4 Theory and requirements
5 ------ --- ------------
7 There are two main problem converting a CVS repository to SVN:
9 - CVS does not record enough information to determine what actually
10 happened to a repository. For example, CVS does not record:
12 - Which file modifications were part of the same commit
14 - The timestamp of tag and branch creations
16 - Exactly which revision was the base of a branch (there is
17 ambiguity between x.y, x.y.2.0, x.y.4.0, etc.)
19 - When the default branch was changed (for example, from a vendor
20 branch back to trunk).
22 - The timestamps in a CVS archive are not reliable. It can easily
23 happen that timestamps are not even monotonic, and large errors (for
24 example due to a failing server clock battery) are not unusual.
26 The absolutely crucial, sine qua non requirement of a conversion is
27 that the dependency relationships within a file be honored, mainly:
29 - A revision depends on its predecessor
31 - A branch creation depends on the revision from which it branched,
32 and commits on the branch depend on the branch creation
34 - A tag creation depends on the revision being tagged
36 These dependencies are reliably defined in the CVS repository, and
37 they trump all others, so they are the scaffolding of the conversion.
39 Moreover, it is highly desirable that the timestamps of the SVN
40 commits be monotonically increasing.
42 Within these constraints we also want the results of the conversion to
43 resemble the history of the CVS repository as closely as possible.
44 For example, the set of file changes grouped together in an SVN commit
45 should be the same as the files changed within the corresponding CVS
46 commit, insofar as that can be achieved in a manner that is consistent
47 with the dependency requirements. And the SVN commit timestamps
48 should recreate the time of the CVS commit as far as possible without
49 violating the monotonicity requirement.
51 The basic idea of the conversion is this: create the largest
52 conceivable changesets, then split up changesets as necessary to break
53 any cycles in the graph of changeset dependencies. When all cycles
54 have been removed, then do a topological sort of the changesets (with
55 ambiguities resolved using CVS timestamps) to determine a
56 self-consistent changeset commit order.
58 The quality of the conversion (not in terms of correctness, but in
59 terms of minimizing the number of svn commits) is mostly determined by
60 the cleverness of the heuristics used to split up cycles. And all of
61 this has to be affordable, especially in terms of conversion time and
62 RAM usage, for even the largest CVS repositories.
68 A cvs2svn run consists of a number of passes. Each pass saves the
69 data it produces to files on disk, so that a) we don't hold huge
70 amounts of state in memory, and b) the conversion process is
73 The intermediate files are referred to here by the symbolic constants
74 holding their filenames in config.py.
77 CollectRevsPass (formerly called pass1)
80 The goal of this pass is to collect from the CVS files all of the data
81 that will be required for the conversion. If the --use-internal-co
82 option was used, this pass also collects the file delta data; for
83 -use-rcs or -use-cvs, the actual file contents are read again in
86 To collect this data, we walk over the repository, collecting data
87 about the RCS files into an instance of CollectData. Each RCS file is
88 processed with rcsparse.parse(), which invokes callbacks from an
89 instance of cvs2svn's _FileDataCollector class (which is a subclass of
92 While a file is being processed, all of the data for the file (except
93 for contents and log messages) is held in memory. When the file has
94 been read completely, its data is converted into an instance of
95 CVSFileItems, and this instance is manipulated a bit then pickled and
96 stored to CVS_ITEMS_STORE.
98 For each RCS file, the first thing the parser encounters is the
99 administrative header, including the head revision, the principal
100 branch, symbolic names, RCS comments, etc. The main thing that
101 happens here is that _FileDataCollector.define_tag() is invoked on
102 each symbolic name and its attached revision, so all the tags and
103 branches of this file get collected.
105 Next, the parser hits the revision summary section. That's the part
106 of the RCS file that looks like this:
109 date 2002.06.12.04.54.12; author captnmark; state Exp;
115 date 2002.05.28.18.02.11; author captnmark; state Exp;
121 For each revision summary, _FileDataCollector.define_revision() is
122 invoked, recording that revision's metadata in various variables of
123 the _FileDataCollector class instance.
125 Next, the parser encounters the *real* revision data, which has the
126 log messages and file contents. For each revision, it invokes
127 _FileDataCollector.set_revision_info(), which sets some more fields in
130 When the parser is done with the file, _ProjectDataCollector takes the
131 resulting CVSFileItems object and manipulates it to handle some CVS
134 - If the file had a vendor branch, make some adjustments to the
135 file dependency graph to reflect implicit dependencies related to
136 the vendor branch. Also delete the 1.1 revision in the usual
137 case that it doesn't contain any useful information.
139 - If the file was added on a branch rather than on trunk, then
140 delete the "dead" 1.1 revision on trunk in the usual case that it
141 doesn't contain any useful information.
143 - If the file was added on a branch after it already existed on
144 trunk, then recent versions of CVS add an extra "dead" revision
145 on the branch. Remove this revision in the usual case that it
146 doesn't contain any useful information, and sever the branch from
147 trunk (since the branch version is independent of the trunk
150 - If the conversion was started with the --trunk-only option, then
152 1. graft any non-trunk default branch revisions onto trunk
153 (because they affect the history of the default branch), and
155 2. delete all branches and tags and all remaining branch
158 Finally, the CVSFileItems instance is stored to a database and
159 statistics about how symbols were used in the file are recorded.
161 That's it -- the RCS file is done.
163 When every CVS file is done, CollectRevsPass is complete, and:
165 - The basic information about each project is stored to PROJECTS.
167 - The basic information about each file and directory (filename,
168 path, etc) is written as a pickled CVSPath instance to
171 - Information about each symbol seen, along with statistics like
172 how often it was used as a branch or tag, is written as a pickled
173 symbol_statistics._Stat object to SYMBOL_STATISTICS. This
174 includes the following information:
176 ID -- a unique positive identifying integer
178 NAME -- the symbol name
180 TAG_CREATE_COUNT -- the number of times the symbol was used
183 BRANCH_CREATE_COUNT -- the number of times the symbol was
186 BRANCH_COMMIT_COUNT -- the number of files in which there was
187 a commit on a branch with this name.
189 BRANCH_BLOCKERS -- the set of other symbols that ever
190 sprouted from a branch with this name. (A symbol cannot
191 be excluded from the conversion unless all of its
192 blockers are also excluded.)
194 POSSIBLE_PARENTS -- a count of in how many files each other
195 branch could have served as the symbol's source.
197 These data are used to look for inconsistencies in the use of
198 symbols under CVS and to decide which symbols can be excluded or
199 forced to be branches and/or tags. The POSSIBLE_PARENTS data is
200 used to pick the "optimum" parent from which the symbol should
201 sprout in as many files as possible.
203 For a multiproject conversion, distinct symbol records (and IDs)
204 are created for symbols in separate projects, even if they have
205 the same name. This is to prevent symbols in separate projects
206 from being filled at the same time.
208 - Information about each CVS event is converted into a CVSItem
209 instance and stored to CVS_ITEMS_STORE. There are several types
212 CVSRevision -- A specific revision of a specific CVS file.
214 CVSBranch -- The creation of a branch tag in a specific CVS
217 CVSTag -- The creation of a non-branch tag in a specific CVS
220 The CVSItems are grouped into CVSFileItems instances, one per
221 CVSFile. But a multi-file commit will still be scattered all
224 - Selected metadata for each CVS revision, including the author and
225 log message, is written to METADATA_INDEX_TABLE and
226 METADATA_STORE. The purpose is twofold: first, to save space by
227 not having to save this information multiple times, and second
228 because CVSRevisions that have the same metadata are candidates
229 to be combined into an SVN changeset.
231 First, an SHA digest is created for each set of metadata. The
232 digest is constructed so that CVSRevisions that can be combined
233 are all mapped to the same digest. CVSRevisions that were part
234 of a single CVS commit always have a common author and log
235 message, therefore these fields are always included in the
238 - if ctx.cross_project_commits is False, we avoid combining CVS
239 revisions from separate projects by including the project.id in
242 - if ctx.cross_branch_commits is False, we avoid combining CVS
243 revisions from different branches by including the branch name
246 During the database creation phase, the database keeps track of a
249 digest (20-byte string) -> metadata_id (int)
251 to allow the record for a set of metadata to be located
252 efficiently. As data are collected, it stores a map
254 metadata_id (int) -> (author, log_msg,) (tuple)
256 into the database for use in future passes. CVSRevision records
257 include the metadata_id.
259 During this run, each CVSFile, Symbol, CVSItem, and metadata record is
260 assigned an arbitrary unique ID that is used throughout the conversion
267 Encode the cvs revision metadata as UTF-8, ensuring that all entries
268 can be decoded using the chosen encodings. Output the results to
269 METADATA_CLEAN_INDEX_TABLE and METADATA_CLEAN_STORE.
275 Use the symbol statistics collected in CollectRevsPass and any runtime
276 options to determine which symbols should be treated as branches,
277 which as tags, and which should be excluded from the conversion
280 Create SYMBOL_DB, which contains a pickle of a list of TypedSymbol
281 (Branch, Tag, or ExcludedSymbol) instances indicating how each symbol
282 should be processed in the conversion. The IDs used for a TypedSymbol
283 is the same as the ID allocated to the corresponding symbol in
284 CollectRevsPass, so references in CVSItems do not have to be updated.
290 This pass works through the CVSFileItems instances stored in
291 CVS_ITEMS_STORE, processing all of the items from each file as a
292 group. (This is the last pass in which all of the CVSItems for a file
293 are in memory at once.) It does the following things:
295 - Exclude any symbols that CollateSymbolsPass determined should be
296 excluded, and any revisions on such branches. Also delete
297 references from other CVSItems to those that are being deleted.
299 - Transform any branches to tags or vice versa, also depending on
300 the results of CollateSymbolsPass, and fix up the references from
303 - Decide what line of development to use as the parent for each
304 symbol in the file, and adjust the file's dependency tree
307 - For each CVSRevision, record the list of symbols that the
308 revision opens and closes.
310 - Write each surviving CVSRevision to CVS_REVS_DATAFILE. Each line
311 of the file has the format
313 METADATA_ID TIMESTAMP CVS_REVISION
315 where TIMESTAMP is a fixed-width timestamp, and CVS_REVISION is
316 the pickled CVSRevision in a format that does not contain any
317 newlines. These summaries will be sorted in SortRevisionsPass
318 then used by InitializeChangesetsPass to create preliminary
321 - Write the CVSSymbols to CVS_SYMBOLS_DATAFILE. Each line of the
326 where CVS_SYMBOL is the pickled CVSSymbol in a format that does
327 not contain any newlines. This information will be sorted by
328 SYMBOL_ID in SortSymbolsPass then used to create preliminary
335 Sort CVS_REVS_DATAFILE (written by FilterSymbolsPass), creating
336 CVS_REVS_SORTED_DATAFILE. The sort groups items that might be added
337 to the same changeset together and, within a group, sorts revisions by
338 timestamp. This step makes it easy for InitializeChangesetsPass to
339 read the initial draft of RevisionChangesets straight from the file.
345 Sort CVS_SYMBOLS_DATAFILE (written by FilterSymbolsPass), creating
346 CVS_SYMBOLS_SORTED_DATAFILE. The sort groups together symbol items
347 that might be added to the same changeset (though not in anything
348 resembling chronological order). The output of this pass is used by
349 InitializeChangesetsPass.
352 InitializeChangesetsPass
353 ========================
355 This pass creates first-draft changesets, splitting them using
356 COMMIT_THRESHOLD and breaking up any revision changesets that have
357 internal dependencies.
359 The raw material for creating revision changesets is
360 CVS_REVS_SORTED_DATAFILE, which already has CVSRevisions sorted in
361 such a way that potential changesets are grouped together and sorted
362 by date. The contents of this file are read line by line, and the
363 corresponding CVSRevisions are accumulated into a changeset. Whenever
364 the metadata_id changes, or whenever there is a time gap of more than
365 COMMIT_THRESHOLD (currently set to 5 minutes) between CVSRevisions,
366 then a new changeset is started.
368 At this point a revision changeset can have internal dependencies if
369 two commits were made to the same file with the same log message
370 within COMMIT_THRESHOLD of each other. The next job of this pass is
371 to split up changesets in such a way to break such internal
372 dependencies. This is done by sorting the CVSRevisions within a
373 changeset by timestamp, then choosing the split point that breaks the
374 most internal dependencies. This procedure is continued recursively
375 until there are no more dependencies internal to a single changeset.
377 Analogously, the CVSSymbol items from CVS_SYMBOLS_SORTED_DATAFILE are
378 grouped into symbol changesets. (Symbol changesets cannot have
379 internal dependencies, so there is no need to break them up at this
382 Finally, this pass writes a CVSItem database with the CVSItems written
383 in order grouped by the preliminary changeset to which they belong.
384 Even though the preliminary changesets still have to be split up to
385 form final changesets, grouping the CVSItems this way improves the
386 locality of disk accesses and thereby speeds up later passes.
388 The result of this pass is two databases:
390 - CVS_ITEM_TO_CHANGESET, which maps CVSItem ids to the id of the
391 changeset containing the item, and
393 - CHANGESETS_STORE and CHANGESETS_INDEX, which contain the
394 changeset objects themselves, indexed by changeset id.
396 - CVS_ITEMS_SORTED_STORE and CVS_ITEMS_SORTED_INDEX_TABLE, which
397 contain the pickled CVSItems ordered by changeset.
400 BreakRevisionChangesetCyclesPass
401 ================================
403 There can still be cycles in the dependency graph of
404 RevisionChangesets caused by:
406 - Interleaved commits. Since CVS commits are not atomic, it can
407 happen that two commits are in progress at the same time and each
408 alters the same two files, but in different orders. These should
409 be small cycles involving only a few revision changesets. To
410 resolve these cycles, one or more of the RevisionChangesets have
411 to be split up (eventually becoming separate svn commits).
413 - Cycles involving a RevisionChangeset formed by the accidental
414 combination of unrelated items within a short period of time that
415 have the same author and log message. These should also be small
416 cycles involving only a few changesets.
418 The job of this pass is to break up such cycles (those involving only
421 This pass works by building up the graph of revision changesets and
422 their dependencies in memory, then attempting a topological sort of
423 the changesets. Whenever the topological sort stalls, that implies
424 the existence of a cycle, one of which can easily be determined. This
425 cycle is broken through the use of heuristics that try to determine an
426 "efficient" way of splitting one or more of the changesets that are
429 The new RevisionChangesets are written to
430 CVS_ITEM_TO_CHANGESET_REVBROKEN, CHANGESETS_REVBROKEN_STORE, and
431 CHANGESETS_REVBROKEN_INDEX, along with the unmodified
432 SymbolChangesets. These files are in the same format as the analogous
433 files produced by InitializeChangesetsPass.
436 RevisionTopologicalSortPass
437 ===========================
439 Topologically sort the RevisionChangesets, thereby picking the order
440 in which the RevisionChangesets will be committed. (Since the
441 previous pass eliminated any dependency cycles, this sort is
442 guaranteed to succeed.) Ambiguities in the topological sort are
443 resolved using the changesets' timestamps. Then simplify the
444 changeset graph into a linear chain by converting each
445 RevisionChangeset into an OrderedChangeset that stores dependency
446 links only to its commit-order predecessor and successor. This
447 simplified graph enforces the commit order that resulted from the
448 topological sort, even after the SymbolChangesets are added back into
449 the graph later. Store the OrderedChangesets into
450 CHANGESETS_REVSORTED_STORE and CHANGESETS_REVSORTED_INDEX along with
451 the unmodified SymbolChangesets.
454 BreakSymbolChangesetCyclesPass
455 ==============================
457 It is possible for there to be cycles in the graph of SymbolChangesets
460 - Split creation of branches. It is possible that branch A depends
461 on branch B in one file, but B depends on A in another file.
462 These cycles can be large, but they only involve
465 Break up such dependency loops. Output the results to
466 CVS_ITEM_TO_CHANGESET_SYMBROKEN, CHANGESETS_SYMBROKEN_STORE, and
467 CHANGESETS_SYMBROKEN_INDEX.
470 BreakAllChangesetCyclesPass
471 ===========================
473 The complete changeset graph (including both RevisionChangesets and
474 BranchChangesets) can still have dependency cycles cause by:
476 - Split creation of branches. The same branch tag can be added to
477 different files at completely different times. It is possible
478 that the revision that was branched later depends on a
479 RevisionChangeset that involves a file on the branch that was
480 created earlier. These cycles can be large, but they always
481 involve a SymbolChangeset. To resolve these cycles, the
482 SymbolChangeset is split up into two changesets.
484 In fact, tag changesets do not have to be considered--CVSTags cannot
485 participate in dependency cycles because no other CVSItem can depend
488 Since the input of this pass has been through
489 RevisionTopologicalSortPass, all revision cycles have already been
490 broken up and the order that the RevisionChangesets will be committed
491 has been determined. In this pass, the complete changeset graph is
492 created in memory, including the linear list of OrderedChangesets from
493 RevisionTopologicalSortPass plus all of the symbol changesets.
494 Because this pass doesn't break up any OrderedChangesets, it is
495 constrained to finding places within the revision changeset sequence
496 in which the symbol changeset commits can be inserted.
498 The new changesets are written to CVS_ITEM_TO_CHANGESET_ALLBROKEN,
499 CHANGESETS_ALLBROKEN_STORE, and CHANGESETS_ALLBROKEN_INDEX, which are
500 in the same format as the analogous files produced by
501 InitializeChangesetsPass.
507 Now that the earlier passes have broken up any dependency cycles among
508 the changesets, it is possible to order all of the changesets in such
509 a way that all of a changeset's dependencies are committed before the
510 changeset itself. This pass does so by again building up the graph of
511 changesets in memory, then at each step picking a changeset that has
512 no remaining dependencies and removing it from the graph. Whenever
513 more than one dependency-free changeset is available, symbol
514 changesets are chosen before revision changesets. As changesets are
515 processed, the timestamp sequence is ensured to be monotonic by the
516 simple expedient of adjusting retrograde timestamps to be later than
517 their predecessor. Timestamps that lie in the future, on the other
518 hand, are assumed to be bogus and are adjusted backwards, also to be
519 just later than their predecessor.
521 This pass writes a line to CHANGESETS_SORTED_DATAFILE for each
522 RevisionChangeset, in the order that the changesets should be
523 committed. Each lines contains
525 CHANGESET_ID TIMESTAMP
527 where CHANGESET_ID is the id of the changeset in the
528 CHANGESETS_ALLBROKEN_* databases and TIMESTAMP is the timstamp that
529 should be assigned to it when it is committed. Both values are
530 written in hexadecimal.
533 CreateRevsPass (formerly called pass5)
536 This pass generates SVNCommits from Changesets and records symbol
537 openings and closings. (One Changeset can result in multiple
538 SVNCommits, for example if it causes symbols to be filled or copies to
541 This pass does the following:
543 1. Creates a database file to map Subversion revision numbers to
544 SVNCommit instances (SVN_COMMITS_STORE and
545 SVN_COMMITS_INDEX_TABLE). Creates another database file to map CVS
546 Revisions to their Subversion Revision numbers
547 (CVS_REVS_TO_SVN_REVNUMS).
549 2. When a file is copied to a symbolic name in cvs2svn, it is copied
550 from a specific source: either a CVSRevision, or a copy created by
551 a previous CVSBranch of the file. The copy has to be made from an
552 SVN revision that is during the lifetime of the source. The SVN
553 revision when the source was created is called the symbol's
554 "opening", and the SVN revision when it was deleted or overwritten
555 is called the symbol's "closing". In this pass, the
556 SymbolingsLogger class writes out a line to
557 SYMBOL_OPENINGS_CLOSINGS for each symbol opening or closing. Note
558 that some openings do not have closings, namely if the
559 corresponding source is still present at the HEAD revision.
561 The format of each line is:
563 SYMBOL_ID SVN_REVNUM TYPE CVS_SYMBOL_ID
572 Here is what the columns mean:
574 SYMBOL_ID -- The id of the branch or tag that has an opening in
575 this SVN_REVNUM, in hexadecimal.
577 SVN_REVNUM -- The Subversion revision number in which the opening
578 or closing occurred. (There can be multiple openings and
579 closings per SVN_REVNUM).
581 TYPE -- "O" for openings and "C" for closings.
583 CVS_SYMBOL_ID -- The id of the CVSSymbol instance whose opening or
584 closing is being described, in hexadecimal.
586 Each CVSSymbol that tags a non-dead file has exactly one opening
587 and either zero or one closing. The closing, if it exists, always
588 occurs in a later SVN revision than the opening.
590 See SymbolingsLogger for more details.
593 SortSymbolOpeningsClosingsPass (formerly called pass6)
594 ==============================
596 This pass sorts SYMBOL_OPENINGS_CLOSINGS into
597 SYMBOL_OPENINGS_CLOSINGS_SORTED. This orders the file first by symbol
598 ID, and second by Subversion revision number, thus grouping all
599 openings and closings for each symbolic name together.
602 IndexSymbolsPass (formerly called pass7)
605 This pass iterates through all the lines in
606 SYMBOL_OPENINGS_CLOSINGS_SORTED, writing out a pickle file
607 (SYMBOL_OFFSETS_DB) mapping SYMBOL_ID to the file offset in
608 SYMBOL_OPENINGS_CLOSINGS_SORTED where SYMBOL_ID is first encountered.
609 This will allow us to seek to the various offsets in the file and
610 sequentially read only the openings and closings that we need.
613 OutputPass (formerly called pass8)
616 This pass opens the svn-commits database and sequentially plays out
617 all the commits to either a Subversion repository or to a dumpfile.
618 It also decides what sources to use to fill symbols.
620 In --dumpfile mode, the result of this pass is a Subversion repository
621 dumpfile (suitable for input to 'svnadmin load'). The dumpfile is the
622 data's last static stage: last chance to check over the data, run it
623 through svndumpfilter, move the dumpfile to another machine, etc.
625 When not in --dumpfile mode, no full dumpfile is created. Instead,
626 miniature dumpfiles representing a single revisions are created,
627 loaded into the repository, and then removed.
629 In both modes, the dumpfile revisions are created by walking through
630 the SVN_COMMITS_* database.
632 The database in MIRROR_NODES_STORE and MIRROR_NODES_INDEX_TABLE holds
633 a skeletal mirror of the repository structure at each SVN revision.
634 This mirror keeps track of which files existed on each LOD, but does
635 not record any file contents. cvs2svn requires this information to
636 decide which paths to copy when filling branches and tags.
638 When .cvsignore files are modified, cvs2svn computes the corresponding
639 svn:ignore properties and applies the properties to the parent
640 directory. The .cvsignore files themselves are not included in the
641 output unless the --keep-cvsignore option was specified. But in
642 either case, the .cvsignore files are recorded within the repository
643 mirror as if they were being written to disk, to ensure that the
644 containing directory is not pruned if the directory in CVS still
645 contained a .cvsignore file.
648 ===============================
649 Branches and Tags Plan.
650 ===============================
652 This pass is also where tag and branch creation is done. Since
653 subversion does tags and branches by copying from existing revisions
654 (then maybe editing the copy, making subcopies underneath, etc), the
655 big question for cvs2svn is how to achieve the minimum number of
656 operations per creation. For example, if it's possible to get the
657 right tag by just copying revision 53, then it's better to do that
658 than, say, copying revision 51 and then sub-copying in bits of
661 Tags are created as soon as cvs2svn encounters the last CVS Revision
662 that is a source for that tag. The whole tag is created in one
665 Branches are created as soon as all of their prerequisites are in
666 place. If a branch creation had to be broken up due to dependency
667 cycles, then non-final parts are also created as soon as their
668 prerequisites are ready. In such a case, the SymbolChangeset
669 specifies how much of the branch can be created in each step.
671 How just-in-time branch creation works:
673 In order to make the "best" set of copies/deletes when creating a
674 branch, cvs2svn keeps track of two sets of trees while it's making
677 1. A skeleton mirror of the subversion repository, that is, a
678 record of which file existed on which LOD for each SVN revision.
680 2. A tree for each CVS symbolic name, and the svn file/directory
681 revisions from which various parts of that tree could be copied.
683 Each LOD is recorded as a tree using the following schema: unique keys
684 map to marshal.dumps() representations of dictionaries, which in turn
685 map path component names to other unique keys:
687 root_key ==> { entryname1 : entrykey1, entryname2 : entrykey2, ... }
688 entrykey1 ==> { entrynameX : entrykeyX, ... }
689 entrykey2 ==> { entrynameY : entrykeyY, ... }
690 entrykeyX ==> { etc, etc ...}
691 entrykeyY ==> { etc, etc ...}
693 (The leaf nodes -- files -- are represented by None.)
695 The repository mirror allows cvs2svn to remember what paths exist in
698 For details on how branches and tags are created, please see the
699 docstring the SymbolingsLogger class (and its methods).