1
# Copyright (C) 2005, 2006 Canonical Ltd
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
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.
64
# TODO: Change the parameter 'rev_id' to 'revision_id' to be consistent with
65
# the rest of the code; add a deprecation of the old name.
72
from cStringIO import StringIO
75
import bzrlib.errors as errors
76
from bzrlib.errors import (BzrError, PointlessCommit,
80
from bzrlib.osutils import (kind_marker, isdir,isfile, is_inside_any,
81
is_inside_or_parent_of_any,
82
quotefn, sha_file, split_lines)
83
from bzrlib.testament import Testament
84
from bzrlib.trace import mutter, note, warning
85
from bzrlib.xml5 import serializer_v5
86
from bzrlib.inventory import Inventory, ROOT_ID, InventoryEntry
87
from bzrlib import symbol_versioning
88
from bzrlib.symbol_versioning import (deprecated_passed,
91
from bzrlib.workingtree import WorkingTree
94
class NullCommitReporter(object):
95
"""I report on progress of a commit."""
97
def snapshot_change(self, change, path):
100
def completed(self, revno, rev_id):
103
def deleted(self, file_id):
106
def escaped(self, escape_count, message):
109
def missing(self, path):
112
def renamed(self, change, old_path, new_path):
116
class ReportCommitToLog(NullCommitReporter):
118
# this may be more useful if 'note' was replaced by an overridable
119
# method on self, which would allow more trivial subclassing.
120
# alternative, a callable could be passed in, allowing really trivial
121
# reuse for some uis. RBC 20060511
123
def snapshot_change(self, change, path):
124
if change == 'unchanged':
126
note("%s %s", change, path)
128
def completed(self, revno, rev_id):
129
note('Committed revision %d.', revno)
131
def deleted(self, file_id):
132
note('deleted %s', file_id)
134
def escaped(self, escape_count, message):
135
note("replaced %d control characters in message", escape_count)
137
def missing(self, path):
138
note('missing %s', path)
140
def renamed(self, change, old_path, new_path):
141
note('%s %s => %s', change, old_path, new_path)
144
class Commit(object):
145
"""Task of committing a new revision.
147
This is a MethodObject: it accumulates state as the commit is
148
prepared, and then it is discarded. It doesn't represent
149
historical revisions, just the act of recording a new one.
152
Modified to hold a list of files that have been deleted from
153
the working directory; these should be removed from the
159
if reporter is not None:
160
self.reporter = reporter
162
self.reporter = NullCommitReporter()
163
if config is not None:
169
branch=DEPRECATED_PARAMETER, message=None,
175
allow_pointless=True,
183
"""Commit working copy as a new revision.
185
branch -- the deprecated branch to commit to. New callers should pass in
188
message -- the commit message, a mandatory parameter
190
timestamp -- if not None, seconds-since-epoch for a
191
postdated/predated commit.
193
specific_files -- If true, commit only those files.
195
rev_id -- If set, use this as the new revision id.
196
Useful for test or import commands that need to tightly
197
control what revisions are assigned. If you duplicate
198
a revision id that exists elsewhere it is your own fault.
199
If null (default), a time/random revision id is generated.
201
allow_pointless -- If true (default), commit even if nothing
202
has changed and no merges are recorded.
204
strict -- If true, don't allow a commit if the working tree
205
contains unknown files.
207
revprops -- Properties for new revision
208
:param local: Perform a local only commit.
210
mutter('preparing to commit')
212
if deprecated_passed(branch):
213
symbol_versioning.warn("Commit.commit (branch, ...): The branch parameter is "
214
"deprecated as of bzr 0.8. Please use working_tree= instead.",
215
DeprecationWarning, stacklevel=2)
217
self.work_tree = self.branch.bzrdir.open_workingtree()
218
elif working_tree is None:
219
raise BzrError("One of branch and working_tree must be passed into commit().")
221
self.work_tree = working_tree
222
self.branch = self.work_tree.branch
224
raise BzrError("The message keyword parameter is required for commit().")
226
self.bound_branch = None
228
self.master_branch = None
229
self.master_locked = False
231
self.specific_files = specific_files
232
self.allow_pointless = allow_pointless
234
if reporter is None and self.reporter is None:
235
self.reporter = NullCommitReporter()
236
elif reporter is not None:
237
self.reporter = reporter
239
self.work_tree.lock_write()
240
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
242
# Cannot commit with conflicts present.
243
if len(self.work_tree.conflicts())>0:
244
raise ConflictsInTree
246
# setup the bound branch variables as needed.
247
self._check_bound_branch()
249
# check for out of date working trees
251
first_tree_parent = self.work_tree.get_parent_ids()[0]
253
# if there are no parents, treat our parent as 'None'
254
# this is so that we still consier the master branch
255
# - in a checkout scenario the tree may have no
256
# parents but the branch may do.
257
first_tree_parent = None
258
master_last = self.master_branch.last_revision()
259
if (master_last is not None and
260
master_last != first_tree_parent):
261
raise errors.OutOfDateTree(self.work_tree)
264
# raise an exception as soon as we find a single unknown.
265
for unknown in self.work_tree.unknowns():
266
raise StrictCommitFailed()
268
if self.config is None:
269
self.config = self.branch.get_config()
271
if isinstance(message, str):
272
message = message.decode(bzrlib.user_encoding)
273
assert isinstance(message, unicode), type(message)
274
self.message = message
275
self._escape_commit_message()
277
self.work_inv = self.work_tree.inventory
278
self.basis_tree = self.work_tree.basis_tree()
279
self.basis_inv = self.basis_tree.inventory
280
# one to finish, one for rev and inventory, and one for each
281
# inventory entry, and the same for the new inventory.
282
# note that this estimate is too long when we do a partial tree
283
# commit which excludes some new files from being considered.
284
# The estimate is corrected when we populate the new inv.
285
self.pb_total = len(self.work_inv) + 5
288
self._gather_parents()
289
if len(self.parents) > 1 and self.specific_files:
290
raise NotImplementedError('selected-file commit of merges is not supported yet: files %r',
293
self.builder = self.branch.get_commit_builder(self.parents,
294
self.config, timestamp, timezone, committer, revprops, rev_id)
296
self._remove_deleted()
297
self._populate_new_inv()
298
self._report_deletes()
300
self._check_pointless()
302
self._emit_progress_update()
303
# TODO: Now the new inventory is known, check for conflicts and
304
# prompt the user for a commit message.
305
# ADHB 2006-08-08: If this is done, populate_new_inv should not add
306
# weave lines, because nothing should be recorded until it is known
307
# that commit will succeed.
308
self.builder.finish_inventory()
309
self._emit_progress_update()
310
self.rev_id = self.builder.commit(self.message)
311
self._emit_progress_update()
312
# revision data is in the local branch now.
314
# upload revision data to the master.
315
# this will propagate merged revisions too if needed.
316
if self.bound_branch:
317
self.master_branch.repository.fetch(self.branch.repository,
318
revision_id=self.rev_id)
319
# now the master has the revision data
320
# 'commit' to the master first so a timeout here causes the local
321
# branch to be out of date
322
self.master_branch.append_revision(self.rev_id)
324
# and now do the commit locally.
325
self.branch.append_revision(self.rev_id)
327
# if the builder gave us the revisiontree it created back, we
328
# could use it straight away here.
329
# TODO: implement this.
330
self.work_tree.set_parent_trees([(self.rev_id,
331
self.branch.repository.revision_tree(self.rev_id))])
332
# now the work tree is up to date with the branch
334
self.reporter.completed(self.branch.revno(), self.rev_id)
335
if self.config.post_commit() is not None:
336
hooks = self.config.post_commit().split(' ')
337
# this would be nicer with twisted.python.reflect.namedAny
339
result = eval(hook + '(branch, rev_id)',
340
{'branch':self.branch,
342
'rev_id':self.rev_id})
343
self._emit_progress_update()
348
def _check_pointless(self):
349
if self.allow_pointless:
351
# A merge with no effect on files
352
if len(self.parents) > 1:
354
# work around the fact that a newly-initted tree does differ from its
356
if len(self.builder.new_inventory) != len(self.basis_inv):
358
if (len(self.builder.new_inventory) != 1 and
359
self.builder.new_inventory != self.basis_inv):
361
raise PointlessCommit()
363
def _check_bound_branch(self):
364
"""Check to see if the local branch is bound.
366
If it is bound, then most of the commit will actually be
367
done using the remote branch as the target branch.
368
Only at the end will the local branch be updated.
370
if self.local and not self.branch.get_bound_location():
371
raise errors.LocalRequiresBoundBranch()
374
self.master_branch = self.branch.get_master_branch()
376
if not self.master_branch:
377
# make this branch the reference branch for out of date checks.
378
self.master_branch = self.branch
381
# If the master branch is bound, we must fail
382
master_bound_location = self.master_branch.get_bound_location()
383
if master_bound_location:
384
raise errors.CommitToDoubleBoundBranch(self.branch,
385
self.master_branch, master_bound_location)
387
# TODO: jam 20051230 We could automatically push local
388
# commits to the remote branch if they would fit.
389
# But for now, just require remote to be identical
392
# Make sure the local branch is identical to the master
393
master_rh = self.master_branch.revision_history()
394
local_rh = self.branch.revision_history()
395
if local_rh != master_rh:
396
raise errors.BoundBranchOutOfDate(self.branch,
399
# Now things are ready to change the master branch
401
self.bound_branch = self.branch
402
self.master_branch.lock_write()
403
self.master_locked = True
406
"""Cleanup any open locks, progress bars etc."""
407
cleanups = [self._cleanup_bound_branch,
408
self.work_tree.unlock,
410
found_exception = None
411
for cleanup in cleanups:
414
# we want every cleanup to run no matter what.
415
# so we have a catchall here, but we will raise the
416
# last encountered exception up the stack: and
417
# typically this will be useful enough.
420
if found_exception is not None:
421
# don't do a plan raise, because the last exception may have been
422
# trashed, e is our sure-to-work exception even though it loses the
423
# full traceback. XXX: RBC 20060421 perhaps we could check the
424
# exc_info and if its the same one do a plain raise otherwise
425
# 'raise e' as we do now.
428
def _cleanup_bound_branch(self):
429
"""Executed at the end of a try/finally to cleanup a bound branch.
431
If the branch wasn't bound, this is a no-op.
432
If it was, it resents self.branch to the local branch, instead
435
if not self.bound_branch:
437
if self.master_locked:
438
self.master_branch.unlock()
440
def _escape_commit_message(self):
441
"""Replace xml-incompatible control characters."""
442
# FIXME: RBC 20060419 this should be done by the revision
443
# serialiser not by commit. Then we can also add an unescaper
444
# in the deserializer and start roundtripping revision messages
445
# precisely. See repository_implementations/test_repository.py
447
# Python strings can include characters that can't be
448
# represented in well-formed XML; escape characters that
449
# aren't listed in the XML specification
450
# (http://www.w3.org/TR/REC-xml/#NT-Char).
451
self.message, escape_count = re.subn(
452
u'[^\x09\x0A\x0D\u0020-\uD7FF\uE000-\uFFFD]+',
453
lambda match: match.group(0).encode('unicode_escape'),
456
self.reporter.escaped(escape_count, self.message)
458
def _gather_parents(self):
459
"""Record the parents of a merge for merge detection."""
460
# TODO: Make sure that this list doesn't contain duplicate
461
# entries and the order is preserved when doing this.
462
self.parents = self.work_tree.get_parent_ids()
463
self.parent_invs = []
464
for revision in self.parents:
465
if self.branch.repository.has_revision(revision):
466
mutter('commit parent revision {%s}', revision)
467
inventory = self.branch.repository.get_inventory(revision)
468
self.parent_invs.append(inventory)
470
mutter('commit parent ghost revision {%s}', revision)
472
def _remove_deleted(self):
473
"""Remove deleted files from the working inventories.
475
This is done prior to taking the working inventory as the
476
basis for the new committed inventory.
478
This returns true if any files
479
*that existed in the basis inventory* were deleted.
480
Files that were added and deleted
481
in the working copy don't matter.
483
specific = self.specific_files
485
deleted_paths = set()
486
for path, ie in self.work_inv.iter_entries():
487
if is_inside_any(deleted_paths, path):
488
# The tree will delete the required ids recursively.
490
if specific and not is_inside_any(specific, path):
492
if not self.work_tree.has_filename(path):
493
deleted_paths.add(path)
494
self.reporter.missing(path)
495
deleted_ids.append(ie.file_id)
496
self.work_tree.unversion(deleted_ids)
498
def _populate_new_inv(self):
499
"""Build revision inventory.
501
This creates a new empty inventory. Depending on
502
which files are selected for commit, and what is present in the
503
current tree, the new inventory is populated. inventory entries
504
which are candidates for modification have their revision set to
505
None; inventory entries that are carried over untouched have their
506
revision set to their prior value.
508
# ESEPARATIONOFCONCERNS: this function is diffing and using the diff
509
# results to create a new inventory at the same time, which results
510
# in bugs like #46635. Any reason not to use/enhance Tree.changes_from?
512
mutter("Selecting files for commit with filter %s", self.specific_files)
513
entries = self.work_inv.iter_entries()
514
if not self.builder.record_root_entry:
515
symbol_versioning.warn('CommitBuilders should support recording'
516
' the root entry as of bzr 0.10.', DeprecationWarning,
518
self.builder.new_inventory.add(self.basis_inv.root.copy())
520
self._emit_progress_update()
521
for path, new_ie in entries:
522
self._emit_progress_update()
523
file_id = new_ie.file_id
524
# mutter('check %s {%s}', path, file_id)
525
if (not self.specific_files or
526
is_inside_or_parent_of_any(self.specific_files, path)):
527
# mutter('%s selected for commit', path)
531
# mutter('%s not selected for commit', path)
532
if self.basis_inv.has_id(file_id):
533
ie = self.basis_inv[file_id].copy()
535
# this entry is new and not being committed
538
self.builder.record_entry_contents(ie, self.parent_invs,
539
path, self.work_tree)
540
# describe the nature of the change that has occurred relative to
541
# the basis inventory.
542
if (self.basis_inv.has_id(ie.file_id)):
543
basis_ie = self.basis_inv[ie.file_id]
546
change = ie.describe_change(basis_ie, ie)
547
if change in (InventoryEntry.RENAMED,
548
InventoryEntry.MODIFIED_AND_RENAMED):
549
old_path = self.basis_inv.id2path(ie.file_id)
550
self.reporter.renamed(change, old_path, path)
552
self.reporter.snapshot_change(change, path)
554
if not self.specific_files:
557
# ignore removals that don't match filespec
558
for path, new_ie in self.basis_inv.iter_entries():
559
if new_ie.file_id in self.work_inv:
561
if is_inside_any(self.specific_files, path):
565
self.builder.record_entry_contents(ie, self.parent_invs, path,
568
def _emit_progress_update(self):
569
"""Emit an update to the progress bar."""
570
self.pb.update("Committing", self.pb_count, self.pb_total)
573
def _report_deletes(self):
574
for path, ie in self.basis_inv.iter_entries():
575
if ie.file_id not in self.builder.new_inventory:
576
self.reporter.deleted(path)
19
# FIXME: "bzr commit doc/format" commits doc/format.txt!
21
def commit(branch, message,
28
"""Commit working copy as a new revision.
30
The basic approach is to add all the file texts into the
31
store, then the inventory, then make a new revision pointing
32
to that inventory and store that.
34
This is not quite safe if the working copy changes during the
35
commit; for the moment that is simply not allowed. A better
36
approach is to make a temporary copy of the files before
37
computing their hashes, and then add those hashes in turn to
38
the inventory. This should mean at least that there are no
39
broken hash pointers. There is no way we can get a snapshot
40
of the whole directory at an instant. This would also have to
41
be robust against files disappearing, moving, etc. So the
42
whole thing is a bit hard.
44
timestamp -- if not None, seconds-since-epoch for a
45
postdated/predated commit.
48
If true, commit only those files.
51
If set, use this as the new revision id.
52
Useful for test or import commands that need to tightly
53
control what revisions are assigned. If you duplicate
54
a revision id that exists elsewhere it is your own fault.
55
If null (default), a time/random revision id is generated.
60
from bzrlib.osutils import local_time_offset, username
61
from bzrlib.branch import gen_file_id
62
from bzrlib.errors import BzrError
63
from bzrlib.revision import Revision, RevisionReference
64
from bzrlib.trace import mutter, note
65
from bzrlib.xml import pack_xml
70
# First walk over the working inventory; and both update that
71
# and also build a new revision inventory. The revision
72
# inventory needs to hold the text-id, sha1 and size of the
73
# actual file versions committed in the revision. (These are
74
# not present in the working inventory.) We also need to
75
# detect missing/deleted files, and remove them from the
78
work_tree = branch.working_tree()
79
work_inv = work_tree.inventory
80
basis = branch.basis_tree()
81
basis_inv = basis.inventory
84
note('looking for changes...')
86
pending_merges = branch.pending_merges()
88
missing_ids, new_inv = _gather_commit(branch,
95
for file_id in missing_ids:
96
# Any files that have been deleted are now removed from the
97
# working inventory. Files that were not selected for commit
98
# are left as they were in the working inventory and ommitted
99
# from the revision inventory.
101
# have to do this later so we don't mess up the iterator.
102
# since parents may be removed before their children we
105
# FIXME: There's probably a better way to do this; perhaps
106
# the workingtree should know how to filter itbranch.
107
if work_inv.has_id(file_id):
108
del work_inv[file_id]
112
rev_id = _gen_revision_id(time.time())
115
inv_tmp = tempfile.TemporaryFile()
116
pack_xml(new_inv, inv_tmp)
118
branch.inventory_store.add(inv_tmp, inv_id)
119
mutter('new inventory_id is {%s}' % inv_id)
121
# We could also just sha hash the inv_tmp file
122
# however, in the case that branch.inventory_store.add()
123
# ever actually does anything special
124
inv_sha1 = branch.get_inventory_sha1(inv_id)
126
branch._write_inventory(work_inv)
128
if timestamp == None:
129
timestamp = time.time()
131
if committer == None:
132
committer = username()
135
timezone = local_time_offset()
137
mutter("building commit log message")
138
rev = Revision(timestamp=timestamp,
143
inventory_sha1=inv_sha1,
147
precursor_id = branch.last_patch()
149
precursor_sha1 = branch.get_revision_sha1(precursor_id)
150
rev.parents.append(RevisionReference(precursor_id, precursor_sha1))
151
for merge_rev in pending_merges:
152
rev.parents.append(RevisionReference(merge_rev))
154
rev_tmp = tempfile.TemporaryFile()
155
pack_xml(rev, rev_tmp)
157
branch.revision_store.add(rev_tmp, rev_id)
158
mutter("new revision_id is {%s}" % rev_id)
160
## XXX: Everything up to here can simply be orphaned if we abort
161
## the commit; it will leave junk files behind but that doesn't
164
## TODO: Read back the just-generated changeset, and make sure it
165
## applies and recreates the right state.
167
## TODO: Also calculate and store the inventory SHA1
168
mutter("committing patch r%d" % (branch.revno() + 1))
170
branch.append_revision(rev_id)
172
branch.set_pending_merges([])
175
note("commited r%d" % branch.revno())
181
def _gen_revision_id(when):
182
"""Return new revision-id."""
183
from binascii import hexlify
184
from osutils import rand_bytes, compact_date, user_email
186
s = '%s-%s-' % (user_email(), compact_date(when))
187
s += hexlify(rand_bytes(8))
191
def _gather_commit(branch, work_tree, work_inv, basis_inv, specific_files,
193
"""Build inventory preparatory to commit.
195
This adds any changed files into the text store, and sets their
196
test-id, sha and size in the returned inventory appropriately.
199
Modified to hold a list of files that have been deleted from
200
the working directory; these should be removed from the
203
from bzrlib.inventory import Inventory
204
from osutils import isdir, isfile, sha_string, quotefn, \
205
local_time_offset, username, kind_marker, is_inside_any
207
from branch import gen_file_id
208
from errors import BzrError
209
from revision import Revision
210
from bzrlib.trace import mutter, note
215
for path, entry in work_inv.iter_entries():
216
## TODO: Check that the file kind has not changed from the previous
217
## revision of this file (if any).
219
p = branch.abspath(path)
220
file_id = entry.file_id
221
mutter('commit prep file %s, id %r ' % (p, file_id))
223
if specific_files and not is_inside_any(specific_files, path):
224
if basis_inv.has_id(file_id):
225
# carry over with previous state
226
inv.add(basis_inv[file_id].copy())
228
# omit this from committed inventory
232
if not work_tree.has_id(file_id):
234
print('deleted %s%s' % (path, kind_marker(entry.kind)))
235
mutter(" file is missing, removing from inventory")
236
missing_ids.append(file_id)
239
# this is present in the new inventory; may be new, modified or
241
old_ie = basis_inv.has_id(file_id) and basis_inv[file_id]
247
old_kind = old_ie.kind
248
if old_kind != entry.kind:
249
raise BzrError("entry %r changed kind from %r to %r"
250
% (file_id, old_kind, entry.kind))
252
if entry.kind == 'directory':
254
raise BzrError("%s is entered as directory but not a directory"
256
elif entry.kind == 'file':
258
raise BzrError("%s is entered as file but is not a file" % quotefn(p))
260
new_sha1 = work_tree.get_file_sha1(file_id)
263
and old_ie.text_sha1 == new_sha1):
264
## assert content == basis.get_file(file_id).read()
265
entry.text_id = old_ie.text_id
266
entry.text_sha1 = new_sha1
267
entry.text_size = old_ie.text_size
268
mutter(' unchanged from previous text_id {%s}' %
271
content = file(p, 'rb').read()
273
# calculate the sha again, just in case the file contents
274
# changed since we updated the cache
275
entry.text_sha1 = sha_string(content)
276
entry.text_size = len(content)
278
entry.text_id = gen_file_id(entry.name)
279
branch.text_store.add(content, entry.text_id)
280
mutter(' stored with text_id {%s}' % entry.text_id)
283
marked = path + kind_marker(entry.kind)
285
print 'added', marked
286
elif old_ie == entry:
288
elif (old_ie.name == entry.name
289
and old_ie.parent_id == entry.parent_id):
290
print 'modified', marked
292
print 'renamed', marked
294
return missing_ids, inv