1
# Copyright (C) 2006, 2007 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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""DirState objects record the state of a directory and its bzr metadata.
19
Pseudo EBNF grammar for the state file. Fields are separated by NULLs, and
20
lines by NL. The field delimiters are ommitted in the grammar, line delimiters
21
are not - this is done for clarity of reading. All string data is in utf8.
23
MINIKIND = "f" | "d" | "l" | "a" | "r";
26
WHOLE_NUMBER = {digit}, digit;
28
REVISION_ID = a non-empty utf8 string;
30
dirstate format = header line, full checksum, row count, parent details,
31
ghost_details, entries;
32
header line = "#bazaar dirstate flat format 2", NL;
33
full checksum = "adler32: ", ["-"], WHOLE_NUMBER, NL;
34
row count = "num_entries: ", digit, NL;
35
parent_details = WHOLE NUMBER, {REVISION_ID}* NL;
36
ghost_details = WHOLE NUMBER, {REVISION_ID}*, NL;
38
entry = entry_key, current_entry_details, {parent_entry_details};
39
entry_key = dirname, basename, fileid;
40
current_entry_details = common_entry_details, working_entry_details;
41
parent_entry_details = common_entry_details, history_entry_details;
42
common_entry_details = MINIKIND, fingerprint, size, executable
43
working_entry_details = packed_stat
44
history_entry_details = REVISION_ID;
47
fingerprint = a nonempty utf8 sequence with meaning defined by minikind.
49
Given this definition, the following is useful to know:
50
entry (aka row) - all the data for a given key.
51
entry[0]: The key (dirname, basename, fileid)
55
entry[1]: The tree(s) data for this path and id combination.
56
entry[1][0]: The current tree
57
entry[1][1]: The second tree
59
For an entry for a tree, we have (using tree 0 - current tree) to demonstrate:
60
entry[1][0][0]: minikind
61
entry[1][0][1]: fingerprint
63
entry[1][0][3]: executable
64
entry[1][0][4]: packed_stat
66
entry[1][1][4]: revision_id
68
There may be multiple rows at the root, one per id present in the root, so the
69
in memory root row is now:
70
self._dirblocks[0] -> ('', [entry ...]),
71
and the entries in there are
74
entries[0][2]: file_id
75
entries[1][0]: The tree data for the current tree for this fileid at /
79
'r' is a relocated entry: This path is not present in this tree with this id,
80
but the id can be found at another location. The fingerprint is used to
81
point to the target location.
82
'a' is an absent entry: In that tree the id is not present at this path.
83
'd' is a directory entry: This path in this tree is a directory with the
84
current file id. There is no fingerprint for directories.
85
'f' is a file entry: As for directory, but its a file. The fingerprint is a
87
'l' is a symlink entry: As for directory, but a symlink. The fingerprint is the
91
--- Format 1 had the following different definition: ---
92
rows = dirname, NULL, basename, NULL, MINIKIND, NULL, fileid_utf8, NULL,
93
WHOLE NUMBER (* size *), NULL, packed stat, NULL, sha1|symlink target,
95
PARENT ROW = NULL, revision_utf8, NULL, MINIKIND, NULL, dirname, NULL,
96
basename, NULL, WHOLE NUMBER (* size *), NULL, "y" | "n", NULL,
99
PARENT ROW's are emitted for every parent that is not in the ghosts details
100
line. That is, if the parents are foo, bar, baz, and the ghosts are bar, then
101
each row will have a PARENT ROW for foo and baz, but not for bar.
104
In any tree, a kind of 'moved' indicates that the fingerprint field
105
(which we treat as opaque data specific to the 'kind' anyway) has the
106
details for the id of this row in that tree.
108
I'm strongly tempted to add a id->path index as well, but I think that
109
where we need id->path mapping; we also usually read the whole file, so
110
I'm going to skip that for the moment, as we have the ability to locate
111
via bisect any path in any tree, and if we lookup things by path, we can
112
accumulate a id->path mapping as we go, which will tend to match what we
115
I plan to implement this asap, so please speak up now to alter/tweak the
116
design - and once we stabilise on this, I'll update the wiki page for
119
The rationale for all this is that we want fast operations for the
120
common case (diff/status/commit/merge on all files) and extremely fast
121
operations for the less common but still occurs a lot status/diff/commit
122
on specific files). Operations on specific files involve a scan for all
123
the children of a path, *in every involved tree*, which the current
124
format did not accommodate.
128
1) Fast end to end use for bzr's top 5 uses cases. (commmit/diff/status/merge/???)
129
2) fall back current object model as needed.
130
3) scale usably to the largest trees known today - say 50K entries. (mozilla
131
is an example of this)
135
Eventually reuse dirstate objects across locks IFF the dirstate file has not
136
been modified, but will require that we flush/ignore cached stat-hit data
137
because we wont want to restat all files on disk just because a lock was
138
acquired, yet we cannot trust the data after the previous lock was released.
140
Memory representation:
141
vector of all directories, and vector of the childen ?
143
root_entrie = (direntry for root, [parent_direntries_for_root]),
145
('', ['data for achild', 'data for bchild', 'data for cchild'])
146
('dir', ['achild', 'cchild', 'echild'])
148
- single bisect to find N subtrees from a path spec
149
- in-order for serialisation - this is 'dirblock' grouping.
150
- insertion of a file '/a' affects only the '/' child-vector, that is, to
151
insert 10K elements from scratch does not generates O(N^2) memoves of a
152
single vector, rather each individual, which tends to be limited to a
153
manageable number. Will scale badly on trees with 10K entries in a
154
single directory. compare with Inventory.InventoryDirectory which has
155
a dictionary for the children. No bisect capability, can only probe for
156
exact matches, or grab all elements and sorta.
157
- Whats the risk of error here? Once we have the base format being processed
158
we should have a net win regardless of optimality. So we are going to
159
go with what seems reasonably.
162
maybe we should do a test profile of these core structure - 10K simulated searches/lookups/etc?
164
Objects for each row?
165
The lifetime of Dirstate objects is current per lock, but see above for
166
possible extensions. The lifetime of a row from a dirstate is expected to be
167
very short in the optimistic case: which we are optimising for. For instance,
168
subtree status will determine from analysis of the disk data what rows need to
169
be examined at all, and will be able to determine from a single row whether
170
that file has altered or not, so we are aiming to process tens of thousands of
171
entries each second within the dirstate context, before exposing anything to
172
the larger codebase. This suggests we want the time for a single file
173
comparison to be < 0.1 milliseconds. That would give us 10000 paths per second
174
processed, and to scale to 100 thousand we'll another order of magnitude to do
175
that. Now, as the lifetime for all unchanged entries is the time to parse, stat
176
the file on disk, and then immediately discard, the overhead of object creation
177
becomes a significant cost.
179
Figures: Creating a tuple from from 3 elements was profiled at 0.0625
180
microseconds, whereas creating a object which is subclassed from tuple was
181
0.500 microseconds, and creating an object with 3 elements and slots was 3
182
microseconds long. 0.1 milliseconds is 100 microseconds, and ideally we'll get
183
down to 10 microseconds for the total processing - having 33% of that be object
184
creation is a huge overhead. There is a potential cost in using tuples within
185
each row which is that the conditional code to do comparisons may be slower
186
than method invocation, but method invocation is known to be slow due to stack
187
frame creation, so avoiding methods in these tight inner loops in unfortunately
188
desirable. We can consider a pyrex version of this with objects in future if
207
import bzrlib.inventory
208
from bzrlib import osutils
209
from bzrlib.osutils import (
217
class _Bisector(object):
218
"""This just keeps track of information as we are bisecting."""
221
class DirState(object):
222
"""Record directory and metadata state for fast access.
224
A dirstate is a specialised data structure for managing local working
225
tree state information. Its not yet well defined whether it is platform
226
specific, and if it is how we detect/parameterise that.
229
_kind_to_minikind = {'absent':'a', 'file':'f', 'directory':'d', 'relocated':'r', 'symlink':'l'}
230
_minikind_to_kind = {'a':'absent', 'f':'file', 'd':'directory', 'l':'symlink', 'r':'relocated'}
231
_to_yesno = {True:'y', False: 'n'} # TODO profile the performance gain
232
# of using int conversion rather than a dict here. AND BLAME ANDREW IF
235
# TODO: jam 20070221 Make sure we handle when there are duplicated records
236
# (like when we remove + add the same path, or we have a rename)
237
# TODO: jam 20070221 Figure out what to do if we have a record that exceeds
238
# the BISECT_PAGE_SIZE. For now, we just have to make it large enough
239
# that we are sure a single record will always fit.
240
BISECT_PAGE_SIZE = 4096
243
IN_MEMORY_UNMODIFIED = 1
244
IN_MEMORY_MODIFIED = 2
246
# A pack_stat (the x's) that is just noise and will never match the output
249
NULL_PARENT_DETAILS = ('a', '', 0, False, '')
251
def __init__(self, path):
252
"""Create a DirState object.
256
:attr _root_entrie: The root row of the directory/file information,
257
- contains the path to / - '', ''
258
- kind of 'directory',
259
- the file id of the root in utf8
262
- and no sha information.
263
:param path: The path at which the dirstate file on disk should live.
265
# _header_state and _dirblock_state represent the current state
266
# of the dirstate metadata and the per-row data respectiely.
267
# NOT_IN_MEMORY indicates that no data is in memory
268
# IN_MEMORY_UNMODIFIED indicates that what we have in memory
269
# is the same as is on disk
270
# IN_MEMORY_MODIFIED indicates that we have a modified version
271
# of what is on disk.
272
# In future we will add more granularity, for instance _dirblock_state
273
# will probably support partially-in-memory as a separate variable,
274
# allowing for partially-in-memory unmodified and partially-in-memory
276
self._header_state = DirState.NOT_IN_MEMORY
277
self._dirblock_state = DirState.NOT_IN_MEMORY
281
self._state_file = None
282
self._filename = path
283
self._lock_token = None
284
self._end_of_header = None
285
self._bisect_page_size = DirState.BISECT_PAGE_SIZE
287
def add(self, path, file_id, kind, stat, link_or_sha1):
288
"""Add a path to be tracked.
290
:param path: The path within the dirstate - '' is the root, 'foo' is the
291
path foo within the root, 'foo/bar' is the path bar within foo
293
:param file_id: The file id of the path being added.
294
:param kind: The kind of the path.
295
:param stat: The output of os.lstate for the path.
296
:param link_or_sha1: The sha value of the file, or the target of a
297
symlink. '' for directories.
300
# find the block its in.
301
# find the location in the block.
302
# check its not there
304
#------- copied from bzrlib.inventory.make_entry
305
# --- normalized_filename wants a unicode basename only, so get one.
306
dirname, basename = os.path.split(path)
307
# we dont import normalized_filename directly because we want to be
308
# able to change the implementation at runtime for tests.
309
norm_name, can_access = osutils.normalized_filename(basename)
310
if norm_name != basename:
314
raise errors.InvalidNormalization(path)
315
# now that we've normalised, we need the correct utf8 path and
316
# dirname and basename elements. This single encode and split should be
317
# faster than three separate encodes.
318
utf8path = (dirname + '/' + basename).strip('/').encode('utf8')
319
dirname, basename = os.path.split(utf8path)
320
assert file_id.__class__ == str, \
321
"must be a utf8 file_id not %s" % (type(file_id))
322
entry_key = (dirname, basename, file_id)
323
self._read_dirblocks_if_needed()
324
block_index, present = self._find_block_index_from_key(entry_key)
326
# TODO: This test is not complete - an empty directory, or a
327
# directory for a parent tree will fool it.
328
# some parent path has not been added - its an error to add this
330
raise errors.NotVersionedError(path, str(self))
331
block = self._dirblocks[block_index][1]
334
packed_stat = DirState.NULLSTAT
337
packed_stat = pack_stat(stat)
338
parent_info = self._empty_parent_info()
339
minikind = DirState._kind_to_minikind[kind]
341
entry_data = entry_key, [
342
(minikind, link_or_sha1, size, False, packed_stat),
344
elif kind == 'directory':
345
entry_data = entry_key, [
346
(minikind, '', 0, False, packed_stat),
348
elif kind == 'symlink':
349
entry_data = entry_key, [
350
(minikind, link_or_sha1, size, False, packed_stat),
353
raise errors.BzrError('unknown kind %r' % kind)
354
entry_index, present = self._find_entry_index(entry_key, block)
355
assert not present, "basename %r already added" % basename
356
block.insert(entry_index, entry_data)
358
if kind == 'directory':
359
# insert a new dirblock
360
self._ensure_block(block_index, entry_index, utf8path)
361
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
363
def _bisect(self, dir_name_list):
364
"""Bisect through the disk structure for specific rows.
366
:param dir_name_list: A list of (dir, name) pairs.
367
:return: A dict mapping (dir, name) => entry for found entries. Missing
368
entries will not be in the map.
370
self._requires_lock()
371
# We need the file pointer to be right after the initial header block
372
self._read_header_if_needed()
373
# If _dirblock_state was in memory, we should just return info from
374
# there, this function is only meant to handle when we want to read
376
assert self._dirblock_state == DirState.NOT_IN_MEMORY
378
# The disk representation is generally info + '\0\n\0' at the end. But
379
# for bisecting, it is easier to treat this as '\0' + info + '\0\n'
380
# Because it means we can sync on the '\n'
381
state_file = self._state_file
382
file_size = os.fstat(state_file.fileno()).st_size
383
# We end up with 2 extra fields, we should have a trailing '\n' to
384
# ensure that we read the whole record, and we should have a precursur
385
# '' which ensures that we start after the previous '\n'
386
entry_field_count = self._fields_per_entry() + 1
388
low = self._end_of_header
389
high = file_size - 1 # Ignore the final '\0'
390
# Map from (dir, name) => entry
393
# Avoid infinite seeking
394
max_count = 30*len(dir_name_list)
396
# pending is a list of places to look.
397
# each entry is a tuple of low, high, dir_names
398
# low -> the first byte offset to read (inclusive)
399
# high -> the last byte offset (inclusive)
400
# dir_names -> The list of (dir, name) pairs that should be found in
401
# the [low, high] range
402
pending = [(low, high, dir_name_list)]
404
page_size = self._bisect_page_size
406
fields_to_entry = self._get_fields_to_entry()
409
low, high, cur_files = pending.pop()
411
if not cur_files or low >= high:
416
if count > max_count:
417
raise errors.BzrError('Too many seeks, most likely a bug.')
419
mid = max(low, (low+high-page_size)/2)
422
# limit the read size, so we don't end up reading data that we have
424
read_size = min(page_size, (high-mid)+1)
425
block = state_file.read(read_size)
428
entries = block.split('\n')
431
# We didn't find a '\n', so we cannot have found any records.
432
# So put this range back and try again. But we know we have to
433
# increase the page size, because a single read did not contain
434
# a record break (so records must be larger than page_size)
436
pending.append((low, high, cur_files))
439
# Check the first and last entries, in case they are partial, or if
440
# we don't care about the rest of this page
442
first_fields = entries[0].split('\0')
443
if len(first_fields) < entry_field_count:
444
# We didn't get the complete first entry
445
# so move start, and grab the next, which
446
# should be a full entry
447
start += len(entries[0])+1
448
first_fields = entries[1].split('\0')
451
if len(first_fields) <= 2:
452
# We didn't even get a filename here... what do we do?
453
# Try a large page size and repeat this query
455
pending.append((low, high, cur_files))
458
# Find what entries we are looking for, which occur before and
459
# after this first record.
461
first_dir_name = (first_fields[1], first_fields[2])
462
first_loc = bisect.bisect_left(cur_files, first_dir_name)
464
# These exist before the current location
465
pre = cur_files[:first_loc]
466
# These occur after the current location, which may be in the
467
# data we read, or might be after the last entry
468
post = cur_files[first_loc:]
470
if post and len(first_fields) >= entry_field_count:
471
# We have files after the first entry
473
# Parse the last entry
474
last_entry_num = len(entries)-1
475
last_fields = entries[last_entry_num].split('\0')
476
if len(last_fields) < entry_field_count:
477
# The very last hunk was not complete,
478
# read the previous hunk
479
after = mid + len(block) - len(entries[-1])
481
last_fields = entries[last_entry_num].split('\0')
483
after = mid + len(block)
485
last_dir_name = (last_fields[1], last_fields[2])
486
last_loc = bisect.bisect_right(post, last_dir_name)
488
middle_files = post[:last_loc]
489
post = post[last_loc:]
492
# We have files that should occur in this block
493
# (>= first, <= last)
494
# Either we will find them here, or we can mark them as
497
if middle_files[0] == first_dir_name:
498
# We might need to go before this location
499
pre.append(first_dir_name)
500
if middle_files[-1] == last_dir_name:
501
post.insert(0, last_dir_name)
503
# Find out what paths we have
504
paths = {first_dir_name:[first_fields]}
505
# last_dir_name might == first_dir_name so we need to be
506
# careful if we should append rather than overwrite
507
if last_entry_num != first_entry_num:
508
paths.setdefault(last_dir_name, []).append(last_fields)
509
for num in xrange(first_entry_num+1, last_entry_num):
510
# TODO: jam 20070223 We are already splitting here, so
511
# shouldn't we just split the whole thing rather
512
# than doing the split again in add_one_record?
513
fields = entries[num].split('\0')
514
dir_name = (fields[1], fields[2])
515
paths.setdefault(dir_name, []).append(fields)
517
for dir_name in middle_files:
518
for fields in paths.get(dir_name, []):
519
# offset by 1 because of the opening '\0'
520
# consider changing fields_to_entry to avoid the
522
entry = fields_to_entry(fields[1:])
523
found.setdefault(dir_name, []).append(entry)
525
# Now we have split up everything into pre, middle, and post, and
526
# we have handled everything that fell in 'middle'.
527
# We add 'post' first, so that we prefer to seek towards the
528
# beginning, so that we will tend to go as early as we need, and
529
# then only seek forward after that.
531
pending.append((after, high, post))
533
pending.append((low, start-1, pre))
535
# Consider that we may want to return the directory entries in sorted
536
# order. For now, we just return them in whatever order we found them,
537
# and leave it up to the caller if they care if it is ordered or not.
540
def _bisect_dirblocks(self, dir_list):
541
"""Bisect through the disk structure to find entries in given dirs.
543
_bisect_dirblocks is meant to find the contents of directories, which
544
differs from _bisect, which only finds individual entries.
546
:param dir_list: An sorted list of directory names ['', 'dir', 'foo'].
547
:return: A map from dir => entries_for_dir
549
# TODO: jam 20070223 A lot of the bisecting logic could be shared
550
# between this and _bisect. It would require parameterizing the
551
# inner loop with a function, though. We should evaluate the
552
# performance difference.
553
self._requires_lock()
554
# We need the file pointer to be right after the initial header block
555
self._read_header_if_needed()
556
# If _dirblock_state was in memory, we should just return info from
557
# there, this function is only meant to handle when we want to read
559
assert self._dirblock_state == DirState.NOT_IN_MEMORY
561
# The disk representation is generally info + '\0\n\0' at the end. But
562
# for bisecting, it is easier to treat this as '\0' + info + '\0\n'
563
# Because it means we can sync on the '\n'
564
state_file = self._state_file
565
file_size = os.fstat(state_file.fileno()).st_size
566
# We end up with 2 extra fields, we should have a trailing '\n' to
567
# ensure that we read the whole record, and we should have a precursur
568
# '' which ensures that we start after the previous '\n'
569
entry_field_count = self._fields_per_entry() + 1
571
low = self._end_of_header
572
high = file_size - 1 # Ignore the final '\0'
573
# Map from dir => entry
576
# Avoid infinite seeking
577
max_count = 30*len(dir_list)
579
# pending is a list of places to look.
580
# each entry is a tuple of low, high, dir_names
581
# low -> the first byte offset to read (inclusive)
582
# high -> the last byte offset (inclusive)
583
# dirs -> The list of directories that should be found in
584
# the [low, high] range
585
pending = [(low, high, dir_list)]
587
page_size = self._bisect_page_size
589
fields_to_entry = self._get_fields_to_entry()
592
low, high, cur_dirs = pending.pop()
594
if not cur_dirs or low >= high:
599
if count > max_count:
600
raise errors.BzrError('Too many seeks, most likely a bug.')
602
mid = max(low, (low+high-page_size)/2)
605
# limit the read size, so we don't end up reading data that we have
607
read_size = min(page_size, (high-mid)+1)
608
block = state_file.read(read_size)
611
entries = block.split('\n')
614
# We didn't find a '\n', so we cannot have found any records.
615
# So put this range back and try again. But we know we have to
616
# increase the page size, because a single read did not contain
617
# a record break (so records must be larger than page_size)
619
pending.append((low, high, cur_dirs))
622
# Check the first and last entries, in case they are partial, or if
623
# we don't care about the rest of this page
625
first_fields = entries[0].split('\0')
626
if len(first_fields) < entry_field_count:
627
# We didn't get the complete first entry
628
# so move start, and grab the next, which
629
# should be a full entry
630
start += len(entries[0])+1
631
first_fields = entries[1].split('\0')
634
if len(first_fields) <= 1:
635
# We didn't even get a dirname here... what do we do?
636
# Try a large page size and repeat this query
638
pending.append((low, high, cur_dirs))
641
# Find what entries we are looking for, which occur before and
642
# after this first record.
644
first_dir = first_fields[1]
645
first_loc = bisect.bisect_left(cur_dirs, first_dir)
647
# These exist before the current location
648
pre = cur_dirs[:first_loc]
649
# These occur after the current location, which may be in the
650
# data we read, or might be after the last entry
651
post = cur_dirs[first_loc:]
653
if post and len(first_fields) >= entry_field_count:
654
# We have records to look at after the first entry
656
# Parse the last entry
657
last_entry_num = len(entries)-1
658
last_fields = entries[last_entry_num].split('\0')
659
if len(last_fields) < entry_field_count:
660
# The very last hunk was not complete,
661
# read the previous hunk
662
after = mid + len(block) - len(entries[-1])
664
last_fields = entries[last_entry_num].split('\0')
666
after = mid + len(block)
668
last_dir = last_fields[1]
669
last_loc = bisect.bisect_right(post, last_dir)
671
middle_files = post[:last_loc]
672
post = post[last_loc:]
675
# We have files that should occur in this block
676
# (>= first, <= last)
677
# Either we will find them here, or we can mark them as
680
if middle_files[0] == first_dir:
681
# We might need to go before this location
682
pre.append(first_dir)
683
if middle_files[-1] == last_dir:
684
post.insert(0, last_dir)
686
# Find out what paths we have
687
paths = {first_dir:[first_fields]}
688
# last_dir might == first_dir so we need to be
689
# careful if we should append rather than overwrite
690
if last_entry_num != first_entry_num:
691
paths.setdefault(last_dir, []).append(last_fields)
692
for num in xrange(first_entry_num+1, last_entry_num):
693
# TODO: jam 20070223 We are already splitting here, so
694
# shouldn't we just split the whole thing rather
695
# than doing the split again in add_one_record?
696
fields = entries[num].split('\0')
697
paths.setdefault(fields[1], []).append(fields)
699
for cur_dir in middle_files:
700
for fields in paths.get(cur_dir, []):
701
# offset by 1 because of the opening '\0'
702
# consider changing fields_to_entry to avoid the
704
entry = fields_to_entry(fields[1:])
705
found.setdefault(cur_dir, []).append(entry)
707
# Now we have split up everything into pre, middle, and post, and
708
# we have handled everything that fell in 'middle'.
709
# We add 'post' first, so that we prefer to seek towards the
710
# beginning, so that we will tend to go as early as we need, and
711
# then only seek forward after that.
713
pending.append((after, high, post))
715
pending.append((low, start-1, pre))
719
def _bisect_recursive(self, dir_name_list):
720
"""Bisect for entries for all paths and their children.
722
This will use bisect to find all records for the supplied paths. It
723
will then continue to bisect for any records which are marked as
724
directories. (and renames?)
726
:param paths: A sorted list of (dir, name) pairs
727
eg: [('', 'a'), ('', 'f'), ('a/b', 'c')]
728
:return: A dictionary mapping (dir, name, file_id) => [tree_info]
730
# Map from (dir, name, file_id) => [tree_info]
733
found_dir_names = set()
735
# Directories that have been read
736
processed_dirs = set()
737
# Get the ball rolling with the first bisect for all entries.
738
newly_found = self._bisect(dir_name_list)
741
# Directories that need to be read
743
paths_to_search = set()
744
for entry_list in newly_found.itervalues():
745
for dir_name_id, trees_info in entry_list:
746
found[dir_name_id] = trees_info
747
found_dir_names.add(dir_name_id[:2])
749
for tree_info in trees_info:
750
minikind = tree_info[0]
753
# We already processed this one as a directory,
754
# we don't need to do the extra work again.
756
subdir, name, file_id = dir_name_id
757
path = osutils.pathjoin(subdir, name)
759
if path not in processed_dirs:
760
pending_dirs.add(path)
761
elif minikind == 'r':
762
# Rename, we need to directly search the target
763
# which is contained in the fingerprint column
764
dir_name = osutils.split(tree_info[1])
765
if dir_name[0] in pending_dirs:
766
# This entry will be found in the dir search
768
# TODO: We need to check if this entry has
769
# already been found. Otherwise we might be
770
# hitting infinite recursion.
771
if dir_name not in found_dir_names:
772
paths_to_search.add(dir_name)
773
# Now we have a list of paths to look for directly, and
774
# directory blocks that need to be read.
775
# newly_found is mixing the keys between (dir, name) and path
776
# entries, but that is okay, because we only really care about the
778
newly_found = self._bisect(sorted(paths_to_search))
779
newly_found.update(self._bisect_dirblocks(sorted(pending_dirs)))
780
processed_dirs.update(pending_dirs)
783
def _empty_parent_info(self):
784
return [DirState.NULL_PARENT_DETAILS] * (len(self._parents) -
787
def _ensure_block(self, parent_block_index, parent_row_index, dirname):
788
"""Enssure a block for dirname exists.
790
This function exists to let callers which know that there is a
791
directory dirname ensure that the block for it exists. This block can
792
fail to exist because of demand loading, or because a directory had no
793
children. In either case it is not an error. It is however an error to
794
call this if there is no parent entry for the directory, and thus the
795
function requires the coordinates of such an entry to be provided.
797
The root row is special cased and can be indicated with a parent block
800
:param parent_block_index: The index of the block in which dirname's row
802
:param parent_row_index: The index in the parent block where the row
804
:param dirname: The utf8 dirname to ensure there is a block for.
805
:return: The index for the block.
808
# the basename of the directory must be the end of its full name.
809
if not (parent_block_index == -1 and
810
parent_block_index == -1 and dirname == ''):
811
assert dirname.endswith(
812
self._dirblocks[parent_block_index][1][parent_row_index][0][1])
813
block_index, present = self._find_block_index_from_key((dirname, '', ''))
815
## In future, when doing partial parsing, this should load and
816
# populate the entire block.
817
self._dirblocks.insert(block_index, (dirname, []))
820
def _entries_to_current_state(self, new_entries):
821
"""Load new_entries into self.dirblocks.
823
Process new_entries into the current state object, making them the active
826
:param new_entries: A sorted list of entries. This function does not sort
827
to prevent unneeded overhead when callers have a sorted list already.
830
assert new_entries[0][0][0:2] == ('', ''), \
831
"Missing root row %r" % new_entries[0][0]
832
# The two blocks here are deliberate: the root block and the
833
# contents-of-root block.
834
self._dirblocks = [('', []), ('', [])]
835
current_block = self._dirblocks[0][1]
838
append_entry = current_block.append
839
for entry in new_entries:
840
if entry[0][0] != current_dirname:
841
# new block - different dirname
843
current_dirname = entry[0][0]
844
self._dirblocks.append((current_dirname, current_block))
845
append_entry = current_block.append
846
# append the entry to the current block
848
self._split_root_dirblock_into_contents()
850
def _split_root_dirblock_into_contents(self):
851
"""Split the root dirblocks into root and contents-of-root.
853
After parsing by path, we end up with root entries and contents-of-root
854
entries in the same block. This loop splits them out again.
856
# The above loop leaves the "root block" entries mixed with the
857
# "contents-of-root block". But we don't want an if check on
858
# all entries, so instead we just fix it up here.
859
assert self._dirblocks[1] == ('', [])
861
contents_of_root_block = []
862
for entry in self._dirblocks[0][1]:
863
if not entry[0][1]: # This is a root entry
864
root_block.append(entry)
866
contents_of_root_block.append(entry)
867
self._dirblocks[0] = ('', root_block)
868
self._dirblocks[1] = ('', contents_of_root_block)
870
def _entry_to_line(self, entry):
871
"""Serialize entry to a NULL delimited line ready for _get_output_lines.
873
:param entry: An entry_tuple as defined in the module docstring.
875
entire_entry = list(entry[0])
876
for tree_number, tree_data in enumerate(entry[1]):
877
# (minikind, fingerprint, size, executable, tree_specific_string)
878
entire_entry.extend(tree_data)
879
# 3 for the key, 5 for the fields per tree.
880
tree_offset = 3 + tree_number * 5
882
entire_entry[tree_offset + 0] = tree_data[0]
884
entire_entry[tree_offset + 2] = str(tree_data[2])
886
entire_entry[tree_offset + 3] = DirState._to_yesno[tree_data[3]]
887
return '\0'.join(entire_entry)
889
def _fields_per_entry(self):
890
"""How many null separated fields should be in each entry row.
892
Each line now has an extra '\n' field which is not used
893
so we just skip over it
896
+ number of fields per tree_data (5) * tree count
899
tree_count = 1 + self._num_present_parents()
900
return 3 + 5 * tree_count + 1
902
def _find_block(self, key, add_if_missing=False):
903
"""Return the block that key should be present in.
905
:param key: A dirstate entry key.
906
:return: The block tuple.
908
block_index, present = self._find_block_index_from_key(key)
910
if not add_if_missing:
911
# check to see if key is versioned itself - we might want to
912
# add it anyway, because dirs with no entries dont get a
913
# dirblock at parse time.
914
# This is an uncommon branch to take: most dirs have children,
915
# and most code works with versioned paths.
916
parent_base, parent_name = os.path.split(key[0])
917
if not self._get_block_entry_index(parent_base, parent_name, 0)[3]:
918
# some parent path has not been added - its an error to add
920
raise errors.NotVersionedError(key[0:2], str(self))
921
self._dirblocks.insert(block_index, (key[0], []))
922
return self._dirblocks[block_index]
924
def _find_block_index_from_key(self, key):
925
"""Find the dirblock index for a key.
927
:return: The block index, True if the block for the key is present.
929
if key[0:2] == ('', ''):
931
block_index = bisect.bisect_left(self._dirblocks, (key[0], []), 1)
932
# _right returns one-past-where-key is so we have to subtract
933
# one to use it. we use _right here because there are two
934
# '' blocks - the root, and the contents of root
935
# we always have a minimum of 2 in self._dirblocks: root and
936
# root-contents, and for '', we get 2 back, so this is
937
# simple and correct:
938
present = (block_index < len(self._dirblocks) and
939
self._dirblocks[block_index][0] == key[0])
940
return block_index, present
942
def _find_entry_index(self, key, block):
943
"""Find the entry index for a key in a block.
945
:return: The entry index, True if the entry for the key is present.
947
entry_index = bisect.bisect_left(block, (key, []))
948
present = (entry_index < len(block) and
949
block[entry_index][0] == key)
950
return entry_index, present
953
def from_tree(tree, dir_state_filename):
954
"""Create a dirstate from a bzr Tree.
956
:param tree: The tree which should provide parent information and
958
:return: a DirState object which is currently locked for writing.
959
(it was locked by DirState.initialize)
961
result = DirState.initialize(dir_state_filename)
965
parent_ids = tree.get_parent_ids()
966
num_parents = len(parent_ids)
968
for parent_id in parent_ids:
969
parent_tree = tree.branch.repository.revision_tree(parent_id)
970
parent_trees.append((parent_id, parent_tree))
971
parent_tree.lock_read()
972
result.set_parent_trees(parent_trees, [])
973
result.set_state_from_inventory(tree.inventory)
975
for revid, parent_tree in parent_trees:
979
# The caller won't have a chance to unlock this, so make sure we
985
def get_ghosts(self):
986
"""Return a list of the parent tree revision ids that are ghosts."""
987
self._read_header_if_needed()
991
"""Serialise the entire dirstate to a sequence of lines."""
992
if (self._header_state == DirState.IN_MEMORY_UNMODIFIED and
993
self._dirblock_state == DirState.IN_MEMORY_UNMODIFIED):
994
# read whats on disk.
995
self._state_file.seek(0)
996
return self._state_file.readlines()
998
lines.append(self._get_parents_line(self.get_parent_ids()))
999
lines.append(self._get_ghosts_line(self._ghosts))
1000
# append the root line which is special cased
1001
lines.extend(map(self._entry_to_line, self._iter_entries()))
1002
return self._get_output_lines(lines)
1004
def _get_ghosts_line(self, ghost_ids):
1005
"""Create a line for the state file for ghost information."""
1006
return '\0'.join([str(len(ghost_ids))] + ghost_ids)
1008
def _get_parents_line(self, parent_ids):
1009
"""Create a line for the state file for parents information."""
1010
return '\0'.join([str(len(parent_ids))] + parent_ids)
1012
def _get_fields_to_entry(self):
1013
"""Get a function which converts entry fields into a entry record.
1015
This handles size and executable, as well as parent records.
1017
:return: A function which takes a list of fields, and returns an
1018
appropriate record for storing in memory.
1020
# This is intentionally unrolled for performance
1021
num_present_parents = self._num_present_parents()
1022
if num_present_parents == 0:
1023
def fields_to_entry_0_parents(fields, _int=int):
1024
path_name_file_id_key = (fields[0], fields[1], fields[2])
1025
return (path_name_file_id_key, [
1027
fields[3], # minikind
1028
fields[4], # fingerprint
1029
_int(fields[5]), # size
1030
fields[6] == 'y', # executable
1031
fields[7], # packed_stat or revision_id
1033
return fields_to_entry_0_parents
1034
elif num_present_parents == 1:
1035
def fields_to_entry_1_parent(fields, _int=int):
1036
path_name_file_id_key = (fields[0], fields[1], fields[2])
1037
return (path_name_file_id_key, [
1039
fields[3], # minikind
1040
fields[4], # fingerprint
1041
_int(fields[5]), # size
1042
fields[6] == 'y', # executable
1043
fields[7], # packed_stat or revision_id
1046
fields[8], # minikind
1047
fields[9], # fingerprint
1048
_int(fields[10]), # size
1049
fields[11] == 'y', # executable
1050
fields[12], # packed_stat or revision_id
1053
return fields_to_entry_1_parent
1054
elif num_present_parents == 2:
1055
def fields_to_entry_2_parents(fields, _int=int):
1056
path_name_file_id_key = (fields[0], fields[1], fields[2])
1057
return (path_name_file_id_key, [
1059
fields[3], # minikind
1060
fields[4], # fingerprint
1061
_int(fields[5]), # size
1062
fields[6] == 'y', # executable
1063
fields[7], # packed_stat or revision_id
1066
fields[8], # minikind
1067
fields[9], # fingerprint
1068
_int(fields[10]), # size
1069
fields[11] == 'y', # executable
1070
fields[12], # packed_stat or revision_id
1073
fields[13], # minikind
1074
fields[14], # fingerprint
1075
_int(fields[15]), # size
1076
fields[16] == 'y', # executable
1077
fields[17], # packed_stat or revision_id
1080
return fields_to_entry_2_parents
1082
def fields_to_entry_n_parents(fields, _int=int):
1083
path_name_file_id_key = (fields[0], fields[1], fields[2])
1084
trees = [(fields[cur], # minikind
1085
fields[cur+1], # fingerprint
1086
_int(fields[cur+2]), # size
1087
fields[cur+3] == 'y', # executable
1088
fields[cur+4], # stat or revision_id
1089
) for cur in xrange(3, len(fields)-1, 5)]
1090
return path_name_file_id_key, trees
1091
return fields_to_entry_n_parents
1093
def get_parent_ids(self):
1094
"""Return a list of the parent tree ids for the directory state."""
1095
self._read_header_if_needed()
1096
return list(self._parents)
1098
def _get_block_entry_index(self, dirname, basename, tree_index):
1099
"""Get the coordinates for a path in the state structure.
1101
:param dirname: The utf8 dirname to lookup.
1102
:param basename: The utf8 basename to lookup.
1103
:param tree_index: The index of the tree for which this lookup should
1105
:return: A tuple describing where the path is located, or should be
1106
inserted. The tuple contains four fields: the block index, the row
1107
index, anda two booleans are True when the directory is present, and
1108
when the entire path is present. There is no guarantee that either
1109
coordinate is currently reachable unless the found field for it is
1110
True. For instance, a directory not present in the searched tree
1111
may be returned with a value one greater than the current highest
1112
block offset. The directory present field will always be True when
1113
the path present field is True. The directory present field does
1114
NOT indicate that the directory is present in the searched tree,
1115
rather it indicates that there are at least some files in some
1118
self._read_dirblocks_if_needed()
1119
key = dirname, basename, ''
1120
block_index, present = self._find_block_index_from_key(key)
1122
# no such directory - return the dir index and 0 for the row.
1123
return block_index, 0, False, False
1124
block = self._dirblocks[block_index][1] # access the entries only
1125
entry_index, present = self._find_entry_index(key, block)
1126
# linear search through present entries at this path to find the one
1128
while entry_index < len(block) and block[entry_index][0][1] == basename:
1129
if block[entry_index][1][tree_index][0] not in \
1130
('a', 'r'): # absent, relocated
1131
return block_index, entry_index, True, True
1133
return block_index, entry_index, True, False
1135
def _get_entry(self, tree_index, fileid_utf8=None, path_utf8=None):
1136
"""Get the dirstate entry for path in tree tree_index
1138
If either file_id or path is supplied, it is used as the key to lookup.
1139
If both are supplied, the fastest lookup is used, and an error is
1140
raised if they do not both point at the same row.
1142
:param tree_index: The index of the tree we wish to locate this path
1143
in. If the path is present in that tree, the entry containing its
1144
details is returned, otherwise (None, None) is returned
1145
:param fileid_utf8: A utf8 file_id to look up.
1146
:param path_utf8: An utf8 path to be looked up.
1147
:return: The dirstate entry tuple for path, or (None, None)
1149
self._read_dirblocks_if_needed()
1150
if path_utf8 is not None:
1151
assert path_utf8.__class__ == str, 'path_utf8 is not a str: %s %s' % (type(path_utf8), path_utf8)
1152
# path lookups are faster
1153
dirname, basename = os.path.split(path_utf8)
1154
block_index, entry_index, dir_present, file_present = \
1155
self._get_block_entry_index(dirname, basename, tree_index)
1156
if not file_present:
1158
entry = self._dirblocks[block_index][1][entry_index]
1159
assert entry[0][2] and entry[1][tree_index][0] not in ('a', 'r'), 'unversioned entry?!?!'
1161
if entry[0][2] != fileid_utf8:
1162
raise BzrError('integrity error ? : mismatching tree_index, file_id and path')
1165
for entry in self._iter_entries():
1166
if entry[0][2] == fileid_utf8:
1167
if entry[1][tree_index][0] == 'r': # relocated
1168
# look up the real location directly by path
1169
return self._get_entry(tree_index,
1170
fileid_utf8=fileid_utf8,
1171
path_utf8=entry[1][tree_index][1])
1172
if entry[1][tree_index][0] == 'a': # absent
1173
# not in the tree at all.
1179
def initialize(path):
1180
"""Create a new dirstate on path.
1182
The new dirstate will be an empty tree - that is it has no parents,
1183
and only a root node - which has id ROOT_ID.
1185
The object will be write locked when returned to the caller,
1186
unless there was an exception in the writing, in which case it
1189
:param path: The name of the file for the dirstate.
1190
:return: A DirState object.
1192
# This constructs a new DirState object on a path, sets the _state_file
1193
# to a new empty file for that path. It then calls _set_data() with our
1194
# stock empty dirstate information - a root with ROOT_ID, no children,
1195
# and no parents. Finally it calls save() to ensure that this data will
1197
result = DirState(path)
1198
# root dir and root dir contents with no children.
1199
empty_tree_dirblocks = [('', []), ('', [])]
1200
# a new root directory, with a NULLSTAT.
1201
empty_tree_dirblocks[0][1].append(
1202
(('', '', bzrlib.inventory.ROOT_ID), [
1203
('d', '', 0, False, DirState.NULLSTAT),
1207
result._set_data([], empty_tree_dirblocks)
1214
def _inv_entry_to_details(self, inv_entry):
1215
"""Convert an inventory entry (from a revision tree) to state details.
1217
:param inv_entry: An inventory entry whose sha1 and link targets can be
1218
relied upon, and which has a revision set.
1219
:return: A details tuple - the details for a single tree at a path +
1222
kind = inv_entry.kind
1223
minikind = DirState._kind_to_minikind[kind]
1224
tree_data = inv_entry.revision
1225
assert len(tree_data) > 0, 'empty revision for the inv_entry.'
1226
if kind == 'directory':
1230
elif kind == 'symlink':
1231
fingerprint = inv_entry.symlink_target or ''
1234
elif kind == 'file':
1235
fingerprint = inv_entry.text_sha1 or ''
1236
size = inv_entry.text_size or 0
1237
executable = inv_entry.executable
1240
return (minikind, fingerprint, size, executable, tree_data)
1242
def _iter_entries(self):
1243
"""Iterate over all the entries in the dirstate.
1245
Each yelt item is an entry in the standard format described in the
1246
docstring of bzrlib.dirstate.
1248
self._read_dirblocks_if_needed()
1249
for directory in self._dirblocks:
1250
for entry in directory[1]:
1253
def _get_id_index(self):
1254
"""Get an id index of self._dirblocks."""
1256
for key, tree_details in self._iter_entries():
1257
id_index.setdefault(key[2], set()).add(key)
1260
def _get_output_lines(self, lines):
1261
"""format lines for final output.
1263
:param lines: A sequece of lines containing the parents list and the
1266
output_lines = ['#bazaar dirstate flat format 2\n']
1267
lines.append('') # a final newline
1268
inventory_text = '\0\n\0'.join(lines)
1269
output_lines.append('adler32: %s\n' % (zlib.adler32(inventory_text),))
1270
# -3, 1 for num parents, 1 for ghosts, 1 for final newline
1271
num_entries = len(lines)-3
1272
output_lines.append('num_entries: %s\n' % (num_entries,))
1273
output_lines.append(inventory_text)
1276
def _make_deleted_row(self, fileid_utf8, parents):
1277
"""Return a deleted for for fileid_utf8."""
1278
return ('/', 'RECYCLED.BIN', 'file', fileid_utf8, 0, DirState.NULLSTAT,
1281
def _num_present_parents(self):
1282
"""The number of parent entries in each record row."""
1283
return len(self._parents) - len(self._ghosts)
1287
"""Construct a DirState on the file at path path.
1289
:return: An unlocked DirState object, associated with the given path.
1291
result = DirState(path)
1294
def _read_dirblocks_if_needed(self):
1295
"""Read in all the dirblocks from the file if they are not in memory.
1297
This populates self._dirblocks, and sets self._dirblock_state to
1298
IN_MEMORY_UNMODIFIED. It is not currently ready for incremental block
1301
self._read_header_if_needed()
1302
if self._dirblock_state == DirState.NOT_IN_MEMORY:
1303
# the _state_file pointer will be positioned at the start of the
1305
text = self._state_file.read()
1306
# TODO: check the adler checksums. adler_measured = zlib.adler32(text)
1308
fields = text.split('\0')
1309
# Remove the last blank entry
1310
trailing = fields.pop()
1311
assert trailing == ''
1312
# consider turning fields into a tuple.
1314
# skip the first field which is the trailing null from the header.
1316
# Each line now has an extra '\n' field which is not used
1317
# so we just skip over it
1319
# 3 fields for the key
1320
# + number of fields per tree_data (5) * tree count
1322
num_present_parents = self._num_present_parents()
1323
tree_count = 1 + num_present_parents
1324
entry_size = self._fields_per_entry()
1325
expected_field_count = entry_size * self._num_entries
1326
if len(fields) - cur > expected_field_count:
1327
fields = fields[:expected_field_count + cur]
1328
trace.mutter('Unexpectedly long dirstate field count!')
1329
print "XXX: incorrectly truncated dirstate file bug triggered."
1330
field_count = len(fields)
1331
# this checks our adjustment, and also catches file too short.
1332
assert field_count - cur == expected_field_count, \
1333
'field count incorrect %s != %s, entry_size=%s, '\
1334
'num_entries=%s fields=%r' % (
1335
field_count - cur, expected_field_count, entry_size,
1336
self._num_entries, fields)
1338
if num_present_parents == 1:
1339
# Bind external functions to local names
1341
# We access all fields in order, so we can just iterate over
1342
# them. Grab an straight iterator over the fields. (We use an
1343
# iterator because we don't want to do a lot of additions, nor
1344
# do we want to do a lot of slicing)
1345
next = iter(fields).next
1346
# Move the iterator to the current position
1347
for x in xrange(cur):
1349
# The two blocks here are deliberate: the root block and the
1350
# contents-of-root block.
1351
self._dirblocks = [('', []), ('', [])]
1352
current_block = self._dirblocks[0][1]
1353
current_dirname = ''
1354
append_entry = current_block.append
1355
for count in xrange(self._num_entries):
1359
if dirname != current_dirname:
1360
# new block - different dirname
1362
current_dirname = dirname
1363
self._dirblocks.append((current_dirname, current_block))
1364
append_entry = current_block.append
1365
# we know current_dirname == dirname, so re-use it to avoid
1366
# creating new strings
1367
entry = ((current_dirname, name, file_id),
1370
next(), # fingerprint
1371
_int(next()), # size
1372
next() == 'y', # executable
1373
next(), # packed_stat or revision_id
1377
next(), # fingerprint
1378
_int(next()), # size
1379
next() == 'y', # executable
1380
next(), # packed_stat or revision_id
1384
assert trailing == '\n'
1385
# append the entry to the current block
1387
self._split_root_dirblock_into_contents()
1389
fields_to_entry = self._get_fields_to_entry()
1390
entries = [fields_to_entry(fields[pos:pos+entry_size])
1391
for pos in xrange(cur, field_count, entry_size)]
1392
self._entries_to_current_state(entries)
1393
self._dirblock_state = DirState.IN_MEMORY_UNMODIFIED
1395
def _read_header(self):
1396
"""This reads in the metadata header, and the parent ids.
1398
After reading in, the file should be positioned at the null
1399
just before the start of the first record in the file.
1401
:return: (expected adler checksum, number of entries, parent list)
1403
self._read_prelude()
1404
parent_line = self._state_file.readline()
1405
info = parent_line.split('\0')
1406
num_parents = int(info[0])
1407
assert num_parents == len(info)-2, 'incorrect parent info line'
1408
self._parents = info[1:-1]
1410
ghost_line = self._state_file.readline()
1411
info = ghost_line.split('\0')
1412
num_ghosts = int(info[1])
1413
assert num_ghosts == len(info)-3, 'incorrect ghost info line'
1414
self._ghosts = info[2:-1]
1415
self._header_state = DirState.IN_MEMORY_UNMODIFIED
1416
self._end_of_header = self._state_file.tell()
1418
def _read_header_if_needed(self):
1419
"""Read the header of the dirstate file if needed."""
1420
# inline this as it will be called a lot
1421
if not self._lock_token:
1422
raise errors.ObjectNotLocked(self)
1423
if self._header_state == DirState.NOT_IN_MEMORY:
1426
def _read_prelude(self):
1427
"""Read in the prelude header of the dirstate file
1429
This only reads in the stuff that is not connected to the adler
1430
checksum. The position will be correct to read in the rest of
1431
the file and check the checksum after this point.
1432
The next entry in the file should be the number of parents,
1433
and their ids. Followed by a newline.
1435
header = self._state_file.readline()
1436
assert header == '#bazaar dirstate flat format 2\n', \
1437
'invalid header line: %r' % (header,)
1438
adler_line = self._state_file.readline()
1439
assert adler_line.startswith('adler32: '), 'missing adler32 checksum'
1440
self.adler_expected = int(adler_line[len('adler32: '):-1])
1441
num_entries_line = self._state_file.readline()
1442
assert num_entries_line.startswith('num_entries: '), 'missing num_entries line'
1443
self._num_entries = int(num_entries_line[len('num_entries: '):-1])
1446
"""Save any pending changes created during this session.
1448
We reuse the existing file, because that prevents race conditions with
1449
file creation, and we expect to be using oslocks on it in the near
1450
future to prevent concurrent modification and reads - because dirstates
1451
incremental data aggretation is not compatible with reading a modified
1452
file, and replacing a file in use by another process is impossible on
1455
A dirstate in read only mode should be smart enough though to validate
1456
that the file has not changed, and otherwise discard its cache and
1457
start over, to allow for fine grained read lock duration, so 'status'
1458
wont block 'commit' - for example.
1460
if (self._header_state == DirState.IN_MEMORY_MODIFIED or
1461
self._dirblock_state == DirState.IN_MEMORY_MODIFIED):
1462
self._state_file.seek(0)
1463
self._state_file.writelines(self.get_lines())
1464
self._state_file.truncate()
1465
self._state_file.flush()
1466
self._header_state = DirState.IN_MEMORY_UNMODIFIED
1467
self._dirblock_state = DirState.IN_MEMORY_UNMODIFIED
1469
def _set_data(self, parent_ids, dirblocks):
1470
"""Set the full dirstate data in memory.
1472
This is an internal function used to completely replace the objects
1473
in memory state. It puts the dirstate into state 'full-dirty'.
1475
:param parent_ids: A list of parent tree revision ids.
1476
:param dirblocks: A list containing one tuple for each directory in the
1477
tree. Each tuple contains the directory path and a list of entries
1478
found in that directory.
1480
# our memory copy is now authoritative.
1481
self._dirblocks = dirblocks
1482
self._header_state = DirState.IN_MEMORY_MODIFIED
1483
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
1484
self._parents = list(parent_ids)
1486
def set_path_id(self, path, new_id):
1487
"""Change the id of path to new_id in the current working tree.
1489
:param path: The path inside the tree to set - '' is the root, 'foo'
1490
is the path foo in the root.
1491
:param new_id: The new id to assign to the path. This must be a utf8
1492
file id (not unicode, and not None).
1494
# TODO: start warning here.
1495
assert new_id.__class__ == str
1496
self._read_dirblocks_if_needed()
1498
import pdb;pdb.set_trace()
1500
raise NotImplementedError(self.set_path_id)
1501
# TODO: check new id is unique
1502
entry = self._get_entry(0, path_utf8=path)
1503
# mark the old path absent, and insert a new root path
1504
present_parents = len(entry[1]) - 1
1505
self._make_absent(entry)
1506
id_index = self._get_id_index()
1507
self.update_minimal(('', '', new_id), 'directory', present_parents,
1508
path_utf8='', id_index=id_index, packed_stat=entry[1][0][4])
1509
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
1511
def set_parent_trees(self, trees, ghosts):
1512
"""Set the parent trees for the dirstate.
1514
:param trees: A list of revision_id, tree tuples. tree must be provided
1515
even if the revision_id refers to a ghost: supply an empty tree in
1517
:param ghosts: A list of the revision_ids that are ghosts at the time
1520
# TODO: generate a list of parent indexes to preserve to save
1521
# processing specific parent trees. In the common case one tree will
1522
# be preserved - the left most parent.
1523
# TODO: if the parent tree is a dirstate, we might want to walk them
1524
# all by path in parallel for 'optimal' common-case performance.
1525
# generate new root row.
1526
self._read_dirblocks_if_needed()
1527
# TODO future sketch: Examine the existing parents to generate a change
1528
# map and then walk the new parent trees only, mapping them into the
1529
# dirstate. Walk the dirstate at the same time to remove unreferenced
1532
# sketch: loop over all entries in the dirstate, cherry picking
1533
# entries from the parent trees, if they are not ghost trees.
1534
# after we finish walking the dirstate, all entries not in the dirstate
1535
# are deletes, so we want to append them to the end as per the design
1536
# discussions. So do a set difference on ids with the parents to
1537
# get deletes, and add them to the end.
1538
# During the update process we need to answer the following questions:
1539
# - find other keys containing a fileid in order to create cross-path
1540
# links. We dont't trivially use the inventory from other trees
1541
# because this leads to either double touching, or to accessing
1543
# - find other keys containing a path
1544
# We accumulate each entry via this dictionary, including the root
1547
# we could do parallel iterators, but because file id data may be
1548
# scattered throughout, we dont save on index overhead: we have to look
1549
# at everything anyway. We can probably save cycles by reusing parent
1550
# data and doing an incremental update when adding an additional
1551
# parent, but for now the common cases are adding a new parent (merge),
1552
# and replacing completely (commit), and commit is more common: so
1553
# optimise merge later.
1555
# ---- start generation of full tree mapping data
1556
# what trees should we use?
1557
parent_trees = [tree for rev_id, tree in trees if rev_id not in ghosts]
1558
# how many trees do we end up with
1559
parent_count = len(parent_trees)
1561
# one: the current tree
1562
for entry in self._iter_entries():
1563
# skip entries not in the current tree
1564
if entry[1][0][0] in ('a', 'r'): # absent, relocated
1566
by_path[entry[0]] = [entry[1][0]] + \
1567
[DirState.NULL_PARENT_DETAILS] * parent_count
1568
id_index[entry[0][2]] = set([entry[0]])
1570
# now the parent trees:
1571
for tree_index, tree in enumerate(parent_trees):
1572
# the index is off by one, adjust it.
1573
tree_index = tree_index + 1
1574
# when we add new locations for a fileid we need these ranges for
1575
# any fileid in this tree as we set the by_path[id] to:
1576
# already_processed_tree_details + new_details + new_location_suffix
1577
# the suffix is from tree_index+1:parent_count+1.
1578
new_location_suffix = [DirState.NULL_PARENT_DETAILS] * (parent_count - tree_index)
1579
# now stitch in all the entries from this tree
1580
for path, entry in tree.inventory.iter_entries_by_dir():
1581
# here we process each trees details for each item in the tree.
1582
# we first update any existing entries for the id at other paths,
1583
# then we either create or update the entry for the id at the
1584
# right path, and finally we add (if needed) a mapping from
1585
# file_id to this path. We do it in this order to allow us to
1586
# avoid checking all known paths for the id when generating a
1587
# new entry at this path: by adding the id->path mapping last,
1588
# all the mappings are valid and have correct relocation
1589
# records where needed.
1590
file_id = entry.file_id
1591
path_utf8 = path.encode('utf8')
1592
dirname, basename = os.path.split(path_utf8)
1593
new_entry_key = (dirname, basename, file_id)
1594
# tree index consistency: All other paths for this id in this tree
1595
# index must point to the correct path.
1596
for entry_key in id_index.setdefault(file_id, set()):
1597
# TODO:PROFILING: It might be faster to just update
1598
# rather than checking if we need to, and then overwrite
1599
# the one we are located at.
1600
if entry_key != new_entry_key:
1601
# this file id is at a different path in one of the
1602
# other trees, so put absent pointers there
1603
# This is the vertical axis in the matrix, all pointing
1605
by_path[entry_key][tree_index] = ('r', path_utf8, 0, False, '')
1606
# by path consistency: Insert into an existing path record (trivial), or
1607
# add a new one with relocation pointers for the other tree indexes.
1608
if new_entry_key in id_index[file_id]:
1609
# there is already an entry where this data belongs, just insert it.
1610
by_path[new_entry_key][tree_index] = \
1611
self._inv_entry_to_details(entry)
1613
# add relocated entries to the horizontal axis - this row
1614
# mapping from path,id. We need to look up the correct path
1615
# for the indexes from 0 to tree_index -1
1617
for lookup_index in xrange(tree_index):
1618
# boundary case: this is the first occurence of file_id
1619
# so there are no id_indexs, possibly take this out of
1621
if not len(id_index[file_id]):
1622
new_details.append(DirState.NULL_PARENT_DETAILS)
1624
# grab any one entry, use it to find the right path.
1625
# TODO: optimise this to reduce memory use in highly
1626
# fragmented situations by reusing the relocation
1628
a_key = iter(id_index[file_id]).next()
1629
if by_path[a_key][lookup_index][0] in ('r', 'a'):
1630
# its a pointer or missing statement, use it as is.
1631
new_details.append(by_path[a_key][lookup_index])
1633
# we have the right key, make a pointer to it.
1634
real_path = ('/'.join(a_key[0:2])).strip('/')
1635
new_details.append(('r', real_path, 0, False, ''))
1636
new_details.append(self._inv_entry_to_details(entry))
1637
new_details.extend(new_location_suffix)
1638
by_path[new_entry_key] = new_details
1639
id_index[file_id].add(new_entry_key)
1640
# --- end generation of full tree mappings
1642
# sort and output all the entries
1643
new_entries = sorted(by_path.items())
1644
self._entries_to_current_state(new_entries)
1645
self._parents = [rev_id for rev_id, tree in trees]
1646
self._ghosts = list(ghosts)
1647
self._header_state = DirState.IN_MEMORY_MODIFIED
1648
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
1650
def set_state_from_inventory(self, new_inv):
1651
"""Set new_inv as the current state.
1653
This API is called by tree transform, and will usually occur with
1654
existing parent trees.
1656
:param new_inv: The inventory object to set current state from.
1658
self._read_dirblocks_if_needed()
1660
# generate a byid index of the dirstate
1661
id_index = self._get_id_index()
1663
num_present_parents = len(self._parents) - len(self._ghosts)
1664
# incremental algorithm:
1665
# two iterators: current data and new data, both in dirblock order.
1666
new_iterator = new_inv.iter_entries_by_dir()
1667
# we will be modifying the dirstate, so we need a stable iterator. In
1668
# future we might write one, for now we just clone the state into a
1669
# list - which is a shallow copy, so each
1670
old_iterator = iter(list(self._iter_entries()))
1671
# both must have roots so this is safe:
1672
current_new = new_iterator.next()
1673
current_old = old_iterator.next()
1674
def advance(iterator):
1676
return iterator.next()
1677
except StopIteration:
1679
while current_new or current_old:
1680
# skip entries in old that are not really there
1681
if current_old and current_old[1][0][0] in ('r', 'a'):
1682
# relocated or absent
1683
current_old = advance(old_iterator)
1686
# convert new into dirblock style
1687
new_path_utf8 = current_new[0].encode('utf8')
1688
new_dirname, new_basename = os.path.split(new_path_utf8)
1689
new_id = current_new[1].file_id
1690
new_entry_key = (new_dirname, new_basename, new_id)
1692
# for safety disable variables
1693
new_path_utf8 = new_dirname = new_basename = new_id = new_entry_key = None
1694
# 5 cases, we dont have a value that is strictly greater than everything, so
1695
# we make both end conditions explicit
1697
# old is finished: insert current_new into the state.
1698
self.update_minimal(new_entry_key, current_new[1].kind,
1699
num_present_parents, executable=current_new[1].executable,
1700
id_index=id_index, path_utf8=new_path_utf8)
1701
current_new = advance(new_iterator)
1702
elif not current_new:
1704
self._make_absent(current_old, id_index)
1705
current_old = advance(old_iterator)
1706
elif new_entry_key == current_old[0]:
1707
# same - common case
1708
# TODO: update the record if anything significant has changed.
1709
# the minimal required trigger is if the execute bit or cached
1711
kind = DirState._minikind_to_kind[current_old[1][0][0]]
1712
if (current_old[1][0][3] != current_new[1].executable or
1713
kind != current_new[1].kind):
1714
self.update_minimal(current_old[0], current_new[1].kind,
1715
num_present_parents,
1716
executable=current_new[1].executable,
1717
id_index=id_index, path_utf8=new_path_utf8)
1718
# both sides are dealt with, move on
1719
current_old = advance(old_iterator)
1720
current_new = advance(new_iterator)
1721
elif new_entry_key < current_old[0]:
1723
# add a entry for this and advance new
1724
self.update_minimal(new_entry_key, current_new[1].kind,
1725
num_present_parents, executable=current_new[1].executable,
1726
id_index=id_index, path_utf8=new_path_utf8)
1727
current_new = advance(new_iterator)
1730
self._make_absent(current_old, id_index)
1731
current_old = advance(old_iterator)
1732
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
1734
def _make_absent(self, current_old, id_index=None):
1735
"""Mark current_old - an entry - as absent for tree 0.
1737
:param id_index: An index from fileid_utf8 to sets of keys, used by
1738
some functions. If provided it will be updated if needed.
1739
:return: True if this was the last details entry for they entry key:
1740
that is, if the underlying block has had the entry removed, thus
1741
shrinking in legnth.
1743
# build up paths that this id will be left at after the change is made,
1744
# so we can update their cross references in tree 0
1745
all_remaining_keys = set()
1746
# Dont check the working tree, because its going.
1747
for details in current_old[1][1:]:
1748
if details[0] not in ('a', 'r'): # absent, relocated
1749
all_remaining_keys.add(current_old[0])
1750
elif details[0] == 'r': # relocated
1751
# record the key for the real path.
1752
all_remaining_keys.add(tuple(os.path.split(details[1])) + (current_old[0][2],))
1753
# absent rows are not present at any path.
1754
last_reference = current_old[0] not in all_remaining_keys
1756
# the current row consists entire of the current item (being marked
1757
# absent), and relocated or absent entries for the other trees:
1758
# Remove it, its meaningless.
1759
block = self._find_block(current_old[0])
1760
entry_index, present = self._find_entry_index(current_old[0], block[1])
1762
block[1].pop(entry_index)
1763
# if we have an id_index in use, remove this key from it for this id.
1764
if id_index is not None:
1765
id_index[current_old[0][2]].remove(current_old[0])
1766
# update all remaining keys for this id to record it as absent. The
1767
# existing details may either be the record we are making as deleted
1768
# (if there were other trees with the id present at this path), or may
1770
for update_key in all_remaining_keys:
1771
update_block_index, present = \
1772
self._find_block_index_from_key(update_key)
1774
update_entry_index, present = \
1775
self._find_entry_index(update_key, self._dirblocks[update_block_index][1])
1777
update_tree_details = self._dirblocks[update_block_index][1][update_entry_index][1]
1778
# it must not be absent at the moment
1779
assert update_tree_details[0][0] != 'a' # absent
1780
update_tree_details[0] = DirState.NULL_PARENT_DETAILS
1781
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
1782
return last_reference
1784
def update_minimal(self, key, kind, num_present_parents, executable=False,
1785
fingerprint='', packed_stat=None, size=0, id_index=None,
1787
"""Update an entry to the state in tree 0."""
1788
block = self._find_block(key)[1]
1789
if packed_stat is None:
1790
packed_stat = DirState.NULLSTAT
1791
entry_index, present = self._find_entry_index(key, block)
1792
minikind = DirState._kind_to_minikind[kind]
1793
new_details = (minikind, fingerprint, size, executable, packed_stat)
1794
assert id_index is not None, 'need an id index to do updates for now !'
1796
# new entry, synthesis cross reference here,
1797
existing_keys = id_index.setdefault(key[2], set())
1798
if not existing_keys:
1799
# not currently in the state, simplest case
1800
new_entry = key, [new_details] + self._empty_parent_info()
1802
# present at one or more existing other paths.
1803
# grab one of them and use it to generate parent
1804
# relocation/absent entries.
1805
new_entry = key, [new_details]
1806
for other_key in existing_keys:
1807
# change the record at other to be a pointer to this new
1808
# record. The loop looks similar to the change to
1809
# relocations when updating an existing record but its not:
1810
# the test for existing kinds is different: this can be
1811
# factored out to a helper though.
1812
other_block_index, present = self._find_block_index_from_key(other_key)
1814
other_entry_index, present = self._find_entry_index(other_key, self._dirblocks[other_block_index][1])
1816
assert path_utf8 is not None
1817
self._dirblocks[other_block_index][1][other_entry_index][1][0] = \
1818
('r', path_utf8, 0, False, '')
1820
for lookup_index in xrange(1, num_present_parents + 1):
1821
# grab any one entry, use it to find the right path.
1822
# TODO: optimise this to reduce memory use in highly
1823
# fragmented situations by reusing the relocation
1825
update_block_index, present = \
1826
self._find_block_index_from_key(other_key)
1828
update_entry_index, present = \
1829
self._find_entry_index(other_key, self._dirblocks[update_block_index][1])
1831
update_details = self._dirblocks[update_block_index][1][update_entry_index][1][lookup_index]
1832
if update_details[0] in ('r', 'a'): # relocated, absent
1833
# its a pointer or absent in lookup_index's tree, use
1835
new_entry[1].append(update_details)
1837
# we have the right key, make a pointer to it.
1838
pointer_path = os.path.join(*other_key[0:2])
1839
new_entry[1].append(('r', pointer_path, 0, False, ''))
1840
block.insert(entry_index, new_entry)
1841
existing_keys.add(key)
1843
# Does the new state matter?
1844
block[entry_index][1][0] = new_details
1845
# parents cannot be affected by what we do.
1846
# other occurences of this id can be found
1847
# from the id index.
1849
# tree index consistency: All other paths for this id in this tree
1850
# index must point to the correct path. We have to loop here because
1851
# we may have passed entries in the state with this file id already
1852
# that were absent - where parent entries are - and they need to be
1853
# converted to relocated.
1854
assert path_utf8 is not None
1855
for entry_key in id_index.setdefault(key[2], set()):
1856
# TODO:PROFILING: It might be faster to just update
1857
# rather than checking if we need to, and then overwrite
1858
# the one we are located at.
1859
if entry_key != key:
1860
# this file id is at a different path in one of the
1861
# other trees, so put absent pointers there
1862
# This is the vertical axis in the matrix, all pointing
1864
block_index, present = self._find_block_index_from_key(entry_key)
1866
entry_index, present = self._find_entry_index(entry_key, self._dirblocks[block_index][1])
1868
self._dirblocks[block_index][1][entry_index][1][0] = \
1869
('r', path_utf8, 0, False, '')
1870
# add a containing dirblock if needed.
1871
if new_details[0] == 'd':
1872
subdir_key = (os.path.join(*key[0:2]), '', '')
1873
block_index, present = self._find_block_index_from_key(subdir_key)
1875
self._dirblocks.insert(block_index, (subdir_key[0], []))
1877
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
1880
def _wipe_state(self):
1881
"""Forget all state information about the dirstate."""
1882
self._header_state = DirState.NOT_IN_MEMORY
1883
self._dirblock_state = DirState.NOT_IN_MEMORY
1886
self._dirblocks = []
1888
def lock_read(self):
1889
"""Acquire a read lock on the dirstate"""
1890
if self._lock_token is not None:
1891
raise errors.LockContention(self._lock_token)
1892
self._lock_token = lock.ReadLock(self._filename)
1893
self._state_file = self._lock_token.f
1896
def lock_write(self):
1897
"""Acquire a write lock on the dirstate"""
1898
if self._lock_token is not None:
1899
raise errors.LockContention(self._lock_token)
1900
self._lock_token = lock.WriteLock(self._filename)
1901
self._state_file = self._lock_token.f
1905
"""Drop any locks held on the dirstate"""
1906
if self._lock_token is None:
1907
raise errors.LockNotHeld(self)
1908
self._state_file = None
1909
self._lock_token.unlock()
1910
self._lock_token = None
1912
def _requires_lock(self):
1913
"""Checks that a lock is currently held by someone on the dirstate"""
1914
if not self._lock_token:
1915
raise errors.ObjectNotLocked(self)
1917
def pack_stat(st, _encode=base64.encodestring, _pack=struct.pack):
1918
"""Convert stat values into a packed representation."""
1919
# jam 20060614 it isn't really worth removing more entries if we
1920
# are going to leave it in packed form.
1921
# With only st_mtime and st_mode filesize is 5.5M and read time is 275ms
1922
# With all entries filesize is 5.9M and read time is mabye 280ms
1923
# well within the noise margin
1925
# base64.encode always adds a final newline, so strip it off
1926
return _encode(_pack('>llllll'
1927
, st.st_size, st.st_mtime, st.st_ctime
1928
, st.st_dev, st.st_ino, st.st_mode))[:-1]
added
removed
bzrlib/dirstate.py
