1
# Copyright (C) 2006, 2007, 2008 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" | "t";
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 3", NL;
33
full checksum = "crc32: ", ["-"], WHOLE_NUMBER, NL;
34
row count = "num_entries: ", WHOLE_NUMBER, 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
89
't' is a reference to a nested subtree; the fingerprint is the referenced
94
The entries on disk and in memory are ordered according to the following keys:
96
directory, as a list of components
100
--- Format 1 had the following different definition: ---
101
rows = dirname, NULL, basename, NULL, MINIKIND, NULL, fileid_utf8, NULL,
102
WHOLE NUMBER (* size *), NULL, packed stat, NULL, sha1|symlink target,
104
PARENT ROW = NULL, revision_utf8, NULL, MINIKIND, NULL, dirname, NULL,
105
basename, NULL, WHOLE NUMBER (* size *), NULL, "y" | "n", NULL,
108
PARENT ROW's are emitted for every parent that is not in the ghosts details
109
line. That is, if the parents are foo, bar, baz, and the ghosts are bar, then
110
each row will have a PARENT ROW for foo and baz, but not for bar.
113
In any tree, a kind of 'moved' indicates that the fingerprint field
114
(which we treat as opaque data specific to the 'kind' anyway) has the
115
details for the id of this row in that tree.
117
I'm strongly tempted to add a id->path index as well, but I think that
118
where we need id->path mapping; we also usually read the whole file, so
119
I'm going to skip that for the moment, as we have the ability to locate
120
via bisect any path in any tree, and if we lookup things by path, we can
121
accumulate an id->path mapping as we go, which will tend to match what we
124
I plan to implement this asap, so please speak up now to alter/tweak the
125
design - and once we stabilise on this, I'll update the wiki page for
128
The rationale for all this is that we want fast operations for the
129
common case (diff/status/commit/merge on all files) and extremely fast
130
operations for the less common but still occurs a lot status/diff/commit
131
on specific files). Operations on specific files involve a scan for all
132
the children of a path, *in every involved tree*, which the current
133
format did not accommodate.
137
1) Fast end to end use for bzr's top 5 uses cases. (commmit/diff/status/merge/???)
138
2) fall back current object model as needed.
139
3) scale usably to the largest trees known today - say 50K entries. (mozilla
140
is an example of this)
144
Eventually reuse dirstate objects across locks IFF the dirstate file has not
145
been modified, but will require that we flush/ignore cached stat-hit data
146
because we won't want to restat all files on disk just because a lock was
147
acquired, yet we cannot trust the data after the previous lock was released.
149
Memory representation:
150
vector of all directories, and vector of the childen ?
152
root_entrie = (direntry for root, [parent_direntries_for_root]),
154
('', ['data for achild', 'data for bchild', 'data for cchild'])
155
('dir', ['achild', 'cchild', 'echild'])
157
- single bisect to find N subtrees from a path spec
158
- in-order for serialisation - this is 'dirblock' grouping.
159
- insertion of a file '/a' affects only the '/' child-vector, that is, to
160
insert 10K elements from scratch does not generates O(N^2) memoves of a
161
single vector, rather each individual, which tends to be limited to a
162
manageable number. Will scale badly on trees with 10K entries in a
163
single directory. compare with Inventory.InventoryDirectory which has
164
a dictionary for the children. No bisect capability, can only probe for
165
exact matches, or grab all elements and sort.
166
- What's the risk of error here? Once we have the base format being processed
167
we should have a net win regardless of optimality. So we are going to
168
go with what seems reasonable.
171
Maybe we should do a test profile of the core structure - 10K simulated
172
searches/lookups/etc?
174
Objects for each row?
175
The lifetime of Dirstate objects is current per lock, but see above for
176
possible extensions. The lifetime of a row from a dirstate is expected to be
177
very short in the optimistic case: which we are optimising for. For instance,
178
subtree status will determine from analysis of the disk data what rows need to
179
be examined at all, and will be able to determine from a single row whether
180
that file has altered or not, so we are aiming to process tens of thousands of
181
entries each second within the dirstate context, before exposing anything to
182
the larger codebase. This suggests we want the time for a single file
183
comparison to be < 0.1 milliseconds. That would give us 10000 paths per second
184
processed, and to scale to 100 thousand we'll another order of magnitude to do
185
that. Now, as the lifetime for all unchanged entries is the time to parse, stat
186
the file on disk, and then immediately discard, the overhead of object creation
187
becomes a significant cost.
189
Figures: Creating a tuple from from 3 elements was profiled at 0.0625
190
microseconds, whereas creating a object which is subclassed from tuple was
191
0.500 microseconds, and creating an object with 3 elements and slots was 3
192
microseconds long. 0.1 milliseconds is 100 microseconds, and ideally we'll get
193
down to 10 microseconds for the total processing - having 33% of that be object
194
creation is a huge overhead. There is a potential cost in using tuples within
195
each row which is that the conditional code to do comparisons may be slower
196
than method invocation, but method invocation is known to be slow due to stack
197
frame creation, so avoiding methods in these tight inner loops in unfortunately
198
desirable. We can consider a pyrex version of this with objects in future if
207
from stat import S_IEXEC
225
def pack_stat(st, _encode=binascii.b2a_base64, _pack=struct.pack):
226
"""Convert stat values into a packed representation."""
227
# jam 20060614 it isn't really worth removing more entries if we
228
# are going to leave it in packed form.
229
# With only st_mtime and st_mode filesize is 5.5M and read time is 275ms
230
# With all entries, filesize is 5.9M and read time is maybe 280ms
231
# well within the noise margin
233
# base64 encoding always adds a final newline, so strip it off
234
# The current version
235
return _encode(_pack('>LLLLLL'
236
, st.st_size, int(st.st_mtime), int(st.st_ctime)
237
, st.st_dev, st.st_ino & 0xFFFFFFFF, st.st_mode))[:-1]
238
# This is 0.060s / 1.520s faster by not encoding as much information
239
# return _encode(_pack('>LL', int(st.st_mtime), st.st_mode))[:-1]
240
# This is not strictly faster than _encode(_pack())[:-1]
241
# return '%X.%X.%X.%X.%X.%X' % (
242
# st.st_size, int(st.st_mtime), int(st.st_ctime),
243
# st.st_dev, st.st_ino, st.st_mode)
244
# Similar to the _encode(_pack('>LL'))
245
# return '%X.%X' % (int(st.st_mtime), st.st_mode)
248
class DirState(object):
249
"""Record directory and metadata state for fast access.
251
A dirstate is a specialised data structure for managing local working
252
tree state information. Its not yet well defined whether it is platform
253
specific, and if it is how we detect/parameterize that.
255
Dirstates use the usual lock_write, lock_read and unlock mechanisms.
256
Unlike most bzr disk formats, DirStates must be locked for reading, using
257
lock_read. (This is an os file lock internally.) This is necessary
258
because the file can be rewritten in place.
260
DirStates must be explicitly written with save() to commit changes; just
261
unlocking them does not write the changes to disk.
264
_kind_to_minikind = {
270
'tree-reference': 't',
272
_minikind_to_kind = {
278
't': 'tree-reference',
280
_stat_to_minikind = {
285
_to_yesno = {True:'y', False: 'n'} # TODO profile the performance gain
286
# of using int conversion rather than a dict here. AND BLAME ANDREW IF
289
# TODO: jam 20070221 Figure out what to do if we have a record that exceeds
290
# the BISECT_PAGE_SIZE. For now, we just have to make it large enough
291
# that we are sure a single record will always fit.
292
BISECT_PAGE_SIZE = 4096
295
IN_MEMORY_UNMODIFIED = 1
296
IN_MEMORY_MODIFIED = 2
298
# A pack_stat (the x's) that is just noise and will never match the output
301
NULL_PARENT_DETAILS = ('a', '', 0, False, '')
303
HEADER_FORMAT_2 = '#bazaar dirstate flat format 2\n'
304
HEADER_FORMAT_3 = '#bazaar dirstate flat format 3\n'
306
def __init__(self, path):
307
"""Create a DirState object.
309
:param path: The path at which the dirstate file on disk should live.
311
# _header_state and _dirblock_state represent the current state
312
# of the dirstate metadata and the per-row data respectiely.
313
# NOT_IN_MEMORY indicates that no data is in memory
314
# IN_MEMORY_UNMODIFIED indicates that what we have in memory
315
# is the same as is on disk
316
# IN_MEMORY_MODIFIED indicates that we have a modified version
317
# of what is on disk.
318
# In future we will add more granularity, for instance _dirblock_state
319
# will probably support partially-in-memory as a separate variable,
320
# allowing for partially-in-memory unmodified and partially-in-memory
322
self._header_state = DirState.NOT_IN_MEMORY
323
self._dirblock_state = DirState.NOT_IN_MEMORY
324
# If true, an error has been detected while updating the dirstate, and
325
# for safety we're not going to commit to disk.
326
self._changes_aborted = False
330
self._state_file = None
331
self._filename = path
332
self._lock_token = None
333
self._lock_state = None
334
self._id_index = None
335
# a map from packed_stat to sha's.
336
self._packed_stat_index = None
337
self._end_of_header = None
338
self._cutoff_time = None
339
self._split_path_cache = {}
340
self._bisect_page_size = DirState.BISECT_PAGE_SIZE
341
if 'hashcache' in debug.debug_flags:
342
self._sha1_file = self._sha1_file_and_mutter
344
self._sha1_file = osutils.sha_file_by_name
345
# These two attributes provide a simple cache for lookups into the
346
# dirstate in-memory vectors. By probing respectively for the last
347
# block, and for the next entry, we save nearly 2 bisections per path
349
self._last_block_index = None
350
self._last_entry_index = None
354
(self.__class__.__name__, self._filename)
356
def add(self, path, file_id, kind, stat, fingerprint):
357
"""Add a path to be tracked.
359
:param path: The path within the dirstate - '' is the root, 'foo' is the
360
path foo within the root, 'foo/bar' is the path bar within foo
362
:param file_id: The file id of the path being added.
363
:param kind: The kind of the path, as a string like 'file',
365
:param stat: The output of os.lstat for the path.
366
:param fingerprint: The sha value of the file,
367
or the target of a symlink,
368
or the referenced revision id for tree-references,
369
or '' for directories.
372
# find the block its in.
373
# find the location in the block.
374
# check its not there
376
#------- copied from inventory.ensure_normalized_name - keep synced.
377
# --- normalized_filename wants a unicode basename only, so get one.
378
dirname, basename = osutils.split(path)
379
# we dont import normalized_filename directly because we want to be
380
# able to change the implementation at runtime for tests.
381
norm_name, can_access = osutils.normalized_filename(basename)
382
if norm_name != basename:
386
raise errors.InvalidNormalization(path)
387
# you should never have files called . or ..; just add the directory
388
# in the parent, or according to the special treatment for the root
389
if basename == '.' or basename == '..':
390
raise errors.InvalidEntryName(path)
391
# now that we've normalised, we need the correct utf8 path and
392
# dirname and basename elements. This single encode and split should be
393
# faster than three separate encodes.
394
utf8path = (dirname + '/' + basename).strip('/').encode('utf8')
395
dirname, basename = osutils.split(utf8path)
396
assert file_id.__class__ == str, \
397
"must be a utf8 file_id not %s" % (type(file_id))
398
# Make sure the file_id does not exist in this tree
399
file_id_entry = self._get_entry(0, fileid_utf8=file_id)
400
if file_id_entry != (None, None):
401
path = osutils.pathjoin(file_id_entry[0][0], file_id_entry[0][1])
402
kind = DirState._minikind_to_kind[file_id_entry[1][0][0]]
403
info = '%s:%s' % (kind, path)
404
raise errors.DuplicateFileId(file_id, info)
405
first_key = (dirname, basename, '')
406
block_index, present = self._find_block_index_from_key(first_key)
408
# check the path is not in the tree
409
block = self._dirblocks[block_index][1]
410
entry_index, _ = self._find_entry_index(first_key, block)
411
while (entry_index < len(block) and
412
block[entry_index][0][0:2] == first_key[0:2]):
413
if block[entry_index][1][0][0] not in 'ar':
414
# this path is in the dirstate in the current tree.
415
raise Exception, "adding already added path!"
418
# The block where we want to put the file is not present. But it
419
# might be because the directory was empty, or not loaded yet. Look
420
# for a parent entry, if not found, raise NotVersionedError
421
parent_dir, parent_base = osutils.split(dirname)
422
parent_block_idx, parent_entry_idx, _, parent_present = \
423
self._get_block_entry_index(parent_dir, parent_base, 0)
424
if not parent_present:
425
raise errors.NotVersionedError(path, str(self))
426
self._ensure_block(parent_block_idx, parent_entry_idx, dirname)
427
block = self._dirblocks[block_index][1]
428
entry_key = (dirname, basename, file_id)
431
packed_stat = DirState.NULLSTAT
434
packed_stat = pack_stat(stat)
435
parent_info = self._empty_parent_info()
436
minikind = DirState._kind_to_minikind[kind]
438
entry_data = entry_key, [
439
(minikind, fingerprint, size, False, packed_stat),
441
elif kind == 'directory':
442
entry_data = entry_key, [
443
(minikind, '', 0, False, packed_stat),
445
elif kind == 'symlink':
446
entry_data = entry_key, [
447
(minikind, fingerprint, size, False, packed_stat),
449
elif kind == 'tree-reference':
450
entry_data = entry_key, [
451
(minikind, fingerprint, 0, False, packed_stat),
454
raise errors.BzrError('unknown kind %r' % kind)
455
entry_index, present = self._find_entry_index(entry_key, block)
457
block.insert(entry_index, entry_data)
459
assert block[entry_index][1][0][0] == 'a', " %r(%r) already added" % (basename, file_id)
460
block[entry_index][1][0] = entry_data[1][0]
462
if kind == 'directory':
463
# insert a new dirblock
464
self._ensure_block(block_index, entry_index, utf8path)
465
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
467
self._id_index.setdefault(entry_key[2], set()).add(entry_key)
469
def _bisect(self, paths):
470
"""Bisect through the disk structure for specific rows.
472
:param paths: A list of paths to find
473
:return: A dict mapping path => entries for found entries. Missing
474
entries will not be in the map.
475
The list is not sorted, and entries will be populated
476
based on when they were read.
478
self._requires_lock()
479
# We need the file pointer to be right after the initial header block
480
self._read_header_if_needed()
481
# If _dirblock_state was in memory, we should just return info from
482
# there, this function is only meant to handle when we want to read
484
assert self._dirblock_state == DirState.NOT_IN_MEMORY
486
# The disk representation is generally info + '\0\n\0' at the end. But
487
# for bisecting, it is easier to treat this as '\0' + info + '\0\n'
488
# Because it means we can sync on the '\n'
489
state_file = self._state_file
490
file_size = os.fstat(state_file.fileno()).st_size
491
# We end up with 2 extra fields, we should have a trailing '\n' to
492
# ensure that we read the whole record, and we should have a precursur
493
# '' which ensures that we start after the previous '\n'
494
entry_field_count = self._fields_per_entry() + 1
496
low = self._end_of_header
497
high = file_size - 1 # Ignore the final '\0'
498
# Map from (dir, name) => entry
501
# Avoid infinite seeking
502
max_count = 30*len(paths)
504
# pending is a list of places to look.
505
# each entry is a tuple of low, high, dir_names
506
# low -> the first byte offset to read (inclusive)
507
# high -> the last byte offset (inclusive)
508
# dir_names -> The list of (dir, name) pairs that should be found in
509
# the [low, high] range
510
pending = [(low, high, paths)]
512
page_size = self._bisect_page_size
514
fields_to_entry = self._get_fields_to_entry()
517
low, high, cur_files = pending.pop()
519
if not cur_files or low >= high:
524
if count > max_count:
525
raise errors.BzrError('Too many seeks, most likely a bug.')
527
mid = max(low, (low+high-page_size)/2)
530
# limit the read size, so we don't end up reading data that we have
532
read_size = min(page_size, (high-mid)+1)
533
block = state_file.read(read_size)
536
entries = block.split('\n')
539
# We didn't find a '\n', so we cannot have found any records.
540
# So put this range back and try again. But we know we have to
541
# increase the page size, because a single read did not contain
542
# a record break (so records must be larger than page_size)
544
pending.append((low, high, cur_files))
547
# Check the first and last entries, in case they are partial, or if
548
# we don't care about the rest of this page
550
first_fields = entries[0].split('\0')
551
if len(first_fields) < entry_field_count:
552
# We didn't get the complete first entry
553
# so move start, and grab the next, which
554
# should be a full entry
555
start += len(entries[0])+1
556
first_fields = entries[1].split('\0')
559
if len(first_fields) <= 2:
560
# We didn't even get a filename here... what do we do?
561
# Try a large page size and repeat this query
563
pending.append((low, high, cur_files))
566
# Find what entries we are looking for, which occur before and
567
# after this first record.
570
first_path = first_fields[1] + '/' + first_fields[2]
572
first_path = first_fields[2]
573
first_loc = _bisect_path_left(cur_files, first_path)
575
# These exist before the current location
576
pre = cur_files[:first_loc]
577
# These occur after the current location, which may be in the
578
# data we read, or might be after the last entry
579
post = cur_files[first_loc:]
581
if post and len(first_fields) >= entry_field_count:
582
# We have files after the first entry
584
# Parse the last entry
585
last_entry_num = len(entries)-1
586
last_fields = entries[last_entry_num].split('\0')
587
if len(last_fields) < entry_field_count:
588
# The very last hunk was not complete,
589
# read the previous hunk
590
after = mid + len(block) - len(entries[-1])
592
last_fields = entries[last_entry_num].split('\0')
594
after = mid + len(block)
597
last_path = last_fields[1] + '/' + last_fields[2]
599
last_path = last_fields[2]
600
last_loc = _bisect_path_right(post, last_path)
602
middle_files = post[:last_loc]
603
post = post[last_loc:]
606
# We have files that should occur in this block
607
# (>= first, <= last)
608
# Either we will find them here, or we can mark them as
611
if middle_files[0] == first_path:
612
# We might need to go before this location
613
pre.append(first_path)
614
if middle_files[-1] == last_path:
615
post.insert(0, last_path)
617
# Find out what paths we have
618
paths = {first_path:[first_fields]}
619
# last_path might == first_path so we need to be
620
# careful if we should append rather than overwrite
621
if last_entry_num != first_entry_num:
622
paths.setdefault(last_path, []).append(last_fields)
623
for num in xrange(first_entry_num+1, last_entry_num):
624
# TODO: jam 20070223 We are already splitting here, so
625
# shouldn't we just split the whole thing rather
626
# than doing the split again in add_one_record?
627
fields = entries[num].split('\0')
629
path = fields[1] + '/' + fields[2]
632
paths.setdefault(path, []).append(fields)
634
for path in middle_files:
635
for fields in paths.get(path, []):
636
# offset by 1 because of the opening '\0'
637
# consider changing fields_to_entry to avoid the
639
entry = fields_to_entry(fields[1:])
640
found.setdefault(path, []).append(entry)
642
# Now we have split up everything into pre, middle, and post, and
643
# we have handled everything that fell in 'middle'.
644
# We add 'post' first, so that we prefer to seek towards the
645
# beginning, so that we will tend to go as early as we need, and
646
# then only seek forward after that.
648
pending.append((after, high, post))
650
pending.append((low, start-1, pre))
652
# Consider that we may want to return the directory entries in sorted
653
# order. For now, we just return them in whatever order we found them,
654
# and leave it up to the caller if they care if it is ordered or not.
657
def _bisect_dirblocks(self, dir_list):
658
"""Bisect through the disk structure to find entries in given dirs.
660
_bisect_dirblocks is meant to find the contents of directories, which
661
differs from _bisect, which only finds individual entries.
663
:param dir_list: A sorted list of directory names ['', 'dir', 'foo'].
664
:return: A map from dir => entries_for_dir
666
# TODO: jam 20070223 A lot of the bisecting logic could be shared
667
# between this and _bisect. It would require parameterizing the
668
# inner loop with a function, though. We should evaluate the
669
# performance difference.
670
self._requires_lock()
671
# We need the file pointer to be right after the initial header block
672
self._read_header_if_needed()
673
# If _dirblock_state was in memory, we should just return info from
674
# there, this function is only meant to handle when we want to read
676
assert self._dirblock_state == DirState.NOT_IN_MEMORY
678
# The disk representation is generally info + '\0\n\0' at the end. But
679
# for bisecting, it is easier to treat this as '\0' + info + '\0\n'
680
# Because it means we can sync on the '\n'
681
state_file = self._state_file
682
file_size = os.fstat(state_file.fileno()).st_size
683
# We end up with 2 extra fields, we should have a trailing '\n' to
684
# ensure that we read the whole record, and we should have a precursur
685
# '' which ensures that we start after the previous '\n'
686
entry_field_count = self._fields_per_entry() + 1
688
low = self._end_of_header
689
high = file_size - 1 # Ignore the final '\0'
690
# Map from dir => entry
693
# Avoid infinite seeking
694
max_count = 30*len(dir_list)
696
# pending is a list of places to look.
697
# each entry is a tuple of low, high, dir_names
698
# low -> the first byte offset to read (inclusive)
699
# high -> the last byte offset (inclusive)
700
# dirs -> The list of directories that should be found in
701
# the [low, high] range
702
pending = [(low, high, dir_list)]
704
page_size = self._bisect_page_size
706
fields_to_entry = self._get_fields_to_entry()
709
low, high, cur_dirs = pending.pop()
711
if not cur_dirs or low >= high:
716
if count > max_count:
717
raise errors.BzrError('Too many seeks, most likely a bug.')
719
mid = max(low, (low+high-page_size)/2)
722
# limit the read size, so we don't end up reading data that we have
724
read_size = min(page_size, (high-mid)+1)
725
block = state_file.read(read_size)
728
entries = block.split('\n')
731
# We didn't find a '\n', so we cannot have found any records.
732
# So put this range back and try again. But we know we have to
733
# increase the page size, because a single read did not contain
734
# a record break (so records must be larger than page_size)
736
pending.append((low, high, cur_dirs))
739
# Check the first and last entries, in case they are partial, or if
740
# we don't care about the rest of this page
742
first_fields = entries[0].split('\0')
743
if len(first_fields) < entry_field_count:
744
# We didn't get the complete first entry
745
# so move start, and grab the next, which
746
# should be a full entry
747
start += len(entries[0])+1
748
first_fields = entries[1].split('\0')
751
if len(first_fields) <= 1:
752
# We didn't even get a dirname here... what do we do?
753
# Try a large page size and repeat this query
755
pending.append((low, high, cur_dirs))
758
# Find what entries we are looking for, which occur before and
759
# after this first record.
761
first_dir = first_fields[1]
762
first_loc = bisect.bisect_left(cur_dirs, first_dir)
764
# These exist before the current location
765
pre = cur_dirs[:first_loc]
766
# These occur after the current location, which may be in the
767
# data we read, or might be after the last entry
768
post = cur_dirs[first_loc:]
770
if post and len(first_fields) >= entry_field_count:
771
# We have records to look at after the first entry
773
# Parse the last entry
774
last_entry_num = len(entries)-1
775
last_fields = entries[last_entry_num].split('\0')
776
if len(last_fields) < entry_field_count:
777
# The very last hunk was not complete,
778
# read the previous hunk
779
after = mid + len(block) - len(entries[-1])
781
last_fields = entries[last_entry_num].split('\0')
783
after = mid + len(block)
785
last_dir = last_fields[1]
786
last_loc = bisect.bisect_right(post, last_dir)
788
middle_files = post[:last_loc]
789
post = post[last_loc:]
792
# We have files that should occur in this block
793
# (>= first, <= last)
794
# Either we will find them here, or we can mark them as
797
if middle_files[0] == first_dir:
798
# We might need to go before this location
799
pre.append(first_dir)
800
if middle_files[-1] == last_dir:
801
post.insert(0, last_dir)
803
# Find out what paths we have
804
paths = {first_dir:[first_fields]}
805
# last_dir might == first_dir so we need to be
806
# careful if we should append rather than overwrite
807
if last_entry_num != first_entry_num:
808
paths.setdefault(last_dir, []).append(last_fields)
809
for num in xrange(first_entry_num+1, last_entry_num):
810
# TODO: jam 20070223 We are already splitting here, so
811
# shouldn't we just split the whole thing rather
812
# than doing the split again in add_one_record?
813
fields = entries[num].split('\0')
814
paths.setdefault(fields[1], []).append(fields)
816
for cur_dir in middle_files:
817
for fields in paths.get(cur_dir, []):
818
# offset by 1 because of the opening '\0'
819
# consider changing fields_to_entry to avoid the
821
entry = fields_to_entry(fields[1:])
822
found.setdefault(cur_dir, []).append(entry)
824
# Now we have split up everything into pre, middle, and post, and
825
# we have handled everything that fell in 'middle'.
826
# We add 'post' first, so that we prefer to seek towards the
827
# beginning, so that we will tend to go as early as we need, and
828
# then only seek forward after that.
830
pending.append((after, high, post))
832
pending.append((low, start-1, pre))
836
def _bisect_recursive(self, paths):
837
"""Bisect for entries for all paths and their children.
839
This will use bisect to find all records for the supplied paths. It
840
will then continue to bisect for any records which are marked as
841
directories. (and renames?)
843
:param paths: A sorted list of (dir, name) pairs
844
eg: [('', 'a'), ('', 'f'), ('a/b', 'c')]
845
:return: A dictionary mapping (dir, name, file_id) => [tree_info]
847
# Map from (dir, name, file_id) => [tree_info]
850
found_dir_names = set()
852
# Directories that have been read
853
processed_dirs = set()
854
# Get the ball rolling with the first bisect for all entries.
855
newly_found = self._bisect(paths)
858
# Directories that need to be read
860
paths_to_search = set()
861
for entry_list in newly_found.itervalues():
862
for dir_name_id, trees_info in entry_list:
863
found[dir_name_id] = trees_info
864
found_dir_names.add(dir_name_id[:2])
866
for tree_info in trees_info:
867
minikind = tree_info[0]
870
# We already processed this one as a directory,
871
# we don't need to do the extra work again.
873
subdir, name, file_id = dir_name_id
874
path = osutils.pathjoin(subdir, name)
876
if path not in processed_dirs:
877
pending_dirs.add(path)
878
elif minikind == 'r':
879
# Rename, we need to directly search the target
880
# which is contained in the fingerprint column
881
dir_name = osutils.split(tree_info[1])
882
if dir_name[0] in pending_dirs:
883
# This entry will be found in the dir search
885
if dir_name not in found_dir_names:
886
paths_to_search.add(tree_info[1])
887
# Now we have a list of paths to look for directly, and
888
# directory blocks that need to be read.
889
# newly_found is mixing the keys between (dir, name) and path
890
# entries, but that is okay, because we only really care about the
892
newly_found = self._bisect(sorted(paths_to_search))
893
newly_found.update(self._bisect_dirblocks(sorted(pending_dirs)))
894
processed_dirs.update(pending_dirs)
897
def _discard_merge_parents(self):
898
"""Discard any parents trees beyond the first.
900
Note that if this fails the dirstate is corrupted.
902
After this function returns the dirstate contains 2 trees, neither of
905
self._read_header_if_needed()
906
parents = self.get_parent_ids()
909
# only require all dirblocks if we are doing a full-pass removal.
910
self._read_dirblocks_if_needed()
911
dead_patterns = set([('a', 'r'), ('a', 'a'), ('r', 'r'), ('r', 'a')])
912
def iter_entries_removable():
913
for block in self._dirblocks:
914
deleted_positions = []
915
for pos, entry in enumerate(block[1]):
917
if (entry[1][0][0], entry[1][1][0]) in dead_patterns:
918
deleted_positions.append(pos)
919
if deleted_positions:
920
if len(deleted_positions) == len(block[1]):
923
for pos in reversed(deleted_positions):
925
# if the first parent is a ghost:
926
if parents[0] in self.get_ghosts():
927
empty_parent = [DirState.NULL_PARENT_DETAILS]
928
for entry in iter_entries_removable():
929
entry[1][1:] = empty_parent
931
for entry in iter_entries_removable():
935
self._parents = [parents[0]]
936
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
937
self._header_state = DirState.IN_MEMORY_MODIFIED
939
def _empty_parent_info(self):
940
return [DirState.NULL_PARENT_DETAILS] * (len(self._parents) -
943
def _ensure_block(self, parent_block_index, parent_row_index, dirname):
944
"""Ensure a block for dirname exists.
946
This function exists to let callers which know that there is a
947
directory dirname ensure that the block for it exists. This block can
948
fail to exist because of demand loading, or because a directory had no
949
children. In either case it is not an error. It is however an error to
950
call this if there is no parent entry for the directory, and thus the
951
function requires the coordinates of such an entry to be provided.
953
The root row is special cased and can be indicated with a parent block
956
:param parent_block_index: The index of the block in which dirname's row
958
:param parent_row_index: The index in the parent block where the row
960
:param dirname: The utf8 dirname to ensure there is a block for.
961
:return: The index for the block.
963
if dirname == '' and parent_row_index == 0 and parent_block_index == 0:
964
# This is the signature of the root row, and the
965
# contents-of-root row is always index 1
967
# the basename of the directory must be the end of its full name.
968
if not (parent_block_index == -1 and
969
parent_block_index == -1 and dirname == ''):
970
assert dirname.endswith(
971
self._dirblocks[parent_block_index][1][parent_row_index][0][1])
972
block_index, present = self._find_block_index_from_key((dirname, '', ''))
974
## In future, when doing partial parsing, this should load and
975
# populate the entire block.
976
self._dirblocks.insert(block_index, (dirname, []))
979
def _entries_to_current_state(self, new_entries):
980
"""Load new_entries into self.dirblocks.
982
Process new_entries into the current state object, making them the active
983
state. The entries are grouped together by directory to form dirblocks.
985
:param new_entries: A sorted list of entries. This function does not sort
986
to prevent unneeded overhead when callers have a sorted list already.
989
assert new_entries[0][0][0:2] == ('', ''), \
990
"Missing root row %r" % (new_entries[0][0],)
991
# The two blocks here are deliberate: the root block and the
992
# contents-of-root block.
993
self._dirblocks = [('', []), ('', [])]
994
current_block = self._dirblocks[0][1]
997
append_entry = current_block.append
998
for entry in new_entries:
999
if entry[0][0] != current_dirname:
1000
# new block - different dirname
1002
current_dirname = entry[0][0]
1003
self._dirblocks.append((current_dirname, current_block))
1004
append_entry = current_block.append
1005
# append the entry to the current block
1007
self._split_root_dirblock_into_contents()
1009
def _split_root_dirblock_into_contents(self):
1010
"""Split the root dirblocks into root and contents-of-root.
1012
After parsing by path, we end up with root entries and contents-of-root
1013
entries in the same block. This loop splits them out again.
1015
# The above loop leaves the "root block" entries mixed with the
1016
# "contents-of-root block". But we don't want an if check on
1017
# all entries, so instead we just fix it up here.
1018
assert self._dirblocks[1] == ('', [])
1020
contents_of_root_block = []
1021
for entry in self._dirblocks[0][1]:
1022
if not entry[0][1]: # This is a root entry
1023
root_block.append(entry)
1025
contents_of_root_block.append(entry)
1026
self._dirblocks[0] = ('', root_block)
1027
self._dirblocks[1] = ('', contents_of_root_block)
1029
def _entry_to_line(self, entry):
1030
"""Serialize entry to a NULL delimited line ready for _get_output_lines.
1032
:param entry: An entry_tuple as defined in the module docstring.
1034
entire_entry = list(entry[0])
1035
for tree_number, tree_data in enumerate(entry[1]):
1036
# (minikind, fingerprint, size, executable, tree_specific_string)
1037
entire_entry.extend(tree_data)
1038
# 3 for the key, 5 for the fields per tree.
1039
tree_offset = 3 + tree_number * 5
1041
entire_entry[tree_offset + 0] = tree_data[0]
1043
entire_entry[tree_offset + 2] = str(tree_data[2])
1045
entire_entry[tree_offset + 3] = DirState._to_yesno[tree_data[3]]
1046
return '\0'.join(entire_entry)
1048
def _fields_per_entry(self):
1049
"""How many null separated fields should be in each entry row.
1051
Each line now has an extra '\n' field which is not used
1052
so we just skip over it
1054
3 fields for the key
1055
+ number of fields per tree_data (5) * tree count
1058
tree_count = 1 + self._num_present_parents()
1059
return 3 + 5 * tree_count + 1
1061
def _find_block(self, key, add_if_missing=False):
1062
"""Return the block that key should be present in.
1064
:param key: A dirstate entry key.
1065
:return: The block tuple.
1067
block_index, present = self._find_block_index_from_key(key)
1069
if not add_if_missing:
1070
# check to see if key is versioned itself - we might want to
1071
# add it anyway, because dirs with no entries dont get a
1072
# dirblock at parse time.
1073
# This is an uncommon branch to take: most dirs have children,
1074
# and most code works with versioned paths.
1075
parent_base, parent_name = osutils.split(key[0])
1076
if not self._get_block_entry_index(parent_base, parent_name, 0)[3]:
1077
# some parent path has not been added - its an error to add
1079
raise errors.NotVersionedError(key[0:2], str(self))
1080
self._dirblocks.insert(block_index, (key[0], []))
1081
return self._dirblocks[block_index]
1083
def _find_block_index_from_key(self, key):
1084
"""Find the dirblock index for a key.
1086
:return: The block index, True if the block for the key is present.
1088
if key[0:2] == ('', ''):
1091
if (self._last_block_index is not None and
1092
self._dirblocks[self._last_block_index][0] == key[0]):
1093
return self._last_block_index, True
1096
block_index = bisect_dirblock(self._dirblocks, key[0], 1,
1097
cache=self._split_path_cache)
1098
# _right returns one-past-where-key is so we have to subtract
1099
# one to use it. we use _right here because there are two
1100
# '' blocks - the root, and the contents of root
1101
# we always have a minimum of 2 in self._dirblocks: root and
1102
# root-contents, and for '', we get 2 back, so this is
1103
# simple and correct:
1104
present = (block_index < len(self._dirblocks) and
1105
self._dirblocks[block_index][0] == key[0])
1106
self._last_block_index = block_index
1107
# Reset the entry index cache to the beginning of the block.
1108
self._last_entry_index = -1
1109
return block_index, present
1111
def _find_entry_index(self, key, block):
1112
"""Find the entry index for a key in a block.
1114
:return: The entry index, True if the entry for the key is present.
1116
len_block = len(block)
1118
if self._last_entry_index is not None:
1120
entry_index = self._last_entry_index + 1
1121
# A hit is when the key is after the last slot, and before or
1122
# equal to the next slot.
1123
if ((entry_index > 0 and block[entry_index - 1][0] < key) and
1124
key <= block[entry_index][0]):
1125
self._last_entry_index = entry_index
1126
present = (block[entry_index][0] == key)
1127
return entry_index, present
1130
entry_index = bisect.bisect_left(block, (key, []))
1131
present = (entry_index < len_block and
1132
block[entry_index][0] == key)
1133
self._last_entry_index = entry_index
1134
return entry_index, present
1137
def from_tree(tree, dir_state_filename):
1138
"""Create a dirstate from a bzr Tree.
1140
:param tree: The tree which should provide parent information and
1142
:return: a DirState object which is currently locked for writing.
1143
(it was locked by DirState.initialize)
1145
result = DirState.initialize(dir_state_filename)
1149
parent_ids = tree.get_parent_ids()
1150
num_parents = len(parent_ids)
1152
for parent_id in parent_ids:
1153
parent_tree = tree.branch.repository.revision_tree(parent_id)
1154
parent_trees.append((parent_id, parent_tree))
1155
parent_tree.lock_read()
1156
result.set_parent_trees(parent_trees, [])
1157
result.set_state_from_inventory(tree.inventory)
1159
for revid, parent_tree in parent_trees:
1160
parent_tree.unlock()
1163
# The caller won't have a chance to unlock this, so make sure we
1169
def update_basis_by_delta(self, delta, new_revid):
1170
"""Update the parents of this tree after a commit.
1172
This gives the tree one parent, with revision id new_revid. The
1173
inventory delta is applied to the current basis tree to generate the
1174
inventory for the parent new_revid, and all other parent trees are
1177
Note that an exception during the operation of this method will leave
1178
the dirstate in a corrupt state where it should not be saved.
1180
Finally, we expect all changes to be synchronising the basis tree with
1183
:param new_revid: The new revision id for the trees parent.
1184
:param delta: An inventory delta (see apply_inventory_delta) describing
1185
the changes from the current left most parent revision to new_revid.
1187
self._read_dirblocks_if_needed()
1188
self._discard_merge_parents()
1189
if self._ghosts != []:
1190
raise NotImplementedError(self.update_basis_by_delta)
1191
if len(self._parents) == 0:
1192
# setup a blank tree, the most simple way.
1193
empty_parent = DirState.NULL_PARENT_DETAILS
1194
for entry in self._iter_entries():
1195
entry[1].append(empty_parent)
1196
self._parents.append(new_revid)
1198
self._parents[0] = new_revid
1200
delta = sorted(delta, reverse=True)
1204
# The paths this function accepts are unicode and must be encoded as we
1206
encode = cache_utf8.encode
1207
inv_to_entry = self._inv_entry_to_details
1208
# delta is now (deletes, changes), (adds) in reverse lexographical
1210
# deletes in reverse lexographic order are safe to process in situ.
1211
# renames are not, as a rename from any path could go to a path
1212
# lexographically lower, so we transform renames into delete, add pairs,
1213
# expanding them recursively as needed.
1214
# At the same time, to reduce interface friction we convert the input
1215
# inventory entries to dirstate.
1216
root_only = ('', '')
1217
for old_path, new_path, file_id, inv_entry in delta:
1218
if old_path is None:
1219
adds.append((None, encode(new_path), file_id,
1220
inv_to_entry(inv_entry), True))
1221
elif new_path is None:
1222
deletes.append((encode(old_path), None, file_id, None, True))
1223
elif (old_path, new_path) != root_only:
1225
# Because renames must preserve their children we must have
1226
# processed all relocations and removes before hand. The sort
1227
# order ensures we've examined the child paths, but we also
1228
# have to execute the removals, or the split to an add/delete
1229
# pair will result in the deleted item being reinserted, or
1230
# renamed items being reinserted twice - and possibly at the
1231
# wrong place. Splitting into a delete/add pair also simplifies
1232
# the handling of entries with ('f', ...), ('r' ...) because
1233
# the target of the 'r' is old_path here, and we add that to
1234
# deletes, meaning that the add handler does not need to check
1235
# for 'r' items on every pass.
1236
self._update_basis_apply_deletes(deletes)
1238
new_path_utf8 = encode(new_path)
1239
# Split into an add/delete pair recursively.
1240
adds.append((None, new_path_utf8, file_id,
1241
inv_to_entry(inv_entry), False))
1242
# Expunge deletes that we've seen so that deleted/renamed
1243
# children of a rename directory are handled correctly.
1244
new_deletes = reversed(list(self._iter_child_entries(1,
1246
# Remove the current contents of the tree at orig_path, and
1247
# reinsert at the correct new path.
1248
for entry in new_deletes:
1250
source_path = entry[0][0] + '/' + entry[0][1]
1252
source_path = entry[0][1]
1253
target_path = new_path_utf8 + source_path[len(old_path):]
1254
adds.append((None, target_path, entry[0][2], entry[1][1], False))
1256
(source_path, target_path, entry[0][2], None, False))
1258
(encode(old_path), new_path, file_id, None, False))
1260
# changes to just the root should not require remove/insertion
1262
changes.append((encode(old_path), encode(new_path), file_id,
1263
inv_to_entry(inv_entry)))
1265
# Finish expunging deletes/first half of renames.
1266
self._update_basis_apply_deletes(deletes)
1267
# Reinstate second half of renames and new paths.
1268
self._update_basis_apply_adds(adds)
1269
# Apply in-situ changes.
1270
self._update_basis_apply_changes(changes)
1272
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
1273
self._header_state = DirState.IN_MEMORY_MODIFIED
1274
self._id_index = None
1277
def _update_basis_apply_adds(self, adds):
1278
"""Apply a sequence of adds to tree 1 during update_basis_by_delta.
1280
They may be adds, or renames that have been split into add/delete
1283
:param adds: A sequence of adds. Each add is a tuple:
1284
(None, new_path_utf8, file_id, (entry_details), real_add). real_add
1285
is False when the add is the second half of a remove-and-reinsert
1286
pair created to handle renames and deletes.
1288
# Adds are accumulated partly from renames, so can be in any input
1291
# adds is now in lexographic order, which places all parents before
1292
# their children, so we can process it linearly.
1294
for old_path, new_path, file_id, new_details, real_add in adds:
1295
assert old_path is None
1296
# the entry for this file_id must be in tree 0.
1297
entry = self._get_entry(0, file_id, new_path)
1298
if entry[0] is None or entry[0][2] != file_id:
1299
self._changes_aborted = True
1300
raise errors.InconsistentDelta(new_path, file_id,
1301
'working tree does not contain new entry')
1302
if real_add and entry[1][1][0] not in absent:
1303
self._changes_aborted = True
1304
raise errors.InconsistentDelta(new_path, file_id,
1305
'The entry was considered to be a genuinely new record,'
1306
' but there was already an old record for it.')
1307
# We don't need to update the target of an 'r' because the handling
1308
# of renames turns all 'r' situations into a delete at the original
1310
entry[1][1] = new_details
1312
def _update_basis_apply_changes(self, changes):
1313
"""Apply a sequence of changes to tree 1 during update_basis_by_delta.
1315
:param adds: A sequence of changes. Each change is a tuple:
1316
(path_utf8, path_utf8, file_id, (entry_details))
1319
for old_path, new_path, file_id, new_details in changes:
1320
assert old_path == new_path
1321
# the entry for this file_id must be in tree 0.
1322
entry = self._get_entry(0, file_id, new_path)
1323
if entry[0] is None or entry[0][2] != file_id:
1324
self._changes_aborted = True
1325
raise errors.InconsistentDelta(new_path, file_id,
1326
'working tree does not contain new entry')
1327
if (entry[1][0][0] in absent or
1328
entry[1][1][0] in absent):
1329
self._changes_aborted = True
1330
raise errors.InconsistentDelta(new_path, file_id,
1331
'changed considered absent')
1332
entry[1][1] = new_details
1334
def _update_basis_apply_deletes(self, deletes):
1335
"""Apply a sequence of deletes to tree 1 during update_basis_by_delta.
1337
They may be deletes, or renames that have been split into add/delete
1340
:param deletes: A sequence of deletes. Each delete is a tuple:
1341
(old_path_utf8, new_path_utf8, file_id, None, real_delete).
1342
real_delete is True when the desired outcome is an actual deletion
1343
rather than the rename handling logic temporarily deleting a path
1344
during the replacement of a parent.
1346
null = DirState.NULL_PARENT_DETAILS
1347
for old_path, new_path, file_id, _, real_delete in deletes:
1349
assert new_path is None
1351
assert new_path is not None
1352
# the entry for this file_id must be in tree 1.
1353
dirname, basename = osutils.split(old_path)
1354
block_index, entry_index, dir_present, file_present = \
1355
self._get_block_entry_index(dirname, basename, 1)
1356
if not file_present:
1357
self._changes_aborted = True
1358
raise errors.InconsistentDelta(old_path, file_id,
1359
'basis tree does not contain removed entry')
1360
entry = self._dirblocks[block_index][1][entry_index]
1361
if entry[0][2] != file_id:
1362
self._changes_aborted = True
1363
raise errors.InconsistentDelta(old_path, file_id,
1364
'mismatched file_id in tree 1')
1366
if entry[1][0][0] != 'a':
1367
self._changes_aborted = True
1368
raise errors.InconsistentDelta(old_path, file_id,
1369
'This was marked as a real delete, but the WT state'
1370
' claims that it still exists and is versioned.')
1371
del self._dirblocks[block_index][1][entry_index]
1373
if entry[1][0][0] == 'a':
1374
self._changes_aborted = True
1375
raise errors.InconsistentDelta(old_path, file_id,
1376
'The entry was considered a rename, but the source path'
1377
' is marked as absent.')
1378
# For whatever reason, we were asked to rename an entry
1379
# that was originally marked as deleted. This could be
1380
# because we are renaming the parent directory, and the WT
1381
# current state has the file marked as deleted.
1382
elif entry[1][0][0] == 'r':
1383
# implement the rename
1384
del self._dirblocks[block_index][1][entry_index]
1386
# it is being resurrected here, so blank it out temporarily.
1387
self._dirblocks[block_index][1][entry_index][1][1] = null
1389
def update_entry(self, entry, abspath, stat_value,
1390
_stat_to_minikind=_stat_to_minikind,
1391
_pack_stat=pack_stat):
1392
"""Update the entry based on what is actually on disk.
1394
:param entry: This is the dirblock entry for the file in question.
1395
:param abspath: The path on disk for this file.
1396
:param stat_value: (optional) if we already have done a stat on the
1398
:return: The sha1 hexdigest of the file (40 bytes) or link target of a
1402
minikind = _stat_to_minikind[stat_value.st_mode & 0170000]
1406
packed_stat = _pack_stat(stat_value)
1407
(saved_minikind, saved_link_or_sha1, saved_file_size,
1408
saved_executable, saved_packed_stat) = entry[1][0]
1410
if (minikind == saved_minikind
1411
and packed_stat == saved_packed_stat):
1412
# The stat hasn't changed since we saved, so we can re-use the
1417
# size should also be in packed_stat
1418
if saved_file_size == stat_value.st_size:
1419
return saved_link_or_sha1
1421
# If we have gotten this far, that means that we need to actually
1422
# process this entry.
1425
link_or_sha1 = self._sha1_file(abspath)
1426
executable = self._is_executable(stat_value.st_mode,
1428
if self._cutoff_time is None:
1429
self._sha_cutoff_time()
1430
if (stat_value.st_mtime < self._cutoff_time
1431
and stat_value.st_ctime < self._cutoff_time):
1432
entry[1][0] = ('f', link_or_sha1, stat_value.st_size,
1433
executable, packed_stat)
1435
entry[1][0] = ('f', '', stat_value.st_size,
1436
executable, DirState.NULLSTAT)
1437
elif minikind == 'd':
1439
entry[1][0] = ('d', '', 0, False, packed_stat)
1440
if saved_minikind != 'd':
1441
# This changed from something into a directory. Make sure we
1442
# have a directory block for it. This doesn't happen very
1443
# often, so this doesn't have to be super fast.
1444
block_index, entry_index, dir_present, file_present = \
1445
self._get_block_entry_index(entry[0][0], entry[0][1], 0)
1446
self._ensure_block(block_index, entry_index,
1447
osutils.pathjoin(entry[0][0], entry[0][1]))
1448
elif minikind == 'l':
1449
link_or_sha1 = self._read_link(abspath, saved_link_or_sha1)
1450
if self._cutoff_time is None:
1451
self._sha_cutoff_time()
1452
if (stat_value.st_mtime < self._cutoff_time
1453
and stat_value.st_ctime < self._cutoff_time):
1454
entry[1][0] = ('l', link_or_sha1, stat_value.st_size,
1457
entry[1][0] = ('l', '', stat_value.st_size,
1458
False, DirState.NULLSTAT)
1459
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
1462
def _sha_cutoff_time(self):
1463
"""Return cutoff time.
1465
Files modified more recently than this time are at risk of being
1466
undetectably modified and so can't be cached.
1468
# Cache the cutoff time as long as we hold a lock.
1469
# time.time() isn't super expensive (approx 3.38us), but
1470
# when you call it 50,000 times it adds up.
1471
# For comparison, os.lstat() costs 7.2us if it is hot.
1472
self._cutoff_time = int(time.time()) - 3
1473
return self._cutoff_time
1475
def _lstat(self, abspath, entry):
1476
"""Return the os.lstat value for this path."""
1477
return os.lstat(abspath)
1479
def _sha1_file_and_mutter(self, abspath):
1480
# when -Dhashcache is turned on, this is monkey-patched in to log
1482
trace.mutter("dirstate sha1 " + abspath)
1483
return osutils.sha_file_by_name(abspath)
1485
def _is_executable(self, mode, old_executable):
1486
"""Is this file executable?"""
1487
return bool(S_IEXEC & mode)
1489
def _is_executable_win32(self, mode, old_executable):
1490
"""On win32 the executable bit is stored in the dirstate."""
1491
return old_executable
1493
if sys.platform == 'win32':
1494
_is_executable = _is_executable_win32
1496
def _read_link(self, abspath, old_link):
1497
"""Read the target of a symlink"""
1498
# TODO: jam 200700301 On Win32, this could just return the value
1499
# already in memory. However, this really needs to be done at a
1500
# higher level, because there either won't be anything on disk,
1501
# or the thing on disk will be a file.
1502
return os.readlink(abspath)
1504
def get_ghosts(self):
1505
"""Return a list of the parent tree revision ids that are ghosts."""
1506
self._read_header_if_needed()
1509
def get_lines(self):
1510
"""Serialise the entire dirstate to a sequence of lines."""
1511
if (self._header_state == DirState.IN_MEMORY_UNMODIFIED and
1512
self._dirblock_state == DirState.IN_MEMORY_UNMODIFIED):
1513
# read whats on disk.
1514
self._state_file.seek(0)
1515
return self._state_file.readlines()
1517
lines.append(self._get_parents_line(self.get_parent_ids()))
1518
lines.append(self._get_ghosts_line(self._ghosts))
1519
# append the root line which is special cased
1520
lines.extend(map(self._entry_to_line, self._iter_entries()))
1521
return self._get_output_lines(lines)
1523
def _get_ghosts_line(self, ghost_ids):
1524
"""Create a line for the state file for ghost information."""
1525
return '\0'.join([str(len(ghost_ids))] + ghost_ids)
1527
def _get_parents_line(self, parent_ids):
1528
"""Create a line for the state file for parents information."""
1529
return '\0'.join([str(len(parent_ids))] + parent_ids)
1531
def _get_fields_to_entry(self):
1532
"""Get a function which converts entry fields into a entry record.
1534
This handles size and executable, as well as parent records.
1536
:return: A function which takes a list of fields, and returns an
1537
appropriate record for storing in memory.
1539
# This is intentionally unrolled for performance
1540
num_present_parents = self._num_present_parents()
1541
if num_present_parents == 0:
1542
def fields_to_entry_0_parents(fields, _int=int):
1543
path_name_file_id_key = (fields[0], fields[1], fields[2])
1544
return (path_name_file_id_key, [
1546
fields[3], # minikind
1547
fields[4], # fingerprint
1548
_int(fields[5]), # size
1549
fields[6] == 'y', # executable
1550
fields[7], # packed_stat or revision_id
1552
return fields_to_entry_0_parents
1553
elif num_present_parents == 1:
1554
def fields_to_entry_1_parent(fields, _int=int):
1555
path_name_file_id_key = (fields[0], fields[1], fields[2])
1556
return (path_name_file_id_key, [
1558
fields[3], # minikind
1559
fields[4], # fingerprint
1560
_int(fields[5]), # size
1561
fields[6] == 'y', # executable
1562
fields[7], # packed_stat or revision_id
1565
fields[8], # minikind
1566
fields[9], # fingerprint
1567
_int(fields[10]), # size
1568
fields[11] == 'y', # executable
1569
fields[12], # packed_stat or revision_id
1572
return fields_to_entry_1_parent
1573
elif num_present_parents == 2:
1574
def fields_to_entry_2_parents(fields, _int=int):
1575
path_name_file_id_key = (fields[0], fields[1], fields[2])
1576
return (path_name_file_id_key, [
1578
fields[3], # minikind
1579
fields[4], # fingerprint
1580
_int(fields[5]), # size
1581
fields[6] == 'y', # executable
1582
fields[7], # packed_stat or revision_id
1585
fields[8], # minikind
1586
fields[9], # fingerprint
1587
_int(fields[10]), # size
1588
fields[11] == 'y', # executable
1589
fields[12], # packed_stat or revision_id
1592
fields[13], # minikind
1593
fields[14], # fingerprint
1594
_int(fields[15]), # size
1595
fields[16] == 'y', # executable
1596
fields[17], # packed_stat or revision_id
1599
return fields_to_entry_2_parents
1601
def fields_to_entry_n_parents(fields, _int=int):
1602
path_name_file_id_key = (fields[0], fields[1], fields[2])
1603
trees = [(fields[cur], # minikind
1604
fields[cur+1], # fingerprint
1605
_int(fields[cur+2]), # size
1606
fields[cur+3] == 'y', # executable
1607
fields[cur+4], # stat or revision_id
1608
) for cur in xrange(3, len(fields)-1, 5)]
1609
return path_name_file_id_key, trees
1610
return fields_to_entry_n_parents
1612
def get_parent_ids(self):
1613
"""Return a list of the parent tree ids for the directory state."""
1614
self._read_header_if_needed()
1615
return list(self._parents)
1617
def _get_block_entry_index(self, dirname, basename, tree_index):
1618
"""Get the coordinates for a path in the state structure.
1620
:param dirname: The utf8 dirname to lookup.
1621
:param basename: The utf8 basename to lookup.
1622
:param tree_index: The index of the tree for which this lookup should
1624
:return: A tuple describing where the path is located, or should be
1625
inserted. The tuple contains four fields: the block index, the row
1626
index, the directory is present (boolean), the entire path is
1627
present (boolean). There is no guarantee that either
1628
coordinate is currently reachable unless the found field for it is
1629
True. For instance, a directory not present in the searched tree
1630
may be returned with a value one greater than the current highest
1631
block offset. The directory present field will always be True when
1632
the path present field is True. The directory present field does
1633
NOT indicate that the directory is present in the searched tree,
1634
rather it indicates that there are at least some files in some
1637
self._read_dirblocks_if_needed()
1638
key = dirname, basename, ''
1639
block_index, present = self._find_block_index_from_key(key)
1641
# no such directory - return the dir index and 0 for the row.
1642
return block_index, 0, False, False
1643
block = self._dirblocks[block_index][1] # access the entries only
1644
entry_index, present = self._find_entry_index(key, block)
1645
# linear search through entries at this path to find the one
1647
while entry_index < len(block) and block[entry_index][0][1] == basename:
1648
if block[entry_index][1][tree_index][0] not in 'ar':
1649
# neither absent or relocated
1650
return block_index, entry_index, True, True
1652
return block_index, entry_index, True, False
1654
def _get_entry(self, tree_index, fileid_utf8=None, path_utf8=None):
1655
"""Get the dirstate entry for path in tree tree_index.
1657
If either file_id or path is supplied, it is used as the key to lookup.
1658
If both are supplied, the fastest lookup is used, and an error is
1659
raised if they do not both point at the same row.
1661
:param tree_index: The index of the tree we wish to locate this path
1662
in. If the path is present in that tree, the entry containing its
1663
details is returned, otherwise (None, None) is returned
1664
0 is the working tree, higher indexes are successive parent
1666
:param fileid_utf8: A utf8 file_id to look up.
1667
:param path_utf8: An utf8 path to be looked up.
1668
:return: The dirstate entry tuple for path, or (None, None)
1670
self._read_dirblocks_if_needed()
1671
if path_utf8 is not None:
1672
assert path_utf8.__class__ == str, ('path_utf8 is not a str: %s %s'
1673
% (type(path_utf8), path_utf8))
1674
# path lookups are faster
1675
dirname, basename = osutils.split(path_utf8)
1676
block_index, entry_index, dir_present, file_present = \
1677
self._get_block_entry_index(dirname, basename, tree_index)
1678
if not file_present:
1680
entry = self._dirblocks[block_index][1][entry_index]
1681
assert entry[0][2] and entry[1][tree_index][0] not in ('a', 'r'), 'unversioned entry?!?!'
1683
if entry[0][2] != fileid_utf8:
1684
self._changes_aborted = True
1685
raise errors.BzrError('integrity error ? : mismatching'
1686
' tree_index, file_id and path')
1689
assert fileid_utf8 is not None
1690
possible_keys = self._get_id_index().get(fileid_utf8, None)
1691
if not possible_keys:
1693
for key in possible_keys:
1694
block_index, present = \
1695
self._find_block_index_from_key(key)
1696
# strange, probably indicates an out of date
1697
# id index - for now, allow this.
1700
# WARNING: DO not change this code to use _get_block_entry_index
1701
# as that function is not suitable: it does not use the key
1702
# to lookup, and thus the wrong coordinates are returned.
1703
block = self._dirblocks[block_index][1]
1704
entry_index, present = self._find_entry_index(key, block)
1706
entry = self._dirblocks[block_index][1][entry_index]
1707
if entry[1][tree_index][0] in 'fdlt':
1708
# this is the result we are looking for: the
1709
# real home of this file_id in this tree.
1711
if entry[1][tree_index][0] == 'a':
1712
# there is no home for this entry in this tree
1714
assert entry[1][tree_index][0] == 'r', \
1715
"entry %r has invalid minikind %r for tree %r" \
1717
entry[1][tree_index][0],
1719
real_path = entry[1][tree_index][1]
1720
return self._get_entry(tree_index, fileid_utf8=fileid_utf8,
1721
path_utf8=real_path)
1725
def initialize(cls, path):
1726
"""Create a new dirstate on path.
1728
The new dirstate will be an empty tree - that is it has no parents,
1729
and only a root node - which has id ROOT_ID.
1731
:param path: The name of the file for the dirstate.
1732
:return: A write-locked DirState object.
1734
# This constructs a new DirState object on a path, sets the _state_file
1735
# to a new empty file for that path. It then calls _set_data() with our
1736
# stock empty dirstate information - a root with ROOT_ID, no children,
1737
# and no parents. Finally it calls save() to ensure that this data will
1740
# root dir and root dir contents with no children.
1741
empty_tree_dirblocks = [('', []), ('', [])]
1742
# a new root directory, with a NULLSTAT.
1743
empty_tree_dirblocks[0][1].append(
1744
(('', '', inventory.ROOT_ID), [
1745
('d', '', 0, False, DirState.NULLSTAT),
1749
result._set_data([], empty_tree_dirblocks)
1756
def _inv_entry_to_details(self, inv_entry):
1757
"""Convert an inventory entry (from a revision tree) to state details.
1759
:param inv_entry: An inventory entry whose sha1 and link targets can be
1760
relied upon, and which has a revision set.
1761
:return: A details tuple - the details for a single tree at a path +
1764
kind = inv_entry.kind
1765
minikind = DirState._kind_to_minikind[kind]
1766
tree_data = inv_entry.revision
1767
assert tree_data, 'empty revision for the inv_entry %s.' % \
1769
if kind == 'directory':
1773
elif kind == 'symlink':
1774
fingerprint = inv_entry.symlink_target or ''
1777
elif kind == 'file':
1778
fingerprint = inv_entry.text_sha1 or ''
1779
size = inv_entry.text_size or 0
1780
executable = inv_entry.executable
1781
elif kind == 'tree-reference':
1782
fingerprint = inv_entry.reference_revision or ''
1786
raise Exception("can't pack %s" % inv_entry)
1787
return (minikind, fingerprint, size, executable, tree_data)
1789
def _iter_child_entries(self, tree_index, path_utf8):
1790
"""Iterate over all the entries that are children of path_utf.
1792
This only returns entries that are present (not in 'a', 'r') in
1793
tree_index. tree_index data is not refreshed, so if tree 0 is used,
1794
results may differ from that obtained if paths were statted to
1795
determine what ones were directories.
1797
Asking for the children of a non-directory will return an empty
1801
next_pending_dirs = [path_utf8]
1803
while next_pending_dirs:
1804
pending_dirs = next_pending_dirs
1805
next_pending_dirs = []
1806
for path in pending_dirs:
1807
block_index, present = self._find_block_index_from_key(
1809
if block_index == 0:
1811
if len(self._dirblocks) == 1:
1812
# asked for the children of the root with no other
1816
# children of a non-directory asked for.
1818
block = self._dirblocks[block_index]
1819
for entry in block[1]:
1820
kind = entry[1][tree_index][0]
1821
if kind not in absent:
1825
path = entry[0][0] + '/' + entry[0][1]
1828
next_pending_dirs.append(path)
1830
def _iter_entries(self):
1831
"""Iterate over all the entries in the dirstate.
1833
Each yelt item is an entry in the standard format described in the
1834
docstring of bzrlib.dirstate.
1836
self._read_dirblocks_if_needed()
1837
for directory in self._dirblocks:
1838
for entry in directory[1]:
1841
def _get_id_index(self):
1842
"""Get an id index of self._dirblocks."""
1843
if self._id_index is None:
1845
for key, tree_details in self._iter_entries():
1846
id_index.setdefault(key[2], set()).add(key)
1847
self._id_index = id_index
1848
return self._id_index
1850
def _get_output_lines(self, lines):
1851
"""Format lines for final output.
1853
:param lines: A sequence of lines containing the parents list and the
1856
output_lines = [DirState.HEADER_FORMAT_3]
1857
lines.append('') # a final newline
1858
inventory_text = '\0\n\0'.join(lines)
1859
output_lines.append('crc32: %s\n' % (zlib.crc32(inventory_text),))
1860
# -3, 1 for num parents, 1 for ghosts, 1 for final newline
1861
num_entries = len(lines)-3
1862
output_lines.append('num_entries: %s\n' % (num_entries,))
1863
output_lines.append(inventory_text)
1866
def _make_deleted_row(self, fileid_utf8, parents):
1867
"""Return a deleted row for fileid_utf8."""
1868
return ('/', 'RECYCLED.BIN', 'file', fileid_utf8, 0, DirState.NULLSTAT,
1871
def _num_present_parents(self):
1872
"""The number of parent entries in each record row."""
1873
return len(self._parents) - len(self._ghosts)
1877
"""Construct a DirState on the file at path path.
1879
:return: An unlocked DirState object, associated with the given path.
1881
result = DirState(path)
1884
def _read_dirblocks_if_needed(self):
1885
"""Read in all the dirblocks from the file if they are not in memory.
1887
This populates self._dirblocks, and sets self._dirblock_state to
1888
IN_MEMORY_UNMODIFIED. It is not currently ready for incremental block
1891
self._read_header_if_needed()
1892
if self._dirblock_state == DirState.NOT_IN_MEMORY:
1893
_read_dirblocks(self)
1895
def _read_header(self):
1896
"""This reads in the metadata header, and the parent ids.
1898
After reading in, the file should be positioned at the null
1899
just before the start of the first record in the file.
1901
:return: (expected crc checksum, number of entries, parent list)
1903
self._read_prelude()
1904
parent_line = self._state_file.readline()
1905
info = parent_line.split('\0')
1906
num_parents = int(info[0])
1907
assert num_parents == len(info)-2, 'incorrect parent info line'
1908
self._parents = info[1:-1]
1910
ghost_line = self._state_file.readline()
1911
info = ghost_line.split('\0')
1912
num_ghosts = int(info[1])
1913
assert num_ghosts == len(info)-3, 'incorrect ghost info line'
1914
self._ghosts = info[2:-1]
1915
self._header_state = DirState.IN_MEMORY_UNMODIFIED
1916
self._end_of_header = self._state_file.tell()
1918
def _read_header_if_needed(self):
1919
"""Read the header of the dirstate file if needed."""
1920
# inline this as it will be called a lot
1921
if not self._lock_token:
1922
raise errors.ObjectNotLocked(self)
1923
if self._header_state == DirState.NOT_IN_MEMORY:
1926
def _read_prelude(self):
1927
"""Read in the prelude header of the dirstate file.
1929
This only reads in the stuff that is not connected to the crc
1930
checksum. The position will be correct to read in the rest of
1931
the file and check the checksum after this point.
1932
The next entry in the file should be the number of parents,
1933
and their ids. Followed by a newline.
1935
header = self._state_file.readline()
1936
assert header == DirState.HEADER_FORMAT_3, \
1937
'invalid header line: %r' % (header,)
1938
crc_line = self._state_file.readline()
1939
assert crc_line.startswith('crc32: '), 'missing crc32 checksum'
1940
self.crc_expected = int(crc_line[len('crc32: '):-1])
1941
num_entries_line = self._state_file.readline()
1942
assert num_entries_line.startswith('num_entries: '), 'missing num_entries line'
1943
self._num_entries = int(num_entries_line[len('num_entries: '):-1])
1945
def sha1_from_stat(self, path, stat_result, _pack_stat=pack_stat):
1946
"""Find a sha1 given a stat lookup."""
1947
return self._get_packed_stat_index().get(_pack_stat(stat_result), None)
1949
def _get_packed_stat_index(self):
1950
"""Get a packed_stat index of self._dirblocks."""
1951
if self._packed_stat_index is None:
1953
for key, tree_details in self._iter_entries():
1954
if tree_details[0][0] == 'f':
1955
index[tree_details[0][4]] = tree_details[0][1]
1956
self._packed_stat_index = index
1957
return self._packed_stat_index
1960
"""Save any pending changes created during this session.
1962
We reuse the existing file, because that prevents race conditions with
1963
file creation, and use oslocks on it to prevent concurrent modification
1964
and reads - because dirstate's incremental data aggregation is not
1965
compatible with reading a modified file, and replacing a file in use by
1966
another process is impossible on Windows.
1968
A dirstate in read only mode should be smart enough though to validate
1969
that the file has not changed, and otherwise discard its cache and
1970
start over, to allow for fine grained read lock duration, so 'status'
1971
wont block 'commit' - for example.
1973
if self._changes_aborted:
1974
# Should this be a warning? For now, I'm expecting that places that
1975
# mark it inconsistent will warn, making a warning here redundant.
1976
trace.mutter('Not saving DirState because '
1977
'_changes_aborted is set.')
1979
if (self._header_state == DirState.IN_MEMORY_MODIFIED or
1980
self._dirblock_state == DirState.IN_MEMORY_MODIFIED):
1982
grabbed_write_lock = False
1983
if self._lock_state != 'w':
1984
grabbed_write_lock, new_lock = self._lock_token.temporary_write_lock()
1985
# Switch over to the new lock, as the old one may be closed.
1986
# TODO: jam 20070315 We should validate the disk file has
1987
# not changed contents. Since temporary_write_lock may
1988
# not be an atomic operation.
1989
self._lock_token = new_lock
1990
self._state_file = new_lock.f
1991
if not grabbed_write_lock:
1992
# We couldn't grab a write lock, so we switch back to a read one
1995
self._state_file.seek(0)
1996
self._state_file.writelines(self.get_lines())
1997
self._state_file.truncate()
1998
self._state_file.flush()
1999
self._header_state = DirState.IN_MEMORY_UNMODIFIED
2000
self._dirblock_state = DirState.IN_MEMORY_UNMODIFIED
2002
if grabbed_write_lock:
2003
self._lock_token = self._lock_token.restore_read_lock()
2004
self._state_file = self._lock_token.f
2005
# TODO: jam 20070315 We should validate the disk file has
2006
# not changed contents. Since restore_read_lock may
2007
# not be an atomic operation.
2009
def _set_data(self, parent_ids, dirblocks):
2010
"""Set the full dirstate data in memory.
2012
This is an internal function used to completely replace the objects
2013
in memory state. It puts the dirstate into state 'full-dirty'.
2015
:param parent_ids: A list of parent tree revision ids.
2016
:param dirblocks: A list containing one tuple for each directory in the
2017
tree. Each tuple contains the directory path and a list of entries
2018
found in that directory.
2020
# our memory copy is now authoritative.
2021
self._dirblocks = dirblocks
2022
self._header_state = DirState.IN_MEMORY_MODIFIED
2023
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
2024
self._parents = list(parent_ids)
2025
self._id_index = None
2026
self._packed_stat_index = None
2028
def set_path_id(self, path, new_id):
2029
"""Change the id of path to new_id in the current working tree.
2031
:param path: The path inside the tree to set - '' is the root, 'foo'
2032
is the path foo in the root.
2033
:param new_id: The new id to assign to the path. This must be a utf8
2034
file id (not unicode, and not None).
2036
assert new_id.__class__ == str, \
2037
"path_id %r is not a plain string" % (new_id,)
2038
self._read_dirblocks_if_needed()
2040
# TODO: logic not written
2041
raise NotImplementedError(self.set_path_id)
2042
# TODO: check new id is unique
2043
entry = self._get_entry(0, path_utf8=path)
2044
if entry[0][2] == new_id:
2045
# Nothing to change.
2047
# mark the old path absent, and insert a new root path
2048
self._make_absent(entry)
2049
self.update_minimal(('', '', new_id), 'd',
2050
path_utf8='', packed_stat=entry[1][0][4])
2051
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
2052
if self._id_index is not None:
2053
self._id_index.setdefault(new_id, set()).add(entry[0])
2055
def set_parent_trees(self, trees, ghosts):
2056
"""Set the parent trees for the dirstate.
2058
:param trees: A list of revision_id, tree tuples. tree must be provided
2059
even if the revision_id refers to a ghost: supply an empty tree in
2061
:param ghosts: A list of the revision_ids that are ghosts at the time
2064
# TODO: generate a list of parent indexes to preserve to save
2065
# processing specific parent trees. In the common case one tree will
2066
# be preserved - the left most parent.
2067
# TODO: if the parent tree is a dirstate, we might want to walk them
2068
# all by path in parallel for 'optimal' common-case performance.
2069
# generate new root row.
2070
self._read_dirblocks_if_needed()
2071
# TODO future sketch: Examine the existing parents to generate a change
2072
# map and then walk the new parent trees only, mapping them into the
2073
# dirstate. Walk the dirstate at the same time to remove unreferenced
2076
# sketch: loop over all entries in the dirstate, cherry picking
2077
# entries from the parent trees, if they are not ghost trees.
2078
# after we finish walking the dirstate, all entries not in the dirstate
2079
# are deletes, so we want to append them to the end as per the design
2080
# discussions. So do a set difference on ids with the parents to
2081
# get deletes, and add them to the end.
2082
# During the update process we need to answer the following questions:
2083
# - find other keys containing a fileid in order to create cross-path
2084
# links. We dont't trivially use the inventory from other trees
2085
# because this leads to either double touching, or to accessing
2087
# - find other keys containing a path
2088
# We accumulate each entry via this dictionary, including the root
2091
# we could do parallel iterators, but because file id data may be
2092
# scattered throughout, we dont save on index overhead: we have to look
2093
# at everything anyway. We can probably save cycles by reusing parent
2094
# data and doing an incremental update when adding an additional
2095
# parent, but for now the common cases are adding a new parent (merge),
2096
# and replacing completely (commit), and commit is more common: so
2097
# optimise merge later.
2099
# ---- start generation of full tree mapping data
2100
# what trees should we use?
2101
parent_trees = [tree for rev_id, tree in trees if rev_id not in ghosts]
2102
# how many trees do we end up with
2103
parent_count = len(parent_trees)
2105
# one: the current tree
2106
for entry in self._iter_entries():
2107
# skip entries not in the current tree
2108
if entry[1][0][0] in 'ar': # absent, relocated
2110
by_path[entry[0]] = [entry[1][0]] + \
2111
[DirState.NULL_PARENT_DETAILS] * parent_count
2112
id_index[entry[0][2]] = set([entry[0]])
2114
# now the parent trees:
2115
for tree_index, tree in enumerate(parent_trees):
2116
# the index is off by one, adjust it.
2117
tree_index = tree_index + 1
2118
# when we add new locations for a fileid we need these ranges for
2119
# any fileid in this tree as we set the by_path[id] to:
2120
# already_processed_tree_details + new_details + new_location_suffix
2121
# the suffix is from tree_index+1:parent_count+1.
2122
new_location_suffix = [DirState.NULL_PARENT_DETAILS] * (parent_count - tree_index)
2123
# now stitch in all the entries from this tree
2124
for path, entry in tree.inventory.iter_entries_by_dir():
2125
# here we process each trees details for each item in the tree.
2126
# we first update any existing entries for the id at other paths,
2127
# then we either create or update the entry for the id at the
2128
# right path, and finally we add (if needed) a mapping from
2129
# file_id to this path. We do it in this order to allow us to
2130
# avoid checking all known paths for the id when generating a
2131
# new entry at this path: by adding the id->path mapping last,
2132
# all the mappings are valid and have correct relocation
2133
# records where needed.
2134
file_id = entry.file_id
2135
path_utf8 = path.encode('utf8')
2136
dirname, basename = osutils.split(path_utf8)
2137
new_entry_key = (dirname, basename, file_id)
2138
# tree index consistency: All other paths for this id in this tree
2139
# index must point to the correct path.
2140
for entry_key in id_index.setdefault(file_id, set()):
2141
# TODO:PROFILING: It might be faster to just update
2142
# rather than checking if we need to, and then overwrite
2143
# the one we are located at.
2144
if entry_key != new_entry_key:
2145
# this file id is at a different path in one of the
2146
# other trees, so put absent pointers there
2147
# This is the vertical axis in the matrix, all pointing
2149
by_path[entry_key][tree_index] = ('r', path_utf8, 0, False, '')
2150
# by path consistency: Insert into an existing path record (trivial), or
2151
# add a new one with relocation pointers for the other tree indexes.
2152
if new_entry_key in id_index[file_id]:
2153
# there is already an entry where this data belongs, just insert it.
2154
by_path[new_entry_key][tree_index] = \
2155
self._inv_entry_to_details(entry)
2157
# add relocated entries to the horizontal axis - this row
2158
# mapping from path,id. We need to look up the correct path
2159
# for the indexes from 0 to tree_index -1
2161
for lookup_index in xrange(tree_index):
2162
# boundary case: this is the first occurence of file_id
2163
# so there are no id_indexs, possibly take this out of
2165
if not len(id_index[file_id]):
2166
new_details.append(DirState.NULL_PARENT_DETAILS)
2168
# grab any one entry, use it to find the right path.
2169
# TODO: optimise this to reduce memory use in highly
2170
# fragmented situations by reusing the relocation
2172
a_key = iter(id_index[file_id]).next()
2173
if by_path[a_key][lookup_index][0] in ('r', 'a'):
2174
# its a pointer or missing statement, use it as is.
2175
new_details.append(by_path[a_key][lookup_index])
2177
# we have the right key, make a pointer to it.
2178
real_path = ('/'.join(a_key[0:2])).strip('/')
2179
new_details.append(('r', real_path, 0, False, ''))
2180
new_details.append(self._inv_entry_to_details(entry))
2181
new_details.extend(new_location_suffix)
2182
by_path[new_entry_key] = new_details
2183
id_index[file_id].add(new_entry_key)
2184
# --- end generation of full tree mappings
2186
# sort and output all the entries
2187
new_entries = self._sort_entries(by_path.items())
2188
self._entries_to_current_state(new_entries)
2189
self._parents = [rev_id for rev_id, tree in trees]
2190
self._ghosts = list(ghosts)
2191
self._header_state = DirState.IN_MEMORY_MODIFIED
2192
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
2193
self._id_index = id_index
2195
def _sort_entries(self, entry_list):
2196
"""Given a list of entries, sort them into the right order.
2198
This is done when constructing a new dirstate from trees - normally we
2199
try to keep everything in sorted blocks all the time, but sometimes
2200
it's easier to sort after the fact.
2203
# sort by: directory parts, file name, file id
2204
return entry[0][0].split('/'), entry[0][1], entry[0][2]
2205
return sorted(entry_list, key=_key)
2207
def set_state_from_inventory(self, new_inv):
2208
"""Set new_inv as the current state.
2210
This API is called by tree transform, and will usually occur with
2211
existing parent trees.
2213
:param new_inv: The inventory object to set current state from.
2215
if 'evil' in debug.debug_flags:
2216
trace.mutter_callsite(1,
2217
"set_state_from_inventory called; please mutate the tree instead")
2218
self._read_dirblocks_if_needed()
2220
# Two iterators: current data and new data, both in dirblock order.
2221
# We zip them together, which tells about entries that are new in the
2222
# inventory, or removed in the inventory, or present in both and
2225
# You might think we could just synthesize a new dirstate directly
2226
# since we're processing it in the right order. However, we need to
2227
# also consider there may be any number of parent trees and relocation
2228
# pointers, and we don't want to duplicate that here.
2229
new_iterator = new_inv.iter_entries_by_dir()
2230
# we will be modifying the dirstate, so we need a stable iterator. In
2231
# future we might write one, for now we just clone the state into a
2232
# list - which is a shallow copy.
2233
old_iterator = iter(list(self._iter_entries()))
2234
# both must have roots so this is safe:
2235
current_new = new_iterator.next()
2236
current_old = old_iterator.next()
2237
def advance(iterator):
2239
return iterator.next()
2240
except StopIteration:
2242
while current_new or current_old:
2243
# skip entries in old that are not really there
2244
if current_old and current_old[1][0][0] in 'ar':
2245
# relocated or absent
2246
current_old = advance(old_iterator)
2249
# convert new into dirblock style
2250
new_path_utf8 = current_new[0].encode('utf8')
2251
new_dirname, new_basename = osutils.split(new_path_utf8)
2252
new_id = current_new[1].file_id
2253
new_entry_key = (new_dirname, new_basename, new_id)
2254
current_new_minikind = \
2255
DirState._kind_to_minikind[current_new[1].kind]
2256
if current_new_minikind == 't':
2257
fingerprint = current_new[1].reference_revision or ''
2259
# We normally only insert or remove records, or update
2260
# them when it has significantly changed. Then we want to
2261
# erase its fingerprint. Unaffected records should
2262
# normally not be updated at all.
2265
# for safety disable variables
2266
new_path_utf8 = new_dirname = new_basename = new_id = \
2267
new_entry_key = None
2268
# 5 cases, we dont have a value that is strictly greater than everything, so
2269
# we make both end conditions explicit
2271
# old is finished: insert current_new into the state.
2272
self.update_minimal(new_entry_key, current_new_minikind,
2273
executable=current_new[1].executable,
2274
path_utf8=new_path_utf8, fingerprint=fingerprint)
2275
current_new = advance(new_iterator)
2276
elif not current_new:
2278
self._make_absent(current_old)
2279
current_old = advance(old_iterator)
2280
elif new_entry_key == current_old[0]:
2281
# same - common case
2282
# We're looking at the same path and id in both the dirstate
2283
# and inventory, so just need to update the fields in the
2284
# dirstate from the one in the inventory.
2285
# TODO: update the record if anything significant has changed.
2286
# the minimal required trigger is if the execute bit or cached
2288
if (current_old[1][0][3] != current_new[1].executable or
2289
current_old[1][0][0] != current_new_minikind):
2290
self.update_minimal(current_old[0], current_new_minikind,
2291
executable=current_new[1].executable,
2292
path_utf8=new_path_utf8, fingerprint=fingerprint)
2293
# both sides are dealt with, move on
2294
current_old = advance(old_iterator)
2295
current_new = advance(new_iterator)
2296
elif (cmp_by_dirs(new_dirname, current_old[0][0]) < 0
2297
or (new_dirname == current_old[0][0]
2298
and new_entry_key[1:] < current_old[0][1:])):
2300
# add a entry for this and advance new
2301
self.update_minimal(new_entry_key, current_new_minikind,
2302
executable=current_new[1].executable,
2303
path_utf8=new_path_utf8, fingerprint=fingerprint)
2304
current_new = advance(new_iterator)
2306
# we've advanced past the place where the old key would be,
2307
# without seeing it in the new list. so it must be gone.
2308
self._make_absent(current_old)
2309
current_old = advance(old_iterator)
2310
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
2311
self._id_index = None
2312
self._packed_stat_index = None
2314
def _make_absent(self, current_old):
2315
"""Mark current_old - an entry - as absent for tree 0.
2317
:return: True if this was the last details entry for the entry key:
2318
that is, if the underlying block has had the entry removed, thus
2319
shrinking in length.
2321
# build up paths that this id will be left at after the change is made,
2322
# so we can update their cross references in tree 0
2323
all_remaining_keys = set()
2324
# Dont check the working tree, because it's going.
2325
for details in current_old[1][1:]:
2326
if details[0] not in 'ar': # absent, relocated
2327
all_remaining_keys.add(current_old[0])
2328
elif details[0] == 'r': # relocated
2329
# record the key for the real path.
2330
all_remaining_keys.add(tuple(osutils.split(details[1])) + (current_old[0][2],))
2331
# absent rows are not present at any path.
2332
last_reference = current_old[0] not in all_remaining_keys
2334
# the current row consists entire of the current item (being marked
2335
# absent), and relocated or absent entries for the other trees:
2336
# Remove it, its meaningless.
2337
block = self._find_block(current_old[0])
2338
entry_index, present = self._find_entry_index(current_old[0], block[1])
2339
assert present, 'could not find entry for %s' % (current_old,)
2340
block[1].pop(entry_index)
2341
# if we have an id_index in use, remove this key from it for this id.
2342
if self._id_index is not None:
2343
self._id_index[current_old[0][2]].remove(current_old[0])
2344
# update all remaining keys for this id to record it as absent. The
2345
# existing details may either be the record we are marking as deleted
2346
# (if there were other trees with the id present at this path), or may
2348
for update_key in all_remaining_keys:
2349
update_block_index, present = \
2350
self._find_block_index_from_key(update_key)
2351
assert present, 'could not find block for %s' % (update_key,)
2352
update_entry_index, present = \
2353
self._find_entry_index(update_key, self._dirblocks[update_block_index][1])
2354
assert present, 'could not find entry for %s' % (update_key,)
2355
update_tree_details = self._dirblocks[update_block_index][1][update_entry_index][1]
2356
# it must not be absent at the moment
2357
assert update_tree_details[0][0] != 'a' # absent
2358
update_tree_details[0] = DirState.NULL_PARENT_DETAILS
2359
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
2360
return last_reference
2362
def update_minimal(self, key, minikind, executable=False, fingerprint='',
2363
packed_stat=None, size=0, path_utf8=None):
2364
"""Update an entry to the state in tree 0.
2366
This will either create a new entry at 'key' or update an existing one.
2367
It also makes sure that any other records which might mention this are
2370
:param key: (dir, name, file_id) for the new entry
2371
:param minikind: The type for the entry ('f' == 'file', 'd' ==
2373
:param executable: Should the executable bit be set?
2374
:param fingerprint: Simple fingerprint for new entry: sha1 for files,
2375
referenced revision id for subtrees, etc.
2376
:param packed_stat: Packed stat value for new entry.
2377
:param size: Size information for new entry
2378
:param path_utf8: key[0] + '/' + key[1], just passed in to avoid doing
2381
If packed_stat and fingerprint are not given, they're invalidated in
2384
block = self._find_block(key)[1]
2385
if packed_stat is None:
2386
packed_stat = DirState.NULLSTAT
2387
# XXX: Some callers pass '' as the packed_stat, and it seems to be
2388
# sometimes present in the dirstate - this seems oddly inconsistent.
2390
entry_index, present = self._find_entry_index(key, block)
2391
new_details = (minikind, fingerprint, size, executable, packed_stat)
2392
id_index = self._get_id_index()
2394
# new entry, synthesis cross reference here,
2395
existing_keys = id_index.setdefault(key[2], set())
2396
if not existing_keys:
2397
# not currently in the state, simplest case
2398
new_entry = key, [new_details] + self._empty_parent_info()
2400
# present at one or more existing other paths.
2401
# grab one of them and use it to generate parent
2402
# relocation/absent entries.
2403
new_entry = key, [new_details]
2404
for other_key in existing_keys:
2405
# change the record at other to be a pointer to this new
2406
# record. The loop looks similar to the change to
2407
# relocations when updating an existing record but its not:
2408
# the test for existing kinds is different: this can be
2409
# factored out to a helper though.
2410
other_block_index, present = self._find_block_index_from_key(other_key)
2411
assert present, 'could not find block for %s' % (other_key,)
2412
other_entry_index, present = self._find_entry_index(other_key,
2413
self._dirblocks[other_block_index][1])
2414
assert present, 'could not find entry for %s' % (other_key,)
2415
assert path_utf8 is not None
2416
self._dirblocks[other_block_index][1][other_entry_index][1][0] = \
2417
('r', path_utf8, 0, False, '')
2419
num_present_parents = self._num_present_parents()
2420
for lookup_index in xrange(1, num_present_parents + 1):
2421
# grab any one entry, use it to find the right path.
2422
# TODO: optimise this to reduce memory use in highly
2423
# fragmented situations by reusing the relocation
2425
update_block_index, present = \
2426
self._find_block_index_from_key(other_key)
2427
assert present, 'could not find block for %s' % (other_key,)
2428
update_entry_index, present = \
2429
self._find_entry_index(other_key, self._dirblocks[update_block_index][1])
2430
assert present, 'could not find entry for %s' % (other_key,)
2431
update_details = self._dirblocks[update_block_index][1][update_entry_index][1][lookup_index]
2432
if update_details[0] in 'ar': # relocated, absent
2433
# its a pointer or absent in lookup_index's tree, use
2435
new_entry[1].append(update_details)
2437
# we have the right key, make a pointer to it.
2438
pointer_path = osutils.pathjoin(*other_key[0:2])
2439
new_entry[1].append(('r', pointer_path, 0, False, ''))
2440
block.insert(entry_index, new_entry)
2441
existing_keys.add(key)
2443
# Does the new state matter?
2444
block[entry_index][1][0] = new_details
2445
# parents cannot be affected by what we do.
2446
# other occurences of this id can be found
2447
# from the id index.
2449
# tree index consistency: All other paths for this id in this tree
2450
# index must point to the correct path. We have to loop here because
2451
# we may have passed entries in the state with this file id already
2452
# that were absent - where parent entries are - and they need to be
2453
# converted to relocated.
2454
assert path_utf8 is not None
2455
for entry_key in id_index.setdefault(key[2], set()):
2456
# TODO:PROFILING: It might be faster to just update
2457
# rather than checking if we need to, and then overwrite
2458
# the one we are located at.
2459
if entry_key != key:
2460
# this file id is at a different path in one of the
2461
# other trees, so put absent pointers there
2462
# This is the vertical axis in the matrix, all pointing
2464
block_index, present = self._find_block_index_from_key(entry_key)
2466
entry_index, present = self._find_entry_index(entry_key, self._dirblocks[block_index][1])
2468
self._dirblocks[block_index][1][entry_index][1][0] = \
2469
('r', path_utf8, 0, False, '')
2470
# add a containing dirblock if needed.
2471
if new_details[0] == 'd':
2472
subdir_key = (osutils.pathjoin(*key[0:2]), '', '')
2473
block_index, present = self._find_block_index_from_key(subdir_key)
2475
self._dirblocks.insert(block_index, (subdir_key[0], []))
2477
self._dirblock_state = DirState.IN_MEMORY_MODIFIED
2479
def _validate(self):
2480
"""Check that invariants on the dirblock are correct.
2482
This can be useful in debugging; it shouldn't be necessary in
2485
This must be called with a lock held.
2487
# NOTE: This must always raise AssertionError not just assert,
2488
# otherwise it may not behave properly under python -O
2490
# TODO: All entries must have some content that's not 'a' or 'r',
2491
# otherwise it could just be removed.
2493
# TODO: All relocations must point directly to a real entry.
2495
# TODO: No repeated keys.
2498
from pprint import pformat
2499
self._read_dirblocks_if_needed()
2500
if len(self._dirblocks) > 0:
2501
if not self._dirblocks[0][0] == '':
2502
raise AssertionError(
2503
"dirblocks don't start with root block:\n" + \
2505
if len(self._dirblocks) > 1:
2506
if not self._dirblocks[1][0] == '':
2507
raise AssertionError(
2508
"dirblocks missing root directory:\n" + \
2510
# the dirblocks are sorted by their path components, name, and dir id
2511
dir_names = [d[0].split('/')
2512
for d in self._dirblocks[1:]]
2513
if dir_names != sorted(dir_names):
2514
raise AssertionError(
2515
"dir names are not in sorted order:\n" + \
2516
pformat(self._dirblocks) + \
2519
for dirblock in self._dirblocks:
2520
# within each dirblock, the entries are sorted by filename and
2522
for entry in dirblock[1]:
2523
if dirblock[0] != entry[0][0]:
2524
raise AssertionError(
2526
"doesn't match directory name in\n%r" %
2527
(entry, pformat(dirblock)))
2528
if dirblock[1] != sorted(dirblock[1]):
2529
raise AssertionError(
2530
"dirblock for %r is not sorted:\n%s" % \
2531
(dirblock[0], pformat(dirblock)))
2533
def check_valid_parent():
2534
"""Check that the current entry has a valid parent.
2536
This makes sure that the parent has a record,
2537
and that the parent isn't marked as "absent" in the
2538
current tree. (It is invalid to have a non-absent file in an absent
2541
if entry[0][0:2] == ('', ''):
2542
# There should be no parent for the root row
2544
parent_entry = self._get_entry(tree_index, path_utf8=entry[0][0])
2545
if parent_entry == (None, None):
2546
raise AssertionError(
2547
"no parent entry for: %s in tree %s"
2548
% (this_path, tree_index))
2549
if parent_entry[1][tree_index][0] != 'd':
2550
raise AssertionError(
2551
"Parent entry for %s is not marked as a valid"
2552
" directory. %s" % (this_path, parent_entry,))
2554
# For each file id, for each tree: either
2555
# the file id is not present at all; all rows with that id in the
2556
# key have it marked as 'absent'
2557
# OR the file id is present under exactly one name; any other entries
2558
# that mention that id point to the correct name.
2560
# We check this with a dict per tree pointing either to the present
2561
# name, or None if absent.
2562
tree_count = self._num_present_parents() + 1
2563
id_path_maps = [dict() for i in range(tree_count)]
2564
# Make sure that all renamed entries point to the correct location.
2565
for entry in self._iter_entries():
2566
file_id = entry[0][2]
2567
this_path = osutils.pathjoin(entry[0][0], entry[0][1])
2568
if len(entry[1]) != tree_count:
2569
raise AssertionError(
2570
"wrong number of entry details for row\n%s" \
2571
",\nexpected %d" % \
2572
(pformat(entry), tree_count))
2573
absent_positions = 0
2574
for tree_index, tree_state in enumerate(entry[1]):
2575
this_tree_map = id_path_maps[tree_index]
2576
minikind = tree_state[0]
2577
if minikind in 'ar':
2578
absent_positions += 1
2579
# have we seen this id before in this column?
2580
if file_id in this_tree_map:
2581
previous_path, previous_loc = this_tree_map[file_id]
2582
# any later mention of this file must be consistent with
2583
# what was said before
2585
if previous_path is not None:
2586
raise AssertionError(
2587
"file %s is absent in row %r but also present " \
2589
(file_id, entry, previous_path))
2590
elif minikind == 'r':
2591
target_location = tree_state[1]
2592
if previous_path != target_location:
2593
raise AssertionError(
2594
"file %s relocation in row %r but also at %r" \
2595
% (file_id, entry, previous_path))
2597
# a file, directory, etc - may have been previously
2598
# pointed to by a relocation, which must point here
2599
if previous_path != this_path:
2600
raise AssertionError(
2601
"entry %r inconsistent with previous path %r "
2603
(entry, previous_path, previous_loc))
2604
check_valid_parent()
2607
# absent; should not occur anywhere else
2608
this_tree_map[file_id] = None, this_path
2609
elif minikind == 'r':
2610
# relocation, must occur at expected location
2611
this_tree_map[file_id] = tree_state[1], this_path
2613
this_tree_map[file_id] = this_path, this_path
2614
check_valid_parent()
2615
if absent_positions == tree_count:
2616
raise AssertionError(
2617
"entry %r has no data for any tree." % (entry,))
2619
def _wipe_state(self):
2620
"""Forget all state information about the dirstate."""
2621
self._header_state = DirState.NOT_IN_MEMORY
2622
self._dirblock_state = DirState.NOT_IN_MEMORY
2623
self._changes_aborted = False
2626
self._dirblocks = []
2627
self._id_index = None
2628
self._packed_stat_index = None
2629
self._end_of_header = None
2630
self._cutoff_time = None
2631
self._split_path_cache = {}
2633
def lock_read(self):
2634
"""Acquire a read lock on the dirstate."""
2635
if self._lock_token is not None:
2636
raise errors.LockContention(self._lock_token)
2637
# TODO: jam 20070301 Rather than wiping completely, if the blocks are
2638
# already in memory, we could read just the header and check for
2639
# any modification. If not modified, we can just leave things
2641
self._lock_token = lock.ReadLock(self._filename)
2642
self._lock_state = 'r'
2643
self._state_file = self._lock_token.f
2646
def lock_write(self):
2647
"""Acquire a write lock on the dirstate."""
2648
if self._lock_token is not None:
2649
raise errors.LockContention(self._lock_token)
2650
# TODO: jam 20070301 Rather than wiping completely, if the blocks are
2651
# already in memory, we could read just the header and check for
2652
# any modification. If not modified, we can just leave things
2654
self._lock_token = lock.WriteLock(self._filename)
2655
self._lock_state = 'w'
2656
self._state_file = self._lock_token.f
2660
"""Drop any locks held on the dirstate."""
2661
if self._lock_token is None:
2662
raise errors.LockNotHeld(self)
2663
# TODO: jam 20070301 Rather than wiping completely, if the blocks are
2664
# already in memory, we could read just the header and check for
2665
# any modification. If not modified, we can just leave things
2667
self._state_file = None
2668
self._lock_state = None
2669
self._lock_token.unlock()
2670
self._lock_token = None
2671
self._split_path_cache = {}
2673
def _requires_lock(self):
2674
"""Check that a lock is currently held by someone on the dirstate."""
2675
if not self._lock_token:
2676
raise errors.ObjectNotLocked(self)
2679
# Try to load the compiled form if possible
2681
from bzrlib._dirstate_helpers_c import (
2682
_read_dirblocks_c as _read_dirblocks,
2683
bisect_dirblock_c as bisect_dirblock,
2684
_bisect_path_left_c as _bisect_path_left,
2685
_bisect_path_right_c as _bisect_path_right,
2686
cmp_by_dirs_c as cmp_by_dirs,
2689
from bzrlib._dirstate_helpers_py import (
2690
_read_dirblocks_py as _read_dirblocks,
2691
bisect_dirblock_py as bisect_dirblock,
2692
_bisect_path_left_py as _bisect_path_left,
2693
_bisect_path_right_py as _bisect_path_right,
2694
cmp_by_dirs_py as cmp_by_dirs,
added
removed
bzrlib/dirstate.py
