1
# Copyright (C) 2008, 2009 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
"""Core compression logic for compressing streams of related files."""
19
from itertools import izip
20
from cStringIO import StringIO
40
from bzrlib.graph import Graph
41
from bzrlib.btree_index import BTreeBuilder
42
from bzrlib.lru_cache import LRUSizeCache
43
from bzrlib.tsort import topo_sort
44
from bzrlib.versionedfile import (
47
ChunkedContentFactory,
48
FulltextContentFactory,
52
_USE_LZMA = False and (pylzma is not None)
54
# osutils.sha_string('')
55
_null_sha1 = 'da39a3ee5e6b4b0d3255bfef95601890afd80709'
58
def sort_gc_optimal(parent_map):
59
"""Sort and group the keys in parent_map into groupcompress order.
61
groupcompress is defined (currently) as reverse-topological order, grouped
64
:return: A sorted-list of keys
66
# groupcompress ordering is approximately reverse topological,
67
# properly grouped by file-id.
69
for item in parent_map.iteritems():
71
if isinstance(key, str) or len(key) == 1:
76
per_prefix_map[prefix].append(item)
78
per_prefix_map[prefix] = [item]
81
for prefix in sorted(per_prefix_map):
82
present_keys.extend(reversed(topo_sort(per_prefix_map[prefix])))
86
# The max zlib window size is 32kB, so if we set 'max_size' output of the
87
# decompressor to the requested bytes + 32kB, then we should guarantee
88
# num_bytes coming out.
89
_ZLIB_DECOMP_WINDOW = 32*1024
91
class GroupCompressBlock(object):
92
"""An object which maintains the internal structure of the compressed data.
94
This tracks the meta info (start of text, length, type, etc.)
97
# Group Compress Block v1 Zlib
98
GCB_HEADER = 'gcb1z\n'
99
# Group Compress Block v1 Lzma
100
GCB_LZ_HEADER = 'gcb1l\n'
101
GCB_KNOWN_HEADERS = (GCB_HEADER, GCB_LZ_HEADER)
104
# map by key? or just order in file?
105
self._compressor_name = None
106
self._z_content = None
107
self._z_content_decompressor = None
108
self._z_content_length = None
109
self._content_length = None
113
# This is the maximum number of bytes this object will reference if
114
# everything is decompressed. However, if we decompress less than
115
# everything... (this would cause some problems for LRUSizeCache)
116
return self._content_length + self._z_content_length
118
def _ensure_content(self, num_bytes=None):
119
"""Make sure that content has been expanded enough.
121
:param num_bytes: Ensure that we have extracted at least num_bytes of
122
content. If None, consume everything
124
# TODO: If we re-use the same content block at different times during
125
# get_record_stream(), it is possible that the first pass will
126
# get inserted, triggering an extract/_ensure_content() which
127
# will get rid of _z_content. And then the next use of the block
128
# will try to access _z_content (to send it over the wire), and
129
# fail because it is already extracted. Consider never releasing
130
# _z_content because of this.
131
if num_bytes is None:
132
num_bytes = self._content_length
133
elif (self._content_length is not None
134
and num_bytes > self._content_length):
135
raise AssertionError(
136
'requested num_bytes (%d) > content length (%d)'
137
% (num_bytes, self._content_length))
138
# Expand the content if required
139
if self._content is None:
140
if self._z_content is None:
141
raise AssertionError('No content to decompress')
142
if self._z_content == '':
144
elif self._compressor_name == 'lzma':
145
# We don't do partial lzma decomp yet
146
self._content = pylzma.decompress(self._z_content)
147
elif self._compressor_name == 'zlib':
148
# Start a zlib decompressor
149
if num_bytes is None:
150
self._content = zlib.decompress(self._z_content)
152
self._z_content_decompressor = zlib.decompressobj()
153
# Seed the decompressor with the uncompressed bytes, so
154
# that the rest of the code is simplified
155
self._content = self._z_content_decompressor.decompress(
156
self._z_content, num_bytes + _ZLIB_DECOMP_WINDOW)
158
raise AssertionError('Unknown compressor: %r'
159
% self._compressor_name)
160
# Any bytes remaining to be decompressed will be in the decompressors
163
# Do we have enough bytes already?
164
if num_bytes is not None and len(self._content) >= num_bytes:
166
if num_bytes is None and self._z_content_decompressor is None:
167
# We must have already decompressed everything
169
# If we got this far, and don't have a decompressor, something is wrong
170
if self._z_content_decompressor is None:
171
raise AssertionError(
172
'No decompressor to decompress %d bytes' % num_bytes)
173
remaining_decomp = self._z_content_decompressor.unconsumed_tail
174
if num_bytes is None:
176
# We don't know how much is left, but we'll decompress it all
177
self._content += self._z_content_decompressor.decompress(
179
# Note: There's what I consider a bug in zlib.decompressobj
180
# If you pass back in the entire unconsumed_tail, only
181
# this time you don't pass a max-size, it doesn't
182
# change the unconsumed_tail back to None/''.
183
# However, we know we are done with the whole stream
184
self._z_content_decompressor = None
185
# XXX: Why is this the only place in this routine we set this?
186
self._content_length = len(self._content)
188
if not remaining_decomp:
189
raise AssertionError('Nothing left to decompress')
190
needed_bytes = num_bytes - len(self._content)
191
# We always set max_size to 32kB over the minimum needed, so that
192
# zlib will give us as much as we really want.
193
# TODO: If this isn't good enough, we could make a loop here,
194
# that keeps expanding the request until we get enough
195
self._content += self._z_content_decompressor.decompress(
196
remaining_decomp, needed_bytes + _ZLIB_DECOMP_WINDOW)
197
if len(self._content) < num_bytes:
198
raise AssertionError('%d bytes wanted, only %d available'
199
% (num_bytes, len(self._content)))
200
if not self._z_content_decompressor.unconsumed_tail:
201
# The stream is finished
202
self._z_content_decompressor = None
204
def _parse_bytes(self, bytes, pos):
205
"""Read the various lengths from the header.
207
This also populates the various 'compressed' buffers.
209
:return: The position in bytes just after the last newline
211
# At present, we have 2 integers for the compressed and uncompressed
212
# content. In base10 (ascii) 14 bytes can represent > 1TB, so to avoid
213
# checking too far, cap the search to 14 bytes.
214
pos2 = bytes.index('\n', pos, pos + 14)
215
self._z_content_length = int(bytes[pos:pos2])
217
pos2 = bytes.index('\n', pos, pos + 14)
218
self._content_length = int(bytes[pos:pos2])
220
if len(bytes) != (pos + self._z_content_length):
221
# XXX: Define some GCCorrupt error ?
222
raise AssertionError('Invalid bytes: (%d) != %d + %d' %
223
(len(bytes), pos, self._z_content_length))
224
self._z_content = bytes[pos:]
227
def from_bytes(cls, bytes):
229
if bytes[:6] not in cls.GCB_KNOWN_HEADERS:
230
raise ValueError('bytes did not start with any of %r'
231
% (cls.GCB_KNOWN_HEADERS,))
232
# XXX: why not testing the whole header ?
234
out._compressor_name = 'zlib'
235
elif bytes[4] == 'l':
236
out._compressor_name = 'lzma'
238
raise ValueError('unknown compressor: %r' % (bytes,))
239
out._parse_bytes(bytes, 6)
242
def extract(self, key, start, end, sha1=None):
243
"""Extract the text for a specific key.
245
:param key: The label used for this content
246
:param sha1: TODO (should we validate only when sha1 is supplied?)
247
:return: The bytes for the content
249
if start == end == 0:
251
self._ensure_content(end)
252
# The bytes are 'f' or 'd' for the type, then a variable-length
253
# base128 integer for the content size, then the actual content
254
# We know that the variable-length integer won't be longer than 5
255
# bytes (it takes 5 bytes to encode 2^32)
256
c = self._content[start]
261
raise ValueError('Unknown content control code: %s'
264
content_len, len_len = decode_base128_int(
265
self._content[start + 1:start + 6])
266
content_start = start + 1 + len_len
267
if end != content_start + content_len:
268
raise ValueError('end != len according to field header'
269
' %s != %s' % (end, content_start + content_len))
271
bytes = self._content[content_start:end]
273
bytes = apply_delta_to_source(self._content, content_start, end)
276
def set_content(self, content):
277
"""Set the content of this block."""
278
self._content_length = len(content)
279
self._content = content
280
self._z_content = None
283
"""Encode the information into a byte stream."""
284
compress = zlib.compress
286
compress = pylzma.compress
287
if self._z_content is None:
288
if self._content is None:
289
raise AssertionError('Nothing to compress')
290
self._z_content = compress(self._content)
291
self._z_content_length = len(self._z_content)
293
header = self.GCB_LZ_HEADER
295
header = self.GCB_HEADER
297
'%d\n%d\n' % (self._z_content_length, self._content_length),
300
return ''.join(chunks)
302
def _dump(self, include_text=False):
303
"""Take this block, and spit out a human-readable structure.
305
:param include_text: Inserts also include text bits, chose whether you
306
want this displayed in the dump or not.
307
:return: A dump of the given block. The layout is something like:
308
[('f', length), ('d', delta_length, text_length, [delta_info])]
309
delta_info := [('i', num_bytes, text), ('c', offset, num_bytes),
312
self._ensure_content()
315
while pos < self._content_length:
316
kind = self._content[pos]
318
if kind not in ('f', 'd'):
319
raise ValueError('invalid kind character: %r' % (kind,))
320
content_len, len_len = decode_base128_int(
321
self._content[pos:pos + 5])
323
if content_len + pos > self._content_length:
324
raise ValueError('invalid content_len %d for record @ pos %d'
325
% (content_len, pos - len_len - 1))
326
if kind == 'f': # Fulltext
327
result.append(('f', content_len))
328
elif kind == 'd': # Delta
329
delta_content = self._content[pos:pos+content_len]
331
# The first entry in a delta is the decompressed length
332
decomp_len, delta_pos = decode_base128_int(delta_content)
333
result.append(('d', content_len, decomp_len, delta_info))
335
while delta_pos < content_len:
336
c = ord(delta_content[delta_pos])
340
delta_pos) = decode_copy_instruction(delta_content, c,
342
delta_info.append(('c', offset, length))
343
measured_len += length
346
txt = delta_content[delta_pos:delta_pos+c]
349
delta_info.append(('i', c, txt))
352
if delta_pos != content_len:
353
raise ValueError('Delta consumed a bad number of bytes:'
354
' %d != %d' % (delta_pos, content_len))
355
if measured_len != decomp_len:
356
raise ValueError('Delta claimed fulltext was %d bytes, but'
357
' extraction resulted in %d bytes'
358
% (decomp_len, measured_len))
363
class _LazyGroupCompressFactory(object):
364
"""Yield content from a GroupCompressBlock on demand."""
366
def __init__(self, key, parents, manager, start, end, first):
367
"""Create a _LazyGroupCompressFactory
369
:param key: The key of just this record
370
:param parents: The parents of this key (possibly None)
371
:param gc_block: A GroupCompressBlock object
372
:param start: Offset of the first byte for this record in the
374
:param end: Offset of the byte just after the end of this record
375
(ie, bytes = content[start:end])
376
:param first: Is this the first Factory for the given block?
379
self.parents = parents
381
# Note: This attribute coupled with Manager._factories creates a
382
# reference cycle. Perhaps we would rather use a weakref(), or
383
# find an appropriate time to release the ref. After the first
384
# get_bytes_as call? After Manager.get_record_stream() returns
386
self._manager = manager
388
self.storage_kind = 'groupcompress-block'
390
self.storage_kind = 'groupcompress-block-ref'
396
return '%s(%s, first=%s)' % (self.__class__.__name__,
397
self.key, self._first)
399
def get_bytes_as(self, storage_kind):
400
if storage_kind == self.storage_kind:
402
# wire bytes, something...
403
return self._manager._wire_bytes()
406
if storage_kind in ('fulltext', 'chunked'):
407
if self._bytes is None:
408
# Grab and cache the raw bytes for this entry
409
# and break the ref-cycle with _manager since we don't need it
411
self._manager._prepare_for_extract()
412
block = self._manager._block
413
self._bytes = block.extract(self.key, self._start, self._end)
414
# There are code paths that first extract as fulltext, and then
415
# extract as storage_kind (smart fetch). So we don't break the
416
# refcycle here, but instead in manager.get_record_stream()
417
# self._manager = None
418
if storage_kind == 'fulltext':
422
raise errors.UnavailableRepresentation(self.key, storage_kind,
426
class _LazyGroupContentManager(object):
427
"""This manages a group of _LazyGroupCompressFactory objects."""
429
def __init__(self, block):
431
# We need to preserve the ordering
435
def add_factory(self, key, parents, start, end):
436
if not self._factories:
440
# Note that this creates a reference cycle....
441
factory = _LazyGroupCompressFactory(key, parents, self,
442
start, end, first=first)
443
# max() works here, but as a function call, doing a compare seems to be
444
# significantly faster, timeit says 250ms for max() and 100ms for the
446
if end > self._last_byte:
447
self._last_byte = end
448
self._factories.append(factory)
450
def get_record_stream(self):
451
"""Get a record for all keys added so far."""
452
for factory in self._factories:
454
# Break the ref-cycle
455
factory._bytes = None
456
factory._manager = None
457
# TODO: Consider setting self._factories = None after the above loop,
458
# as it will break the reference cycle
460
def _trim_block(self, last_byte):
461
"""Create a new GroupCompressBlock, with just some of the content."""
462
# None of the factories need to be adjusted, because the content is
463
# located in an identical place. Just that some of the unreferenced
464
# trailing bytes are stripped
465
trace.mutter('stripping trailing bytes from groupcompress block'
466
' %d => %d', self._block._content_length, last_byte)
467
new_block = GroupCompressBlock()
468
self._block._ensure_content(last_byte)
469
new_block.set_content(self._block._content[:last_byte])
470
self._block = new_block
472
def _rebuild_block(self):
473
"""Create a new GroupCompressBlock with only the referenced texts."""
474
compressor = GroupCompressor()
476
old_length = self._block._content_length
478
for factory in self._factories:
479
bytes = factory.get_bytes_as('fulltext')
480
(found_sha1, start_point, end_point,
481
type) = compressor.compress(factory.key, bytes, factory.sha1)
482
# Now update this factory with the new offsets, etc
483
factory.sha1 = found_sha1
484
factory._start = start_point
485
factory._end = end_point
486
self._last_byte = end_point
487
new_block = compressor.flush()
488
# TODO: Should we check that new_block really *is* smaller than the old
489
# block? It seems hard to come up with a method that it would
490
# expand, since we do full compression again. Perhaps based on a
491
# request that ends up poorly ordered?
492
delta = time.time() - tstart
493
self._block = new_block
494
trace.mutter('creating new compressed block on-the-fly in %.3fs'
495
' %d bytes => %d bytes', delta, old_length,
496
self._block._content_length)
498
def _prepare_for_extract(self):
499
"""A _LazyGroupCompressFactory is about to extract to fulltext."""
500
# We expect that if one child is going to fulltext, all will be. This
501
# helps prevent all of them from extracting a small amount at a time.
502
# Which in itself isn't terribly expensive, but resizing 2MB 32kB at a
503
# time (self._block._content) is a little expensive.
504
self._block._ensure_content(self._last_byte)
506
def _check_rebuild_block(self):
507
"""Check to see if our block should be repacked."""
510
for factory in self._factories:
511
total_bytes_used += factory._end - factory._start
512
last_byte_used = max(last_byte_used, factory._end)
513
# If we are using most of the bytes from the block, we have nothing
514
# else to check (currently more that 1/2)
515
if total_bytes_used * 2 >= self._block._content_length:
517
# Can we just strip off the trailing bytes? If we are going to be
518
# transmitting more than 50% of the front of the content, go ahead
519
if total_bytes_used * 2 > last_byte_used:
520
self._trim_block(last_byte_used)
523
# We are using a small amount of the data, and it isn't just packed
524
# nicely at the front, so rebuild the content.
525
# Note: This would be *nicer* as a strip-data-from-group, rather than
526
# building it up again from scratch
527
# It might be reasonable to consider the fulltext sizes for
528
# different bits when deciding this, too. As you may have a small
529
# fulltext, and a trivial delta, and you are just trading around
530
# for another fulltext. If we do a simple 'prune' you may end up
531
# expanding many deltas into fulltexts, as well.
532
# If we build a cheap enough 'strip', then we could try a strip,
533
# if that expands the content, we then rebuild.
534
self._rebuild_block()
536
def _wire_bytes(self):
537
"""Return a byte stream suitable for transmitting over the wire."""
538
self._check_rebuild_block()
539
# The outer block starts with:
540
# 'groupcompress-block\n'
541
# <length of compressed key info>\n
542
# <length of uncompressed info>\n
543
# <length of gc block>\n
546
lines = ['groupcompress-block\n']
547
# The minimal info we need is the key, the start offset, and the
548
# parents. The length and type are encoded in the record itself.
549
# However, passing in the other bits makes it easier. The list of
550
# keys, and the start offset, the length
552
# 1 line with parents, '' for ()
553
# 1 line for start offset
554
# 1 line for end byte
556
for factory in self._factories:
557
key_bytes = '\x00'.join(factory.key)
558
parents = factory.parents
560
parent_bytes = 'None:'
562
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
563
record_header = '%s\n%s\n%d\n%d\n' % (
564
key_bytes, parent_bytes, factory._start, factory._end)
565
header_lines.append(record_header)
566
# TODO: Can we break the refcycle at this point and set
567
# factory._manager = None?
568
header_bytes = ''.join(header_lines)
570
header_bytes_len = len(header_bytes)
571
z_header_bytes = zlib.compress(header_bytes)
573
z_header_bytes_len = len(z_header_bytes)
574
block_bytes = self._block.to_bytes()
575
lines.append('%d\n%d\n%d\n' % (z_header_bytes_len, header_bytes_len,
577
lines.append(z_header_bytes)
578
lines.append(block_bytes)
579
del z_header_bytes, block_bytes
580
return ''.join(lines)
583
def from_bytes(cls, bytes):
584
# TODO: This does extra string copying, probably better to do it a
586
(storage_kind, z_header_len, header_len,
587
block_len, rest) = bytes.split('\n', 4)
589
if storage_kind != 'groupcompress-block':
590
raise ValueError('Unknown storage kind: %s' % (storage_kind,))
591
z_header_len = int(z_header_len)
592
if len(rest) < z_header_len:
593
raise ValueError('Compressed header len shorter than all bytes')
594
z_header = rest[:z_header_len]
595
header_len = int(header_len)
596
header = zlib.decompress(z_header)
597
if len(header) != header_len:
598
raise ValueError('invalid length for decompressed bytes')
600
block_len = int(block_len)
601
if len(rest) != z_header_len + block_len:
602
raise ValueError('Invalid length for block')
603
block_bytes = rest[z_header_len:]
605
# So now we have a valid GCB, we just need to parse the factories that
607
header_lines = header.split('\n')
609
last = header_lines.pop()
611
raise ValueError('header lines did not end with a trailing'
613
if len(header_lines) % 4 != 0:
614
raise ValueError('The header was not an even multiple of 4 lines')
615
block = GroupCompressBlock.from_bytes(block_bytes)
618
for start in xrange(0, len(header_lines), 4):
620
key = tuple(header_lines[start].split('\x00'))
621
parents_line = header_lines[start+1]
622
if parents_line == 'None:':
625
parents = tuple([tuple(segment.split('\x00'))
626
for segment in parents_line.split('\t')
628
start_offset = int(header_lines[start+2])
629
end_offset = int(header_lines[start+3])
630
result.add_factory(key, parents, start_offset, end_offset)
634
def network_block_to_records(storage_kind, bytes, line_end):
635
if storage_kind != 'groupcompress-block':
636
raise ValueError('Unknown storage kind: %s' % (storage_kind,))
637
manager = _LazyGroupContentManager.from_bytes(bytes)
638
return manager.get_record_stream()
641
class _CommonGroupCompressor(object):
644
"""Create a GroupCompressor."""
649
self.labels_deltas = {}
650
self._delta_index = None # Set by the children
651
self._block = GroupCompressBlock()
653
def compress(self, key, bytes, expected_sha, nostore_sha=None, soft=False):
654
"""Compress lines with label key.
656
:param key: A key tuple. It is stored in the output
657
for identification of the text during decompression. If the last
658
element is 'None' it is replaced with the sha1 of the text -
660
:param bytes: The bytes to be compressed
661
:param expected_sha: If non-None, the sha the lines are believed to
662
have. During compression the sha is calculated; a mismatch will
664
:param nostore_sha: If the computed sha1 sum matches, we will raise
665
ExistingContent rather than adding the text.
666
:param soft: Do a 'soft' compression. This means that we require larger
667
ranges to match to be considered for a copy command.
669
:return: The sha1 of lines, the start and end offsets in the delta, and
670
the type ('fulltext' or 'delta').
672
:seealso VersionedFiles.add_lines:
674
if not bytes: # empty, like a dir entry, etc
675
if nostore_sha == _null_sha1:
676
raise errors.ExistingContent()
677
return _null_sha1, 0, 0, 'fulltext'
678
# we assume someone knew what they were doing when they passed it in
679
if expected_sha is not None:
682
sha1 = osutils.sha_string(bytes)
683
if nostore_sha is not None:
684
if sha1 == nostore_sha:
685
raise errors.ExistingContent()
687
key = key[:-1] + ('sha1:' + sha1,)
689
start, end, type = self._compress(key, bytes, len(bytes) / 2, soft)
690
return sha1, start, end, type
692
def _compress(self, key, bytes, max_delta_size, soft=False):
693
"""Compress lines with label key.
695
:param key: A key tuple. It is stored in the output for identification
696
of the text during decompression.
698
:param bytes: The bytes to be compressed
700
:param max_delta_size: The size above which we issue a fulltext instead
703
:param soft: Do a 'soft' compression. This means that we require larger
704
ranges to match to be considered for a copy command.
706
:return: The sha1 of lines, the start and end offsets in the delta, and
707
the type ('fulltext' or 'delta').
709
raise NotImplementedError(self._compress)
711
def extract(self, key):
712
"""Extract a key previously added to the compressor.
714
:param key: The key to extract.
715
:return: An iterable over bytes and the sha1.
717
(start_byte, start_chunk, end_byte, end_chunk) = self.labels_deltas[key]
718
delta_chunks = self.chunks[start_chunk:end_chunk]
719
stored_bytes = ''.join(delta_chunks)
720
if stored_bytes[0] == 'f':
721
fulltext_len, offset = decode_base128_int(stored_bytes[1:10])
722
data_len = fulltext_len + 1 + offset
723
if data_len != len(stored_bytes):
724
raise ValueError('Index claimed fulltext len, but stored bytes'
726
% (len(stored_bytes), data_len))
727
bytes = stored_bytes[offset + 1:]
729
# XXX: This is inefficient at best
730
source = ''.join(self.chunks[:start_chunk])
731
if stored_bytes[0] != 'd':
732
raise ValueError('Unknown content kind, bytes claim %s'
733
% (stored_bytes[0],))
734
delta_len, offset = decode_base128_int(stored_bytes[1:10])
735
data_len = delta_len + 1 + offset
736
if data_len != len(stored_bytes):
737
raise ValueError('Index claimed delta len, but stored bytes'
739
% (len(stored_bytes), data_len))
740
bytes = apply_delta(source, stored_bytes[offset + 1:])
741
bytes_sha1 = osutils.sha_string(bytes)
742
return bytes, bytes_sha1
745
"""Finish this group, creating a formatted stream.
747
After calling this, the compressor should no longer be used
749
content = ''.join(self.chunks)
751
self._delta_index = None
752
self._block.set_content(content)
756
"""Call this if you want to 'revoke' the last compression.
758
After this, the data structures will be rolled back, but you cannot do
761
self._delta_index = None
762
del self.chunks[self._last[0]:]
763
self.endpoint = self._last[1]
767
"""Return the overall compression ratio."""
768
return float(self.input_bytes) / float(self.endpoint)
771
class PythonGroupCompressor(_CommonGroupCompressor):
774
"""Create a GroupCompressor.
776
Used only if the pyrex version is not available.
778
super(PythonGroupCompressor, self).__init__()
779
self._delta_index = LinesDeltaIndex([])
780
# The actual content is managed by LinesDeltaIndex
781
self.chunks = self._delta_index.lines
783
def _compress(self, key, bytes, max_delta_size, soft=False):
784
"""see _CommonGroupCompressor._compress"""
785
input_len = len(bytes)
786
new_lines = osutils.split_lines(bytes)
787
out_lines, index_lines = self._delta_index.make_delta(
788
new_lines, bytes_length=input_len, soft=soft)
789
delta_length = sum(map(len, out_lines))
790
if delta_length > max_delta_size:
791
# The delta is longer than the fulltext, insert a fulltext
793
out_lines = ['f', encode_base128_int(input_len)]
794
out_lines.extend(new_lines)
795
index_lines = [False, False]
796
index_lines.extend([True] * len(new_lines))
798
# this is a worthy delta, output it
801
# Update the delta_length to include those two encoded integers
802
out_lines[1] = encode_base128_int(delta_length)
804
start = self.endpoint
805
chunk_start = len(self.chunks)
806
self._last = (chunk_start, self.endpoint)
807
self._delta_index.extend_lines(out_lines, index_lines)
808
self.endpoint = self._delta_index.endpoint
809
self.input_bytes += input_len
810
chunk_end = len(self.chunks)
811
self.labels_deltas[key] = (start, chunk_start,
812
self.endpoint, chunk_end)
813
return start, self.endpoint, type
816
class PyrexGroupCompressor(_CommonGroupCompressor):
817
"""Produce a serialised group of compressed texts.
819
It contains code very similar to SequenceMatcher because of having a similar
820
task. However some key differences apply:
821
- there is no junk, we want a minimal edit not a human readable diff.
822
- we don't filter very common lines (because we don't know where a good
823
range will start, and after the first text we want to be emitting minmal
825
- we chain the left side, not the right side
826
- we incrementally update the adjacency matrix as new lines are provided.
827
- we look for matches in all of the left side, so the routine which does
828
the analagous task of find_longest_match does not need to filter on the
833
super(PyrexGroupCompressor, self).__init__()
834
self._delta_index = DeltaIndex()
836
def _compress(self, key, bytes, max_delta_size, soft=False):
837
"""see _CommonGroupCompressor._compress"""
838
input_len = len(bytes)
839
# By having action/label/sha1/len, we can parse the group if the index
840
# was ever destroyed, we have the key in 'label', we know the final
841
# bytes are valid from sha1, and we know where to find the end of this
842
# record because of 'len'. (the delta record itself will store the
843
# total length for the expanded record)
844
# 'len: %d\n' costs approximately 1% increase in total data
845
# Having the labels at all costs us 9-10% increase, 38% increase for
846
# inventory pages, and 5.8% increase for text pages
847
# new_chunks = ['label:%s\nsha1:%s\n' % (label, sha1)]
848
if self._delta_index._source_offset != self.endpoint:
849
raise AssertionError('_source_offset != endpoint'
850
' somehow the DeltaIndex got out of sync with'
852
delta = self._delta_index.make_delta(bytes, max_delta_size)
855
enc_length = encode_base128_int(len(bytes))
856
len_mini_header = 1 + len(enc_length)
857
self._delta_index.add_source(bytes, len_mini_header)
858
new_chunks = ['f', enc_length, bytes]
861
enc_length = encode_base128_int(len(delta))
862
len_mini_header = 1 + len(enc_length)
863
new_chunks = ['d', enc_length, delta]
864
self._delta_index.add_delta_source(delta, len_mini_header)
866
start = self.endpoint
867
chunk_start = len(self.chunks)
868
# Now output these bytes
869
self._output_chunks(new_chunks)
870
self.input_bytes += input_len
871
chunk_end = len(self.chunks)
872
self.labels_deltas[key] = (start, chunk_start,
873
self.endpoint, chunk_end)
874
if not self._delta_index._source_offset == self.endpoint:
875
raise AssertionError('the delta index is out of sync'
876
'with the output lines %s != %s'
877
% (self._delta_index._source_offset, self.endpoint))
878
return start, self.endpoint, type
880
def _output_chunks(self, new_chunks):
881
"""Output some chunks.
883
:param new_chunks: The chunks to output.
885
self._last = (len(self.chunks), self.endpoint)
886
endpoint = self.endpoint
887
self.chunks.extend(new_chunks)
888
endpoint += sum(map(len, new_chunks))
889
self.endpoint = endpoint
892
def make_pack_factory(graph, delta, keylength):
893
"""Create a factory for creating a pack based groupcompress.
895
This is only functional enough to run interface tests, it doesn't try to
896
provide a full pack environment.
898
:param graph: Store a graph.
899
:param delta: Delta compress contents.
900
:param keylength: How long should keys be.
902
def factory(transport):
907
graph_index = BTreeBuilder(reference_lists=ref_length,
908
key_elements=keylength)
909
stream = transport.open_write_stream('newpack')
910
writer = pack.ContainerWriter(stream.write)
912
index = _GCGraphIndex(graph_index, lambda:True, parents=parents,
913
add_callback=graph_index.add_nodes)
914
access = knit._DirectPackAccess({})
915
access.set_writer(writer, graph_index, (transport, 'newpack'))
916
result = GroupCompressVersionedFiles(index, access, delta)
917
result.stream = stream
918
result.writer = writer
923
def cleanup_pack_group(versioned_files):
924
versioned_files.writer.end()
925
versioned_files.stream.close()
928
class GroupCompressVersionedFiles(VersionedFiles):
929
"""A group-compress based VersionedFiles implementation."""
931
def __init__(self, index, access, delta=True):
932
"""Create a GroupCompressVersionedFiles object.
934
:param index: The index object storing access and graph data.
935
:param access: The access object storing raw data.
936
:param delta: Whether to delta compress or just entropy compress.
939
self._access = access
941
self._unadded_refs = {}
942
self._group_cache = LRUSizeCache(max_size=50*1024*1024)
943
self._fallback_vfs = []
945
def add_lines(self, key, parents, lines, parent_texts=None,
946
left_matching_blocks=None, nostore_sha=None, random_id=False,
948
"""Add a text to the store.
950
:param key: The key tuple of the text to add.
951
:param parents: The parents key tuples of the text to add.
952
:param lines: A list of lines. Each line must be a bytestring. And all
953
of them except the last must be terminated with \n and contain no
954
other \n's. The last line may either contain no \n's or a single
955
terminating \n. If the lines list does meet this constraint the add
956
routine may error or may succeed - but you will be unable to read
957
the data back accurately. (Checking the lines have been split
958
correctly is expensive and extremely unlikely to catch bugs so it
959
is not done at runtime unless check_content is True.)
960
:param parent_texts: An optional dictionary containing the opaque
961
representations of some or all of the parents of version_id to
962
allow delta optimisations. VERY IMPORTANT: the texts must be those
963
returned by add_lines or data corruption can be caused.
964
:param left_matching_blocks: a hint about which areas are common
965
between the text and its left-hand-parent. The format is
966
the SequenceMatcher.get_matching_blocks format.
967
:param nostore_sha: Raise ExistingContent and do not add the lines to
968
the versioned file if the digest of the lines matches this.
969
:param random_id: If True a random id has been selected rather than
970
an id determined by some deterministic process such as a converter
971
from a foreign VCS. When True the backend may choose not to check
972
for uniqueness of the resulting key within the versioned file, so
973
this should only be done when the result is expected to be unique
975
:param check_content: If True, the lines supplied are verified to be
976
bytestrings that are correctly formed lines.
977
:return: The text sha1, the number of bytes in the text, and an opaque
978
representation of the inserted version which can be provided
979
back to future add_lines calls in the parent_texts dictionary.
981
self._index._check_write_ok()
982
self._check_add(key, lines, random_id, check_content)
984
# The caller might pass None if there is no graph data, but kndx
985
# indexes can't directly store that, so we give them
986
# an empty tuple instead.
988
# double handling for now. Make it work until then.
989
length = sum(map(len, lines))
990
record = ChunkedContentFactory(key, parents, None, lines)
991
sha1 = list(self._insert_record_stream([record], random_id=random_id,
992
nostore_sha=nostore_sha))[0]
993
return sha1, length, None
995
def add_fallback_versioned_files(self, a_versioned_files):
996
"""Add a source of texts for texts not present in this knit.
998
:param a_versioned_files: A VersionedFiles object.
1000
self._fallback_vfs.append(a_versioned_files)
1002
def annotate(self, key):
1003
"""See VersionedFiles.annotate."""
1005
parent_map = self.get_parent_map([key])
1007
raise errors.RevisionNotPresent(key, self)
1008
if parent_map[key] is not None:
1009
search = graph._make_breadth_first_searcher([key])
1013
present, ghosts = search.next_with_ghosts()
1014
except StopIteration:
1016
keys.update(present)
1017
parent_map = self.get_parent_map(keys)
1020
parent_map = {key:()}
1021
# So we used Graph(self) to load the parent_map, but now that we have
1022
# it, we can just query the parent map directly, so create a new Graph
1024
graph = _mod_graph.Graph(_mod_graph.DictParentsProvider(parent_map))
1025
head_cache = _mod_graph.FrozenHeadsCache(graph)
1027
reannotate = annotate.reannotate
1028
for record in self.get_record_stream(keys, 'topological', True):
1030
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
1031
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
1032
parent_cache[key] = list(
1033
reannotate(parent_lines, lines, key, None, head_cache))
1034
return parent_cache[key]
1036
def check(self, progress_bar=None):
1037
"""See VersionedFiles.check()."""
1039
for record in self.get_record_stream(keys, 'unordered', True):
1040
record.get_bytes_as('fulltext')
1042
def _check_add(self, key, lines, random_id, check_content):
1043
"""check that version_id and lines are safe to add."""
1044
version_id = key[-1]
1045
if version_id is not None:
1046
if osutils.contains_whitespace(version_id):
1047
raise errors.InvalidRevisionId(version_id, self)
1048
self.check_not_reserved_id(version_id)
1049
# TODO: If random_id==False and the key is already present, we should
1050
# probably check that the existing content is identical to what is
1051
# being inserted, and otherwise raise an exception. This would make
1052
# the bundle code simpler.
1054
self._check_lines_not_unicode(lines)
1055
self._check_lines_are_lines(lines)
1057
def get_parent_map(self, keys):
1058
"""Get a map of the graph parents of keys.
1060
:param keys: The keys to look up parents for.
1061
:return: A mapping from keys to parents. Absent keys are absent from
1064
return self._get_parent_map_with_sources(keys)[0]
1066
def _get_parent_map_with_sources(self, keys):
1067
"""Get a map of the parents of keys.
1069
:param keys: The keys to look up parents for.
1070
:return: A tuple. The first element is a mapping from keys to parents.
1071
Absent keys are absent from the mapping. The second element is a
1072
list with the locations each key was found in. The first element
1073
is the in-this-knit parents, the second the first fallback source,
1077
sources = [self._index] + self._fallback_vfs
1080
for source in sources:
1083
new_result = source.get_parent_map(missing)
1084
source_results.append(new_result)
1085
result.update(new_result)
1086
missing.difference_update(set(new_result))
1087
return result, source_results
1089
def _get_block(self, index_memo):
1090
read_memo = index_memo[0:3]
1093
block = self._group_cache[read_memo]
1096
zdata = self._access.get_raw_records([read_memo]).next()
1097
# decompress - whole thing - this is not a bug, as it
1098
# permits caching. We might want to store the partially
1099
# decompresed group and decompress object, so that recent
1100
# texts are not penalised by big groups.
1101
block = GroupCompressBlock.from_bytes(zdata)
1102
self._group_cache[read_memo] = block
1104
# print len(zdata), len(plain)
1105
# parse - requires split_lines, better to have byte offsets
1106
# here (but not by much - we only split the region for the
1107
# recipe, and we often want to end up with lines anyway.
1110
def get_missing_compression_parent_keys(self):
1111
"""Return the keys of missing compression parents.
1113
Missing compression parents occur when a record stream was missing
1114
basis texts, or a index was scanned that had missing basis texts.
1116
# GroupCompress cannot currently reference texts that are not in the
1117
# group, so this is valid for now
1120
def get_record_stream(self, keys, ordering, include_delta_closure):
1121
"""Get a stream of records for keys.
1123
:param keys: The keys to include.
1124
:param ordering: Either 'unordered' or 'topological'. A topologically
1125
sorted stream has compression parents strictly before their
1127
:param include_delta_closure: If True then the closure across any
1128
compression parents will be included (in the opaque data).
1129
:return: An iterator of ContentFactory objects, each of which is only
1130
valid until the iterator is advanced.
1132
# keys might be a generator
1133
orig_keys = list(keys)
1137
if (not self._index.has_graph
1138
and ordering in ('topological', 'groupcompress')):
1139
# Cannot topological order when no graph has been stored.
1140
# but we allow 'as-requested' or 'unordered'
1141
ordering = 'unordered'
1143
remaining_keys = keys
1146
keys = set(remaining_keys)
1147
for content_factory in self._get_remaining_record_stream(keys,
1148
orig_keys, ordering, include_delta_closure):
1149
remaining_keys.discard(content_factory.key)
1150
yield content_factory
1152
except errors.RetryWithNewPacks, e:
1153
self._access.reload_or_raise(e)
1155
def _find_from_fallback(self, missing):
1156
"""Find whatever keys you can from the fallbacks.
1158
:param missing: A set of missing keys. This set will be mutated as keys
1159
are found from a fallback_vfs
1160
:return: (parent_map, key_to_source_map, source_results)
1161
parent_map the overall key => parent_keys
1162
key_to_source_map a dict from {key: source}
1163
source_results a list of (source: keys)
1166
key_to_source_map = {}
1168
for source in self._fallback_vfs:
1171
source_parents = source.get_parent_map(missing)
1172
parent_map.update(source_parents)
1173
source_parents = list(source_parents)
1174
source_results.append((source, source_parents))
1175
key_to_source_map.update((key, source) for key in source_parents)
1176
missing.difference_update(source_parents)
1177
return parent_map, key_to_source_map, source_results
1179
def _get_ordered_source_keys(self, ordering, parent_map, key_to_source_map):
1180
"""Get the (source, [keys]) list.
1182
The returned objects should be in the order defined by 'ordering',
1183
which can weave between different sources.
1184
:param ordering: Must be one of 'topological' or 'groupcompress'
1185
:return: List of [(source, [keys])] tuples, such that all keys are in
1186
the defined order, regardless of source.
1188
if ordering == 'topological':
1189
present_keys = topo_sort(parent_map)
1191
# ordering == 'groupcompress'
1192
# XXX: This only optimizes for the target ordering. We may need
1193
# to balance that with the time it takes to extract
1194
# ordering, by somehow grouping based on
1195
# locations[key][0:3]
1196
present_keys = sort_gc_optimal(parent_map)
1197
# Now group by source:
1199
current_source = None
1200
for key in present_keys:
1201
source = key_to_source_map.get(key, self)
1202
if source is not current_source:
1203
source_keys.append((source, []))
1204
current_source = source
1205
source_keys[-1][1].append(key)
1208
def _get_as_requested_source_keys(self, orig_keys, locations, unadded_keys,
1211
current_source = None
1212
for key in orig_keys:
1213
if key in locations or key in unadded_keys:
1215
elif key in key_to_source_map:
1216
source = key_to_source_map[key]
1219
if source is not current_source:
1220
source_keys.append((source, []))
1221
current_source = source
1222
source_keys[-1][1].append(key)
1225
def _get_io_ordered_source_keys(self, locations, unadded_keys,
1228
# This is the group the bytes are stored in, followed by the
1229
# location in the group
1230
return locations[key][0]
1231
present_keys = sorted(locations.iterkeys(), key=get_group)
1232
# We don't have an ordering for keys in the in-memory object, but
1233
# lets process the in-memory ones first.
1234
present_keys = list(unadded_keys) + present_keys
1235
# Now grab all of the ones from other sources
1236
source_keys = [(self, present_keys)]
1237
source_keys.extend(source_result)
1240
def _get_remaining_record_stream(self, keys, orig_keys, ordering,
1241
include_delta_closure):
1242
"""Get a stream of records for keys.
1244
:param keys: The keys to include.
1245
:param ordering: one of 'unordered', 'topological', 'groupcompress' or
1247
:param include_delta_closure: If True then the closure across any
1248
compression parents will be included (in the opaque data).
1249
:return: An iterator of ContentFactory objects, each of which is only
1250
valid until the iterator is advanced.
1253
locations = self._index.get_build_details(keys)
1254
unadded_keys = set(self._unadded_refs).intersection(keys)
1255
missing = keys.difference(locations)
1256
missing.difference_update(unadded_keys)
1257
(fallback_parent_map, key_to_source_map,
1258
source_result) = self._find_from_fallback(missing)
1259
if ordering in ('topological', 'groupcompress'):
1260
# would be better to not globally sort initially but instead
1261
# start with one key, recurse to its oldest parent, then grab
1262
# everything in the same group, etc.
1263
parent_map = dict((key, details[2]) for key, details in
1264
locations.iteritems())
1265
for key in unadded_keys:
1266
parent_map[key] = self._unadded_refs[key]
1267
parent_map.update(fallback_parent_map)
1268
source_keys = self._get_ordered_source_keys(ordering, parent_map,
1270
elif ordering == 'as-requested':
1271
source_keys = self._get_as_requested_source_keys(orig_keys,
1272
locations, unadded_keys, key_to_source_map)
1274
# We want to yield the keys in a semi-optimal (read-wise) ordering.
1275
# Otherwise we thrash the _group_cache and destroy performance
1276
source_keys = self._get_io_ordered_source_keys(locations,
1277
unadded_keys, source_result)
1279
yield AbsentContentFactory(key)
1281
last_read_memo = None
1282
# TODO: This works fairly well at batching up existing groups into a
1283
# streamable format, and possibly allowing for taking one big
1284
# group and splitting it when it isn't fully utilized.
1285
# However, it doesn't allow us to find under-utilized groups and
1286
# combine them into a bigger group on the fly.
1287
# (Consider the issue with how chk_map inserts texts
1288
# one-at-a-time.) This could be done at insert_record_stream()
1289
# time, but it probably would decrease the number of
1290
# bytes-on-the-wire for fetch.
1291
for source, keys in source_keys:
1294
if key in self._unadded_refs:
1295
if manager is not None:
1296
for factory in manager.get_record_stream():
1298
last_read_memo = manager = None
1299
bytes, sha1 = self._compressor.extract(key)
1300
parents = self._unadded_refs[key]
1301
yield FulltextContentFactory(key, parents, sha1, bytes)
1303
index_memo, _, parents, (method, _) = locations[key]
1304
read_memo = index_memo[0:3]
1305
if last_read_memo != read_memo:
1306
# We are starting a new block. If we have a
1307
# manager, we have found everything that fits for
1308
# now, so yield records
1309
if manager is not None:
1310
for factory in manager.get_record_stream():
1312
# Now start a new manager
1313
block = self._get_block(index_memo)
1314
manager = _LazyGroupContentManager(block)
1315
last_read_memo = read_memo
1316
start, end = index_memo[3:5]
1317
manager.add_factory(key, parents, start, end)
1319
if manager is not None:
1320
for factory in manager.get_record_stream():
1322
last_read_memo = manager = None
1323
for record in source.get_record_stream(keys, ordering,
1324
include_delta_closure):
1326
if manager is not None:
1327
for factory in manager.get_record_stream():
1330
def get_sha1s(self, keys):
1331
"""See VersionedFiles.get_sha1s()."""
1333
for record in self.get_record_stream(keys, 'unordered', True):
1334
if record.sha1 != None:
1335
result[record.key] = record.sha1
1337
if record.storage_kind != 'absent':
1338
result[record.key] = osutils.sha_string(
1339
record.get_bytes_as('fulltext'))
1342
def insert_record_stream(self, stream):
1343
"""Insert a record stream into this container.
1345
:param stream: A stream of records to insert.
1347
:seealso VersionedFiles.get_record_stream:
1349
# XXX: Setting random_id=True makes
1350
# test_insert_record_stream_existing_keys fail for groupcompress and
1351
# groupcompress-nograph, this needs to be revisited while addressing
1352
# 'bzr branch' performance issues.
1353
for _ in self._insert_record_stream(stream, random_id=False):
1356
def _insert_record_stream(self, stream, random_id=False, nostore_sha=None,
1358
"""Internal core to insert a record stream into this container.
1360
This helper function has a different interface than insert_record_stream
1361
to allow add_lines to be minimal, but still return the needed data.
1363
:param stream: A stream of records to insert.
1364
:param nostore_sha: If the sha1 of a given text matches nostore_sha,
1365
raise ExistingContent, rather than committing the new text.
1366
:param reuse_blocks: If the source is streaming from
1367
groupcompress-blocks, just insert the blocks as-is, rather than
1368
expanding the texts and inserting again.
1369
:return: An iterator over the sha1 of the inserted records.
1370
:seealso insert_record_stream:
1374
def get_adapter(adapter_key):
1376
return adapters[adapter_key]
1378
adapter_factory = adapter_registry.get(adapter_key)
1379
adapter = adapter_factory(self)
1380
adapters[adapter_key] = adapter
1382
# This will go up to fulltexts for gc to gc fetching, which isn't
1384
self._compressor = GroupCompressor()
1385
self._unadded_refs = {}
1388
bytes = self._compressor.flush().to_bytes()
1389
index, start, length = self._access.add_raw_records(
1390
[(None, len(bytes))], bytes)[0]
1392
for key, reads, refs in keys_to_add:
1393
nodes.append((key, "%d %d %s" % (start, length, reads), refs))
1394
self._index.add_records(nodes, random_id=random_id)
1395
self._unadded_refs = {}
1397
self._compressor = GroupCompressor()
1400
max_fulltext_len = 0
1401
max_fulltext_prefix = None
1402
insert_manager = None
1405
# XXX: TODO: remove this, it is just for safety checking for now
1406
inserted_keys = set()
1407
for record in stream:
1408
# Raise an error when a record is missing.
1409
if record.storage_kind == 'absent':
1410
raise errors.RevisionNotPresent(record.key, self)
1412
if record.key in inserted_keys:
1413
trace.note('Insert claimed random_id=True,'
1414
' but then inserted %r two times', record.key)
1416
inserted_keys.add(record.key)
1418
# If the reuse_blocks flag is set, check to see if we can just
1419
# copy a groupcompress block as-is.
1420
if record.storage_kind == 'groupcompress-block':
1421
# Insert the raw block into the target repo
1422
insert_manager = record._manager
1423
insert_manager._check_rebuild_block()
1424
bytes = record._manager._block.to_bytes()
1425
_, start, length = self._access.add_raw_records(
1426
[(None, len(bytes))], bytes)[0]
1429
block_length = length
1430
if record.storage_kind in ('groupcompress-block',
1431
'groupcompress-block-ref'):
1432
if insert_manager is None:
1433
raise AssertionError('No insert_manager set')
1434
value = "%d %d %d %d" % (block_start, block_length,
1435
record._start, record._end)
1436
nodes = [(record.key, value, (record.parents,))]
1437
# TODO: Consider buffering up many nodes to be added, not
1438
# sure how much overhead this has, but we're seeing
1439
# ~23s / 120s in add_records calls
1440
self._index.add_records(nodes, random_id=random_id)
1443
bytes = record.get_bytes_as('fulltext')
1444
except errors.UnavailableRepresentation:
1445
adapter_key = record.storage_kind, 'fulltext'
1446
adapter = get_adapter(adapter_key)
1447
bytes = adapter.get_bytes(record)
1448
if len(record.key) > 1:
1449
prefix = record.key[0]
1450
soft = (prefix == last_prefix)
1454
if max_fulltext_len < len(bytes):
1455
max_fulltext_len = len(bytes)
1456
max_fulltext_prefix = prefix
1457
(found_sha1, start_point, end_point,
1458
type) = self._compressor.compress(record.key,
1459
bytes, record.sha1, soft=soft,
1460
nostore_sha=nostore_sha)
1461
# delta_ratio = float(len(bytes)) / (end_point - start_point)
1462
# Check if we want to continue to include that text
1463
if (prefix == max_fulltext_prefix
1464
and end_point < 2 * max_fulltext_len):
1465
# As long as we are on the same file_id, we will fill at least
1466
# 2 * max_fulltext_len
1467
start_new_block = False
1468
elif end_point > 4*1024*1024:
1469
start_new_block = True
1470
elif (prefix is not None and prefix != last_prefix
1471
and end_point > 2*1024*1024):
1472
start_new_block = True
1474
start_new_block = False
1475
last_prefix = prefix
1477
self._compressor.pop_last()
1479
max_fulltext_len = len(bytes)
1480
(found_sha1, start_point, end_point,
1481
type) = self._compressor.compress(record.key, bytes,
1483
if record.key[-1] is None:
1484
key = record.key[:-1] + ('sha1:' + found_sha1,)
1487
self._unadded_refs[key] = record.parents
1489
keys_to_add.append((key, '%d %d' % (start_point, end_point),
1491
if len(keys_to_add):
1493
self._compressor = None
1495
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1496
"""Iterate over the lines in the versioned files from keys.
1498
This may return lines from other keys. Each item the returned
1499
iterator yields is a tuple of a line and a text version that that line
1500
is present in (not introduced in).
1502
Ordering of results is in whatever order is most suitable for the
1503
underlying storage format.
1505
If a progress bar is supplied, it may be used to indicate progress.
1506
The caller is responsible for cleaning up progress bars (because this
1510
* Lines are normalised by the underlying store: they will all have \n
1512
* Lines are returned in arbitrary order.
1514
:return: An iterator over (line, key).
1517
pb = progress.DummyProgress()
1520
# we don't care about inclusions, the caller cares.
1521
# but we need to setup a list of records to visit.
1522
# we need key, position, length
1523
for key_idx, record in enumerate(self.get_record_stream(keys,
1524
'unordered', True)):
1525
# XXX: todo - optimise to use less than full texts.
1527
pb.update('Walking content', key_idx, total)
1528
if record.storage_kind == 'absent':
1529
raise errors.RevisionNotPresent(key, self)
1530
lines = osutils.split_lines(record.get_bytes_as('fulltext'))
1533
pb.update('Walking content', total, total)
1536
"""See VersionedFiles.keys."""
1537
if 'evil' in debug.debug_flags:
1538
trace.mutter_callsite(2, "keys scales with size of history")
1539
sources = [self._index] + self._fallback_vfs
1541
for source in sources:
1542
result.update(source.keys())
1546
class _GCGraphIndex(object):
1547
"""Mapper from GroupCompressVersionedFiles needs into GraphIndex storage."""
1549
def __init__(self, graph_index, is_locked, parents=True,
1550
add_callback=None, track_external_parent_refs=False):
1551
"""Construct a _GCGraphIndex on a graph_index.
1553
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1554
:param is_locked: A callback, returns True if the index is locked and
1556
:param parents: If True, record knits parents, if not do not record
1558
:param add_callback: If not None, allow additions to the index and call
1559
this callback with a list of added GraphIndex nodes:
1560
[(node, value, node_refs), ...]
1561
:param track_external_parent_refs: As keys are added, keep track of the
1562
keys they reference, so that we can query get_missing_parents(),
1565
self._add_callback = add_callback
1566
self._graph_index = graph_index
1567
self._parents = parents
1568
self.has_graph = parents
1569
self._is_locked = is_locked
1570
if track_external_parent_refs:
1571
self._key_dependencies = knit._KeyRefs()
1573
self._key_dependencies = None
1575
def add_records(self, records, random_id=False):
1576
"""Add multiple records to the index.
1578
This function does not insert data into the Immutable GraphIndex
1579
backing the KnitGraphIndex, instead it prepares data for insertion by
1580
the caller and checks that it is safe to insert then calls
1581
self._add_callback with the prepared GraphIndex nodes.
1583
:param records: a list of tuples:
1584
(key, options, access_memo, parents).
1585
:param random_id: If True the ids being added were randomly generated
1586
and no check for existence will be performed.
1588
if not self._add_callback:
1589
raise errors.ReadOnlyError(self)
1590
# we hope there are no repositories with inconsistent parentage
1595
for (key, value, refs) in records:
1596
if not self._parents:
1600
raise KnitCorrupt(self,
1601
"attempt to add node with parents "
1602
"in parentless index.")
1605
keys[key] = (value, refs)
1608
present_nodes = self._get_entries(keys)
1609
for (index, key, value, node_refs) in present_nodes:
1610
if node_refs != keys[key][1]:
1611
raise errors.KnitCorrupt(self, "inconsistent details in add_records"
1612
": %s %s" % ((value, node_refs), keys[key]))
1618
for key, (value, node_refs) in keys.iteritems():
1619
result.append((key, value, node_refs))
1621
for key, (value, node_refs) in keys.iteritems():
1622
result.append((key, value))
1624
key_dependencies = self._key_dependencies
1625
if key_dependencies is not None and self._parents:
1626
for key, value, refs in records:
1628
key_dependencies.add_references(key, parents)
1629
self._add_callback(records)
1631
def _check_read(self):
1632
"""Raise an exception if reads are not permitted."""
1633
if not self._is_locked():
1634
raise errors.ObjectNotLocked(self)
1636
def _check_write_ok(self):
1637
"""Raise an exception if writes are not permitted."""
1638
if not self._is_locked():
1639
raise errors.ObjectNotLocked(self)
1641
def _get_entries(self, keys, check_present=False):
1642
"""Get the entries for keys.
1644
Note: Callers are responsible for checking that the index is locked
1645
before calling this method.
1647
:param keys: An iterable of index key tuples.
1652
for node in self._graph_index.iter_entries(keys):
1654
found_keys.add(node[1])
1656
# adapt parentless index to the rest of the code.
1657
for node in self._graph_index.iter_entries(keys):
1658
yield node[0], node[1], node[2], ()
1659
found_keys.add(node[1])
1661
missing_keys = keys.difference(found_keys)
1663
raise RevisionNotPresent(missing_keys.pop(), self)
1665
def get_parent_map(self, keys):
1666
"""Get a map of the parents of keys.
1668
:param keys: The keys to look up parents for.
1669
:return: A mapping from keys to parents. Absent keys are absent from
1673
nodes = self._get_entries(keys)
1677
result[node[1]] = node[3][0]
1680
result[node[1]] = None
1683
def get_missing_parents(self):
1684
"""Return the keys of missing parents."""
1685
# Copied from _KnitGraphIndex.get_missing_parents
1686
# We may have false positives, so filter those out.
1687
self._key_dependencies.add_keys(
1688
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
1689
return frozenset(self._key_dependencies.get_unsatisfied_refs())
1691
def get_build_details(self, keys):
1692
"""Get the various build details for keys.
1694
Ghosts are omitted from the result.
1696
:param keys: An iterable of keys.
1697
:return: A dict of key:
1698
(index_memo, compression_parent, parents, record_details).
1700
opaque structure to pass to read_records to extract the raw
1703
Content that this record is built upon, may be None
1705
Logical parents of this node
1707
extra information about the content which needs to be passed to
1708
Factory.parse_record
1712
entries = self._get_entries(keys)
1713
for entry in entries:
1715
if not self._parents:
1718
parents = entry[3][0]
1720
result[key] = (self._node_to_position(entry),
1721
None, parents, (method, None))
1725
"""Get all the keys in the collection.
1727
The keys are not ordered.
1730
return [node[1] for node in self._graph_index.iter_all_entries()]
1732
def _node_to_position(self, node):
1733
"""Convert an index value to position details."""
1734
bits = node[2].split(' ')
1735
# It would be nice not to read the entire gzip.
1736
start = int(bits[0])
1738
basis_end = int(bits[2])
1739
delta_end = int(bits[3])
1740
return node[0], start, stop, basis_end, delta_end
1742
def scan_unvalidated_index(self, graph_index):
1743
"""Inform this _GCGraphIndex that there is an unvalidated index.
1745
This allows this _GCGraphIndex to keep track of any missing
1746
compression parents we may want to have filled in to make those
1749
:param graph_index: A GraphIndex
1751
if self._key_dependencies is not None:
1752
# Add parent refs from graph_index (and discard parent refs that
1753
# the graph_index has).
1754
add_refs = self._key_dependencies.add_references
1755
for node in graph_index.iter_all_entries():
1756
add_refs(node[1], node[3][0])
1760
from bzrlib._groupcompress_py import (
1762
apply_delta_to_source,
1765
decode_copy_instruction,
1769
from bzrlib._groupcompress_pyx import (
1771
apply_delta_to_source,
1776
GroupCompressor = PyrexGroupCompressor
1778
GroupCompressor = PythonGroupCompressor
added
removed
bzrlib/groupcompress.py
