« back to all changes in this revision
Viewing changes to bzrlib/commit.py
- browse files at revision 1185.11.1
- compare with another revision
- download diff
- download tarball
- view history from revision 1185.11.1
- Committer: John Arbash Meinel
- Date: 2005-09-15 21:35:53 UTC
- mfrom: (907.1.57)
- mto: (1393.2.1)
- mto: This revision was merged to the branch mainline in revision 1396.
- Revision ID: john@arbash-meinel.com-20050915213552-a6c83a5ef1e20897
(broken) Transport work is merged in. Tests do not pass yet.
- files added:
- bzrlib/meta_store.py
- files removed:
- NEWS.developers
- files renamed:
- contrib/newinventory.py => bzrlib/newinventory.py
- bzrlib/upgrade.py => tools/history2weaves.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
# 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
)
78
from bzrlib.revision import Revision
79
from bzrlib.trace import mutter, note, warning
80
from bzrlib.xml5 import serializer_v5
81
from bzrlib.inventory import Inventory, ROOT_ID
82
from bzrlib.weave import Weave
83
from bzrlib.weavefile import read_weave, write_weave_v5
84
from bzrlib.atomicfile import AtomicFile
85
86
87
def commit(*args, **kwargs):
88
"""Commit a new revision to a branch.
89
90
Function-style interface for convenience of old callers.
91
92
New code should use the Commit class instead.
93
"""
94
## XXX: Remove this in favor of Branch.commit?
95
Commit().commit(*args, **kwargs)
96
97
98
class NullCommitReporter(object):
99
"""I report on progress of a commit."""
100
def added(self, path):
101
pass
102
103
def removed(self, path):
104
pass
105
106
def renamed(self, old_path, new_path):
107
pass
108
109
110
class ReportCommitToLog(NullCommitReporter):
111
def added(self, path):
112
note('added %s', path)
113
114
def removed(self, path):
115
note('removed %s', path)
116
117
def renamed(self, old_path, new_path):
118
note('renamed %s => %s', old_path, new_path)
119
120
121
class Commit(object):
122
"""Task of committing a new revision.
123
124
This is a MethodObject: it accumulates state as the commit is
125
prepared, and then it is discarded. It doesn't represent
126
historical revisions, just the act of recording a new one.
127
128
missing_ids
129
Modified to hold a list of files that have been deleted from
130
the working directory; these should be removed from the
131
working inventory.
132
"""
133
def __init__(self,
134
reporter=None):
135
if reporter is not None:
136
self.reporter = reporter
137
else:
138
self.reporter = NullCommitReporter()
139
140
141
def commit(self,
142
branch, message,
143
timestamp=None,
144
timezone=None,
145
committer=None,
146
specific_files=None,
147
rev_id=None,
148
allow_pointless=True,
149
verbose=False):
150
"""Commit working copy as a new revision.
151
152
timestamp -- if not None, seconds-since-epoch for a
153
postdated/predated commit.
154
155
specific_files -- If true, commit only those files.
156
157
rev_id -- If set, use this as the new revision id.
158
Useful for test or import commands that need to tightly
159
control what revisions are assigned. If you duplicate
160
a revision id that exists elsewhere it is your own fault.
161
If null (default), a time/random revision id is generated.
162
163
allow_pointless -- If true (default), commit even if nothing
164
has changed and no merges are recorded.
165
"""
166
mutter('preparing to commit')
167
168
self.branch = branch
169
self.weave_store = branch.weave_store
170
self.rev_id = rev_id
171
self.specific_files = specific_files
172
self.allow_pointless = allow_pointless
173
174
if timestamp is None:
175
self.timestamp = time.time()
176
else:
177
self.timestamp = long(timestamp)
178
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 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
179
119
if rev_id is None:
180
self.rev_id = _gen_revision_id(self.branch, self.timestamp)
181
else:
182
self.rev_id = rev_id
183
184
if committer is None:
185
self.committer = username(self.branch)
186
else:
187
assert isinstance(committer, basestring), type(committer)
188
self.committer = committer
189
190
if timezone is None:
191
self.timezone = local_time_offset()
192
else:
193
self.timezone = int(timezone)
194
195
assert isinstance(message, basestring), type(message)
196
self.message = message
197
self._escape_commit_message()
198
199
self.branch.lock_write()
200
try:
201
self.work_tree = self.branch.working_tree()
202
self.work_inv = self.work_tree.inventory
203
self.basis_tree = self.branch.basis_tree()
204
self.basis_inv = self.basis_tree.inventory
205
206
self._gather_parents()
207
if len(self.parents) > 1 and self.specific_files:
208
raise NotImplementedError('selected-file commit of merges is not supported yet')
209
self._check_parents_present()
210
211
self._remove_deleted()
212
self._populate_new_inv()
213
self._store_snapshot()
214
self._report_deletes()
215
216
if not (self.allow_pointless
217
or len(self.parents) > 1
218
or self.new_inv != self.basis_inv):
219
raise PointlessCommit()
220
221
self._record_inventory()
222
self._make_revision()
223
note('committed r%d {%s}', (self.branch.revno() + 1),
224
self.rev_id)
225
self.branch.append_revision(self.rev_id)
226
self.branch.set_pending_merges([])
227
finally:
228
self.branch.unlock()
229
230
def _record_inventory(self):
231
"""Store the inventory for the new revision."""
232
inv_text = serializer_v5.write_inventory_to_string(self.new_inv)
233
self.inv_sha1 = sha_string(inv_text)
234
s = self.branch.control_weaves
235
s.add_text('inventory', self.rev_id,
236
split_lines(inv_text), self.present_parents)
237
238
def _escape_commit_message(self):
239
"""Replace xml-incompatible control characters."""
240
# Python strings can include characters that can't be
241
# represented in well-formed XML; escape characters that
242
# aren't listed in the XML specification
243
# (http://www.w3.org/TR/REC-xml/#NT-Char).
244
if isinstance(self.message, unicode):
245
char_pattern = u'[^\x09\x0A\x0D\u0020-\uD7FF\uE000-\uFFFD]'
246
else:
247
# Use a regular 'str' as pattern to avoid having re.subn
248
# return 'unicode' results.
249
char_pattern = '[^x09\x0A\x0D\x20-\xFF]'
250
self.message, escape_count = re.subn(
251
char_pattern,
252
lambda match: match.group(0).encode('unicode_escape'),
253
self.message)
254
if escape_count:
255
note("replaced %d control characters in message", escape_count)
256
257
def _gather_parents(self):
258
"""Record the parents of a merge for merge detection."""
259
pending_merges = self.branch.pending_merges()
260
self.parents = []
261
self.parent_invs = []
262
self.present_parents = []
263
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()
264
157
if precursor_id:
265
self.parents.append(precursor_id)
266
self.parents += pending_merges
267
for revision in self.parents:
268
if self.branch.has_revision(revision):
269
self.parent_invs.append(self.branch.get_inventory(revision))
270
self.present_parents.append(revision)
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))
271
162
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
if parent_id == self.branch.last_revision():
277
warning("parent is pissing %r", parent_id)
278
raise HistoryMissing(self.branch, 'revision', parent_id)
279
else:
280
mutter("commit will ghost revision %r", parent_id)
281
282
def _make_revision(self):
283
"""Record a new revision object for this commit."""
284
self.rev = Revision(timestamp=self.timestamp,
285
timezone=self.timezone,
286
committer=self.committer,
287
message=self.message,
288
inventory_sha1=self.inv_sha1,
289
revision_id=self.rev_id)
290
self.rev.parent_ids = self.parents
291
rev_tmp = StringIO()
292
serializer_v5.write_revision(self.rev, rev_tmp)
163
rev_tmp = tempfile.TemporaryFile()
164
serializer_v4.write_revision(rev, rev_tmp)
293
165
rev_tmp.seek(0)
294
self.branch.revision_store.add(rev_tmp, self.rev_id)
295
mutter('new revision_id is {%s}', self.rev_id)
296
297
def _remove_deleted(self):
298
"""Remove deleted files from the working inventories.
299
300
This is done prior to taking the working inventory as the
301
basis for the new committed inventory.
302
303
This returns true if any files
304
*that existed in the basis inventory* were deleted.
305
Files that were added and deleted
306
in the working copy don't matter.
307
"""
308
specific = self.specific_files
309
deleted_ids = []
310
for path, ie in self.work_inv.iter_entries():
311
if specific and not is_inside_any(specific, path):
312
continue
313
if not self.work_tree.has_filename(path):
314
note('missing %s', path)
315
deleted_ids.append(ie.file_id)
316
if deleted_ids:
317
for file_id in deleted_ids:
318
if file_id in self.work_inv:
319
del self.work_inv[file_id]
320
self.branch._write_inventory(self.work_inv)
321
322
def _store_snapshot(self):
323
"""Pass over inventory and record a snapshot.
324
325
Entries get a new revision when they are modified in
326
any way, which includes a merge with a new set of
327
parents that have the same entry.
328
"""
329
# XXX: Need to think more here about when the user has
330
# made a specific decision on a particular value -- c.f.
331
# mark-merge.
332
for path, ie in self.new_inv.iter_entries():
333
previous_entries = ie.find_previous_heads(
334
self.parent_invs,
335
self.weave_store.get_weave_or_empty(ie.file_id))
336
if ie.revision is None:
337
change = ie.snapshot(self.rev_id, path, previous_entries,
338
self.work_tree, self.weave_store)
339
else:
340
change = "unchanged"
341
note("%s %s", change, path)
342
343
def _populate_new_inv(self):
344
"""Build revision inventory.
345
346
This creates a new empty inventory. Depending on
347
which files are selected for commit, and what is present in the
348
current tree, the new inventory is populated. inventory entries
349
which are candidates for modification have their revision set to
350
None; inventory entries that are carried over untouched have their
351
revision set to their prior value.
352
"""
353
mutter("Selecting files for commit with filter %s", self.specific_files)
354
self.new_inv = Inventory()
355
for path, new_ie in self.work_inv.iter_entries():
356
file_id = new_ie.file_id
357
mutter('check %s {%s}', path, new_ie.file_id)
358
if self.specific_files:
359
if not is_inside_any(self.specific_files, path):
360
mutter('%s not selected for commit', path)
361
self._carry_entry(file_id)
362
continue
363
else:
364
# this is selected, ensure its parents are too.
365
parent_id = new_ie.parent_id
366
while parent_id != ROOT_ID:
367
if not self.new_inv.has_id(parent_id):
368
ie = self._select_entry(self.work_inv[parent_id])
369
mutter('%s selected for commit because of %s',
370
self.new_inv.id2path(parent_id), path)
371
372
ie = self.new_inv[parent_id]
373
if ie.revision is not None:
374
ie.revision = None
375
mutter('%s selected for commit because of %s',
376
self.new_inv.id2path(parent_id), path)
377
parent_id = ie.parent_id
378
mutter('%s selected for commit', path)
379
self._select_entry(new_ie)
380
381
def _select_entry(self, new_ie):
382
"""Make new_ie be considered for committing."""
383
ie = new_ie.copy()
384
ie.revision = None
385
self.new_inv.add(ie)
386
return ie
387
388
def _carry_entry(self, file_id):
389
"""Carry the file unchanged from the basis revision."""
390
if self.basis_inv.has_id(file_id):
391
self.new_inv.add(self.basis_inv[file_id].copy())
392
393
def _report_deletes(self):
394
for file_id in self.basis_inv:
395
if file_id not in self.new_inv:
396
note('deleted %s', self.basis_inv.id2path(file_id))
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
191
397
192
398
193
def _gen_revision_id(branch, when):
399
194
"""Return new revision-id."""
195
from binascii import hexlify
196
from bzrlib.osutils import rand_bytes, compact_date, user_email
197
400
198
s = '%s-%s-' % (user_email(branch), compact_date(when))
401
199
s += hexlify(rand_bytes(8))
402
200
return s
201
202
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
elif old_ie != entry:
314
any_changes = True
315
316
return missing_ids, inv, any_changes
317
318
Loggerhead is a web-based interface for Breezy
Version: 2.0.1