« back to all changes in this revision
Viewing changes to bzrlib/commit.py
- browse files at revision 1417.1.4
- compare with another revision
- download diff
- download tarball
- view history from revision 1417.1.4
- Committer: Robert Collins
- Date: 2005-10-06 22:15:52 UTC
- mfrom: (1185.13.2)
- mto: This revision was merged to the branch mainline in revision 1420.
- Revision ID: robertc@robertcollins.net-20051006221552-9b15c96fa504e0ad
merge from upstream
- 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
- tools/history2weaves.py => bzrlib/upgrade.py
- bzrlib/newinventory.py => contrib/newinventory.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
19
# FIXME: "bzr commit doc/format" commits doc/format.txt!
20
21
def commit(branch, message,
22
timestamp=None,
23
timezone=None,
24
committer=None,
25
verbose=True,
26
specific_files=None,
27
rev_id=None,
28
allow_pointless=True):
29
"""Commit working copy as a new revision.
30
31
The basic approach is to add all the file texts into the
32
store, then the inventory, then make a new revision pointing
33
to that inventory and store that.
34
35
This is not quite safe if the working copy changes during the
36
commit; for the moment that is simply not allowed. A better
37
approach is to make a temporary copy of the files before
38
computing their hashes, and then add those hashes in turn to
39
the inventory. This should mean at least that there are no
40
broken hash pointers. There is no way we can get a snapshot
41
of the whole directory at an instant. This would also have to
42
be robust against files disappearing, moving, etc. So the
43
whole thing is a bit hard.
44
45
This raises PointlessCommit if there are no changes, no new merges,
46
and allow_pointless is false.
47
48
timestamp -- if not None, seconds-since-epoch for a
49
postdated/predated commit.
50
51
specific_files
52
If true, commit only those files.
53
54
rev_id
55
If set, use this as the new revision id.
56
Useful for test or import commands that need to tightly
57
control what revisions are assigned. If you duplicate
58
a revision id that exists elsewhere it is your own fault.
59
If null (default), a time/random revision id is generated.
60
"""
61
62
import time, tempfile
63
64
from bzrlib.osutils import local_time_offset, username
65
from bzrlib.branch import gen_file_id
66
from bzrlib.errors import BzrError, PointlessCommit
67
from bzrlib.revision import Revision, RevisionReference
68
from bzrlib.trace import mutter, note
69
from bzrlib.xml import pack_xml
70
71
branch.lock_write()
72
73
try:
74
# First walk over the working inventory; and both update that
75
# and also build a new revision inventory. The revision
76
# inventory needs to hold the text-id, sha1 and size of the
77
# actual file versions committed in the revision. (These are
78
# not present in the working inventory.) We also need to
79
# detect missing/deleted files, and remove them from the
80
# working inventory.
81
82
work_tree = branch.working_tree()
83
work_inv = work_tree.inventory
84
basis = branch.basis_tree()
85
basis_inv = basis.inventory
86
87
if verbose:
88
# note('looking for changes...')
89
# print 'looking for changes...'
90
# disabled; should be done at a higher level
91
pass
92
93
pending_merges = branch.pending_merges()
94
95
missing_ids, new_inv, any_changes = \
96
_gather_commit(branch,
97
work_tree,
98
work_inv,
99
basis_inv,
100
specific_files,
101
verbose)
102
103
if not (any_changes or allow_pointless or pending_merges):
104
raise PointlessCommit()
105
106
for file_id in missing_ids:
107
# Any files that have been deleted are now removed from the
108
# working inventory. Files that were not selected for commit
109
# are left as they were in the working inventory and ommitted
110
# from the revision inventory.
111
112
# have to do this later so we don't mess up the iterator.
113
# since parents may be removed before their children we
114
# have to test.
115
116
# FIXME: There's probably a better way to do this; perhaps
117
# the workingtree should know how to filter itbranch.
118
if work_inv.has_id(file_id):
119
del work_inv[file_id]
120
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.
58
59
60
import os
61
import re
62
import sys
63
import time
64
import pdb
65
66
from binascii import hexlify
67
from cStringIO import StringIO
68
69
from bzrlib.osutils import (local_time_offset, username,
70
rand_bytes, compact_date, user_email,
71
kind_marker, is_inside_any, quotefn,
72
sha_string, sha_strings, sha_file, isdir, isfile,
73
split_lines)
74
from bzrlib.branch import gen_file_id
75
from bzrlib.errors import (BzrError, PointlessCommit,
76
HistoryMissing,
77
ConflictsInTree
78
)
79
from bzrlib.revision import Revision
80
from bzrlib.trace import mutter, note, warning
81
from bzrlib.xml5 import serializer_v5
82
from bzrlib.inventory import Inventory, ROOT_ID
83
from bzrlib.weave import Weave
84
from bzrlib.weavefile import read_weave, write_weave_v5
85
from bzrlib.atomicfile import AtomicFile
86
87
88
def commit(*args, **kwargs):
89
"""Commit a new revision to a branch.
90
91
Function-style interface for convenience of old callers.
92
93
New code should use the Commit class instead.
94
"""
95
## XXX: Remove this in favor of Branch.commit?
96
Commit().commit(*args, **kwargs)
97
98
99
class NullCommitReporter(object):
100
"""I report on progress of a commit."""
101
102
def snapshot_change(self, change, path):
103
pass
104
105
def completed(self, revno, rev_id):
106
pass
107
108
def deleted(self, file_id):
109
pass
110
111
def escaped(self, escape_count, message):
112
pass
113
114
def missing(self, path):
115
pass
116
117
class ReportCommitToLog(NullCommitReporter):
118
119
def snapshot_change(self, change, path):
120
note("%s %s", change, path)
121
122
def completed(self, revno, rev_id):
123
note('committed r%d {%s}', revno, rev_id)
124
125
def deleted(self, file_id):
126
note('deleted %s', file_id)
127
128
def escaped(self, escape_count, message):
129
note("replaced %d control characters in message", escape_count)
130
131
def missing(self, path):
132
note('missing %s', path)
133
134
class Commit(object):
135
"""Task of committing a new revision.
136
137
This is a MethodObject: it accumulates state as the commit is
138
prepared, and then it is discarded. It doesn't represent
139
historical revisions, just the act of recording a new one.
140
141
missing_ids
142
Modified to hold a list of files that have been deleted from
143
the working directory; these should be removed from the
144
working inventory.
145
"""
146
def __init__(self,
147
reporter=None):
148
if reporter is not None:
149
self.reporter = reporter
150
else:
151
self.reporter = NullCommitReporter()
152
153
154
def commit(self,
155
branch, message,
156
timestamp=None,
157
timezone=None,
158
committer=None,
159
specific_files=None,
160
rev_id=None,
161
allow_pointless=True,
162
verbose=False):
163
"""Commit working copy as a new revision.
164
165
timestamp -- if not None, seconds-since-epoch for a
166
postdated/predated commit.
167
168
specific_files -- If true, commit only those files.
169
170
rev_id -- If set, use this as the new revision id.
171
Useful for test or import commands that need to tightly
172
control what revisions are assigned. If you duplicate
173
a revision id that exists elsewhere it is your own fault.
174
If null (default), a time/random revision id is generated.
175
176
allow_pointless -- If true (default), commit even if nothing
177
has changed and no merges are recorded.
178
"""
179
mutter('preparing to commit')
180
181
self.branch = branch
182
self.weave_store = branch.weave_store
183
self.rev_id = rev_id
184
self.specific_files = specific_files
185
self.allow_pointless = allow_pointless
186
187
if timestamp is None:
188
self.timestamp = time.time()
189
else:
190
self.timestamp = long(timestamp)
191
121
192
if rev_id is None:
122
rev_id = _gen_revision_id(branch, time.time())
123
inv_id = rev_id
124
125
inv_tmp = tempfile.TemporaryFile()
126
pack_xml(new_inv, inv_tmp)
127
inv_tmp.seek(0)
128
branch.inventory_store.add(inv_tmp, inv_id)
129
mutter('new inventory_id is {%s}' % inv_id)
130
131
# We could also just sha hash the inv_tmp file
132
# however, in the case that branch.inventory_store.add()
133
# ever actually does anything special
134
inv_sha1 = branch.get_inventory_sha1(inv_id)
135
136
branch._write_inventory(work_inv)
137
138
if timestamp == None:
139
timestamp = time.time()
140
141
if committer == None:
142
committer = username(branch)
143
144
if timezone == None:
145
timezone = local_time_offset()
146
147
mutter("building commit log message")
148
rev = Revision(timestamp=timestamp,
149
timezone=timezone,
150
committer=committer,
151
message = message,
152
inventory_id=inv_id,
153
inventory_sha1=inv_sha1,
154
revision_id=rev_id)
155
156
rev.parents = []
157
precursor_id = branch.last_patch()
193
self.rev_id = _gen_revision_id(self.branch, self.timestamp)
194
else:
195
self.rev_id = rev_id
196
197
if committer is None:
198
self.committer = username(self.branch)
199
else:
200
assert isinstance(committer, basestring), type(committer)
201
self.committer = committer
202
203
if timezone is None:
204
self.timezone = local_time_offset()
205
else:
206
self.timezone = int(timezone)
207
208
assert isinstance(message, basestring), type(message)
209
self.message = message
210
self._escape_commit_message()
211
212
self.branch.lock_write()
213
try:
214
self.work_tree = self.branch.working_tree()
215
self.work_inv = self.work_tree.inventory
216
self.basis_tree = self.branch.basis_tree()
217
self.basis_inv = self.basis_tree.inventory
218
219
self._gather_parents()
220
if len(self.parents) > 1 and self.specific_files:
221
raise NotImplementedError('selected-file commit of merges is not supported yet')
222
self._check_parents_present()
223
224
self._remove_deleted()
225
self._populate_new_inv()
226
self._store_snapshot()
227
self._report_deletes()
228
229
if not (self.allow_pointless
230
or len(self.parents) > 1
231
or self.new_inv != self.basis_inv):
232
raise PointlessCommit()
233
234
if len(list(self.work_tree.iter_conflicts()))>0:
235
raise ConflictsInTree
236
237
self._record_inventory()
238
self._make_revision()
239
self.reporter.completed(self.branch.revno()+1, self.rev_id)
240
self.branch.append_revision(self.rev_id)
241
self.branch.set_pending_merges([])
242
finally:
243
self.branch.unlock()
244
245
def _record_inventory(self):
246
"""Store the inventory for the new revision."""
247
inv_text = serializer_v5.write_inventory_to_string(self.new_inv)
248
self.inv_sha1 = sha_string(inv_text)
249
s = self.branch.control_weaves
250
s.add_text('inventory', self.rev_id,
251
split_lines(inv_text), self.present_parents)
252
253
def _escape_commit_message(self):
254
"""Replace xml-incompatible control characters."""
255
# Python strings can include characters that can't be
256
# represented in well-formed XML; escape characters that
257
# aren't listed in the XML specification
258
# (http://www.w3.org/TR/REC-xml/#NT-Char).
259
if isinstance(self.message, unicode):
260
char_pattern = u'[^\x09\x0A\x0D\u0020-\uD7FF\uE000-\uFFFD]'
261
else:
262
# Use a regular 'str' as pattern to avoid having re.subn
263
# return 'unicode' results.
264
char_pattern = '[^x09\x0A\x0D\x20-\xFF]'
265
self.message, escape_count = re.subn(
266
char_pattern,
267
lambda match: match.group(0).encode('unicode_escape'),
268
self.message)
269
if escape_count:
270
self.reporter.escaped(escape_count, self.message)
271
272
def _gather_parents(self):
273
"""Record the parents of a merge for merge detection."""
274
pending_merges = self.branch.pending_merges()
275
self.parents = []
276
self.parent_invs = []
277
self.present_parents = []
278
precursor_id = self.branch.last_revision()
158
279
if precursor_id:
159
precursor_sha1 = branch.get_revision_sha1(precursor_id)
160
rev.parents.append(RevisionReference(precursor_id, precursor_sha1))
161
for merge_rev in pending_merges:
162
rev.parents.append(RevisionReference(merge_rev))
280
self.parents.append(precursor_id)
281
self.parents += pending_merges
282
for revision in self.parents:
283
if self.branch.has_revision(revision):
284
self.parent_invs.append(self.branch.get_inventory(revision))
285
self.present_parents.append(revision)
163
286
164
rev_tmp = tempfile.TemporaryFile()
165
pack_xml(rev, rev_tmp)
287
def _check_parents_present(self):
288
for parent_id in self.parents:
289
mutter('commit parent revision {%s}', parent_id)
290
if not self.branch.has_revision(parent_id):
291
if parent_id == self.branch.last_revision():
292
warning("parent is missing %r", parent_id)
293
raise HistoryMissing(self.branch, 'revision', parent_id)
294
else:
295
mutter("commit will ghost revision %r", parent_id)
296
297
def _make_revision(self):
298
"""Record a new revision object for this commit."""
299
self.rev = Revision(timestamp=self.timestamp,
300
timezone=self.timezone,
301
committer=self.committer,
302
message=self.message,
303
inventory_sha1=self.inv_sha1,
304
revision_id=self.rev_id)
305
self.rev.parent_ids = self.parents
306
rev_tmp = StringIO()
307
serializer_v5.write_revision(self.rev, rev_tmp)
166
308
rev_tmp.seek(0)
167
branch.revision_store.add(rev_tmp, rev_id)
168
mutter("new revision_id is {%s}" % rev_id)
169
170
## XXX: Everything up to here can simply be orphaned if we abort
171
## the commit; it will leave junk files behind but that doesn't
172
## matter.
173
174
## TODO: Read back the just-generated changeset, and make sure it
175
## applies and recreates the right state.
176
177
## TODO: Also calculate and store the inventory SHA1
178
mutter("committing patch r%d" % (branch.revno() + 1))
179
180
branch.append_revision(rev_id)
181
182
branch.set_pending_merges([])
183
184
if verbose:
185
# disabled; should go through logging
186
# note("commited r%d" % branch.revno())
187
# print ("commited r%d" % branch.revno())
188
pass
189
finally:
190
branch.unlock()
191
192
309
self.branch.revision_store.add(rev_tmp, self.rev_id)
310
mutter('new revision_id is {%s}', self.rev_id)
311
312
def _remove_deleted(self):
313
"""Remove deleted files from the working inventories.
314
315
This is done prior to taking the working inventory as the
316
basis for the new committed inventory.
317
318
This returns true if any files
319
*that existed in the basis inventory* were deleted.
320
Files that were added and deleted
321
in the working copy don't matter.
322
"""
323
specific = self.specific_files
324
deleted_ids = []
325
for path, ie in self.work_inv.iter_entries():
326
if specific and not is_inside_any(specific, path):
327
continue
328
if not self.work_tree.has_filename(path):
329
self.reporter.missing(path)
330
deleted_ids.append((path, ie.file_id))
331
if deleted_ids:
332
deleted_ids.sort(reverse=True)
333
for path, file_id in deleted_ids:
334
del self.work_inv[file_id]
335
self.branch._write_inventory(self.work_inv)
336
337
def _store_snapshot(self):
338
"""Pass over inventory and record a snapshot.
339
340
Entries get a new revision when they are modified in
341
any way, which includes a merge with a new set of
342
parents that have the same entry.
343
"""
344
# XXX: Need to think more here about when the user has
345
# made a specific decision on a particular value -- c.f.
346
# mark-merge.
347
for path, ie in self.new_inv.iter_entries():
348
previous_entries = ie.find_previous_heads(
349
self.parent_invs,
350
self.weave_store.get_weave_or_empty(ie.file_id))
351
if ie.revision is None:
352
change = ie.snapshot(self.rev_id, path, previous_entries,
353
self.work_tree, self.weave_store)
354
else:
355
change = "unchanged"
356
self.reporter.snapshot_change(change, path)
357
358
def _populate_new_inv(self):
359
"""Build revision inventory.
360
361
This creates a new empty inventory. Depending on
362
which files are selected for commit, and what is present in the
363
current tree, the new inventory is populated. inventory entries
364
which are candidates for modification have their revision set to
365
None; inventory entries that are carried over untouched have their
366
revision set to their prior value.
367
"""
368
mutter("Selecting files for commit with filter %s", self.specific_files)
369
self.new_inv = Inventory()
370
for path, new_ie in self.work_inv.iter_entries():
371
file_id = new_ie.file_id
372
mutter('check %s {%s}', path, new_ie.file_id)
373
if self.specific_files:
374
if not is_inside_any(self.specific_files, path):
375
mutter('%s not selected for commit', path)
376
self._carry_entry(file_id)
377
continue
378
else:
379
# this is selected, ensure its parents are too.
380
parent_id = new_ie.parent_id
381
while parent_id != ROOT_ID:
382
if not self.new_inv.has_id(parent_id):
383
ie = self._select_entry(self.work_inv[parent_id])
384
mutter('%s selected for commit because of %s',
385
self.new_inv.id2path(parent_id), path)
386
387
ie = self.new_inv[parent_id]
388
if ie.revision is not None:
389
ie.revision = None
390
mutter('%s selected for commit because of %s',
391
self.new_inv.id2path(parent_id), path)
392
parent_id = ie.parent_id
393
mutter('%s selected for commit', path)
394
self._select_entry(new_ie)
395
396
def _select_entry(self, new_ie):
397
"""Make new_ie be considered for committing."""
398
ie = new_ie.copy()
399
ie.revision = None
400
self.new_inv.add(ie)
401
return ie
402
403
def _carry_entry(self, file_id):
404
"""Carry the file unchanged from the basis revision."""
405
if self.basis_inv.has_id(file_id):
406
self.new_inv.add(self.basis_inv[file_id].copy())
407
408
def _report_deletes(self):
409
for file_id in self.basis_inv:
410
if file_id not in self.new_inv:
411
self.reporter.deleted(self.basis_inv.id2path(file_id))
193
412
194
413
def _gen_revision_id(branch, when):
195
414
"""Return new revision-id."""
196
from binascii import hexlify
197
from bzrlib.osutils import rand_bytes, compact_date, user_email
198
199
415
s = '%s-%s-' % (user_email(branch), compact_date(when))
200
416
s += hexlify(rand_bytes(8))
201
417
return s
202
203
204
def _gather_commit(branch, work_tree, work_inv, basis_inv, specific_files,
205
verbose):
206
"""Build inventory preparatory to commit.
207
208
Returns missing_ids, new_inv, any_changes.
209
210
This adds any changed files into the text store, and sets their
211
test-id, sha and size in the returned inventory appropriately.
212
213
missing_ids
214
Modified to hold a list of files that have been deleted from
215
the working directory; these should be removed from the
216
working inventory.
217
"""
218
from bzrlib.inventory import Inventory
219
from bzrlib.osutils import isdir, isfile, sha_string, quotefn, \
220
local_time_offset, username, kind_marker, is_inside_any
221
222
from bzrlib.branch import gen_file_id
223
from bzrlib.errors import BzrError
224
from bzrlib.revision import Revision
225
from bzrlib.trace import mutter, note
226
227
any_changes = False
228
inv = Inventory(work_inv.root.file_id)
229
missing_ids = []
230
231
for path, entry in work_inv.iter_entries():
232
## TODO: Check that the file kind has not changed from the previous
233
## revision of this file (if any).
234
235
p = branch.abspath(path)
236
file_id = entry.file_id
237
mutter('commit prep file %s, id %r ' % (p, file_id))
238
239
if specific_files and not is_inside_any(specific_files, path):
240
mutter(' skipping file excluded from commit')
241
if basis_inv.has_id(file_id):
242
# carry over with previous state
243
inv.add(basis_inv[file_id].copy())
244
else:
245
# omit this from committed inventory
246
pass
247
continue
248
249
if not work_tree.has_id(file_id):
250
if verbose:
251
print('deleted %s%s' % (path, kind_marker(entry.kind)))
252
any_changes = True
253
mutter(" file is missing, removing from inventory")
254
missing_ids.append(file_id)
255
continue
256
257
# this is present in the new inventory; may be new, modified or
258
# unchanged.
259
old_ie = basis_inv.has_id(file_id) and basis_inv[file_id]
260
261
entry = entry.copy()
262
inv.add(entry)
263
264
if old_ie:
265
old_kind = old_ie.kind
266
if old_kind != entry.kind:
267
raise BzrError("entry %r changed kind from %r to %r"
268
% (file_id, old_kind, entry.kind))
269
270
if entry.kind == 'directory':
271
if not isdir(p):
272
raise BzrError("%s is entered as directory but not a directory"
273
% quotefn(p))
274
elif entry.kind == 'file':
275
if not isfile(p):
276
raise BzrError("%s is entered as file but is not a file" % quotefn(p))
277
278
new_sha1 = work_tree.get_file_sha1(file_id)
279
280
if (old_ie
281
and old_ie.text_sha1 == new_sha1):
282
## assert content == basis.get_file(file_id).read()
283
entry.text_id = old_ie.text_id
284
entry.text_sha1 = new_sha1
285
entry.text_size = old_ie.text_size
286
mutter(' unchanged from previous text_id {%s}' %
287
entry.text_id)
288
else:
289
content = file(p, 'rb').read()
290
291
# calculate the sha again, just in case the file contents
292
# changed since we updated the cache
293
entry.text_sha1 = sha_string(content)
294
entry.text_size = len(content)
295
296
entry.text_id = gen_file_id(entry.name)
297
branch.text_store.add(content, entry.text_id)
298
mutter(' stored with text_id {%s}' % entry.text_id)
299
300
if verbose:
301
marked = path + kind_marker(entry.kind)
302
if not old_ie:
303
print 'added', marked
304
any_changes = True
305
elif old_ie == entry:
306
pass # unchanged
307
elif (old_ie.name == entry.name
308
and old_ie.parent_id == entry.parent_id):
309
print 'modified', marked
310
any_changes = True
311
else:
312
print 'renamed', marked
313
any_changes = True
314
315
return missing_ids, inv, any_changes
316
317
Loggerhead is a web-based interface for Breezy
Version: 2.0.1