120
119
:param num_bytes: Ensure that we have extracted at least num_bytes of
121
120
content. If None, consume everything
123
if self._content_length is None:
124
raise AssertionError('self._content_length should never be None')
122
# TODO: If we re-use the same content block at different times during
123
# get_record_stream(), it is possible that the first pass will
124
# get inserted, triggering an extract/_ensure_content() which
125
# will get rid of _z_content. And then the next use of the block
126
# will try to access _z_content (to send it over the wire), and
127
# fail because it is already extracted. Consider never releasing
128
# _z_content because of this.
125
129
if num_bytes is None:
126
130
num_bytes = self._content_length
127
131
elif (self._content_length is not None
144
148
self._content = pylzma.decompress(self._z_content)
145
149
elif self._compressor_name == 'zlib':
146
150
# Start a zlib decompressor
147
if num_bytes * 4 > self._content_length * 3:
148
# If we are requesting more that 3/4ths of the content,
149
# just extract the whole thing in a single pass
150
num_bytes = self._content_length
151
if num_bytes is None:
151
152
self._content = zlib.decompress(self._z_content)
153
154
self._z_content_decompressor = zlib.decompressobj()
164
163
# 'unconsumed_tail'
166
165
# Do we have enough bytes already?
167
if len(self._content) >= num_bytes:
166
if num_bytes is not None and len(self._content) >= num_bytes:
168
if num_bytes is None and self._z_content_decompressor is None:
169
# We must have already decompressed everything
169
171
# If we got this far, and don't have a decompressor, something is wrong
170
172
if self._z_content_decompressor is None:
171
173
raise AssertionError(
172
174
'No decompressor to decompress %d bytes' % num_bytes)
173
175
remaining_decomp = self._z_content_decompressor.unconsumed_tail
174
if not remaining_decomp:
175
raise AssertionError('Nothing left to decompress')
176
needed_bytes = num_bytes - len(self._content)
177
# We always set max_size to 32kB over the minimum needed, so that
178
# zlib will give us as much as we really want.
179
# TODO: If this isn't good enough, we could make a loop here,
180
# that keeps expanding the request until we get enough
181
self._content += self._z_content_decompressor.decompress(
182
remaining_decomp, needed_bytes + _ZLIB_DECOMP_WINDOW)
183
if len(self._content) < num_bytes:
184
raise AssertionError('%d bytes wanted, only %d available'
185
% (num_bytes, len(self._content)))
186
if not self._z_content_decompressor.unconsumed_tail:
187
# The stream is finished
188
self._z_content_decompressor = None
176
if num_bytes is None:
178
# We don't know how much is left, but we'll decompress it all
179
self._content += self._z_content_decompressor.decompress(
181
# Note: There's what I consider a bug in zlib.decompressobj
182
# If you pass back in the entire unconsumed_tail, only
183
# this time you don't pass a max-size, it doesn't
184
# change the unconsumed_tail back to None/''.
185
# However, we know we are done with the whole stream
186
self._z_content_decompressor = None
187
# XXX: Why is this the only place in this routine we set this?
188
self._content_length = len(self._content)
190
if not remaining_decomp:
191
raise AssertionError('Nothing left to decompress')
192
needed_bytes = num_bytes - len(self._content)
193
# We always set max_size to 32kB over the minimum needed, so that
194
# zlib will give us as much as we really want.
195
# TODO: If this isn't good enough, we could make a loop here,
196
# that keeps expanding the request until we get enough
197
self._content += self._z_content_decompressor.decompress(
198
remaining_decomp, needed_bytes + _ZLIB_DECOMP_WINDOW)
199
if len(self._content) < num_bytes:
200
raise AssertionError('%d bytes wanted, only %d available'
201
% (num_bytes, len(self._content)))
202
if not self._z_content_decompressor.unconsumed_tail:
203
# The stream is finished
204
self._z_content_decompressor = None
190
206
def _parse_bytes(self, bytes, pos):
191
207
"""Read the various lengths from the header.
1267
1283
return self.get_record_stream(keys, 'unordered', True)
1269
def clear_cache(self):
1270
"""See VersionedFiles.clear_cache()"""
1271
self._group_cache.clear()
1272
self._index._graph_index.clear_cache()
1273
self._index._int_cache.clear()
1275
1285
def _check_add(self, key, lines, random_id, check_content):
1276
1286
"""check that version_id and lines are safe to add."""
1277
1287
version_id = key[-1]
1747
1757
key = record.key
1748
1758
self._unadded_refs[key] = record.parents
1749
1759
yield found_sha1
1750
as_st = static_tuple.StaticTuple.from_sequence
1751
if record.parents is not None:
1752
parents = as_st([as_st(p) for p in record.parents])
1755
refs = static_tuple.StaticTuple(parents)
1756
keys_to_add.append((key, '%d %d' % (start_point, end_point), refs))
1760
keys_to_add.append((key, '%d %d' % (start_point, end_point),
1757
1762
if len(keys_to_add):
1759
1764
self._compressor = None
1812
class _GCBuildDetails(object):
1813
"""A blob of data about the build details.
1815
This stores the minimal data, which then allows compatibility with the old
1816
api, without taking as much memory.
1819
__slots__ = ('_index', '_group_start', '_group_end', '_basis_end',
1820
'_delta_end', '_parents')
1823
compression_parent = None
1825
def __init__(self, parents, position_info):
1826
self._parents = parents
1827
(self._index, self._group_start, self._group_end, self._basis_end,
1828
self._delta_end) = position_info
1831
return '%s(%s, %s)' % (self.__class__.__name__,
1832
self.index_memo, self._parents)
1835
def index_memo(self):
1836
return (self._index, self._group_start, self._group_end,
1837
self._basis_end, self._delta_end)
1840
def record_details(self):
1841
return static_tuple.StaticTuple(self.method, None)
1843
def __getitem__(self, offset):
1844
"""Compatibility thunk to act like a tuple."""
1846
return self.index_memo
1848
return self.compression_parent # Always None
1850
return self._parents
1852
return self.record_details
1854
raise IndexError('offset out of range')
1860
1817
class _GCGraphIndex(object):
1861
1818
"""Mapper from GroupCompressVersionedFiles needs into GraphIndex storage."""
1931
1885
if not random_id:
1932
1886
present_nodes = self._get_entries(keys)
1933
1887
for (index, key, value, node_refs) in present_nodes:
1934
# Sometimes these are passed as a list rather than a tuple
1935
node_refs = static_tuple.as_tuples(node_refs)
1936
passed = static_tuple.as_tuples(keys[key])
1937
if node_refs != passed[1]:
1938
details = '%s %s %s' % (key, (value, node_refs), passed)
1888
if node_refs != keys[key][1]:
1889
details = '%s %s %s' % (key, (value, node_refs), keys[key])
1939
1890
if self._inconsistency_fatal:
1940
1891
raise errors.KnitCorrupt(self, "inconsistent details"
1941
1892
" in add_records: %s" %
2073
2025
"""Convert an index value to position details."""
2074
2026
bits = node[2].split(' ')
2075
2027
# It would be nice not to read the entire gzip.
2076
# start and stop are put into _int_cache because they are very common.
2077
# They define the 'group' that an entry is in, and many groups can have
2078
# thousands of objects.
2079
# Branching Launchpad, for example, saves ~600k integers, at 12 bytes
2080
# each, or about 7MB. Note that it might be even more when you consider
2081
# how PyInt is allocated in separate slabs. And you can't return a slab
2082
# to the OS if even 1 int on it is in use. Note though that Python uses
2083
# a LIFO when re-using PyInt slots, which might cause more
2085
2028
start = int(bits[0])
2086
start = self._int_cache.setdefault(start, start)
2087
2029
stop = int(bits[1])
2088
stop = self._int_cache.setdefault(stop, stop)
2089
2030
basis_end = int(bits[2])
2090
2031
delta_end = int(bits[3])
2091
# We can't use StaticTuple here, because node[0] is a BTreeGraphIndex
2093
return (node[0], start, stop, basis_end, delta_end)
2032
return node[0], start, stop, basis_end, delta_end
2095
2034
def scan_unvalidated_index(self, graph_index):
2096
2035
"""Inform this _GCGraphIndex that there is an unvalidated index.