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]
120
rev_id = _gen_revision_id(time.time())
123
inv_tmp = tempfile.TemporaryFile()
124
pack_xml(new_inv, inv_tmp)
18
## XXX: Can we do any better about making interrupted commits change
21
## XXX: If we merged two versions of a file then we still need to
22
## create a new version representing that merge, even if it didn't
23
## change from the parent.
25
## TODO: Read back the just-generated changeset, and make sure it
26
## applies and recreates the right state.
29
## This is not quite safe if the working copy changes during the
30
## commit; for the moment that is simply not allowed. A better
31
## approach is to make a temporary copy of the files before
32
## computing their hashes, and then add those hashes in turn to
33
## the inventory. This should mean at least that there are no
34
## broken hash pointers. There is no way we can get a snapshot
35
## of the whole directory at an instant. This would also have to
36
## be robust against files disappearing, moving, etc. So the
37
## whole thing is a bit hard.
39
## The newly committed revision is going to have a shape corresponding
40
## to that of the working inventory. Files that are not in the
41
## working tree and that were in the predecessor are reported as
42
## removed -- this can include files that were either removed from the
43
## inventory or deleted in the working tree. If they were only
44
## deleted from disk, they are removed from the working inventory.
46
## We then consider the remaining entries, which will be in the new
47
## version. Directory entries are simply copied across. File entries
48
## must be checked to see if a new version of the file should be
49
## recorded. For each parent revision inventory, we check to see what
50
## version of the file was present. If the file was present in at
51
## least one tree, and if it was the same version in all the trees,
52
## then we can just refer to that version. Otherwise, a new version
53
## representing the merger of the file versions must be added.
65
from binascii import hexlify
66
from cStringIO import StringIO
68
from bzrlib.osutils import (local_time_offset, username,
69
rand_bytes, compact_date, user_email,
70
kind_marker, is_inside_any, quotefn,
71
sha_string, sha_strings, sha_file, isdir, isfile)
72
from bzrlib.branch import gen_file_id, INVENTORY_FILEID, ANCESTRY_FILEID
73
from bzrlib.errors import BzrError, PointlessCommit
74
from bzrlib.revision import Revision, RevisionReference
75
from bzrlib.trace import mutter, note
76
from bzrlib.xml5 import serializer_v5
77
from bzrlib.inventory import Inventory
78
from bzrlib.delta import compare_trees
79
from bzrlib.weave import Weave
80
from bzrlib.weavefile import read_weave, write_weave_v5
81
from bzrlib.atomicfile import AtomicFile
84
def commit(*args, **kwargs):
85
"""Commit a new revision to a branch.
87
Function-style interface for convenience of old callers.
89
New code should use the Commit class instead.
91
Commit().commit(*args, **kwargs)
94
class NullCommitReporter(object):
95
"""I report on progress of a commit."""
96
def added(self, path):
99
def removed(self, path):
102
def renamed(self, old_path, new_path):
106
class ReportCommitToLog(NullCommitReporter):
107
def added(self, path):
108
note('added %s', path)
110
def removed(self, path):
111
note('removed %s', path)
113
def renamed(self, old_path, new_path):
114
note('renamed %s => %s', old_path, new_path)
117
class Commit(object):
118
"""Task of committing a new revision.
120
This is a MethodObject: it accumulates state as the commit is
121
prepared, and then it is discarded. It doesn't represent
122
historical revisions, just the act of recording a new one.
125
Modified to hold a list of files that have been deleted from
126
the working directory; these should be removed from the
131
if reporter is not None:
132
self.reporter = reporter
134
self.reporter = NullCommitReporter()
144
allow_pointless=True):
145
"""Commit working copy as a new revision.
147
The basic approach is to add all the file texts into the
148
store, then the inventory, then make a new revision pointing
149
to that inventory and store that.
151
This raises PointlessCommit if there are no changes, no new merges,
152
and allow_pointless is false.
154
timestamp -- if not None, seconds-since-epoch for a
155
postdated/predated commit.
158
If true, commit only those files.
161
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.
170
self.specific_files = specific_files
171
self.allow_pointless = allow_pointless
173
if timestamp is None:
174
self.timestamp = time.time()
176
self.timestamp = long(timestamp)
178
if committer is None:
179
self.committer = username(self.branch)
181
assert isinstance(committer, basestring), type(committer)
182
self.committer = committer
185
self.timezone = local_time_offset()
187
self.timezone = int(timezone)
189
assert isinstance(message, basestring), type(message)
190
self.message = message
192
self.branch.lock_write()
194
# First walk over the working inventory; and both update that
195
# and also build a new revision inventory. The revision
196
# inventory needs to hold the text-id, sha1 and size of the
197
# actual file versions committed in the revision. (These are
198
# not present in the working inventory.) We also need to
199
# detect missing/deleted files, and remove them from the
202
self.work_tree = self.branch.working_tree()
203
self.work_inv = self.work_tree.inventory
204
self.basis_tree = self.branch.basis_tree()
205
self.basis_inv = self.basis_tree.inventory
207
self._gather_parents()
209
if self.rev_id is None:
210
self.rev_id = _gen_revision_id(self.branch, time.time())
212
self._remove_deletions()
214
# TODO: update hashcache
215
self.delta = compare_trees(self.basis_tree, self.work_tree,
216
specific_files=self.specific_files)
218
if not (self.delta.has_changed()
219
or self.allow_pointless
220
or len(self.parents) != 1):
221
raise PointlessCommit()
223
self.new_inv = self.basis_inv.copy()
225
## FIXME: Don't write to stdout!
226
self.delta.show(sys.stdout)
228
self._remove_deleted()
231
self.branch._write_inventory(self.work_inv)
232
self._record_inventory()
233
self._record_ancestry()
235
self._make_revision()
236
note('committted r%d {%s}', (self.branch.revno() + 1),
238
self.branch.append_revision(self.rev_id)
239
self.branch.set_pending_merges([])
245
def _remove_deletions(self):
246
"""Remove deleted files from the working inventory."""
251
def _record_inventory(self):
252
"""Store the inventory for the new revision."""
254
serializer_v5.write_inventory(self.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()
256
self.inv_sha1 = sha_string(inv_tmp.getvalue())
257
inv_lines = inv_tmp.readlines()
258
self.branch.weave_store.add_text(INVENTORY_FILEID, self.rev_id,
259
inv_lines, self.parents)
262
def _record_ancestry(self):
263
"""Append merged revision ancestry to the ancestry file."""
264
if len(self.parents) > 1:
265
raise NotImplementedError("sorry, can't commit merges yet")
266
w = self.branch.weave_store.get_weave_or_empty(ANCESTRY_FILEID)
268
lines = w.get(w.lookup(self.parents[0]))
271
lines.append(self.rev_id + '\n')
272
parent_idxs = map(w.lookup, self.parents)
273
w.add(self.rev_id, parent_idxs, lines)
274
self.branch.weave_store.put_weave(ANCESTRY_FILEID, w)
277
def _gather_parents(self):
278
pending_merges = self.branch.pending_merges()
280
raise NotImplementedError("sorry, can't commit merges to the weave format yet")
282
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))
284
self.parents.append(precursor_id)
285
self.parents += pending_merges
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.parents = map(RevisionReference, self.parents)
162
297
rev_tmp = tempfile.TemporaryFile()
163
pack_xml(rev, rev_tmp)
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)
301
mutter('new revision_id is {%s}', self.rev_id)
304
def _remove_deleted(self):
305
"""Remove deleted files from the working and stored inventories."""
306
for path, id, kind in self.delta.removed:
307
if self.work_inv.has_id(id):
308
del self.work_inv[id]
309
if self.new_inv.has_id(id):
314
def _store_files(self):
315
"""Store new texts of modified/added files."""
316
# We must make sure that directories are added before anything
317
# inside them is added. the files within the delta report are
318
# sorted by path so we know the directory will come before its
320
for path, file_id, kind in self.delta.added:
322
ie = self.work_inv[file_id].copy()
325
self._store_file_text(file_id)
327
for path, file_id, kind in self.delta.modified:
330
self._store_file_text(file_id)
332
for old_path, new_path, file_id, kind, text_modified in self.delta.renamed:
335
if not text_modified:
337
self._store_file_text(file_id)
340
def _store_file_text(self, file_id):
341
"""Store updated text for one modified or added file."""
342
note('store new text for {%s} in revision {%s}',
343
file_id, self.rev_id)
344
new_lines = self.work_tree.get_file(file_id).readlines()
345
if file_id in self.new_inv: # was in basis inventory
346
ie = self.new_inv[file_id]
347
assert ie.file_id == file_id
348
assert file_id in self.basis_inv
349
assert self.basis_inv[file_id].kind == 'file'
350
old_version = self.basis_inv[file_id].text_version
351
file_parents = [old_version]
352
else: # new in this revision
353
ie = self.work_inv[file_id].copy()
355
assert file_id not in self.basis_inv
357
assert ie.kind == 'file'
358
self._add_text_to_weave(file_id, new_lines, file_parents)
359
# make a new inventory entry for this file, using whatever
360
# it had in the working copy, plus details on the new text
361
ie.text_sha1 = sha_strings(new_lines)
362
ie.text_size = sum(map(len, new_lines))
363
ie.text_version = self.rev_id
364
ie.entry_version = self.rev_id
367
def _add_text_to_weave(self, file_id, new_lines, parents):
368
if file_id.startswith('__'):
369
raise ValueError('illegal file-id %r for text file' % file_id)
370
self.branch.weave_store.add_text(file_id, self.rev_id, new_lines, parents)
373
def _gen_revision_id(branch, when):
190
374
"""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))
375
s = '%s-%s-' % (user_email(branch), compact_date(when))
195
376
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