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
# 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.
70
from binascii import hexlify
71
from cStringIO import StringIO
73
from bzrlib.osutils import (local_time_offset, username,
74
rand_bytes, compact_date, user_email,
75
kind_marker, is_inside_any, quotefn,
76
sha_string, sha_strings, sha_file, isdir, isfile,
78
from bzrlib.branch import gen_file_id
79
from bzrlib.errors import (BzrError, PointlessCommit,
82
from bzrlib.revision import Revision
83
from bzrlib.trace import mutter, note, warning
84
from bzrlib.xml5 import serializer_v5
85
from bzrlib.inventory import Inventory
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."""
104
def added(self, path):
107
def removed(self, path):
110
def renamed(self, old_path, new_path):
114
class ReportCommitToLog(NullCommitReporter):
115
def added(self, path):
116
note('added %s', path)
118
def removed(self, path):
119
note('removed %s', path)
121
def renamed(self, old_path, new_path):
122
note('renamed %s => %s', old_path, new_path)
125
class Commit(object):
126
"""Task of committing a new revision.
128
This is a MethodObject: it accumulates state as the commit is
129
prepared, and then it is discarded. It doesn't represent
130
historical revisions, just the act of recording a new one.
133
Modified to hold a list of files that have been deleted from
134
the working directory; these should be removed from the
139
if reporter is not None:
140
self.reporter = reporter
142
self.reporter = NullCommitReporter()
152
allow_pointless=True,
154
"""Commit working copy as a new revision.
156
timestamp -- if not None, seconds-since-epoch for a
157
postdated/predated commit.
159
specific_files -- If true, commit only those files.
161
rev_id -- If set, use this as the new revision id.
162
Useful for test or import commands that need to tightly
163
control what revisions are assigned. If you duplicate
164
a revision id that exists elsewhere it is your own fault.
165
If null (default), a time/random revision id is generated.
167
allow_pointless -- If true (default), commit even if nothing
168
has changed and no merges are recorded.
170
mutter('preparing to commit')
173
self.weave_store = branch.weave_store
175
self.specific_files = specific_files
176
self.allow_pointless = allow_pointless
178
if timestamp is None:
179
self.timestamp = time.time()
181
self.timestamp = long(timestamp)
119
183
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()
184
self.rev_id = _gen_revision_id(self.branch, self.timestamp)
188
if committer is None:
189
self.committer = username(self.branch)
191
assert isinstance(committer, basestring), type(committer)
192
self.committer = committer
195
self.timezone = local_time_offset()
197
self.timezone = int(timezone)
199
assert isinstance(message, basestring), type(message)
200
self.message = message
202
self.branch.lock_write()
204
self.work_tree = self.branch.working_tree()
205
self.work_inv = self.work_tree.inventory
206
self.basis_tree = self.branch.basis_tree()
207
self.basis_inv = self.basis_tree.inventory
209
self._gather_parents()
210
if len(self.parents) > 1 and self.specific_files:
211
raise NotImplementedError('selected-file commit of merges is not supported yet')
212
self._check_parents_present()
214
self._remove_deleted()
215
self.new_inv = Inventory()
216
self._store_entries()
217
self._report_deletes()
218
self._set_name_versions()
220
if not (self.allow_pointless
221
or len(self.parents) > 1
222
or self.new_inv != self.basis_inv):
223
raise PointlessCommit()
225
self._record_inventory()
226
self._record_ancestry()
227
self._make_revision()
228
note('committed r%d {%s}', (self.branch.revno() + 1),
230
self.branch.append_revision(self.rev_id)
231
self.branch.set_pending_merges([])
237
def _record_inventory(self):
238
"""Store the inventory for the new revision."""
239
inv_text = serializer_v5.write_inventory_to_string(self.new_inv)
240
self.inv_sha1 = sha_string(inv_text)
241
s = self.branch.control_weaves
242
s.add_text('inventory', self.rev_id,
243
split_lines(inv_text), self.parents)
246
def _record_ancestry(self):
247
"""Append merged revision ancestry to the ancestry file.
249
This should be the merged ancestry of all parents, plus the
251
s = self.branch.control_weaves
252
w = s.get_weave_or_empty('ancestry')
253
lines = self._make_ancestry(w)
254
w.add(self.rev_id, self.parents, lines)
255
s.put_weave('ancestry', w)
258
def _make_ancestry(self, ancestry_weave):
259
"""Return merged ancestry lines.
261
The lines are revision-ids followed by newlines."""
262
parent_ancestries = [ancestry_weave.get(p) for p in self.parents]
263
new_lines = merge_ancestry_lines(self.rev_id, parent_ancestries)
264
mutter('merged ancestry of {%s}:\n%s', self.rev_id, ''.join(new_lines))
268
def _gather_parents(self):
269
pending_merges = self.branch.pending_merges()
271
self.parent_trees = []
272
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))
162
rev_tmp = tempfile.TemporaryFile()
163
pack_xml(rev, rev_tmp)
274
self.parents.append(precursor_id)
275
self.parent_trees.append(self.basis_tree)
276
self.parents += pending_merges
277
self.parent_trees.extend(map(self.branch.revision_tree, pending_merges))
280
def _check_parents_present(self):
281
for parent_id in self.parents:
282
mutter('commit parent revision {%s}', parent_id)
283
if not self.branch.has_revision(parent_id):
284
warning("can't commit a merge from an absent parent")
285
raise HistoryMissing(self.branch, 'revision', parent_id)
288
def _make_revision(self):
289
"""Record a new revision object for this commit."""
290
self.rev = Revision(timestamp=self.timestamp,
291
timezone=self.timezone,
292
committer=self.committer,
293
message=self.message,
294
inventory_sha1=self.inv_sha1,
295
revision_id=self.rev_id)
296
self.rev.parent_ids = self.parents
298
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):
300
self.branch.revision_store.add(rev_tmp, self.rev_id, compressed=False)
301
mutter('new revision_id is {%s}', self.rev_id)
304
def _remove_deleted(self):
305
"""Remove deleted files from the working inventories.
307
This is done prior to taking the working inventory as the
308
basis for the new committed inventory.
310
This returns true if any files
311
*that existed in the basis inventory* were deleted.
312
Files that were added and deleted
313
in the working copy don't matter.
315
specific = self.specific_files
317
for path, ie in self.work_inv.iter_entries():
318
if specific and not is_inside_any(specific, path):
320
if not self.work_tree.has_filename(path):
321
note('missing %s', path)
322
deleted_ids.append(ie.file_id)
324
for file_id in deleted_ids:
325
del self.work_inv[file_id]
326
self.branch._write_inventory(self.work_inv)
329
def _find_file_parents(self, file_id):
330
"""Return the text versions and hashes for all file parents.
332
Returned as a map from text version to inventory entry.
334
This is a set containing the file versions in all parents
335
revisions containing the file. If the file is new, the set
338
for tree in self.parent_trees:
339
if file_id in tree.inventory:
340
ie = tree.inventory[file_id]
341
assert ie.kind == 'file'
342
assert ie.file_id == file_id
343
if ie.text_version in r:
344
assert r[ie.text_version] == ie
346
r[ie.text_version] = ie
350
def _set_name_versions(self):
351
"""Pass over inventory and mark new entry version as needed.
353
Files get a new name version when they are new, have a
354
different parent, or a different name from in the
355
basis inventory, or if the file is in a different place
356
to any of the parents."""
357
# XXX: Need to think more here about when the user has
358
# made a specific decision on a particular value -- c.f.
360
for path, ie in self.new_inv.iter_entries():
363
for parent_tree in self.parent_trees:
364
parent_inv = parent_tree.inventory
365
if file_id not in parent_inv:
367
parent_ie = parent_inv[file_id]
368
if parent_ie.parent_id != ie.parent_id:
371
elif parent_ie.name != ie.name:
374
elif old_version is None:
375
old_version = parent_ie.name_version
376
elif old_version != parent_ie.name_version:
380
pass # so far so good
381
if old_version is None:
382
mutter('new name_version for {%s}', file_id)
383
ie.name_version = self.rev_id
385
mutter('name_version for {%s} inherited as {%s}',
386
file_id, old_version)
387
ie.name_version = old_version
390
def _store_entries(self):
391
"""Build revision inventory and store modified files.
393
This is called with new_inv a new empty inventory. Depending on
394
which files are selected for commit, and which ones have
395
been modified or merged, new inventory entries are built
396
based on the working and parent inventories.
398
As a side-effect this stores new text versions for committed
399
files with text changes or merges.
401
Each entry can have one of several things happen:
403
carry_file -- carried from the previous version (if not
406
commit_nonfile -- no text to worry about
408
commit_old_text -- same text, may have moved
410
commit_file -- new text version
412
for path, new_ie in self.work_inv.iter_entries():
413
file_id = new_ie.file_id
414
mutter('check %s {%s}', path, new_ie.file_id)
415
if self.specific_files:
416
if not is_inside_any(self.specific_files, path):
417
mutter('%s not selected for commit', path)
418
self._carry_file(file_id)
420
if new_ie.kind != 'file':
421
self._commit_nonfile(file_id)
424
file_parents = self._find_file_parents(file_id)
425
mutter('parents of %s are %r', path, file_parents)
426
if len(file_parents) == 1:
427
parent_ie = file_parents.values()[0]
428
wc_sha1 = self.work_tree.get_file_sha1(file_id)
429
if parent_ie.text_sha1 == wc_sha1:
430
# text not changed or merged
431
self._commit_old_text(file_id, parent_ie)
433
# file is either new, or a file merge; need to record
435
if len(file_parents) > 1:
436
note('merged %s', path)
437
elif len(file_parents) == 0:
438
note('added %s', path)
440
note('modified %s', path)
441
self._commit_file(new_ie, file_id, file_parents)
444
def _commit_nonfile(self, file_id):
445
self.new_inv.add(self.work_inv[file_id].copy())
448
def _carry_file(self, file_id):
449
"""Carry the file unchanged from the basis revision."""
450
if self.basis_inv.has_id(file_id):
451
self.new_inv.add(self.basis_inv[file_id].copy())
454
def _commit_old_text(self, file_id, parent_ie):
455
"""Keep the same text as last time, but possibly a different name."""
456
ie = self.work_inv[file_id].copy()
457
ie.text_version = parent_ie.text_version
458
ie.text_size = parent_ie.text_size
459
ie.text_sha1 = parent_ie.text_sha1
463
def _report_deletes(self):
464
for file_id in self.basis_inv:
465
if file_id not in self.new_inv:
466
note('deleted %s', self.basis_inv.id2path(file_id))
469
def _commit_file(self, new_ie, file_id, file_parents):
470
mutter('store new text for {%s} in revision {%s}',
471
file_id, self.rev_id)
472
new_lines = self.work_tree.get_file(file_id).readlines()
473
self._add_text_to_weave(file_id, new_lines, file_parents)
474
new_ie.text_version = self.rev_id
475
new_ie.text_sha1 = sha_strings(new_lines)
476
new_ie.text_size = sum(map(len, new_lines))
477
self.new_inv.add(new_ie)
480
def _add_text_to_weave(self, file_id, new_lines, parents):
481
self.weave_store.add_text(file_id, self.rev_id, new_lines, parents)
484
def _gen_revision_id(branch, when):
190
485
"""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))
486
s = '%s-%s-' % (user_email(branch), compact_date(when))
195
487
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
493
def merge_ancestry_lines(rev_id, ancestries):
494
"""Return merged ancestry lines.
496
rev_id -- id of the new revision
498
ancestries -- a sequence of ancestries for parent revisions,
499
as newline-terminated line lists.
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
501
if len(ancestries) == 0:
502
return [rev_id + '\n']
503
seen = set(ancestries[0])
504
ancs = ancestries[0][:]
505
for parent_ancestry in ancestries[1:]:
506
for line in parent_ancestry:
507
assert line[-1] == '\n'