Basics

FTPHost instances can be generated with the following call:

host = ftputil.FTPHost(host, user, password, account,
                       session_factory=ftplib.FTP)

The first four parameters are strings with the same meaning as for the FTP class in the ftplib module.

Session factories

The keyword argument session_factory may be used to generate FTP connections with other factories than the default ftplib.FTP. For example, the M2Crypto distribution uses a secure FTP class which is derived from ftplib.FTP.

In fact, all positional and keyword arguments other than session_factory are passed to the factory to generate a new background session. This happens for every remote file that is opened; see below.

This functionality of the constructor also allows to wrap ftplib.FTP objects to do something that wouldn't be possible with the ftplib.FTP constructor alone.

As an example, assume you want to connect to another than the default port, but ftplib.FTP only offers this by means of its connect method, not via its constructor. The solution is to use a wrapper class:

import ftplib
import ftputil

EXAMPLE_PORT = 50001

class MySession(ftplib.FTP):
    def __init__(self, host, userid, password, port):
        """Act like ftplib.FTP's constructor but connect to another port."""
        ftplib.FTP.__init__(self)
        self.connect(host, port)
        self.login(userid, password)

# try not to use MySession() as factory, - use the class itself
host = ftputil.FTPHost(host, userid, password,
                       port=EXAMPLE_PORT, session_factory=MySession)
# use `host` as usual

On login, the format of the directory listings (needed for stat'ing files and directories) should be determined automatically. If not, please file a bug report.

Support for the with statement

If you are sure that all the users of your code use at least Python 2.5, you can use Python's with statement:

# not needed for Python 2.6 and later
from __future__ import with_statement

import ftputil

with ftputil.FTPHost(host, user, password) as host:
    print host.listdir(host.curdir)

After the with block, the FTPHost instance and the associated FTP sessions will be closed automatically.

If something goes wrong during the FTPHost construction or in the body of the with statement, the instance is closed as well. Exceptions will be propagated (as with try ... finally).

Attributes

• curdir, pardir, sep

  are strings which denote the current and the parent directory on the remote server. sep holds the path separator. Though RFC 959 (File Transfer Protocol) notes that these values may depend on the FTP server implementation, the Unix variants seem to work well in practice, even for non-Unix servers.

Remote file system navigation

• getcwd()

  returns the absolute current directory on the remote host. This method acts similar to os.getcwd.

• chdir(directory)

  sets the current directory on the FTP server. This resembles os.chdir, as you may have expected.

Uploading and downloading files

• upload(source, target, mode='')

  copies a local source file (given by a filename, i. e. a string) to the remote host under the name target. Both source and target may be absolute paths or relative to their corresponding current directory (on the local or the remote host, respectively). The mode may be "" or "a" for ASCII uploads or "b" for binary uploads. ASCII mode is the default, similar to regular local file objects.

• download(source, target, mode='')

  performs a download from the remote source to a target file. Both source and target are strings. Most of the description of the upload method applies here, too.

• upload_if_newer(source, target, mode='')

  is similar to the upload method. The only difference is that the upload is only invoked if the time of the last modification for the source file is more recent than that of the target file or the target doesn't exist at all. If an upload actually happened, the return value is a true value, else a false value.

  Note that this method only checks the existence and/or the modification time of the source and target file; it can't recognize a change in the transfer mode, e. g.

  # transfer in ASCII mode
  host.upload_if_newer('source_file', 'target_file', 'a')
  # won't transfer the file again, which is bad!
  host.upload_if_newer('source_file', 'target_file', 'b')
  

  Similarly, if a transfer is interrupted, the remote file will have a newer modification time than the local file, and thus the transfer won't be repeated if upload_if_newer is used a second time. There are at least two possibilities after a failed upload:

  • use upload instead of upload_if_newer, or
  • remove the incomplete target file with FTPHost.remove, then use upload or upload_if_newer to transfer it again.

  If it seems that a file is uploaded unnecessarily or not when it should, read the subsection on time shift settings.

• download_if_newer(source, target, mode='')

  corresponds to upload_if_newer but performs a download from the server to the local host. Read the descriptions of download and upload_if_newer for more. If a download actually happened, the return value is a true value, else a false value.

  If it seems that a file is downloaded unnecessarily or not when it should, read the subsection on time zone correction.

Time zone correction

If the client where ftputil runs and the server have a different understanding of their local times, this has to be taken into account for upload_if_newer and download_if_newer to work correctly.

Note that even if the client and the server are in the same time zone (or even on the same computer), the time shift value (see below) may be different from zero. For example, my computer is set to use local time whereas the server running on the very same host insists on using UTC time.

• set_time_shift(time_shift)

  sets the so-called time shift value, measured in seconds. The time shift is the difference between the local time of the server and the local time of the client at a given moment, i. e. by definition

  time_shift = server_time - client_time
  

  Setting this value is important for upload_if_newer and download_if_newer to work correctly even if the time zone of the FTP server differs from that of the client. Note that the time shift value can be negative.

  If the time shift value is invalid, e. g. no multiple of a full hour or its absolute value larger than 24 hours, a TimeShiftError is raised.

  See also synchronize_times for a way to set the time shift with a simple method call.

• time_shift()

  returns the currently-set time shift value. See set_time_shift above for its definition.

• synchronize_times()

  synchronizes the local times of the server and the client, so that upload_if_newer and download_if_newer work as expected, even if the client and the server use different time zones. For this to work, all of the following conditions must be true:

  • The connection between server and client is established.
  • The client has write access to the directory that is current when synchronize_times is called.

  If you can't fulfill these conditions, you can nevertheless set the time shift value explicitly with set_time_shift. Trying to call synchronize_times if the above conditions aren't met results in a TimeShiftError exception.

Creating and removing directories

• mkdir(path, [mode])

  makes the given directory on the remote host. This doesn't construct "intermediate" directories which don't already exist. The mode parameter is ignored; this is for compatibility with os.mkdir if an FTPHost object is passed into a function instead of the os module. See the explanation in the subsection Exception hierarchy.

• makedirs(path, [mode])

  works similar to mkdir (see above), but also makes intermediate directories like os.makedirs. The mode parameter is only there for compatibility with os.makedirs and is ignored.

• rmdir(path)

  removes the given remote directory. If it's not empty, raise a PermanentError.

• rmtree(path, ignore_errors=False, onerror=None)

  removes the given remote, possibly non-empty, directory tree. The interface of this method is rather complex, in favor of compatibility with shutil.rmtree.

  If ignore_errors is set to a true value, errors are ignored. If ignore_errors is a false value and onerror isn't set, all exceptions occurring during the tree iteration and processing are raised. These exceptions are all of type PermanentError.

  To distinguish between different kinds of errors, pass in a callable for onerror. This callable must accept three arguments: func, path and exc_info. func is a bound method object, for example your_host_object.listdir. path is the path that was the recent argument of the respective method (listdir, remove, rmdir). exc_info is the exception info as it is gotten from sys.exc_info.

  The code of rmtree is taken from Python's shutil module and adapted for ftputil.

Removing files and links

• remove(path)

  removes a file or link on the remote host, similar to os.remove.

• unlink(path)

  is an alias for remove.

Retrieving information about directories, files and links

• listdir(path)

  returns a list containing the names of the files and directories in the given path, similar to os.listdir. The special names . and .. are not in the list.

The methods lstat and stat (and some others) rely on the directory listing format used by the FTP server. When connecting to a host, FTPHost's constructor tries to guess the right format, which succeeds in most cases. However, if you get strange results or ParserError exceptions by a mere lstat call, please file a bug report.

If lstat or stat yield wrong modification dates or times, look at the methods that deal with time zone differences (time zone correction).

• lstat(path)

  returns an object similar to that from os.lstat. This is a "tuple" with additional attributes; see the documentation of the os module for details.

  The result is derived by parsing the output of a DIR command on the server. Therefore, the result from FTPHost.lstat can not contain more information than the received text. In particular:

  • User and group ids can only be determined as strings, not as numbers, and that only if the server supplies them. This is usually the case with Unix servers but maybe not for other FTP server programs.

  • Values for the time of the last modification may be rough, depending on the information from the server. For timestamps older than a year, this usually means that the precision of the modification timestamp value is not better than days. For newer files, the information may be accurate to a minute.

  • Links can only be recognized on servers that provide this information in the DIR output.

  • Stat attributes that can't be determined at all are set to

    None. For example, a line of a directory listing may not contain the date/time of a directory's last modification.

  • There's a special problem with stat'ing the root directory. (Stat'ing things in the root directory is fine though.) In this case, a RootDirError is raised. This has to do with the algorithm used by (l)stat, and I know of no approach which mends this problem.

  Currently, ftputil recognizes the common Unix-style and Microsoft/DOS-style directory formats. If you need to parse output from another server type, please write to the ftputil mailing list. You may consider to write your own parser.

• stat(path)

  returns stat information also for files which are pointed to by a link. This method follows multiple links until a regular file or directory is found. If an infinite link chain is encountered or the target of the last link in the chain doesn't exist, a PermanentError is raised.

FTPHost objects contain an attribute named path, similar to os.path. The following methods can be applied to the remote host with the same semantics as for os.path:

abspath(path)
basename(path)
commonprefix(path_list)
dirname(path)
exists(path)
getmtime(path)
getsize(path)
isabs(path)
isdir(path)
isfile(path)
islink(path)
join(path1, path2, ...)
normcase(path)
normpath(path)
split(path)
splitdrive(path)
splitext(path)
walk(path, func, arg)

Like Python's counterparts under os.path, ftputil's is... methods return False if they can't find the path given by their argument.

Local caching of file system information

Many of the above methods need access to the remote file system to obtain data on directories and files. To get the most recent data, each call to lstat, stat, exists, getmtime etc. would require to fetch a directory listing from the server, which can make the program very slow. This effect is more pronounced for operations which mostly scan the file system rather than transferring file data.

For this reason, ftputil by default saves the results from directory listings locally and reuses those results. This reduces network accesses and so speeds up the software a lot. However, since data is more rarely fetched from the server, the risk of obsolete data also increases. This will be discussed below.

Caching can be controlled -- if necessary at all -- via the stat_cache object in an FTPHost's namespace. For example, after calling

host = ftputil.FTPHost(host, user, password, account,
                       session_factory=ftplib.FTP)

the cache can be accessed as host.stat_cache.

While ftputil usually manages the cache quite well, there are two possible reasons that may suggest modifying cache parameters. The first is when the number of possible entries is too low. You may notice that when you are processing very large directories, e. g. containing more than 1000 directories or files, and the program becomes much slower than before. It's common for code to read a directory with listdir and then process the found directories and files. For this application, it's a good rule of thumb to set the cache size to somewhat more than the number of directory entries fetched with listdir. This is done by the resize method:

host.stat_cache.resize(2000)

where the argument is the maximum number of lstat results to store (the default is 1000). Note that each path on the server, e. g. "/home/schwa/some_dir", corresponds to a single cache entry. Methods like exists or getmtime all derive their results from a previously fetched lstat result.

The value 2000 above means that the cache will hold at most 2000 entries. If more are about to be stored, the entries which haven't been used for the longest time will be deleted to make place for newer entries.

Caching is so effective because it reduces network accesses. This can also be a disadvantage if the file system data on the remote server changes after a stat result has been retrieved; the client, when looking at the cached stat data, will use obsolete information.

There are two ways to get such out-of-date stat data. The first happens when an FTPHost instance modifies a file path for which it has a cache entry, e. g. by calling remove or rmdir. Such changes are handled transparently; the path will be deleted from the cache. A different matter are changes unknown to the FTPHost object which inspects its cache. Obviously, for example, these are changes by programs running on the remote host. On the other hand, cache inconsistencies can also occur if two FTPHost objects change a file system simultaneously:

host1 = ftputil.FTPHost(server, user1, password1)
host2 = ftputil.FTPHost(server, user1, password1)
try:
    stat_result1 = host1.stat("some_file")
    stat_result2 = host2.stat("some_file")
    host2.remove("some_file")
    # `host1` will still see the obsolete cache entry!
    print host1.stat("some_file")
    # will raise an exception since an `FTPHost` object
    #  knows of its own changes
    print host2.stat("some_file")
finally:
    host1.close()
    host2.close()

At first sight, it may appear to be a good idea to have a shared cache among several FTPHost objects. After some thinking, this turns out to be very error-prone. For example, it won't help with different processes using ftputil. So, if you have to deal with concurrent write/read accesses to a server, you have to handle them explicitly.

The most useful tool for this is the invalidate method. In the example above, it could be used like this:

host1 = ftputil.FTPHost(server, user1, password1)
host2 = ftputil.FTPHost(server, user1, password1)
try:
    stat_result1 = host1.stat("some_file")
    stat_result2 = host2.stat("some_file")
    host2.remove("some_file")
    # invalidate using an absolute path
    absolute_path = host1.path.abspath(
                    host1.path.join(host1.curdir, "some_file"))
    host1.stat_cache.invalidate(absolute_path)
    # will now raise an exception as it should
    print host1.stat("some_file")
    # would raise an exception since an `FTPHost` object
    #  knows of its own changes, even without `invalidate`
    print host2.stat("some_file")
finally:
    host1.close()
    host2.close()

The method invalidate can be used on any absolute path, be it a directory, a file or a link.

By default, the cache entries (if not replaced by newer ones) are stored for an infinite time. That is, if you start your Python process using ftputil and let it run for three days a stat call may still access cache data that old. To avoid this, you can set the max_age attribute:

host = ftputil.FTPHost(server, user, password)
host.stat_cache.max_age = 60 * 60  # = 3600 seconds

This sets the maximum age of entries in the cache to an hour. This means any entry older won't be retrieved from the cache but its data instead fetched again from the remote host and then again stored for up to an hour. To reset max_age to the default of unlimited age, i. e. cache entries never expire, use None as value.

If you are certain that the cache will be in the way, you can disable and later re-enable it completely with disable and enable:

host = ftputil.FTPHost(server, user, password)
host.stat_cache.disable()
...
host.stat_cache.enable()

During that time, the cache won't be used; all data will be fetched from the network. After enabling the cache, its entries will be the same as when the cache was disabled, that is, entries won't get updated with newer data during this period. Note that even when the cache is disabled, the file system data in the code can become inconsistent:

host = ftputil.FTPHost(server, user, password)
host.stat_cache.disable()
if host.path.exists("some_file"):
    mtime = host.path.getmtime("some_file")

In that case, the file some_file may have been removed by another process between the calls to exists and getmtime!

Iteration over directories

• walk(top, topdown=True, onerror=None)

  iterates over a directory tree, similar to os.walk. Actually, FTPHost.walk uses the code from Python with just the necessary modifications, so see the linked documentation.

• path.walk(path, func, arg)

  Similar to os.path.walk, the walk method in FTPHost.path can be used, though FTPHost.walk is probably easier to use.

Other methods

• close()

  closes the connection to the remote host. After this, no more interaction with the FTP server is possible without using a new FTPHost object.

• rename(source, target)

  renames the source file (or directory) on the FTP server.

• chmod(path, mode)

  sets the access mode (permission flags) for the given path. The mode is an integer as returned for the mode by the stat and lstat methods. Be careful: Usually, mode values are written as octal numbers, for example 0755 to make a directory readable and writable for the owner, but not writable for the group and others. If you want to use such octal values, rely on Python's support for them:

  host.chmod("some_directory", 0755)
  

  Note the leading zero.

  Not all FTP servers support the chmod command. In case of an exception, how do you know if the path doesn't exist or if the command itself is invalid? If the FTP server complies with RFC 959, it should return a status code 502 if the SITE CHMOD command isn't allowed. ftputil maps this special error response to a CommandNotImplementedError which is derived from PermanentError.

  So you need to code like this:

  host = ftputil.FTPHost(server, user, password)
  try:
      host.chmod("some_file", 0644)
  except ftp_error.CommandNotImplementedError:
      # chmod not supported
      ...
  except ftp_error.PermanentError:
      # possibly a non-existent file
      ...
  

  Because the CommandNotImplementedError is more specific, you have to test for it first.

• copyfileobj(source, target, length=64*1024)

  copies the contents from the file-like object source to the file-like object target. The only difference to shutil.copyfileobj is the default buffer size. Note that arbitrary file-like objects can be used as arguments (e. g. local files, remote FTP files). See File-like objects for construction and use of remote file-like objects.

• set_parser(parser)

  sets a custom parser for FTP directories. Note that you have to pass in a parser instance, not the class.

  An extra section shows how to write own parsers if the default parsers in ftputil don't work for you. Possibly you are lucky and someone has already written a parser you can use. Please ask on the mailing list.

Basics

FTPFile objects are returned by a call to FTPHost.file or FTPHost.open, never use the constructor directly.

• FTPHost.file(path, mode='r')

  returns a file-like object that refers to the path on the remote host. This path may be absolute or relative to the current directory on the remote host (this directory can be determined with the getcwd method). As with local file objects the default mode is "r", i. e. reading text files. Valid modes are "r", "rb", "w", and "wb".

• FTPHost.open(path, mode='r')

  is an alias for file (see above).

Support for the with statement

If you are sure that all the users of your code use at least Python 2.5, you can use Python's with statement with the FTPFile constructor:

# not needed for Python 2.6 and later
from __future__ import with_statement

import ftputil

# get an ``FTPHost`` object from somewhere
...

with host.file("new_file", "w") as f:
    f.write("This is some text.")

At the end of the with block, the file will be closed automatically.

If something goes wrong during the construction of the file or in the body of the with statement, the file will be closed as well. Exceptions will be propagated as with try ... finally.