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
def commit(branch, message,
26
allow_pointless=True):
27
"""Commit working copy as a new revision.
29
The basic approach is to add all the file texts into the
30
store, then the inventory, then make a new revision pointing
31
to that inventory and store that.
33
This is not quite safe if the working copy changes during the
34
commit; for the moment that is simply not allowed. A better
35
approach is to make a temporary copy of the files before
36
computing their hashes, and then add those hashes in turn to
37
the inventory. This should mean at least that there are no
38
broken hash pointers. There is no way we can get a snapshot
39
of the whole directory at an instant. This would also have to
40
be robust against files disappearing, moving, etc. So the
41
whole thing is a bit hard.
43
This raises PointlessCommit if there are no changes, no new merges,
44
and allow_pointless is false.
46
timestamp -- if not None, seconds-since-epoch for a
47
postdated/predated commit.
50
If true, commit only those files.
53
If set, use this as the new revision id.
54
Useful for test or import commands that need to tightly
55
control what revisions are assigned. If you duplicate
56
a revision id that exists elsewhere it is your own fault.
57
If null (default), a time/random revision id is generated.
60
import time, tempfile, re
62
from bzrlib.osutils import local_time_offset, username
63
from bzrlib.branch import gen_file_id
64
from bzrlib.errors import BzrError, PointlessCommit
65
from bzrlib.revision import Revision, RevisionReference
66
from bzrlib.trace import mutter, note
67
from bzrlib.xml import serializer_v4
72
# First walk over the working inventory; and both update that
73
# and also build a new revision inventory. The revision
74
# inventory needs to hold the text-id, sha1 and size of the
75
# actual file versions committed in the revision. (These are
76
# not present in the working inventory.) We also need to
77
# detect missing/deleted files, and remove them from the
80
work_tree = branch.working_tree()
81
work_inv = work_tree.inventory
82
basis = branch.basis_tree()
83
basis_inv = basis.inventory
86
# note('looking for changes...')
87
# print 'looking for changes...'
88
# disabled; should be done at a higher level
91
pending_merges = branch.pending_merges()
93
missing_ids, new_inv, any_changes = \
94
_gather_commit(branch,
101
if not (any_changes or allow_pointless or pending_merges):
102
raise PointlessCommit()
104
for file_id in missing_ids:
105
# Any files that have been deleted are now removed from the
106
# working inventory. Files that were not selected for commit
107
# are left as they were in the working inventory and ommitted
108
# from the revision inventory.
110
# have to do this later so we don't mess up the iterator.
111
# since parents may be removed before their children we
114
# FIXME: There's probably a better way to do this; perhaps
115
# the workingtree should know how to filter itbranch.
116
if work_inv.has_id(file_id):
117
del work_inv[file_id]
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
# This code requires all merge parents to be present in the branch.
48
# We could relax this but for the sake of simplicity the constraint is
49
# here for now. It's not totally clear to me how we'd know which file
50
# need new text versions if some parents are absent. -- mbp 20050915
52
# TODO: Rather than mashing together the ancestry and storing it back,
53
# perhaps the weave should have single method which does it all in one
54
# go, avoiding a lot of redundant work.
56
# TODO: Perhaps give a warning if one of the revisions marked as
57
# merged is already in the ancestry, and then don't record it as a
60
# TODO: If the file is newly merged but unchanged from the version it
61
# merges from, then it should still be reported as newly added
62
# relative to the basis revision.
71
from binascii import hexlify
72
from cStringIO import StringIO
74
from bzrlib.osutils import (local_time_offset, username,
75
rand_bytes, compact_date, user_email,
76
kind_marker, is_inside_any, quotefn,
77
sha_string, sha_strings, sha_file, isdir, isfile,
79
from bzrlib.branch import gen_file_id
80
from bzrlib.errors import (BzrError, PointlessCommit,
83
from bzrlib.revision import Revision
84
from bzrlib.trace import mutter, note, warning
85
from bzrlib.xml5 import serializer_v5
86
from bzrlib.inventory import Inventory
87
from bzrlib.weave import Weave
88
from bzrlib.weavefile import read_weave, write_weave_v5
89
from bzrlib.atomicfile import AtomicFile
92
def commit(*args, **kwargs):
93
"""Commit a new revision to a branch.
95
Function-style interface for convenience of old callers.
97
New code should use the Commit class instead.
99
## XXX: Remove this in favor of Branch.commit?
100
Commit().commit(*args, **kwargs)
103
class NullCommitReporter(object):
104
"""I report on progress of a commit."""
105
def added(self, path):
108
def removed(self, path):
111
def renamed(self, old_path, new_path):
115
class ReportCommitToLog(NullCommitReporter):
116
def added(self, path):
117
note('added %s', path)
119
def removed(self, path):
120
note('removed %s', path)
122
def renamed(self, old_path, new_path):
123
note('renamed %s => %s', old_path, new_path)
126
class Commit(object):
127
"""Task of committing a new revision.
129
This is a MethodObject: it accumulates state as the commit is
130
prepared, and then it is discarded. It doesn't represent
131
historical revisions, just the act of recording a new one.
134
Modified to hold a list of files that have been deleted from
135
the working directory; these should be removed from the
140
if reporter is not None:
141
self.reporter = reporter
143
self.reporter = NullCommitReporter()
153
allow_pointless=True,
155
"""Commit working copy as a new revision.
157
timestamp -- if not None, seconds-since-epoch for a
158
postdated/predated commit.
160
specific_files -- If true, commit only those files.
162
rev_id -- If set, use this as the new revision id.
163
Useful for test or import commands that need to tightly
164
control what revisions are assigned. If you duplicate
165
a revision id that exists elsewhere it is your own fault.
166
If null (default), a time/random revision id is generated.
168
allow_pointless -- If true (default), commit even if nothing
169
has changed and no merges are recorded.
171
mutter('preparing to commit')
174
self.weave_store = branch.weave_store
176
self.specific_files = specific_files
177
self.allow_pointless = allow_pointless
179
if timestamp is None:
180
self.timestamp = time.time()
182
self.timestamp = long(timestamp)
119
184
if rev_id is None:
120
rev_id = _gen_revision_id(branch, time.time())
123
inv_tmp = tempfile.TemporaryFile()
125
serializer_v4.write_inventory(new_inv, inv_tmp)
127
branch.inventory_store.add(inv_tmp, inv_id)
128
mutter('new inventory_id is {%s}' % inv_id)
130
# We could also just sha hash the inv_tmp file
131
# however, in the case that branch.inventory_store.add()
132
# ever actually does anything special
133
inv_sha1 = branch.get_inventory_sha1(inv_id)
135
branch._write_inventory(work_inv)
137
if timestamp == None:
138
timestamp = time.time()
140
if committer == None:
141
committer = username(branch)
144
timezone = local_time_offset()
146
mutter("building commit log message")
185
self.rev_id = _gen_revision_id(self.branch, self.timestamp)
189
if committer is None:
190
self.committer = username(self.branch)
192
assert isinstance(committer, basestring), type(committer)
193
self.committer = committer
196
self.timezone = local_time_offset()
198
self.timezone = int(timezone)
200
assert isinstance(message, basestring), type(message)
201
self.message = message
202
self._escape_commit_message()
204
self.branch.lock_write()
206
self.work_tree = self.branch.working_tree()
207
self.work_inv = self.work_tree.inventory
208
self.basis_tree = self.branch.basis_tree()
209
self.basis_inv = self.basis_tree.inventory
211
self._gather_parents()
212
if len(self.parents) > 1 and self.specific_files:
213
raise NotImplementedError('selected-file commit of merges is not supported yet')
214
self._check_parents_present()
216
self._remove_deleted()
217
self.new_inv = Inventory()
218
self._store_entries()
219
self._report_deletes()
220
self._set_name_versions()
222
if not (self.allow_pointless
223
or len(self.parents) > 1
224
or self.new_inv != self.basis_inv):
225
raise PointlessCommit()
227
self._record_inventory()
228
self._record_ancestry()
229
self._make_revision()
230
note('committed r%d {%s}', (self.branch.revno() + 1),
232
self.branch.append_revision(self.rev_id)
233
self.branch.set_pending_merges([])
239
def _record_inventory(self):
240
"""Store the inventory for the new revision."""
241
inv_text = serializer_v5.write_inventory_to_string(self.new_inv)
242
self.inv_sha1 = sha_string(inv_text)
243
s = self.branch.control_weaves
244
s.add_text('inventory', self.rev_id,
245
split_lines(inv_text), self.parents)
247
def _escape_commit_message(self):
248
"""Replace xml-incompatible control characters."""
147
249
# Python strings can include characters that can't be
148
250
# represented in well-formed XML; escape characters that
149
251
# aren't listed in the XML specification
150
252
# (http://www.w3.org/TR/REC-xml/#NT-Char).
151
if isinstance(message, unicode):
253
if isinstance(self.message, unicode):
152
254
char_pattern = u'[^\x09\x0A\x0D\u0020-\uD7FF\uE000-\uFFFD]'
154
256
# Use a regular 'str' as pattern to avoid having re.subn
155
257
# return 'unicode' results.
156
258
char_pattern = '[^x09\x0A\x0D\x20-\xFF]'
157
message, escape_count = re.subn(
259
self.message, escape_count = re.subn(
159
261
lambda match: match.group(0).encode('unicode_escape'),
162
264
note("replaced %d control characters in message", escape_count)
163
rev = Revision(timestamp=timestamp,
168
inventory_sha1=inv_sha1,
172
precursor_id = branch.last_patch()
266
def _record_ancestry(self):
267
"""Append merged revision ancestry to the ancestry file.
269
This should be the merged ancestry of all parents, plus the
271
s = self.branch.control_weaves
272
w = s.get_weave_or_empty('ancestry')
273
lines = self._make_ancestry(w)
274
w.add(self.rev_id, self.parents, lines)
275
s.put_weave('ancestry', w)
278
def _make_ancestry(self, ancestry_weave):
279
"""Return merged ancestry lines.
281
The lines are revision-ids followed by newlines."""
282
parent_ancestries = [ancestry_weave.get(p) for p in self.parents]
283
new_lines = merge_ancestry_lines(self.rev_id, parent_ancestries)
284
mutter('merged ancestry of {%s}:\n%s', self.rev_id, ''.join(new_lines))
288
def _gather_parents(self):
289
pending_merges = self.branch.pending_merges()
291
self.parent_trees = []
292
precursor_id = self.branch.last_revision()
174
precursor_sha1 = branch.get_revision_sha1(precursor_id)
175
rev.parents.append(RevisionReference(precursor_id, precursor_sha1))
176
for merge_rev in pending_merges:
177
rev.parents.append(RevisionReference(merge_rev))
179
rev_tmp = tempfile.TemporaryFile()
180
serializer_v4.write_revision(rev, rev_tmp)
294
self.parents.append(precursor_id)
295
self.parent_trees.append(self.basis_tree)
296
self.parents += pending_merges
297
self.parent_trees.extend(map(self.branch.revision_tree, pending_merges))
300
def _check_parents_present(self):
301
for parent_id in self.parents:
302
mutter('commit parent revision {%s}', parent_id)
303
if not self.branch.has_revision(parent_id):
304
warning("can't commit a merge from an absent parent")
305
raise HistoryMissing(self.branch, 'revision', parent_id)
308
def _make_revision(self):
309
"""Record a new revision object for this commit."""
310
self.rev = Revision(timestamp=self.timestamp,
311
timezone=self.timezone,
312
committer=self.committer,
313
message=self.message,
314
inventory_sha1=self.inv_sha1,
315
revision_id=self.rev_id)
316
self.rev.parent_ids = self.parents
318
serializer_v5.write_revision(self.rev, rev_tmp)
182
branch.revision_store.add(rev_tmp, rev_id)
183
mutter("new revision_id is {%s}" % rev_id)
185
## XXX: Everything up to here can simply be orphaned if we abort
186
## the commit; it will leave junk files behind but that doesn't
189
## TODO: Read back the just-generated changeset, and make sure it
190
## applies and recreates the right state.
192
## TODO: Also calculate and store the inventory SHA1
193
mutter("committing patch r%d" % (branch.revno() + 1))
195
branch.append_revision(rev_id)
197
branch.set_pending_merges([])
200
# disabled; should go through logging
201
# note("commited r%d" % branch.revno())
202
# print ("commited r%d" % branch.revno())
320
self.branch.revision_store.add(rev_tmp, self.rev_id, compressed=True)
321
mutter('new revision_id is {%s}', self.rev_id)
324
def _remove_deleted(self):
325
"""Remove deleted files from the working inventories.
327
This is done prior to taking the working inventory as the
328
basis for the new committed inventory.
330
This returns true if any files
331
*that existed in the basis inventory* were deleted.
332
Files that were added and deleted
333
in the working copy don't matter.
335
specific = self.specific_files
337
for path, ie in self.work_inv.iter_entries():
338
if specific and not is_inside_any(specific, path):
340
if not self.work_tree.has_filename(path):
341
note('missing %s', path)
342
deleted_ids.append(ie.file_id)
344
for file_id in deleted_ids:
345
del self.work_inv[file_id]
346
self.branch._write_inventory(self.work_inv)
349
def _find_file_parents(self, file_id):
350
"""Return the text versions and hashes for all file parents.
352
Returned as a map from text version to inventory entry.
354
This is a set containing the file versions in all parents
355
revisions containing the file. If the file is new, the set
358
for tree in self.parent_trees:
359
if file_id in tree.inventory:
360
ie = tree.inventory[file_id]
361
assert ie.kind == 'file'
362
assert ie.file_id == file_id
363
if ie.text_version in r:
364
assert r[ie.text_version] == ie
366
r[ie.text_version] = ie
370
def _set_name_versions(self):
371
"""Pass over inventory and mark new entry version as needed.
373
Files get a new name version when they are new, have a
374
different parent, or a different name from in the
375
basis inventory, or if the file is in a different place
376
to any of the parents."""
377
# XXX: Need to think more here about when the user has
378
# made a specific decision on a particular value -- c.f.
380
for path, ie in self.new_inv.iter_entries():
381
compatible_priors = set()
383
for previous_tree in self.parent_trees:
385
previous_inv = previous_tree.inventory
386
if file_id not in previous_inv:
388
previous_ie = previous_inv[file_id]
389
if ie.compatible_for_commit(previous_ie):
390
mutter("found compatible previous entry")
391
compatible_priors.add(previous_ie.name_version)
392
if len(compatible_priors) != 1:
393
mutter('new name_version for {%s}', file_id)
394
ie.name_version = self.rev_id
396
ie.name_version = compatible_priors.pop()
397
mutter('name_version for {%s} inherited as {%s}',
398
file_id, ie.name_version)
400
def _store_entries(self):
401
"""Build revision inventory and store modified files.
403
This is called with new_inv a new empty inventory. Depending on
404
which files are selected for commit, and which ones have
405
been modified or merged, new inventory entries are built
406
based on the working and parent inventories.
408
As a side-effect this stores new text versions for committed
409
files with text changes or merges.
411
Each entry can have one of several things happen:
413
carry_file -- carried from the previous version (if not
416
commit_nonfile -- no text to worry about
418
commit_old_text -- same text, may have moved
420
commit_file -- new text version
422
for path, new_ie in self.work_inv.iter_entries():
423
file_id = new_ie.file_id
424
mutter('check %s {%s}', path, new_ie.file_id)
425
if self.specific_files:
426
if not is_inside_any(self.specific_files, path):
427
mutter('%s not selected for commit', path)
428
self._carry_file(file_id)
430
if new_ie.kind == 'symlink':
431
new_ie.read_symlink_target(self.branch.abspath(path))
432
if new_ie.kind != 'file':
433
self._commit_nonfile(file_id)
436
file_parents = self._find_file_parents(file_id)
437
mutter('parents of %s are %r', path, file_parents)
438
if len(file_parents) == 1:
439
parent_ie = file_parents.values()[0]
440
wc_sha1 = self.work_tree.get_file_sha1(file_id)
441
if parent_ie.text_sha1 == wc_sha1:
442
# text not changed or merged
443
self._commit_old_text(file_id, parent_ie)
445
# file is either new, or a file merge; need to record
447
if len(file_parents) > 1:
448
note('merged %s', path)
449
elif len(file_parents) == 0:
450
note('added %s', path)
452
note('modified %s', path)
453
self._commit_file(new_ie, file_id, file_parents)
456
def _commit_nonfile(self, file_id):
457
self.new_inv.add(self.work_inv[file_id].copy())
460
def _carry_file(self, file_id):
461
"""Carry the file unchanged from the basis revision."""
462
if self.basis_inv.has_id(file_id):
463
self.new_inv.add(self.basis_inv[file_id].copy())
466
def _commit_old_text(self, file_id, parent_ie):
467
"""Keep the same text as last time, but possibly a different name."""
468
ie = self.work_inv[file_id].copy()
469
ie.text_version = parent_ie.text_version
470
ie.text_size = parent_ie.text_size
471
ie.text_sha1 = parent_ie.text_sha1
475
def _report_deletes(self):
476
for file_id in self.basis_inv:
477
if file_id not in self.new_inv:
478
note('deleted %s', self.basis_inv.id2path(file_id))
481
def _commit_file(self, new_ie, file_id, file_parents):
482
mutter('store new text for {%s} in revision {%s}',
483
file_id, self.rev_id)
484
new_lines = self.work_tree.get_file(file_id).readlines()
485
self._add_text_to_weave(file_id, new_lines, file_parents)
486
new_ie.text_version = self.rev_id
487
new_ie.text_sha1 = sha_strings(new_lines)
488
new_ie.text_size = sum(map(len, new_lines))
489
self.new_inv.add(new_ie)
492
def _add_text_to_weave(self, file_id, new_lines, parents):
493
self.weave_store.add_text(file_id, self.rev_id, new_lines, parents)
209
496
def _gen_revision_id(branch, when):
210
497
"""Return new revision-id."""
211
from binascii import hexlify
212
from bzrlib.osutils import rand_bytes, compact_date, user_email
214
498
s = '%s-%s-' % (user_email(branch), compact_date(when))
215
499
s += hexlify(rand_bytes(8))
219
def _gather_commit(branch, work_tree, work_inv, basis_inv, specific_files,
221
"""Build inventory preparatory to commit.
223
Returns missing_ids, new_inv, any_changes.
225
This adds any changed files into the text store, and sets their
226
test-id, sha and size in the returned inventory appropriately.
229
Modified to hold a list of files that have been deleted from
230
the working directory; these should be removed from the
505
def merge_ancestry_lines(rev_id, ancestries):
506
"""Return merged ancestry lines.
508
rev_id -- id of the new revision
510
ancestries -- a sequence of ancestries for parent revisions,
511
as newline-terminated line lists.
233
from bzrlib.inventory import Inventory
234
from bzrlib.osutils import isdir, isfile, islink, sha_string, quotefn, \
235
local_time_offset, username, kind_marker, is_inside_any
237
from bzrlib.branch import gen_file_id
238
from bzrlib.errors import BzrError
239
from bzrlib.revision import Revision
240
from bzrlib.trace import mutter, note
243
inv = Inventory(work_inv.root.file_id)
246
for path, entry in work_inv.iter_entries():
247
## TODO: Check that the file kind has not changed from the previous
248
## revision of this file (if any).
250
abspath = branch.abspath(path)
251
file_id = entry.file_id
252
mutter('commit prep file %s, id %r ' % (abspath, file_id))
254
if specific_files and not is_inside_any(specific_files, path):
255
mutter(' skipping file excluded from commit')
256
if basis_inv.has_id(file_id):
257
# carry over with previous state
258
inv.add(basis_inv[file_id].copy())
260
# omit this from committed inventory
264
if not work_tree.has_id(file_id):
266
print('deleted %s%s' % (path, kind_marker(entry.kind)))
268
mutter(" file is missing, removing from inventory")
269
missing_ids.append(file_id)
272
# this is present in the new inventory; may be new, modified or
274
old_ie = basis_inv.has_id(file_id) and basis_inv[file_id]
280
old_kind = old_ie.kind
281
if old_kind != entry.kind:
282
raise BzrError("entry %r changed kind from %r to %r - not "
284
% (file_id, old_kind, entry.kind))
286
if entry.kind == 'directory':
287
if not isdir(abspath):
288
raise BzrError("%s is entered as directory but not a directory"
290
elif entry.kind == 'file':
291
if not isfile(abspath):
292
raise BzrError("%s is entered as file but is not a file" % quotefn(abspath))
294
new_sha1 = work_tree.get_file_sha1(file_id)
297
and old_ie.text_sha1 == new_sha1):
298
## assert content == basis.get_file(file_id).read()
299
entry.text_id = old_ie.text_id
300
entry.text_sha1 = new_sha1
301
entry.text_size = old_ie.text_size
302
mutter(' unchanged from previous text_id {%s}' %
305
content = file(abspath, 'rb').read()
307
# calculate the sha again, just in case the file contents
308
# changed since we updated the cache
309
entry.text_sha1 = sha_string(content)
310
entry.text_size = len(content)
312
entry.text_id = gen_file_id(entry.name)
313
branch.text_store.add(content, entry.text_id)
314
mutter(' stored with text_id {%s}' % entry.text_id)
315
elif entry.kind == 'symlink':
316
if not islink(abspath):
317
raise BzrError("%s is entered as link but is not a link"
320
entry.read_symlink_target(abspath)
323
and old_ie.symlink_target == entry.symlink_target):
324
mutter(' unchanged from previous symlink {%s}' %
325
entry.symlink_target)
327
mutter(' changed to value {%s}' % entry.symlink_target)
330
marked = path + kind_marker(entry.kind)
332
print 'added', marked
334
elif old_ie == entry:
336
elif (old_ie.name == entry.name
337
and old_ie.parent_id == entry.parent_id):
338
print 'modified', marked
341
print 'renamed', marked
343
elif old_ie != entry:
346
return missing_ids, inv, any_changes
513
if len(ancestries) == 0:
514
return [rev_id + '\n']
515
seen = set(ancestries[0])
516
ancs = ancestries[0][:]
517
for parent_ancestry in ancestries[1:]:
518
for line in parent_ancestry:
519
assert line[-1] == '\n'