20
17
"""Versioned text file storage api."""
19
from __future__ import absolute_import
22
21
from copy import copy
23
from cStringIO import StringIO
26
25
from zlib import adler32
28
from bzrlib.lazy_import import lazy_import
27
from ..lazy_import import lazy_import
29
28
lazy_import(globals(), """
35
33
graph as _mod_graph,
45
from bzrlib.graph import DictParentsProvider, Graph, StackedParentsProvider
46
from bzrlib.transport.memory import MemoryTransport
40
from breezy.bzr import (
48
from bzrlib.registry import Registry
49
from bzrlib.symbol_versioning import *
50
from bzrlib.textmerge import TextMerge
51
from bzrlib import bencode
46
from ..registry import Registry
47
from ..sixish import (
53
from ..textmerge import TextMerge
54
56
adapter_registry = Registry()
55
adapter_registry.register_lazy(('knit-delta-gz', 'fulltext'), 'bzrlib.knit',
57
adapter_registry.register_lazy(('knit-delta-gz', 'fulltext'), 'breezy.bzr.knit',
56
58
'DeltaPlainToFullText')
57
adapter_registry.register_lazy(('knit-ft-gz', 'fulltext'), 'bzrlib.knit',
59
adapter_registry.register_lazy(('knit-ft-gz', 'fulltext'), 'breezy.bzr.knit',
58
60
'FTPlainToFullText')
59
61
adapter_registry.register_lazy(('knit-annotated-delta-gz', 'knit-delta-gz'),
60
'bzrlib.knit', 'DeltaAnnotatedToUnannotated')
62
'breezy.bzr.knit', 'DeltaAnnotatedToUnannotated')
61
63
adapter_registry.register_lazy(('knit-annotated-delta-gz', 'fulltext'),
62
'bzrlib.knit', 'DeltaAnnotatedToFullText')
64
'breezy.bzr.knit', 'DeltaAnnotatedToFullText')
63
65
adapter_registry.register_lazy(('knit-annotated-ft-gz', 'knit-ft-gz'),
64
'bzrlib.knit', 'FTAnnotatedToUnannotated')
66
'breezy.bzr.knit', 'FTAnnotatedToUnannotated')
65
67
adapter_registry.register_lazy(('knit-annotated-ft-gz', 'fulltext'),
66
'bzrlib.knit', 'FTAnnotatedToFullText')
68
'breezy.bzr.knit', 'FTAnnotatedToFullText')
67
69
# adapter_registry.register_lazy(('knit-annotated-ft-gz', 'chunked'),
68
# 'bzrlib.knit', 'FTAnnotatedToChunked')
70
# 'breezy.bzr.knit', 'FTAnnotatedToChunked')
71
73
class ContentFactory(object):
213
class _MPDiffGenerator(object):
214
"""Pull out the functionality for generating mp_diffs."""
216
def __init__(self, vf, keys):
218
# This is the order the keys were requested in
219
self.ordered_keys = tuple(keys)
220
# keys + their parents, what we need to compute the diffs
221
self.needed_keys = ()
222
# Map from key: mp_diff
224
# Map from key: parents_needed (may have ghosts)
226
# Parents that aren't present
227
self.ghost_parents = ()
228
# Map from parent_key => number of children for this text
230
# Content chunks that are cached while we still need them
233
def _find_needed_keys(self):
234
"""Find the set of keys we need to request.
236
This includes all the original keys passed in, and the non-ghost
237
parents of those keys.
239
:return: (needed_keys, refcounts)
240
needed_keys is the set of all texts we need to extract
241
refcounts is a dict of {key: num_children} letting us know when we
242
no longer need to cache a given parent text
244
# All the keys and their parents
245
needed_keys = set(self.ordered_keys)
246
parent_map = self.vf.get_parent_map(needed_keys)
247
self.parent_map = parent_map
248
# TODO: Should we be using a different construct here? I think this
249
# uses difference_update internally, and we expect the result to
251
missing_keys = needed_keys.difference(parent_map)
253
raise errors.RevisionNotPresent(list(missing_keys)[0], self.vf)
254
# Parents that might be missing. They are allowed to be ghosts, but we
255
# should check for them
257
setdefault = refcounts.setdefault
259
for child_key, parent_keys in viewitems(parent_map):
261
# parent_keys may be None if a given VersionedFile claims to
262
# not support graph operations.
264
just_parents.update(parent_keys)
265
needed_keys.update(parent_keys)
266
for p in parent_keys:
267
refcounts[p] = setdefault(p, 0) + 1
268
just_parents.difference_update(parent_map)
269
# Remove any parents that are actually ghosts from the needed set
270
self.present_parents = set(self.vf.get_parent_map(just_parents))
271
self.ghost_parents = just_parents.difference(self.present_parents)
272
needed_keys.difference_update(self.ghost_parents)
273
self.needed_keys = needed_keys
274
self.refcounts = refcounts
275
return needed_keys, refcounts
277
def _compute_diff(self, key, parent_lines, lines):
278
"""Compute a single mp_diff, and store it in self._diffs"""
279
if len(parent_lines) > 0:
280
# XXX: _extract_blocks is not usefully defined anywhere...
281
# It was meant to extract the left-parent diff without
282
# having to recompute it for Knit content (pack-0.92,
283
# etc). That seems to have regressed somewhere
284
left_parent_blocks = self.vf._extract_blocks(key,
285
parent_lines[0], lines)
287
left_parent_blocks = None
288
diff = multiparent.MultiParent.from_lines(lines,
289
parent_lines, left_parent_blocks)
290
self.diffs[key] = diff
292
def _process_one_record(self, key, this_chunks):
294
if key in self.parent_map:
295
# This record should be ready to diff, since we requested
296
# content in 'topological' order
297
parent_keys = self.parent_map.pop(key)
298
# If a VersionedFile claims 'no-graph' support, then it may return
299
# None for any parent request, so we replace it with an empty tuple
300
if parent_keys is None:
303
for p in parent_keys:
304
# Alternatively we could check p not in self.needed_keys, but
305
# ghost_parents should be tiny versus huge
306
if p in self.ghost_parents:
308
refcount = self.refcounts[p]
309
if refcount == 1: # Last child reference
310
self.refcounts.pop(p)
311
parent_chunks = self.chunks.pop(p)
313
self.refcounts[p] = refcount - 1
314
parent_chunks = self.chunks[p]
315
p_lines = osutils.chunks_to_lines(parent_chunks)
316
# TODO: Should we cache the line form? We did the
317
# computation to get it, but storing it this way will
318
# be less memory efficient...
319
parent_lines.append(p_lines)
321
lines = osutils.chunks_to_lines(this_chunks)
322
# Since we needed the lines, we'll go ahead and cache them this way
324
self._compute_diff(key, parent_lines, lines)
326
# Is this content required for any more children?
327
if key in self.refcounts:
328
self.chunks[key] = this_chunks
330
def _extract_diffs(self):
331
needed_keys, refcounts = self._find_needed_keys()
332
for record in self.vf.get_record_stream(needed_keys,
333
'topological', True):
334
if record.storage_kind == 'absent':
335
raise errors.RevisionNotPresent(record.key, self.vf)
336
self._process_one_record(record.key,
337
record.get_bytes_as('chunked'))
339
def compute_diffs(self):
340
self._extract_diffs()
341
dpop = self.diffs.pop
342
return [dpop(k) for k in self.ordered_keys]
209
345
class VersionedFile(object):
210
346
"""Versioned text file storage.
401
541
for version, parent_ids, expected_sha1, mpdiff in records:
402
542
needed_parents.update(p for p in parent_ids
403
543
if not mpvf.has_version(p))
404
present_parents = set(self.get_parent_map(needed_parents).keys())
544
present_parents = set(self.get_parent_map(needed_parents))
405
545
for parent_id, lines in zip(present_parents,
406
self._get_lf_split_line_list(present_parents)):
546
self._get_lf_split_line_list(present_parents)):
407
547
mpvf.add_version(lines, parent_id, [])
408
for (version, parent_ids, expected_sha1, mpdiff), lines in\
409
zip(records, mpvf.get_line_list(versions)):
548
for (version, parent_ids, expected_sha1, mpdiff), lines in zip(
549
records, mpvf.get_line_list(versions)):
410
550
if len(parent_ids) == 1:
411
551
left_matching_blocks = list(mpdiff.get_matching_blocks(0,
412
552
mpvf.get_diff(parent_ids[0]).num_lines()))
744
882
This mapper is for use with a transport based backend.
747
_safe = "abcdefghijklmnopqrstuvwxyz0123456789-_@,."
885
_safe = bytearray(b"abcdefghijklmnopqrstuvwxyz0123456789-_@,.")
749
887
def _escape(self, prefix):
750
888
"""Turn a key element into a filesystem safe string.
752
This is similar to a plain urllib.quote, except
890
This is similar to a plain urlutils.quote, except
753
891
it uses specific safe characters, so that it doesn't
754
892
have to translate a lot of valid file ids.
756
894
# @ does not get escaped. This is because it is a valid
757
895
# filesystem character we use all the time, and it looks
758
896
# a lot better than seeing %40 all the time.
759
r = [((c in self._safe) and c or ('%%%02x' % ord(c)))
897
r = [(c in self._safe) and chr(c) or ('%%%02x' % c)
898
for c in bytearray(prefix)]
899
return ''.join(r).encode('ascii')
763
901
def _unescape(self, basename):
764
902
"""Escaped names are easily unescaped by urlutils."""
765
return urllib.unquote(basename)
903
return urlutils.unquote(basename)
768
906
def make_versioned_files_factory(versioned_file_factory, mapper):
790
928
The keyspace is expressed via simple tuples. Any instance of VersionedFiles
791
929
may have a different length key-size, but that size will be constant for
792
all texts added to or retrieved from it. For instance, bzrlib uses
930
all texts added to or retrieved from it. For instance, breezy uses
793
931
instances with a key-size of 2 for storing user files in a repository, with
794
932
the first element the fileid, and the second the version of that file.
796
934
The use of tuples allows a single code base to support several different
797
935
uses with only the mapping logic changing from instance to instance.
937
:ivar _immediate_fallback_vfs: For subclasses that support stacking,
938
this is a list of other VersionedFiles immediately underneath this
939
one. They may in turn each have further fallbacks.
800
942
def add_lines(self, key, parents, lines, parent_texts=None,
837
979
raise NotImplementedError(self.add_lines)
839
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
840
"""Add a text to the store.
842
This is a private function for use by CommitBuilder.
844
:param key: The key tuple of the text to add. If the last element is
845
None, a CHK string will be generated during the addition.
846
:param parents: The parents key tuples of the text to add.
847
:param text: A string containing the text to be committed.
848
:param nostore_sha: Raise ExistingContent and do not add the lines to
849
the versioned file if the digest of the lines matches this.
850
:param random_id: If True a random id has been selected rather than
851
an id determined by some deterministic process such as a converter
852
from a foreign VCS. When True the backend may choose not to check
853
for uniqueness of the resulting key within the versioned file, so
854
this should only be done when the result is expected to be unique
856
:param check_content: If True, the lines supplied are verified to be
857
bytestrings that are correctly formed lines.
858
:return: The text sha1, the number of bytes in the text, and an opaque
859
representation of the inserted version which can be provided
860
back to future _add_text calls in the parent_texts dictionary.
862
# The default implementation just thunks over to .add_lines(),
863
# inefficient, but it works.
864
return self.add_lines(key, parents, osutils.split_lines(text),
865
nostore_sha=nostore_sha,
869
981
def add_mpdiffs(self, records):
870
982
"""Add mpdiffs to this VersionedFile.
1048
1160
def make_mpdiffs(self, keys):
1049
1161
"""Create multiparent diffs for specified keys."""
1050
keys_order = tuple(keys)
1051
keys = frozenset(keys)
1052
knit_keys = set(keys)
1053
parent_map = self.get_parent_map(keys)
1054
for parent_keys in parent_map.itervalues():
1056
knit_keys.update(parent_keys)
1057
missing_keys = keys - set(parent_map)
1059
raise errors.RevisionNotPresent(list(missing_keys)[0], self)
1060
# We need to filter out ghosts, because we can't diff against them.
1061
maybe_ghosts = knit_keys - keys
1062
ghosts = maybe_ghosts - set(self.get_parent_map(maybe_ghosts))
1063
knit_keys.difference_update(ghosts)
1065
chunks_to_lines = osutils.chunks_to_lines
1066
for record in self.get_record_stream(knit_keys, 'topological', True):
1067
lines[record.key] = chunks_to_lines(record.get_bytes_as('chunked'))
1068
# line_block_dict = {}
1069
# for parent, blocks in record.extract_line_blocks():
1070
# line_blocks[parent] = blocks
1071
# line_blocks[record.key] = line_block_dict
1073
for key in keys_order:
1075
parents = parent_map[key] or []
1076
# Note that filtering knit_keys can lead to a parent difference
1077
# between the creation and the application of the mpdiff.
1078
parent_lines = [lines[p] for p in parents if p in knit_keys]
1079
if len(parent_lines) > 0:
1080
left_parent_blocks = self._extract_blocks(key, parent_lines[0],
1083
left_parent_blocks = None
1084
diffs.append(multiparent.MultiParent.from_lines(target,
1085
parent_lines, left_parent_blocks))
1162
generator = _MPDiffGenerator(self, keys)
1163
return generator.compute_diffs()
1165
def get_annotator(self):
1166
return annotate.Annotator(self)
1088
1168
missing_keys = index._missing_keys_from_parent_map
1090
1170
def _extract_blocks(self, version_id, source, target):
1173
def _transitive_fallbacks(self):
1174
"""Return the whole stack of fallback versionedfiles.
1176
This VersionedFiles may have a list of fallbacks, but it doesn't
1177
necessarily know about the whole stack going down, and it can't know
1178
at open time because they may change after the objects are opened.
1181
for a_vfs in self._immediate_fallback_vfs:
1182
all_fallbacks.append(a_vfs)
1183
all_fallbacks.extend(a_vfs._transitive_fallbacks())
1184
return all_fallbacks
1094
1187
class ThunkedVersionedFiles(VersionedFiles):
1095
1188
"""Storage for many versioned files thunked onto a 'VersionedFile' class.
1404
class VersionedFilesWithFallbacks(VersionedFiles):
1406
def without_fallbacks(self):
1407
"""Return a clone of this object without any fallbacks configured."""
1408
raise NotImplementedError(self.without_fallbacks)
1410
def add_fallback_versioned_files(self, a_versioned_files):
1411
"""Add a source of texts for texts not present in this knit.
1413
:param a_versioned_files: A VersionedFiles object.
1415
raise NotImplementedError(self.add_fallback_versioned_files)
1417
def get_known_graph_ancestry(self, keys):
1418
"""Get a KnownGraph instance with the ancestry of keys."""
1419
parent_map, missing_keys = self._index.find_ancestry(keys)
1420
for fallback in self._transitive_fallbacks():
1421
if not missing_keys:
1423
(f_parent_map, f_missing_keys) = fallback._index.find_ancestry(
1425
parent_map.update(f_parent_map)
1426
missing_keys = f_missing_keys
1427
kg = _mod_graph.KnownGraph(parent_map)
1314
1431
class _PlanMergeVersionedFile(VersionedFiles):
1315
1432
"""A VersionedFile for uncommitted and committed texts.
1745
class NoDupeAddLinesDecorator(object):
1746
"""Decorator for a VersionedFiles that skips doing an add_lines if the key
1750
def __init__(self, store):
1753
def add_lines(self, key, parents, lines, parent_texts=None,
1754
left_matching_blocks=None, nostore_sha=None, random_id=False,
1755
check_content=True):
1756
"""See VersionedFiles.add_lines.
1758
This implementation may return None as the third element of the return
1759
value when the original store wouldn't.
1762
raise NotImplementedError(
1763
"NoDupeAddLinesDecorator.add_lines does not implement the "
1764
"nostore_sha behaviour.")
1766
sha1 = osutils.sha_strings(lines)
1767
key = (b"sha1:" + sha1,)
1770
if key in self._store.get_parent_map([key]):
1771
# This key has already been inserted, so don't do it again.
1773
sha1 = osutils.sha_strings(lines)
1774
return sha1, sum(map(len, lines)), None
1775
return self._store.add_lines(key, parents, lines,
1776
parent_texts=parent_texts,
1777
left_matching_blocks=left_matching_blocks,
1778
nostore_sha=nostore_sha, random_id=random_id,
1779
check_content=check_content)
1781
def __getattr__(self, name):
1782
return getattr(self._store, name)
1627
1785
def network_bytes_to_kind_and_offset(network_bytes):
1628
1786
"""Strip of a record kind from the front of network_bytes.
1630
1788
:param network_bytes: The bytes of a record.
1631
1789
:return: A tuple (storage_kind, offset_of_remaining_bytes)
1633
line_end = network_bytes.find('\n')
1634
storage_kind = network_bytes[:line_end]
1791
line_end = network_bytes.find(b'\n')
1792
storage_kind = network_bytes[:line_end].decode('ascii')
1635
1793
return storage_kind, line_end + 1
1720
1878
for prefix in sorted(per_prefix_map):
1721
1879
present_keys.extend(reversed(tsort.topo_sort(per_prefix_map[prefix])))
1722
1880
return present_keys
1883
class _KeyRefs(object):
1885
def __init__(self, track_new_keys=False):
1886
# dict mapping 'key' to 'set of keys referring to that key'
1889
# set remembering all new keys
1890
self.new_keys = set()
1892
self.new_keys = None
1898
self.new_keys.clear()
1900
def add_references(self, key, refs):
1901
# Record the new references
1902
for referenced in refs:
1904
needed_by = self.refs[referenced]
1906
needed_by = self.refs[referenced] = set()
1908
# Discard references satisfied by the new key
1911
def get_new_keys(self):
1912
return self.new_keys
1914
def get_unsatisfied_refs(self):
1915
return self.refs.keys()
1917
def _satisfy_refs_for_key(self, key):
1921
# No keys depended on this key. That's ok.
1924
def add_key(self, key):
1925
# satisfy refs for key, and remember that we've seen this key.
1926
self._satisfy_refs_for_key(key)
1927
if self.new_keys is not None:
1928
self.new_keys.add(key)
1930
def satisfy_refs_for_keys(self, keys):
1932
self._satisfy_refs_for_key(key)
1934
def get_referrers(self):
1935
return set(itertools.chain.from_iterable(viewvalues(self.refs)))