« back to all changes in this revision
Viewing changes to bzrlib/commit.py
- browse files at revision 1181
- compare with another revision
- download diff
- download tarball
- view history from revision 1181
- Committer: Martin Pool
- Date: 2005-09-05 09:11:03 UTC
- Revision ID: mbp@sourcefrog.net-20050905091103-1e51e146be0f08b4
- add test for deserialization from a canned XML inventory
- files removed:
- bzrlib/selftest/test_ancestry.py
- files renamed:
- contrib/newinventory.py => bzrlib/newinventory.py
- bzrlib/selftest/test_weave.py => tools/testweave.py
- files modified:
- .bzrignore
added
removed
bzrlib/commit.py
1
1
# Copyright (C) 2005 Canonical Ltd
2
#
2
3
3
# This program is free software; you can redistribute it and/or modify
4
4
# it under the terms of the GNU General Public License as published by
5
5
# the Free Software Foundation; either version 2 of the License, or
6
6
# (at your option) any later version.
7
#
7
8
8
# This program is distributed in the hope that it will be useful,
9
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
11
# GNU General Public License for more details.
12
#
12
13
13
# You should have received a copy of the GNU General Public License
14
14
# along with this program; if not, write to the Free Software
15
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
16
16
17
17
18
# XXX: Can we do any better about making interrupted commits change
19
# nothing? Perhaps the best approach is to integrate commit of
20
# AtomicFiles with releasing the lock on the Branch.
21
22
# TODO: Separate 'prepare' phase where we find a list of potentially
23
# committed files. We then can then pause the commit to prompt for a
24
# commit message, knowing the summary will be the same as what's
25
# actually used for the commit. (But perhaps simpler to simply get
26
# the tree status, then use that for a selective commit?)
27
28
# The newly committed revision is going to have a shape corresponding
29
# to that of the working inventory. Files that are not in the
30
# working tree and that were in the predecessor are reported as
31
# removed --- this can include files that were either removed from the
32
# inventory or deleted in the working tree. If they were only
33
# deleted from disk, they are removed from the working inventory.
34
35
# We then consider the remaining entries, which will be in the new
36
# version. Directory entries are simply copied across. File entries
37
# must be checked to see if a new version of the file should be
38
# recorded. For each parent revision inventory, we check to see what
39
# version of the file was present. If the file was present in at
40
# least one tree, and if it was the same version in all the trees,
41
# then we can just refer to that version. Otherwise, a new version
42
# representing the merger of the file versions must be added.
43
44
# TODO: Update hashcache before and after - or does the WorkingTree
45
# look after that?
46
47
# This code requires all merge parents to be present in the branch.
48
# We could relax this but for the sake of simplicity the constraint is
49
# here for now. It's not totally clear to me how we'd know which file
50
# need new text versions if some parents are absent. -- mbp 20050915
51
52
53
import os
54
import sys
55
import time
56
import pdb
57
58
from binascii import hexlify
59
from cStringIO import StringIO
60
61
from bzrlib.osutils import (local_time_offset, username,
62
rand_bytes, compact_date, user_email,
63
kind_marker, is_inside_any, quotefn,
64
sha_string, sha_strings, sha_file, isdir, isfile,
65
split_lines)
66
from bzrlib.branch import gen_file_id, INVENTORY_FILEID, ANCESTRY_FILEID
67
from bzrlib.errors import (BzrError, PointlessCommit,
68
HistoryMissing,
69
)
70
from bzrlib.revision import Revision, RevisionReference
71
from bzrlib.trace import mutter, note, warning
72
from bzrlib.xml5 import serializer_v5
73
from bzrlib.inventory import Inventory
74
from bzrlib.weave import Weave
75
from bzrlib.weavefile import read_weave, write_weave_v5
76
from bzrlib.atomicfile import AtomicFile
77
78
79
def commit(*args, **kwargs):
80
"""Commit a new revision to a branch.
81
82
Function-style interface for convenience of old callers.
83
84
New code should use the Commit class instead.
85
"""
86
## XXX: Remove this in favor of Branch.commit?
87
Commit().commit(*args, **kwargs)
88
89
90
class NullCommitReporter(object):
91
"""I report on progress of a commit."""
92
def added(self, path):
93
pass
94
95
def removed(self, path):
96
pass
97
98
def renamed(self, old_path, new_path):
99
pass
100
101
102
class ReportCommitToLog(NullCommitReporter):
103
def added(self, path):
104
note('added %s', path)
105
106
def removed(self, path):
107
note('removed %s', path)
108
109
def renamed(self, old_path, new_path):
110
note('renamed %s => %s', old_path, new_path)
111
112
113
class Commit(object):
114
"""Task of committing a new revision.
115
116
This is a MethodObject: it accumulates state as the commit is
117
prepared, and then it is discarded. It doesn't represent
118
historical revisions, just the act of recording a new one.
119
120
missing_ids
121
Modified to hold a list of files that have been deleted from
122
the working directory; these should be removed from the
123
working inventory.
124
"""
125
def __init__(self,
126
reporter=None):
127
if reporter is not None:
128
self.reporter = reporter
129
else:
130
self.reporter = NullCommitReporter()
131
132
133
def commit(self,
134
branch, message,
135
timestamp=None,
136
timezone=None,
137
committer=None,
138
specific_files=None,
139
rev_id=None,
140
allow_pointless=True,
141
verbose=False):
142
"""Commit working copy as a new revision.
143
144
timestamp -- if not None, seconds-since-epoch for a
145
postdated/predated commit.
146
147
specific_files -- If true, commit only those files.
148
149
rev_id -- If set, use this as the new revision id.
150
Useful for test or import commands that need to tightly
151
control what revisions are assigned. If you duplicate
152
a revision id that exists elsewhere it is your own fault.
153
If null (default), a time/random revision id is generated.
154
155
allow_pointless -- If true (default), commit even if nothing
156
has changed and no merges are recorded.
157
"""
158
mutter('preparing to commit')
159
160
self.branch = branch
161
self.weave_store = branch.weave_store
162
self.rev_id = rev_id
163
self.specific_files = specific_files
164
self.allow_pointless = allow_pointless
165
166
if timestamp is None:
167
self.timestamp = time.time()
168
else:
169
self.timestamp = long(timestamp)
170
18
19
def commit(branch, message,
20
timestamp=None,
21
timezone=None,
22
committer=None,
23
verbose=True,
24
specific_files=None,
25
rev_id=None,
26
allow_pointless=True):
27
"""Commit working copy as a new revision.
28
29
The basic approach is to add all the file texts into the
30
store, then the inventory, then make a new revision pointing
31
to that inventory and store that.
32
33
This is not quite safe if the working copy changes during the
34
commit; for the moment that is simply not allowed. A better
35
approach is to make a temporary copy of the files before
36
computing their hashes, and then add those hashes in turn to
37
the inventory. This should mean at least that there are no
38
broken hash pointers. There is no way we can get a snapshot
39
of the whole directory at an instant. This would also have to
40
be robust against files disappearing, moving, etc. So the
41
whole thing is a bit hard.
42
43
This raises PointlessCommit if there are no changes, no new merges,
44
and allow_pointless is false.
45
46
timestamp -- if not None, seconds-since-epoch for a
47
postdated/predated commit.
48
49
specific_files
50
If true, commit only those files.
51
52
rev_id
53
If set, use this as the new revision id.
54
Useful for test or import commands that need to tightly
55
control what revisions are assigned. If you duplicate
56
a revision id that exists elsewhere it is your own fault.
57
If null (default), a time/random revision id is generated.
58
"""
59
60
import time, tempfile
61
62
from bzrlib.osutils import local_time_offset, username
63
from bzrlib.branch import gen_file_id
64
from bzrlib.errors import BzrError, PointlessCommit
65
from bzrlib.revision import Revision, RevisionReference
66
from bzrlib.trace import mutter, note
67
from bzrlib.xml import pack_xml, serializer_v4
68
69
branch.lock_write()
70
71
try:
72
# First walk over the working inventory; and both update that
73
# and also build a new revision inventory. The revision
74
# inventory needs to hold the text-id, sha1 and size of the
75
# actual file versions committed in the revision. (These are
76
# not present in the working inventory.) We also need to
77
# detect missing/deleted files, and remove them from the
78
# working inventory.
79
80
work_tree = branch.working_tree()
81
work_inv = work_tree.inventory
82
basis = branch.basis_tree()
83
basis_inv = basis.inventory
84
85
if verbose:
86
# note('looking for changes...')
87
# print 'looking for changes...'
88
# disabled; should be done at a higher level
89
pass
90
91
pending_merges = branch.pending_merges()
92
93
missing_ids, new_inv, any_changes = \
94
_gather_commit(branch,
95
work_tree,
96
work_inv,
97
basis_inv,
98
specific_files,
99
verbose)
100
101
if not (any_changes or allow_pointless or pending_merges):
102
raise PointlessCommit()
103
104
for file_id in missing_ids:
105
# Any files that have been deleted are now removed from the
106
# working inventory. Files that were not selected for commit
107
# are left as they were in the working inventory and ommitted
108
# from the revision inventory.
109
110
# have to do this later so we don't mess up the iterator.
111
# since parents may be removed before their children we
112
# have to test.
113
114
# FIXME: There's probably a better way to do this; perhaps
115
# the workingtree should know how to filter itbranch.
116
if work_inv.has_id(file_id):
117
del work_inv[file_id]
118
171
119
if rev_id is None:
172
self.rev_id = _gen_revision_id(self.branch, self.timestamp)
173
else:
174
self.rev_id = rev_id
175
176
if committer is None:
177
self.committer = username(self.branch)
178
else:
179
assert isinstance(committer, basestring), type(committer)
180
self.committer = committer
181
182
if timezone is None:
183
self.timezone = local_time_offset()
184
else:
185
self.timezone = int(timezone)
186
187
assert isinstance(message, basestring), type(message)
188
self.message = message
189
190
self.branch.lock_write()
191
try:
192
self.work_tree = self.branch.working_tree()
193
self.work_inv = self.work_tree.inventory
194
self.basis_tree = self.branch.basis_tree()
195
self.basis_inv = self.basis_tree.inventory
196
197
self._gather_parents()
198
self._check_parents_present()
199
200
self._remove_deleted()
201
self.new_inv = Inventory()
202
self._store_entries()
203
self._report_deletes()
204
205
if not (self.allow_pointless
206
or len(self.parents) > 1
207
or self.new_inv != self.basis_inv):
208
raise PointlessCommit()
209
210
self._record_inventory()
211
self._record_ancestry()
212
self._make_revision()
213
note('committed r%d {%s}', (self.branch.revno() + 1),
214
self.rev_id)
215
self.branch.append_revision(self.rev_id)
216
self.branch.set_pending_merges([])
217
finally:
218
self.branch.unlock()
219
220
221
222
def _record_inventory(self):
223
"""Store the inventory for the new revision."""
224
inv_text = serializer_v5.write_inventory_to_string(self.new_inv)
225
self.inv_sha1 = sha_string(inv_text)
226
self.weave_store.add_text(INVENTORY_FILEID, self.rev_id,
227
split_lines(inv_text), self.parents)
228
229
230
def _record_ancestry(self):
231
"""Append merged revision ancestry to the ancestry file.
232
233
This should be the merged ancestry of all parents, plus the
234
new revision id."""
235
w = self.weave_store.get_weave_or_empty(ANCESTRY_FILEID)
236
lines = self._merge_ancestry_lines(w)
237
w.add(self.rev_id, self.parents, lines)
238
self.weave_store.put_weave(ANCESTRY_FILEID, w)
239
240
241
def _merge_ancestry_lines(self, ancestry_weave):
242
"""Return merged ancestry lines.
243
244
The lines are revision-ids followed by newlines."""
245
seen = set()
246
ancs = []
247
for parent_id in self.parents:
248
for line in ancestry_weave.get(parent_id):
249
assert line[-1] == '\n'
250
if line not in seen:
251
ancs.append(line)
252
seen.add(line)
253
r = self.rev_id + '\n'
254
assert r not in ancs
255
ancs.append(r)
256
mutter('merged ancestry of {%s}:\n%s', self.rev_id, ''.join(ancs))
257
return ancs
258
259
260
def _gather_parents(self):
261
pending_merges = self.branch.pending_merges()
262
self.parents = []
263
self.parent_trees = []
264
precursor_id = self.branch.last_revision()
120
rev_id = _gen_revision_id(branch, time.time())
121
inv_id = rev_id
122
123
inv_tmp = tempfile.TemporaryFile()
124
125
serializer_v4.write_inventory(new_inv, inv_tmp)
126
inv_tmp.seek(0)
127
branch.inventory_store.add(inv_tmp, inv_id)
128
mutter('new inventory_id is {%s}' % inv_id)
129
130
# We could also just sha hash the inv_tmp file
131
# however, in the case that branch.inventory_store.add()
132
# ever actually does anything special
133
inv_sha1 = branch.get_inventory_sha1(inv_id)
134
135
branch._write_inventory(work_inv)
136
137
if timestamp == None:
138
timestamp = time.time()
139
140
if committer == None:
141
committer = username(branch)
142
143
if timezone == None:
144
timezone = local_time_offset()
145
146
mutter("building commit log message")
147
rev = Revision(timestamp=timestamp,
148
timezone=timezone,
149
committer=committer,
150
message = message,
151
inventory_id=inv_id,
152
inventory_sha1=inv_sha1,
153
revision_id=rev_id)
154
155
rev.parents = []
156
precursor_id = branch.last_patch()
265
157
if precursor_id:
266
self.parents.append(precursor_id)
267
self.parent_trees.append(self.basis_tree)
268
self.parents += pending_merges
269
self.parent_trees.extend(map(self.branch.revision_tree, pending_merges))
270
271
272
def _check_parents_present(self):
273
for parent_id in self.parents:
274
mutter('commit parent revision {%s}', parent_id)
275
if not self.branch.has_revision(parent_id):
276
warning("can't commit a merge from an absent parent")
277
raise HistoryMissing(self.branch, 'revision', parent_id)
278
279
280
def _make_revision(self):
281
"""Record a new revision object for this commit."""
282
self.rev = Revision(timestamp=self.timestamp,
283
timezone=self.timezone,
284
committer=self.committer,
285
message=self.message,
286
inventory_sha1=self.inv_sha1,
287
revision_id=self.rev_id)
288
self.rev.parents = map(RevisionReference, self.parents)
289
rev_tmp = StringIO()
290
serializer_v5.write_revision(self.rev, rev_tmp)
158
precursor_sha1 = branch.get_revision_sha1(precursor_id)
159
rev.parents.append(RevisionReference(precursor_id, precursor_sha1))
160
for merge_rev in pending_merges:
161
rev.parents.append(RevisionReference(merge_rev))
162
163
rev_tmp = tempfile.TemporaryFile()
164
pack_xml(rev, rev_tmp)
291
165
rev_tmp.seek(0)
292
self.branch.revision_store.add(rev_tmp, self.rev_id)
293
mutter('new revision_id is {%s}', self.rev_id)
294
295
296
def _remove_deleted(self):
297
"""Remove deleted files from the working inventories.
298
299
This is done prior to taking the working inventory as the
300
basis for the new committed inventory.
301
302
This returns true if any files
303
*that existed in the basis inventory* were deleted.
304
Files that were added and deleted
305
in the working copy don't matter.
306
"""
307
specific = self.specific_files
308
deleted_ids = []
309
for path, ie in self.work_inv.iter_entries():
310
if specific and not is_inside_any(specific, path):
311
continue
312
if not self.work_tree.has_filename(path):
313
note('missing %s', path)
314
deleted_ids.append(ie.file_id)
315
if deleted_ids:
316
for file_id in deleted_ids:
317
del self.work_inv[file_id]
318
self.branch._write_inventory(self.work_inv)
319
320
321
def _find_file_parents(self, file_id):
322
"""Return the text versions and hashes for all file parents.
323
324
Returned as a map from text version to inventory entry.
325
326
This is a set containing the file versions in all parents
327
revisions containing the file. If the file is new, the set
328
will be empty."""
329
r = {}
330
for tree in self.parent_trees:
331
if file_id in tree.inventory:
332
ie = tree.inventory[file_id]
333
assert ie.kind == 'file'
334
assert ie.file_id == file_id
335
if ie.text_version in r:
336
assert r[ie.text_version] == ie
337
else:
338
r[ie.text_version] = ie
339
return r
340
341
342
def _store_entries(self):
343
"""Build revision inventory and store modified files.
344
345
This is called with new_inv a new empty inventory. Depending on
346
which files are selected for commit, and which ones have
347
been modified or merged, new inventory entries are built
348
based on the working and parent inventories.
349
350
As a side-effect this stores new text versions for committed
351
files with text changes or merges.
352
353
Each entry can have one of several things happen:
354
355
carry_file -- carried from the previous version (if not
356
selected for commit)
357
358
commit_nonfile -- no text to worry about
359
360
commit_old_text -- same text, may have moved
361
362
commit_file -- new text version
363
"""
364
for path, new_ie in self.work_inv.iter_entries():
365
file_id = new_ie.file_id
366
mutter('check %s {%s}', path, new_ie.file_id)
367
if self.specific_files:
368
if not is_inside_any(self.specific_files, path):
369
mutter('%s not selected for commit', path)
370
self._carry_file(file_id)
371
continue
372
if new_ie.kind != 'file':
373
self._commit_nonfile(file_id)
374
continue
375
376
file_parents = self._find_file_parents(file_id)
377
if len(file_parents) == 1:
378
parent_ie = file_parents.values()[0]
379
wc_sha1 = self.work_tree.get_file_sha1(file_id)
380
if parent_ie.text_sha1 == wc_sha1:
381
# text not changed or merged
382
self._commit_old_text(file_id, parent_ie)
383
continue
384
385
mutter('parents of %s are %r', path, file_parents)
386
387
# file is either new, or a file merge; need to record
388
# a new version
389
if len(file_parents) > 1:
390
note('merged %s', path)
391
elif len(file_parents) == 0:
392
note('added %s', path)
393
else:
394
note('modified %s', path)
395
self._commit_file(new_ie, file_id, file_parents)
396
397
398
def _commit_nonfile(self, file_id):
399
self.new_inv.add(self.work_inv[file_id].copy())
400
401
402
def _carry_file(self, file_id):
403
"""Carry the file unchanged from the basis revision."""
404
if self.basis_inv.has_id(file_id):
405
self.new_inv.add(self.basis_inv[file_id].copy())
406
407
408
def _commit_old_text(self, file_id, parent_ie):
409
"""Keep the same text as last time, but possibly a different name."""
410
ie = self.work_inv[file_id].copy()
411
ie.text_version = parent_ie.text_version
412
ie.text_size = parent_ie.text_size
413
ie.text_sha1 = parent_ie.text_sha1
414
self.new_inv.add(ie)
415
416
417
def _report_deletes(self):
418
for file_id in self.basis_inv:
419
if file_id not in self.new_inv:
420
note('deleted %s', self.basis_inv.id2path(file_id))
421
422
423
def _commit_file(self, new_ie, file_id, file_parents):
424
mutter('store new text for {%s} in revision {%s}',
425
file_id, self.rev_id)
426
new_lines = self.work_tree.get_file(file_id).readlines()
427
self._add_text_to_weave(file_id, new_lines, file_parents)
428
new_ie.text_version = self.rev_id
429
new_ie.text_sha1 = sha_strings(new_lines)
430
new_ie.text_size = sum(map(len, new_lines))
431
self.new_inv.add(new_ie)
432
433
434
def _add_text_to_weave(self, file_id, new_lines, parents):
435
if file_id.startswith('__'):
436
raise ValueError('illegal file-id %r for text file' % file_id)
437
self.weave_store.add_text(file_id, self.rev_id, new_lines, parents)
166
branch.revision_store.add(rev_tmp, rev_id)
167
mutter("new revision_id is {%s}" % rev_id)
168
169
## XXX: Everything up to here can simply be orphaned if we abort
170
## the commit; it will leave junk files behind but that doesn't
171
## matter.
172
173
## TODO: Read back the just-generated changeset, and make sure it
174
## applies and recreates the right state.
175
176
## TODO: Also calculate and store the inventory SHA1
177
mutter("committing patch r%d" % (branch.revno() + 1))
178
179
branch.append_revision(rev_id)
180
181
branch.set_pending_merges([])
182
183
if verbose:
184
# disabled; should go through logging
185
# note("commited r%d" % branch.revno())
186
# print ("commited r%d" % branch.revno())
187
pass
188
finally:
189
branch.unlock()
190
438
191
439
192
440
193
def _gen_revision_id(branch, when):
441
194
"""Return new revision-id."""
195
from binascii import hexlify
196
from bzrlib.osutils import rand_bytes, compact_date, user_email
197
442
198
s = '%s-%s-' % (user_email(branch), compact_date(when))
443
199
s += hexlify(rand_bytes(8))
444
200
return s
445
201
446
202
447
448
203
def _gather_commit(branch, work_tree, work_inv, basis_inv, specific_files,
204
verbose):
205
"""Build inventory preparatory to commit.
206
207
Returns missing_ids, new_inv, any_changes.
208
209
This adds any changed files into the text store, and sets their
210
test-id, sha and size in the returned inventory appropriately.
211
212
missing_ids
213
Modified to hold a list of files that have been deleted from
214
the working directory; these should be removed from the
215
working inventory.
216
"""
217
from bzrlib.inventory import Inventory
218
from bzrlib.osutils import isdir, isfile, sha_string, quotefn, \
219
local_time_offset, username, kind_marker, is_inside_any
220
221
from bzrlib.branch import gen_file_id
222
from bzrlib.errors import BzrError
223
from bzrlib.revision import Revision
224
from bzrlib.trace import mutter, note
225
226
any_changes = False
227
inv = Inventory(work_inv.root.file_id)
228
missing_ids = []
229
230
for path, entry in work_inv.iter_entries():
231
## TODO: Check that the file kind has not changed from the previous
232
## revision of this file (if any).
233
234
p = branch.abspath(path)
235
file_id = entry.file_id
236
mutter('commit prep file %s, id %r ' % (p, file_id))
237
238
if specific_files and not is_inside_any(specific_files, path):
239
mutter(' skipping file excluded from commit')
240
if basis_inv.has_id(file_id):
241
# carry over with previous state
242
inv.add(basis_inv[file_id].copy())
243
else:
244
# omit this from committed inventory
245
pass
246
continue
247
248
if not work_tree.has_id(file_id):
249
if verbose:
250
print('deleted %s%s' % (path, kind_marker(entry.kind)))
251
any_changes = True
252
mutter(" file is missing, removing from inventory")
253
missing_ids.append(file_id)
254
continue
255
256
# this is present in the new inventory; may be new, modified or
257
# unchanged.
258
old_ie = basis_inv.has_id(file_id) and basis_inv[file_id]
259
260
entry = entry.copy()
261
inv.add(entry)
262
263
if old_ie:
264
old_kind = old_ie.kind
265
if old_kind != entry.kind:
266
raise BzrError("entry %r changed kind from %r to %r"
267
% (file_id, old_kind, entry.kind))
268
269
if entry.kind == 'directory':
270
if not isdir(p):
271
raise BzrError("%s is entered as directory but not a directory"
272
% quotefn(p))
273
elif entry.kind == 'file':
274
if not isfile(p):
275
raise BzrError("%s is entered as file but is not a file" % quotefn(p))
276
277
new_sha1 = work_tree.get_file_sha1(file_id)
278
279
if (old_ie
280
and old_ie.text_sha1 == new_sha1):
281
## assert content == basis.get_file(file_id).read()
282
entry.text_id = old_ie.text_id
283
entry.text_sha1 = new_sha1
284
entry.text_size = old_ie.text_size
285
mutter(' unchanged from previous text_id {%s}' %
286
entry.text_id)
287
else:
288
content = file(p, 'rb').read()
289
290
# calculate the sha again, just in case the file contents
291
# changed since we updated the cache
292
entry.text_sha1 = sha_string(content)
293
entry.text_size = len(content)
294
295
entry.text_id = gen_file_id(entry.name)
296
branch.text_store.add(content, entry.text_id)
297
mutter(' stored with text_id {%s}' % entry.text_id)
298
299
if verbose:
300
marked = path + kind_marker(entry.kind)
301
if not old_ie:
302
print 'added', marked
303
any_changes = True
304
elif old_ie == entry:
305
pass # unchanged
306
elif (old_ie.name == entry.name
307
and old_ie.parent_id == entry.parent_id):
308
print 'modified', marked
309
any_changes = True
310
else:
311
print 'renamed', marked
312
any_changes = True
313
314
return missing_ids, inv, any_changes
315
316
Loggerhead is a web-based interface for Breezy
Version: 2.0.1