hgext/convert/subversion.py

xdiff: add a preprocessing step that trims files

xdiff has a `xdl_trim_ends` step that removes common lines, unmatchable
lines. That is in theory good, but happens too late - after splitting,
hashing, and adjusting the hash values so they are unique. Those splitting,
hashing and adjusting hash values steps could have noticeable overhead.

Diffing two large files with minor (one-line-ish) changes are not uncommon.
In that case, the raw performance of those preparation steps seriously
matter. Even allocating an O(N) array and storing line offsets to it is
expensive.  Therefore my previous attempts [1] [2] cannot be good enough
since they do not remove the O(N) array assignment.

This patch adds a preprocessing step - `xdl_trim_files` that runs before
other preprocessing steps. It counts common prefix and suffix and lines in
them (needed for displaying line number), without doing anything else.

Testing with a crafted large (169MB) file, with minor change:

```
  open('a','w').write(''.join('%s\n' % (i % 100000) for i in xrange(30000000) if i != 6000000))
  open('b','w').write(''.join('%s\n' % (i % 100000) for i in xrange(30000000) if i != 6003000))
```

Running xdiff by a simple binary [3], this patch improves the xdiff perf by
more than 10x for the above case:

```
  # xdiff before this patch
  2.41s user 1.13s system 98% cpu 3.592 total
  # xdiff after this patch
  0.14s user 0.16s system 98% cpu 0.309 total
  # gnu diffutils
  0.12s user 0.15s system 98% cpu 0.272 total
  # (best of 20 runs)
```

It's still slightly slower than GNU diffutils. But it's pretty close now.

Testing with real repo data:

For the whole repo, this patch makes xdiff 25% faster:

```
  # hg perfbdiff --count 100 --alldata -c d334afc585e2 --blocks [--xdiff]
  # xdiff, after
  ! wall 0.058861 comb 0.050000 user 0.050000 sys 0.000000 (best of 100)
  # xdiff, before
  ! wall 0.077816 comb 0.080000 user 0.080000 sys 0.000000 (best of 91)
  # bdiff
  ! wall 0.117473 comb 0.120000 user 0.120000 sys 0.000000 (best of 67)
```

For files that are long (ex. commands.py), the speedup is more than 3x, very
significant:

```
  # hg perfbdiff --count 3000 --blocks commands.py.i 1 [--xdiff]
  # xdiff, after
  ! wall 0.690583 comb 0.690000 user 0.690000 sys 0.000000 (best of 12)
  # xdiff, before
  ! wall 2.240361 comb 2.210000 user 2.210000 sys 0.000000 (best of 4)
  # bdiff
  ! wall 2.469852 comb 2.440000 user 2.440000 sys 0.000000 (best of 4)
```

[1]: https://phab.mercurial-scm.org/D2631
[2]: https://phab.mercurial-scm.org/D2634
[3]:

```
// Code to run xdiff from command line. No proper error handling.
#include <stdlib.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include "mercurial/thirdparty/xdiff/xdiff.h"
#define ensure(x) if (!(x)) exit(255);
mmfile_t readfile(const char *path) {
  struct stat st; int fd = open(path, O_RDONLY);
  fstat(fd, &st); mmfile_t file = { malloc(st.st_size), st.st_size };
  ensure(read(fd, file.ptr, st.st_size) == st.st_size); close(fd);
  return file;
}
int main(int argc, char const *argv[]) {
  mmfile_t a = readfile(argv[1]), b = readfile(argv[2]);
  xpparam_t xpp = {0}; xdemitconf_t xecfg = {0}; xdemitcb_t ecb = {0};
  xdl_diff(&a, &b, &xpp, &xecfg, &ecb);
  return 0;
}
```

Differential Revision: https://phab.mercurial-scm.org/D2686

# Subversion 1.4/1.5 Python API backend
#
# Copyright(C) 2007 Daniel Holth et al
from __future__ import absolute_import

import os
import re
import tempfile
import xml.dom.minidom

from mercurial.i18n import _
from mercurial import (
    encoding,
    error,
    pycompat,
    util,
    vfs as vfsmod,
)
from mercurial.utils import dateutil

from . import common

pickle = util.pickle
stringio = util.stringio
propertycache = util.propertycache
urlerr = util.urlerr
urlreq = util.urlreq

commandline = common.commandline
commit = common.commit
converter_sink = common.converter_sink
converter_source = common.converter_source
decodeargs = common.decodeargs
encodeargs = common.encodeargs
makedatetimestamp = common.makedatetimestamp
mapfile = common.mapfile
MissingTool = common.MissingTool
NoRepo = common.NoRepo

# Subversion stuff. Works best with very recent Python SVN bindings
# e.g. SVN 1.5 or backports. Thanks to the bzr folks for enhancing
# these bindings.

try:
    import svn
    import svn.client
    import svn.core
    import svn.ra
    import svn.delta
    from . import transport
    import warnings
    warnings.filterwarnings('ignore',
            module='svn.core',
            category=DeprecationWarning)
    svn.core.SubversionException # trigger import to catch error

except ImportError:
    svn = None

class SvnPathNotFound(Exception):
    pass

def revsplit(rev):
    """Parse a revision string and return (uuid, path, revnum).
    >>> revsplit(b'svn:a2147622-4a9f-4db4-a8d3-13562ff547b2'
    ...          b'/proj%20B/mytrunk/mytrunk@1')
    ('a2147622-4a9f-4db4-a8d3-13562ff547b2', '/proj%20B/mytrunk/mytrunk', 1)
    >>> revsplit(b'svn:8af66a51-67f5-4354-b62c-98d67cc7be1d@1')
    ('', '', 1)
    >>> revsplit(b'@7')
    ('', '', 7)
    >>> revsplit(b'7')
    ('', '', 0)
    >>> revsplit(b'bad')
    ('', '', 0)
    """
    parts = rev.rsplit('@', 1)
    revnum = 0
    if len(parts) > 1:
        revnum = int(parts[1])
    parts = parts[0].split('/', 1)
    uuid = ''
    mod = ''
    if len(parts) > 1 and parts[0].startswith('svn:'):
        uuid = parts[0][4:]
        mod = '/' + parts[1]
    return uuid, mod, revnum

def quote(s):
    # As of svn 1.7, many svn calls expect "canonical" paths. In
    # theory, we should call svn.core.*canonicalize() on all paths
    # before passing them to the API.  Instead, we assume the base url
    # is canonical and copy the behaviour of svn URL encoding function
    # so we can extend it safely with new components. The "safe"
    # characters were taken from the "svn_uri__char_validity" table in
    # libsvn_subr/path.c.
    return urlreq.quote(s, "!$&'()*+,-./:=@_~")

def geturl(path):
    try:
        return svn.client.url_from_path(svn.core.svn_path_canonicalize(path))
    except svn.core.SubversionException:
        # svn.client.url_from_path() fails with local repositories
        pass
    if os.path.isdir(path):
        path = os.path.normpath(os.path.abspath(path))
        if pycompat.iswindows:
            path = '/' + util.normpath(path)
        # Module URL is later compared with the repository URL returned
        # by svn API, which is UTF-8.
        path = encoding.tolocal(path)
        path = 'file://%s' % quote(path)
    return svn.core.svn_path_canonicalize(path)

def optrev(number):
    optrev = svn.core.svn_opt_revision_t()
    optrev.kind = svn.core.svn_opt_revision_number
    optrev.value.number = number
    return optrev

class changedpath(object):
    def __init__(self, p):
        self.copyfrom_path = p.copyfrom_path
        self.copyfrom_rev = p.copyfrom_rev
        self.action = p.action

def get_log_child(fp, url, paths, start, end, limit=0,
                  discover_changed_paths=True, strict_node_history=False):
    protocol = -1
    def receiver(orig_paths, revnum, author, date, message, pool):
        paths = {}
        if orig_paths is not None:
            for k, v in orig_paths.iteritems():
                paths[k] = changedpath(v)
        pickle.dump((paths, revnum, author, date, message),
                    fp, protocol)

    try:
        # Use an ra of our own so that our parent can consume
        # our results without confusing the server.
        t = transport.SvnRaTransport(url=url)
        svn.ra.get_log(t.ra, paths, start, end, limit,
                       discover_changed_paths,
                       strict_node_history,
                       receiver)
    except IOError:
        # Caller may interrupt the iteration
        pickle.dump(None, fp, protocol)
    except Exception as inst:
        pickle.dump(util.forcebytestr(inst), fp, protocol)
    else:
        pickle.dump(None, fp, protocol)
    fp.flush()
    # With large history, cleanup process goes crazy and suddenly
    # consumes *huge* amount of memory. The output file being closed,
    # there is no need for clean termination.
    os._exit(0)

def debugsvnlog(ui, **opts):
    """Fetch SVN log in a subprocess and channel them back to parent to
    avoid memory collection issues.
    """
    if svn is None:
        raise error.Abort(_('debugsvnlog could not load Subversion python '
                           'bindings'))

    args = decodeargs(ui.fin.read())
    get_log_child(ui.fout, *args)

class logstream(object):
    """Interruptible revision log iterator."""
    def __init__(self, stdout):
        self._stdout = stdout

    def __iter__(self):
        while True:
            try:
                entry = pickle.load(self._stdout)
            except EOFError:
                raise error.Abort(_('Mercurial failed to run itself, check'
                                   ' hg executable is in PATH'))
            try:
                orig_paths, revnum, author, date, message = entry
            except (TypeError, ValueError):
                if entry is None:
                    break
                raise error.Abort(_("log stream exception '%s'") % entry)
            yield entry

    def close(self):
        if self._stdout:
            self._stdout.close()
            self._stdout = None

class directlogstream(list):
    """Direct revision log iterator.
    This can be used for debugging and development but it will probably leak
    memory and is not suitable for real conversions."""
    def __init__(self, url, paths, start, end, limit=0,
                  discover_changed_paths=True, strict_node_history=False):

        def receiver(orig_paths, revnum, author, date, message, pool):
            paths = {}
            if orig_paths is not None:
                for k, v in orig_paths.iteritems():
                    paths[k] = changedpath(v)
            self.append((paths, revnum, author, date, message))

        # Use an ra of our own so that our parent can consume
        # our results without confusing the server.
        t = transport.SvnRaTransport(url=url)
        svn.ra.get_log(t.ra, paths, start, end, limit,
                       discover_changed_paths,
                       strict_node_history,
                       receiver)

    def close(self):
        pass

# Check to see if the given path is a local Subversion repo. Verify this by
# looking for several svn-specific files and directories in the given
# directory.
def filecheck(ui, path, proto):
    for x in ('locks', 'hooks', 'format', 'db'):
        if not os.path.exists(os.path.join(path, x)):
            return False
    return True

# Check to see if a given path is the root of an svn repo over http. We verify
# this by requesting a version-controlled URL we know can't exist and looking
# for the svn-specific "not found" XML.
def httpcheck(ui, path, proto):
    try:
        opener = urlreq.buildopener()
        rsp = opener.open('%s://%s/!svn/ver/0/.svn' % (proto, path), 'rb')
        data = rsp.read()
    except urlerr.httperror as inst:
        if inst.code != 404:
            # Except for 404 we cannot know for sure this is not an svn repo
            ui.warn(_('svn: cannot probe remote repository, assume it could '
                      'be a subversion repository. Use --source-type if you '
                      'know better.\n'))
            return True
        data = inst.fp.read()
    except Exception:
        # Could be urlerr.urlerror if the URL is invalid or anything else.
        return False
    return '<m:human-readable errcode="160013">' in data

protomap = {'http': httpcheck,
            'https': httpcheck,
            'file': filecheck,
            }
def issvnurl(ui, url):
    try:
        proto, path = url.split('://', 1)
        if proto == 'file':
            if (pycompat.iswindows and path[:1] == '/'
                  and path[1:2].isalpha() and path[2:6].lower() == '%3a/'):
                path = path[:2] + ':/' + path[6:]
            path = urlreq.url2pathname(path)
    except ValueError:
        proto = 'file'
        path = os.path.abspath(url)
    if proto == 'file':
        path = util.pconvert(path)
    check = protomap.get(proto, lambda *args: False)
    while '/' in path:
        if check(ui, path, proto):
            return True
        path = path.rsplit('/', 1)[0]
    return False

# SVN conversion code stolen from bzr-svn and tailor
#
# Subversion looks like a versioned filesystem, branches structures
# are defined by conventions and not enforced by the tool. First,
# we define the potential branches (modules) as "trunk" and "branches"
# children directories. Revisions are then identified by their
# module and revision number (and a repository identifier).
#
# The revision graph is really a tree (or a forest). By default, a
# revision parent is the previous revision in the same module. If the
# module directory is copied/moved from another module then the
# revision is the module root and its parent the source revision in
# the parent module. A revision has at most one parent.
#
class svn_source(converter_source):
    def __init__(self, ui, repotype, url, revs=None):
        super(svn_source, self).__init__(ui, repotype, url, revs=revs)

        if not (url.startswith('svn://') or url.startswith('svn+ssh://') or
                (os.path.exists(url) and
                 os.path.exists(os.path.join(url, '.svn'))) or
                issvnurl(ui, url)):
            raise NoRepo(_("%s does not look like a Subversion repository")
                         % url)
        if svn is None:
            raise MissingTool(_('could not load Subversion python bindings'))

        try:
            version = svn.core.SVN_VER_MAJOR, svn.core.SVN_VER_MINOR
            if version < (1, 4):
                raise MissingTool(_('Subversion python bindings %d.%d found, '
                                    '1.4 or later required') % version)
        except AttributeError:
            raise MissingTool(_('Subversion python bindings are too old, 1.4 '
                                'or later required'))

        self.lastrevs = {}

        latest = None
        try:
            # Support file://path@rev syntax. Useful e.g. to convert
            # deleted branches.
            at = url.rfind('@')
            if at >= 0:
                latest = int(url[at + 1:])
                url = url[:at]
        except ValueError:
            pass
        self.url = geturl(url)
        self.encoding = 'UTF-8' # Subversion is always nominal UTF-8
        try:
            self.transport = transport.SvnRaTransport(url=self.url)
            self.ra = self.transport.ra
            self.ctx = self.transport.client
            self.baseurl = svn.ra.get_repos_root(self.ra)
            # Module is either empty or a repository path starting with
            # a slash and not ending with a slash.
            self.module = urlreq.unquote(self.url[len(self.baseurl):])
            self.prevmodule = None
            self.rootmodule = self.module
            self.commits = {}
            self.paths = {}
            self.uuid = svn.ra.get_uuid(self.ra)
        except svn.core.SubversionException:
            ui.traceback()
            svnversion = '%d.%d.%d' % (svn.core.SVN_VER_MAJOR,
                                       svn.core.SVN_VER_MINOR,
                                       svn.core.SVN_VER_MICRO)
            raise NoRepo(_("%s does not look like a Subversion repository "
                           "to libsvn version %s")
                         % (self.url, svnversion))

        if revs:
            if len(revs) > 1:
                raise error.Abort(_('subversion source does not support '
                                   'specifying multiple revisions'))
            try:
                latest = int(revs[0])
            except ValueError:
                raise error.Abort(_('svn: revision %s is not an integer') %
                                 revs[0])

        trunkcfg = self.ui.config('convert', 'svn.trunk')
        if trunkcfg is None:
            trunkcfg = 'trunk'
        self.trunkname = trunkcfg.strip('/')
        self.startrev = self.ui.config('convert', 'svn.startrev')
        try:
            self.startrev = int(self.startrev)
            if self.startrev < 0:
                self.startrev = 0
        except ValueError:
            raise error.Abort(_('svn: start revision %s is not an integer')
                             % self.startrev)

        try:
            self.head = self.latest(self.module, latest)
        except SvnPathNotFound:
            self.head = None
        if not self.head:
            raise error.Abort(_('no revision found in module %s')
                             % self.module)
        self.last_changed = self.revnum(self.head)

        self._changescache = (None, None)

        if os.path.exists(os.path.join(url, '.svn/entries')):
            self.wc = url
        else:
            self.wc = None
        self.convertfp = None

    def setrevmap(self, revmap):
        lastrevs = {}
        for revid in revmap:
            uuid, module, revnum = revsplit(revid)
            lastrevnum = lastrevs.setdefault(module, revnum)
            if revnum > lastrevnum:
                lastrevs[module] = revnum
        self.lastrevs = lastrevs

    def exists(self, path, optrev):
        try:
            svn.client.ls(self.url.rstrip('/') + '/' + quote(path),
                                 optrev, False, self.ctx)
            return True
        except svn.core.SubversionException:
            return False

    def getheads(self):

        def isdir(path, revnum):
            kind = self._checkpath(path, revnum)
            return kind == svn.core.svn_node_dir

        def getcfgpath(name, rev):
            cfgpath = self.ui.config('convert', 'svn.' + name)
            if cfgpath is not None and cfgpath.strip() == '':
                return None
            path = (cfgpath or name).strip('/')
            if not self.exists(path, rev):
                if self.module.endswith(path) and name == 'trunk':
                    # we are converting from inside this directory
                    return None
                if cfgpath:
                    raise error.Abort(_('expected %s to be at %r, but not found'
                                       ) % (name, path))
                return None
            self.ui.note(_('found %s at %r\n') % (name, path))
            return path

        rev = optrev(self.last_changed)
        oldmodule = ''
        trunk = getcfgpath('trunk', rev)
        self.tags = getcfgpath('tags', rev)
        branches = getcfgpath('branches', rev)

        # If the project has a trunk or branches, we will extract heads
        # from them. We keep the project root otherwise.
        if trunk:
            oldmodule = self.module or ''
            self.module += '/' + trunk
            self.head = self.latest(self.module, self.last_changed)
            if not self.head:
                raise error.Abort(_('no revision found in module %s')
                                 % self.module)

        # First head in the list is the module's head
        self.heads = [self.head]
        if self.tags is not None:
            self.tags = '%s/%s' % (oldmodule , (self.tags or 'tags'))

        # Check if branches bring a few more heads to the list
        if branches:
            rpath = self.url.strip('/')
            branchnames = svn.client.ls(rpath + '/' + quote(branches),
                                        rev, False, self.ctx)
            for branch in sorted(branchnames):
                module = '%s/%s/%s' % (oldmodule, branches, branch)
                if not isdir(module, self.last_changed):
                    continue
                brevid = self.latest(module, self.last_changed)
                if not brevid:
                    self.ui.note(_('ignoring empty branch %s\n') % branch)
                    continue
                self.ui.note(_('found branch %s at %d\n') %
                             (branch, self.revnum(brevid)))
                self.heads.append(brevid)

        if self.startrev and self.heads:
            if len(self.heads) > 1:
                raise error.Abort(_('svn: start revision is not supported '
                                   'with more than one branch'))
            revnum = self.revnum(self.heads[0])
            if revnum < self.startrev:
                raise error.Abort(
                    _('svn: no revision found after start revision %d')
                                 % self.startrev)

        return self.heads

    def _getchanges(self, rev, full):
        (paths, parents) = self.paths[rev]
        copies = {}
        if parents:
            files, self.removed, copies = self.expandpaths(rev, paths, parents)
        if full or not parents:
            # Perform a full checkout on roots
            uuid, module, revnum = revsplit(rev)
            entries = svn.client.ls(self.baseurl + quote(module),
                                    optrev(revnum), True, self.ctx)
            files = [n for n, e in entries.iteritems()
                     if e.kind == svn.core.svn_node_file]
            self.removed = set()

        files.sort()
        files = zip(files, [rev] * len(files))
        return (files, copies)

    def getchanges(self, rev, full):
        # reuse cache from getchangedfiles
        if self._changescache[0] == rev and not full:
            (files, copies) = self._changescache[1]
        else:
            (files, copies) = self._getchanges(rev, full)
            # caller caches the result, so free it here to release memory
            del self.paths[rev]
        return (files, copies, set())

    def getchangedfiles(self, rev, i):
        # called from filemap - cache computed values for reuse in getchanges
        (files, copies) = self._getchanges(rev, False)
        self._changescache = (rev, (files, copies))
        return [f[0] for f in files]

    def getcommit(self, rev):
        if rev not in self.commits:
            uuid, module, revnum = revsplit(rev)
            self.module = module
            self.reparent(module)
            # We assume that:
            # - requests for revisions after "stop" come from the
            # revision graph backward traversal. Cache all of them
            # down to stop, they will be used eventually.
            # - requests for revisions before "stop" come to get
            # isolated branches parents. Just fetch what is needed.
            stop = self.lastrevs.get(module, 0)
            if revnum < stop:
                stop = revnum + 1
            self._fetch_revisions(revnum, stop)
            if rev not in self.commits:
                raise error.Abort(_('svn: revision %s not found') % revnum)
        revcommit = self.commits[rev]
        # caller caches the result, so free it here to release memory
        del self.commits[rev]
        return revcommit

    def checkrevformat(self, revstr, mapname='splicemap'):
        """ fails if revision format does not match the correct format"""
        if not re.match(r'svn:[0-9a-f]{8,8}-[0-9a-f]{4,4}-'
                              r'[0-9a-f]{4,4}-[0-9a-f]{4,4}-[0-9a-f]'
                              r'{12,12}(.*)\@[0-9]+$',revstr):
            raise error.Abort(_('%s entry %s is not a valid revision'
                               ' identifier') % (mapname, revstr))

    def numcommits(self):
        return int(self.head.rsplit('@', 1)[1]) - self.startrev

    def gettags(self):
        tags = {}
        if self.tags is None:
            return tags

        # svn tags are just a convention, project branches left in a
        # 'tags' directory. There is no other relationship than
        # ancestry, which is expensive to discover and makes them hard
        # to update incrementally.  Worse, past revisions may be
        # referenced by tags far away in the future, requiring a deep
        # history traversal on every calculation.  Current code
        # performs a single backward traversal, tracking moves within
        # the tags directory (tag renaming) and recording a new tag
        # everytime a project is copied from outside the tags
        # directory. It also lists deleted tags, this behaviour may
        # change in the future.
        pendings = []
        tagspath = self.tags
        start = svn.ra.get_latest_revnum(self.ra)
        stream = self._getlog([self.tags], start, self.startrev)
        try:
            for entry in stream:
                origpaths, revnum, author, date, message = entry
                if not origpaths:
                    origpaths = []
                copies = [(e.copyfrom_path, e.copyfrom_rev, p) for p, e
                          in origpaths.iteritems() if e.copyfrom_path]
                # Apply moves/copies from more specific to general
                copies.sort(reverse=True)

                srctagspath = tagspath
                if copies and copies[-1][2] == tagspath:
                    # Track tags directory moves
                    srctagspath = copies.pop()[0]

                for source, sourcerev, dest in copies:
                    if not dest.startswith(tagspath + '/'):
                        continue
                    for tag in pendings:
                        if tag[0].startswith(dest):
                            tagpath = source + tag[0][len(dest):]
                            tag[:2] = [tagpath, sourcerev]
                            break
                    else:
                        pendings.append([source, sourcerev, dest])

                # Filter out tags with children coming from different
                # parts of the repository like:
                # /tags/tag.1 (from /trunk:10)
                # /tags/tag.1/foo (from /branches/foo:12)
                # Here/tags/tag.1 discarded as well as its children.
                # It happens with tools like cvs2svn. Such tags cannot
                # be represented in mercurial.
                addeds = dict((p, e.copyfrom_path) for p, e
                              in origpaths.iteritems()
                              if e.action == 'A' and e.copyfrom_path)
                badroots = set()
                for destroot in addeds:
                    for source, sourcerev, dest in pendings:
                        if (not dest.startswith(destroot + '/')
                            or source.startswith(addeds[destroot] + '/')):
                            continue
                        badroots.add(destroot)
                        break

                for badroot in badroots:
                    pendings = [p for p in pendings if p[2] != badroot
                                and not p[2].startswith(badroot + '/')]

                # Tell tag renamings from tag creations
                renamings = []
                for source, sourcerev, dest in pendings:
                    tagname = dest.split('/')[-1]
                    if source.startswith(srctagspath):
                        renamings.append([source, sourcerev, tagname])
                        continue
                    if tagname in tags:
                        # Keep the latest tag value
                        continue
                    # From revision may be fake, get one with changes
                    try:
                        tagid = self.latest(source, sourcerev)
                        if tagid and tagname not in tags:
                            tags[tagname] = tagid
                    except SvnPathNotFound:
                        # It happens when we are following directories
                        # we assumed were copied with their parents
                        # but were really created in the tag
                        # directory.
                        pass
                pendings = renamings
                tagspath = srctagspath
        finally:
            stream.close()
        return tags

    def converted(self, rev, destrev):
        if not self.wc:
            return
        if self.convertfp is None:
            self.convertfp = open(os.path.join(self.wc, '.svn', 'hg-shamap'),
                                  'ab')
        self.convertfp.write(util.tonativeeol('%s %d\n'
                                              % (destrev, self.revnum(rev))))
        self.convertfp.flush()

    def revid(self, revnum, module=None):
        return 'svn:%s%s@%s' % (self.uuid, module or self.module, revnum)

    def revnum(self, rev):
        return int(rev.split('@')[-1])

    def latest(self, path, stop=None):
        """Find the latest revid affecting path, up to stop revision
        number. If stop is None, default to repository latest
        revision. It may return a revision in a different module,
        since a branch may be moved without a change being
        reported. Return None if computed module does not belong to
        rootmodule subtree.
        """
        def findchanges(path, start, stop=None):
            stream = self._getlog([path], start, stop or 1)
            try:
                for entry in stream:
                    paths, revnum, author, date, message = entry
                    if stop is None and paths:
                        # We do not know the latest changed revision,
                        # keep the first one with changed paths.
                        break
                    if revnum <= stop:
                        break

                    for p in paths:
                        if (not path.startswith(p) or
                            not paths[p].copyfrom_path):
                            continue
                        newpath = paths[p].copyfrom_path + path[len(p):]
                        self.ui.debug("branch renamed from %s to %s at %d\n" %
                                      (path, newpath, revnum))
                        path = newpath
                        break
                if not paths:
                    revnum = None
                return revnum, path
            finally:
                stream.close()

        if not path.startswith(self.rootmodule):
            # Requests on foreign branches may be forbidden at server level
            self.ui.debug('ignoring foreign branch %r\n' % path)
            return None

        if stop is None:
            stop = svn.ra.get_latest_revnum(self.ra)
        try:
            prevmodule = self.reparent('')
            dirent = svn.ra.stat(self.ra, path.strip('/'), stop)
            self.reparent(prevmodule)
        except svn.core.SubversionException:
            dirent = None
        if not dirent:
            raise SvnPathNotFound(_('%s not found up to revision %d')
                                  % (path, stop))

        # stat() gives us the previous revision on this line of
        # development, but it might be in *another module*. Fetch the
        # log and detect renames down to the latest revision.
        revnum, realpath = findchanges(path, stop, dirent.created_rev)
        if revnum is None:
            # Tools like svnsync can create empty revision, when
            # synchronizing only a subtree for instance. These empty
            # revisions created_rev still have their original values
            # despite all changes having disappeared and can be
            # returned by ra.stat(), at least when stating the root
            # module. In that case, do not trust created_rev and scan
            # the whole history.
            revnum, realpath = findchanges(path, stop)
            if revnum is None:
                self.ui.debug('ignoring empty branch %r\n' % realpath)
                return None

        if not realpath.startswith(self.rootmodule):
            self.ui.debug('ignoring foreign branch %r\n' % realpath)
            return None
        return self.revid(revnum, realpath)

    def reparent(self, module):
        """Reparent the svn transport and return the previous parent."""
        if self.prevmodule == module:
            return module
        svnurl = self.baseurl + quote(module)
        prevmodule = self.prevmodule
        if prevmodule is None:
            prevmodule = ''
        self.ui.debug("reparent to %s\n" % svnurl)
        svn.ra.reparent(self.ra, svnurl)
        self.prevmodule = module
        return prevmodule

    def expandpaths(self, rev, paths, parents):
        changed, removed = set(), set()
        copies = {}

        new_module, revnum = revsplit(rev)[1:]
        if new_module != self.module:
            self.module = new_module
            self.reparent(self.module)

        for i, (path, ent) in enumerate(paths):
            self.ui.progress(_('scanning paths'), i, item=path,
                             total=len(paths), unit=_('paths'))
            entrypath = self.getrelpath(path)

            kind = self._checkpath(entrypath, revnum)
            if kind == svn.core.svn_node_file:
                changed.add(self.recode(entrypath))
                if not ent.copyfrom_path or not parents:
                    continue
                # Copy sources not in parent revisions cannot be
                # represented, ignore their origin for now
                pmodule, prevnum = revsplit(parents[0])[1:]
                if ent.copyfrom_rev < prevnum:
                    continue
                copyfrom_path = self.getrelpath(ent.copyfrom_path, pmodule)
                if not copyfrom_path:
                    continue
                self.ui.debug("copied to %s from %s@%s\n" %
                              (entrypath, copyfrom_path, ent.copyfrom_rev))
                copies[self.recode(entrypath)] = self.recode(copyfrom_path)
            elif kind == 0: # gone, but had better be a deleted *file*
                self.ui.debug("gone from %s\n" % ent.copyfrom_rev)
                pmodule, prevnum = revsplit(parents[0])[1:]
                parentpath = pmodule + "/" + entrypath
                fromkind = self._checkpath(entrypath, prevnum, pmodule)

                if fromkind == svn.core.svn_node_file:
                    removed.add(self.recode(entrypath))
                elif fromkind == svn.core.svn_node_dir:
                    oroot = parentpath.strip('/')
                    nroot = path.strip('/')
                    children = self._iterfiles(oroot, prevnum)
                    for childpath in children:
                        childpath = childpath.replace(oroot, nroot)
                        childpath = self.getrelpath("/" + childpath, pmodule)
                        if childpath:
                            removed.add(self.recode(childpath))
                else:
                    self.ui.debug('unknown path in revision %d: %s\n' % \
                                  (revnum, path))
            elif kind == svn.core.svn_node_dir:
                if ent.action == 'M':
                    # If the directory just had a prop change,
                    # then we shouldn't need to look for its children.
                    continue
                if ent.action == 'R' and parents:
                    # If a directory is replacing a file, mark the previous
                    # file as deleted
                    pmodule, prevnum = revsplit(parents[0])[1:]
                    pkind = self._checkpath(entrypath, prevnum, pmodule)
                    if pkind == svn.core.svn_node_file:
                        removed.add(self.recode(entrypath))
                    elif pkind == svn.core.svn_node_dir:
                        # We do not know what files were kept or removed,
                        # mark them all as changed.
                        for childpath in self._iterfiles(pmodule, prevnum):
                            childpath = self.getrelpath("/" + childpath)
                            if childpath:
                                changed.add(self.recode(childpath))

                for childpath in self._iterfiles(path, revnum):
                    childpath = self.getrelpath("/" + childpath)
                    if childpath:
                        changed.add(self.recode(childpath))

                # Handle directory copies
                if not ent.copyfrom_path or not parents:
                    continue
                # Copy sources not in parent revisions cannot be
                # represented, ignore their origin for now
                pmodule, prevnum = revsplit(parents[0])[1:]
                if ent.copyfrom_rev < prevnum:
                    continue
                copyfrompath = self.getrelpath(ent.copyfrom_path, pmodule)
                if not copyfrompath:
                    continue
                self.ui.debug("mark %s came from %s:%d\n"
                              % (path, copyfrompath, ent.copyfrom_rev))
                children = self._iterfiles(ent.copyfrom_path, ent.copyfrom_rev)
                for childpath in children:
                    childpath = self.getrelpath("/" + childpath, pmodule)
                    if not childpath:
                        continue
                    copytopath = path + childpath[len(copyfrompath):]
                    copytopath = self.getrelpath(copytopath)
                    copies[self.recode(copytopath)] = self.recode(childpath)

        self.ui.progress(_('scanning paths'), None)
        changed.update(removed)
        return (list(changed), removed, copies)

    def _fetch_revisions(self, from_revnum, to_revnum):
        if from_revnum < to_revnum:
            from_revnum, to_revnum = to_revnum, from_revnum

        self.child_cset = None

        def parselogentry(orig_paths, revnum, author, date, message):
            """Return the parsed commit object or None, and True if
            the revision is a branch root.
            """
            self.ui.debug("parsing revision %d (%d changes)\n" %
                          (revnum, len(orig_paths)))

            branched = False
            rev = self.revid(revnum)
            # branch log might return entries for a parent we already have

            if rev in self.commits or revnum < to_revnum:
                return None, branched

            parents = []
            # check whether this revision is the start of a branch or part
            # of a branch renaming
            orig_paths = sorted(orig_paths.iteritems())
            root_paths = [(p, e) for p, e in orig_paths
                          if self.module.startswith(p)]
            if root_paths:
                path, ent = root_paths[-1]
                if ent.copyfrom_path:
                    branched = True
                    newpath = ent.copyfrom_path + self.module[len(path):]
                    # ent.copyfrom_rev may not be the actual last revision
                    previd = self.latest(newpath, ent.copyfrom_rev)
                    if previd is not None:
                        prevmodule, prevnum = revsplit(previd)[1:]
                        if prevnum >= self.startrev:
                            parents = [previd]
                            self.ui.note(
                                _('found parent of branch %s at %d: %s\n') %
                                (self.module, prevnum, prevmodule))
                else:
                    self.ui.debug("no copyfrom path, don't know what to do.\n")

            paths = []
            # filter out unrelated paths
            for path, ent in orig_paths:
                if self.getrelpath(path) is None:
                    continue
                paths.append((path, ent))

            # Example SVN datetime. Includes microseconds.
            # ISO-8601 conformant
            # '2007-01-04T17:35:00.902377Z'
            date = dateutil.parsedate(date[:19] + " UTC", ["%Y-%m-%dT%H:%M:%S"])
            if self.ui.configbool('convert', 'localtimezone'):
                date = makedatetimestamp(date[0])

            if message:
                log = self.recode(message)
            else:
                log = ''

            if author:
                author = self.recode(author)
            else:
                author = ''

            try:
                branch = self.module.split("/")[-1]
                if branch == self.trunkname:
                    branch = None
            except IndexError:
                branch = None

            cset = commit(author=author,
                          date=dateutil.datestr(date, '%Y-%m-%d %H:%M:%S %1%2'),
                          desc=log,
                          parents=parents,
                          branch=branch,
                          rev=rev)

            self.commits[rev] = cset
            # The parents list is *shared* among self.paths and the
            # commit object. Both will be updated below.
            self.paths[rev] = (paths, cset.parents)
            if self.child_cset and not self.child_cset.parents:
                self.child_cset.parents[:] = [rev]
            self.child_cset = cset
            return cset, branched

        self.ui.note(_('fetching revision log for "%s" from %d to %d\n') %
                     (self.module, from_revnum, to_revnum))

        try:
            firstcset = None
            lastonbranch = False
            stream = self._getlog([self.module], from_revnum, to_revnum)
            try:
                for entry in stream:
                    paths, revnum, author, date, message = entry
                    if revnum < self.startrev:
                        lastonbranch = True
                        break
                    if not paths:
                        self.ui.debug('revision %d has no entries\n' % revnum)
                        # If we ever leave the loop on an empty
                        # revision, do not try to get a parent branch
                        lastonbranch = lastonbranch or revnum == 0
                        continue
                    cset, lastonbranch = parselogentry(paths, revnum, author,
                                                       date, message)
                    if cset:
                        firstcset = cset
                    if lastonbranch:
                        break
            finally:
                stream.close()

            if not lastonbranch and firstcset and not firstcset.parents:
                # The first revision of the sequence (the last fetched one)
                # has invalid parents if not a branch root. Find the parent
                # revision now, if any.
                try:
                    firstrevnum = self.revnum(firstcset.rev)
                    if firstrevnum > 1:
                        latest = self.latest(self.module, firstrevnum - 1)
                        if latest:
                            firstcset.parents.append(latest)
                except SvnPathNotFound:
                    pass
        except svn.core.SubversionException as xxx_todo_changeme:
            (inst, num) = xxx_todo_changeme.args
            if num == svn.core.SVN_ERR_FS_NO_SUCH_REVISION:
                raise error.Abort(_('svn: branch has no revision %s')
                                 % to_revnum)
            raise

    def getfile(self, file, rev):
        # TODO: ra.get_file transmits the whole file instead of diffs.
        if file in self.removed:
            return None, None
        mode = ''
        try:
            new_module, revnum = revsplit(rev)[1:]
            if self.module != new_module:
                self.module = new_module
                self.reparent(self.module)
            io = stringio()
            info = svn.ra.get_file(self.ra, file, revnum, io)
            data = io.getvalue()
            # ra.get_file() seems to keep a reference on the input buffer
            # preventing collection. Release it explicitly.
            io.close()
            if isinstance(info, list):
                info = info[-1]
            mode = ("svn:executable" in info) and 'x' or ''
            mode = ("svn:special" in info) and 'l' or mode
        except svn.core.SubversionException as e:
            notfound = (svn.core.SVN_ERR_FS_NOT_FOUND,
                svn.core.SVN_ERR_RA_DAV_PATH_NOT_FOUND)
            if e.apr_err in notfound: # File not found
                return None, None
            raise
        if mode == 'l':
            link_prefix = "link "
            if data.startswith(link_prefix):
                data = data[len(link_prefix):]
        return data, mode

    def _iterfiles(self, path, revnum):
        """Enumerate all files in path at revnum, recursively."""
        path = path.strip('/')
        pool = svn.core.Pool()
        rpath = '/'.join([self.baseurl, quote(path)]).strip('/')
        entries = svn.client.ls(rpath, optrev(revnum), True, self.ctx, pool)
        if path:
            path += '/'
        return ((path + p) for p, e in entries.iteritems()
                if e.kind == svn.core.svn_node_file)

    def getrelpath(self, path, module=None):
        if module is None:
            module = self.module
        # Given the repository url of this wc, say
        #   "http://server/plone/CMFPlone/branches/Plone-2_0-branch"
        # extract the "entry" portion (a relative path) from what
        # svn log --xml says, i.e.
        #   "/CMFPlone/branches/Plone-2_0-branch/tests/PloneTestCase.py"
        # that is to say "tests/PloneTestCase.py"
        if path.startswith(module):
            relative = path.rstrip('/')[len(module):]
            if relative.startswith('/'):
                return relative[1:]
            elif relative == '':
                return relative

        # The path is outside our tracked tree...
        self.ui.debug('%r is not under %r, ignoring\n' % (path, module))
        return None

    def _checkpath(self, path, revnum, module=None):
        if module is not None:
            prevmodule = self.reparent('')
            path = module + '/' + path
        try:
            # ra.check_path does not like leading slashes very much, it leads
            # to PROPFIND subversion errors
            return svn.ra.check_path(self.ra, path.strip('/'), revnum)
        finally:
            if module is not None:
                self.reparent(prevmodule)

    def _getlog(self, paths, start, end, limit=0, discover_changed_paths=True,
                strict_node_history=False):
        # Normalize path names, svn >= 1.5 only wants paths relative to
        # supplied URL
        relpaths = []
        for p in paths:
            if not p.startswith('/'):
                p = self.module + '/' + p
            relpaths.append(p.strip('/'))
        args = [self.baseurl, relpaths, start, end, limit,
                discover_changed_paths, strict_node_history]
        # developer config: convert.svn.debugsvnlog
        if not self.ui.configbool('convert', 'svn.debugsvnlog'):
            return directlogstream(*args)
        arg = encodeargs(args)
        hgexe = util.hgexecutable()
        cmd = '%s debugsvnlog' % util.shellquote(hgexe)
        stdin, stdout = util.popen2(util.quotecommand(cmd))
        stdin.write(arg)
        try:
            stdin.close()
        except IOError:
            raise error.Abort(_('Mercurial failed to run itself, check'
                               ' hg executable is in PATH'))
        return logstream(stdout)

pre_revprop_change = '''#!/bin/sh

REPOS="$1"
REV="$2"
USER="$3"
PROPNAME="$4"
ACTION="$5"

if [ "$ACTION" = "M" -a "$PROPNAME" = "svn:log" ]; then exit 0; fi
if [ "$ACTION" = "A" -a "$PROPNAME" = "hg:convert-branch" ]; then exit 0; fi
if [ "$ACTION" = "A" -a "$PROPNAME" = "hg:convert-rev" ]; then exit 0; fi

echo "Changing prohibited revision property" >&2
exit 1
'''

class svn_sink(converter_sink, commandline):
    commit_re = re.compile(r'Committed revision (\d+).', re.M)
    uuid_re = re.compile(r'Repository UUID:\s*(\S+)', re.M)

    def prerun(self):
        if self.wc:
            os.chdir(self.wc)

    def postrun(self):
        if self.wc:
            os.chdir(self.cwd)

    def join(self, name):
        return os.path.join(self.wc, '.svn', name)

    def revmapfile(self):
        return self.join('hg-shamap')

    def authorfile(self):
        return self.join('hg-authormap')

    def __init__(self, ui, repotype, path):

        converter_sink.__init__(self, ui, repotype, path)
        commandline.__init__(self, ui, 'svn')
        self.delete = []
        self.setexec = []
        self.delexec = []
        self.copies = []
        self.wc = None
        self.cwd = pycompat.getcwd()

        created = False
        if os.path.isfile(os.path.join(path, '.svn', 'entries')):
            self.wc = os.path.realpath(path)
            self.run0('update')
        else:
            if not re.search(br'^(file|http|https|svn|svn\+ssh)\://', path):
                path = os.path.realpath(path)
                if os.path.isdir(os.path.dirname(path)):
                    if not os.path.exists(os.path.join(path, 'db', 'fs-type')):
                        ui.status(_('initializing svn repository %r\n') %
                                  os.path.basename(path))
                        commandline(ui, 'svnadmin').run0('create', path)
                        created = path
                    path = util.normpath(path)
                    if not path.startswith('/'):
                        path = '/' + path
                    path = 'file://' + path

            wcpath = os.path.join(pycompat.getcwd(), os.path.basename(path) +
                                '-wc')
            ui.status(_('initializing svn working copy %r\n')
                      % os.path.basename(wcpath))
            self.run0('checkout', path, wcpath)

            self.wc = wcpath
        self.opener = vfsmod.vfs(self.wc)
        self.wopener = vfsmod.vfs(self.wc)
        self.childmap = mapfile(ui, self.join('hg-childmap'))
        if util.checkexec(self.wc):
            self.is_exec = util.isexec
        else:
            self.is_exec = None

        if created:
            hook = os.path.join(created, 'hooks', 'pre-revprop-change')
            fp = open(hook, 'wb')
            fp.write(pre_revprop_change)
            fp.close()
            util.setflags(hook, False, True)

        output = self.run0('info')
        self.uuid = self.uuid_re.search(output).group(1).strip()

    def wjoin(self, *names):
        return os.path.join(self.wc, *names)

    @propertycache
    def manifest(self):
        # As of svn 1.7, the "add" command fails when receiving
        # already tracked entries, so we have to track and filter them
        # ourselves.
        m = set()
        output = self.run0('ls', recursive=True, xml=True)
        doc = xml.dom.minidom.parseString(output)
        for e in doc.getElementsByTagName('entry'):
            for n in e.childNodes:
                if n.nodeType != n.ELEMENT_NODE or n.tagName != 'name':
                    continue
                name = ''.join(c.data for c in n.childNodes
                               if c.nodeType == c.TEXT_NODE)
                # Entries are compared with names coming from
                # mercurial, so bytes with undefined encoding. Our
                # best bet is to assume they are in local
                # encoding. They will be passed to command line calls
                # later anyway, so they better be.
                m.add(encoding.unitolocal(name))
                break
        return m

    def putfile(self, filename, flags, data):
        if 'l' in flags:
            self.wopener.symlink(data, filename)
        else:
            try:
                if os.path.islink(self.wjoin(filename)):
                    os.unlink(filename)
            except OSError:
                pass
            self.wopener.write(filename, data)

            if self.is_exec:
                if self.is_exec(self.wjoin(filename)):
                    if 'x' not in flags:
                        self.delexec.append(filename)
                else:
                    if 'x' in flags:
                        self.setexec.append(filename)
                util.setflags(self.wjoin(filename), False, 'x' in flags)

    def _copyfile(self, source, dest):
        # SVN's copy command pukes if the destination file exists, but
        # our copyfile method expects to record a copy that has
        # already occurred.  Cross the semantic gap.
        wdest = self.wjoin(dest)
        exists = os.path.lexists(wdest)
        if exists:
            fd, tempname = tempfile.mkstemp(
                prefix='hg-copy-', dir=os.path.dirname(wdest))
            os.close(fd)
            os.unlink(tempname)
            os.rename(wdest, tempname)
        try:
            self.run0('copy', source, dest)
        finally:
            self.manifest.add(dest)
            if exists:
                try:
                    os.unlink(wdest)
                except OSError:
                    pass
                os.rename(tempname, wdest)

    def dirs_of(self, files):
        dirs = set()
        for f in files:
            if os.path.isdir(self.wjoin(f)):
                dirs.add(f)
            i = len(f)
            for i in iter(lambda: f.rfind('/', 0, i), -1):
                dirs.add(f[:i])
        return dirs

    def add_dirs(self, files):
        add_dirs = [d for d in sorted(self.dirs_of(files))
                    if d not in self.manifest]
        if add_dirs:
            self.manifest.update(add_dirs)
            self.xargs(add_dirs, 'add', non_recursive=True, quiet=True)
        return add_dirs

    def add_files(self, files):
        files = [f for f in files if f not in self.manifest]
        if files:
            self.manifest.update(files)
            self.xargs(files, 'add', quiet=True)
        return files

    def addchild(self, parent, child):
        self.childmap[parent] = child

    def revid(self, rev):
        return u"svn:%s@%s" % (self.uuid, rev)

    def putcommit(self, files, copies, parents, commit, source, revmap, full,
                  cleanp2):
        for parent in parents:
            try:
                return self.revid(self.childmap[parent])
            except KeyError:
                pass

        # Apply changes to working copy
        for f, v in files:
            data, mode = source.getfile(f, v)
            if data is None:
                self.delete.append(f)
            else:
                self.putfile(f, mode, data)
                if f in copies:
                    self.copies.append([copies[f], f])
        if full:
            self.delete.extend(sorted(self.manifest.difference(files)))
        files = [f[0] for f in files]

        entries = set(self.delete)
        files = frozenset(files)
        entries.update(self.add_dirs(files.difference(entries)))
        if self.copies:
            for s, d in self.copies:
                self._copyfile(s, d)
            self.copies = []
        if self.delete:
            self.xargs(self.delete, 'delete')
            for f in self.delete:
                self.manifest.remove(f)
            self.delete = []
        entries.update(self.add_files(files.difference(entries)))
        if self.delexec:
            self.xargs(self.delexec, 'propdel', 'svn:executable')
            self.delexec = []
        if self.setexec:
            self.xargs(self.setexec, 'propset', 'svn:executable', '*')
            self.setexec = []

        fd, messagefile = tempfile.mkstemp(prefix='hg-convert-')
        fp = os.fdopen(fd, pycompat.sysstr('wb'))
        fp.write(util.tonativeeol(commit.desc))
        fp.close()
        try:
            output = self.run0('commit',
                               username=util.shortuser(commit.author),
                               file=messagefile,
                               encoding='utf-8')
            try:
                rev = self.commit_re.search(output).group(1)
            except AttributeError:
                if parents and not files:
                    return parents[0]
                self.ui.warn(_('unexpected svn output:\n'))
                self.ui.warn(output)
                raise error.Abort(_('unable to cope with svn output'))
            if commit.rev:
                self.run('propset', 'hg:convert-rev', commit.rev,
                         revprop=True, revision=rev)
            if commit.branch and commit.branch != 'default':
                self.run('propset', 'hg:convert-branch', commit.branch,
                         revprop=True, revision=rev)
            for parent in parents:
                self.addchild(parent, rev)
            return self.revid(rev)
        finally:
            os.unlink(messagefile)

    def puttags(self, tags):
        self.ui.warn(_('writing Subversion tags is not yet implemented\n'))
        return None, None

    def hascommitfrommap(self, rev):
        # We trust that revisions referenced in a map still is present
        # TODO: implement something better if necessary and feasible
        return True

    def hascommitforsplicemap(self, rev):
        # This is not correct as one can convert to an existing subversion
        # repository and childmap would not list all revisions. Too bad.
        if rev in self.childmap:
            return True
        raise error.Abort(_('splice map revision %s not found in subversion '
                           'child map (revision lookups are not implemented)')
                         % rev)