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
"""Commit working copy as a new revision.
28
The basic approach is to add all the file texts into the
29
store, then the inventory, then make a new revision pointing
30
to that inventory and store that.
32
This is not quite safe if the working copy changes during the
33
commit; for the moment that is simply not allowed. A better
34
approach is to make a temporary copy of the files before
35
computing their hashes, and then add those hashes in turn to
36
the inventory. This should mean at least that there are no
37
broken hash pointers. There is no way we can get a snapshot
38
of the whole directory at an instant. This would also have to
39
be robust against files disappearing, moving, etc. So the
40
whole thing is a bit hard.
42
timestamp -- if not None, seconds-since-epoch for a
43
postdated/predated commit.
46
If true, commit only those files.
49
If set, use this as the new revision id.
50
Useful for test or import commands that need to tightly
51
control what revisions are assigned. If you duplicate
52
a revision id that exists elsewhere it is your own fault.
53
If null (default), a time/random revision id is generated.
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.osutils import (local_time_offset,
76
rand_bytes, compact_date,
77
kind_marker, is_inside_any, quotefn,
78
sha_string, sha_strings, sha_file, isdir, isfile,
80
from bzrlib.branch import gen_file_id
82
from bzrlib.errors import (BzrError, PointlessCommit,
87
from bzrlib.revision import Revision
88
from bzrlib.trace import mutter, note, warning
89
from bzrlib.xml5 import serializer_v5
90
from bzrlib.inventory import Inventory, ROOT_ID
91
from bzrlib.weave import Weave
92
from bzrlib.weavefile import read_weave, write_weave_v5
93
from bzrlib.atomicfile import AtomicFile
96
def commit(*args, **kwargs):
97
"""Commit a new revision to a branch.
99
Function-style interface for convenience of old callers.
101
New code should use the Commit class instead.
56
import os, time, tempfile
58
from inventory import Inventory
59
from osutils import isdir, isfile, sha_string, quotefn, \
60
local_time_offset, username, kind_marker, is_inside_any
103
## XXX: Remove this in favor of Branch.commit?
104
Commit().commit(*args, **kwargs)
107
class NullCommitReporter(object):
108
"""I report on progress of a commit."""
110
def snapshot_change(self, change, path):
113
def completed(self, revno, rev_id):
116
def deleted(self, file_id):
119
def escaped(self, escape_count, message):
122
def missing(self, path):
125
class ReportCommitToLog(NullCommitReporter):
127
def snapshot_change(self, change, path):
128
note("%s %s", change, path)
130
def completed(self, revno, rev_id):
131
note('committed r%d {%s}', revno, rev_id)
62
from branch import gen_file_id
63
from errors import BzrError
64
from revision import Revision
65
from trace import mutter, note
67
branch._need_writelock()
69
# First walk over the working inventory; and both update that
70
# and also build a new revision inventory. The revision
71
# inventory needs to hold the text-id, sha1 and size of the
72
# actual file versions committed in the revision. (These are
73
# not present in the working inventory.) We also need to
74
# detect missing/deleted files, and remove them from the
77
work_tree = branch.working_tree()
78
work_inv = work_tree.inventory
80
basis = branch.basis_tree()
81
basis_inv = basis.inventory
85
note('looking for changes...')
87
for path, entry in work_inv.iter_entries():
88
## TODO: Check that the file kind has not changed from the previous
89
## revision of this file (if any).
93
p = branch.abspath(path)
94
file_id = entry.file_id
95
mutter('commit prep file %s, id %r ' % (p, file_id))
97
if specific_files and not is_inside_any(specific_files, path):
98
if basis_inv.has_id(file_id):
99
# carry over with previous state
100
inv.add(basis_inv[file_id].copy())
102
# omit this from committed inventory
106
if not work_tree.has_id(file_id):
108
print('deleted %s%s' % (path, kind_marker(entry.kind)))
109
mutter(" file is missing, removing from inventory")
110
missing_ids.append(file_id)
115
if basis_inv.has_id(file_id):
116
old_kind = basis_inv[file_id].kind
117
if old_kind != entry.kind:
118
raise BzrError("entry %r changed kind from %r to %r"
119
% (file_id, old_kind, entry.kind))
121
if entry.kind == 'directory':
123
raise BzrError("%s is entered as directory but not a directory"
125
elif entry.kind == 'file':
127
raise BzrError("%s is entered as file but is not a file" % quotefn(p))
129
new_sha1 = work_tree.get_file_sha1(file_id)
131
old_ie = basis_inv.has_id(file_id) and basis_inv[file_id]
133
and old_ie.text_sha1 == new_sha1):
134
## assert content == basis.get_file(file_id).read()
135
entry.text_id = old_ie.text_id
136
entry.text_sha1 = new_sha1
137
entry.text_size = old_ie.text_size
138
mutter(' unchanged from previous text_id {%s}' %
141
content = file(p, 'rb').read()
143
# calculate the sha again, just in case the file contents
144
# changed since we updated the cache
145
entry.text_sha1 = sha_string(content)
146
entry.text_size = len(content)
148
entry.text_id = gen_file_id(entry.name)
149
branch.text_store.add(content, entry.text_id)
150
mutter(' stored with text_id {%s}' % entry.text_id)
153
print('added %s' % path)
154
elif (old_ie.name == entry.name
155
and old_ie.parent_id == entry.parent_id):
156
print('modified %s' % path)
158
print('renamed %s' % path)
161
for file_id in missing_ids:
162
# Any files that have been deleted are now removed from the
163
# working inventory. Files that were not selected for commit
164
# are left as they were in the working inventory and ommitted
165
# from the revision inventory.
167
# have to do this later so we don't mess up the iterator.
168
# since parents may be removed before their children we
171
# FIXME: There's probably a better way to do this; perhaps
172
# the workingtree should know how to filter itbranch.
173
if work_inv.has_id(file_id):
174
del work_inv[file_id]
178
rev_id = _gen_revision_id(time.time())
181
inv_tmp = tempfile.TemporaryFile()
182
inv.write_xml(inv_tmp)
184
branch.inventory_store.add(inv_tmp, inv_id)
185
mutter('new inventory_id is {%s}' % inv_id)
187
branch._write_inventory(work_inv)
189
if timestamp == None:
190
timestamp = time.time()
192
if committer == None:
193
committer = username()
196
timezone = local_time_offset()
198
mutter("building commit log message")
199
rev = Revision(timestamp=timestamp,
202
precursor = branch.last_patch(),
207
rev_tmp = tempfile.TemporaryFile()
208
rev.write_xml(rev_tmp)
210
branch.revision_store.add(rev_tmp, rev_id)
211
mutter("new revision_id is {%s}" % rev_id)
213
## XXX: Everything up to here can simply be orphaned if we abort
214
## the commit; it will leave junk files behind but that doesn't
217
## TODO: Read back the just-generated changeset, and make sure it
218
## applies and recreates the right state.
220
## TODO: Also calculate and store the inventory SHA1
221
mutter("committing patch r%d" % (branch.revno() + 1))
223
branch.append_revision(rev_id)
226
note("commited r%d" % branch.revno())
230
def _gen_revision_id(when):
133
def deleted(self, file_id):
134
note('deleted %s', file_id)
136
def escaped(self, escape_count, message):
137
note("replaced %d control characters in message", escape_count)
139
def missing(self, path):
140
note('missing %s', path)
142
class Commit(object):
143
"""Task of committing a new revision.
145
This is a MethodObject: it accumulates state as the commit is
146
prepared, and then it is discarded. It doesn't represent
147
historical revisions, just the act of recording a new one.
150
Modified to hold a list of files that have been deleted from
151
the working directory; these should be removed from the
156
if reporter is not None:
157
self.reporter = reporter
159
self.reporter = NullCommitReporter()
169
allow_pointless=True,
173
"""Commit working copy as a new revision.
175
timestamp -- if not None, seconds-since-epoch for a
176
postdated/predated commit.
178
specific_files -- If true, commit only those files.
180
rev_id -- If set, use this as the new revision id.
181
Useful for test or import commands that need to tightly
182
control what revisions are assigned. If you duplicate
183
a revision id that exists elsewhere it is your own fault.
184
If null (default), a time/random revision id is generated.
186
allow_pointless -- If true (default), commit even if nothing
187
has changed and no merges are recorded.
189
strict -- If true, don't allow a commit if the working tree
190
contains unknown files.
192
revprops -- Properties for new revision
194
mutter('preparing to commit')
197
self.weave_store = branch.weave_store
199
self.specific_files = specific_files
200
self.allow_pointless = allow_pointless
201
self.revprops = revprops
203
if strict and branch.unknowns():
204
raise StrictCommitFailed()
206
if timestamp is None:
207
self.timestamp = time.time()
209
self.timestamp = long(timestamp)
211
config = bzrlib.config.BranchConfig(self.branch)
213
self.rev_id = _gen_revision_id(config, self.timestamp)
217
if committer is None:
218
self.committer = config.username()
220
assert isinstance(committer, basestring), type(committer)
221
self.committer = committer
224
self.timezone = local_time_offset()
226
self.timezone = int(timezone)
228
assert isinstance(message, basestring), type(message)
229
self.message = message
230
self._escape_commit_message()
232
self.branch.lock_write()
234
self.work_tree = self.branch.working_tree()
235
self.work_inv = self.work_tree.inventory
236
self.basis_tree = self.branch.basis_tree()
237
self.basis_inv = self.basis_tree.inventory
239
self._gather_parents()
240
if len(self.parents) > 1 and self.specific_files:
241
raise NotImplementedError('selected-file commit of merges is not supported yet')
242
self._check_parents_present()
244
self._remove_deleted()
245
self._populate_new_inv()
246
self._store_snapshot()
247
self._report_deletes()
249
if not (self.allow_pointless
250
or len(self.parents) > 1
251
or self.new_inv != self.basis_inv):
252
raise PointlessCommit()
254
if len(list(self.work_tree.iter_conflicts()))>0:
255
raise ConflictsInTree
257
self._record_inventory()
258
self._make_revision()
259
self.reporter.completed(self.branch.revno()+1, self.rev_id)
260
self.branch.append_revision(self.rev_id)
261
self.branch.set_pending_merges([])
265
def _record_inventory(self):
266
"""Store the inventory for the new revision."""
267
inv_text = serializer_v5.write_inventory_to_string(self.new_inv)
268
self.inv_sha1 = sha_string(inv_text)
269
s = self.branch.control_weaves
270
s.add_text('inventory', self.rev_id,
271
split_lines(inv_text), self.present_parents,
272
self.branch.get_transaction())
274
def _escape_commit_message(self):
275
"""Replace xml-incompatible control characters."""
276
# Python strings can include characters that can't be
277
# represented in well-formed XML; escape characters that
278
# aren't listed in the XML specification
279
# (http://www.w3.org/TR/REC-xml/#NT-Char).
280
if isinstance(self.message, unicode):
281
char_pattern = u'[^\x09\x0A\x0D\u0020-\uD7FF\uE000-\uFFFD]'
283
# Use a regular 'str' as pattern to avoid having re.subn
284
# return 'unicode' results.
285
char_pattern = '[^x09\x0A\x0D\x20-\xFF]'
286
self.message, escape_count = re.subn(
288
lambda match: match.group(0).encode('unicode_escape'),
291
self.reporter.escaped(escape_count, self.message)
293
def _gather_parents(self):
294
"""Record the parents of a merge for merge detection."""
295
pending_merges = self.branch.pending_merges()
297
self.parent_invs = []
298
self.present_parents = []
299
precursor_id = self.branch.last_revision()
301
self.parents.append(precursor_id)
302
self.parents += pending_merges
303
for revision in self.parents:
304
if self.branch.has_revision(revision):
305
self.parent_invs.append(self.branch.get_inventory(revision))
306
self.present_parents.append(revision)
308
def _check_parents_present(self):
309
for parent_id in self.parents:
310
mutter('commit parent revision {%s}', parent_id)
311
if not self.branch.has_revision(parent_id):
312
if parent_id == self.branch.last_revision():
313
warning("parent is missing %r", parent_id)
314
raise HistoryMissing(self.branch, 'revision', parent_id)
316
mutter("commit will ghost revision %r", parent_id)
318
def _make_revision(self):
319
"""Record a new revision object for this commit."""
320
self.rev = Revision(timestamp=self.timestamp,
321
timezone=self.timezone,
322
committer=self.committer,
323
message=self.message,
324
inventory_sha1=self.inv_sha1,
325
revision_id=self.rev_id,
326
properties=self.revprops)
327
self.rev.parent_ids = self.parents
329
serializer_v5.write_revision(self.rev, rev_tmp)
331
self.branch.revision_store.add(rev_tmp, self.rev_id)
332
mutter('new revision_id is {%s}', self.rev_id)
334
def _remove_deleted(self):
335
"""Remove deleted files from the working inventories.
337
This is done prior to taking the working inventory as the
338
basis for the new committed inventory.
340
This returns true if any files
341
*that existed in the basis inventory* were deleted.
342
Files that were added and deleted
343
in the working copy don't matter.
345
specific = self.specific_files
347
for path, ie in self.work_inv.iter_entries():
348
if specific and not is_inside_any(specific, path):
350
if not self.work_tree.has_filename(path):
351
self.reporter.missing(path)
352
deleted_ids.append((path, ie.file_id))
354
deleted_ids.sort(reverse=True)
355
for path, file_id in deleted_ids:
356
del self.work_inv[file_id]
357
self.branch._write_inventory(self.work_inv)
359
def _store_snapshot(self):
360
"""Pass over inventory and record a snapshot.
362
Entries get a new revision when they are modified in
363
any way, which includes a merge with a new set of
364
parents that have the same entry.
366
# XXX: Need to think more here about when the user has
367
# made a specific decision on a particular value -- c.f.
369
for path, ie in self.new_inv.iter_entries():
370
previous_entries = ie.find_previous_heads(
372
self.weave_store.get_weave_or_empty(ie.file_id,
373
self.branch.get_transaction()))
374
if ie.revision is None:
375
change = ie.snapshot(self.rev_id, path, previous_entries,
376
self.work_tree, self.weave_store,
377
self.branch.get_transaction())
380
self.reporter.snapshot_change(change, path)
382
def _populate_new_inv(self):
383
"""Build revision inventory.
385
This creates a new empty inventory. Depending on
386
which files are selected for commit, and what is present in the
387
current tree, the new inventory is populated. inventory entries
388
which are candidates for modification have their revision set to
389
None; inventory entries that are carried over untouched have their
390
revision set to their prior value.
392
mutter("Selecting files for commit with filter %s", self.specific_files)
393
self.new_inv = Inventory()
394
for path, new_ie in self.work_inv.iter_entries():
395
file_id = new_ie.file_id
396
mutter('check %s {%s}', path, new_ie.file_id)
397
if self.specific_files:
398
if not is_inside_any(self.specific_files, path):
399
mutter('%s not selected for commit', path)
400
self._carry_entry(file_id)
403
# this is selected, ensure its parents are too.
404
parent_id = new_ie.parent_id
405
while parent_id != ROOT_ID:
406
if not self.new_inv.has_id(parent_id):
407
ie = self._select_entry(self.work_inv[parent_id])
408
mutter('%s selected for commit because of %s',
409
self.new_inv.id2path(parent_id), path)
411
ie = self.new_inv[parent_id]
412
if ie.revision is not None:
414
mutter('%s selected for commit because of %s',
415
self.new_inv.id2path(parent_id), path)
416
parent_id = ie.parent_id
417
mutter('%s selected for commit', path)
418
self._select_entry(new_ie)
420
def _select_entry(self, new_ie):
421
"""Make new_ie be considered for committing."""
427
def _carry_entry(self, file_id):
428
"""Carry the file unchanged from the basis revision."""
429
if self.basis_inv.has_id(file_id):
430
self.new_inv.add(self.basis_inv[file_id].copy())
432
def _report_deletes(self):
433
for file_id in self.basis_inv:
434
if file_id not in self.new_inv:
435
self.reporter.deleted(self.basis_inv.id2path(file_id))
437
def _gen_revision_id(config, when):
231
438
"""Return new revision-id."""
232
from binascii import hexlify
233
from osutils import rand_bytes, compact_date, user_email
235
s = '%s-%s-' % (user_email(), compact_date(when))
439
s = '%s-%s-' % (config.user_email(), compact_date(when))
236
440
s += hexlify(rand_bytes(8))
added
removed
bzrlib/commit.py
