view svnwrap/svn_swig_wrapper.py @ 225:2117cb0118fe

Get rid of .hg/svn/last_rev: We now calculate the last known revision by iterating over all known revisions and finding the highest number. Theoretically, we might be able to simply read the latest entry, but in practice, that's a bug waiting to happen. For instance, we might want to achieve compatibility with '.hg/shamap' as generated by the ConvertExtension, and it not only cannot offer a guarantee of linearity, but it also allows more than one conversion to source exists. I'd say we have other problems to care about until this turns up as a hotspot in profiling. Such as why we leak circa 100MB of memory per 1000 revisions converted ;)
author Dan Villiom Podlaski Christiansen <danchr@gmail.com>
date Fri, 27 Mar 2009 01:09:36 +0100
parents 2165461d2dd8
children f71af18c4379
line wrap: on
line source

import cStringIO
import getpass
import os
import shutil
import sys
import tempfile
import hashlib

from svn import client
from svn import core
from svn import delta
from svn import ra

def version():
    return '%d.%d.%d' % (core.SVN_VER_MAJOR, core.SVN_VER_MINOR, core.SVN_VER_MICRO)

if (core.SVN_VER_MAJOR, core.SVN_VER_MINOR, core.SVN_VER_MICRO) < (1, 5, 0): #pragma: no cover
    raise ImportError, 'You must have Subversion 1.5.0 or newer and matching SWIG bindings.'

class SubversionRepoCanNotReplay(Exception):
    """Exception raised when the svn server is too old to have replay.
    """

class SubversionRepoCanNotDiff(Exception):
    """Exception raised when the svn API diff3() command cannot be used
    """

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

svn_config = core.svn_config_get_config(None)
class RaCallbacks(ra.Callbacks):
    def open_tmp_file(self, pool): #pragma: no cover
        (fd, fn) = tempfile.mkstemp()
        os.close(fd)
        return fn

    def get_client_string(self, pool):
        return 'hgsubversion'


def user_pass_prompt(realm, default_username, ms, pool): #pragma: no cover
    creds = core.svn_auth_cred_simple_t()
    creds.may_save = ms
    if default_username:
        sys.stderr.write('Auth realm: %s\n' % (realm,))
        creds.username = default_username
    else:
        sys.stderr.write('Auth realm: %s\n' % (realm,))
        sys.stderr.write('Username: ')
        sys.stderr.flush()
        creds.username = sys.stdin.readline().strip()
    creds.password = getpass.getpass('Password for %s: ' % creds.username)
    return creds

def _create_auth_baton(pool):
    """Create a Subversion authentication baton. """
    # Give the client context baton a suite of authentication
    # providers.h
    platform_specific = ['svn_auth_get_gnome_keyring_simple_provider',
                       'svn_auth_get_gnome_keyring_ssl_client_cert_pw_provider',
                         'svn_auth_get_keychain_simple_provider',
                         'svn_auth_get_keychain_ssl_client_cert_pw_provider',
                         'svn_auth_get_kwallet_simple_provider',
                         'svn_auth_get_kwallet_ssl_client_cert_pw_provider',
                         'svn_auth_get_ssl_client_cert_file_provider',
                         'svn_auth_get_windows_simple_provider',
                         'svn_auth_get_windows_ssl_server_trust_provider',
                         ]

    providers = []

    for p in platform_specific:
        if hasattr(core, p):
            try:
                providers.append(getattr(core, p)())
            except RuntimeError:
                pass

    providers += [
        client.get_simple_provider(),
        client.get_username_provider(),
        client.get_ssl_client_cert_file_provider(),
        client.get_ssl_client_cert_pw_file_provider(),
        client.get_ssl_server_trust_file_provider(),
        client.get_simple_prompt_provider(user_pass_prompt, 2),
        ]

    return core.svn_auth_open(providers, pool)


class Revision(object):
    """Wrapper for a Subversion revision.
    """
    def __init__(self, revnum, author, message, date, paths, strip_path=''):
        self.revnum, self.author, self.message = revnum, author, message
        # TODO parse this into a datetime
        self.date = date
        self.paths = {}
        if paths:
            for p in paths:
                self.paths[p[len(strip_path):]] = paths[p]

    def __str__(self):
        return 'r%d by %s' % (self.revnum, self.author)

_svntypes = {
    core.svn_node_dir: 'd',
    core.svn_node_file: 'f',
    }

class SubversionRepo(object):
    """Wrapper for a Subversion repository.

    This uses the SWIG Python bindings, and will only work on svn >= 1.4.
    It takes a required param, the URL.
    """
    def __init__(self, url='', username=''):
        self.svn_url = url
        self.uname = username
        self.auth_baton_pool = core.Pool()
        self.auth_baton = _create_auth_baton(self.auth_baton_pool)

        self.init_ra_and_client()
        self.uuid = ra.get_uuid(self.ra, self.pool)
        repo_root = ra.get_repos_root(self.ra, self.pool)
        # *will* have a leading '/', would not if we used get_repos_root2
        self.subdir = url[len(repo_root):]
        if not self.subdir or self.subdir[-1] != '/':
            self.subdir += '/'
        self.hasdiff3 = True

    def init_ra_and_client(self):
        """Initializes the RA and client layers, because sometimes getting
        unified diffs runs the remote server out of open files.
        """
        # while we're in here we'll recreate our pool
        self.pool = core.Pool()
        self.client_context = client.create_context()

        self.client_context.auth_baton = self.auth_baton
        self.client_context.config = svn_config
        callbacks = RaCallbacks()
        callbacks.auth_baton = self.auth_baton
        self.callbacks = callbacks
        self.ra = ra.open2(self.svn_url.encode('utf-8'), callbacks,
                           svn_config, self.pool)

    def HEAD(self):
        return ra.get_latest_revnum(self.ra, self.pool)
    HEAD = property(HEAD)

    def START(self):
        return 0
    START = property(START)

    def branches(self):
        """Get the branches defined in this repo assuming a standard layout.
        """
        branches = self.list_dir('branches').keys()
        branch_info = {}
        head=self.HEAD
        for b in branches:
            b_path = 'branches/%s' %b
            hist_gen = self.fetch_history_at_paths([b_path], stop=head)
            hist = hist_gen.next()
            source, source_rev = self._get_copy_source(b_path, cached_head=head)
            # This if statement guards against projects that have non-ancestral
            # branches by not listing them has branches
            # Note that they probably are really ancestrally related, but there
            # is just no way for us to know how.
            if source is not None and source_rev is not None:
                branch_info[b] = (source, source_rev, hist.revnum)
        return branch_info
    branches = property(branches)

    def tags(self):
        """Get the current tags in this repo assuming a standard layout.

        This returns a dictionary of tag: (source path, source rev)
        """
        return self.tags_at_rev(self.HEAD)
    tags = property(tags)

    def tags_at_rev(self, revision):
        try:
            tags = self.list_dir('tags', revision=revision).keys()
        except core.SubversionException, e:
            if e.apr_err == core.SVN_ERR_FS_NOT_FOUND:
                return {}
            raise
        tag_info = {}
        for t in tags:
            tag_info[t] = self._get_copy_source('tags/%s' % t,
                                                cached_head=revision)
        return tag_info

    def _get_copy_source(self, path, cached_head=None):
        """Get copy revision for the given path, assuming it was meant to be
        a copy of the entire tree.
        """
        if not cached_head:
            cached_head = self.HEAD
        hist_gen = self.fetch_history_at_paths([path], stop=cached_head)
        hist = hist_gen.next()
        if hist.paths[path].copyfrom_path is None:
            return None, None
        source = hist.paths[path].copyfrom_path
        source_rev = 0
        for p in hist.paths:
            if hist.paths[p].copyfrom_rev:
                # We assume that the revision of the source tree as it was
                # copied was actually the revision of the highest revision
                # copied item. This could be wrong, but in practice it will
                # *probably* be correct
                if source_rev < hist.paths[p].copyfrom_rev:
                    source_rev = hist.paths[p].copyfrom_rev
        source = source[len(self.subdir):]
        return source, source_rev

    def list_dir(self, dir, revision=None):
        """List the contents of a server-side directory.

        Returns a dict-like object with one dict key per directory entry.

        Args:
          dir: the directory to list, no leading slash
          rev: the revision at which to list the directory, defaults to HEAD
        """
        if dir[-1] == '/':
            dir = dir[:-1]
        if revision is None:
            revision = self.HEAD
        r = ra.get_dir2(self.ra, dir, revision, core.SVN_DIRENT_KIND, self.pool)
        folders, props, junk = r
        return folders

    def revisions(self, start=None, chunk_size=1000):
        """Load the history of this repo.

        This is LAZY. It returns a generator, and fetches a small number
        of revisions at a time.

        The reason this is lazy is so that you can use the same repo object
        to perform RA calls to get deltas.
        """
        # NB: you'd think this would work, but you'd be wrong. I'm pretty
        # convinced there must be some kind of svn bug here.
        #return self.fetch_history_at_paths(['tags', 'trunk', 'branches'],
        #                                   start=start)
        # this does the same thing, but at the repo root + filtering. It's
        # kind of tough cookies, sadly.
        for r in self.fetch_history_at_paths([''], start=start,
                                             chunk_size=chunk_size):
            should_yield = False
            i = 0
            paths = list(r.paths.keys())
            while i < len(paths) and not should_yield:
                p = paths[i]
                if (p.startswith('trunk') or p.startswith('tags')
                    or p.startswith('branches')):
                    should_yield = True
                i += 1
            if should_yield:
                yield r


    def fetch_history_at_paths(self, paths, start=None, stop=None,
                               chunk_size=1000):
        revisions = []
        def callback(paths, revnum, author, date, message, pool):
            r = Revision(revnum, author, message, date, paths,
                         strip_path=self.subdir)
            revisions.append(r)
        if not start:
            start = self.START
        if not stop:
            stop = self.HEAD
        while stop > start:
            ra.get_log(self.ra,
                       paths,
                       start+1,
                       stop,
                       chunk_size, #limit of how many log messages to load
                       True, # don't need to know changed paths
                       True, # stop on copies
                       callback,
                       self.pool)
            if len(revisions) < chunk_size:
                # this means there was no history for the path, so force the
                # loop to exit
                start = stop
            else:
                start = revisions[-1].revnum
            while len(revisions) > 0:
                yield revisions[0]
                revisions.pop(0)

    def commit(self, paths, message, file_data, base_revision, addeddirs,
               deleteddirs, properties, copies):
        """Commits the appropriate targets from revision in editor's store.
        """
        self.init_ra_and_client()
        commit_info = []
        def commit_cb(_commit_info, pool):
            commit_info.append(_commit_info)
        editor, edit_baton = ra.get_commit_editor2(self.ra,
                                                   message,
                                                   commit_cb,
                                                   None,
                                                   False,
                                                   self.pool)
        checksum = []
        # internal dir batons can fall out of scope and get GCed before svn is
        # done with them. This prevents that (credit to gvn for the idea).
        batons = [edit_baton, ]
        def driver_cb(parent, path, pool):
            if not parent:
                bat = editor.open_root(edit_baton, base_revision, self.pool)
                batons.append(bat)
                return bat
            if path in deleteddirs:
                bat = editor.delete_entry(path, base_revision, parent, pool)
                batons.append(bat)
                return bat
            if path not in file_data:
                if path in addeddirs:
                    bat = editor.add_directory(path, parent, None, -1, pool)
                else:
                    bat = editor.open_directory(path, parent, base_revision, pool)
                batons.append(bat)
                props = properties.get(path, {})
                if 'svn:externals' in props:
                    value = props['svn:externals']
                    editor.change_dir_prop(bat, 'svn:externals', value, pool)
                return bat
            base_text, new_text, action = file_data[path]
            compute_delta = True
            if action == 'modify':
                baton = editor.open_file(path, parent, base_revision, pool)
            elif action == 'add':
                try:
                    frompath, fromrev = copies.get(path, (None, -1))
                    if frompath:
                        frompath = self.svn_url + '/' + frompath
                    baton = editor.add_file(path, parent, frompath, fromrev, pool)
                except (core.SubversionException, TypeError), e: #pragma: no cover
                    print e.message
                    raise
            elif action == 'delete':
                baton = editor.delete_entry(path, base_revision, parent, pool)
                compute_delta = False

            if path in properties:
                if properties[path].get('svn:special', None):
                    new_text = 'link %s' % new_text
                for p, v in properties[path].iteritems():
                    editor.change_file_prop(baton, p, v)

            if compute_delta:
                handler, wh_baton = editor.apply_textdelta(baton, None,
                                                           self.pool)

                txdelta_stream = delta.svn_txdelta(
                    cStringIO.StringIO(base_text), cStringIO.StringIO(new_text),
                    self.pool)
                delta.svn_txdelta_send_txstream(txdelta_stream, handler,
                                                wh_baton, pool)

                # TODO pass md5(new_text) instead of None
                editor.close_file(baton, None, pool)

        delta.path_driver(editor, edit_baton, base_revision, paths, driver_cb,
                          self.pool)
        editor.close_edit(edit_baton, self.pool)

    def get_replay(self, revision, editor, oldest_rev_i_have=0):
        # this method has a tendency to chew through RAM if you don't re-init
        self.init_ra_and_client()
        e_ptr, e_baton = delta.make_editor(editor)
        try:
            ra.replay(self.ra, revision, oldest_rev_i_have, True, e_ptr,
                      e_baton, self.pool)
        except core.SubversionException, e: #pragma: no cover
            # can I depend on this number being constant?
            if (e.apr_err == core.SVN_ERR_RA_NOT_IMPLEMENTED or
                e.apr_err == core.SVN_ERR_UNSUPPORTED_FEATURE):
                raise SubversionRepoCanNotReplay, ('This Subversion server '
                   'is older than 1.4.0, and cannot satisfy replay requests.')
            else:
                raise

    def get_unified_diff(self, path, revision, other_path=None, other_rev=None,
                         deleted=True, ignore_type=False):
        """Gets a unidiff of path at revision against revision-1.
        """
        if not self.hasdiff3:
            raise SubversionRepoCanNotDiff()
        # works around an svn server keeping too many open files (observed
        # in an svnserve from the 1.2 era)
        self.init_ra_and_client()

        assert path[0] != '/'
        url = self.svn_url + '/' + path
        url2 = url
        if other_path is not None:
            url2 = self.svn_url + '/' + other_path
        if other_rev is None:
            other_rev = revision - 1
        old_cwd = os.getcwd()
        tmpdir = tempfile.mkdtemp('svnwrap_temp')
        out, err = None, None
        try:
            # hot tip: the swig bridge doesn't like StringIO for these bad boys
            out_path = os.path.join(tmpdir, 'diffout')
            error_path = os.path.join(tmpdir, 'differr')
            out = open(out_path, 'w')
            err = open(error_path, 'w')
            try:
                client.diff3([], url2, optrev(other_rev), url, optrev(revision),
                             True, True, deleted, ignore_type, 'UTF-8', out, err,
                             self.client_context, self.pool)
            except core.SubversionException, e:
                # "Can't write to stream: The handle is invalid."
                # This error happens systematically under Windows, possibly
                # related to file handles being non-write shareable by default.
                if e.apr_err != 720006:
                    raise
                self.hasdiff3 = False
                raise SubversionRepoCanNotDiff()
            out.close()
            err.close()
            out, err = None, None
            assert len(open(error_path).read()) == 0
            diff = open(out_path).read()
            return diff
        finally:
            if out: out.close()
            if err: err.close()
            shutil.rmtree(tmpdir)
            os.chdir(old_cwd)

    def get_file(self, path, revision):
        """Return content and mode of file at given path and revision.

        "link " prefix is dropped from symlink content. Mode is 'x' if
        file is executable, 'l' if a symlink, the empty string
        otherwise. If the file does not exist at this revision, raise
        IOError.
        """
        mode = ''
        try:
            out = cStringIO.StringIO()
            info = ra.get_file(self.ra, path, revision, out)
            data = out.getvalue()
            out.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 core.SubversionException, e:
            notfound = (core.SVN_ERR_FS_NOT_FOUND,
                        core.SVN_ERR_RA_DAV_PATH_NOT_FOUND)
            if e.apr_err in notfound: # File not found
                raise IOError()
            raise
        if mode  == 'l':
            linkprefix = "link "
            if data.startswith(linkprefix):
                data = data[len(linkprefix):]
        return data, mode

    def list_props(self, path, revision):
        """Return a mapping of property names to values, raise IOError if
        specified path does not exist.
        """
        self.init_ra_and_client()
        rev = optrev(revision)
        rpath = (self.svn_url + '/' + path.strip('/')).strip('/')
        try:
            pl = client.proplist2(rpath, rev, rev, False,
                                  self.client_context, self.pool)
        except core.SubversionException, e:
            # Specified path does not exist at this revision
            if e.apr_err == core.SVN_ERR_NODE_UNKNOWN_KIND:
                raise IOError()
            raise
        if not pl:
            return {}
        return pl[0][1]

    def fetch_all_files_to_dir(self, path, revision, checkout_path):
        rev = optrev(revision)
        client.export3(self.svn_url+'/'+path, checkout_path, rev,
                       rev, True, True, True, 'LF', # should be 'CRLF' on win32
                       self.client_context, self.pool)

    def list_files(self, dirpath, revision):
        """List the content of a directory at a given revision, recursively.

        Yield tuples (path, kind) where 'path' is the entry path relatively to
        'dirpath' and 'kind' is 'f' if the entry is a file, 'd' if it is a
        directory. Raise IOError if the directory cannot be found at given
        revision.
        """
        dirpath = dirpath.strip('/')
        pool = core.Pool()
        rpath = '/'.join([self.svn_url, dirpath]).strip('/')
        rev = optrev(revision)
        try:
            entries = client.ls(rpath, rev, True, self.client_context, pool)
        except core.SubversionException, e:
            if e.apr_err == core.SVN_ERR_FS_NOT_FOUND:
                raise IOError('%s cannot be found at r%d' % (dirpath, revision))
            raise
        for path, e in entries.iteritems():
            kind = _svntypes.get(e.kind)
            yield path, kind

    def checkpath(self, path, revision):
        """Return the entry type at the given revision, 'f', 'd' or None
        if the entry does not exist.
        """
        kind = ra.check_path(self.ra, path.strip('/'), revision)
        return _svntypes.get(kind)