21
31
that has merged into it. As the first step of a merge, pull, or
22
32
branch operation we copy history from the source into the destination
35
The copying is done in a slightly complicated order. We don't want to
36
add a revision to the store until everything it refers to is also
37
stored, so that if a revision is present we can totally recreate it.
38
However, we can't know what files are included in a revision until we
39
read its inventory. Therefore, we first pull the XML and hold it in
40
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().
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))
43
# TODO: Avoid repeatedly opening weaves so many times.
45
# XXX: This doesn't handle ghost (not present in branch) revisions at
46
# all yet. I'm not sure they really should be supported.
48
# TODO: This doesn't handle revisions which may be present but not
49
# merged into the last revision.
51
# - get a list of revisions that need to be pulled in
52
# - for each one, pull in that revision file
53
# and get the inventory, and store the inventory with right
55
# - and get the ancestry, and store that with right parents too
56
# - and keep a note of all file ids and version seen
57
# - then go through all files; for each one get the weave,
58
# and add in all file versions
62
def greedy_fetch(to_branch, from_branch, revision, pb):
63
f = Fetcher(to_branch, from_branch, revision, pb)
64
return f.count_copied, f.failed_revisions
67
class Fetcher(object):
68
"""Pull history from one branch to another.
71
If set, pull only up to this revision_id.
73
def __init__(self, to_branch, from_branch, revision_limit=None, pb=None):
74
self.to_branch = to_branch
75
self.from_branch = from_branch
76
self.failed_revisions = []
89
78
self.count_total = 0
90
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
# This is a class variable so that the test suite can override it.
186
known_graph_threshold = 100
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.
249
if len(revs) > self.known_graph_threshold:
250
graph = self.source.get_known_graph_ancestry(revs)
251
new_roots_stream = _new_root_data_stream(
252
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
253
return [('texts', new_roots_stream)]
256
def _new_root_data_stream(
257
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
258
"""Generate a texts substream of synthesised root entries.
260
Used in fetches that do rich-root upgrades.
262
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
263
the root entries to create.
264
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
265
calculating the parents. If a parent rev_id is not found here then it
266
will be recalculated.
267
:param parent_map: a parent map for all the revisions in
269
:param graph: a graph to use instead of repo.get_graph().
271
for root_key in root_keys_to_create:
272
root_id, rev_id = root_key
273
parent_keys = _parent_keys_for_root_version(
274
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
275
yield versionedfile.FulltextContentFactory(
276
root_key, parent_keys, None, '')
279
def _parent_keys_for_root_version(
280
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
281
"""Get the parent keys for a given root id.
283
A helper function for _new_root_data_stream.
285
# Include direct parents of the revision, but only if they used the same
286
# root_id and are heads.
287
rev_parents = parent_map[rev_id]
289
for parent_id in rev_parents:
290
if parent_id == NULL_REVISION:
292
if parent_id not in rev_id_to_root_id_map:
293
# We probably didn't read this revision, go spend the extra effort
296
tree = repo.revision_tree(parent_id)
297
except errors.NoSuchRevision:
298
# Ghost, fill out rev_id_to_root_id in case we encounter this
300
# But set parent_root_id to None since we don't really know
301
parent_root_id = None
303
parent_root_id = tree.get_root_id()
304
rev_id_to_root_id_map[parent_id] = None
306
# rev_id_to_root_id_map[parent_id] = parent_root_id
307
# memory consumption maybe?
309
parent_root_id = rev_id_to_root_id_map[parent_id]
310
if root_id == parent_root_id:
311
# With stacking we _might_ want to refer to a non-local revision,
312
# but this code path only applies when we have the full content
313
# available, so ghosts really are ghosts, not just the edge of
315
parent_ids.append(parent_id)
317
# root_id may be in the parent anyway.
319
tree = repo.revision_tree(parent_id)
320
except errors.NoSuchRevision:
321
# ghost, can't refer to it.
325
parent_ids.append(tree.inventory[root_id].revision)
326
except errors.NoSuchId:
329
# Drop non-head parents
331
graph = repo.get_graph()
332
heads = graph.heads(parent_ids)
334
for parent_id in parent_ids:
335
if parent_id in heads and parent_id not in selected_ids:
336
selected_ids.append(parent_id)
337
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
80
self.pb = bzrlib.ui.ui_factory.progress_bar()
83
self.revision_limit = self._find_revision_limit(revision_limit)
84
revs_to_fetch = self._compare_ancestries()
85
self._copy_revisions(revs_to_fetch)
89
def _find_revision_limit(self, revision_limit):
90
"""Find the limiting source revision.
92
Every ancestor of that revision will be merged across.
94
Returns the revision_id, or returns None if there's no history
95
in the source branch."""
96
self.pb.update('get source history')
97
from_history = self.from_branch.revision_history()
98
self.pb.update('get destination history')
100
if revision_limit not in from_history:
101
raise NoSuchRevision(self.from_branch, revision_limit)
103
return revision_limit
105
return from_history[-1]
107
return None # no history in the source branch
110
def _compare_ancestries(self):
111
"""Get a list of revisions that must be copied.
113
That is, every revision that's in the ancestry of the source
114
branch and not in the destination branch."""
115
self.pb.update('get source ancestry')
116
self.from_ancestry = self.from_branch.get_ancestry(self.revision_limit)
118
dest_last_rev = self.to_branch.last_revision()
119
self.pb.update('get destination ancestry')
121
dest_ancestry = self.to_branch.get_ancestry(dest_last_rev)
124
ss = set(dest_ancestry)
126
for rev_id in self.from_ancestry:
128
to_fetch.append(rev_id)
129
mutter('need to get revision {%s}', rev_id)
130
mutter('need to get %d revisions in total', len(to_fetch))
131
self.count_total = len(to_fetch)
136
def _copy_revisions(self, revs_to_fetch):
138
for rev_id in revs_to_fetch:
139
self.pb.update('fetch revision', i, self.count_total)
140
self._copy_one_revision(rev_id)
144
def _copy_one_revision(self, rev_id):
145
"""Copy revision and everything referenced by it."""
146
mutter('copying revision {%s}', rev_id)
147
rev_xml = self.from_branch.get_revision_xml(rev_id)
148
inv_xml = self.from_branch.get_inventory_xml(rev_id)
149
rev = serializer_v5.read_revision_from_string(rev_xml)
150
inv = serializer_v5.read_inventory_from_string(inv_xml)
151
assert rev.revision_id == rev_id
152
assert rev.inventory_sha1 == sha_string(inv_xml)
153
mutter(' commiter %s, %d parents',
156
self._copy_new_texts(rev_id, inv)
157
self.to_branch.weave_store.add_text(INVENTORY_FILEID, rev_id,
158
split_lines(inv_xml), rev.parents)
159
self.to_branch.revision_store.add(StringIO(rev_xml), rev_id)
162
def _copy_new_texts(self, rev_id, inv):
163
"""Copy any new texts occuring in this revision."""
164
# TODO: Rather than writing out weaves every time, hold them
165
# in memory until everything's done? But this way is nicer
166
# if it's interrupted.
167
for path, ie in inv.iter_entries():
168
if ie.kind != 'file':
170
if ie.text_version != rev_id:
172
mutter('%s {%s} is changed in this revision',
174
self._copy_one_text(rev_id, ie.file_id)
177
def _copy_one_text(self, rev_id, file_id):
178
"""Copy one file text."""
179
from_weave = self.from_branch.weave_store.get_weave(file_id)
180
from_idx = from_weave.lookup(rev_id)
181
from_parents = map(from_weave.idx_to_name, from_weave.parents(from_idx))
182
text_lines = from_weave.get(from_idx)
183
to_weave = self.to_branch.weave_store.get_weave_or_empty(file_id)
184
to_parents = map(to_weave.lookup, from_parents)
185
# it's ok to add even if the text is already there
186
to_weave.add(rev_id, to_parents, text_lines)
187
self.to_branch.weave_store.put_weave(file_id, to_weave)
added
removed
bzrlib/fetch.py
