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
58
from binascii import hexlify
59
from cStringIO import StringIO
61
from bzrlib.osutils import (local_time_offset, username,
62
rand_bytes, compact_date, user_email,
63
kind_marker, is_inside_any, quotefn,
64
sha_string, sha_strings, sha_file, isdir, isfile,
66
from bzrlib.branch import gen_file_id, INVENTORY_FILEID, ANCESTRY_FILEID
67
from bzrlib.errors import (BzrError, PointlessCommit,
70
from bzrlib.revision import Revision, RevisionReference
71
from bzrlib.trace import mutter, note, warning
72
from bzrlib.xml5 import serializer_v5
73
from bzrlib.inventory import Inventory
74
from bzrlib.weave import Weave
75
from bzrlib.weavefile import read_weave, write_weave_v5
76
from bzrlib.atomicfile import AtomicFile
79
def commit(*args, **kwargs):
80
"""Commit a new revision to a branch.
82
Function-style interface for convenience of old callers.
84
New code should use the Commit class instead.
86
Commit().commit(*args, **kwargs)
89
class NullCommitReporter(object):
90
"""I report on progress of a commit."""
91
def added(self, path):
94
def removed(self, path):
97
def renamed(self, old_path, new_path):
101
class ReportCommitToLog(NullCommitReporter):
102
def added(self, path):
103
note('added %s', path)
105
def removed(self, path):
106
note('removed %s', path)
108
def renamed(self, old_path, new_path):
109
note('renamed %s => %s', old_path, new_path)
112
class Commit(object):
113
"""Task of committing a new revision.
115
This is a MethodObject: it accumulates state as the commit is
116
prepared, and then it is discarded. It doesn't represent
117
historical revisions, just the act of recording a new one.
120
Modified to hold a list of files that have been deleted from
121
the working directory; these should be removed from the
126
if reporter is not None:
127
self.reporter = reporter
129
self.reporter = NullCommitReporter()
139
allow_pointless=True):
140
"""Commit working copy as a new revision.
142
timestamp -- if not None, seconds-since-epoch for a
143
postdated/predated commit.
145
specific_files -- If true, commit only those files.
147
rev_id -- If set, use this as the new revision id.
148
Useful for test or import commands that need to tightly
149
control what revisions are assigned. If you duplicate
150
a revision id that exists elsewhere it is your own fault.
151
If null (default), a time/random revision id is generated.
153
allow_pointless -- If true (default), commit even if nothing
154
has changed and no merges are recorded.
156
self.any_changes = False
159
self.weave_store = branch.weave_store
161
self.specific_files = specific_files
162
self.allow_pointless = allow_pointless
164
if timestamp is None:
165
self.timestamp = time.time()
167
self.timestamp = long(timestamp)
119
169
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()
170
self.rev_id = _gen_revision_id(self.branch, self.timestamp)
174
if committer is None:
175
self.committer = username(self.branch)
177
assert isinstance(committer, basestring), type(committer)
178
self.committer = committer
181
self.timezone = local_time_offset()
183
self.timezone = int(timezone)
185
assert isinstance(message, basestring), type(message)
186
self.message = message
188
self.branch.lock_write()
190
self.work_tree = self.branch.working_tree()
191
self.work_inv = self.work_tree.inventory
192
self.basis_tree = self.branch.basis_tree()
193
self.basis_inv = self.basis_tree.inventory
195
self._gather_parents()
197
self._remove_deleted()
198
self.new_inv = Inventory()
200
self._report_deletes()
202
if not (self.allow_pointless
203
or len(self.parents) != 1
204
or self.new_inv != self.basis_inv):
205
raise PointlessCommit()
207
self._record_inventory()
208
self._record_ancestry()
209
self._make_revision()
210
note('committed r%d {%s}', (self.branch.revno() + 1),
212
self.branch.append_revision(self.rev_id)
213
self.branch.set_pending_merges([])
219
def _record_inventory(self):
220
"""Store the inventory for the new revision."""
221
inv_text = serializer_v5.write_inventory_to_string(self.new_inv)
222
self.inv_sha1 = sha_string(inv_text)
223
self.weave_store.add_text(INVENTORY_FILEID, self.rev_id,
224
split_lines(inv_text), self.parents)
227
def _record_ancestry(self):
228
"""Append merged revision ancestry to the ancestry file."""
229
w = self.weave_store.get_weave_or_empty(ANCESTRY_FILEID)
231
lines = w.get(w.lookup(self.parents[0]))
234
lines.append(self.rev_id + '\n')
235
parent_idxs = map(w.lookup, self.parents)
236
w.add(self.rev_id, parent_idxs, lines)
237
self.weave_store.put_weave(ANCESTRY_FILEID, w)
240
def _gather_parents(self):
241
pending_merges = self.branch.pending_merges()
243
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)
245
self.parents.append(precursor_id)
246
self.parents += pending_merges
247
for parent_id in self.parents:
248
if not self.branch.has_revision(parent_id):
249
warning("can't commit a merge from an absent parent")
250
raise HistoryMissing(self.branch, 'revision', parent_id)
251
self.parent_trees = map(self.branch.revision_tree, self.parents)
254
def _make_revision(self):
255
"""Record a new revision object for this commit."""
256
self.rev = Revision(timestamp=self.timestamp,
257
timezone=self.timezone,
258
committer=self.committer,
259
message=self.message,
260
inventory_sha1=self.inv_sha1,
261
revision_id=self.rev_id)
262
self.rev.parents = map(RevisionReference, self.parents)
264
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):
266
self.branch.revision_store.add(rev_tmp, self.rev_id)
267
mutter('new revision_id is {%s}', self.rev_id)
270
def _remove_deleted(self):
271
"""Remove deleted files from the working inventories.
273
This is done prior to taking the working inventory as the
274
basis for the new committed inventory.
276
This returns true if any files
277
*that existed in the basis inventory* were deleted.
278
Files that were added and deleted
279
in the working copy don't matter.
281
specific = self.specific_files
283
for path, ie in self.work_inv.iter_entries():
284
if specific and not is_inside_any(specific, path):
286
if not self.work_tree.has_filename(path):
287
note('missing %s', path)
288
deleted_ids.append(ie.file_id)
290
for file_id in deleted_ids:
291
del self.work_inv[file_id]
292
self.branch._write_inventory(self.work_inv)
295
def _find_file_parents(self, file_id):
296
"""Return the text versions and hashes for all file parents.
298
Returned as a map from text version to text sha1.
300
This is a set containing the file versions in all parents
301
revisions containing the file. If the file is new, the set
304
for tree in self.parent_trees:
305
if file_id in tree.inventory:
306
ie = tree.inventory[file_id]
307
assert ie.kind == 'file'
308
assert ie.file_id == file_id
309
if ie.text_version in r:
310
assert r[ie.text_version] == ie.text_sha1
312
r[ie.text_version] = ie.text_sha1
316
def _store_files(self):
317
"""Store new texts of modified/added files.
319
This is called with new_inv set to a copy of the working
320
inventory, with deleted/removed files already cut out. So
321
this code only needs to deal with setting text versions, and
322
possibly recording new file texts."""
323
for path, new_ie in self.work_inv.iter_entries():
324
file_id = new_ie.file_id
325
mutter('check %s {%s}', path, new_ie.file_id)
326
if self.specific_files:
327
if not is_inside_any(self.specific_files, path):
328
mutter('%s not selected for commit', path)
329
self._carry_file(file_id)
331
if new_ie.kind != 'file':
332
self._commit_nonfile(file_id)
334
file_parents = self._find_file_parents(file_id)
335
wc_sha1 = self.work_tree.get_file_sha1(file_id)
336
if (len(file_parents) == 1
337
and file_parents.values()[0] == wc_sha1):
338
# not changed or merged
339
self._carry_file(file_id)
342
mutter('parents of %s are %r', path, file_parents)
344
# file is either new, or a file merge; need to record
346
if len(file_parents) > 1:
347
note('merged %s', path)
348
elif len(file_parents) == 0:
349
note('added %s', path)
351
note('modified %s', path)
352
self._commit_file(new_ie, file_id, file_parents)
355
def _commit_nonfile(self, file_id):
356
self.new_inv.add(self.work_inv[file_id].copy())
359
def _carry_file(self, file_id):
360
"""Keep a file in the same state as in the basis."""
361
if self.basis_inv.has_id(file_id):
362
self.new_inv.add(self.basis_inv[file_id].copy())
365
def _report_deletes(self):
366
for file_id in self.basis_inv:
367
if file_id not in self.new_inv:
368
note('deleted %s', self.basis_inv.id2path(file_id))
371
def _commit_file(self, new_ie, file_id, file_parents):
372
mutter('store new text for {%s} in revision {%s}',
373
file_id, self.rev_id)
374
new_lines = self.work_tree.get_file(file_id).readlines()
375
self._add_text_to_weave(file_id, new_lines, file_parents)
376
new_ie.text_version = self.rev_id
377
new_ie.text_sha1 = sha_strings(new_lines)
378
new_ie.text_size = sum(map(len, new_lines))
379
self.new_inv.add(new_ie)
382
def _add_text_to_weave(self, file_id, new_lines, parents):
383
if file_id.startswith('__'):
384
raise ValueError('illegal file-id %r for text file' % file_id)
385
self.weave_store.add_text(file_id, self.rev_id, new_lines, parents)
388
def _gen_revision_id(branch, when):
190
389
"""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))
390
s = '%s-%s-' % (user_email(branch), compact_date(when))
195
391
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
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