The MUSCLE Python API closely resembles the C api but is slightly more brief. Currently, only Python 2.x is supported. It is enabled by default on the latest version of MUSCLE, and can be forced to be enabled by setting

$ cd [MUSCLE_SRC]/build
$ export MUSCLE_BUILD_OPTIONS="-DBUILD_PYTHON=ON"
$ ./build.sh [MUSCLE_INSTALL_DIR]

It will then be built as the Python module. By doing

$ source [MUSCLE_INSTALL_DIR]/etc/muscle.profile

it gets included in the Python module path. The latest information about the module can always be found by doing

$ python
import muscle
help(muscle)

The API includes the following functions:

muscle.init(sys.argv)
muscle.finalize()

muscle.send(portName, data, [muscle.datatype])
muscle.receive(portName, [muscle.datatype])

muscle.will_stop()
muscle.kernel_name()
muscle.tmp_path()
muscle.has_property(propName)
muscle.get_property(propName)

muscle.log(msg)
muscle.fine(msg)
muscle.warning(msg)
muscle.severe(msg)

These functions behave just like their C++ counterparts, except init, send and receive. For init, first run import sys and then muscle.init(sys.argv). Send will accept a sequence of objects, given a muscle datatype. The following datatypes are recognized:

Object	Python data type	C data type
muscle.int	sequence of int	32-bit int
muscle.long	sequence of int	64-bit long
muscle.float	sequence of double	32-bit float
muscle.double	sequence of double	64-bit double
muscle.boolean	sequence of bool	bool
muscle.raw	bytearray	char *
muscle.string	str	const char *

The first five data types take any sequence, as long as the Python datatype that is contained matches. In some cases, it may be necessary to recreate a NumPy? array so that the canonical Python datatypes are used. For example, with int:

import numpy as np
import muscle

data = np.zeros(100, np.int)
muscle.send([int(d) for d in data], 'out', muscle.int)