« back to all changes in this revision
Viewing changes to bzrlib/commit.py
- browse files at revision 1185.3.3
- compare with another revision
- download diff
- download tarball
- view history from revision 1185.3.3
- Committer: Martin Pool
- Date: 2005-09-12 01:29:49 UTC
- mto: (1092.2.12) (974.1.76) (1185.8.2)
- mto: This revision was merged to the branch mainline in revision 1390.
- Revision ID: mbp@sourcefrog.net-20050912012949-73a539d3f2542173
- patch from mpe to automatically add parent directories
- files removed:
- bzrlib/selftest/test_ancestry.py
- files renamed:
- contrib/newinventory.py => bzrlib/newinventory.py
- files modified:
- NEWS
added
removed
bzrlib/commit.py
15
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
16
16
17
17
18
## XXX: Can we do any better about making interrupted commits change
19
## nothing?
20
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.
24
25
## TODO: Read back the just-generated changeset, and make sure it
26
## applies and recreates the right state.
27
28
29
30
31
import os
32
import sys
33
import time
34
import tempfile
35
import sha
36
37
from binascii import hexlify
38
from cStringIO import StringIO
39
40
from bzrlib.osutils import (local_time_offset, username,
41
rand_bytes, compact_date, user_email,
42
kind_marker, is_inside_any, quotefn,
43
sha_string, sha_file, isdir, isfile)
44
from bzrlib.branch import gen_file_id, INVENTORY_FILEID, ANCESTRY_FILEID
45
from bzrlib.errors import BzrError, PointlessCommit
46
from bzrlib.revision import Revision, RevisionReference
47
from bzrlib.trace import mutter, note
48
from bzrlib.xml5 import serializer_v5
49
from bzrlib.inventory import Inventory
50
from bzrlib.delta import compare_trees
51
from bzrlib.weave import Weave
52
from bzrlib.weavefile import read_weave, write_weave_v5
53
from bzrlib.atomicfile import AtomicFile
54
55
56
def commit(*args, **kwargs):
57
"""Commit a new revision to a branch.
58
59
Function-style interface for convenience of old callers.
60
61
New code should use the Commit class instead.
62
"""
63
Commit().commit(*args, **kwargs)
64
65
66
class NullCommitReporter(object):
67
"""I report on progress of a commit."""
68
def added(self, path):
69
pass
70
71
def removed(self, path):
72
pass
73
74
def renamed(self, old_path, new_path):
75
pass
76
77
78
class ReportCommitToLog(NullCommitReporter):
79
def added(self, path):
80
note('added %s', path)
81
82
def removed(self, path):
83
note('removed %s', path)
84
85
def renamed(self, old_path, new_path):
86
note('renamed %s => %s', old_path, new_path)
87
88
89
class Commit(object):
90
"""Task of committing a new revision.
91
92
This is a MethodObject: it accumulates state as the commit is
93
prepared, and then it is discarded. It doesn't represent
94
historical revisions, just the act of recording a new one.
95
96
missing_ids
97
Modified to hold a list of files that have been deleted from
98
the working directory; these should be removed from the
99
working inventory.
100
"""
101
def __init__(self,
102
reporter=None):
103
if reporter is not None:
104
self.reporter = reporter
105
else:
106
self.reporter = NullCommitReporter()
107
18
19
def commit(branch, message,
20
timestamp=None,
21
timezone=None,
22
committer=None,
23
verbose=True,
24
specific_files=None,
25
rev_id=None,
26
allow_pointless=True):
27
"""Commit working copy as a new revision.
28
29
The basic approach is to add all the file texts into the
30
store, then the inventory, then make a new revision pointing
31
to that inventory and store that.
32
33
This is not quite safe if the working copy changes during the
34
commit; for the moment that is simply not allowed. A better
35
approach is to make a temporary copy of the files before
36
computing their hashes, and then add those hashes in turn to
37
the inventory. This should mean at least that there are no
38
broken hash pointers. There is no way we can get a snapshot
39
of the whole directory at an instant. This would also have to
40
be robust against files disappearing, moving, etc. So the
41
whole thing is a bit hard.
42
43
This raises PointlessCommit if there are no changes, no new merges,
44
and allow_pointless is false.
45
46
timestamp -- if not None, seconds-since-epoch for a
47
postdated/predated commit.
48
49
specific_files
50
If true, commit only those files.
51
52
rev_id
53
If set, use this as the new revision id.
54
Useful for test or import commands that need to tightly
55
control what revisions are assigned. If you duplicate
56
a revision id that exists elsewhere it is your own fault.
57
If null (default), a time/random revision id is generated.
58
"""
59
60
import time, tempfile
61
62
from bzrlib.osutils import local_time_offset, username
63
from bzrlib.branch import gen_file_id
64
from bzrlib.errors import BzrError, PointlessCommit
65
from bzrlib.revision import Revision, RevisionReference
66
from bzrlib.trace import mutter, note
67
from bzrlib.xml import serializer_v4
68
69
branch.lock_write()
70
71
try:
72
# First walk over the working inventory; and both update that
73
# and also build a new revision inventory. The revision
74
# inventory needs to hold the text-id, sha1 and size of the
75
# actual file versions committed in the revision. (These are
76
# not present in the working inventory.) We also need to
77
# detect missing/deleted files, and remove them from the
78
# working inventory.
79
80
work_tree = branch.working_tree()
81
work_inv = work_tree.inventory
82
basis = branch.basis_tree()
83
basis_inv = basis.inventory
84
85
if verbose:
86
# note('looking for changes...')
87
# print 'looking for changes...'
88
# disabled; should be done at a higher level
89
pass
90
91
pending_merges = branch.pending_merges()
92
93
missing_ids, new_inv, any_changes = \
94
_gather_commit(branch,
95
work_tree,
96
work_inv,
97
basis_inv,
98
specific_files,
99
verbose)
100
101
if not (any_changes or allow_pointless or pending_merges):
102
raise PointlessCommit()
103
104
for file_id in missing_ids:
105
# Any files that have been deleted are now removed from the
106
# working inventory. Files that were not selected for commit
107
# are left as they were in the working inventory and ommitted
108
# from the revision inventory.
109
110
# have to do this later so we don't mess up the iterator.
111
# since parents may be removed before their children we
112
# have to test.
113
114
# FIXME: There's probably a better way to do this; perhaps
115
# the workingtree should know how to filter itbranch.
116
if work_inv.has_id(file_id):
117
del work_inv[file_id]
118
119
if rev_id is None:
120
rev_id = _gen_revision_id(branch, time.time())
121
inv_id = rev_id
122
123
inv_tmp = tempfile.TemporaryFile()
108
124
109
def commit(self,
110
branch, message,
111
timestamp=None,
112
timezone=None,
113
committer=None,
114
specific_files=None,
115
rev_id=None,
116
allow_pointless=True):
117
"""Commit working copy as a new revision.
118
119
The basic approach is to add all the file texts into the
120
store, then the inventory, then make a new revision pointing
121
to that inventory and store that.
122
123
This is not quite safe if the working copy changes during the
124
commit; for the moment that is simply not allowed. A better
125
approach is to make a temporary copy of the files before
126
computing their hashes, and then add those hashes in turn to
127
the inventory. This should mean at least that there are no
128
broken hash pointers. There is no way we can get a snapshot
129
of the whole directory at an instant. This would also have to
130
be robust against files disappearing, moving, etc. So the
131
whole thing is a bit hard.
132
133
This raises PointlessCommit if there are no changes, no new merges,
134
and allow_pointless is false.
135
136
timestamp -- if not None, seconds-since-epoch for a
137
postdated/predated commit.
138
139
specific_files
140
If true, commit only those files.
141
142
rev_id
143
If set, use this as the new revision id.
144
Useful for test or import commands that need to tightly
145
control what revisions are assigned. If you duplicate
146
a revision id that exists elsewhere it is your own fault.
147
If null (default), a time/random revision id is generated.
148
"""
149
150
self.branch = branch
151
self.branch.lock_write()
152
self.rev_id = rev_id
153
self.specific_files = specific_files
154
self.allow_pointless = allow_pointless
155
156
if timestamp is None:
157
self.timestamp = time.time()
158
else:
159
self.timestamp = long(timestamp)
160
161
if committer is None:
162
self.committer = username(self.branch)
163
else:
164
assert isinstance(committer, basestring), type(committer)
165
self.committer = committer
166
167
if timezone is None:
168
self.timezone = local_time_offset()
169
else:
170
self.timezone = int(timezone)
171
172
assert isinstance(message, basestring), type(message)
173
self.message = message
174
175
try:
176
# First walk over the working inventory; and both update that
177
# and also build a new revision inventory. The revision
178
# inventory needs to hold the text-id, sha1 and size of the
179
# actual file versions committed in the revision. (These are
180
# not present in the working inventory.) We also need to
181
# detect missing/deleted files, and remove them from the
182
# working inventory.
183
184
self.work_tree = self.branch.working_tree()
185
self.work_inv = self.work_tree.inventory
186
self.basis_tree = self.branch.basis_tree()
187
self.basis_inv = self.basis_tree.inventory
188
189
self._gather_parents()
190
191
if self.rev_id is None:
192
self.rev_id = _gen_revision_id(self.branch, time.time())
193
194
# TODO: update hashcache
195
self.delta = compare_trees(self.basis_tree, self.work_tree,
196
specific_files=self.specific_files)
197
198
if not (self.delta.has_changed()
199
or self.allow_pointless
200
or len(self.parents) != 1):
201
raise PointlessCommit()
202
203
self.new_inv = self.basis_inv.copy()
204
205
self.delta.show(sys.stdout)
206
207
self._remove_deleted()
208
self._store_files()
209
210
self.branch._write_inventory(self.work_inv)
211
self._record_inventory()
212
self._record_ancestry()
213
214
self._make_revision()
215
note('committted r%d {%s}', (self.branch.revno() + 1),
216
self.rev_id)
217
self.branch.append_revision(self.rev_id)
218
self.branch.set_pending_merges([])
219
finally:
220
self.branch.unlock()
221
222
223
def _record_inventory(self):
224
"""Store the inventory for the new revision."""
225
inv_tmp = StringIO()
226
serializer_v5.write_inventory(self.new_inv, inv_tmp)
125
serializer_v4.write_inventory(new_inv, inv_tmp)
227
126
inv_tmp.seek(0)
228
self.inv_sha1 = sha_string(inv_tmp.getvalue())
229
inv_lines = inv_tmp.readlines()
230
self.branch.weave_store.add_text(INVENTORY_FILEID, self.rev_id,
231
inv_lines, self.parents)
232
233
234
def _record_ancestry(self):
235
"""Append merged revision ancestry to the ancestry file."""
236
if len(self.parents) > 1:
237
raise NotImplementedError("sorry, can't commit merges yet")
238
w = self.branch.weave_store.get_weave_or_empty(ANCESTRY_FILEID)
239
if self.parents:
240
lines = w.get(w.lookup(self.parents[0]))
241
else:
242
lines = []
243
lines.append(self.rev_id + '\n')
244
parent_idxs = map(w.lookup, self.parents)
245
w.add(self.rev_id, parent_idxs, lines)
246
self.branch.weave_store.put_weave(ANCESTRY_FILEID, w)
247
248
249
def _gather_parents(self):
250
pending_merges = self.branch.pending_merges()
251
if pending_merges:
252
raise NotImplementedError("sorry, can't commit merges to the weave format yet")
253
self.parents = []
254
precursor_id = self.branch.last_patch()
127
branch.inventory_store.add(inv_tmp, inv_id)
128
mutter('new inventory_id is {%s}' % inv_id)
129
130
# We could also just sha hash the inv_tmp file
131
# however, in the case that branch.inventory_store.add()
132
# ever actually does anything special
133
inv_sha1 = branch.get_inventory_sha1(inv_id)
134
135
branch._write_inventory(work_inv)
136
137
if timestamp == None:
138
timestamp = time.time()
139
140
if committer == None:
141
committer = username(branch)
142
143
if timezone == None:
144
timezone = local_time_offset()
145
146
mutter("building commit log message")
147
rev = Revision(timestamp=timestamp,
148
timezone=timezone,
149
committer=committer,
150
message = message,
151
inventory_id=inv_id,
152
inventory_sha1=inv_sha1,
153
revision_id=rev_id)
154
155
rev.parents = []
156
precursor_id = branch.last_patch()
255
157
if precursor_id:
256
self.parents.append(precursor_id)
257
self.parents += pending_merges
258
259
260
def _make_revision(self):
261
"""Record a new revision object for this commit."""
262
self.rev = Revision(timestamp=self.timestamp,
263
timezone=self.timezone,
264
committer=self.committer,
265
message=self.message,
266
inventory_sha1=self.inv_sha1,
267
revision_id=self.rev_id)
268
self.rev.parents = map(RevisionReference, self.parents)
158
precursor_sha1 = branch.get_revision_sha1(precursor_id)
159
rev.parents.append(RevisionReference(precursor_id, precursor_sha1))
160
for merge_rev in pending_merges:
161
rev.parents.append(RevisionReference(merge_rev))
162
269
163
rev_tmp = tempfile.TemporaryFile()
270
serializer_v5.write_revision(self.rev, rev_tmp)
164
serializer_v4.write_revision(rev, rev_tmp)
271
165
rev_tmp.seek(0)
272
self.branch.revision_store.add(rev_tmp, self.rev_id)
273
mutter('new revision_id is {%s}', self.rev_id)
274
275
276
def _remove_deleted(self):
277
"""Remove deleted files from the working and stored inventories."""
278
for path, id, kind in self.delta.removed:
279
if self.work_inv.has_id(id):
280
del self.work_inv[id]
281
if self.new_inv.has_id(id):
282
del self.new_inv[id]
283
284
285
286
def _store_files(self):
287
"""Store new texts of modified/added files."""
288
# We must make sure that directories are added before anything
289
# inside them is added. the files within the delta report are
290
# sorted by path so we know the directory will come before its
291
# contents.
292
for path, file_id, kind in self.delta.added:
293
if kind != 'file':
294
ie = self.work_inv[file_id].copy()
295
self.new_inv.add(ie)
296
else:
297
self._store_file_text(file_id)
298
299
for path, file_id, kind in self.delta.modified:
300
if kind != 'file':
301
continue
302
self._store_file_text(file_id)
303
304
for old_path, new_path, file_id, kind, text_modified in self.delta.renamed:
305
if kind != 'file':
306
continue
307
if not text_modified:
308
continue
309
self._store_file_text(file_id)
310
311
312
def _store_file_text(self, file_id):
313
"""Store updated text for one modified or added file."""
314
note('store new text for {%s} in revision {%s}',
315
file_id, self.rev_id)
316
new_lines = self.work_tree.get_file(file_id).readlines()
317
if file_id in self.new_inv: # was in basis inventory
318
ie = self.new_inv[file_id]
319
assert ie.file_id == file_id
320
assert file_id in self.basis_inv
321
assert self.basis_inv[file_id].kind == 'file'
322
old_version = self.basis_inv[file_id].text_version
323
file_parents = [old_version]
324
else: # new in this revision
325
ie = self.work_inv[file_id].copy()
326
self.new_inv.add(ie)
327
assert file_id not in self.basis_inv
328
file_parents = []
329
assert ie.kind == 'file'
330
self._add_text_to_weave(file_id, new_lines, file_parents)
331
# make a new inventory entry for this file, using whatever
332
# it had in the working copy, plus details on the new text
333
ie.text_sha1 = _sha_strings(new_lines)
334
ie.text_size = sum(map(len, new_lines))
335
ie.text_version = self.rev_id
336
ie.entry_version = self.rev_id
337
338
339
def _add_text_to_weave(self, file_id, new_lines, parents):
340
if file_id.startswith('__'):
341
raise ValueError('illegal file-id %r for text file' % file_id)
342
self.branch.weave_store.add_text(file_id, self.rev_id, new_lines, parents)
166
branch.revision_store.add(rev_tmp, rev_id)
167
mutter("new revision_id is {%s}" % rev_id)
168
169
## XXX: Everything up to here can simply be orphaned if we abort
170
## the commit; it will leave junk files behind but that doesn't
171
## matter.
172
173
## TODO: Read back the just-generated changeset, and make sure it
174
## applies and recreates the right state.
175
176
## TODO: Also calculate and store the inventory SHA1
177
mutter("committing patch r%d" % (branch.revno() + 1))
178
179
branch.append_revision(rev_id)
180
181
branch.set_pending_merges([])
182
183
if verbose:
184
# disabled; should go through logging
185
# note("commited r%d" % branch.revno())
186
# print ("commited r%d" % branch.revno())
187
pass
188
finally:
189
branch.unlock()
190
343
191
344
192
345
193
def _gen_revision_id(branch, when):
346
194
"""Return new revision-id."""
195
from binascii import hexlify
196
from bzrlib.osutils import rand_bytes, compact_date, user_email
197
347
198
s = '%s-%s-' % (user_email(branch), compact_date(when))
348
199
s += hexlify(rand_bytes(8))
349
200
return s
350
201
351
202
352
def _sha_strings(strings):
353
"""Return the sha-1 of concatenation of strings"""
354
s = sha.new()
355
map(s.update, strings)
356
return s.hexdigest()
203
def _gather_commit(branch, work_tree, work_inv, basis_inv, specific_files,
204
verbose):
205
"""Build inventory preparatory to commit.
206
207
Returns missing_ids, new_inv, any_changes.
208
209
This adds any changed files into the text store, and sets their
210
test-id, sha and size in the returned inventory appropriately.
211
212
missing_ids
213
Modified to hold a list of files that have been deleted from
214
the working directory; these should be removed from the
215
working inventory.
216
"""
217
from bzrlib.inventory import Inventory
218
from bzrlib.osutils import isdir, isfile, sha_string, quotefn, \
219
local_time_offset, username, kind_marker, is_inside_any
220
221
from bzrlib.branch import gen_file_id
222
from bzrlib.errors import BzrError
223
from bzrlib.revision import Revision
224
from bzrlib.trace import mutter, note
225
226
any_changes = False
227
inv = Inventory(work_inv.root.file_id)
228
missing_ids = []
229
230
for path, entry in work_inv.iter_entries():
231
## TODO: Check that the file kind has not changed from the previous
232
## revision of this file (if any).
233
234
p = branch.abspath(path)
235
file_id = entry.file_id
236
mutter('commit prep file %s, id %r ' % (p, file_id))
237
238
if specific_files and not is_inside_any(specific_files, path):
239
mutter(' skipping file excluded from commit')
240
if basis_inv.has_id(file_id):
241
# carry over with previous state
242
inv.add(basis_inv[file_id].copy())
243
else:
244
# omit this from committed inventory
245
pass
246
continue
247
248
if not work_tree.has_id(file_id):
249
if verbose:
250
print('deleted %s%s' % (path, kind_marker(entry.kind)))
251
any_changes = True
252
mutter(" file is missing, removing from inventory")
253
missing_ids.append(file_id)
254
continue
255
256
# this is present in the new inventory; may be new, modified or
257
# unchanged.
258
old_ie = basis_inv.has_id(file_id) and basis_inv[file_id]
259
260
entry = entry.copy()
261
inv.add(entry)
262
263
if old_ie:
264
old_kind = old_ie.kind
265
if old_kind != entry.kind:
266
raise BzrError("entry %r changed kind from %r to %r"
267
% (file_id, old_kind, entry.kind))
268
269
if entry.kind == 'directory':
270
if not isdir(p):
271
raise BzrError("%s is entered as directory but not a directory"
272
% quotefn(p))
273
elif entry.kind == 'file':
274
if not isfile(p):
275
raise BzrError("%s is entered as file but is not a file" % quotefn(p))
276
277
new_sha1 = work_tree.get_file_sha1(file_id)
278
279
if (old_ie
280
and old_ie.text_sha1 == new_sha1):
281
## assert content == basis.get_file(file_id).read()
282
entry.text_id = old_ie.text_id
283
entry.text_sha1 = new_sha1
284
entry.text_size = old_ie.text_size
285
mutter(' unchanged from previous text_id {%s}' %
286
entry.text_id)
287
else:
288
content = file(p, 'rb').read()
289
290
# calculate the sha again, just in case the file contents
291
# changed since we updated the cache
292
entry.text_sha1 = sha_string(content)
293
entry.text_size = len(content)
294
295
entry.text_id = gen_file_id(entry.name)
296
branch.text_store.add(content, entry.text_id)
297
mutter(' stored with text_id {%s}' % entry.text_id)
298
299
if verbose:
300
marked = path + kind_marker(entry.kind)
301
if not old_ie:
302
print 'added', marked
303
any_changes = True
304
elif old_ie == entry:
305
pass # unchanged
306
elif (old_ie.name == entry.name
307
and old_ie.parent_id == entry.parent_id):
308
print 'modified', marked
309
any_changes = True
310
else:
311
print 'renamed', marked
312
any_changes = True
313
314
return missing_ids, inv, any_changes
315
316
Loggerhead is a web-based interface for Breezy
Version: 2.0.1