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.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
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.weave import Weave
94
from bzrlib.weavefile import read_weave, write_weave_v5
95
from bzrlib.atomicfile import AtomicFile
98
def commit(*args, **kwargs):
99
"""Commit a new revision to a branch.
101
Function-style interface for convenience of old callers.
103
New code should use the Commit class instead.
105
## XXX: Remove this in favor of Branch.commit?
106
Commit().commit(*args, **kwargs)
109
class NullCommitReporter(object):
110
"""I report on progress of a commit."""
112
def snapshot_change(self, change, path):
115
def completed(self, revno, rev_id):
118
def deleted(self, file_id):
121
def escaped(self, escape_count, message):
124
def missing(self, path):
127
class ReportCommitToLog(NullCommitReporter):
129
def snapshot_change(self, change, path):
130
note("%s %s", change, path)
132
def completed(self, revno, rev_id):
133
note('committed r%d {%s}', revno, rev_id)
135
def deleted(self, file_id):
136
note('deleted %s', file_id)
138
def escaped(self, escape_count, message):
139
note("replaced %d control characters in message", escape_count)
141
def missing(self, path):
142
note('missing %s', path)
144
class Commit(object):
145
"""Task of committing a new revision.
147
This is a MethodObject: it accumulates state as the commit is
148
prepared, and then it is discarded. It doesn't represent
149
historical revisions, just the act of recording a new one.
152
Modified to hold a list of files that have been deleted from
153
the working directory; these should be removed from the
159
if reporter is not None:
160
self.reporter = reporter
162
self.reporter = NullCommitReporter()
163
if config is not None:
175
allow_pointless=True,
179
"""Commit working copy as a new revision.
181
timestamp -- if not None, seconds-since-epoch for a
182
postdated/predated commit.
184
specific_files -- If true, commit only those files.
186
rev_id -- If set, use this as the new revision id.
187
Useful for test or import commands that need to tightly
188
control what revisions are assigned. If you duplicate
189
a revision id that exists elsewhere it is your own fault.
190
If null (default), a time/random revision id is generated.
192
allow_pointless -- If true (default), commit even if nothing
193
has changed and no merges are recorded.
195
strict -- If true, don't allow a commit if the working tree
196
contains unknown files.
198
revprops -- Properties for new revision
200
mutter('preparing to commit')
203
self.weave_store = branch.weave_store
205
self.specific_files = specific_files
206
self.allow_pointless = allow_pointless
207
self.revprops = revprops
209
if strict and branch.unknowns():
210
raise StrictCommitFailed()
212
if timestamp is None:
213
self.timestamp = time.time()
215
self.timestamp = long(timestamp)
217
if self.config is None:
218
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...')
89
# print 'looking for changes...'
90
# disabled; should be done at a higher level
93
pending_merges = branch.pending_merges()
95
missing_ids, new_inv, any_changes = \
96
_gather_commit(branch,
103
if not (any_changes or allow_pointless or pending_merges):
104
raise PointlessCommit()
106
for file_id in missing_ids:
107
# Any files that have been deleted are now removed from the
108
# working inventory. Files that were not selected for commit
109
# are left as they were in the working inventory and ommitted
110
# from the revision inventory.
112
# have to do this later so we don't mess up the iterator.
113
# since parents may be removed before their children we
116
# FIXME: There's probably a better way to do this; perhaps
117
# the workingtree should know how to filter itbranch.
118
if work_inv.has_id(file_id):
119
del work_inv[file_id]
220
121
if rev_id is None:
221
self.rev_id = _gen_revision_id(self.config, self.timestamp)
225
if committer is None:
226
self.committer = self.config.username()
228
assert isinstance(committer, basestring), type(committer)
229
self.committer = committer
232
self.timezone = local_time_offset()
234
self.timezone = int(timezone)
236
assert isinstance(message, basestring), type(message)
237
self.message = message
238
self._escape_commit_message()
240
self.branch.lock_write()
242
self.work_tree = self.branch.working_tree()
243
self.work_inv = self.work_tree.inventory
244
self.basis_tree = self.branch.basis_tree()
245
self.basis_inv = self.basis_tree.inventory
247
self._gather_parents()
248
if len(self.parents) > 1 and self.specific_files:
249
raise NotImplementedError('selected-file commit of merges is not supported yet')
250
self._check_parents_present()
252
self._remove_deleted()
253
self._populate_new_inv()
254
self._store_snapshot()
255
self._report_deletes()
257
if not (self.allow_pointless
258
or len(self.parents) > 1
259
or self.new_inv != self.basis_inv):
260
raise PointlessCommit()
262
if len(list(self.work_tree.iter_conflicts()))>0:
263
raise ConflictsInTree
265
self._record_inventory()
266
self._make_revision()
267
self.reporter.completed(self.branch.revno()+1, self.rev_id)
268
self.branch.append_revision(self.rev_id)
269
self.branch.set_pending_merges([])
273
def _record_inventory(self):
274
"""Store the inventory for the new revision."""
275
inv_text = serializer_v5.write_inventory_to_string(self.new_inv)
276
self.inv_sha1 = sha_string(inv_text)
277
s = self.branch.control_weaves
278
s.add_text('inventory', self.rev_id,
279
split_lines(inv_text), self.present_parents,
280
self.branch.get_transaction())
282
def _escape_commit_message(self):
283
"""Replace xml-incompatible control characters."""
284
# Python strings can include characters that can't be
285
# represented in well-formed XML; escape characters that
286
# aren't listed in the XML specification
287
# (http://www.w3.org/TR/REC-xml/#NT-Char).
288
if isinstance(self.message, unicode):
289
char_pattern = u'[^\x09\x0A\x0D\u0020-\uD7FF\uE000-\uFFFD]'
291
# Use a regular 'str' as pattern to avoid having re.subn
292
# return 'unicode' results.
293
char_pattern = '[^x09\x0A\x0D\x20-\xFF]'
294
self.message, escape_count = re.subn(
296
lambda match: match.group(0).encode('unicode_escape'),
299
self.reporter.escaped(escape_count, self.message)
301
def _gather_parents(self):
302
"""Record the parents of a merge for merge detection."""
303
pending_merges = self.branch.pending_merges()
305
self.parent_invs = []
306
self.present_parents = []
307
precursor_id = self.branch.last_revision()
122
rev_id = _gen_revision_id(branch, time.time())
125
inv_tmp = tempfile.TemporaryFile()
126
pack_xml(new_inv, inv_tmp)
128
branch.inventory_store.add(inv_tmp, inv_id)
129
mutter('new inventory_id is {%s}' % inv_id)
131
# We could also just sha hash the inv_tmp file
132
# however, in the case that branch.inventory_store.add()
133
# ever actually does anything special
134
inv_sha1 = branch.get_inventory_sha1(inv_id)
136
branch._write_inventory(work_inv)
138
if timestamp == None:
139
timestamp = time.time()
141
if committer == None:
142
committer = username(branch)
145
timezone = local_time_offset()
147
mutter("building commit log message")
148
rev = Revision(timestamp=timestamp,
153
inventory_sha1=inv_sha1,
157
precursor_id = branch.last_patch()
309
self.parents.append(precursor_id)
310
self.parents += pending_merges
311
for revision in self.parents:
312
if self.branch.has_revision(revision):
313
self.parent_invs.append(self.branch.get_inventory(revision))
314
self.present_parents.append(revision)
159
precursor_sha1 = branch.get_revision_sha1(precursor_id)
160
rev.parents.append(RevisionReference(precursor_id, precursor_sha1))
161
for merge_rev in pending_merges:
162
rev.parents.append(RevisionReference(merge_rev))
316
def _check_parents_present(self):
317
for parent_id in self.parents:
318
mutter('commit parent revision {%s}', parent_id)
319
if not self.branch.has_revision(parent_id):
320
if parent_id == self.branch.last_revision():
321
warning("parent is missing %r", parent_id)
322
raise HistoryMissing(self.branch, 'revision', parent_id)
324
mutter("commit will ghost revision %r", parent_id)
326
def _make_revision(self):
327
"""Record a new revision object for this commit."""
328
self.rev = Revision(timestamp=self.timestamp,
329
timezone=self.timezone,
330
committer=self.committer,
331
message=self.message,
332
inventory_sha1=self.inv_sha1,
333
revision_id=self.rev_id,
334
properties=self.revprops)
335
self.rev.parent_ids = self.parents
337
serializer_v5.write_revision(self.rev, rev_tmp)
164
rev_tmp = tempfile.TemporaryFile()
165
pack_xml(rev, rev_tmp)
339
if self.config.signature_needed():
340
plaintext = Testament(self.rev, self.new_inv).as_short_text()
341
self.branch.store_revision_signature(gpg.GPGStrategy(self.config),
342
plaintext, self.rev_id)
343
self.branch.revision_store.add(rev_tmp, self.rev_id)
344
mutter('new revision_id is {%s}', self.rev_id)
346
def _remove_deleted(self):
347
"""Remove deleted files from the working inventories.
349
This is done prior to taking the working inventory as the
350
basis for the new committed inventory.
352
This returns true if any files
353
*that existed in the basis inventory* were deleted.
354
Files that were added and deleted
355
in the working copy don't matter.
357
specific = self.specific_files
359
for path, ie in self.work_inv.iter_entries():
360
if specific and not is_inside_any(specific, path):
362
if not self.work_tree.has_filename(path):
363
self.reporter.missing(path)
364
deleted_ids.append((path, ie.file_id))
366
deleted_ids.sort(reverse=True)
367
for path, file_id in deleted_ids:
368
del self.work_inv[file_id]
369
self.branch._write_inventory(self.work_inv)
371
def _store_snapshot(self):
372
"""Pass over inventory and record a snapshot.
374
Entries get a new revision when they are modified in
375
any way, which includes a merge with a new set of
376
parents that have the same entry.
378
# XXX: Need to think more here about when the user has
379
# made a specific decision on a particular value -- c.f.
381
for path, ie in self.new_inv.iter_entries():
382
previous_entries = ie.find_previous_heads(
384
self.weave_store.get_weave_or_empty(ie.file_id,
385
self.branch.get_transaction()))
386
if ie.revision is None:
387
change = ie.snapshot(self.rev_id, path, previous_entries,
388
self.work_tree, self.weave_store,
389
self.branch.get_transaction())
392
self.reporter.snapshot_change(change, path)
394
def _populate_new_inv(self):
395
"""Build revision inventory.
397
This creates a new empty inventory. Depending on
398
which files are selected for commit, and what is present in the
399
current tree, the new inventory is populated. inventory entries
400
which are candidates for modification have their revision set to
401
None; inventory entries that are carried over untouched have their
402
revision set to their prior value.
404
mutter("Selecting files for commit with filter %s", self.specific_files)
405
self.new_inv = Inventory()
406
for path, new_ie in self.work_inv.iter_entries():
407
file_id = new_ie.file_id
408
mutter('check %s {%s}', path, new_ie.file_id)
409
if self.specific_files:
410
if not is_inside_any(self.specific_files, path):
411
mutter('%s not selected for commit', path)
412
self._carry_entry(file_id)
415
# this is selected, ensure its parents are too.
416
parent_id = new_ie.parent_id
417
while parent_id != ROOT_ID:
418
if not self.new_inv.has_id(parent_id):
419
ie = self._select_entry(self.work_inv[parent_id])
420
mutter('%s selected for commit because of %s',
421
self.new_inv.id2path(parent_id), path)
423
ie = self.new_inv[parent_id]
424
if ie.revision is not None:
426
mutter('%s selected for commit because of %s',
427
self.new_inv.id2path(parent_id), path)
428
parent_id = ie.parent_id
429
mutter('%s selected for commit', path)
430
self._select_entry(new_ie)
432
def _select_entry(self, new_ie):
433
"""Make new_ie be considered for committing."""
439
def _carry_entry(self, file_id):
440
"""Carry the file unchanged from the basis revision."""
441
if self.basis_inv.has_id(file_id):
442
self.new_inv.add(self.basis_inv[file_id].copy())
444
def _report_deletes(self):
445
for file_id in self.basis_inv:
446
if file_id not in self.new_inv:
447
self.reporter.deleted(self.basis_inv.id2path(file_id))
449
def _gen_revision_id(config, when):
167
branch.revision_store.add(rev_tmp, rev_id)
168
mutter("new revision_id is {%s}" % rev_id)
170
## XXX: Everything up to here can simply be orphaned if we abort
171
## the commit; it will leave junk files behind but that doesn't
174
## TODO: Read back the just-generated changeset, and make sure it
175
## applies and recreates the right state.
177
## TODO: Also calculate and store the inventory SHA1
178
mutter("committing patch r%d" % (branch.revno() + 1))
180
branch.append_revision(rev_id)
182
branch.set_pending_merges([])
185
# disabled; should go through logging
186
# note("commited r%d" % branch.revno())
187
# print ("commited r%d" % branch.revno())
194
def _gen_revision_id(branch, when):
450
195
"""Return new revision-id."""
451
s = '%s-%s-' % (config.user_email(), compact_date(when))
196
from binascii import hexlify
197
from bzrlib.osutils import rand_bytes, compact_date, user_email
199
s = '%s-%s-' % (user_email(branch), compact_date(when))
452
200
s += hexlify(rand_bytes(8))
204
def _gather_commit(branch, work_tree, work_inv, basis_inv, specific_files,
206
"""Build inventory preparatory to commit.
208
Returns missing_ids, new_inv, any_changes.
210
This adds any changed files into the text store, and sets their
211
test-id, sha and size in the returned inventory appropriately.
214
Modified to hold a list of files that have been deleted from
215
the working directory; these should be removed from the
218
from bzrlib.inventory import Inventory
219
from bzrlib.osutils import isdir, isfile, sha_string, quotefn, \
220
local_time_offset, username, kind_marker, is_inside_any
222
from bzrlib.branch import gen_file_id
223
from bzrlib.errors import BzrError
224
from bzrlib.revision import Revision
225
from bzrlib.trace import mutter, note
228
inv = Inventory(work_inv.root.file_id)
231
for path, entry in work_inv.iter_entries():
232
## TODO: Check that the file kind has not changed from the previous
233
## revision of this file (if any).
235
p = branch.abspath(path)
236
file_id = entry.file_id
237
mutter('commit prep file %s, id %r ' % (p, file_id))
239
if specific_files and not is_inside_any(specific_files, path):
240
mutter(' skipping file excluded from commit')
241
if basis_inv.has_id(file_id):
242
# carry over with previous state
243
inv.add(basis_inv[file_id].copy())
245
# omit this from committed inventory
249
if not work_tree.has_id(file_id):
251
print('deleted %s%s' % (path, kind_marker(entry.kind)))
253
mutter(" file is missing, removing from inventory")
254
missing_ids.append(file_id)
257
# this is present in the new inventory; may be new, modified or
259
old_ie = basis_inv.has_id(file_id) and basis_inv[file_id]
265
old_kind = old_ie.kind
266
if old_kind != entry.kind:
267
raise BzrError("entry %r changed kind from %r to %r"
268
% (file_id, old_kind, entry.kind))
270
if entry.kind == 'directory':
272
raise BzrError("%s is entered as directory but not a directory"
274
elif entry.kind == 'file':
276
raise BzrError("%s is entered as file but is not a file" % quotefn(p))
278
new_sha1 = work_tree.get_file_sha1(file_id)
281
and old_ie.text_sha1 == new_sha1):
282
## assert content == basis.get_file(file_id).read()
283
entry.text_id = old_ie.text_id
284
entry.text_sha1 = new_sha1
285
entry.text_size = old_ie.text_size
286
mutter(' unchanged from previous text_id {%s}' %
289
content = file(p, 'rb').read()
291
# calculate the sha again, just in case the file contents
292
# changed since we updated the cache
293
entry.text_sha1 = sha_string(content)
294
entry.text_size = len(content)
296
entry.text_id = gen_file_id(entry.name)
297
branch.text_store.add(content, entry.text_id)
298
mutter(' stored with text_id {%s}' % entry.text_id)
301
marked = path + kind_marker(entry.kind)
303
print 'added', marked
305
elif old_ie == entry:
307
elif (old_ie.name == entry.name
308
and old_ie.parent_id == entry.parent_id):
309
print 'modified', marked
312
print 'renamed', marked
315
return missing_ids, inv, any_changes