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.symbol_versioning import *
94
from bzrlib.weave import Weave
95
from bzrlib.weavefile import read_weave, write_weave_v5
96
from bzrlib.workingtree import WorkingTree
99
@deprecated_function(zero_seven)
100
def commit(*args, **kwargs):
101
"""Commit a new revision to a branch.
103
Function-style interface for convenience of old callers.
105
New code should use the Commit class instead.
107
## XXX: Remove this in favor of Branch.commit?
108
Commit().commit(*args, **kwargs)
111
class NullCommitReporter(object):
112
"""I report on progress of a commit."""
114
def snapshot_change(self, change, path):
117
def completed(self, revno, rev_id):
120
def deleted(self, file_id):
123
def escaped(self, escape_count, message):
126
def missing(self, path):
130
class ReportCommitToLog(NullCommitReporter):
132
def snapshot_change(self, change, path):
133
note("%s %s", change, path)
135
def completed(self, revno, rev_id):
136
note('committed r%d {%s}', revno, rev_id)
138
def deleted(self, file_id):
139
note('deleted %s', file_id)
141
def escaped(self, escape_count, message):
142
note("replaced %d control characters in message", escape_count)
144
def missing(self, path):
145
note('missing %s', path)
148
class Commit(object):
149
"""Task of committing a new revision.
151
This is a MethodObject: it accumulates state as the commit is
152
prepared, and then it is discarded. It doesn't represent
153
historical revisions, just the act of recording a new one.
156
Modified to hold a list of files that have been deleted from
157
the working directory; these should be removed from the
163
if reporter is not None:
164
self.reporter = reporter
166
self.reporter = NullCommitReporter()
167
if config is not None:
173
branch=DEPRECATED_PARAMETER, message=None,
179
allow_pointless=True,
184
"""Commit working copy as a new revision.
186
branch -- the deprecated branch to commit to. New callers should pass in
189
message -- the commit message, a mandatory parameter
191
timestamp -- if not None, seconds-since-epoch for a
192
postdated/predated commit.
194
specific_files -- If true, commit only those files.
196
rev_id -- If set, use this as the new revision id.
197
Useful for test or import commands that need to tightly
198
control what revisions are assigned. If you duplicate
199
a revision id that exists elsewhere it is your own fault.
200
If null (default), a time/random revision id is generated.
202
allow_pointless -- If true (default), commit even if nothing
203
has changed and no merges are recorded.
205
strict -- If true, don't allow a commit if the working tree
206
contains unknown files.
208
revprops -- Properties for new revision
210
mutter('preparing to commit')
212
if deprecated_passed(branch):
213
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 = WorkingTree(branch.base, branch)
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.weave_store = self.branch.repository.weave_store
228
self.specific_files = specific_files
229
self.allow_pointless = allow_pointless
230
self.revprops = {'branch-nick': self.branch.nick}
232
self.revprops.update(revprops)
235
# raise an exception as soon as we find a single unknown.
236
for unknown in self.work_tree.unknowns():
237
raise StrictCommitFailed()
239
if timestamp is None:
240
self.timestamp = time.time()
242
self.timestamp = long(timestamp)
244
if self.config is None:
245
self.config = bzrlib.config.BranchConfig(self.branch)
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]
247
119
if rev_id is None:
248
self.rev_id = _gen_revision_id(self.config, self.timestamp)
252
if committer is None:
253
self.committer = self.config.username()
255
assert isinstance(committer, basestring), type(committer)
256
self.committer = committer
259
self.timezone = local_time_offset()
261
self.timezone = int(timezone)
263
if isinstance(message, str):
264
message = message.decode(bzrlib.user_encoding)
265
assert isinstance(message, unicode), type(message)
266
self.message = message
267
self._escape_commit_message()
269
self.branch.lock_write()
271
self.work_inv = self.work_tree.inventory
272
self.basis_tree = self.work_tree.basis_tree()
273
self.basis_inv = self.basis_tree.inventory
275
self._gather_parents()
276
if len(self.parents) > 1 and self.specific_files:
277
raise NotImplementedError('selected-file commit of merges is not supported yet')
278
self._check_parents_present()
280
self._remove_deleted()
281
self._populate_new_inv()
282
self._store_snapshot()
283
self._report_deletes()
285
if not (self.allow_pointless
286
or len(self.parents) > 1
287
or self.new_inv != self.basis_inv):
288
raise PointlessCommit()
290
if len(list(self.work_tree.iter_conflicts()))>0:
291
raise ConflictsInTree
293
self._record_inventory()
294
self._make_revision()
295
self.work_tree.set_pending_merges([])
296
self.branch.append_revision(self.rev_id)
297
if len(self.parents):
298
precursor = self.parents[0]
301
self.work_tree.set_last_revision(self.rev_id, precursor)
302
self.reporter.completed(self.branch.revno()+1, self.rev_id)
303
if self.config.post_commit() is not None:
304
hooks = self.config.post_commit().split(' ')
305
# this would be nicer with twisted.python.reflect.namedAny
307
result = eval(hook + '(branch, rev_id)',
308
{'branch':self.branch,
310
'rev_id':self.rev_id})
314
def _record_inventory(self):
315
"""Store the inventory for the new revision."""
316
inv_text = serializer_v5.write_inventory_to_string(self.new_inv)
317
self.inv_sha1 = sha_string(inv_text)
318
s = self.branch.repository.control_weaves
319
s.add_text('inventory', self.rev_id,
320
split_lines(inv_text), self.present_parents,
321
self.branch.get_transaction())
323
def _escape_commit_message(self):
324
"""Replace xml-incompatible control characters."""
325
# Python strings can include characters that can't be
326
# represented in well-formed XML; escape characters that
327
# aren't listed in the XML specification
328
# (http://www.w3.org/TR/REC-xml/#NT-Char).
329
self.message, escape_count = re.subn(
330
u'[^\x09\x0A\x0D\u0020-\uD7FF\uE000-\uFFFD]+',
331
lambda match: match.group(0).encode('unicode_escape'),
334
self.reporter.escaped(escape_count, self.message)
336
def _gather_parents(self):
337
"""Record the parents of a merge for merge detection."""
338
pending_merges = self.work_tree.pending_merges()
340
self.parent_invs = []
341
self.present_parents = []
342
precursor_id = self.branch.last_revision()
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()
344
self.parents.append(precursor_id)
345
self.parents += pending_merges
346
for revision in self.parents:
347
if self.branch.repository.has_revision(revision):
348
inventory = self.branch.repository.get_inventory(revision)
349
self.parent_invs.append(inventory)
350
self.present_parents.append(revision)
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))
352
def _check_parents_present(self):
353
for parent_id in self.parents:
354
mutter('commit parent revision {%s}', parent_id)
355
if not self.branch.repository.has_revision(parent_id):
356
if parent_id == self.branch.last_revision():
357
warning("parent is missing %r", parent_id)
358
raise HistoryMissing(self.branch, 'revision', parent_id)
360
mutter("commit will ghost revision %r", parent_id)
362
def _make_revision(self):
363
"""Record a new revision object for this commit."""
364
self.rev = Revision(timestamp=self.timestamp,
365
timezone=self.timezone,
366
committer=self.committer,
367
message=self.message,
368
inventory_sha1=self.inv_sha1,
369
revision_id=self.rev_id,
370
properties=self.revprops)
371
self.rev.parent_ids = self.parents
373
serializer_v5.write_revision(self.rev, rev_tmp)
162
rev_tmp = tempfile.TemporaryFile()
163
pack_xml(rev, rev_tmp)
375
if self.config.signature_needed():
376
plaintext = Testament(self.rev, self.new_inv).as_short_text()
377
self.branch.repository.store_revision_signature(
378
gpg.GPGStrategy(self.config), plaintext, self.rev_id)
379
self.branch.repository.revision_store.add(rev_tmp, self.rev_id)
380
mutter('new revision_id is {%s}', self.rev_id)
382
def _remove_deleted(self):
383
"""Remove deleted files from the working inventories.
385
This is done prior to taking the working inventory as the
386
basis for the new committed inventory.
388
This returns true if any files
389
*that existed in the basis inventory* were deleted.
390
Files that were added and deleted
391
in the working copy don't matter.
393
specific = self.specific_files
395
for path, ie in self.work_inv.iter_entries():
396
if specific and not is_inside_any(specific, path):
398
if not self.work_tree.has_filename(path):
399
self.reporter.missing(path)
400
deleted_ids.append((path, ie.file_id))
402
deleted_ids.sort(reverse=True)
403
for path, file_id in deleted_ids:
404
del self.work_inv[file_id]
405
self.work_tree._write_inventory(self.work_inv)
407
def _store_snapshot(self):
408
"""Pass over inventory and record a snapshot.
410
Entries get a new revision when they are modified in
411
any way, which includes a merge with a new set of
412
parents that have the same entry.
414
# XXX: Need to think more here about when the user has
415
# made a specific decision on a particular value -- c.f.
417
for path, ie in self.new_inv.iter_entries():
418
previous_entries = ie.find_previous_heads(
420
self.weave_store.get_weave_prelude_or_empty(ie.file_id,
421
self.branch.get_transaction()))
422
if ie.revision is None:
423
change = ie.snapshot(self.rev_id, path, previous_entries,
424
self.work_tree, self.weave_store,
425
self.branch.get_transaction())
428
self.reporter.snapshot_change(change, path)
430
def _populate_new_inv(self):
431
"""Build revision inventory.
433
This creates a new empty inventory. Depending on
434
which files are selected for commit, and what is present in the
435
current tree, the new inventory is populated. inventory entries
436
which are candidates for modification have their revision set to
437
None; inventory entries that are carried over untouched have their
438
revision set to their prior value.
440
mutter("Selecting files for commit with filter %s", self.specific_files)
441
self.new_inv = Inventory()
442
for path, new_ie in self.work_inv.iter_entries():
443
file_id = new_ie.file_id
444
mutter('check %s {%s}', path, new_ie.file_id)
445
if self.specific_files:
446
if not is_inside_any(self.specific_files, path):
447
mutter('%s not selected for commit', path)
448
self._carry_entry(file_id)
451
# this is selected, ensure its parents are too.
452
parent_id = new_ie.parent_id
453
while parent_id != ROOT_ID:
454
if not self.new_inv.has_id(parent_id):
455
ie = self._select_entry(self.work_inv[parent_id])
456
mutter('%s selected for commit because of %s',
457
self.new_inv.id2path(parent_id), path)
459
ie = self.new_inv[parent_id]
460
if ie.revision is not None:
462
mutter('%s selected for commit because of %s',
463
self.new_inv.id2path(parent_id), path)
464
parent_id = ie.parent_id
465
mutter('%s selected for commit', path)
466
self._select_entry(new_ie)
468
def _select_entry(self, new_ie):
469
"""Make new_ie be considered for committing."""
475
def _carry_entry(self, file_id):
476
"""Carry the file unchanged from the basis revision."""
477
if self.basis_inv.has_id(file_id):
478
self.new_inv.add(self.basis_inv[file_id].copy())
480
def _report_deletes(self):
481
for file_id in self.basis_inv:
482
if file_id not in self.new_inv:
483
self.reporter.deleted(self.basis_inv.id2path(file_id))
485
def _gen_revision_id(config, when):
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):
486
190
"""Return new revision-id."""
487
s = '%s-%s-' % (config.user_email(), compact_date(when))
191
from binascii import hexlify
192
from osutils import rand_bytes, compact_date, user_email
194
s = '%s-%s-' % (user_email(), compact_date(when))
488
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
added
removed
bzrlib/commit.py
