« back to all changes in this revision
Viewing changes to bzrlib/commit.py
- browse files at revision 1457.1.5
- compare with another revision
- download diff
- download tarball
- view history from revision 1457.1.5
- Committer: Robert Collins
- Date: 2005-10-21 04:02:58 UTC
- mto: This revision was merged to the branch mainline in revision 1503.
- Revision ID: robertc@robertcollins.net-20051021040258-84a5db1be67b30b9
make build-api have a hasbang
- files added:
- NEWS.developers
- bzrlib/store
- bzrlib/transport
- files removed:
- bzrlib/meta_store.py
- files renamed:
- tools/testweave.py => bzrlib/selftest/test_weave.py
- bzrlib/store.py => bzrlib/store/__init__.py
- bzrlib/weavestore.py => bzrlib/store/weave.py
- tools/history2weaves.py => bzrlib/upgrade.py
- files modified:
- .bzrignore
- bzr *
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?
20
21
## XXX: If we merged two versions of a file then we still need to
22
## create a new version representing that merge, even if it didn't
23
## change from the parent.
24
25
## TODO: Read back the just-generated changeset, and make sure it
26
## applies and recreates the right state.
27
28
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
# TODO: Rather than mashing together the ancestry and storing it back,
48
# perhaps the weave should have single method which does it all in one
49
# go, avoiding a lot of redundant work.
50
51
# TODO: Perhaps give a warning if one of the revisions marked as
52
# merged is already in the ancestry, and then don't record it as a
53
# distinct parent.
54
55
# TODO: If the file is newly merged but unchanged from the version it
56
# merges from, then it should still be reported as newly added
57
# relative to the basis revision.
29
58
30
59
31
60
import os
61
import re
32
62
import sys
33
63
import time
34
import tempfile
35
import sha
64
import pdb
36
65
37
66
from binascii import hexlify
38
67
from cStringIO import StringIO
39
68
40
from bzrlib.osutils import (local_time_offset, username,
41
rand_bytes, compact_date, user_email,
69
from bzrlib.osutils import (local_time_offset,
70
rand_bytes, compact_date,
42
71
kind_marker, is_inside_any, quotefn,
43
sha_string, sha_file, isdir, isfile)
72
sha_string, sha_strings, sha_file, isdir, isfile,
73
split_lines)
44
74
from bzrlib.branch import gen_file_id
45
from bzrlib.errors import BzrError, PointlessCommit
46
from bzrlib.revision import Revision, RevisionReference
47
from bzrlib.trace import mutter, note
75
import bzrlib.config
76
from bzrlib.errors import (BzrError, PointlessCommit,
77
HistoryMissing,
78
ConflictsInTree
79
)
80
from bzrlib.revision import Revision
81
from bzrlib.trace import mutter, note, warning
48
82
from bzrlib.xml5 import serializer_v5
49
from bzrlib.inventory import Inventory
50
from bzrlib.delta import compare_trees
83
from bzrlib.inventory import Inventory, ROOT_ID
51
84
from bzrlib.weave import Weave
52
85
from bzrlib.weavefile import read_weave, write_weave_v5
53
86
from bzrlib.atomicfile import AtomicFile
60
93
61
94
New code should use the Commit class instead.
62
95
"""
96
## XXX: Remove this in favor of Branch.commit?
63
97
Commit().commit(*args, **kwargs)
64
98
65
99
66
100
class NullCommitReporter(object):
67
101
"""I report on progress of a commit."""
68
def added(self, path):
69
pass
70
71
def removed(self, path):
72
pass
73
74
def renamed(self, old_path, new_path):
75
pass
76
102
103
def snapshot_change(self, change, path):
104
pass
105
106
def completed(self, revno, rev_id):
107
pass
108
109
def deleted(self, file_id):
110
pass
111
112
def escaped(self, escape_count, message):
113
pass
114
115
def missing(self, path):
116
pass
77
117
78
118
class ReportCommitToLog(NullCommitReporter):
79
def added(self, path):
80
note('added %s', path)
81
82
def removed(self, path):
83
note('removed %s', path)
84
85
def renamed(self, old_path, new_path):
86
note('renamed %s => %s', old_path, new_path)
87
119
120
def snapshot_change(self, change, path):
121
note("%s %s", change, path)
122
123
def completed(self, revno, rev_id):
124
note('committed r%d {%s}', revno, rev_id)
125
126
def deleted(self, file_id):
127
note('deleted %s', file_id)
128
129
def escaped(self, escape_count, message):
130
note("replaced %d control characters in message", escape_count)
131
132
def missing(self, path):
133
note('missing %s', path)
88
134
89
135
class Commit(object):
90
136
"""Task of committing a new revision.
113
159
committer=None,
114
160
specific_files=None,
115
161
rev_id=None,
116
allow_pointless=True):
162
allow_pointless=True,
163
verbose=False,
164
revprops=None):
117
165
"""Commit working copy as a new revision.
118
166
119
The basic approach is to add all the file texts into the
120
store, then the inventory, then make a new revision pointing
121
to that inventory and store that.
122
123
This is not quite safe if the working copy changes during the
124
commit; for the moment that is simply not allowed. A better
125
approach is to make a temporary copy of the files before
126
computing their hashes, and then add those hashes in turn to
127
the inventory. This should mean at least that there are no
128
broken hash pointers. There is no way we can get a snapshot
129
of the whole directory at an instant. This would also have to
130
be robust against files disappearing, moving, etc. So the
131
whole thing is a bit hard.
132
133
This raises PointlessCommit if there are no changes, no new merges,
134
and allow_pointless is false.
135
136
167
timestamp -- if not None, seconds-since-epoch for a
137
168
postdated/predated commit.
138
169
139
specific_files
140
If true, commit only those files.
170
specific_files -- If true, commit only those files.
141
171
142
rev_id
143
If set, use this as the new revision id.
172
rev_id -- If set, use this as the new revision id.
144
173
Useful for test or import commands that need to tightly
145
174
control what revisions are assigned. If you duplicate
146
175
a revision id that exists elsewhere it is your own fault.
147
176
If null (default), a time/random revision id is generated.
177
178
allow_pointless -- If true (default), commit even if nothing
179
has changed and no merges are recorded.
180
181
revprops -- Properties for new revision
148
182
"""
183
mutter('preparing to commit')
149
184
150
185
self.branch = branch
151
self.branch.lock_write()
186
self.weave_store = branch.weave_store
152
187
self.rev_id = rev_id
153
188
self.specific_files = specific_files
154
189
self.allow_pointless = allow_pointless
190
self.revprops = revprops
155
191
156
192
if timestamp is None:
157
193
self.timestamp = time.time()
158
194
else:
159
195
self.timestamp = long(timestamp)
160
196
197
config = bzrlib.config.BranchConfig(self.branch)
198
if rev_id is None:
199
self.rev_id = _gen_revision_id(config, self.timestamp)
200
else:
201
self.rev_id = rev_id
202
161
203
if committer is None:
162
self.committer = username(self.branch)
204
self.committer = config.username()
163
205
else:
164
206
assert isinstance(committer, basestring), type(committer)
165
207
self.committer = committer
171
213
172
214
assert isinstance(message, basestring), type(message)
173
215
self.message = message
216
self._escape_commit_message()
174
217
218
self.branch.lock_write()
175
219
try:
176
# First walk over the working inventory; and both update that
177
# and also build a new revision inventory. The revision
178
# inventory needs to hold the text-id, sha1 and size of the
179
# actual file versions committed in the revision. (These are
180
# not present in the working inventory.) We also need to
181
# detect missing/deleted files, and remove them from the
182
# working inventory.
183
184
220
self.work_tree = self.branch.working_tree()
185
221
self.work_inv = self.work_tree.inventory
186
222
self.basis_tree = self.branch.basis_tree()
187
223
self.basis_inv = self.basis_tree.inventory
188
224
189
self.pending_merges = self.branch.pending_merges()
190
if self.pending_merges:
191
raise NotImplementedError("sorry, can't commit merges to the weave format yet")
225
self._gather_parents()
226
if len(self.parents) > 1 and self.specific_files:
227
raise NotImplementedError('selected-file commit of merges is not supported yet')
228
self._check_parents_present()
192
229
193
if self.rev_id is None:
194
self.rev_id = _gen_revision_id(self.branch, time.time())
195
196
# todo: update hashcache
197
self.delta = compare_trees(self.basis_tree, self.work_tree,
198
specific_files=self.specific_files)
199
200
if not (self.delta.has_changed()
201
or self.allow_pointless
202
or self.pending_merges):
230
self._remove_deleted()
231
self._populate_new_inv()
232
self._store_snapshot()
233
self._report_deletes()
234
235
if not (self.allow_pointless
236
or len(self.parents) > 1
237
or self.new_inv != self.basis_inv):
203
238
raise PointlessCommit()
204
239
205
self.new_inv = self.basis_inv.copy()
206
207
self.delta.show(sys.stdout)
208
209
self._remove_deleted()
210
self._store_files()
211
212
self.branch._write_inventory(self.work_inv)
240
if len(list(self.work_tree.iter_conflicts()))>0:
241
raise ConflictsInTree
242
213
243
self._record_inventory()
214
215
244
self._make_revision()
216
note('committted r%d {%s}', (self.branch.revno() + 1),
217
self.rev_id)
245
self.reporter.completed(self.branch.revno()+1, self.rev_id)
218
246
self.branch.append_revision(self.rev_id)
219
247
self.branch.set_pending_merges([])
220
248
finally:
221
249
self.branch.unlock()
222
250
223
224
251
def _record_inventory(self):
225
inv_tmp = StringIO()
226
serializer_v5.write_inventory(self.new_inv, inv_tmp)
227
self.inv_sha1 = sha_string(inv_tmp.getvalue())
228
inv_tmp.seek(0)
229
self.branch.inventory_store.add(inv_tmp, self.rev_id)
230
231
252
"""Store the inventory for the new revision."""
253
inv_text = serializer_v5.write_inventory_to_string(self.new_inv)
254
self.inv_sha1 = sha_string(inv_text)
255
s = self.branch.control_weaves
256
s.add_text('inventory', self.rev_id,
257
split_lines(inv_text), self.present_parents,
258
self.branch.get_transaction())
259
260
def _escape_commit_message(self):
261
"""Replace xml-incompatible control characters."""
262
# Python strings can include characters that can't be
263
# represented in well-formed XML; escape characters that
264
# aren't listed in the XML specification
265
# (http://www.w3.org/TR/REC-xml/#NT-Char).
266
if isinstance(self.message, unicode):
267
char_pattern = u'[^\x09\x0A\x0D\u0020-\uD7FF\uE000-\uFFFD]'
268
else:
269
# Use a regular 'str' as pattern to avoid having re.subn
270
# return 'unicode' results.
271
char_pattern = '[^x09\x0A\x0D\x20-\xFF]'
272
self.message, escape_count = re.subn(
273
char_pattern,
274
lambda match: match.group(0).encode('unicode_escape'),
275
self.message)
276
if escape_count:
277
self.reporter.escaped(escape_count, self.message)
278
279
def _gather_parents(self):
280
"""Record the parents of a merge for merge detection."""
281
pending_merges = self.branch.pending_merges()
282
self.parents = []
283
self.parent_invs = []
284
self.present_parents = []
285
precursor_id = self.branch.last_revision()
286
if precursor_id:
287
self.parents.append(precursor_id)
288
self.parents += pending_merges
289
for revision in self.parents:
290
if self.branch.has_revision(revision):
291
self.parent_invs.append(self.branch.get_inventory(revision))
292
self.present_parents.append(revision)
293
294
def _check_parents_present(self):
295
for parent_id in self.parents:
296
mutter('commit parent revision {%s}', parent_id)
297
if not self.branch.has_revision(parent_id):
298
if parent_id == self.branch.last_revision():
299
warning("parent is missing %r", parent_id)
300
raise HistoryMissing(self.branch, 'revision', parent_id)
301
else:
302
mutter("commit will ghost revision %r", parent_id)
303
232
304
def _make_revision(self):
233
305
"""Record a new revision object for this commit."""
234
306
self.rev = Revision(timestamp=self.timestamp,
236
308
committer=self.committer,
237
309
message=self.message,
238
310
inventory_sha1=self.inv_sha1,
239
revision_id=self.rev_id)
240
241
self.rev.parents = []
242
precursor_id = self.branch.last_patch()
243
if precursor_id:
244
self.rev.parents.append(RevisionReference(precursor_id))
245
for merge_rev in self.pending_merges:
246
rev.parents.append(RevisionReference(merge_rev))
247
248
rev_tmp = tempfile.TemporaryFile()
311
revision_id=self.rev_id,
312
properties=self.revprops)
313
self.rev.parent_ids = self.parents
314
rev_tmp = StringIO()
249
315
serializer_v5.write_revision(self.rev, rev_tmp)
250
316
rev_tmp.seek(0)
251
317
self.branch.revision_store.add(rev_tmp, self.rev_id)
252
318
mutter('new revision_id is {%s}', self.rev_id)
253
319
254
255
320
def _remove_deleted(self):
256
"""Remove deleted files from the working and stored inventories."""
257
for path, id, kind in self.delta.removed:
258
if self.work_inv.has_id(id):
259
del self.work_inv[id]
260
if self.new_inv.has_id(id):
261
del self.new_inv[id]
262
263
264
265
def _store_files(self):
266
"""Store new texts of modified/added files."""
267
# We must make sure that directories are added before anything
268
# inside them is added. the files within the delta report are
269
# sorted by path so we know the directory will come before its
270
# contents.
271
for path, file_id, kind in self.delta.added:
272
if kind != 'file':
273
ie = self.work_inv[file_id].copy()
274
self.new_inv.add(ie)
321
"""Remove deleted files from the working inventories.
322
323
This is done prior to taking the working inventory as the
324
basis for the new committed inventory.
325
326
This returns true if any files
327
*that existed in the basis inventory* were deleted.
328
Files that were added and deleted
329
in the working copy don't matter.
330
"""
331
specific = self.specific_files
332
deleted_ids = []
333
for path, ie in self.work_inv.iter_entries():
334
if specific and not is_inside_any(specific, path):
335
continue
336
if not self.work_tree.has_filename(path):
337
self.reporter.missing(path)
338
deleted_ids.append((path, ie.file_id))
339
if deleted_ids:
340
deleted_ids.sort(reverse=True)
341
for path, file_id in deleted_ids:
342
del self.work_inv[file_id]
343
self.branch._write_inventory(self.work_inv)
344
345
def _store_snapshot(self):
346
"""Pass over inventory and record a snapshot.
347
348
Entries get a new revision when they are modified in
349
any way, which includes a merge with a new set of
350
parents that have the same entry.
351
"""
352
# XXX: Need to think more here about when the user has
353
# made a specific decision on a particular value -- c.f.
354
# mark-merge.
355
for path, ie in self.new_inv.iter_entries():
356
previous_entries = ie.find_previous_heads(
357
self.parent_invs,
358
self.weave_store.get_weave_or_empty(ie.file_id,
359
self.branch.get_transaction()))
360
if ie.revision is None:
361
change = ie.snapshot(self.rev_id, path, previous_entries,
362
self.work_tree, self.weave_store,
363
self.branch.get_transaction())
275
364
else:
276
self._store_file_text(file_id)
277
278
for path, file_id, kind in self.delta.modified:
279
if kind != 'file':
280
continue
281
self._store_file_text(file_id)
282
283
for old_path, new_path, file_id, kind, text_modified in self.delta.renamed:
284
if kind != 'file':
285
continue
286
if not text_modified:
287
continue
288
self._store_file_text(file_id)
289
290
291
def _store_file_text(self, file_id):
292
"""Store updated text for one modified or added file."""
293
note('store new text for {%s} in revision {%s}',
294
file_id, self.rev_id)
295
new_lines = self.work_tree.get_file(file_id).readlines()
296
if file_id in self.new_inv: # was in basis inventory
297
ie = self.new_inv[file_id]
298
assert ie.file_id == file_id
299
assert file_id in self.basis_inv
300
assert self.basis_inv[file_id].kind == 'file'
301
old_version = self.basis_inv[file_id].text_version
302
file_parents = [old_version]
303
else: # new in this revision
304
ie = self.work_inv[file_id].copy()
305
self.new_inv.add(ie)
306
assert file_id not in self.basis_inv
307
file_parents = []
308
assert ie.kind == 'file'
309
self._add_text_to_weave(file_id, new_lines, file_parents)
310
# make a new inventory entry for this file, using whatever
311
# it had in the working copy, plus details on the new text
312
ie.text_sha1 = _sha_strings(new_lines)
313
ie.text_size = sum(map(len, new_lines))
314
ie.text_version = self.rev_id
315
ie.entry_version = self.rev_id
316
317
318
def _add_text_to_weave(self, file_id, new_lines, parents):
319
weave_fn = self.branch.controlfilename(['weaves', file_id+'.weave'])
320
if os.path.exists(weave_fn):
321
w = read_weave(file(weave_fn, 'rb'))
322
else:
323
w = Weave()
324
# XXX: Should set the appropriate parents by looking for this file_id
325
# in all revision parents
326
parent_idxs = map(w.lookup, parents)
327
w.add(self.rev_id, parent_idxs, new_lines)
328
af = AtomicFile(weave_fn)
329
try:
330
write_weave_v5(w, af)
331
af.commit()
332
finally:
333
af.close()
334
335
336
def _gen_revision_id(branch, when):
365
change = "unchanged"
366
self.reporter.snapshot_change(change, path)
367
368
def _populate_new_inv(self):
369
"""Build revision inventory.
370
371
This creates a new empty inventory. Depending on
372
which files are selected for commit, and what is present in the
373
current tree, the new inventory is populated. inventory entries
374
which are candidates for modification have their revision set to
375
None; inventory entries that are carried over untouched have their
376
revision set to their prior value.
377
"""
378
mutter("Selecting files for commit with filter %s", self.specific_files)
379
self.new_inv = Inventory()
380
for path, new_ie in self.work_inv.iter_entries():
381
file_id = new_ie.file_id
382
mutter('check %s {%s}', path, new_ie.file_id)
383
if self.specific_files:
384
if not is_inside_any(self.specific_files, path):
385
mutter('%s not selected for commit', path)
386
self._carry_entry(file_id)
387
continue
388
else:
389
# this is selected, ensure its parents are too.
390
parent_id = new_ie.parent_id
391
while parent_id != ROOT_ID:
392
if not self.new_inv.has_id(parent_id):
393
ie = self._select_entry(self.work_inv[parent_id])
394
mutter('%s selected for commit because of %s',
395
self.new_inv.id2path(parent_id), path)
396
397
ie = self.new_inv[parent_id]
398
if ie.revision is not None:
399
ie.revision = None
400
mutter('%s selected for commit because of %s',
401
self.new_inv.id2path(parent_id), path)
402
parent_id = ie.parent_id
403
mutter('%s selected for commit', path)
404
self._select_entry(new_ie)
405
406
def _select_entry(self, new_ie):
407
"""Make new_ie be considered for committing."""
408
ie = new_ie.copy()
409
ie.revision = None
410
self.new_inv.add(ie)
411
return ie
412
413
def _carry_entry(self, file_id):
414
"""Carry the file unchanged from the basis revision."""
415
if self.basis_inv.has_id(file_id):
416
self.new_inv.add(self.basis_inv[file_id].copy())
417
418
def _report_deletes(self):
419
for file_id in self.basis_inv:
420
if file_id not in self.new_inv:
421
self.reporter.deleted(self.basis_inv.id2path(file_id))
422
423
def _gen_revision_id(config, when):
337
424
"""Return new revision-id."""
338
s = '%s-%s-' % (user_email(branch), compact_date(when))
425
s = '%s-%s-' % (config.user_email(), compact_date(when))
339
426
s += hexlify(rand_bytes(8))
340
427
return s
341
342
343
def _sha_strings(strings):
344
"""Return the sha-1 of concatenation of strings"""
345
s = sha.new()
346
map(s.update, strings)
347
return s.hexdigest()
Loggerhead is a web-based interface for Breezy
Version: 2.0.1