« back to all changes in this revision
Viewing changes to bzrlib/branch.py
- browse files at revision 521
- compare with another revision
- download diff
- download tarball
- view history from revision 521
- Committer: Martin Pool
- Date: 2005-05-17 07:01:47 UTC
- Revision ID: mbp@sourcefrog.net-20050517070147-c38da17418ea6711
- Add patch to give symlink support
- files added:
- bzrlib/atomicfile.py
- patches
- files removed:
- bzrlib/tests.py
- files modified:
- .bzrignore
added
removed
bzrlib/branch.py
24
24
import bzrlib
25
25
from inventory import Inventory
26
26
from trace import mutter, note
27
from tree import Tree, EmptyTree, RevisionTree, WorkingTree
27
from tree import Tree, EmptyTree, RevisionTree
28
28
from inventory import InventoryEntry, Inventory
29
29
from osutils import isdir, quotefn, isfile, uuid, sha_file, username, \
30
30
format_date, compact_date, pumpfile, user_email, rand_bytes, splitpath, \
33
33
from revision import Revision
34
34
from errors import bailout, BzrError
35
35
from textui import show_status
36
from diff import diff_trees
37
36
38
37
BZR_BRANCH_FORMAT = "Bazaar-NG branch, format 0.0.4\n"
39
38
## TODO: Maybe include checks for common corruption of newlines, etc?
40
39
41
40
42
41
42
def find_branch(f, **args):
43
if f and (f.startswith('http://') or f.startswith('https://')):
44
import remotebranch
45
return remotebranch.RemoteBranch(f, **args)
46
else:
47
return Branch(f, **args)
48
49
43
50
def find_branch_root(f=None):
44
51
"""Find the branch root enclosing f, or pwd.
45
52
53
f may be a filename or a URL.
54
46
55
It is not necessary that f exists.
47
56
48
57
Basically we keep looking up until we find the control directory or
53
62
f = os.path.realpath(f)
54
63
else:
55
64
f = os.path.abspath(f)
65
if not os.path.exists(f):
66
raise BzrError('%r does not exist' % f)
67
56
68
57
69
orig_f = f
58
70
73
85
class Branch:
74
86
"""Branch holding a history of revisions.
75
87
76
TODO: Perhaps use different stores for different classes of object,
77
so that we can keep track of how much space each one uses,
78
or garbage-collect them.
79
80
TODO: Add a RemoteBranch subclass. For the basic case of read-only
81
HTTP access this should be very easy by,
82
just redirecting controlfile access into HTTP requests.
83
We would need a RemoteStore working similarly.
84
85
TODO: Keep the on-disk branch locked while the object exists.
86
87
TODO: mkdir() method.
88
base
89
Base directory of the branch.
88
90
"""
89
def __init__(self, base, init=False, find_root=True):
91
_lockmode = None
92
93
def __init__(self, base, init=False, find_root=True, lock_mode='w'):
90
94
"""Create new branch object at a particular location.
91
95
92
96
base -- Base directory for the branch.
113
117
['use "bzr init" to initialize a new working tree',
114
118
'current bzr can only operate from top-of-tree'])
115
119
self._check_format()
120
self.lock(lock_mode)
116
121
117
122
self.text_store = ImmutableStore(self.controlfilename('text-store'))
118
123
self.revision_store = ImmutableStore(self.controlfilename('revision-store'))
126
131
__repr__ = __str__
127
132
128
133
134
135
def lock(self, mode='w'):
136
"""Lock the on-disk branch, excluding other processes."""
137
try:
138
import fcntl, errno
139
140
if mode == 'w':
141
lm = fcntl.LOCK_EX
142
om = os.O_WRONLY | os.O_CREAT
143
elif mode == 'r':
144
lm = fcntl.LOCK_SH
145
om = os.O_RDONLY
146
else:
147
raise BzrError("invalid locking mode %r" % mode)
148
149
try:
150
lockfile = os.open(self.controlfilename('branch-lock'), om)
151
except OSError, e:
152
if e.errno == errno.ENOENT:
153
# might not exist on branches from <0.0.4
154
self.controlfile('branch-lock', 'w').close()
155
lockfile = os.open(self.controlfilename('branch-lock'), om)
156
else:
157
raise e
158
159
fcntl.lockf(lockfile, lm)
160
def unlock():
161
fcntl.lockf(lockfile, fcntl.LOCK_UN)
162
os.close(lockfile)
163
self._lockmode = None
164
self.unlock = unlock
165
self._lockmode = mode
166
except ImportError:
167
warning("please write a locking method for platform %r" % sys.platform)
168
def unlock():
169
self._lockmode = None
170
self.unlock = unlock
171
self._lockmode = mode
172
173
174
def _need_readlock(self):
175
if self._lockmode not in ['r', 'w']:
176
raise BzrError('need read lock on branch, only have %r' % self._lockmode)
177
178
def _need_writelock(self):
179
if self._lockmode not in ['w']:
180
raise BzrError('need write lock on branch, only have %r' % self._lockmode)
181
182
129
183
def abspath(self, name):
130
184
"""Return absolute filename for something in the branch"""
131
185
return os.path.join(self.base, name)
158
212
and binary. binary files are untranslated byte streams. Text
159
213
control files are stored with Unix newlines and in UTF-8, even
160
214
if the platform or locale defaults are different.
215
216
Controlfiles should almost never be opened in write mode but
217
rather should be atomically copied and replaced using atomicfile.
161
218
"""
162
219
163
220
fn = self.controlfilename(file_or_path)
184
241
for d in ('text-store', 'inventory-store', 'revision-store'):
185
242
os.mkdir(self.controlfilename(d))
186
243
for f in ('revision-history', 'merged-patches',
187
'pending-merged-patches', 'branch-name'):
244
'pending-merged-patches', 'branch-name',
245
'branch-lock'):
188
246
self.controlfile(f, 'w').write('')
189
247
mutter('created control directory in ' + self.base)
190
248
Inventory().write_xml(self.controlfile('inventory','w'))
211
269
212
270
def read_working_inventory(self):
213
271
"""Read the working inventory."""
272
self._need_readlock()
214
273
before = time.time()
215
274
# ElementTree does its own conversion from UTF-8, so open in
216
275
# binary.
226
285
That is to say, the inventory describing changes underway, that
227
286
will be committed to the next revision.
228
287
"""
288
self._need_writelock()
229
289
## TODO: factor out to atomicfile? is rename safe on windows?
230
290
## TODO: Maybe some kind of clean/dirty marker on inventory?
231
291
tmpfname = self.controlfilename('inventory.tmp')
243
303
"""Inventory for the working copy.""")
244
304
245
305
246
def add(self, files, verbose=False):
306
def add(self, files, verbose=False, ids=None):
247
307
"""Make files versioned.
248
308
249
309
Note that the command line normally calls smart_add instead.
262
322
TODO: Adding a directory should optionally recurse down and
263
323
add all non-ignored children. Perhaps do that in a
264
324
higher-level method.
265
266
>>> b = ScratchBranch(files=['foo'])
267
>>> 'foo' in b.unknowns()
268
True
269
>>> b.show_status()
270
? foo
271
>>> b.add('foo')
272
>>> 'foo' in b.unknowns()
273
False
274
>>> bool(b.inventory.path2id('foo'))
275
True
276
>>> b.show_status()
277
A foo
278
279
>>> b.add('foo')
280
Traceback (most recent call last):
281
...
282
BzrError: ('foo is already versioned', [])
283
284
>>> b.add(['nothere'])
285
Traceback (most recent call last):
286
BzrError: ('cannot add: not a regular file or directory: nothere', [])
287
325
"""
326
self._need_writelock()
288
327
289
328
# TODO: Re-adding a file that is removed in the working copy
290
329
# should probably put it back with the previous ID.
291
330
if isinstance(files, types.StringTypes):
331
assert(ids is None or isinstance(ids, types.StringTypes))
292
332
files = [files]
333
if ids is not None:
334
ids = [ids]
335
336
if ids is None:
337
ids = [None] * len(files)
338
else:
339
assert(len(ids) == len(files))
293
340
294
341
inv = self.read_working_inventory()
295
for f in files:
342
for f,file_id in zip(files, ids):
296
343
if is_control_file(f):
297
344
bailout("cannot add control file %s" % quotefn(f))
298
345
312
359
if kind != 'file' and kind != 'directory':
313
360
bailout('cannot add: not a regular file or directory: %s' % quotefn(f))
314
361
315
file_id = gen_file_id(f)
362
if file_id is None:
363
file_id = gen_file_id(f)
316
364
inv.add_path(f, kind=kind, file_id=file_id)
317
365
318
366
if verbose:
325
373
326
374
def print_file(self, file, revno):
327
375
"""Print `file` to stdout."""
376
self._need_readlock()
328
377
tree = self.revision_tree(self.lookup_revision(revno))
329
378
# use inventory as it was in that revision
330
379
file_id = tree.inventory.path2id(file)
340
389
341
390
TODO: Refuse to remove modified files unless --force is given?
342
391
343
>>> b = ScratchBranch(files=['foo'])
344
>>> b.add('foo')
345
>>> b.inventory.has_filename('foo')
346
True
347
>>> b.remove('foo')
348
>>> b.working_tree().has_filename('foo')
349
True
350
>>> b.inventory.has_filename('foo')
351
False
352
353
>>> b = ScratchBranch(files=['foo'])
354
>>> b.add('foo')
355
>>> b.commit('one')
356
>>> b.remove('foo')
357
>>> b.commit('two')
358
>>> b.inventory.has_filename('foo')
359
False
360
>>> b.basis_tree().has_filename('foo')
361
False
362
>>> b.working_tree().has_filename('foo')
363
True
364
365
392
TODO: Do something useful with directories.
366
393
367
394
TODO: Should this remove the text or not? Tough call; not
371
398
"""
372
399
## TODO: Normalize names
373
400
## TODO: Remove nested loops; better scalability
401
self._need_writelock()
374
402
375
403
if isinstance(files, types.StringTypes):
376
404
files = [files]
395
423
396
424
self._write_inventory(inv)
397
425
426
def set_inventory(self, new_inventory_list):
427
inv = Inventory()
428
for path, file_id, parent, kind in new_inventory_list:
429
name = os.path.basename(path)
430
if name == "":
431
continue
432
inv.add(InventoryEntry(file_id, name, kind, parent))
433
self._write_inventory(inv)
434
398
435
399
436
def unknowns(self):
400
437
"""Return all unknown files.
415
452
return self.working_tree().unknowns()
416
453
417
454
418
def commit(self, message, timestamp=None, timezone=None,
419
committer=None,
420
verbose=False):
421
"""Commit working copy as a new revision.
422
423
The basic approach is to add all the file texts into the
424
store, then the inventory, then make a new revision pointing
425
to that inventory and store that.
426
427
This is not quite safe if the working copy changes during the
428
commit; for the moment that is simply not allowed. A better
429
approach is to make a temporary copy of the files before
430
computing their hashes, and then add those hashes in turn to
431
the inventory. This should mean at least that there are no
432
broken hash pointers. There is no way we can get a snapshot
433
of the whole directory at an instant. This would also have to
434
be robust against files disappearing, moving, etc. So the
435
whole thing is a bit hard.
436
437
timestamp -- if not None, seconds-since-epoch for a
438
postdated/predated commit.
439
"""
440
441
## TODO: Show branch names
442
443
# TODO: Don't commit if there are no changes, unless forced?
444
445
# First walk over the working inventory; and both update that
446
# and also build a new revision inventory. The revision
447
# inventory needs to hold the text-id, sha1 and size of the
448
# actual file versions committed in the revision. (These are
449
# not present in the working inventory.) We also need to
450
# detect missing/deleted files, and remove them from the
451
# working inventory.
452
453
work_inv = self.read_working_inventory()
454
inv = Inventory()
455
basis = self.basis_tree()
456
basis_inv = basis.inventory
457
missing_ids = []
458
for path, entry in work_inv.iter_entries():
459
## TODO: Cope with files that have gone missing.
460
461
## TODO: Check that the file kind has not changed from the previous
462
## revision of this file (if any).
463
464
entry = entry.copy()
465
466
p = self.abspath(path)
467
file_id = entry.file_id
468
mutter('commit prep file %s, id %r ' % (p, file_id))
469
470
if not os.path.exists(p):
471
mutter(" file is missing, removing from inventory")
472
if verbose:
473
show_status('D', entry.kind, quotefn(path))
474
missing_ids.append(file_id)
475
continue
476
477
# TODO: Handle files that have been deleted
478
479
# TODO: Maybe a special case for empty files? Seems a
480
# waste to store them many times.
481
482
inv.add(entry)
483
484
if basis_inv.has_id(file_id):
485
old_kind = basis_inv[file_id].kind
486
if old_kind != entry.kind:
487
bailout("entry %r changed kind from %r to %r"
488
% (file_id, old_kind, entry.kind))
489
490
if entry.kind == 'directory':
491
if not isdir(p):
492
bailout("%s is entered as directory but not a directory" % quotefn(p))
493
elif entry.kind == 'file':
494
if not isfile(p):
495
bailout("%s is entered as file but is not a file" % quotefn(p))
496
497
content = file(p, 'rb').read()
498
499
entry.text_sha1 = sha_string(content)
500
entry.text_size = len(content)
501
502
old_ie = basis_inv.has_id(file_id) and basis_inv[file_id]
503
if (old_ie
504
and (old_ie.text_size == entry.text_size)
505
and (old_ie.text_sha1 == entry.text_sha1)):
506
## assert content == basis.get_file(file_id).read()
507
entry.text_id = basis_inv[file_id].text_id
508
mutter(' unchanged from previous text_id {%s}' %
509
entry.text_id)
510
511
else:
512
entry.text_id = gen_file_id(entry.name)
513
self.text_store.add(content, entry.text_id)
514
mutter(' stored with text_id {%s}' % entry.text_id)
515
if verbose:
516
if not old_ie:
517
state = 'A'
518
elif (old_ie.name == entry.name
519
and old_ie.parent_id == entry.parent_id):
520
state = 'M'
521
else:
522
state = 'R'
523
524
show_status(state, entry.kind, quotefn(path))
525
526
for file_id in missing_ids:
527
# have to do this later so we don't mess up the iterator.
528
# since parents may be removed before their children we
529
# have to test.
530
531
# FIXME: There's probably a better way to do this; perhaps
532
# the workingtree should know how to filter itself.
533
if work_inv.has_id(file_id):
534
del work_inv[file_id]
535
536
537
inv_id = rev_id = _gen_revision_id(time.time())
538
539
inv_tmp = tempfile.TemporaryFile()
540
inv.write_xml(inv_tmp)
541
inv_tmp.seek(0)
542
self.inventory_store.add(inv_tmp, inv_id)
543
mutter('new inventory_id is {%s}' % inv_id)
544
545
self._write_inventory(work_inv)
546
547
if timestamp == None:
548
timestamp = time.time()
549
550
if committer == None:
551
committer = username()
552
553
if timezone == None:
554
timezone = local_time_offset()
555
556
mutter("building commit log message")
557
rev = Revision(timestamp=timestamp,
558
timezone=timezone,
559
committer=committer,
560
precursor = self.last_patch(),
561
message = message,
562
inventory_id=inv_id,
563
revision_id=rev_id)
564
565
rev_tmp = tempfile.TemporaryFile()
566
rev.write_xml(rev_tmp)
567
rev_tmp.seek(0)
568
self.revision_store.add(rev_tmp, rev_id)
569
mutter("new revision_id is {%s}" % rev_id)
570
571
## XXX: Everything up to here can simply be orphaned if we abort
572
## the commit; it will leave junk files behind but that doesn't
573
## matter.
574
575
## TODO: Read back the just-generated changeset, and make sure it
576
## applies and recreates the right state.
577
578
## TODO: Also calculate and store the inventory SHA1
579
mutter("committing patch r%d" % (self.revno() + 1))
580
581
582
self.append_revision(rev_id)
583
584
if verbose:
585
note("commited r%d" % self.revno())
586
587
588
455
def append_revision(self, revision_id):
589
456
mutter("add {%s} to revision-history" % revision_id)
590
457
rev_history = self.revision_history()
606
473
607
474
def get_revision(self, revision_id):
608
475
"""Return the Revision object for a named revision"""
476
self._need_readlock()
609
477
r = Revision.read_xml(self.revision_store[revision_id])
610
478
assert r.revision_id == revision_id
611
479
return r
617
485
TODO: Perhaps for this and similar methods, take a revision
618
486
parameter which can be either an integer revno or a
619
487
string hash."""
488
self._need_readlock()
620
489
i = Inventory.read_xml(self.inventory_store[inventory_id])
621
490
return i
622
491
623
492
624
493
def get_revision_inventory(self, revision_id):
625
494
"""Return inventory of a past revision."""
495
self._need_readlock()
626
496
if revision_id == None:
627
497
return Inventory()
628
498
else:
635
505
>>> ScratchBranch().revision_history()
636
506
[]
637
507
"""
508
self._need_readlock()
638
509
return [l.rstrip('\r\n') for l in self.controlfile('revision-history', 'r').readlines()]
639
510
640
511
512
def enum_history(self, direction):
513
"""Return (revno, revision_id) for history of branch.
514
515
direction
516
'forward' is from earliest to latest
517
'reverse' is from latest to earliest
518
"""
519
rh = self.revision_history()
520
if direction == 'forward':
521
i = 1
522
for rid in rh:
523
yield i, rid
524
i += 1
525
elif direction == 'reverse':
526
i = len(rh)
527
while i > 0:
528
yield i, rh[i-1]
529
i -= 1
530
else:
531
raise BzrError('invalid history direction %r' % direction)
532
533
641
534
def revno(self):
642
535
"""Return current revision number for this branch.
643
536
644
537
That is equivalent to the number of revisions committed to
645
538
this branch.
646
647
>>> b = ScratchBranch()
648
>>> b.revno()
649
0
650
>>> b.commit('no foo')
651
>>> b.revno()
652
1
653
539
"""
654
540
return len(self.revision_history())
655
541
656
542
657
543
def last_patch(self):
658
544
"""Return last patch hash, or None if no history.
659
660
>>> ScratchBranch().last_patch() == None
661
True
662
545
"""
663
546
ph = self.revision_history()
664
547
if ph:
665
548
return ph[-1]
666
549
else:
667
550
return None
551
552
553
def commit(self, *args, **kw):
554
"""Deprecated"""
555
from bzrlib.commit import commit
556
commit(self, *args, **kw)
668
557
669
558
670
559
def lookup_revision(self, revno):
684
573
685
574
`revision_id` may be None for the null revision, in which case
686
575
an `EmptyTree` is returned."""
687
576
self._need_readlock()
688
577
if revision_id == None:
689
578
return EmptyTree()
690
579
else:
694
583
695
584
def working_tree(self):
696
585
"""Return a `Tree` for the working copy."""
586
from workingtree import WorkingTree
697
587
return WorkingTree(self.base, self.read_working_inventory())
698
588
699
589
701
591
"""Return `Tree` object for last revision.
702
592
703
593
If there are no revisions yet, return an `EmptyTree`.
704
705
>>> b = ScratchBranch(files=['foo'])
706
>>> b.basis_tree().has_filename('foo')
707
False
708
>>> b.working_tree().has_filename('foo')
709
True
710
>>> b.add('foo')
711
>>> b.commit('add foo')
712
>>> b.basis_tree().has_filename('foo')
713
True
714
594
"""
715
595
r = self.last_patch()
716
596
if r == None:
720
600
721
601
722
602
723
def write_log(self, show_timezone='original', verbose=False):
724
"""Write out human-readable log of commits to this branch
725
726
utc -- If true, show dates in universal time, not local time."""
727
## TODO: Option to choose either original, utc or local timezone
728
revno = 1
729
precursor = None
730
for p in self.revision_history():
731
print '-' * 40
732
print 'revno:', revno
733
## TODO: Show hash if --id is given.
734
##print 'revision-hash:', p
735
rev = self.get_revision(p)
736
print 'committer:', rev.committer
737
print 'timestamp: %s' % (format_date(rev.timestamp, rev.timezone or 0,
738
show_timezone))
739
740
## opportunistic consistency check, same as check_patch_chaining
741
if rev.precursor != precursor:
742
bailout("mismatched precursor!")
743
744
print 'message:'
745
if not rev.message:
746
print ' (no message)'
747
else:
748
for l in rev.message.split('\n'):
749
print ' ' + l
750
751
if verbose == True and precursor != None:
752
print 'changed files:'
753
tree = self.revision_tree(p)
754
prevtree = self.revision_tree(precursor)
755
756
for file_state, fid, old_name, new_name, kind in \
757
diff_trees(prevtree, tree, ):
758
if file_state == 'A' or file_state == 'M':
759
show_status(file_state, kind, new_name)
760
elif file_state == 'D':
761
show_status(file_state, kind, old_name)
762
elif file_state == 'R':
763
show_status(file_state, kind,
764
old_name + ' => ' + new_name)
765
766
revno += 1
767
precursor = p
768
769
770
603
def rename_one(self, from_rel, to_rel):
771
604
"""Rename one file.
772
605
773
606
This can change the directory or the filename or both.
774
"""
607
"""
608
self._need_writelock()
775
609
tree = self.working_tree()
776
610
inv = tree.inventory
777
611
if not tree.has_filename(from_rel):
826
660
Note that to_name is only the last component of the new name;
827
661
this doesn't change the directory.
828
662
"""
663
self._need_writelock()
829
664
## TODO: Option to move IDs only
830
665
assert not isinstance(from_paths, basestring)
831
666
tree = self.working_tree()
876
711
877
712
878
713
879
def show_status(self, show_all=False):
880
"""Display single-line status for non-ignored working files.
881
882
The list is show sorted in order by file name.
883
884
>>> b = ScratchBranch(files=['foo', 'foo~'])
885
>>> b.show_status()
886
? foo
887
>>> b.add('foo')
888
>>> b.show_status()
889
A foo
890
>>> b.commit("add foo")
891
>>> b.show_status()
892
>>> os.unlink(b.abspath('foo'))
893
>>> b.show_status()
894
D foo
895
896
TODO: Get state for single files.
897
"""
898
899
# We have to build everything into a list first so that it can
900
# sorted by name, incorporating all the different sources.
901
902
# FIXME: Rather than getting things in random order and then sorting,
903
# just step through in order.
904
905
# Interesting case: the old ID for a file has been removed,
906
# but a new file has been created under that name.
907
908
old = self.basis_tree()
909
new = self.working_tree()
910
911
for fs, fid, oldname, newname, kind in diff_trees(old, new):
912
if fs == 'R':
913
show_status(fs, kind,
914
oldname + ' => ' + newname)
915
elif fs == 'A' or fs == 'M':
916
show_status(fs, kind, newname)
917
elif fs == 'D':
918
show_status(fs, kind, oldname)
919
elif fs == '.':
920
if show_all:
921
show_status(fs, kind, newname)
922
elif fs == 'I':
923
if show_all:
924
show_status(fs, kind, newname)
925
elif fs == '?':
926
show_status(fs, kind, newname)
927
else:
928
bailout("weird file state %r" % ((fs, fid),))
929
930
931
714
932
715
class ScratchBranch(Branch):
933
716
"""Special test class: a branch that cleans up after itself.
936
719
>>> isdir(b.base)
937
720
True
938
721
>>> bd = b.base
939
>>> del b
722
>>> b.destroy()
940
723
>>> isdir(bd)
941
724
False
942
725
"""
956
739
957
740
958
741
def __del__(self):
742
self.destroy()
743
744
def destroy(self):
959
745
"""Destroy the test branch, removing the scratch directory."""
960
746
try:
747
mutter("delete ScratchBranch %s" % self.base)
961
748
shutil.rmtree(self.base)
962
except OSError:
749
except OSError, e:
963
750
# Work around for shutil.rmtree failing on Windows when
964
751
# readonly files are encountered
752
mutter("hit exception in destroying ScratchBranch: %s" % e)
965
753
for root, dirs, files in os.walk(self.base, topdown=False):
966
754
for name in files:
967
755
os.chmod(os.path.join(root, name), 0700)
968
756
shutil.rmtree(self.base)
757
self.base = None
969
758
970
759
971
760
988
777
989
778
990
779
991
def _gen_revision_id(when):
992
"""Return new revision-id."""
993
s = '%s-%s-' % (user_email(), compact_date(when))
994
s += hexlify(rand_bytes(8))
995
return s
996
997
998
780
def gen_file_id(name):
999
781
"""Return new file id.
1000
782
Loggerhead is a web-based interface for Breezy
Version: 2.0.1