}', file_id)
added
removed
bzrlib/fetch.py
