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
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]
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.
66
from binascii import hexlify
67
from cStringIO import StringIO
69
from bzrlib.osutils import (local_time_offset,
70
rand_bytes, compact_date,
71
kind_marker, is_inside_any, quotefn,
72
sha_string, sha_strings, sha_file, isdir, isfile,
74
from bzrlib.branch import gen_file_id
76
from bzrlib.errors import (BzrError, PointlessCommit,
80
import bzrlib.gpg as gpg
81
from bzrlib.revision import Revision
82
from bzrlib.testament import Testament
83
from bzrlib.trace import mutter, note, warning
84
from bzrlib.xml5 import serializer_v5
85
from bzrlib.inventory import Inventory, ROOT_ID
86
from bzrlib.weave import Weave
87
from bzrlib.weavefile import read_weave, write_weave_v5
88
from bzrlib.atomicfile import AtomicFile
91
def commit(*args, **kwargs):
92
"""Commit a new revision to a branch.
94
Function-style interface for convenience of old callers.
96
New code should use the Commit class instead.
98
## XXX: Remove this in favor of Branch.commit?
99
Commit().commit(*args, **kwargs)
102
class NullCommitReporter(object):
103
"""I report on progress of a commit."""
105
def snapshot_change(self, change, path):
108
def completed(self, revno, rev_id):
111
def deleted(self, file_id):
114
def escaped(self, escape_count, message):
117
def missing(self, path):
120
class ReportCommitToLog(NullCommitReporter):
122
def snapshot_change(self, change, path):
123
note("%s %s", change, path)
125
def completed(self, revno, rev_id):
126
note('committed r%d {%s}', revno, rev_id)
128
def deleted(self, file_id):
129
note('deleted %s', file_id)
131
def escaped(self, escape_count, message):
132
note("replaced %d control characters in message", escape_count)
134
def missing(self, path):
135
note('missing %s', path)
137
class Commit(object):
138
"""Task of committing a new revision.
140
This is a MethodObject: it accumulates state as the commit is
141
prepared, and then it is discarded. It doesn't represent
142
historical revisions, just the act of recording a new one.
145
Modified to hold a list of files that have been deleted from
146
the working directory; these should be removed from the
152
if reporter is not None:
153
self.reporter = reporter
155
self.reporter = NullCommitReporter()
156
if config is not None:
168
allow_pointless=True,
171
"""Commit working copy as a new revision.
173
timestamp -- if not None, seconds-since-epoch for a
174
postdated/predated commit.
176
specific_files -- If true, commit only those files.
178
rev_id -- If set, use this as the new revision id.
179
Useful for test or import commands that need to tightly
180
control what revisions are assigned. If you duplicate
181
a revision id that exists elsewhere it is your own fault.
182
If null (default), a time/random revision id is generated.
184
allow_pointless -- If true (default), commit even if nothing
185
has changed and no merges are recorded.
187
revprops -- Properties for new revision
189
mutter('preparing to commit')
192
self.weave_store = branch.weave_store
194
self.specific_files = specific_files
195
self.allow_pointless = allow_pointless
196
self.revprops = revprops
198
if timestamp is None:
199
self.timestamp = time.time()
201
self.timestamp = long(timestamp)
203
if self.config is None:
204
self.config = bzrlib.config.BranchConfig(self.branch)
119
206
if rev_id is None:
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()
207
self.rev_id = _gen_revision_id(self.config, self.timestamp)
211
if committer is None:
212
self.committer = self.config.username()
214
assert isinstance(committer, basestring), type(committer)
215
self.committer = committer
218
self.timezone = local_time_offset()
220
self.timezone = int(timezone)
222
assert isinstance(message, basestring), type(message)
223
self.message = message
224
self._escape_commit_message()
226
self.branch.lock_write()
228
self.work_tree = self.branch.working_tree()
229
self.work_inv = self.work_tree.inventory
230
self.basis_tree = self.branch.basis_tree()
231
self.basis_inv = self.basis_tree.inventory
233
self._gather_parents()
234
if len(self.parents) > 1 and self.specific_files:
235
raise NotImplementedError('selected-file commit of merges is not supported yet')
236
self._check_parents_present()
238
self._remove_deleted()
239
self._populate_new_inv()
240
self._store_snapshot()
241
self._report_deletes()
243
if not (self.allow_pointless
244
or len(self.parents) > 1
245
or self.new_inv != self.basis_inv):
246
raise PointlessCommit()
248
if len(list(self.work_tree.iter_conflicts()))>0:
249
raise ConflictsInTree
251
self._record_inventory()
252
self._make_revision()
253
self.reporter.completed(self.branch.revno()+1, self.rev_id)
254
self.branch.append_revision(self.rev_id)
255
self.branch.set_pending_merges([])
259
def _record_inventory(self):
260
"""Store the inventory for the new revision."""
261
inv_text = serializer_v5.write_inventory_to_string(self.new_inv)
262
self.inv_sha1 = sha_string(inv_text)
263
s = self.branch.control_weaves
264
s.add_text('inventory', self.rev_id,
265
split_lines(inv_text), self.present_parents,
266
self.branch.get_transaction())
268
def _escape_commit_message(self):
269
"""Replace xml-incompatible control characters."""
270
# Python strings can include characters that can't be
271
# represented in well-formed XML; escape characters that
272
# aren't listed in the XML specification
273
# (http://www.w3.org/TR/REC-xml/#NT-Char).
274
if isinstance(self.message, unicode):
275
char_pattern = u'[^\x09\x0A\x0D\u0020-\uD7FF\uE000-\uFFFD]'
277
# Use a regular 'str' as pattern to avoid having re.subn
278
# return 'unicode' results.
279
char_pattern = '[^x09\x0A\x0D\x20-\xFF]'
280
self.message, escape_count = re.subn(
282
lambda match: match.group(0).encode('unicode_escape'),
285
self.reporter.escaped(escape_count, self.message)
287
def _gather_parents(self):
288
"""Record the parents of a merge for merge detection."""
289
pending_merges = self.branch.pending_merges()
291
self.parent_invs = []
292
self.present_parents = []
293
precursor_id = self.branch.last_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))
295
self.parents.append(precursor_id)
296
self.parents += pending_merges
297
for revision in self.parents:
298
if self.branch.has_revision(revision):
299
self.parent_invs.append(self.branch.get_inventory(revision))
300
self.present_parents.append(revision)
162
rev_tmp = tempfile.TemporaryFile()
163
pack_xml(rev, rev_tmp)
302
def _check_parents_present(self):
303
for parent_id in self.parents:
304
mutter('commit parent revision {%s}', parent_id)
305
if not self.branch.has_revision(parent_id):
306
if parent_id == self.branch.last_revision():
307
warning("parent is missing %r", parent_id)
308
raise HistoryMissing(self.branch, 'revision', parent_id)
310
mutter("commit will ghost revision %r", parent_id)
312
def _make_revision(self):
313
"""Record a new revision object for this commit."""
314
self.rev = Revision(timestamp=self.timestamp,
315
timezone=self.timezone,
316
committer=self.committer,
317
message=self.message,
318
inventory_sha1=self.inv_sha1,
319
revision_id=self.rev_id,
320
properties=self.revprops)
321
self.rev.parent_ids = self.parents
323
serializer_v5.write_revision(self.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):
325
if self.config.signature_needed():
326
plaintext = Testament(self.rev, self.new_inv).as_short_text()
327
self.branch.store_revision_signature(gpg.GPGStrategy(self.config),
328
plaintext, self.rev_id)
329
self.branch.revision_store.add(rev_tmp, self.rev_id)
330
mutter('new revision_id is {%s}', self.rev_id)
332
def _remove_deleted(self):
333
"""Remove deleted files from the working inventories.
335
This is done prior to taking the working inventory as the
336
basis for the new committed inventory.
338
This returns true if any files
339
*that existed in the basis inventory* were deleted.
340
Files that were added and deleted
341
in the working copy don't matter.
343
specific = self.specific_files
345
for path, ie in self.work_inv.iter_entries():
346
if specific and not is_inside_any(specific, path):
348
if not self.work_tree.has_filename(path):
349
self.reporter.missing(path)
350
deleted_ids.append((path, ie.file_id))
352
deleted_ids.sort(reverse=True)
353
for path, file_id in deleted_ids:
354
del self.work_inv[file_id]
355
self.branch._write_inventory(self.work_inv)
357
def _store_snapshot(self):
358
"""Pass over inventory and record a snapshot.
360
Entries get a new revision when they are modified in
361
any way, which includes a merge with a new set of
362
parents that have the same entry.
364
# XXX: Need to think more here about when the user has
365
# made a specific decision on a particular value -- c.f.
367
for path, ie in self.new_inv.iter_entries():
368
previous_entries = ie.find_previous_heads(
370
self.weave_store.get_weave_or_empty(ie.file_id,
371
self.branch.get_transaction()))
372
if ie.revision is None:
373
change = ie.snapshot(self.rev_id, path, previous_entries,
374
self.work_tree, self.weave_store,
375
self.branch.get_transaction())
378
self.reporter.snapshot_change(change, path)
380
def _populate_new_inv(self):
381
"""Build revision inventory.
383
This creates a new empty inventory. Depending on
384
which files are selected for commit, and what is present in the
385
current tree, the new inventory is populated. inventory entries
386
which are candidates for modification have their revision set to
387
None; inventory entries that are carried over untouched have their
388
revision set to their prior value.
390
mutter("Selecting files for commit with filter %s", self.specific_files)
391
self.new_inv = Inventory()
392
for path, new_ie in self.work_inv.iter_entries():
393
file_id = new_ie.file_id
394
mutter('check %s {%s}', path, new_ie.file_id)
395
if self.specific_files:
396
if not is_inside_any(self.specific_files, path):
397
mutter('%s not selected for commit', path)
398
self._carry_entry(file_id)
401
# this is selected, ensure its parents are too.
402
parent_id = new_ie.parent_id
403
while parent_id != ROOT_ID:
404
if not self.new_inv.has_id(parent_id):
405
ie = self._select_entry(self.work_inv[parent_id])
406
mutter('%s selected for commit because of %s',
407
self.new_inv.id2path(parent_id), path)
409
ie = self.new_inv[parent_id]
410
if ie.revision is not None:
412
mutter('%s selected for commit because of %s',
413
self.new_inv.id2path(parent_id), path)
414
parent_id = ie.parent_id
415
mutter('%s selected for commit', path)
416
self._select_entry(new_ie)
418
def _select_entry(self, new_ie):
419
"""Make new_ie be considered for committing."""
425
def _carry_entry(self, file_id):
426
"""Carry the file unchanged from the basis revision."""
427
if self.basis_inv.has_id(file_id):
428
self.new_inv.add(self.basis_inv[file_id].copy())
430
def _report_deletes(self):
431
for file_id in self.basis_inv:
432
if file_id not in self.new_inv:
433
self.reporter.deleted(self.basis_inv.id2path(file_id))
435
def _gen_revision_id(config, when):
190
436
"""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))
437
s = '%s-%s-' % (config.user_email(), compact_date(when))
195
438
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