« back to all changes in this revision
Viewing changes to bzrlib/commit.py
- browse files at revision 1220
- compare with another revision
- download diff
- download tarball
- view history from revision 1220
- Committer: Martin Pool
- Date: 2005-09-13 01:40:45 UTC
- Revision ID: mbp@sourcefrog.net-20050913014045-49f65739c33df949
- v5 xml no longer stores redundant inventory_id
- files added:
- bzrlib/builtins.py
- files removed:
- bzrlib/plugins/checkperms
- files renamed:
- bzrlib/newinventory.py => contrib/newinventory.py
- files modified:
- .bzrignore
added
removed
bzrlib/commit.py
15
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
16
16
17
17
18
19
# FIXME: "bzr commit doc/format" commits doc/format.txt!
20
21
def commit(branch, message,
22
timestamp=None,
23
timezone=None,
24
committer=None,
25
verbose=True,
26
specific_files=None,
27
rev_id=None,
28
allow_pointless=True):
29
"""Commit working copy as a new revision.
30
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.
34
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.
44
45
This raises PointlessCommit if there are no changes, no new merges,
46
and allow_pointless is false.
47
48
timestamp -- if not None, seconds-since-epoch for a
49
postdated/predated commit.
50
51
specific_files
52
If true, commit only those files.
53
54
rev_id
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.
60
"""
61
62
import time, tempfile
63
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
70
71
branch.lock_write()
72
73
try:
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
80
# working inventory.
81
82
work_tree = branch.working_tree()
83
work_inv = work_tree.inventory
84
basis = branch.basis_tree()
85
basis_inv = basis.inventory
86
87
if verbose:
88
# note('looking for changes...')
89
# print 'looking for changes...'
90
# disabled; should be done at a higher level
91
pass
92
93
pending_merges = branch.pending_merges()
94
95
missing_ids, new_inv, any_changes = \
96
_gather_commit(branch,
97
work_tree,
98
work_inv,
99
basis_inv,
100
specific_files,
101
verbose)
102
103
if not (any_changes or allow_pointless or pending_merges):
104
raise PointlessCommit()
105
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.
111
112
# have to do this later so we don't mess up the iterator.
113
# since parents may be removed before their children we
114
# have to test.
115
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]
120
121
if rev_id is None:
122
rev_id = _gen_revision_id(branch, time.time())
123
inv_id = rev_id
124
125
inv_tmp = tempfile.TemporaryFile()
126
pack_xml(new_inv, inv_tmp)
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
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
108
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.pending_merges = self.branch.pending_merges()
190
if self.pending_merges:
191
raise NotImplementedError("sorry, can't commit merges to the weave format yet")
192
193
if self.rev_id is None:
194
self.rev_id = _gen_revision_id(self.branch, time.time())
195
196
# todo: update hashcache
197
self.delta = compare_trees(self.basis_tree, self.work_tree,
198
specific_files=self.specific_files)
199
200
if not (self.delta.has_changed()
201
or self.allow_pointless
202
or self.pending_merges):
203
raise PointlessCommit()
204
205
self.new_inv = self.basis_inv.copy()
206
207
self.delta.show(sys.stdout)
208
209
self._remove_deleted()
210
self._store_files()
211
212
self.branch._write_inventory(self.work_inv)
213
self._record_inventory()
214
215
self._make_revision()
216
note('committted r%d {%s}', (self.branch.revno() + 1),
217
self.rev_id)
218
self.branch.append_revision(self.rev_id)
219
self.branch.set_pending_merges([])
220
finally:
221
self.branch.unlock()
222
223
224
def _record_inventory(self):
225
inv_tmp = StringIO()
226
serializer_v5.write_inventory(self.new_inv, inv_tmp)
227
self.inv_sha1 = sha_string(inv_tmp.getvalue())
127
228
inv_tmp.seek(0)
128
branch.inventory_store.add(inv_tmp, inv_id)
129
mutter('new inventory_id is {%s}' % inv_id)
130
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)
135
136
branch._write_inventory(work_inv)
137
138
if timestamp == None:
139
timestamp = time.time()
140
141
if committer == None:
142
committer = username(branch)
143
144
if timezone == None:
145
timezone = local_time_offset()
146
147
mutter("building commit log message")
148
rev = Revision(timestamp=timestamp,
149
timezone=timezone,
150
committer=committer,
151
message = message,
152
inventory_id=inv_id,
153
inventory_sha1=inv_sha1,
154
revision_id=rev_id)
155
156
rev.parents = []
157
precursor_id = branch.last_patch()
229
self.branch.inventory_store.add(inv_tmp, self.rev_id)
230
231
232
def _make_revision(self):
233
"""Record a new revision object for this commit."""
234
self.rev = Revision(timestamp=self.timestamp,
235
timezone=self.timezone,
236
committer=self.committer,
237
message=self.message,
238
inventory_sha1=self.inv_sha1,
239
revision_id=self.rev_id)
240
241
self.rev.parents = []
242
precursor_id = self.branch.last_patch()
158
243
if precursor_id:
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))
244
self.rev.parents.append(RevisionReference(precursor_id))
245
for merge_rev in self.pending_merges:
246
rev.parents.append(RevisionReference(merge_rev))
163
247
164
248
rev_tmp = tempfile.TemporaryFile()
165
pack_xml(rev, rev_tmp)
249
serializer_v5.write_revision(self.rev, rev_tmp)
166
250
rev_tmp.seek(0)
167
branch.revision_store.add(rev_tmp, rev_id)
168
mutter("new revision_id is {%s}" % rev_id)
169
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
172
## matter.
173
174
## TODO: Read back the just-generated changeset, and make sure it
175
## applies and recreates the right state.
176
177
## TODO: Also calculate and store the inventory SHA1
178
mutter("committing patch r%d" % (branch.revno() + 1))
179
180
branch.append_revision(rev_id)
181
182
branch.set_pending_merges([])
183
184
if verbose:
185
# disabled; should go through logging
186
# note("commited r%d" % branch.revno())
187
# print ("commited r%d" % branch.revno())
188
pass
189
finally:
190
branch.unlock()
191
251
self.branch.revision_store.add(rev_tmp, self.rev_id)
252
mutter('new revision_id is {%s}', self.rev_id)
253
254
255
def _remove_deleted(self):
256
"""Remove deleted files from the working and stored inventories."""
257
for path, id, kind in self.delta.removed:
258
if self.work_inv.has_id(id):
259
del self.work_inv[id]
260
if self.new_inv.has_id(id):
261
del self.new_inv[id]
262
263
264
265
def _store_files(self):
266
"""Store new texts of modified/added files."""
267
# We must make sure that directories are added before anything
268
# inside them is added. the files within the delta report are
269
# sorted by path so we know the directory will come before its
270
# contents.
271
for path, file_id, kind in self.delta.added:
272
if kind != 'file':
273
ie = self.work_inv[file_id].copy()
274
self.new_inv.add(ie)
275
else:
276
self._store_file_text(file_id)
277
278
for path, file_id, kind in self.delta.modified:
279
if kind != 'file':
280
continue
281
self._store_file_text(file_id)
282
283
for old_path, new_path, file_id, kind, text_modified in self.delta.renamed:
284
if kind != 'file':
285
continue
286
if not text_modified:
287
continue
288
self._store_file_text(file_id)
289
290
291
def _store_file_text(self, file_id):
292
"""Store updated text for one modified or added file."""
293
note('store new text for {%s} in revision {%s}',
294
file_id, self.rev_id)
295
new_lines = self.work_tree.get_file(file_id).readlines()
296
if file_id in self.new_inv: # was in basis inventory
297
ie = self.new_inv[file_id]
298
assert ie.file_id == file_id
299
assert file_id in self.basis_inv
300
assert self.basis_inv[file_id].kind == 'file'
301
old_version = self.basis_inv[file_id].text_version
302
file_parents = [old_version]
303
else: # new in this revision
304
ie = self.work_inv[file_id].copy()
305
self.new_inv.add(ie)
306
assert file_id not in self.basis_inv
307
file_parents = []
308
assert ie.kind == 'file'
309
self._add_text_to_weave(file_id, new_lines, file_parents)
310
# make a new inventory entry for this file, using whatever
311
# it had in the working copy, plus details on the new text
312
ie.text_sha1 = _sha_strings(new_lines)
313
ie.text_size = sum(map(len, new_lines))
314
ie.text_version = self.rev_id
315
ie.entry_version = self.rev_id
316
317
318
def _add_text_to_weave(self, file_id, new_lines, parents):
319
weave_fn = self.branch.controlfilename(['weaves', file_id+'.weave'])
320
if os.path.exists(weave_fn):
321
w = read_weave(file(weave_fn, 'rb'))
322
else:
323
w = Weave()
324
# XXX: Should set the appropriate parents by looking for this file_id
325
# in all revision parents
326
parent_idxs = map(w.lookup, parents)
327
w.add(self.rev_id, parent_idxs, new_lines)
328
af = AtomicFile(weave_fn)
329
try:
330
write_weave_v5(w, af)
331
af.commit()
332
finally:
333
af.close()
192
334
193
335
194
336
def _gen_revision_id(branch, when):
195
337
"""Return new revision-id."""
196
from binascii import hexlify
197
from bzrlib.osutils import rand_bytes, compact_date, user_email
198
199
338
s = '%s-%s-' % (user_email(branch), compact_date(when))
200
339
s += hexlify(rand_bytes(8))
201
340
return s
202
341
203
342
204
def _gather_commit(branch, work_tree, work_inv, basis_inv, specific_files,
205
verbose):
206
"""Build inventory preparatory to commit.
207
208
Returns missing_ids, new_inv, any_changes.
209
210
This adds any changed files into the text store, and sets their
211
test-id, sha and size in the returned inventory appropriately.
212
213
missing_ids
214
Modified to hold a list of files that have been deleted from
215
the working directory; these should be removed from the
216
working inventory.
217
"""
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
221
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
226
227
any_changes = False
228
inv = Inventory(work_inv.root.file_id)
229
missing_ids = []
230
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).
234
235
p = branch.abspath(path)
236
file_id = entry.file_id
237
mutter('commit prep file %s, id %r ' % (p, file_id))
238
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())
244
else:
245
# omit this from committed inventory
246
pass
247
continue
248
249
if not work_tree.has_id(file_id):
250
if verbose:
251
print('deleted %s%s' % (path, kind_marker(entry.kind)))
252
any_changes = True
253
mutter(" file is missing, removing from inventory")
254
missing_ids.append(file_id)
255
continue
256
257
# this is present in the new inventory; may be new, modified or
258
# unchanged.
259
old_ie = basis_inv.has_id(file_id) and basis_inv[file_id]
260
261
entry = entry.copy()
262
inv.add(entry)
263
264
if old_ie:
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))
269
270
if entry.kind == 'directory':
271
if not isdir(p):
272
raise BzrError("%s is entered as directory but not a directory"
273
% quotefn(p))
274
elif entry.kind == 'file':
275
if not isfile(p):
276
raise BzrError("%s is entered as file but is not a file" % quotefn(p))
277
278
new_sha1 = work_tree.get_file_sha1(file_id)
279
280
if (old_ie
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}' %
287
entry.text_id)
288
else:
289
content = file(p, 'rb').read()
290
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)
295
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)
299
300
if verbose:
301
marked = path + kind_marker(entry.kind)
302
if not old_ie:
303
print 'added', marked
304
any_changes = True
305
elif old_ie == entry:
306
pass # unchanged
307
elif (old_ie.name == entry.name
308
and old_ie.parent_id == entry.parent_id):
309
print 'modified', marked
310
any_changes = True
311
else:
312
print 'renamed', marked
313
any_changes = True
314
315
return missing_ids, inv, any_changes
316
317
343
def _sha_strings(strings):
344
"""Return the sha-1 of concatenation of strings"""
345
s = sha.new()
346
map(s.update, strings)
347
return s.hexdigest()
Loggerhead is a web-based interface for Breezy
Version: 2.0.1