32
21
that has merged into it. As the first step of a merge, pull, or
33
22
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.
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
28
from bzrlib.lazy_import import lazy_import
29
lazy_import(globals(), """
43
from bzrlib.revision import NULL_REVISION
44
from bzrlib.trace import mutter
47
class RepoFetcher(object):
48
"""Pull revisions and texts from one repository to another.
50
This should not be used directly, it's essential a object to encapsulate
51
the logic in InterRepository.fetch().
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 = []
54
def __init__(self, to_repository, from_repository, last_revision=None,
55
find_ghosts=True, fetch_spec=None):
56
"""Create a repo fetcher.
58
:param last_revision: If set, try to limit to the data this 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.
64
self.to_repository = to_repository
65
self.from_repository = from_repository
66
self.sink = to_repository._get_sink()
67
# must not mutate self._last_revision as its potentially a shared instance
68
self._last_revision = last_revision
69
self._fetch_spec = fetch_spec
70
self.find_ghosts = find_ghosts
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)
78
self.from_repository.unlock()
81
"""Primary worker function.
83
This initialises all the needed variables, and then fetches the
84
requested revisions, finally clearing the progress bar.
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))
99
92
self.count_total = 0
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):
115
self.last_revision = self._find_last_revision(last_revision)
116
except NoSuchRevision, e:
117
mutter('failed getting last revision: %s', e)
118
raise InstallFailed([last_revision])
119
mutter('fetch up to rev {%s}', self.last_revision)
120
if (self.last_revision is not None and
121
self.to_branch.has_revision(self.last_revision)):
124
revs_to_fetch = self._compare_ancestries()
126
raise InstallFailed([self.last_revision])
127
self._copy_revisions(revs_to_fetch)
128
self.new_ancestry = revs_to_fetch
130
def _find_last_revision(self, last_revision):
131
"""Find the limiting source revision.
133
Every ancestor of that revision will be merged across.
135
Returns the revision_id, or returns None if there's no history
136
in the source branch."""
137
self.pb.update('get source history')
138
from_history = self.from_branch.revision_history()
139
self.pb.update('get destination history')
141
self.from_branch.get_revision(last_revision)
144
return from_history[-1]
146
return None # no history in the source branch
149
def _compare_ancestries(self):
150
"""Get a list of revisions that must be copied.
152
That is, every revision that's in the ancestry of the source
153
branch and not in the destination branch."""
154
self.pb.update('get source ancestry')
155
self.from_ancestry = self.from_branch.get_ancestry(self.last_revision)
157
dest_last_rev = self.to_branch.last_revision()
158
self.pb.update('get destination ancestry')
160
dest_ancestry = self.to_branch.get_ancestry(dest_last_rev)
163
ss = set(dest_ancestry)
165
for rev_id in self.from_ancestry:
167
to_fetch.append(rev_id)
168
mutter('need to get revision {%s}', rev_id)
169
mutter('need to get %d revisions in total', len(to_fetch))
170
self.count_total = len(to_fetch)
173
def _copy_revisions(self, revs_to_fetch):
175
for rev_id in revs_to_fetch:
179
if self.to_branch.has_revision(rev_id):
181
self.pb.update('fetch revision', i, self.count_total)
182
self._copy_one_revision(rev_id)
183
self.count_copied += 1
186
def _copy_one_revision(self, rev_id):
187
"""Copy revision and everything referenced by it."""
188
mutter('copying revision {%s}', rev_id)
189
rev_xml = self.from_branch.get_revision_xml(rev_id)
190
inv_xml = self.from_branch.get_inventory_xml(rev_id)
191
rev = serializer_v5.read_revision_from_string(rev_xml)
192
inv = serializer_v5.read_inventory_from_string(inv_xml)
193
assert rev.revision_id == rev_id
194
assert rev.inventory_sha1 == sha_string(inv_xml)
195
mutter(' commiter %s, %d parents',
198
self._copy_new_texts(rev_id, inv)
199
parents = rev.parent_ids
200
for parent in parents:
201
if not self.to_branch.has_revision(parent):
202
parents.pop(parents.index(parent))
203
self._copy_inventory(rev_id, inv_xml, parents)
204
self.to_branch.revision_store.add(StringIO(rev_xml), rev_id)
205
mutter('copied revision %s', rev_id)
208
def _copy_inventory(self, rev_id, inv_xml, parent_ids):
209
self.to_control.add_text('inventory', rev_id,
210
split_lines(inv_xml), parent_ids,
211
self.to_branch.get_transaction())
213
def _copy_new_texts(self, rev_id, inv):
214
"""Copy any new texts occuring in this revision."""
215
# TODO: Rather than writing out weaves every time, hold them
216
# in memory until everything's done? But this way is nicer
217
# if it's interrupted.
218
for path, ie in inv.iter_entries():
219
if ie.revision != rev_id:
221
mutter('%s {%s} is changed in this revision',
223
self._copy_one_weave(rev_id, ie.file_id)
226
def _copy_one_weave(self, rev_id, file_id):
227
"""Copy one file weave."""
228
mutter('copy file {%s} modified in {%s}', file_id, rev_id)
229
if file_id in self.copied_file_ids:
230
mutter('file {%s} already copied', file_id)
232
from_weave = self.from_weaves.get_weave(file_id,
233
self.from_branch.get_transaction())
234
to_weave = self.to_weaves.get_weave_or_empty(file_id,
235
self.to_branch.get_transaction())
237
to_weave.join(from_weave)
238
except errors.WeaveParentMismatch:
239
to_weave.reweave(from_weave)
240
self.to_weaves.put_weave(file_id, to_weave,
241
self.to_branch.get_transaction())
242
self.count_weaves += 1
243
self.copied_file_ids.add(file_id)
244
mutter('copied file {%s}', file_id)
93
self.file_ids_names = {}
94
pb = ui.ui_factory.nested_progress_bar()
95
pb.show_pct = pb.show_count = False
97
pb.update("Finding revisions", 0, 2)
98
search = self._revids_to_fetch()
101
pb.update("Fetching revisions", 1, 2)
102
self._fetch_everything_for_search(search)
106
def _fetch_everything_for_search(self, search):
107
"""Fetch all data for the given set of revisions."""
108
# The first phase is "file". We pass the progress bar for it directly
109
# into item_keys_introduced_by, which has more information about how
110
# that phase is progressing than we do. Progress updates for the other
111
# phases are taken care of in this function.
112
# XXX: there should be a clear owner of the progress reporting. Perhaps
113
# item_keys_introduced_by should have a richer API than it does at the
114
# moment, so that it can feed the progress information back to this
116
if (self.from_repository._format.rich_root_data and
117
not self.to_repository._format.rich_root_data):
118
raise errors.IncompatibleRepositories(
119
self.from_repository, self.to_repository,
120
"different rich-root support")
121
pb = ui.ui_factory.nested_progress_bar()
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, [])
131
if self.to_repository._fallback_repositories:
133
self._parent_inventories(search.get_keys()))
135
pb.update("Missing keys")
136
stream = source.get_stream_for_missing_keys(missing_keys)
137
pb.update("Inserting missing keys")
138
resume_tokens, missing_keys = self.sink.insert_stream(
139
stream, from_format, resume_tokens)
141
raise AssertionError(
142
"second push failed to complete a fetch %r." % (
145
raise AssertionError(
146
"second push failed to commit the fetch %r." % (
148
pb.update("Finishing stream")
153
def _revids_to_fetch(self):
154
"""Determines the exact revisions needed from self.from_repository to
155
install self._last_revision in self.to_repository.
157
If no revisions need to be fetched, then this just returns None.
159
if self._fetch_spec is not None:
160
return self._fetch_spec
161
mutter('fetch up to rev {%s}', self._last_revision)
162
if self._last_revision is NULL_REVISION:
163
# explicit limit of no revisions needed
165
return self.to_repository.search_missing_revision_ids(
166
self.from_repository, self._last_revision,
167
find_ghosts=self.find_ghosts)
169
def _parent_inventories(self, revision_ids):
170
# Find all the parent revisions referenced by the stream, but
171
# not present in the stream, and make sure we send their
173
parent_maps = self.to_repository.get_parent_map(revision_ids)
175
map(parents.update, parent_maps.itervalues())
176
parents.discard(NULL_REVISION)
177
parents.difference_update(revision_ids)
178
missing_keys = set(('inventories', rev_id) for rev_id in parents)
182
class Inter1and2Helper(object):
183
"""Helper for operations that convert data from model 1 and 2
185
This is for use by fetchers and converters.
188
def __init__(self, source):
191
:param source: The repository data comes from
195
def iter_rev_trees(self, revs):
196
"""Iterate through RevisionTrees efficiently.
198
Additionally, the inventory's revision_id is set if unset.
200
Trees are retrieved in batches of 100, and then yielded in the order
203
:param revs: A list of revision ids
205
# In case that revs is not a list.
208
for tree in self.source.revision_trees(revs[:100]):
209
if tree.inventory.revision_id is None:
210
tree.inventory.revision_id = tree.get_revision_id()
214
def _find_root_ids(self, revs, parent_map, graph):
216
for tree in self.iter_rev_trees(revs):
217
revision_id = tree.inventory.root.revision
218
root_id = tree.get_root_id()
219
revision_root[revision_id] = root_id
220
# Find out which parents we don't already know root ids for
222
for revision_parents in parent_map.itervalues():
223
parents.update(revision_parents)
224
parents.difference_update(revision_root.keys() + [NULL_REVISION])
225
# Limit to revisions present in the versionedfile
226
parents = graph.get_parent_map(parents).keys()
227
for tree in self.iter_rev_trees(parents):
228
root_id = tree.get_root_id()
229
revision_root[tree.get_revision_id()] = root_id
232
def generate_root_texts(self, revs):
233
"""Generate VersionedFiles for all root ids.
235
:param revs: the revisions to include
237
graph = self.source.get_graph()
238
parent_map = graph.get_parent_map(revs)
239
rev_order = tsort.topo_sort(parent_map)
240
rev_id_to_root_id = self._find_root_ids(revs, parent_map, graph)
241
root_id_order = [(rev_id_to_root_id[rev_id], rev_id) for rev_id in
243
# Guaranteed stable, this groups all the file id operations together
244
# retaining topological order within the revisions of a file id.
245
# File id splits and joins would invalidate this, but they don't exist
246
# yet, and are unlikely to in non-rich-root environments anyway.
247
root_id_order.sort(key=operator.itemgetter(0))
248
# Create a record stream containing the roots to create.
250
# XXX: not covered by tests, should have a flag to always run
251
# this. -- mbp 20100129
252
graph = _get_rich_root_heads_graph(self.source, revs)
253
new_roots_stream = _new_root_data_stream(
254
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
255
return [('texts', new_roots_stream)]
258
def _get_rich_root_heads_graph(source_repo, revision_ids):
259
"""Get a Graph object suitable for asking heads() for new rich roots."""
260
st = static_tuple.StaticTuple
261
revision_keys = [st(r_id).intern() for r_id in revision_ids]
262
known_graph = source_repo.revisions.get_known_graph_ancestry(
264
return _mod_graph.GraphThunkIdsToKeys(known_graph)
267
def _new_root_data_stream(
268
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
269
"""Generate a texts substream of synthesised root entries.
271
Used in fetches that do rich-root upgrades.
273
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
274
the root entries to create.
275
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
276
calculating the parents. If a parent rev_id is not found here then it
277
will be recalculated.
278
:param parent_map: a parent map for all the revisions in
280
:param graph: a graph to use instead of repo.get_graph().
282
for root_key in root_keys_to_create:
283
root_id, rev_id = root_key
284
parent_keys = _parent_keys_for_root_version(
285
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
286
yield versionedfile.FulltextContentFactory(
287
root_key, parent_keys, None, '')
290
def _parent_keys_for_root_version(
291
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
292
"""Get the parent keys for a given root id.
294
A helper function for _new_root_data_stream.
296
# Include direct parents of the revision, but only if they used the same
297
# root_id and are heads.
298
rev_parents = parent_map[rev_id]
300
for parent_id in rev_parents:
301
if parent_id == NULL_REVISION:
303
if parent_id not in rev_id_to_root_id_map:
304
# We probably didn't read this revision, go spend the extra effort
307
tree = repo.revision_tree(parent_id)
308
except errors.NoSuchRevision:
309
# Ghost, fill out rev_id_to_root_id in case we encounter this
311
# But set parent_root_id to None since we don't really know
312
parent_root_id = None
314
parent_root_id = tree.get_root_id()
315
rev_id_to_root_id_map[parent_id] = None
317
# rev_id_to_root_id_map[parent_id] = parent_root_id
318
# memory consumption maybe?
320
parent_root_id = rev_id_to_root_id_map[parent_id]
321
if root_id == parent_root_id:
322
# With stacking we _might_ want to refer to a non-local revision,
323
# but this code path only applies when we have the full content
324
# available, so ghosts really are ghosts, not just the edge of
326
parent_ids.append(parent_id)
328
# root_id may be in the parent anyway.
330
tree = repo.revision_tree(parent_id)
331
except errors.NoSuchRevision:
332
# ghost, can't refer to it.
336
parent_ids.append(tree.inventory[root_id].revision)
337
except errors.NoSuchId:
340
# Drop non-head parents
342
graph = repo.get_graph()
343
heads = graph.heads(parent_ids)
345
for parent_id in parent_ids:
346
if parent_id in heads and parent_id not in selected_ids:
347
selected_ids.append(parent_id)
348
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
added
removed
bzrlib/fetch.py
