1
1
# Copyright (C) 2005 Canonical Ltd
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.
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.
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
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.
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?)
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.
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.
44
# TODO: Update hashcache before and after - or does the WorkingTree
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.
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
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.
59
# TODO: Do checks that the tree can be committed *before* running the
60
# editor; this should include checks for a pointless commit and for
61
# unknown or missing files.
63
# TODO: If commit fails, leave the message in a file somewhere.
72
from binascii import hexlify
73
from cStringIO import StringIO
75
from bzrlib.atomicfile import AtomicFile
76
from bzrlib.osutils import (local_time_offset,
77
rand_bytes, compact_date,
78
kind_marker, is_inside_any, quotefn,
79
sha_string, sha_strings, sha_file, isdir, isfile,
82
from bzrlib.errors import (BzrError, PointlessCommit,
87
import bzrlib.gpg as gpg
88
from bzrlib.revision import Revision
89
from bzrlib.testament import Testament
90
from bzrlib.trace import mutter, note, warning
91
from bzrlib.xml5 import serializer_v5
92
from bzrlib.inventory import Inventory, ROOT_ID
93
from bzrlib.weave import Weave
94
from bzrlib.weavefile import read_weave, write_weave_v5
95
from bzrlib.workingtree import WorkingTree
98
def commit(*args, **kwargs):
99
"""Commit a new revision to a branch.
101
Function-style interface for convenience of old callers.
103
New code should use the Commit class instead.
105
## XXX: Remove this in favor of Branch.commit?
106
Commit().commit(*args, **kwargs)
109
class NullCommitReporter(object):
110
"""I report on progress of a commit."""
112
def snapshot_change(self, change, path):
115
def completed(self, revno, rev_id):
118
def deleted(self, file_id):
121
def escaped(self, escape_count, message):
124
def missing(self, path):
128
class ReportCommitToLog(NullCommitReporter):
130
def snapshot_change(self, change, path):
131
note("%s %s", change, path)
133
def completed(self, revno, rev_id):
134
note('committed r%d {%s}', revno, rev_id)
136
def deleted(self, file_id):
137
note('deleted %s', file_id)
139
def escaped(self, escape_count, message):
140
note("replaced %d control characters in message", escape_count)
142
def missing(self, path):
143
note('missing %s', path)
146
class Commit(object):
147
"""Task of committing a new revision.
149
This is a MethodObject: it accumulates state as the commit is
150
prepared, and then it is discarded. It doesn't represent
151
historical revisions, just the act of recording a new one.
154
Modified to hold a list of files that have been deleted from
155
the working directory; these should be removed from the
161
if reporter is not None:
162
self.reporter = reporter
164
self.reporter = NullCommitReporter()
165
if config is not None:
177
allow_pointless=True,
181
"""Commit working copy as a new revision.
183
timestamp -- if not None, seconds-since-epoch for a
184
postdated/predated commit.
186
specific_files -- If true, commit only those files.
188
rev_id -- If set, use this as the new revision id.
189
Useful for test or import commands that need to tightly
190
control what revisions are assigned. If you duplicate
191
a revision id that exists elsewhere it is your own fault.
192
If null (default), a time/random revision id is generated.
194
allow_pointless -- If true (default), commit even if nothing
195
has changed and no merges are recorded.
197
strict -- If true, don't allow a commit if the working tree
198
contains unknown files.
200
revprops -- Properties for new revision
202
mutter('preparing to commit')
205
self.weave_store = branch.repository.weave_store
207
self.specific_files = specific_files
208
self.allow_pointless = allow_pointless
209
self.revprops = {'branch-nick': branch.nick}
211
self.revprops.update(revprops)
212
self.work_tree = WorkingTree(branch.base, branch)
215
# raise an exception as soon as we find a single unknown.
216
for unknown in self.work_tree.unknowns():
217
raise StrictCommitFailed()
219
if timestamp is None:
220
self.timestamp = time.time()
222
self.timestamp = long(timestamp)
224
if self.config is None:
225
self.config = bzrlib.config.BranchConfig(self.branch)
19
def commit(branch, message,
26
allow_pointless=True):
27
"""Commit working copy as a new revision.
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.
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.
43
This raises PointlessCommit if there are no changes, no new merges,
44
and allow_pointless is false.
46
timestamp -- if not None, seconds-since-epoch for a
47
postdated/predated commit.
50
If true, commit only those files.
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.
60
import time, tempfile, re
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
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
80
work_tree = branch.working_tree()
81
work_inv = work_tree.inventory
82
basis = branch.basis_tree()
83
basis_inv = basis.inventory
86
# note('looking for changes...')
87
# print 'looking for changes...'
88
# disabled; should be done at a higher level
91
pending_merges = branch.pending_merges()
93
missing_ids, new_inv, any_changes = \
94
_gather_commit(branch,
101
if not (any_changes or allow_pointless or pending_merges):
102
raise PointlessCommit()
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.
110
# have to do this later so we don't mess up the iterator.
111
# since parents may be removed before their children we
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]
227
119
if rev_id is None:
228
self.rev_id = _gen_revision_id(self.config, self.timestamp)
232
if committer is None:
233
self.committer = self.config.username()
235
assert isinstance(committer, basestring), type(committer)
236
self.committer = committer
239
self.timezone = local_time_offset()
241
self.timezone = int(timezone)
243
if isinstance(message, str):
244
message = message.decode(bzrlib.user_encoding)
245
assert isinstance(message, unicode), type(message)
246
self.message = message
247
self._escape_commit_message()
249
self.branch.lock_write()
251
self.work_inv = self.work_tree.inventory
252
self.basis_tree = self.branch.basis_tree()
253
self.basis_inv = self.basis_tree.inventory
255
self._gather_parents()
256
if len(self.parents) > 1 and self.specific_files:
257
raise NotImplementedError('selected-file commit of merges is not supported yet')
258
self._check_parents_present()
260
self._remove_deleted()
261
self._populate_new_inv()
262
self._store_snapshot()
263
self._report_deletes()
265
if not (self.allow_pointless
266
or len(self.parents) > 1
267
or self.new_inv != self.basis_inv):
268
raise PointlessCommit()
270
if len(list(self.work_tree.iter_conflicts()))>0:
271
raise ConflictsInTree
273
self._record_inventory()
274
self._make_revision()
275
self.work_tree.set_pending_merges([])
276
self.branch.append_revision(self.rev_id)
277
self.reporter.completed(self.branch.revno()+1, self.rev_id)
278
if self.config.post_commit() is not None:
279
hooks = self.config.post_commit().split(' ')
280
# this would be nicer with twisted.python.reflect.namedAny
282
result = eval(hook + '(branch, rev_id)',
283
{'branch':self.branch,
285
'rev_id':self.rev_id})
289
def _record_inventory(self):
290
"""Store the inventory for the new revision."""
291
inv_text = serializer_v5.write_inventory_to_string(self.new_inv)
292
self.inv_sha1 = sha_string(inv_text)
293
s = self.branch.repository.control_weaves
294
s.add_text('inventory', self.rev_id,
295
split_lines(inv_text), self.present_parents,
296
self.branch.get_transaction())
298
def _escape_commit_message(self):
299
"""Replace xml-incompatible control characters."""
120
rev_id = _gen_revision_id(branch, time.time())
123
inv_tmp = tempfile.TemporaryFile()
125
serializer_v4.write_inventory(new_inv, inv_tmp)
127
branch.inventory_store.add(inv_tmp, inv_id)
128
mutter('new inventory_id is {%s}' % inv_id)
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)
135
branch._write_inventory(work_inv)
137
if timestamp == None:
138
timestamp = time.time()
140
if committer == None:
141
committer = username(branch)
144
timezone = local_time_offset()
146
mutter("building commit log message")
300
147
# Python strings can include characters that can't be
301
148
# represented in well-formed XML; escape characters that
302
149
# aren't listed in the XML specification
303
150
# (http://www.w3.org/TR/REC-xml/#NT-Char).
304
self.message, escape_count = re.subn(
305
u'[^\x09\x0A\x0D\u0020-\uD7FF\uE000-\uFFFD]+',
151
if isinstance(message, unicode):
152
char_pattern = u'[^\x09\x0A\x0D\u0020-\uD7FF\uE000-\uFFFD]'
154
# Use a regular 'str' as pattern to avoid having re.subn
155
# return 'unicode' results.
156
char_pattern = '[^x09\x0A\x0D\x20-\xFF]'
157
message, escape_count = re.subn(
306
159
lambda match: match.group(0).encode('unicode_escape'),
309
self.reporter.escaped(escape_count, self.message)
162
note("replaced %d control characters in message", escape_count)
163
rev = Revision(timestamp=timestamp,
168
inventory_sha1=inv_sha1,
311
def _gather_parents(self):
312
"""Record the parents of a merge for merge detection."""
313
pending_merges = self.work_tree.pending_merges()
315
self.parent_invs = []
316
self.present_parents = []
317
precursor_id = self.branch.last_revision()
172
precursor_id = branch.last_patch()
319
self.parents.append(precursor_id)
320
self.parents += pending_merges
321
for revision in self.parents:
322
if self.branch.repository.has_revision(revision):
323
inventory = self.branch.repository.get_inventory(revision)
324
self.parent_invs.append(inventory)
325
self.present_parents.append(revision)
174
precursor_sha1 = branch.get_revision_sha1(precursor_id)
175
rev.parents.append(RevisionReference(precursor_id, precursor_sha1))
176
for merge_rev in pending_merges:
177
rev.parents.append(RevisionReference(merge_rev))
327
def _check_parents_present(self):
328
for parent_id in self.parents:
329
mutter('commit parent revision {%s}', parent_id)
330
if not self.branch.repository.has_revision(parent_id):
331
if parent_id == self.branch.last_revision():
332
warning("parent is missing %r", parent_id)
333
raise HistoryMissing(self.branch, 'revision', parent_id)
335
mutter("commit will ghost revision %r", parent_id)
337
def _make_revision(self):
338
"""Record a new revision object for this commit."""
339
self.rev = Revision(timestamp=self.timestamp,
340
timezone=self.timezone,
341
committer=self.committer,
342
message=self.message,
343
inventory_sha1=self.inv_sha1,
344
revision_id=self.rev_id,
345
properties=self.revprops)
346
self.rev.parent_ids = self.parents
348
serializer_v5.write_revision(self.rev, rev_tmp)
179
rev_tmp = tempfile.TemporaryFile()
180
serializer_v4.write_revision(rev, rev_tmp)
350
if self.config.signature_needed():
351
plaintext = Testament(self.rev, self.new_inv).as_short_text()
352
self.branch.repository.store_revision_signature(
353
gpg.GPGStrategy(self.config), plaintext, self.rev_id)
354
self.branch.repository.revision_store.add(rev_tmp, self.rev_id)
355
mutter('new revision_id is {%s}', self.rev_id)
357
def _remove_deleted(self):
358
"""Remove deleted files from the working inventories.
360
This is done prior to taking the working inventory as the
361
basis for the new committed inventory.
363
This returns true if any files
364
*that existed in the basis inventory* were deleted.
365
Files that were added and deleted
366
in the working copy don't matter.
368
specific = self.specific_files
370
for path, ie in self.work_inv.iter_entries():
371
if specific and not is_inside_any(specific, path):
373
if not self.work_tree.has_filename(path):
374
self.reporter.missing(path)
375
deleted_ids.append((path, ie.file_id))
377
deleted_ids.sort(reverse=True)
378
for path, file_id in deleted_ids:
379
del self.work_inv[file_id]
380
self.work_tree._write_inventory(self.work_inv)
382
def _store_snapshot(self):
383
"""Pass over inventory and record a snapshot.
385
Entries get a new revision when they are modified in
386
any way, which includes a merge with a new set of
387
parents that have the same entry.
389
# XXX: Need to think more here about when the user has
390
# made a specific decision on a particular value -- c.f.
392
for path, ie in self.new_inv.iter_entries():
393
previous_entries = ie.find_previous_heads(
395
self.weave_store.get_weave_prelude_or_empty(ie.file_id,
396
self.branch.get_transaction()))
397
if ie.revision is None:
398
change = ie.snapshot(self.rev_id, path, previous_entries,
399
self.work_tree, self.weave_store,
400
self.branch.get_transaction())
403
self.reporter.snapshot_change(change, path)
405
def _populate_new_inv(self):
406
"""Build revision inventory.
408
This creates a new empty inventory. Depending on
409
which files are selected for commit, and what is present in the
410
current tree, the new inventory is populated. inventory entries
411
which are candidates for modification have their revision set to
412
None; inventory entries that are carried over untouched have their
413
revision set to their prior value.
415
mutter("Selecting files for commit with filter %s", self.specific_files)
416
self.new_inv = Inventory()
417
for path, new_ie in self.work_inv.iter_entries():
418
file_id = new_ie.file_id
419
mutter('check %s {%s}', path, new_ie.file_id)
420
if self.specific_files:
421
if not is_inside_any(self.specific_files, path):
422
mutter('%s not selected for commit', path)
423
self._carry_entry(file_id)
426
# this is selected, ensure its parents are too.
427
parent_id = new_ie.parent_id
428
while parent_id != ROOT_ID:
429
if not self.new_inv.has_id(parent_id):
430
ie = self._select_entry(self.work_inv[parent_id])
431
mutter('%s selected for commit because of %s',
432
self.new_inv.id2path(parent_id), path)
434
ie = self.new_inv[parent_id]
435
if ie.revision is not None:
437
mutter('%s selected for commit because of %s',
438
self.new_inv.id2path(parent_id), path)
439
parent_id = ie.parent_id
440
mutter('%s selected for commit', path)
441
self._select_entry(new_ie)
443
def _select_entry(self, new_ie):
444
"""Make new_ie be considered for committing."""
450
def _carry_entry(self, file_id):
451
"""Carry the file unchanged from the basis revision."""
452
if self.basis_inv.has_id(file_id):
453
self.new_inv.add(self.basis_inv[file_id].copy())
455
def _report_deletes(self):
456
for file_id in self.basis_inv:
457
if file_id not in self.new_inv:
458
self.reporter.deleted(self.basis_inv.id2path(file_id))
460
def _gen_revision_id(config, when):
182
branch.revision_store.add(rev_tmp, rev_id)
183
mutter("new revision_id is {%s}" % rev_id)
185
## XXX: Everything up to here can simply be orphaned if we abort
186
## the commit; it will leave junk files behind but that doesn't
189
## TODO: Read back the just-generated changeset, and make sure it
190
## applies and recreates the right state.
192
## TODO: Also calculate and store the inventory SHA1
193
mutter("committing patch r%d" % (branch.revno() + 1))
195
branch.append_revision(rev_id)
197
branch.set_pending_merges([])
200
# disabled; should go through logging
201
# note("commited r%d" % branch.revno())
202
# print ("commited r%d" % branch.revno())
209
def _gen_revision_id(branch, when):
461
210
"""Return new revision-id."""
462
s = '%s-%s-' % (config.user_email(), compact_date(when))
211
from binascii import hexlify
212
from bzrlib.osutils import rand_bytes, compact_date, user_email
214
s = '%s-%s-' % (user_email(branch), compact_date(when))
463
215
s += hexlify(rand_bytes(8))
219
def _gather_commit(branch, work_tree, work_inv, basis_inv, specific_files,
221
"""Build inventory preparatory to commit.
223
Returns missing_ids, new_inv, any_changes.
225
This adds any changed files into the text store, and sets their
226
test-id, sha and size in the returned inventory appropriately.
229
Modified to hold a list of files that have been deleted from
230
the working directory; these should be removed from the
233
from bzrlib.inventory import Inventory
234
from bzrlib.osutils import isdir, isfile, sha_string, quotefn, \
235
local_time_offset, username, kind_marker, is_inside_any
237
from bzrlib.branch import gen_file_id
238
from bzrlib.errors import BzrError
239
from bzrlib.revision import Revision
240
from bzrlib.trace import mutter, note
243
inv = Inventory(work_inv.root.file_id)
246
for path, entry in work_inv.iter_entries():
247
## TODO: Check that the file kind has not changed from the previous
248
## revision of this file (if any).
250
p = branch.abspath(path)
251
file_id = entry.file_id
252
mutter('commit prep file %s, id %r ' % (p, file_id))
254
if specific_files and not is_inside_any(specific_files, path):
255
mutter(' skipping file excluded from commit')
256
if basis_inv.has_id(file_id):
257
# carry over with previous state
258
inv.add(basis_inv[file_id].copy())
260
# omit this from committed inventory
264
if not work_tree.has_id(file_id):
266
print('deleted %s%s' % (path, kind_marker(entry.kind)))
268
mutter(" file is missing, removing from inventory")
269
missing_ids.append(file_id)
272
# this is present in the new inventory; may be new, modified or
274
old_ie = basis_inv.has_id(file_id) and basis_inv[file_id]
280
old_kind = old_ie.kind
281
if old_kind != entry.kind:
282
raise BzrError("entry %r changed kind from %r to %r"
283
% (file_id, old_kind, entry.kind))
285
if entry.kind == 'directory':
287
raise BzrError("%s is entered as directory but not a directory"
289
elif entry.kind == 'file':
291
raise BzrError("%s is entered as file but is not a file" % quotefn(p))
293
new_sha1 = work_tree.get_file_sha1(file_id)
296
and old_ie.text_sha1 == new_sha1):
297
## assert content == basis.get_file(file_id).read()
298
entry.text_id = old_ie.text_id
299
entry.text_sha1 = new_sha1
300
entry.text_size = old_ie.text_size
301
mutter(' unchanged from previous text_id {%s}' %
304
content = file(p, 'rb').read()
306
# calculate the sha again, just in case the file contents
307
# changed since we updated the cache
308
entry.text_sha1 = sha_string(content)
309
entry.text_size = len(content)
311
entry.text_id = gen_file_id(entry.name)
312
branch.text_store.add(content, entry.text_id)
313
mutter(' stored with text_id {%s}' % entry.text_id)
316
marked = path + kind_marker(entry.kind)
318
print 'added', marked
320
elif old_ie == entry:
322
elif (old_ie.name == entry.name
323
and old_ie.parent_id == entry.parent_id):
324
print 'modified', marked
327
print 'renamed', marked
329
elif old_ie != entry:
332
return missing_ids, inv, any_changes
added
removed
bzrlib/commit.py
