libtorrent python binding

building

libtorrent can be built as a python module.

The best way to build the python bindings is using setup.py. This invokes b2 under the hood, so you must have all of libtorrent's build dependencies installed.

If you just want to build the shared library python extension without python packaging semantics, you can also invoke b2 directly.

prerequisites

Whether building with setup.py or directly invoking b2, you must install the build prerequisites on your system:

1. All the build prerequisites for the main libtorrent library, including boost libraries and b2, and your building toolchain (gcc, visual studio, etc).
2. Boost.Python, if not otherwise included in your boost installation
3. Python 3.6+. Older versions may work, but are not tested.

environment variables

b2 is very sensitive to environment variables. At least the following are required:

1. BOOST_ROOT
2. BOOST_BUILD_PATH

b2 is also known to reference dozens of other environment variables when detecting toolsets. Keep this in mind if you are building in an isolation environment like tox.

building with setup.py

By default, setup.py will invoke b2 to build libtorrent:

python setup.py build

setup.py is a normal distutils-based setup script.

To install into your python environment:

python setup.py install

To build a binary wheel package:

python -m pip install wheel
python setup.py bdist_wheel

build for a different python version

setup.py will target the running interpreter. To build for different python versions, you must change how you invoke setup.py:

# build for python3.6
python3.6 setup.py build
# build for python3.7
python3.7 setup.py build

customizing the build

You can customize the build by passing options to the build_ext step of setup.py by passing arguments directly to b2 via --b2-args=:

python setup.py build_ext --b2-args="toolset=msvc-14.2 linkflags=-L../../src/.libs"

For a full list of b2 build options, see libtorrent build features.

Here, it's important to note that build_ext has no "memory" of the build config and arguments you passed to it before. This is different from the way distutils normally works. Consider:

python setup.py build_ext --b2-args="optimization=space"
# the following will build with DEFAULT optimization
python setup.py install

In order to customize the build and run other steps like installation, you should run the steps inline with build_ext:

python setup.py build_ext --b2-args="optimization=space" install

building with b2

You will need to update your user-config.jam so b2 can find your python installation.

b2 has some auto-detection capabilities. You may be able to do just this:

using python : 3.6 ;

However you may need to specify full paths. On windows, it make look like this:

using python : 3.6 : C:/Users/<UserName>/AppData/Local/Programs/Python/Python36 : C:/Users/<UserName>/AppData/Local/Programs/Python/Python36/include : C:/Users/<UserName>/AppData/Local/Programs/Python/Python36/libs ;

Or on Linux, like this:

using python : 3.6 : /usr/bin/python3.6 : /usr/include/python3.6 : /usr/lib/python3.6 ;

Note that b2's python path detection is known to only work for global python installations. It is known to be broken for virtualenvs or pyenv. If you are using pyenv to manage your python versions, you must specify full include and library paths yourself.

invoking b2

Build the bindings like so:

cd bindings/python
b2 release python=3.6 address-model=64

Note that address-model should match the python installation you are building for.

For other build features, see libtorrent build options.

static linking

A python module is a shared library. Specifying link=static when building the binding won't work, as it would try to produce a static library.

Instead, control whether the libtorrent main library or boost is linked statically with libtorrent-link=static and boost-link=static respectively.

By default both are built and linked as shared libraries.

Building and linking boost as static library is only possibly by building it from source. Specify the BOOST_ROOT environment variable to point to the root directory of the boost source distribution.

For example, to build a self-contained python module:

b2 release python=3.6 libtorrent-link=static boost-link=static

helper targets

There are some targets for placing the build artifact in a helpful location:

$ b2 release python=3.6 stage_module stage_dependencies

This will produce a libtorrent python module in the current directory (file name extension depends on operating system). The libraries the python module depends on will be copied into ./dependencies.

To install the python module, build it with the following command:

b2 release python=3.6 install_module

By default the module will be installed to the python user site. This can be changed with the python-install-scope feature. The valid values are user (default) and system. e.g.:

b2 release python=3.6 install_module python-install-scope=system

To specify a custom installation path for the python module, specify the desired path with the python-install-path feature. e.g.:

b2 release python=3.6 install_module python-install-path=/home/foobar/python-site/

using libtorrent in python

The python interface is nearly identical to the C++ interface. Please refer to the library reference. The main differences are:

asio::tcp::endpoint

The endpoint type is represented as a tuple of a string (as the address) and an int for the port number. E.g. ("127.0.0.1", 6881) represents the localhost port 6881.

lt::time_duration

The time duration is represented as a number of seconds in a regular integer.

The following functions takes a reference to a container that is filled with entries by the function. The python equivalent of these functions instead returns a list of entries.

• torrent_handle::get_peer_info
• torrent_handle::file_progress
• torrent_handle::get_download_queue
• torrent_handle::piece_availability

create_torrent::add_node() takes two arguments, one string and one integer, instead of a pair. The string is the address and the integer is the port.

session::apply_settings() accepts a dictionary with keys matching the names of settings in settings_pack. When calling apply_settings, the dictionary does not need to have every settings set, keys that are not present are not updated.

To get a python dictionary of the settings, call session::get_settings.

Retrieving session statistics in Python is more convenient than that in C++. The statistics are stored as an array in session_stats_alert, which will be posted after calling post_session_stats() in the session object. In order to interpret the statistics array, in C++ it is required to call session_stats_metrics() to get the indices of these metrics, while in Python it can be done using session_stats_alert.values["NAME_OF_METRIC"], where NAME_OF_METRIC is the name of a metric.

set_alert_notify

The set_alert_notify() function is not compatible with python. Since it requires locking the GIL from within the libtorrent thread, to call the callback, it can cause a deadlock with the main thread.

Instead, use the python-specific set_alert_fd() which takes a file descriptor that will have 1 byte written to it to notify the client that there are new alerts to be popped.

The file descriptor should be set to non-blocking mode. If writing to the file/sending to the socket blocks, libtorrent's internal thread will stall.

This can be used with socket.socketpair(), for example. The file descriptor is what fileno() returns on a socket.

Example

For an example python program, see client.py in the bindings/python directory.

A very simple example usage of the module would be something like this: