1 # (Be in -*- python -*- mode.)
3 # ====================================================================
4 # Copyright (c) 2000-2009 CollabNet. All rights reserved.
6 # This software is licensed as described in the file COPYING, which
7 # you should have received as part of this distribution. The terms
8 # are also available at http://subversion.tigris.org/license-1.html.
9 # If newer versions of this license are posted there, you may use a
10 # newer version instead, at your option.
12 # This software consists of voluntary contributions made by many
13 # individuals. For exact contribution history, see the revision
14 # history and logs, available at http://cvs2svn.tigris.org/.
15 # ====================================================================
17 """This module contains database facilities used by cvs2svn."""
21 from hashlib
import md5
23 from md5
import new
as md5
26 from cvs2svn_lib
import config
27 from cvs2svn_lib
.common
import FatalError
28 from cvs2svn_lib
.common
import InternalError
29 from cvs2svn_lib
.common
import path_split
30 from cvs2svn_lib
.context
import Ctx
31 from cvs2svn_lib
.cvs_path
import CVSDirectory
32 from cvs2svn_lib
.cvs_path
import CVSFile
33 from cvs2svn_lib
.svn_repository_delegate
import SVNRepositoryDelegate
34 from cvs2svn_lib
.apple_single_filter
import get_maybe_apple_single
37 # Things that can happen to a file.
42 # A mapping from the value of the svn:eol-style property to the EOL
43 # string that should appear in a dumpfile:
44 EOL_STYLE_REPLACEMENTS
= {
52 def canonicalize_eol(text
, eol
):
53 """Replace any end-of-line sequences in TEXT with the string EOL."""
55 text
= text
.replace('\r\n', '\n')
56 text
= text
.replace('\r', '\n')
58 text
= text
.replace('\n', eol
)
62 class DumpfileDelegate(SVNRepositoryDelegate
):
63 """Create a Subversion dumpfile."""
65 def __init__(self
, revision_reader
, dumpfile_path
):
66 """Return a new DumpfileDelegate instance, attached to a dumpfile
67 DUMPFILE_PATH, using Ctx().cvs_filename_decoder()."""
69 self
._revision
_reader
= revision_reader
70 self
.dumpfile_path
= dumpfile_path
72 self
.dumpfile
= open(self
.dumpfile_path
, 'wb')
73 self
._write
_dumpfile
_header
(self
.dumpfile
)
75 # A set of the basic project infrastructure project directories
76 # that have been created so far, as SVN paths. (The root
77 # directory is considered to be present at initialization.) This
78 # includes all of the LOD paths, and all of their parent
80 self
._basic
_directories
= set([''])
82 def _write_dumpfile_header(self
, dumpfile
):
83 # Initialize the dumpfile with the standard headers.
85 # Since the CVS repository doesn't have a UUID, and the Subversion
86 # repository will be created with one anyway, we don't specify a
87 # UUID in the dumpflie
88 dumpfile
.write('SVN-fs-dump-format-version: 2\n\n')
90 def _utf8_path(self
, path
):
91 """Return a copy of PATH encoded in UTF-8."""
93 # Convert each path component separately (as they may each use
94 # different encodings).
97 Ctx().cvs_filename_decoder(piece
).encode('utf8')
98 for piece
in path
.split('/')
102 "Unable to convert a path '%s' to internal encoding.\n"
103 "Consider rerunning with one or more '--encoding' parameters or\n"
104 "with '--fallback-encoding'."
108 def _string_for_props(properties
):
109 """Return PROPERTIES in the form needed for the dumpfile."""
112 for (k
, v
) in sorted(properties
.iteritems()):
113 if k
.startswith('_'):
114 # Such properties are for internal use only.
117 # None indicates that the property should be left unset.
120 prop_strings
.append('K %d\n%s\nV %d\n%s\n' % (len(k
), k
, len(v
), v
))
122 prop_strings
.append('PROPS-END\n')
124 return ''.join(prop_strings
)
126 def start_commit(self
, revnum
, revprops
):
127 """Emit the start of SVN_COMMIT (an SVNCommit)."""
129 self
.revision
= revnum
131 # The start of a new commit typically looks like this:
134 # Prop-content-length: 129
135 # Content-length: 129
140 # Log message for revision 1.
148 # 2003-04-22T22:57:58.132837Z
151 # Notice that the length headers count everything -- not just the
152 # length of the data but also the lengths of the lengths, including
153 # the 'K ' or 'V ' prefixes.
155 # The reason there are both Prop-content-length and Content-length
156 # is that the former includes just props, while the latter includes
157 # everything. That's the generic header form for any entity in a
158 # dumpfile. But since revisions only have props, the two lengths
159 # are always the same for revisions.
161 # Calculate the output needed for the property definitions.
162 all_prop_strings
= self
._string
_for
_props
(revprops
)
163 total_len
= len(all_prop_strings
)
165 # Print the revision header and revprops
167 'Revision-number: %d\n'
168 'Prop-content-length: %d\n'
169 'Content-length: %d\n'
173 % (self
.revision
, total_len
, total_len
, all_prop_strings
)
176 def end_commit(self
):
179 def _make_any_dir(self
, path
):
180 """Emit the creation of directory PATH."""
188 % self
._utf
8_path
(path
)
191 def _register_basic_directory(self
, path
, create
):
192 """Register the creation of PATH if it is not already there.
194 Create any parent directories that do not already exist. If
195 CREATE is set, also create PATH if it doesn't already exist. This
196 method should only be used for the LOD paths and the directories
197 containing them, not for directories within an LOD path."""
199 if path
not in self
._basic
_directories
:
200 # Make sure that the parent directory is present:
201 self
._register
_basic
_directory
(path_split(path
)[0], True)
203 self
._make
_any
_dir
(path
)
204 self
._basic
_directories
.add(path
)
206 def initialize_project(self
, project
):
207 """Create any initial directories for the project.
209 The trunk, tags, and branches directories directories are created
210 the first time the project is seen. Be sure not to create parent
211 directories that already exist (e.g., because two directories
212 share part of their paths either within or across projects)."""
214 for path
in project
.get_initial_directories():
215 self
._register
_basic
_directory
(path
, True)
217 def initialize_lod(self
, lod
):
218 lod_path
= lod
.get_path()
220 self
._register
_basic
_directory
(lod_path
, True)
222 def mkdir(self
, lod
, cvs_directory
):
223 self
._make
_any
_dir
(lod
.get_path(cvs_directory
.cvs_path
))
225 def _add_or_change_path(self
, cvs_rev
, op
):
226 """Emit the addition or change corresponding to CVS_REV.
228 OP is either the constant OP_ADD or OP_CHANGE."""
230 assert op
in [OP_ADD
, OP_CHANGE
]
232 # The property handling here takes advantage of an undocumented
233 # but IMHO consistent feature of the Subversion dumpfile-loading
234 # code. When a node's properties aren't mentioned (that is, the
235 # "Prop-content-length:" header is absent, no properties are
236 # listed at all, and there is no "PROPS-END\n" line) then no
237 # change is made to the node's properties.
239 # This is consistent with the way dumpfiles behave w.r.t. text
240 # content changes, so I'm comfortable relying on it. If you
241 # commit a change to *just* the properties of some node that
242 # already has text contents from a previous revision, then in the
243 # dumpfile output for the prop change, no "Text-content-length:"
244 # nor "Text-content-md5:" header will be present, and the text of
245 # the file will not be given. But this does not cause the file's
246 # text to be erased! It simply remains unchanged.
248 # This works out great for cvs2svn, due to lucky coincidences:
250 # For files, we set most properties in the first revision and
251 # never change them. (The only exception is the 'cvs2svn:cvs-rev'
252 # property.) If 'cvs2svn:cvs-rev' is not being used, then there
253 # is no need to remember the full set of properties on a given
254 # file once we've set it.
256 # For directories, the only property we set is "svn:ignore", and
257 # while we may change it after the first revision, we always do so
258 # based on the contents of a ".cvsignore" file -- in other words,
259 # CVS is doing the remembering for us, so we still don't have to
260 # preserve the previous value of the property ourselves.
262 # Calculate the (sorted-by-name) property string and length, if any.
263 svn_props
= cvs_rev
.get_properties()
264 if cvs_rev
.properties_changed
:
265 prop_contents
= self
._string
_for
_props
(svn_props
)
266 props_header
= 'Prop-content-length: %d\n' % len(prop_contents
)
271 data
= self
._revision
_reader
.get_content(cvs_rev
)
273 # Convert all EOLs to LFs if neccessary
274 eol_style
= svn_props
.get('svn:eol-style', None)
276 eol
= EOL_STYLE_REPLACEMENTS
[eol_style
]
277 data
= canonicalize_eol(data
, eol
)
279 # treat .cvsignore as a directory property
280 dir_path
, basename
= path_split(cvs_rev
.get_svn_path())
281 if basename
== '.cvsignore':
282 ignore_vals
= generate_ignores(data
)
283 ignore_contents
= '\n'.join(ignore_vals
)
285 ignore_contents
+= '\n'
286 ignore_contents
= ('K 10\nsvn:ignore\nV %d\n%s\n' % \
287 (len(ignore_contents
), ignore_contents
))
288 ignore_contents
+= 'PROPS-END\n'
289 ignore_len
= len(ignore_contents
)
291 # write headers, then props
295 'Node-action: change\n'
296 'Prop-content-length: %d\n'
297 'Content-length: %d\n'
300 % (self
._utf
8_path
(dir_path
),
301 ignore_len
, ignore_len
, ignore_contents
)
303 if not Ctx().keep_cvsignore
:
310 '%s' # no property header if no props
311 % (self
._utf
8_path
(cvs_rev
.get_svn_path()), op
, props_header
)
314 pos
= self
.dumpfile
.tell()
316 content_header_fmt
= (
317 'Text-content-length: %16d\n'
318 'Text-content-md5: %32s\n'
319 'Content-length: %16d\n'
323 self
.dumpfile
.write(content_header_fmt
% (0, '', 0,))
326 self
.dumpfile
.write(prop_contents
)
328 # Insert the rev contents, calculating length and checksum.
330 checksum
.update(data
)
332 self
.dumpfile
.write(data
)
334 # Go back to overwrite the length and checksum headers with the
335 # correct values. The content length is the length of property
336 # data, text data, and any metadata around/inside around them:
337 self
.dumpfile
.seek(pos
, 0)
340 % (length
, checksum
.hexdigest(), length
+ len(prop_contents
),)
343 # Jump back to the end of the stream
344 self
.dumpfile
.seek(0, 2)
346 # This record is done (write two newlines -- one to terminate
347 # contents that weren't themselves newline-termination, one to
348 # provide a blank line for readability.
349 self
.dumpfile
.write('\n\n')
351 def add_path(self
, cvs_rev
):
352 """Emit the addition corresponding to CVS_REV, a CVSRevisionAdd."""
354 self
._add
_or
_change
_path
(cvs_rev
, OP_ADD
)
356 def change_path(self
, cvs_rev
):
357 """Emit the change corresponding to CVS_REV, a CVSRevisionChange."""
359 self
._add
_or
_change
_path
(cvs_rev
, OP_CHANGE
)
361 def delete_lod(self
, lod
):
362 """Emit the deletion of LOD."""
366 'Node-action: delete\n'
368 % (self
._utf
8_path
(lod
.get_path()),)
370 self
._basic
_directories
.remove(lod
.get_path())
372 def delete_path(self
, lod
, cvs_path
):
373 dir_path
, basename
= path_split(lod
.get_path(cvs_path
.get_cvs_path()))
374 if basename
== '.cvsignore':
375 # When a .cvsignore file is deleted, the directory's svn:ignore
376 # property needs to be deleted.
377 ignore_contents
= 'PROPS-END\n'
378 ignore_len
= len(ignore_contents
)
380 # write headers, then props
384 'Node-action: change\n'
385 'Prop-content-length: %d\n'
386 'Content-length: %d\n'
389 % (self
._utf
8_path
(dir_path
),
390 ignore_len
, ignore_len
, ignore_contents
)
392 if not Ctx().keep_cvsignore
:
397 'Node-action: delete\n'
399 % (self
._utf
8_path
(lod
.get_path(cvs_path
.cvs_path
)),)
402 def copy_lod(self
, src_lod
, dest_lod
, src_revnum
):
403 # Register the main LOD directory, and create parent directories
405 self
._register
_basic
_directory
(dest_lod
.get_path(), False)
411 'Node-copyfrom-rev: %d\n'
412 'Node-copyfrom-path: %s\n'
414 % (self
._utf
8_path
(dest_lod
.get_path()),
415 src_revnum
, self
._utf
8_path
(src_lod
.get_path()))
418 def copy_path(self
, cvs_path
, src_lod
, dest_lod
, src_revnum
):
419 if isinstance(cvs_path
, CVSFile
):
421 if cvs_path
.basename
== '.cvsignore':
422 # FIXME: Here we have to adjust the containing directory's
423 # svn:ignore property to reflect the addition of the
424 # .cvsignore file to the LOD! This is awkward because we
425 # don't have the contents of the .cvsignore file available.
426 if not Ctx().keep_cvsignore
:
428 elif isinstance(cvs_path
, CVSDirectory
):
431 raise InternalError()
437 'Node-copyfrom-rev: %d\n'
438 'Node-copyfrom-path: %s\n'
441 self
._utf
8_path
(dest_lod
.get_path(cvs_path
.cvs_path
)),
444 self
._utf
8_path
(src_lod
.get_path(cvs_path
.cvs_path
))
449 """Perform any cleanup necessary after all revisions have been
452 self
.dumpfile
.close()
455 def generate_ignores(raw_ignore_val
):
457 for ignore
in raw_ignore_val
.split():
458 # Reset the list if we encounter a '!'
459 # See http://cvsbook.red-bean.com/cvsbook.html#cvsignore
463 ignore_vals
.append(ignore
)