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
from bzrlib.revision import Revision
81
from bzrlib.trace import mutter, note, warning
82
from bzrlib.xml5 import serializer_v5
83
from bzrlib.inventory import Inventory, ROOT_ID
84
from bzrlib.weave import Weave
85
from bzrlib.weavefile import read_weave, write_weave_v5
86
from bzrlib.atomicfile import AtomicFile
89
def commit(*args, **kwargs):
90
"""Commit a new revision to a branch.
92
Function-style interface for convenience of old callers.
94
New code should use the Commit class instead.
96
## XXX: Remove this in favor of Branch.commit?
97
Commit().commit(*args, **kwargs)
100
class NullCommitReporter(object):
101
"""I report on progress of a commit."""
103
def snapshot_change(self, change, path):
106
def completed(self, revno, rev_id):
109
def deleted(self, file_id):
112
def escaped(self, escape_count, message):
115
def missing(self, path):
118
class ReportCommitToLog(NullCommitReporter):
120
def snapshot_change(self, change, path):
121
note("%s %s", change, path)
123
def completed(self, revno, rev_id):
124
note('committed r%d {%s}', revno, rev_id)
126
def deleted(self, file_id):
127
note('deleted %s', file_id)
129
def escaped(self, escape_count, message):
130
note("replaced %d control characters in message", escape_count)
132
def missing(self, path):
133
note('missing %s', path)
135
class Commit(object):
136
"""Task of committing a new revision.
138
This is a MethodObject: it accumulates state as the commit is
139
prepared, and then it is discarded. It doesn't represent
140
historical revisions, just the act of recording a new one.
143
Modified to hold a list of files that have been deleted from
144
the working directory; these should be removed from the
149
if reporter is not None:
150
self.reporter = reporter
152
self.reporter = NullCommitReporter()
162
allow_pointless=True,
164
"""Commit working copy as a new revision.
166
timestamp -- if not None, seconds-since-epoch for a
167
postdated/predated commit.
169
specific_files -- If true, commit only those files.
171
rev_id -- If set, use this as the new revision id.
172
Useful for test or import commands that need to tightly
173
control what revisions are assigned. If you duplicate
174
a revision id that exists elsewhere it is your own fault.
175
If null (default), a time/random revision id is generated.
177
allow_pointless -- If true (default), commit even if nothing
178
has changed and no merges are recorded.
180
mutter('preparing to commit')
183
self.weave_store = branch.weave_store
185
self.specific_files = specific_files
186
self.allow_pointless = allow_pointless
188
if timestamp is None:
189
self.timestamp = time.time()
191
self.timestamp = long(timestamp)
193
config = bzrlib.config.BranchConfig(self.branch)
118
194
if rev_id is None:
119
rev_id = _gen_revision_id(branch, time.time())
122
inv_tmp = tempfile.TemporaryFile()
123
pack_xml(new_inv, inv_tmp)
125
branch.inventory_store.add(inv_tmp, inv_id)
126
mutter('new inventory_id is {%s}' % inv_id)
128
# We could also just sha hash the inv_tmp file
129
# however, in the case that branch.inventory_store.add()
130
# ever actually does anything special
131
inv_sha1 = branch.get_inventory_sha1(inv_id)
133
branch._write_inventory(work_inv)
135
if timestamp == None:
136
timestamp = time.time()
138
if committer == None:
139
committer = username(branch)
142
timezone = local_time_offset()
144
mutter("building commit log message")
145
rev = Revision(timestamp=timestamp,
150
inventory_sha1=inv_sha1,
154
precursor_id = branch.last_patch()
195
self.rev_id = _gen_revision_id(config, self.timestamp)
199
if committer is None:
200
self.committer = config.username()
202
assert isinstance(committer, basestring), type(committer)
203
self.committer = committer
206
self.timezone = local_time_offset()
208
self.timezone = int(timezone)
210
assert isinstance(message, basestring), type(message)
211
self.message = message
212
self._escape_commit_message()
214
self.branch.lock_write()
216
self.work_tree = self.branch.working_tree()
217
self.work_inv = self.work_tree.inventory
218
self.basis_tree = self.branch.basis_tree()
219
self.basis_inv = self.basis_tree.inventory
221
self._gather_parents()
222
if len(self.parents) > 1 and self.specific_files:
223
raise NotImplementedError('selected-file commit of merges is not supported yet')
224
self._check_parents_present()
226
self._remove_deleted()
227
self._populate_new_inv()
228
self._store_snapshot()
229
self._report_deletes()
231
if not (self.allow_pointless
232
or len(self.parents) > 1
233
or self.new_inv != self.basis_inv):
234
raise PointlessCommit()
236
if len(list(self.work_tree.iter_conflicts()))>0:
237
raise ConflictsInTree
239
self._record_inventory()
240
self._make_revision()
241
self.reporter.completed(self.branch.revno()+1, self.rev_id)
242
self.branch.append_revision(self.rev_id)
243
self.branch.set_pending_merges([])
247
def _record_inventory(self):
248
"""Store the inventory for the new revision."""
249
inv_text = serializer_v5.write_inventory_to_string(self.new_inv)
250
self.inv_sha1 = sha_string(inv_text)
251
s = self.branch.control_weaves
252
s.add_text('inventory', self.rev_id,
253
split_lines(inv_text), self.present_parents,
254
self.branch.get_transaction())
256
def _escape_commit_message(self):
257
"""Replace xml-incompatible control characters."""
258
# Python strings can include characters that can't be
259
# represented in well-formed XML; escape characters that
260
# aren't listed in the XML specification
261
# (http://www.w3.org/TR/REC-xml/#NT-Char).
262
if isinstance(self.message, unicode):
263
char_pattern = u'[^\x09\x0A\x0D\u0020-\uD7FF\uE000-\uFFFD]'
265
# Use a regular 'str' as pattern to avoid having re.subn
266
# return 'unicode' results.
267
char_pattern = '[^x09\x0A\x0D\x20-\xFF]'
268
self.message, escape_count = re.subn(
270
lambda match: match.group(0).encode('unicode_escape'),
273
self.reporter.escaped(escape_count, self.message)
275
def _gather_parents(self):
276
"""Record the parents of a merge for merge detection."""
277
pending_merges = self.branch.pending_merges()
279
self.parent_invs = []
280
self.present_parents = []
281
precursor_id = self.branch.last_revision()
156
precursor_sha1 = branch.get_revision_sha1(precursor_id)
157
rev.parents.append(RevisionReference(precursor_id, precursor_sha1))
158
for merge_rev in pending_merges:
159
rev.parents.append(RevisionReference(merge_rev))
283
self.parents.append(precursor_id)
284
self.parents += pending_merges
285
for revision in self.parents:
286
if self.branch.has_revision(revision):
287
self.parent_invs.append(self.branch.get_inventory(revision))
288
self.present_parents.append(revision)
161
rev_tmp = tempfile.TemporaryFile()
162
pack_xml(rev, rev_tmp)
290
def _check_parents_present(self):
291
for parent_id in self.parents:
292
mutter('commit parent revision {%s}', parent_id)
293
if not self.branch.has_revision(parent_id):
294
if parent_id == self.branch.last_revision():
295
warning("parent is missing %r", parent_id)
296
raise HistoryMissing(self.branch, 'revision', parent_id)
298
mutter("commit will ghost revision %r", parent_id)
300
def _make_revision(self):
301
"""Record a new revision object for this commit."""
302
self.rev = Revision(timestamp=self.timestamp,
303
timezone=self.timezone,
304
committer=self.committer,
305
message=self.message,
306
inventory_sha1=self.inv_sha1,
307
revision_id=self.rev_id)
308
self.rev.parent_ids = self.parents
310
serializer_v5.write_revision(self.rev, rev_tmp)
164
branch.revision_store.add(rev_tmp, rev_id)
165
mutter("new revision_id is {%s}" % rev_id)
167
## XXX: Everything up to here can simply be orphaned if we abort
168
## the commit; it will leave junk files behind but that doesn't
171
## TODO: Read back the just-generated changeset, and make sure it
172
## applies and recreates the right state.
174
## TODO: Also calculate and store the inventory SHA1
175
mutter("committing patch r%d" % (branch.revno() + 1))
177
branch.append_revision(rev_id)
179
branch.set_pending_merges([])
182
note("commited r%d" % branch.revno())
188
def _gen_revision_id(branch, when):
312
self.branch.revision_store.add(rev_tmp, self.rev_id)
313
mutter('new revision_id is {%s}', self.rev_id)
315
def _remove_deleted(self):
316
"""Remove deleted files from the working inventories.
318
This is done prior to taking the working inventory as the
319
basis for the new committed inventory.
321
This returns true if any files
322
*that existed in the basis inventory* were deleted.
323
Files that were added and deleted
324
in the working copy don't matter.
326
specific = self.specific_files
328
for path, ie in self.work_inv.iter_entries():
329
if specific and not is_inside_any(specific, path):
331
if not self.work_tree.has_filename(path):
332
self.reporter.missing(path)
333
deleted_ids.append((path, ie.file_id))
335
deleted_ids.sort(reverse=True)
336
for path, file_id in deleted_ids:
337
del self.work_inv[file_id]
338
self.branch._write_inventory(self.work_inv)
340
def _store_snapshot(self):
341
"""Pass over inventory and record a snapshot.
343
Entries get a new revision when they are modified in
344
any way, which includes a merge with a new set of
345
parents that have the same entry.
347
# XXX: Need to think more here about when the user has
348
# made a specific decision on a particular value -- c.f.
350
for path, ie in self.new_inv.iter_entries():
351
previous_entries = ie.find_previous_heads(
353
self.weave_store.get_weave_or_empty(ie.file_id,
354
self.branch.get_transaction()))
355
if ie.revision is None:
356
change = ie.snapshot(self.rev_id, path, previous_entries,
357
self.work_tree, self.weave_store,
358
self.branch.get_transaction())
361
self.reporter.snapshot_change(change, path)
363
def _populate_new_inv(self):
364
"""Build revision inventory.
366
This creates a new empty inventory. Depending on
367
which files are selected for commit, and what is present in the
368
current tree, the new inventory is populated. inventory entries
369
which are candidates for modification have their revision set to
370
None; inventory entries that are carried over untouched have their
371
revision set to their prior value.
373
mutter("Selecting files for commit with filter %s", self.specific_files)
374
self.new_inv = Inventory()
375
for path, new_ie in self.work_inv.iter_entries():
376
file_id = new_ie.file_id
377
mutter('check %s {%s}', path, new_ie.file_id)
378
if self.specific_files:
379
if not is_inside_any(self.specific_files, path):
380
mutter('%s not selected for commit', path)
381
self._carry_entry(file_id)
384
# this is selected, ensure its parents are too.
385
parent_id = new_ie.parent_id
386
while parent_id != ROOT_ID:
387
if not self.new_inv.has_id(parent_id):
388
ie = self._select_entry(self.work_inv[parent_id])
389
mutter('%s selected for commit because of %s',
390
self.new_inv.id2path(parent_id), path)
392
ie = self.new_inv[parent_id]
393
if ie.revision is not None:
395
mutter('%s selected for commit because of %s',
396
self.new_inv.id2path(parent_id), path)
397
parent_id = ie.parent_id
398
mutter('%s selected for commit', path)
399
self._select_entry(new_ie)
401
def _select_entry(self, new_ie):
402
"""Make new_ie be considered for committing."""
408
def _carry_entry(self, file_id):
409
"""Carry the file unchanged from the basis revision."""
410
if self.basis_inv.has_id(file_id):
411
self.new_inv.add(self.basis_inv[file_id].copy())
413
def _report_deletes(self):
414
for file_id in self.basis_inv:
415
if file_id not in self.new_inv:
416
self.reporter.deleted(self.basis_inv.id2path(file_id))
418
def _gen_revision_id(config, when):
189
419
"""Return new revision-id."""
190
from binascii import hexlify
191
from bzrlib.osutils import rand_bytes, compact_date, user_email
193
s = '%s-%s-' % (user_email(branch), compact_date(when))
420
s = '%s-%s-' % (config.user_email(), compact_date(when))
194
421
s += hexlify(rand_bytes(8))
198
def _gather_commit(branch, work_tree, work_inv, basis_inv, specific_files,
200
"""Build inventory preparatory to commit.
202
Returns missing_ids, new_inv, any_changes.
204
This adds any changed files into the text store, and sets their
205
test-id, sha and size in the returned inventory appropriately.
208
Modified to hold a list of files that have been deleted from
209
the working directory; these should be removed from the
212
from bzrlib.inventory import Inventory
213
from bzrlib.osutils import isdir, isfile, sha_string, quotefn, \
214
local_time_offset, username, kind_marker, is_inside_any
216
from bzrlib.branch import gen_file_id
217
from bzrlib.errors import BzrError
218
from bzrlib.revision import Revision
219
from bzrlib.trace import mutter, note
222
inv = Inventory(work_inv.root.file_id)
225
for path, entry in work_inv.iter_entries():
226
## TODO: Check that the file kind has not changed from the previous
227
## revision of this file (if any).
229
p = branch.abspath(path)
230
file_id = entry.file_id
231
mutter('commit prep file %s, id %r ' % (p, file_id))
233
if specific_files and not is_inside_any(specific_files, path):
234
mutter(' skipping file excluded from commit')
235
if basis_inv.has_id(file_id):
236
# carry over with previous state
237
inv.add(basis_inv[file_id].copy())
239
# omit this from committed inventory
243
if not work_tree.has_id(file_id):
245
print('deleted %s%s' % (path, kind_marker(entry.kind)))
247
mutter(" file is missing, removing from inventory")
248
missing_ids.append(file_id)
251
# this is present in the new inventory; may be new, modified or
253
old_ie = basis_inv.has_id(file_id) and basis_inv[file_id]
259
old_kind = old_ie.kind
260
if old_kind != entry.kind:
261
raise BzrError("entry %r changed kind from %r to %r"
262
% (file_id, old_kind, entry.kind))
264
if entry.kind == 'directory':
266
raise BzrError("%s is entered as directory but not a directory"
268
elif entry.kind == 'file':
270
raise BzrError("%s is entered as file but is not a file" % quotefn(p))
272
new_sha1 = work_tree.get_file_sha1(file_id)
275
and old_ie.text_sha1 == new_sha1):
276
## assert content == basis.get_file(file_id).read()
277
entry.text_id = old_ie.text_id
278
entry.text_sha1 = new_sha1
279
entry.text_size = old_ie.text_size
280
mutter(' unchanged from previous text_id {%s}' %
283
content = file(p, 'rb').read()
285
# calculate the sha again, just in case the file contents
286
# changed since we updated the cache
287
entry.text_sha1 = sha_string(content)
288
entry.text_size = len(content)
290
entry.text_id = gen_file_id(entry.name)
291
branch.text_store.add(content, entry.text_id)
292
mutter(' stored with text_id {%s}' % entry.text_id)
295
marked = path + kind_marker(entry.kind)
297
print 'added', marked
299
elif old_ie == entry:
301
elif (old_ie.name == entry.name
302
and old_ie.parent_id == entry.parent_id):
303
print 'modified', marked
306
print 'renamed', marked
309
return missing_ids, inv, any_changes