convert: readd --filemap

To handle merges correctly, this revision adds a filemap_source class
that wraps a converter_source and does the work necessary to calculate
the subgraph we're interested in.

The wrapped converter_source must provide a new getchangedfiles method
that, given a revision rev, and an index N, returns the list of files
that are different in rev and its Nth parent.

The implementation depends on the ability to skip some revisions and to
change the parents field of the commit objects that we returned earlier.

To make the conversion restartable, we assume the revisons in the
revmapfile are topologically sorted.

hgext/convert/__init__.py

@@ -1,404 +1,409
 # convert.py Foreign SCM converter
 #
 # Copyright 2005-2007 Matt Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms
 # of the GNU General Public License, incorporated herein by reference.
 
 from common import NoRepo, SKIPREV, converter_source, converter_sink
 from cvs import convert_cvs
 from darcs import darcs_source
 from git import convert_git
 from hg import mercurial_source, mercurial_sink
 from subversion import convert_svn, debugsvnlog
+import filemap
 
 import os, shutil
 from mercurial import hg, ui, util, commands
 from mercurial.i18n import _
 
 commands.norepo += " convert debugsvnlog"
 
 converters = [convert_cvs, convert_git, convert_svn, mercurial_source,
               mercurial_sink, darcs_source]
 
 def convertsource(ui, path, **opts):
     for c in converters:
         try:
             return c.getcommit and c(ui, path, **opts)
         except (AttributeError, NoRepo):
             pass
     raise util.Abort('%s: unknown repository type' % path)
 
 def convertsink(ui, path):
     if not os.path.isdir(path):
         raise util.Abort("%s: not a directory" % path)
     for c in converters:
         try:
             return c.putcommit and c(ui, path)
         except (AttributeError, NoRepo):
             pass
     raise util.Abort('%s: unknown repository type' % path)
 
 class converter(object):
     def __init__(self, ui, source, dest, revmapfile, opts):
 
         self.source = source
         self.dest = dest
         self.ui = ui
         self.opts = opts
         self.commitcache = {}
         self.revmapfile = revmapfile
         self.revmapfilefd = None
         self.authors = {}
         self.authorfile = None
 
         self.maporder = []
         self.map = {}
         try:
             origrevmapfile = open(self.revmapfile, 'r')
             for l in origrevmapfile:
                 sv, dv = l[:-1].split()
                 if sv not in self.map:
                     self.maporder.append(sv)
                 self.map[sv] = dv
             origrevmapfile.close()
         except IOError:
             pass
 
         # Read first the dst author map if any
         authorfile = self.dest.authorfile()
         if authorfile and os.path.exists(authorfile):
             self.readauthormap(authorfile)
         # Extend/Override with new author map if necessary
         if opts.get('authors'):
             self.readauthormap(opts.get('authors'))
             self.authorfile = self.dest.authorfile()
 
     def walktree(self, heads):
         '''Return a mapping that identifies the uncommitted parents of every
         uncommitted changeset.'''
         visit = heads
         known = {}
         parents = {}
         while visit:
             n = visit.pop(0)
             if n in known or n in self.map: continue
             known[n] = 1
             commit = self.cachecommit(n)
             parents[n] = []
             for p in commit.parents:
                 parents[n].append(p)
                 visit.append(p)
 
         return parents
 
     def toposort(self, parents):
         '''Return an ordering such that every uncommitted changeset is
         preceeded by all its uncommitted ancestors.'''
         visit = parents.keys()
         seen = {}
         children = {}
 
         while visit:
             n = visit.pop(0)
             if n in seen: continue
             seen[n] = 1
             # Ensure that nodes without parents are present in the 'children'
             # mapping.
             children.setdefault(n, [])
             for p in parents[n]:
                 if not p in self.map:
                     visit.append(p)
                 children.setdefault(p, []).append(n)
 
         s = []
         removed = {}
         visit = children.keys()
         while visit:
             n = visit.pop(0)
             if n in removed: continue
             dep = 0
             if n in parents:
                 for p in parents[n]:
                     if p in self.map: continue
                     if p not in removed:
                         # we're still dependent
                         visit.append(n)
                         dep = 1
                         break
 
             if not dep:
                 # all n's parents are in the list
                 removed[n] = 1
                 if n not in self.map:
                     s.append(n)
                 if n in children:
                     for c in children[n]:
                         visit.insert(0, c)
 
         if self.opts.get('datesort'):
             depth = {}
             for n in s:
                 depth[n] = 0
                 pl = [p for p in self.commitcache[n].parents
                       if p not in self.map]
                 if pl:
                     depth[n] = max([depth[p] for p in pl]) + 1
 
             s = [(depth[n], self.commitcache[n].date, n) for n in s]
             s.sort()
             s = [e[2] for e in s]
 
         return s
 
     def mapentry(self, src, dst):
         if self.revmapfilefd is None:
             try:
                 self.revmapfilefd = open(self.revmapfile, "a")
             except IOError, (errno, strerror):
                 raise util.Abort("Could not open map file %s: %s, %s\n" % (self.revmapfile, errno, strerror))
         self.map[src] = dst
         self.revmapfilefd.write("%s %s\n" % (src, dst))
         self.revmapfilefd.flush()
 
     def writeauthormap(self):
         authorfile = self.authorfile
         if authorfile:
            self.ui.status('Writing author map file %s\n' % authorfile)
            ofile = open(authorfile, 'w+')
            for author in self.authors:
                ofile.write("%s=%s\n" % (author, self.authors[author]))
            ofile.close()
 
     def readauthormap(self, authorfile):
         afile = open(authorfile, 'r')
         for line in afile:
             try:
                 srcauthor = line.split('=')[0].strip()
                 dstauthor = line.split('=')[1].strip()
                 if srcauthor in self.authors and dstauthor != self.authors[srcauthor]:
                     self.ui.status(
                         'Overriding mapping for author %s, was %s, will be %s\n'
                         % (srcauthor, self.authors[srcauthor], dstauthor))
                 else:
                     self.ui.debug('Mapping author %s to %s\n'
                                   % (srcauthor, dstauthor))
                     self.authors[srcauthor] = dstauthor
             except IndexError:
                 self.ui.warn(
                     'Ignoring bad line in author file map %s: %s\n'
                     % (authorfile, line))
         afile.close()
 
     def cachecommit(self, rev):
         commit = self.source.getcommit(rev)
         commit.author = self.authors.get(commit.author, commit.author)
         self.commitcache[rev] = commit
         return commit
 
     def copy(self, rev):
         commit = self.commitcache[rev]
         do_copies = hasattr(self.dest, 'copyfile')
         filenames = []
 
         changes = self.source.getchanges(rev)
         if isinstance(changes, basestring):
             if changes == SKIPREV:
                 dest = SKIPREV
             else:
                 dest = self.map[changes]
             self.mapentry(rev, dest)
             return
         files, copies = changes
         parents = [self.map[r] for r in commit.parents]
         if commit.parents:
             prev = commit.parents[0]
             if prev not in self.commitcache:
                 self.cachecommit(prev)
             pbranch = self.commitcache[prev].branch
         else:
             pbranch = None
         self.dest.setbranch(commit.branch, pbranch, parents)
         for f, v in files:
             filenames.append(f)
             try:
                 data = self.source.getfile(f, v)
             except IOError, inst:
                 self.dest.delfile(f)
             else:
                 e = self.source.getmode(f, v)
                 self.dest.putfile(f, e, data)
                 if do_copies:
                     if f in copies:
                         copyf = copies[f]
                         # Merely marks that a copy happened.
                         self.dest.copyfile(copyf, f)
 
         newnode = self.dest.putcommit(filenames, parents, commit)
         self.mapentry(rev, newnode)
 
     def convert(self):
         try:
             self.source.before()
             self.dest.before()
             self.source.setrevmap(self.map, self.maporder)
             self.ui.status("scanning source...\n")
             heads = self.source.getheads()
             parents = self.walktree(heads)
             self.ui.status("sorting...\n")
             t = self.toposort(parents)
             num = len(t)
             c = None
 
             self.ui.status("converting...\n")
             for c in t:
                 num -= 1
                 desc = self.commitcache[c].desc
                 if "\n" in desc:
                     desc = desc.splitlines()[0]
                 self.ui.status("%d %s\n" % (num, desc))
                 self.copy(c)
 
             tags = self.source.gettags()
             ctags = {}
             for k in tags:
                 v = tags[k]
                 if self.map.get(v, SKIPREV) != SKIPREV:
                     ctags[k] = self.map[v]
 
             if c and ctags:
                 nrev = self.dest.puttags(ctags)
                 # write another hash correspondence to override the previous
                 # one so we don't end up with extra tag heads
                 if nrev:
                     self.mapentry(c, nrev)
 
             self.writeauthormap()
         finally:
             self.cleanup()
 
     def cleanup(self):
         try:
             self.dest.after()
         finally:
             self.source.after()
         if self.revmapfilefd:
             self.revmapfilefd.close()
 
 def convert(ui, src, dest=None, revmapfile=None, **opts):
     """Convert a foreign SCM repository to a Mercurial one.
 
     Accepted source formats:
     - CVS
     - Darcs
     - git
     - Subversion
 
     Accepted destination formats:
     - Mercurial
 
     If no revision is given, all revisions will be converted. Otherwise,
     convert will only import up to the named revision (given in a format
     understood by the source).
 
     If no destination directory name is specified, it defaults to the
     basename of the source with '-hg' appended.  If the destination
     repository doesn't exist, it will be created.
 
     If <revmapfile> isn't given, it will be put in a default location
     (<dest>/.hg/shamap by default).  The <revmapfile> is a simple text
     file that maps each source commit ID to the destination ID for
     that revision, like so:
     <source ID> <destination ID>
 
     If the file doesn't exist, it's automatically created.  It's updated
     on each commit copied, so convert-repo can be interrupted and can
     be run repeatedly to copy new commits.
 
     The [username mapping] file is a simple text file that maps each source
     commit author to a destination commit author. It is handy for source SCMs
     that use unix logins to identify authors (eg: CVS). One line per author
     mapping and the line format is:
     srcauthor=whatever string you want
 
     The filemap is a file that allows filtering and remapping of files
     and directories.  Comment lines start with '#'.  Each line can
     contain one of the following directives:
 
       include path/to/file
 
       exclude path/to/file
 
       rename from/file to/file
     
     The 'include' directive causes a file, or all files under a
     directory, to be included in the destination repository.  The
     'exclude' directive causes files or directories to be omitted.
     The 'rename' directive renames a file or directory.  To rename
     from a subdirectory into the root of the repository, use '.' as
     the path to rename to.
     """
 
     util._encoding = 'UTF-8'
 
     if not dest:
         dest = hg.defaultdest(src) + "-hg"
         ui.status("assuming destination %s\n" % dest)
 
     # Try to be smart and initalize things when required
     created = False
     if os.path.isdir(dest):
         if len(os.listdir(dest)) > 0:
             try:
                 hg.repository(ui, dest)
                 ui.status("destination %s is a Mercurial repository\n" % dest)
             except hg.RepoError:
                 raise util.Abort(
                     "destination directory %s is not empty.\n"
                     "Please specify an empty directory to be initialized\n"
                     "or an already initialized mercurial repository"
                     % dest)
         else:
             ui.status("initializing destination %s repository\n" % dest)
             hg.repository(ui, dest, create=True)
             created = True
     elif os.path.exists(dest):
         raise util.Abort("destination %s exists and is not a directory" % dest)
     else:
         ui.status("initializing destination %s repository\n" % dest)
         hg.repository(ui, dest, create=True)
         created = True
 
     destc = convertsink(ui, dest)
 
     try:
         srcc = convertsource(ui, src, rev=opts.get('rev'))
     except Exception:
         if created:
             shutil.rmtree(dest, True)
         raise
 
+    fmap = opts.get('filemap')
+    if fmap:
+        srcc = filemap.filemap_source(ui, srcc, fmap)
+
     if not revmapfile:
         try:
             revmapfile = destc.revmapfile()
         except:
             revmapfile = os.path.join(destc, "map")
 
     c = converter(ui, srcc, destc, revmapfile, opts)
     c.convert()
 
 
 cmdtable = {
     "convert":
         (convert,
          [('A', 'authors', '', 'username mapping filename'),
           ('', 'filemap', '', 'remap file names using contents of file'),
           ('r', 'rev', '', 'import up to target revision REV'),
           ('', 'datesort', None, 'try to sort changesets by date')],
          'hg convert [OPTION]... SOURCE [DEST [MAPFILE]]'),
     "debugsvnlog":
         (debugsvnlog,
          [],
          'hg debugsvnlog'),
 }
 

hgext/convert/common.py

@@ -1,156 +1,169
 # common code for the convert extension
 import base64
 import cPickle as pickle
 
 def encodeargs(args):
     def encodearg(s):
         lines = base64.encodestring(s)
         lines = [l.splitlines()[0] for l in lines]
         return ''.join(lines)
 
     s = pickle.dumps(args)
     return encodearg(s)
 
 def decodeargs(s):
     s = base64.decodestring(s)
     return pickle.loads(s)
 
 class NoRepo(Exception): pass
 
 SKIPREV = 'hg-convert-skipped-revision'
 
 class commit(object):
     def __init__(self, author, date, desc, parents, branch=None, rev=None):
         self.author = author
         self.date = date
         self.desc = desc
         self.parents = parents
         self.branch = branch
         self.rev = rev
 
 class converter_source(object):
     """Conversion source interface"""
 
     def __init__(self, ui, path, rev=None):
         """Initialize conversion source (or raise NoRepo("message")
         exception if path is not a valid repository)"""
         self.ui = ui
         self.path = path
         self.rev = rev
 
         self.encoding = 'utf-8'
 
     def before(self):
         pass
 
     def after(self):
         pass
 
     def setrevmap(self, revmap, order):
         """set the map of already-converted revisions
         
         order is a list with the keys from revmap in the order they
         appear in the revision map file."""
         pass
 
     def getheads(self):
         """Return a list of this repository's heads"""
         raise NotImplementedError()
 
     def getfile(self, name, rev):
         """Return file contents as a string"""
         raise NotImplementedError()
 
     def getmode(self, name, rev):
         """Return file mode, eg. '', 'x', or 'l'"""
         raise NotImplementedError()
 
     def getchanges(self, version):
         """Returns a tuple of (files, copies)
         Files is a sorted list of (filename, id) tuples for all files changed
         in version, where id is the source revision id of the file.
 
         copies is a dictionary of dest: source
         """
         raise NotImplementedError()
 
     def getcommit(self, version):
         """Return the commit object for version"""
         raise NotImplementedError()
 
     def gettags(self):
         """Return the tags as a dictionary of name: revision"""
         raise NotImplementedError()
 
     def recode(self, s, encoding=None):
         if not encoding:
             encoding = self.encoding or 'utf-8'
 
         if isinstance(s, unicode):
             return s.encode("utf-8")
         try:
             return s.decode(encoding).encode("utf-8")
         except:
             try:
                 return s.decode("latin-1").encode("utf-8")
             except:
                 return s.decode(encoding, "replace").encode("utf-8")
 
+    def getchangedfiles(self, rev, i):
+        """Return the files changed by rev compared to parent[i].
+    
+        i is an index selecting one of the parents of rev.  The return
+        value should be the list of files that are different in rev and
+        this parent.
+
+        If rev has no parents, i is None.
+    
+        This function is only needed to support --filemap
+        """
+        raise NotImplementedError()
+
 class converter_sink(object):
     """Conversion sink (target) interface"""
 
     def __init__(self, ui, path):
         """Initialize conversion sink (or raise NoRepo("message")
         exception if path is not a valid repository)"""
         raise NotImplementedError()
 
     def getheads(self):
         """Return a list of this repository's heads"""
         raise NotImplementedError()
 
     def revmapfile(self):
         """Path to a file that will contain lines
         source_rev_id sink_rev_id
         mapping equivalent revision identifiers for each system."""
         raise NotImplementedError()
 
     def authorfile(self):
         """Path to a file that will contain lines
         srcauthor=dstauthor
         mapping equivalent authors identifiers for each system."""
         return None
 
     def putfile(self, f, e, data):
         """Put file for next putcommit().
         f: path to file
         e: '', 'x', or 'l' (regular file, executable, or symlink)
         data: file contents"""
         raise NotImplementedError()
 
     def delfile(self, f):
         """Delete file for next putcommit().
         f: path to file"""
         raise NotImplementedError()
 
     def putcommit(self, files, parents, commit):
         """Create a revision with all changed files listed in 'files'
         and having listed parents. 'commit' is a commit object containing
         at a minimum the author, date, and message for this changeset.
         Called after putfile() and delfile() calls. Note that the sink
         repository is not told to update itself to a particular revision
         (or even what that revision would be) before it receives the
         file data."""
         raise NotImplementedError()
 
     def puttags(self, tags):
         """Put tags into sink.
         tags: {tagname: sink_rev_id, ...}"""
         raise NotImplementedError()
 
     def setbranch(self, branch, pbranch, parents):
         """Set the current branch name. Called before the first putfile
         on the branch.
         branch: branch name for subsequent commits
         pbranch: branch name of parent commit
         parents: destination revisions of parent"""
         pass

hgext/convert/filemap.py

@@ -1,94 +1,330
 # Copyright 2007 Bryan O'Sullivan <[e-mail]>
+# Copyright 2007 Alexis S. L. Carvalho <[e-mail]>
 #
 # This software may be used and distributed according to the terms of
 # the GNU General Public License, incorporated herein by reference.
 
 import shlex
 from mercurial.i18n import _
 from mercurial import util
+from common import SKIPREV
 
 def rpairs(name):
     e = len(name)
     while e != -1:
         yield name[:e], name[e+1:]
         e = name.rfind('/', 0, e)
 
 class filemapper(object):
     '''Map and filter filenames when importing.
     A name can be mapped to itself, a new name, or None (omit from new
     repository).'''
 
     def __init__(self, ui, path=None):
         self.ui = ui
         self.include = {}
         self.exclude = {}
         self.rename = {}
         if path:
             if self.parse(path):
                 raise util.Abort(_('errors in filemap'))
 
     def parse(self, path):
         errs = 0
         def check(name, mapping, listname):
             if name in mapping:
                 self.ui.warn(_('%s:%d: %r already in %s list\n') %
                              (lex.infile, lex.lineno, name, listname))
                 return 1
             return 0
         lex = shlex.shlex(open(path), path, True)
         lex.wordchars += '!@#$%^&*()-=+[]{}|;:,./<>?'
         cmd = lex.get_token()
         while cmd:
             if cmd == 'include':
                 name = lex.get_token()
                 errs += check(name, self.exclude, 'exclude')
                 self.include[name] = name
             elif cmd == 'exclude':
                 name = lex.get_token()
                 errs += check(name, self.include, 'include')
                 errs += check(name, self.rename, 'rename')
                 self.exclude[name] = name
             elif cmd == 'rename':
                 src = lex.get_token()
                 dest = lex.get_token()
                 errs += check(src, self.exclude, 'exclude')
                 self.rename[src] = dest
             elif cmd == 'source':
                 errs += self.parse(lex.get_token())
             else:
                 self.ui.warn(_('%s:%d: unknown directive %r\n') %
                              (lex.infile, lex.lineno, cmd))
                 errs += 1
             cmd = lex.get_token()
         return errs
 
     def lookup(self, name, mapping):
         for pre, suf in rpairs(name):
             try:
                 return mapping[pre], pre, suf
             except KeyError, err:
                 pass
         return '', name, ''
 
     def __call__(self, name):
         if self.include:
             inc = self.lookup(name, self.include)[0]
         else:
             inc = name
         if self.exclude:
             exc = self.lookup(name, self.exclude)[0]
         else:
             exc = ''
         if not inc or exc:
             return None
         newpre, pre, suf = self.lookup(name, self.rename)
         if newpre:
             if newpre == '.':
                 return suf
             if suf:
                 return newpre + '/' + suf
             return newpre
         return name
 
     def active(self):
         return bool(self.include or self.exclude or self.rename)
+
+# This class does two additional things compared to a regular source:
+#
+# - Filter and rename files.  This is mostly wrapped by the filemapper
+#   class above. We hide the original filename in the revision that is
+#   returned by getchanges to be able to find things later in getfile
+#   and getmode.
+#
+# - Return only revisions that matter for the files we're interested in.
+#   This involves rewriting the parents of the original revision to
+#   create a graph that is restricted to those revisions.
+#
+#   This set of revisions includes not only revisions that directly
+#   touch files we're interested in, but also merges that merge two
+#   or more interesting revisions.
+
+class filemap_source(object):
+    def __init__(self, ui, baseconverter, filemap):
+        self.ui = ui
+        self.base = baseconverter
+        self.filemapper = filemapper(ui, filemap)
+        self.commits = {}
+        # if a revision rev has parent p in the original revision graph, then
+        # rev will have parent self.parentmap[p] in the restricted graph.
+        self.parentmap = {}
+        # self.wantedancestors[rev] is the set of all ancestors of rev that
+        # are in the restricted graph.
+        self.wantedancestors = {}
+        self.convertedorder = None
+        self._rebuilt = False
+        self.origparents = {}
+
+    def setrevmap(self, revmap, order):
+        # rebuild our state to make things restartable
+        #
+        # To avoid calling getcommit for every revision that has already
+        # been converted, we rebuild only the parentmap, delaying the
+        # rebuild of wantedancestors until we need it (i.e. until a
+        # merge).
+        #
+        # We assume the order argument lists the revisions in
+        # topological order, so that we can infer which revisions were
+        # wanted by previous runs.
+        self._rebuilt = not revmap
+        seen = {SKIPREV: SKIPREV}
+        dummyset = util.set()
+        converted = []
+        for rev in order:
+            mapped = revmap[rev]
+            wanted = mapped not in seen
+            if wanted:
+                seen[mapped] = rev
+                self.parentmap[rev] = rev
+            else:
+                self.parentmap[rev] = seen[mapped]
+            self.wantedancestors[rev] = dummyset
+            arg = seen[mapped]
+            if arg == SKIPREV:
+                arg = None
+            converted.append((rev, wanted, arg))
+        self.convertedorder = converted
+        return self.base.setrevmap(revmap, order)
+
+    def rebuild(self):
+        if self._rebuilt:
+            return True
+        self._rebuilt = True
+        pmap = self.parentmap.copy()
+        self.parentmap.clear()
+        self.wantedancestors.clear()
+        for rev, wanted, arg in self.convertedorder:
+            parents = self.origparents.get(rev)
+            if parents is None:
+                parents = self.base.getcommit(rev).parents
+            if wanted:
+                self.mark_wanted(rev, parents)
+            else:
+                self.mark_not_wanted(rev, arg)
+
+        assert pmap == self.parentmap
+        return True
+
+    def getheads(self):
+        return self.base.getheads()
+
+    def getcommit(self, rev):
+        # We want to save a reference to the commit objects to be able
+        # to rewrite their parents later on.
+        self.commits[rev] = self.base.getcommit(rev)
+        return self.commits[rev]
+
+    def wanted(self, rev, i):
+        # Return True if we're directly interested in rev.
+        #
+        # i is an index selecting one of the parents of rev (if rev
+        # has no parents, i is None).  getchangedfiles will give us
+        # the list of files that are different in rev and in the parent
+        # indicated by i.  If we're interested in any of these files,
+        # we're interested in rev.
+        try:
+            files = self.base.getchangedfiles(rev, i)
+        except NotImplementedError:
+            raise util.Abort(_("source repository doesn't support --filemap"))
+        for f in files:
+            if self.filemapper(f):
+                return True
+        return False
+
+    def mark_not_wanted(self, rev, p):
+        # Mark rev as not interesting and update data structures.
+
+        if p is None:
+            # A root revision. Use SKIPREV to indicate that it doesn't
+            # map to any revision in the restricted graph.  Put SKIPREV
+            # in the set of wanted ancestors to simplify code elsewhere
+            self.parentmap[rev] = SKIPREV
+            self.wantedancestors[rev] = util.set((SKIPREV,))
+            return
+
+        # Reuse the data from our parent.
+        self.parentmap[rev] = self.parentmap[p]
+        self.wantedancestors[rev] = self.wantedancestors[p]
+
+    def mark_wanted(self, rev, parents):
+        # Mark rev ss wanted and update data structures.
+
+        # rev will be in the restricted graph, so children of rev in
+        # the original graph should still have rev as a parent in the
+        # restricted graph.
+        self.parentmap[rev] = rev
+
+        # The set of wanted ancestors of rev is the union of the sets
+        # of wanted ancestors of its parents. Plus rev itself.
+        wrev = util.set()
+        for p in parents:
+            wrev.update(self.wantedancestors[p])
+        wrev.add(rev)
+        self.wantedancestors[rev] = wrev
+
+    def getchanges(self, rev):
+        parents = self.commits[rev].parents
+        if len(parents) > 1:
+            self.rebuild()
+
+        # To decide whether we're interested in rev we:
+        #
+        # - calculate what parents rev will have if it turns out we're
+        #   interested in it.  If it's going to have more than 1 parent,
+        #   we're interested in it.
+        #
+        # - otherwise, we'll compare it with the single parent we found.
+        #   If any of the files we're interested in is different in the
+        #   the two revisions, we're interested in rev.
+
+        # A parent p is interesting if its mapped version (self.parentmap[p]):
+        # - is not SKIPREV
+        # - is still not in the list of parents (we don't want duplicates)
+        # - is not an ancestor of the mapped versions of the other parents
+        mparents = []
+        wp = None
+        for i, p1 in enumerate(parents):
+            mp1 = self.parentmap[p1]
+            if mp1 == SKIPREV or mp1 in mparents:
+                continue
+            for p2 in parents:
+                if p1 == p2 or mp1 == self.parentmap[p2]:
+                    continue
+                if mp1 in self.wantedancestors[p2]:
+                    break
+            else:
+                mparents.append(mp1)
+                wp = i
+
+        if wp is None and parents:
+            wp = 0
+
+        self.origparents[rev] = parents
+
+        if len(mparents) < 2 and not self.wanted(rev, wp):
+            # We don't want this revision.
+            # Update our state and tell the convert process to map this
+            # revision to the same revision its parent as mapped to.
+            p = None
+            if parents:
+                p = parents[wp]
+            self.mark_not_wanted(rev, p)
+            self.convertedorder.append((rev, False, p))
+            return self.parentmap[rev]
+
+        # We want this revision.
+        # Rewrite the parents of the commit object
+        self.commits[rev].parents = mparents
+        self.mark_wanted(rev, parents)
+        self.convertedorder.append((rev, True, None))
+
+        # Get the real changes and do the filtering/mapping.
+        # To be able to get the files later on in getfile and getmode,
+        # we hide the original filename in the rev part of the return
+        # value.
+        changes, copies = self.base.getchanges(rev)
+        newnames = {}
+        files = []
+        for f, r in changes:
+            newf = self.filemapper(f)
+            if newf:
+                files.append((newf, (f, r)))
+                newnames[f] = newf
+
+        ncopies = {}
+        for c in copies:
+            newc = self.filemapper(c)
+            if newc:
+                newsource = self.filemapper(copies[c])
+                if newsource:
+                    ncopies[newc] = newsource
+
+        return files, ncopies
+
+    def getfile(self, name, rev):
+        realname, realrev = rev
+        return self.base.getfile(realname, realrev)
+
+    def getmode(self, name, rev):
+        realname, realrev = rev
+        return self.base.getmode(realname, realrev)
+
+    def gettags(self):
+        return self.base.gettags()
+
+    def before(self):
+        pass
+
+    def after(self):
+        pass