1
# Copyright (C) 2005 Canonical Ltd
1
# Copyright (C) 2005, 2006 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
19
# FIXME: "bzr commit doc/format" commits doc/format.txt!
21
def commit(branch, message,
28
allow_pointless=True):
29
"""Commit working copy as a new revision.
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.
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.
45
This raises PointlessCommit if there are no changes, no new merges,
46
and allow_pointless is false.
48
timestamp -- if not None, seconds-since-epoch for a
49
postdated/predated commit.
52
If true, commit only those files.
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.
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
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
82
work_tree = branch.working_tree()
83
work_inv = work_tree.inventory
84
basis = branch.basis_tree()
85
basis_inv = basis.inventory
88
note('looking for changes...')
90
pending_merges = branch.pending_merges()
92
missing_ids, new_inv, any_changes = \
93
_gather_commit(branch,
100
if not (any_changes or allow_pointless or pending_merges):
101
raise PointlessCommit()
103
for file_id in missing_ids:
104
# Any files that have been deleted are now removed from the
105
# working inventory. Files that were not selected for commit
106
# are left as they were in the working inventory and ommitted
107
# from the revision inventory.
109
# have to do this later so we don't mess up the iterator.
110
# since parents may be removed before their children we
113
# FIXME: There's probably a better way to do this; perhaps
114
# the workingtree should know how to filter itbranch.
115
if work_inv.has_id(file_id):
116
del work_inv[file_id]
120
rev_id = _gen_revision_id(time.time())
123
inv_tmp = tempfile.TemporaryFile()
124
pack_xml(new_inv, inv_tmp)
126
branch.inventory_store.add(inv_tmp, inv_id)
127
mutter('new inventory_id is {%s}' % inv_id)
129
# We could also just sha hash the inv_tmp file
130
# however, in the case that branch.inventory_store.add()
131
# ever actually does anything special
132
inv_sha1 = branch.get_inventory_sha1(inv_id)
134
branch._write_inventory(work_inv)
136
if timestamp == None:
137
timestamp = time.time()
139
if committer == None:
140
committer = username()
143
timezone = local_time_offset()
145
mutter("building commit log message")
146
rev = Revision(timestamp=timestamp,
151
inventory_sha1=inv_sha1,
155
precursor_id = branch.last_patch()
157
precursor_sha1 = branch.get_revision_sha1(precursor_id)
158
rev.parents.append(RevisionReference(precursor_id, precursor_sha1))
159
for merge_rev in pending_merges:
160
rev.parents.append(RevisionReference(merge_rev))
162
rev_tmp = tempfile.TemporaryFile()
163
pack_xml(rev, rev_tmp)
165
branch.revision_store.add(rev_tmp, rev_id)
166
mutter("new revision_id is {%s}" % rev_id)
168
## XXX: Everything up to here can simply be orphaned if we abort
169
## the commit; it will leave junk files behind but that doesn't
172
## TODO: Read back the just-generated changeset, and make sure it
173
## applies and recreates the right state.
175
## TODO: Also calculate and store the inventory SHA1
176
mutter("committing patch r%d" % (branch.revno() + 1))
178
branch.append_revision(rev_id)
180
branch.set_pending_merges([])
183
note("commited r%d" % branch.revno())
189
def _gen_revision_id(when):
190
"""Return new revision-id."""
191
from binascii import hexlify
192
from osutils import rand_bytes, compact_date, user_email
194
s = '%s-%s-' % (user_email(), compact_date(when))
195
s += hexlify(rand_bytes(8))
199
def _gather_commit(branch, work_tree, work_inv, basis_inv, specific_files,
201
"""Build inventory preparatory to commit.
203
Returns missing_ids, new_inv, any_changes.
205
This adds any changed files into the text store, and sets their
206
test-id, sha and size in the returned inventory appropriately.
209
Modified to hold a list of files that have been deleted from
210
the working directory; these should be removed from the
213
from bzrlib.inventory import Inventory
214
from bzrlib.osutils import isdir, isfile, sha_string, quotefn, \
215
local_time_offset, username, kind_marker, is_inside_any
217
from bzrlib.branch import gen_file_id
218
from bzrlib.errors import BzrError
219
from bzrlib.revision import Revision
220
from bzrlib.trace import mutter, note
223
inv = Inventory(work_inv.root.file_id)
226
for path, entry in work_inv.iter_entries():
227
## TODO: Check that the file kind has not changed from the previous
228
## revision of this file (if any).
230
p = branch.abspath(path)
231
file_id = entry.file_id
232
mutter('commit prep file %s, id %r ' % (p, file_id))
234
if specific_files and not is_inside_any(specific_files, path):
235
mutter(' skipping file excluded from commit')
236
if basis_inv.has_id(file_id):
237
# carry over with previous state
238
inv.add(basis_inv[file_id].copy())
240
# omit this from committed inventory
244
if not work_tree.has_id(file_id):
246
print('deleted %s%s' % (path, kind_marker(entry.kind)))
248
mutter(" file is missing, removing from inventory")
249
missing_ids.append(file_id)
252
# this is present in the new inventory; may be new, modified or
254
old_ie = basis_inv.has_id(file_id) and basis_inv[file_id]
260
old_kind = old_ie.kind
261
if old_kind != entry.kind:
262
raise BzrError("entry %r changed kind from %r to %r"
263
% (file_id, old_kind, entry.kind))
265
if entry.kind == 'directory':
267
raise BzrError("%s is entered as directory but not a directory"
269
elif entry.kind == 'file':
271
raise BzrError("%s is entered as file but is not a file" % quotefn(p))
273
new_sha1 = work_tree.get_file_sha1(file_id)
276
and old_ie.text_sha1 == new_sha1):
277
## assert content == basis.get_file(file_id).read()
278
entry.text_id = old_ie.text_id
279
entry.text_sha1 = new_sha1
280
entry.text_size = old_ie.text_size
281
mutter(' unchanged from previous text_id {%s}' %
284
content = file(p, 'rb').read()
286
# calculate the sha again, just in case the file contents
287
# changed since we updated the cache
288
entry.text_sha1 = sha_string(content)
289
entry.text_size = len(content)
291
entry.text_id = gen_file_id(entry.name)
292
branch.text_store.add(content, entry.text_id)
293
mutter(' stored with text_id {%s}' % entry.text_id)
296
marked = path + kind_marker(entry.kind)
298
print 'added', marked
300
elif old_ie == entry:
302
elif (old_ie.name == entry.name
303
and old_ie.parent_id == entry.parent_id):
304
print 'modified', marked
307
print 'renamed', marked
310
return missing_ids, inv, any_changes
18
# XXX: Can we do any better about making interrupted commits change
21
# TODO: Separate 'prepare' phase where we find a list of potentially
22
# committed files. We then can then pause the commit to prompt for a
23
# commit message, knowing the summary will be the same as what's
24
# actually used for the commit. (But perhaps simpler to simply get
25
# the tree status, then use that for a selective commit?)
27
# The newly committed revision is going to have a shape corresponding
28
# to that of the working inventory. Files that are not in the
29
# working tree and that were in the predecessor are reported as
30
# removed --- this can include files that were either removed from the
31
# inventory or deleted in the working tree. If they were only
32
# deleted from disk, they are removed from the working inventory.
34
# We then consider the remaining entries, which will be in the new
35
# version. Directory entries are simply copied across. File entries
36
# must be checked to see if a new version of the file should be
37
# recorded. For each parent revision inventory, we check to see what
38
# version of the file was present. If the file was present in at
39
# least one tree, and if it was the same version in all the trees,
40
# then we can just refer to that version. Otherwise, a new version
41
# representing the merger of the file versions must be added.
43
# TODO: Update hashcache before and after - or does the WorkingTree
46
# TODO: Rather than mashing together the ancestry and storing it back,
47
# perhaps the weave should have single method which does it all in one
48
# go, avoiding a lot of redundant work.
50
# TODO: Perhaps give a warning if one of the revisions marked as
51
# merged is already in the ancestry, and then don't record it as a
54
# TODO: If the file is newly merged but unchanged from the version it
55
# merges from, then it should still be reported as newly added
56
# relative to the basis revision.
58
# TODO: Do checks that the tree can be committed *before* running the
59
# editor; this should include checks for a pointless commit and for
60
# unknown or missing files.
62
# TODO: If commit fails, leave the message in a file somewhere.
70
from cStringIO import StringIO
73
import bzrlib.errors as errors
74
from bzrlib.errors import (BzrError, PointlessCommit,
78
from bzrlib.osutils import (kind_marker, isdir,isfile, is_inside_any,
79
is_inside_or_parent_of_any,
80
quotefn, sha_file, split_lines)
81
from bzrlib.testament import Testament
82
from bzrlib.trace import mutter, note, warning
83
from bzrlib.xml5 import serializer_v5
84
from bzrlib.inventory import Inventory, ROOT_ID, InventoryEntry
85
from bzrlib import symbol_versioning
86
from bzrlib.symbol_versioning import (deprecated_passed,
89
from bzrlib.workingtree import WorkingTree
92
class NullCommitReporter(object):
93
"""I report on progress of a commit."""
95
def snapshot_change(self, change, path):
98
def completed(self, revno, rev_id):
101
def deleted(self, file_id):
104
def escaped(self, escape_count, message):
107
def missing(self, path):
110
def renamed(self, change, old_path, new_path):
114
class ReportCommitToLog(NullCommitReporter):
116
# this may be more useful if 'note' was replaced by an overridable
117
# method on self, which would allow more trivial subclassing.
118
# alternative, a callable could be passed in, allowing really trivial
119
# reuse for some uis. RBC 20060511
121
def snapshot_change(self, change, path):
122
if change == 'unchanged':
124
note("%s %s", change, path)
126
def completed(self, revno, rev_id):
127
note('Committed revision %d.', revno)
129
def deleted(self, file_id):
130
note('deleted %s', file_id)
132
def escaped(self, escape_count, message):
133
note("replaced %d control characters in message", escape_count)
135
def missing(self, path):
136
note('missing %s', path)
138
def renamed(self, change, old_path, new_path):
139
note('%s %s => %s', change, old_path, new_path)
142
class Commit(object):
143
"""Task of committing a new revision.
145
This is a MethodObject: it accumulates state as the commit is
146
prepared, and then it is discarded. It doesn't represent
147
historical revisions, just the act of recording a new one.
150
Modified to hold a list of files that have been deleted from
151
the working directory; these should be removed from the
157
if reporter is not None:
158
self.reporter = reporter
160
self.reporter = NullCommitReporter()
161
if config is not None:
167
branch=DEPRECATED_PARAMETER, message=None,
173
allow_pointless=True,
181
"""Commit working copy as a new revision.
183
branch -- the deprecated branch to commit to. New callers should pass in
186
message -- the commit message, a mandatory parameter
188
timestamp -- if not None, seconds-since-epoch for a
189
postdated/predated commit.
191
specific_files -- If true, commit only those files.
193
rev_id -- If set, use this as the new revision id.
194
Useful for test or import commands that need to tightly
195
control what revisions are assigned. If you duplicate
196
a revision id that exists elsewhere it is your own fault.
197
If null (default), a time/random revision id is generated.
199
allow_pointless -- If true (default), commit even if nothing
200
has changed and no merges are recorded.
202
strict -- If true, don't allow a commit if the working tree
203
contains unknown files.
205
revprops -- Properties for new revision
206
:param local: Perform a local only commit.
208
mutter('preparing to commit')
210
if deprecated_passed(branch):
211
symbol_versioning.warn("Commit.commit (branch, ...): The branch parameter is "
212
"deprecated as of bzr 0.8. Please use working_tree= instead.",
213
DeprecationWarning, stacklevel=2)
215
self.work_tree = self.branch.bzrdir.open_workingtree()
216
elif working_tree is None:
217
raise BzrError("One of branch and working_tree must be passed into commit().")
219
self.work_tree = working_tree
220
self.branch = self.work_tree.branch
222
raise BzrError("The message keyword parameter is required for commit().")
224
self.bound_branch = None
226
self.master_branch = None
227
self.master_locked = False
229
self.specific_files = specific_files
230
self.allow_pointless = allow_pointless
232
if reporter is None and self.reporter is None:
233
self.reporter = NullCommitReporter()
234
elif reporter is not None:
235
self.reporter = reporter
237
self.work_tree.lock_write()
238
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
240
# Cannot commit with conflicts present.
241
if len(self.work_tree.conflicts())>0:
242
raise ConflictsInTree
244
# setup the bound branch variables as needed.
245
self._check_bound_branch()
247
# check for out of date working trees
248
# if we are bound, then self.branch is the master branch and this
249
# test is thus all we need.
250
if self.work_tree.last_revision() != self.master_branch.last_revision():
251
raise errors.OutOfDateTree(self.work_tree)
254
# raise an exception as soon as we find a single unknown.
255
for unknown in self.work_tree.unknowns():
256
raise StrictCommitFailed()
258
if self.config is None:
259
self.config = self.branch.get_config()
261
if isinstance(message, str):
262
message = message.decode(bzrlib.user_encoding)
263
assert isinstance(message, unicode), type(message)
264
self.message = message
265
self._escape_commit_message()
267
self.work_inv = self.work_tree.inventory
268
self.basis_tree = self.work_tree.basis_tree()
269
self.basis_inv = self.basis_tree.inventory
270
# one to finish, one for rev and inventory, and one for each
271
# inventory entry, and the same for the new inventory.
272
# note that this estimate is too long when we do a partial tree
273
# commit which excludes some new files from being considered.
274
# The estimate is corrected when we populate the new inv.
275
self.pb_total = len(self.work_inv) + 5
278
self._gather_parents()
279
if len(self.parents) > 1 and self.specific_files:
280
raise NotImplementedError('selected-file commit of merges is not supported yet: files %r',
282
self._check_parents_present()
283
self.builder = self.branch.get_commit_builder(self.parents,
284
self.config, timestamp, timezone, committer, revprops, rev_id)
286
self._remove_deleted()
287
self._populate_new_inv()
288
self._report_deletes()
290
if not (self.allow_pointless
291
or len(self.parents) > 1
292
or self.builder.new_inventory != self.basis_inv):
293
raise PointlessCommit()
295
self._emit_progress_update()
296
# TODO: Now the new inventory is known, check for conflicts and
297
# prompt the user for a commit message.
298
# ADHB 2006-08-08: If this is done, populate_new_inv should not add
299
# weave lines, because nothing should be recorded until it is known
300
# that commit will succeed.
301
self.builder.finish_inventory()
302
self._emit_progress_update()
303
self.rev_id = self.builder.commit(self.message)
304
self._emit_progress_update()
305
# revision data is in the local branch now.
307
# upload revision data to the master.
308
# this will propagate merged revisions too if needed.
309
if self.bound_branch:
310
self.master_branch.repository.fetch(self.branch.repository,
311
revision_id=self.rev_id)
312
# now the master has the revision data
313
# 'commit' to the master first so a timeout here causes the local
314
# branch to be out of date
315
self.master_branch.append_revision(self.rev_id)
317
# and now do the commit locally.
318
self.branch.append_revision(self.rev_id)
320
self.work_tree.set_pending_merges([])
321
self.work_tree.set_last_revision(self.rev_id)
322
# now the work tree is up to date with the branch
324
self.reporter.completed(self.branch.revno(), self.rev_id)
325
if self.config.post_commit() is not None:
326
hooks = self.config.post_commit().split(' ')
327
# this would be nicer with twisted.python.reflect.namedAny
329
result = eval(hook + '(branch, rev_id)',
330
{'branch':self.branch,
332
'rev_id':self.rev_id})
333
self._emit_progress_update()
338
def _check_bound_branch(self):
339
"""Check to see if the local branch is bound.
341
If it is bound, then most of the commit will actually be
342
done using the remote branch as the target branch.
343
Only at the end will the local branch be updated.
345
if self.local and not self.branch.get_bound_location():
346
raise errors.LocalRequiresBoundBranch()
349
self.master_branch = self.branch.get_master_branch()
351
if not self.master_branch:
352
# make this branch the reference branch for out of date checks.
353
self.master_branch = self.branch
356
# If the master branch is bound, we must fail
357
master_bound_location = self.master_branch.get_bound_location()
358
if master_bound_location:
359
raise errors.CommitToDoubleBoundBranch(self.branch,
360
self.master_branch, master_bound_location)
362
# TODO: jam 20051230 We could automatically push local
363
# commits to the remote branch if they would fit.
364
# But for now, just require remote to be identical
367
# Make sure the local branch is identical to the master
368
master_rh = self.master_branch.revision_history()
369
local_rh = self.branch.revision_history()
370
if local_rh != master_rh:
371
raise errors.BoundBranchOutOfDate(self.branch,
374
# Now things are ready to change the master branch
376
self.bound_branch = self.branch
377
self.master_branch.lock_write()
378
self.master_locked = True
381
"""Cleanup any open locks, progress bars etc."""
382
cleanups = [self._cleanup_bound_branch,
383
self.work_tree.unlock,
385
found_exception = None
386
for cleanup in cleanups:
389
# we want every cleanup to run no matter what.
390
# so we have a catchall here, but we will raise the
391
# last encountered exception up the stack: and
392
# typically this will be useful enough.
395
if found_exception is not None:
396
# don't do a plan raise, because the last exception may have been
397
# trashed, e is our sure-to-work exception even though it loses the
398
# full traceback. XXX: RBC 20060421 perhaps we could check the
399
# exc_info and if its the same one do a plain raise otherwise
400
# 'raise e' as we do now.
403
def _cleanup_bound_branch(self):
404
"""Executed at the end of a try/finally to cleanup a bound branch.
406
If the branch wasn't bound, this is a no-op.
407
If it was, it resents self.branch to the local branch, instead
410
if not self.bound_branch:
412
if self.master_locked:
413
self.master_branch.unlock()
415
def _escape_commit_message(self):
416
"""Replace xml-incompatible control characters."""
417
# FIXME: RBC 20060419 this should be done by the revision
418
# serialiser not by commit. Then we can also add an unescaper
419
# in the deserializer and start roundtripping revision messages
420
# precisely. See repository_implementations/test_repository.py
422
# Python strings can include characters that can't be
423
# represented in well-formed XML; escape characters that
424
# aren't listed in the XML specification
425
# (http://www.w3.org/TR/REC-xml/#NT-Char).
426
self.message, escape_count = re.subn(
427
u'[^\x09\x0A\x0D\u0020-\uD7FF\uE000-\uFFFD]+',
428
lambda match: match.group(0).encode('unicode_escape'),
431
self.reporter.escaped(escape_count, self.message)
433
def _gather_parents(self):
434
"""Record the parents of a merge for merge detection."""
435
# TODO: Make sure that this list doesn't contain duplicate
436
# entries and the order is preserved when doing this.
437
self.parents = self.work_tree.get_parent_ids()
438
self.parent_invs = []
439
for revision in self.parents:
440
if self.branch.repository.has_revision(revision):
441
inventory = self.branch.repository.get_inventory(revision)
442
self.parent_invs.append(inventory)
444
def _check_parents_present(self):
445
for parent_id in self.parents:
446
mutter('commit parent revision {%s}', parent_id)
447
if not self.branch.repository.has_revision(parent_id):
448
if parent_id == self.branch.last_revision():
449
warning("parent is missing %r", parent_id)
450
raise BzrCheckError("branch %s is missing revision {%s}"
451
% (self.branch, parent_id))
453
def _remove_deleted(self):
454
"""Remove deleted files from the working inventories.
456
This is done prior to taking the working inventory as the
457
basis for the new committed inventory.
459
This returns true if any files
460
*that existed in the basis inventory* were deleted.
461
Files that were added and deleted
462
in the working copy don't matter.
464
specific = self.specific_files
466
for path, ie in self.work_inv.iter_entries():
467
if specific and not is_inside_any(specific, path):
469
if not self.work_tree.has_filename(path):
470
self.reporter.missing(path)
471
deleted_ids.append((path, ie.file_id))
473
deleted_ids.sort(reverse=True)
474
for path, file_id in deleted_ids:
475
del self.work_inv[file_id]
476
self.work_tree._write_inventory(self.work_inv)
478
def _populate_new_inv(self):
479
"""Build revision inventory.
481
This creates a new empty inventory. Depending on
482
which files are selected for commit, and what is present in the
483
current tree, the new inventory is populated. inventory entries
484
which are candidates for modification have their revision set to
485
None; inventory entries that are carried over untouched have their
486
revision set to their prior value.
488
# ESEPARATIONOFCONCERNS: this function is diffing and using the diff
489
# results to create a new inventory at the same time, which results
490
# in bugs like #46635. Any reason not to use/enhance Tree.changes_from?
492
mutter("Selecting files for commit with filter %s", self.specific_files)
493
entries = self.work_inv.iter_entries()
494
if not self.builder.record_root_entry:
495
symbol_versioning.warn('CommitBuilders should support recording'
496
' the root entry as of bzr 0.10.', DeprecationWarning,
498
self.builder.new_inventory.add(self.basis_inv.root.copy())
500
self._emit_progress_update()
501
for path, new_ie in entries:
502
self._emit_progress_update()
503
file_id = new_ie.file_id
504
# mutter('check %s {%s}', path, file_id)
505
if (not self.specific_files or
506
is_inside_or_parent_of_any(self.specific_files, path)):
507
# mutter('%s selected for commit', path)
511
# mutter('%s not selected for commit', path)
512
if self.basis_inv.has_id(file_id):
513
ie = self.basis_inv[file_id].copy()
515
# this entry is new and not being committed
518
self.builder.record_entry_contents(ie, self.parent_invs,
519
path, self.work_tree)
520
# describe the nature of the change that has occurred relative to
521
# the basis inventory.
522
if (self.basis_inv.has_id(ie.file_id)):
523
basis_ie = self.basis_inv[ie.file_id]
526
change = ie.describe_change(basis_ie, ie)
527
if change in (InventoryEntry.RENAMED,
528
InventoryEntry.MODIFIED_AND_RENAMED):
529
old_path = self.basis_inv.id2path(ie.file_id)
530
self.reporter.renamed(change, old_path, path)
532
self.reporter.snapshot_change(change, path)
534
if not self.specific_files:
537
# ignore removals that don't match filespec
538
for path, new_ie in self.basis_inv.iter_entries():
539
if new_ie.file_id in self.work_inv:
541
if is_inside_any(self.specific_files, path):
545
self.builder.record_entry_contents(ie, self.parent_invs, path,
548
def _emit_progress_update(self):
549
"""Emit an update to the progress bar."""
550
self.pb.update("Committing", self.pb_count, self.pb_total)
553
def _report_deletes(self):
554
for path, ie in self.basis_inv.iter_entries():
555
if ie.file_id not in self.builder.new_inventory:
556
self.reporter.deleted(path)
added
removed
bzrlib/commit.py
