~bzr-pqm/bzr/bzr.dev

# Copyright (C) 2005, 2006, 2007 Canonical Ltd

#

# This program is free software; you can redistribute it and/or modify

# it under the terms of the GNU General Public License as published by

# the Free Software Foundation; either version 2 of the License, or

# (at your option) any later version.

#

# This program is distributed in the hope that it will be useful,

# but WITHOUT ANY WARRANTY; without even the implied warranty of

# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

# GNU General Public License for more details.

#

# You should have received a copy of the GNU General Public License

# along with this program; if not, write to the Free Software

# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

"""Transport is an abstraction layer to handle file access.

The abstraction is to allow access from the local filesystem, as well

as remote (such as http or sftp).

Transports are constructed from a string, being a URL or (as a degenerate

case) a local filesystem path.  This is typically the top directory of

a bzrdir, repository, or similar object we are interested in working with.

The Transport returned has methods to read, write and manipulate files within

it.

"""

from cStringIO import StringIO

import re

import sys

from bzrlib.lazy_import import lazy_import

lazy_import(globals(), """

import errno

from collections import deque

from copy import deepcopy

from stat import S_ISDIR

import unittest

import urllib

import urlparse

import warnings

import bzrlib

from bzrlib import (

    errors,

    osutils,

    symbol_versioning,

    urlutils,

    )

""")

from bzrlib.symbol_versioning import (

        deprecated_passed,

        deprecated_method,

        deprecated_function,

        DEPRECATED_PARAMETER,

        zero_eight,

        zero_eleven,

        )

from bzrlib.trace import (

    note,

    mutter,

    warning,

    )

from bzrlib import registry

def _get_protocol_handlers():

    """Return a dictionary of {urlprefix: [factory]}"""

    return transport_list_registry

def _set_protocol_handlers(new_handlers):

    """Replace the current protocol handlers dictionary.

    WARNING this will remove all build in protocols. Use with care.

    """

    global transport_list_registry

    transport_list_registry = new_handlers

def _clear_protocol_handlers():

    global transport_list_registry

    transport_list_registry = TransportListRegistry()

def _get_transport_modules():

    """Return a list of the modules providing transports."""

    modules = set()

    for prefix, factory_list in transport_list_registry.iteritems():

        for factory in factory_list:

            if hasattr(factory, "_module_name"):

                modules.add(factory._module_name)

            else:

                modules.add(factory._obj.__module__)

    # Add chroot directly, because there is not handler registered for it.

    modules.add('bzrlib.transport.chroot')

    result = list(modules)

    result.sort()

    return result

class TransportListRegistry(registry.Registry):

    """A registry which simplifies tracking available Transports.

    A registration of a new protocol requires two step:

    1) register the prefix with the function register_transport( )

    2) register the protocol provider with the function

    register_transport_provider( ) ( and the "lazy" variant )

    This in needed because:

    a) a single provider can support multple protcol ( like the ftp

    privider which supports both the ftp:// and the aftp:// protocols )

    b) a single protocol can have multiple providers ( like the http://

    protocol which is supported by both the urllib and pycurl privider )

    """

    def register_transport_provider(self, key, obj):

        self.get(key).insert(0, registry._ObjectGetter(obj))

    def register_lazy_transport_provider(self, key, module_name, member_name):

        self.get(key).insert(0, 

                registry._LazyObjectGetter(module_name, member_name))

    def register_transport(self, key, help=None, info=None):

        self.register(key, [], help, info)

    def set_default_transport(self, key=None):

        """Return either 'key' or the default key if key is None"""

        self._default_key = key

transport_list_registry = TransportListRegistry( )

def register_transport_proto(prefix, help=None, info=None):

    transport_list_registry.register_transport(prefix, help, info)

def register_lazy_transport(prefix, module, classname):

    if not prefix in transport_list_registry:

        register_transport_proto(prefix)

    transport_list_registry.register_lazy_transport_provider(prefix, module, classname)

def register_transport(prefix, klass, override=DEPRECATED_PARAMETER):

    if not prefix in transport_list_registry:

        register_transport_proto(prefix)

    transport_list_registry.register_transport_provider(prefix, klass)

def register_urlparse_netloc_protocol(protocol):

    """Ensure that protocol is setup to be used with urlparse netloc parsing."""

    if protocol not in urlparse.uses_netloc:

        urlparse.uses_netloc.append(protocol)

def unregister_transport(scheme, factory):

    """Unregister a transport."""

    l = transport_list_registry.get(scheme)

    for i in l:

        o = i.get_obj( )

        if o == factory:

            transport_list_registry.get(scheme).remove(i)

            break

    if len(l) == 0:

        transport_list_registry.remove(scheme)

def split_url(url):

    # TODO: jam 20060606 urls should only be ascii, or they should raise InvalidURL

    if isinstance(url, unicode):

        url = url.encode('utf-8')

    (scheme, netloc, path, params,

     query, fragment) = urlparse.urlparse(url, allow_fragments=False)

    username = password = host = port = None

    if '@' in netloc:

        username, host = netloc.split('@', 1)

        if ':' in username:

            username, password = username.split(':', 1)

            password = urllib.unquote(password)

        username = urllib.unquote(username)

    else:

        host = netloc

    if ':' in host:

        host, port = host.rsplit(':', 1)

        try:

            port = int(port)

        except ValueError:

            # TODO: Should this be ConnectionError?

            raise errors.TransportError(

                'invalid port number %s in url:\n%s' % (port, url))

    host = urllib.unquote(host)

    path = urllib.unquote(path)

    return (scheme, username, password, host, port, path)

class _CoalescedOffset(object):

    """A data container for keeping track of coalesced offsets."""

    __slots__ = ['start', 'length', 'ranges']

    def __init__(self, start, length, ranges):

        self.start = start

        self.length = length

        self.ranges = ranges

    def __cmp__(self, other):

        return cmp((self.start, self.length, self.ranges),

                   (other.start, other.length, other.ranges))

class Transport(object):

    """This class encapsulates methods for retrieving or putting a file

    from/to a storage location.

    Most functions have a _multi variant, which allows you to queue up

    multiple requests. They generally have a dumb base implementation 

    which just iterates over the arguments, but smart Transport

    implementations can do pipelining.

    In general implementations should support having a generator or a list

    as an argument (ie always iterate, never index)

    :ivar base: Base URL for the transport; should always end in a slash.

    """

    # implementations can override this if it is more efficient

    # for them to combine larger read chunks together

    _max_readv_combine = 50

    # It is better to read this much more data in order, rather

    # than doing another seek. Even for the local filesystem,

    # there is a benefit in just reading.

    # TODO: jam 20060714 Do some real benchmarking to figure out

    #       where the biggest benefit between combining reads and

    #       and seeking is. Consider a runtime auto-tune.

    _bytes_to_read_before_seek = 0

    def __init__(self, base):

        super(Transport, self).__init__()

        self.base = base

    def _translate_error(self, e, path, raise_generic=True):

        """Translate an IOError or OSError into an appropriate bzr error.

        This handles things like ENOENT, ENOTDIR, EEXIST, and EACCESS

        """

        if getattr(e, 'errno', None) is not None:

            if e.errno in (errno.ENOENT, errno.ENOTDIR):

                raise errors.NoSuchFile(path, extra=e)

            # I would rather use errno.EFOO, but there doesn't seem to be

            # any matching for 267

            # This is the error when doing a listdir on a file:

            # WindowsError: [Errno 267] The directory name is invalid

            if sys.platform == 'win32' and e.errno in (errno.ESRCH, 267):

                raise errors.NoSuchFile(path, extra=e)

            if e.errno == errno.EEXIST:

                raise errors.FileExists(path, extra=e)

            if e.errno == errno.EACCES:

                raise errors.PermissionDenied(path, extra=e)

            if e.errno == errno.ENOTEMPTY:

                raise errors.DirectoryNotEmpty(path, extra=e)

            if e.errno == errno.EBUSY:

                raise errors.ResourceBusy(path, extra=e)

        if raise_generic:

            raise errors.TransportError(orig_error=e)

    def clone(self, offset=None):

        """Return a new Transport object, cloned from the current location,

        using a subdirectory or parent directory. This allows connections 

        to be pooled, rather than a new one needed for each subdir.

        """

        raise NotImplementedError(self.clone)

    def ensure_base(self):

        """Ensure that the directory this transport references exists.

        This will create a directory if it doesn't exist.

        :return: True if the directory was created, False otherwise.

        """

        # The default implementation just uses "Easier to ask for forgiveness

        # than permission". We attempt to create the directory, and just

        # suppress a FileExists exception.

        try:

            self.mkdir('.')

        except errors.FileExists:

            return False

        else:

            return True

    def should_cache(self):

        """Return True if the data pulled across should be cached locally.

        """

        return False

    def _pump(self, from_file, to_file):

        """Most children will need to copy from one file-like 

        object or string to another one.

        This just gives them something easy to call.

        """

        assert not isinstance(from_file, basestring), \

            '_pump should only be called on files not %s' % (type(from_file,))

        osutils.pumpfile(from_file, to_file)

    def _get_total(self, multi):

        """Try to figure out how many entries are in multi,

        but if not possible, return None.

        """

        try:

            return len(multi)

        except TypeError: # We can't tell how many, because relpaths is a generator

            return None

    def _update_pb(self, pb, msg, count, total):

        """Update the progress bar based on the current count

        and total available, total may be None if it was

        not possible to determine.

        """

        if pb is None:

            return

        if total is None:

            pb.update(msg, count, count+1)

        else:

            pb.update(msg, count, total)

    def _iterate_over(self, multi, func, pb, msg, expand=True):

        """Iterate over all entries in multi, passing them to func,

        and update the progress bar as you go along.

        :param expand:  If True, the entries will be passed to the function

                        by expanding the tuple. If False, it will be passed

                        as a single parameter.

        """

        total = self._get_total(multi)

        result = []

        count = 0

        for entry in multi:

            self._update_pb(pb, msg, count, total)

            if expand:

                result.append(func(*entry))

            else:

                result.append(func(entry))

            count += 1

        return tuple(result)

    def abspath(self, relpath):

        """Return the full url to the given relative path.

        :param relpath: a string of a relative path

        """

        # XXX: Robert Collins 20051016 - is this really needed in the public

        # interface ?

        raise NotImplementedError(self.abspath)

    def _combine_paths(self, base_path, relpath):

        """Transform a Transport-relative path to a remote absolute path.

        This does not handle substitution of ~ but does handle '..' and '.'

        components.

        Examples::

            t._combine_paths('/home/sarah', 'project/foo')

                => '/home/sarah/project/foo'

            t._combine_paths('/home/sarah', '../../etc')

                => '/etc'

            t._combine_paths('/home/sarah', '/etc')

                => '/etc'

        :param base_path: urlencoded path for the transport root; typically a 

             URL but need not contain scheme/host/etc.

        :param relpath: relative url string for relative part of remote path.

        :return: urlencoded string for final path.

        """

        # FIXME: share the common code across more transports; variants of

        # this likely occur in http and sftp too.

        #

        # TODO: Also need to consider handling of ~, which might vary between

        # transports?

        if not isinstance(relpath, str):

            raise errors.InvalidURL("not a valid url: %r" % relpath)

        if relpath.startswith('/'):

            base_parts = []

        else:

            base_parts = base_path.split('/')

        if len(base_parts) > 0 and base_parts[-1] == '':

            base_parts = base_parts[:-1]

        for p in relpath.split('/'):

            if p == '..':

                if len(base_parts) == 0:

                    # In most filesystems, a request for the parent

                    # of root, just returns root.

                    continue

                base_parts.pop()

            elif p == '.':

                continue # No-op

            elif p != '':

                base_parts.append(p)

        path = '/'.join(base_parts)

        if not path.startswith('/'):

            path = '/' + path

        return path

    def relpath(self, abspath):

        """Return the local path portion from a given absolute path.

        This default implementation is not suitable for filesystems with

        aliasing, such as that given by symlinks, where a path may not 

        start with our base, but still be a relpath once aliasing is 

        resolved.

        """

        # TODO: This might want to use bzrlib.osutils.relpath

        #       but we have to watch out because of the prefix issues

        if not (abspath == self.base[:-1] or abspath.startswith(self.base)):

            raise errors.PathNotChild(abspath, self.base)

        pl = len(self.base)

        return abspath[pl:].strip('/')

    def local_abspath(self, relpath):

        """Return the absolute path on the local filesystem.

        This function will only be defined for Transports which have a

        physical local filesystem representation.

        """

        raise errors.NotLocalUrl(self.abspath(relpath))

    def has(self, relpath):

        """Does the file relpath exist?

        Note that some transports MAY allow querying on directories, but this

        is not part of the protocol.  In other words, the results of 

        t.has("a_directory_name") are undefined.

        :rtype: bool

        """

        raise NotImplementedError(self.has)

    def has_multi(self, relpaths, pb=None):

        """Return True/False for each entry in relpaths"""

        total = self._get_total(relpaths)

        count = 0

        for relpath in relpaths:

            self._update_pb(pb, 'has', count, total)

            yield self.has(relpath)

            count += 1

    def has_any(self, relpaths):

        """Return True if any of the paths exist."""

        for relpath in relpaths:

            if self.has(relpath):

                return True

        return False

    def iter_files_recursive(self):

        """Iter the relative paths of files in the transports sub-tree.

        *NOTE*: This only lists *files*, not subdirectories!

        As with other listing functions, only some transports implement this,.

        you may check via is_listable to determine if it will.

        """

        raise errors.TransportNotPossible("This transport has not "

                                          "implemented iter_files_recursive "

                                          "(but must claim to be listable "

                                          "to trigger this error).")

    def get(self, relpath):

        """Get the file at the given relative path.

        :param relpath: The relative path to the file

        :rtype: File-like object.

        """

        raise NotImplementedError(self.get)

    def get_bytes(self, relpath):

        """Get a raw string of the bytes for a file at the given location.

        :param relpath: The relative path to the file

        """

        return self.get(relpath).read()

    def get_smart_client(self):

        """Return a smart client for this transport if possible.

        A smart client doesn't imply the presence of a smart server: it implies

        that the smart protocol can be tunnelled via this transport.

        :raises NoSmartServer: if no smart server client is available.

        """

        raise errors.NoSmartServer(self.base)

    def get_smart_medium(self):

        """Return a smart client medium for this transport if possible.

        A smart medium doesn't imply the presence of a smart server: it implies

        that the smart protocol can be tunnelled via this transport.

        :raises NoSmartMedium: if no smart server medium is available.

        """

        raise errors.NoSmartMedium(self)

    def readv(self, relpath, offsets):

        """Get parts of the file at the given relative path.

        :offsets: A list of (offset, size) tuples.

        :return: A list or generator of (offset, data) tuples

        """

        if not offsets:

            return

        fp = self.get(relpath)

        return self._seek_and_read(fp, offsets, relpath)

    def _seek_and_read(self, fp, offsets, relpath='<unknown>'):

        """An implementation of readv that uses fp.seek and fp.read.

        This uses _coalesce_offsets to issue larger reads and fewer seeks.

        :param fp: A file-like object that supports seek() and read(size)

        :param offsets: A list of offsets to be read from the given file.

        :return: yield (pos, data) tuples for each request

        """

        # We are going to iterate multiple times, we need a list

        offsets = list(offsets)

        sorted_offsets = sorted(offsets)

        # turn the list of offsets into a stack

        offset_stack = iter(offsets)

        cur_offset_and_size = offset_stack.next()

        coalesced = self._coalesce_offsets(sorted_offsets,

                               limit=self._max_readv_combine,

                               fudge_factor=self._bytes_to_read_before_seek)

        # Cache the results, but only until they have been fulfilled

        data_map = {}

        for c_offset in coalesced:

            # TODO: jam 20060724 it might be faster to not issue seek if 

            #       we are already at the right location. This should be

            #       benchmarked.

            fp.seek(c_offset.start)

            data = fp.read(c_offset.length)

            if len(data) < c_offset.length:

                raise errors.ShortReadvError(relpath, c_offset.start,

                            c_offset.length, actual=len(data))

            for suboffset, subsize in c_offset.ranges:

                key = (c_offset.start+suboffset, subsize)

                data_map[key] = data[suboffset:suboffset+subsize]

            # Now that we've read some data, see if we can yield anything back

            while cur_offset_and_size in data_map:

                this_data = data_map.pop(cur_offset_and_size)

                yield cur_offset_and_size[0], this_data

                cur_offset_and_size = offset_stack.next()

    @staticmethod

    def _coalesce_offsets(offsets, limit, fudge_factor):

        """Yield coalesced offsets.

        With a long list of neighboring requests, combine them

        into a single large request, while retaining the original

        offsets.

        Turns  [(15, 10), (25, 10)] => [(15, 20, [(0, 10), (10, 10)])]

        :param offsets: A list of (start, length) pairs

        :param limit: Only combine a maximum of this many pairs

                      Some transports penalize multiple reads more than

                      others, and sometimes it is better to return early.

                      0 means no limit

        :param fudge_factor: All transports have some level of 'it is

                better to read some more data and throw it away rather 

                than seek', so collapse if we are 'close enough'

        :return: yield _CoalescedOffset objects, which have members for wher

                to start, how much to read, and how to split those 

                chunks back up

        """

        last_end = None

        cur = _CoalescedOffset(None, None, [])

        for start, size in offsets:

            end = start + size

            if (last_end is not None 

                and start <= last_end + fudge_factor

                and start >= cur.start

                and (limit <= 0 or len(cur.ranges) < limit)):

                cur.length = end - cur.start

                cur.ranges.append((start-cur.start, size))

            else:

                if cur.start is not None:

                    yield cur

                cur = _CoalescedOffset(start, size, [(0, size)])

            last_end = end

        if cur.start is not None:

            yield cur

        return

    def get_multi(self, relpaths, pb=None):

        """Get a list of file-like objects, one for each entry in relpaths.

        :param relpaths: A list of relative paths.

        :param pb:  An optional ProgressBar for indicating percent done.

        :return: A list or generator of file-like objects

        """

        # TODO: Consider having this actually buffer the requests,

        # in the default mode, it probably won't give worse performance,

        # and all children wouldn't have to implement buffering

        total = self._get_total(relpaths)

        count = 0

        for relpath in relpaths:

            self._update_pb(pb, 'get', count, total)

            yield self.get(relpath)

            count += 1

    @deprecated_method(zero_eleven)

    def put(self, relpath, f, mode=None):

        """Copy the file-like object into the location.

        :param relpath: Location to put the contents, relative to base.

        :param f:       File-like object.

        :param mode: The mode for the newly created file, 

                     None means just use the default

        """

        if isinstance(f, str):

            return self.put_bytes(relpath, f, mode=mode)

        else:

            return self.put_file(relpath, f, mode=mode)

    def put_bytes(self, relpath, bytes, mode=None):

        """Atomically put the supplied bytes into the given location.

        :param relpath: The location to put the contents, relative to the

            transport base.

        :param bytes: A bytestring of data.

        :param mode: Create the file with the given mode.

        :return: None

        """

        if not isinstance(bytes, str):

            raise AssertionError(

                'bytes must be a plain string, not %s' % type(bytes))

        return self.put_file(relpath, StringIO(bytes), mode=mode)

    def put_bytes_non_atomic(self, relpath, bytes, mode=None,

                             create_parent_dir=False,

                             dir_mode=None):

        """Copy the string into the target location.

        This function is not strictly safe to use. See 

        Transport.put_bytes_non_atomic for more information.

        :param relpath: The remote location to put the contents.

        :param bytes:   A string object containing the raw bytes to write into

                        the target file.

        :param mode:    Possible access permissions for new file.

                        None means do not set remote permissions.

        :param create_parent_dir: If we cannot create the target file because

                        the parent directory does not exist, go ahead and

                        create it, and then try again.

        :param dir_mode: Possible access permissions for new directories.

        """

        if not isinstance(bytes, str):

            raise AssertionError(

                'bytes must be a plain string, not %s' % type(bytes))

        self.put_file_non_atomic(relpath, StringIO(bytes), mode=mode,

                                 create_parent_dir=create_parent_dir,

                                 dir_mode=dir_mode)

    def put_file(self, relpath, f, mode=None):

        """Copy the file-like object into the location.

        :param relpath: Location to put the contents, relative to base.

        :param f:       File-like object.

        :param mode: The mode for the newly created file,

                     None means just use the default.

        """

        # We would like to mark this as NotImplemented, but most likely

        # transports have defined it in terms of the old api.

        symbol_versioning.warn('Transport %s should implement put_file,'

                               ' rather than implementing put() as of'

                               ' version 0.11.'

                               % (self.__class__.__name__,),

                               DeprecationWarning)

        return self.put(relpath, f, mode=mode)

        #raise NotImplementedError(self.put_file)

    def put_file_non_atomic(self, relpath, f, mode=None,

                            create_parent_dir=False,

                            dir_mode=None):

        """Copy the file-like object into the target location.

        This function is not strictly safe to use. It is only meant to

        be used when you already know that the target does not exist.

        It is not safe, because it will open and truncate the remote

        file. So there may be a time when the file has invalid contents.

        :param relpath: The remote location to put the contents.

        :param f:       File-like object.

        :param mode:    Possible access permissions for new file.

                        None means do not set remote permissions.