~bzr-pqm/bzr/bzr.dev

# Copyright (C) 2005, 2006 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.

"""

import errno

from collections import deque

from copy import deepcopy

from cStringIO import StringIO

import re

from stat import S_ISDIR

import sys

from unittest import TestSuite

import urllib

import urlparse

import warnings

import bzrlib

from bzrlib import (

    errors,

    osutils,

    symbol_versioning,

    urlutils,

    )

from bzrlib.errors import DependencyNotPresent

from bzrlib.osutils import pumpfile

from bzrlib.symbol_versioning import (

        deprecated_passed,

        deprecated_method,

        deprecated_function,

        DEPRECATED_PARAMETER,

        zero_eight,

        zero_eleven,

        )

from bzrlib.trace import mutter, warning

# {prefix: [transport_classes]}

# Transports are inserted onto the list LIFO and tried in order; as a result

# transports provided by plugins are tried first, which is usually what we

# want.

_protocol_handlers = {

}

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

    """Register a transport that can be used to open URLs

    Normally you should use register_lazy_transport, which defers loading the

    implementation until it's actually used, and so avoids pulling in possibly

    large implementation libraries.

    """

    # Note that this code runs very early in library setup -- trace may not be

    # working, etc.

    global _protocol_handlers

    if deprecated_passed(override):

        warnings.warn("register_transport(override) is deprecated")

    _protocol_handlers.setdefault(prefix, []).insert(0, klass)

def register_lazy_transport(scheme, module, classname):

    """Register lazy-loaded transport class.

    When opening a URL with the given scheme, load the module and then

    instantiate the particular class.  

    If the module raises DependencyNotPresent when it's imported, it is

    skipped and another implementation of the protocol is tried.  This is

    intended to be used when the implementation depends on an external

    implementation that may not be present.  If any other error is raised, it

    propagates up and the attempt to open the url fails.

    """

    # TODO: If no implementation of a protocol is available because of missing

    # dependencies, we should perhaps show the message about what dependency

    # was missing.

    def _loader(base):

        mod = __import__(module, globals(), locals(), [classname])

        klass = getattr(mod, classname)

        return klass(base)

    _loader.module = module

    register_transport(scheme, _loader)

def _get_protocol_handlers():

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

    return _protocol_handlers

def _set_protocol_handlers(new_handlers):

    """Replace the current protocol handlers dictionary.

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

    """

    global _protocol_handlers

    _protocol_handlers = new_handlers

def _clear_protocol_handlers():

    global _protocol_handlers

    _protocol_handlers = {}

def _get_transport_modules():

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

    modules = set()

    for prefix, factory_list in _protocol_handlers.items():

        for factory in factory_list:

            if factory.__module__ == "bzrlib.transport":

                # this is a lazy load transport, because no real ones

                # are directlry in bzrlib.transport

                modules.add(factory.module)

            else:

                modules.add(factory.__module__)

    result = list(modules)

    result.sort()

    return result

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 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('%s: invalid port number' % port)

    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)

    """

    # 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 hasattr(e, 'errno'):

            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 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.

        """

        if isinstance(from_file, basestring):

            to_file.write(from_file)

        else:

            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.

        This can be supplied with a string or a list

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

             interface ?

        """

        raise NotImplementedError(self.abspath)

    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.

        """

        # TODO: jam 20060426 Should this raise NotLocalUrl instead?

        raise errors.TransportNotPossible('This is not a LocalTransport,'

            ' so there is no local representation for a path')

    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.

        """

        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

        """

        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 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)

    def _seek_and_read(self, fp, offsets):

        """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)

            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 or string object into the location.

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

        :param f:       File-like or string 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_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_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

        """

        assert isinstance(bytes, str), \

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

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

    @deprecated_method(zero_eleven)

    def put_multi(self, files, mode=None, pb=None):

        """Put a set of files into the location.

        :param files: A list of tuples of relpath, file object [(path1, file1), (path2, file2),...]

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

        :param mode: The mode for the newly created files

        :return: The number of files copied.

        """

        def put(path, f):

            if isinstance(f, str):

                self.put_bytes(path, f, mode=mode)

            else:

                self.put_file(path, f, mode=mode)

        return len(self._iterate_over(files, put, pb, 'put', expand=True))

    def mkdir(self, relpath, mode=None):

        """Create a directory at the given path."""

        raise NotImplementedError(self.mkdir)

    def mkdir_multi(self, relpaths, mode=None, pb=None):

        """Create a group of directories"""

        def mkdir(path):

            self.mkdir(path, mode=mode)

        return len(self._iterate_over(relpaths, mkdir, pb, 'mkdir', expand=False))

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

        """Append the text in the file-like object to the supplied location.

        returns the length of relpath before the content was written to it.

        If the file does not exist, it is created with the supplied mode.

        """

        raise NotImplementedError(self.append)

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

        """Append the text in the string object to the supplied location.

        returns the length of relpath before the content was written to it.

        If the file does not exist, it is created with the supplied mode.

        """

        assert isinstance(bytes, str), \

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

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

    def append_multi(self, files, pb=None):

        """Append the text in each file-like or string object to

        the supplied location.

        :param files: A set of (path, f) entries

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

        """

        return self._iterate_over(files, self.append, pb, 'append', expand=True)

    def copy(self, rel_from, rel_to):

        """Copy the item at rel_from to the location at rel_to.

        Override this for efficiency if a specific transport can do it 

        faster than this default implementation.

        """

        self.put(rel_to, self.get(rel_from))

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

        """Copy a bunch of entries.

        :param relpaths: A list of tuples of the form [(from, to), (from, to),...]

        """

        # This is the non-pipelined implementation, so that

        # implementors don't have to implement everything.

        return self._iterate_over(relpaths, self.copy, pb, 'copy', expand=True)

    def copy_to(self, relpaths, other, mode=None, pb=None):

        """Copy a set of entries from self into another Transport.

        :param relpaths: A list/generator of entries to be copied.

        :param mode: This is the target mode for the newly created files

        TODO: This interface needs to be updated so that the target location

              can be different from the source location.

        """

        # The dummy implementation just does a simple get + put

        def copy_entry(path):

            other.put(path, self.get(path), mode=mode)

        return len(self._iterate_over(relpaths, copy_entry, pb, 'copy_to', expand=False))

    def copy_tree(self, from_relpath, to_relpath):

        """Copy a subtree from one relpath to another.

        If a faster implementation is available, specific transports should 

        implement it.

        """

        source = self.clone(from_relpath)

        self.mkdir(to_relpath)

        target = self.clone(to_relpath)

        files = []

        directories = ['.']

        while directories:

            dir = directories.pop()

            if dir != '.':

                target.mkdir(dir)

            for path in source.list_dir(dir):

                path = dir + '/' + path

                stat = source.stat(path)

                if S_ISDIR(stat.st_mode):

                    directories.append(path)

                else:

                    files.append(path)

        source.copy_to(files, target)

    def rename(self, rel_from, rel_to):

        """Rename a file or directory.

        This *must* fail if the destination is a nonempty directory - it must

        not automatically remove it.  It should raise DirectoryNotEmpty, or

        some other PathError if the case can't be specifically detected.

        If the destination is an empty directory or a file this function may

        either fail or succeed, depending on the underlying transport.  It

        should not attempt to remove the destination if overwriting is not the

        native transport behaviour.  If at all possible the transport should

        ensure that the rename either completes or not, without leaving the

        destination deleted and the new file not moved in place.

        This is intended mainly for use in implementing LockDir.

        """

        # transports may need to override this

        raise NotImplementedError(self.rename)

    def move(self, rel_from, rel_to):

        """Move the item at rel_from to the location at rel_to.

        The destination is deleted if possible, even if it's a non-empty

        directory tree.

        If a transport can directly implement this it is suggested that

        it do so for efficiency.

        """

        if S_ISDIR(self.stat(rel_from).st_mode):

            self.copy_tree(rel_from, rel_to)

            self.delete_tree(rel_from)

        else:

            self.copy(rel_from, rel_to)

            self.delete(rel_from)

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

        """Move a bunch of entries.

        :param relpaths: A list of tuples of the form [(from1, to1), (from2, to2),...]

        """

        return self._iterate_over(relpaths, self.move, pb, 'move', expand=True)