Skip to main content
Join the official Python Developers Survey 2018 and win valuable prizes: Start the survey!

transparent remote/recursive evaluation between Python and other languages

Project description

The Python module bond supports transparent remote/recursive evaluation between Python and another interpreter through automatic call serialization.

In poorer words, a bond lets you call functions in other languages as they were normal Python functions. It also allows other languages to call Python functions as if they were native.

Remote output is also transparently redirected locally, and since the evaluation is performed through a persistent co-process, you can actually spawn interpreters on different hosts through “ssh” efficiently.

bond currently supports PHP, Perl, JavaScript (Node.js) and Python itself.

A simple example

>>> # Let's bond with a PHP interpreter
>>> from bond import make_bond
>>> php = make_bond('PHP')
>>> php.eval_block('echo "Hello world!\n";')
Hello world!

>>> # Make an expensive split function using PHP's explode
>>> split = php.callable('explode')
>>> split(' ', "Hello world splitted by PHP!")
[u'Hello', u'world', u'splitted', u'by', u'PHP!']

>>> # Call Python from PHP
>>> def call_me():
...     print("Hi, this is Python talking!")
>>> php.export(call_me)
>>> php.eval('call_me()')
Hi, this is Python talking!

>>> # Use some remote resources
>>> remote_php = make_bond('PHP', 'ssh remote php')
>>> remote_php.eval_block('function call_me() { echo "Hi from " . system("hostname") . "!"; }')
>>> remote_php.eval('call_me()')
Hi from remote!

>>> # Bridge two worlds!
>>> perl = make_bond('Perl')
>>> php.proxy('explode', perl)
>>> # note: explode is now available to Perl, but still executes in PHP
>>> perl.eval('explode("=", "Mind=blown!")')
[u'Mind', u'blown!']


I needed bond for refactoring a large PHP project, mostly. With bond you can rewrite your program incrementally, while still executing all your existing code unchanged. You can start by rewriting just a single function:

from bond import make_bond
import sys

php = make_bond('PHP')

def new_function(arg)
    # do something here

php.export(new_function, 'function_to_be_replaced')'main', sys.argv)

It turns out that with the same approach you can easily perform remote/parallel computation as well. Nobody stops you from having multiple interpreters at the same time: you can use bond to setup a poor-man’s distributed system with minimal effort:

# setup the workers
from bond import make_bond
hosts = ['host1', 'host2', 'host3']
nodes = [make_bond('Python', 'ssh {} python'.format(host)) for host in hosts]

# load our libraries first
for node in nodes:
    node.eval_block('from library import *')

# execute "do_something" remotely on each worker
from threading import Thread
threads = [Thread(target=lambda:'do_something')) for node in nodes]
for thread in threads: thread.start()

# collect the results
results = [thread.join() for thread in threads]

You can also use bond to mix Python 2/3 code. bond aims to be completely invisible on the remote side: you don’t need bond installed remotely at all.

The wire protocol is simple enough to be extended to any language supporting an interactive interpreter. In fact, the drivers themselves can be used from any other language.



A bond.Bond object is not normally constructed directly, but by using the bond.make_bond() function:

import bond
interpreter = bond.make_bond('language')

The first argument should be the desired language name (“JavaScript”, “PHP”, “Perl”, “Python”). The list of supported languages can be fetched dynamically using bond.list_drivers().

You can override the default interpreter command using the second argument, which allows to specify any shell command to be executed:

import bond
py = bond.make_bond('Python', 'ssh remote python3')

An additional list of arguments to the interpreter can be provided using the third argument, args:

import bond
py = bond.make_bond('Python', 'ssh remote python3', ['-E', '-OO'])

The arguments, contrarily to the command, are automatically quoted.

Some command line arguments may be supplied automatically by the driver to force an interactive shell; for example “-i” is supplied if Python is requested. You can disable default arguments by using def_args=False.

The following keyword arguments are supported:


Working directory for the interpreter (defaults to current working directory).


Environment for the interpreter (defaults to os.environ).


Enable (default) or suppress default, extra command-line arguments to the interpreter.


Defines the timeout for the underlying communication protocol. Note that bond cannot distinguish between a slow call or noise generated while the interpreter is set up. Defaults to 60 seconds.


Accepts a file handle which is used to log the entire communication with the underlying interpreter for debugging purposes.


Enables/disables “transparent exceptions”. If trans_except is enabled, exceptions will be forwarded across the bond using the original data-type. If trans_except is disabled (the default for all languages except Python), then local exceptions will always contain a string representation of the remote exception instead, which avoids serialization errors.

bond.Bond Methods

The resulting bond.Bond class has the following methods:


Evaluate and return the value of a single statement of code in the interpreter.


Execute a “code” block inside the interpreter. Any construct which is legal by the current interpreter is allowed. Nothing is returned.


Terminate the communication with the interpreter.

call(name, *args):

Call a function “name” in the interpreter using the supplied list of arguments *args (apply *args to a callable statement defined by “name”). The arguments are automatically converted to their other language’s counterpart. The return value is captured and converted back to Python as well.


Return a function that calls “name”:

explode = php.callable('explode')
# Now you can call explode as a normal, local function
explode(' ', 'Hello world')

export(func, name):

Export a local function “func” so that can be called on the remote language as “name”. If “name” is not specified, use the local function name directly. Note that “func” must be a function reference, not a function name.

proxy(name, other, remote):

Export a function “name” from the current bond to “other”, named as “remote”. If “remote” is not provided, the same value as “name” is used.


Start an interactive session with the underlying interpreter. By default, all input lines are executed with bond.eval_block(). If “!” is pre-pended, execute a single statement with bond.eval() and print it’s return value. You can continue the statement on multiple lines by leaving a trailing “". Type Ctrl+C to abort a multi-line block without executing it.


All exceptions thrown by the bond module are of base type RuntimeError <= BondException.

Thrown during initialization or unrecoverable errors.
Thrown when the bond exits unexpectedly.
Thrown when an object/exception which is sent or received cannot be serialized by the current protocol. The side attribute can be either “local” (when attempting to send) or “remote” (when receiving). A SerializationException is not fatal.
Thrown for uncaught remote exceptions. The “data” attribute contains either the error message (with trans_except=False) or the remote exception itself (trans_except=True).

Beware that both SerializationException (with side="remote") and RemoteException may actually be originating from uncaught local exceptions when an exported function is called. Pay attention to the error text/data in these cases, as it will contain several nested exceptions.

Language support


Python, as the identity language, has no restriction on data types. Everything is pickled on both sides, including exceptions.


  • Performed locally and remotely using cPickle in Python 2 or pickle in Python 3.
  • Serialization exceptions on the remote side are of base type TypeError <= _BOND_SerializationException.

Python 2 / Python 3:

You can freely mix Python versions between hosts/interpreters (that is: you can run Python 3 code from a Python 2 host and vice-versa). You’ll need to disable transparent exceptions though, as the as the exception hierarchy is different between major versions:

# assuming a python2.7 environment
from bond import make_bond
py = make_bond('Python', 'python3', trans_except=False)



  • The PHP’s >= 5.3 command line interpreter needs to be installed. On Debian/Ubuntu, the required packages is php5-cli.


  • Performed remotely using JSON. Implement the JsonSerializable interface to tweak which/how objects are encoded.
  • Serialization exceptions on the remote side are of base type _BOND_SerializationException. The detailed results of the error can also be retrieved using json_last_error.


  • PHP <= 5.3 doesn’t support the JsonSerializable interface, and thus lacks transparent exceptions.
  • You cannot use call on a built-in function such as “echo”. You have to use a real function instead, like “print”. You can still call “echo” by using eval or eval_block.
  • Unfortunately, you cannot catch “fatal errors” in PHP. If the evaluated code triggers a fatal error it will terminate the bond without appeal. A common example of such errors in PHP is attempting to use an undefined variable or function (which could happen while prototyping).


Perl is a quirky language, due to its syntax. We assume here you’re an experienced Perl developer.


  • Perl >= 5.14 is required, with the following modules:

    • JSON
    • Data::Dump
    • IO::String

    On Debian/Ubuntu, the required packages are libjson-perl libdata-dump-perl and libio-string-perl.


  • Performed remotely using JSON. Implement the TO_JSON method on blessed references to tweak which/how objects are encoded.
  • Serialization exceptions on the remote side are generated by dying with a _BOND_SerializationException @ISA.


  • By default, evaluation is forced in array context, as otherwise most of the built-ins working with arrays would return an useless scalar. Use the “scalar” keyword for the rare cases when you really need it to.

  • You can “call” any function-like statement, as long as the last argument is expected to be an argument list. This allows you to call builtins directly:'map { $_ + 1 }', [1, 2, 3])
  • You can of course “call” a statement that returns any CODE. Meaning that you can call references to functions as long as you dereference them first:'&$fun_ref', ...)'&{ $any->{expression} }', ...)

    Likewise you can “call” objects methods directly:'$object->method', ...)
  • eval_block introduces a new block. Variables declared as “my” will not be visible into a subsequent eval_block. Use a fully qualified name or “our” to define variables that should persist across blocks:

    perl.eval_block('our $variable = 1;')


JavaScript is supported through Node.js.


  • Node.js v0.6.12 and v0.10.29 have been tested. On Debian/Ubuntu, the required package is nodejs.


  • Performed remotely using JSON. Implement the toJSON property to tweak which/how objects are encoded.
  • Serialization exceptions on the remote side are of base type TypeError <= _BOND_SerializationException.


  • Currently the code expects an unix-like environment with /dev/stdin to perform synchronous I/O.

  • Since there’s no distinction between “plain” objects (dictionaries) and any other object, almost everything will be silently serialized. Define a custom “toJSON” property on your “real” objects to control this behavior.

  • When executing a remote JavaScript bond with Node.js <= 0.6, you need to manually invoke the REPL, as follows:

    js = make_bond('JavaScript',
                   "ssh remote node -e 'require\(\\\"repl\\\"\).start\(\)'",

    When executing “node” locally, or when using Node.js >= 0.10, this is not required (the “-i” flag is automatically provided).

Common traits/limitations

  • Except for Python, only basic types (booleans, numbers, strings, lists/arrays and maps/dictionaries) can be transferred between the interpreters.
  • Serialization is performed locally using JSON. Implement a custom JSONEncoder to tweak which/how objects are encoded.
  • If an object that cannot be serialized reaches a “call”, “eval”, or even a non-local return such as an error or exception, it will generate a SerializationException on the local (Python) side.
  • Strings are always UTF-8 encoded.
  • References are implicitly broken as objects are transferred by value. This is obvious, as you’re talking with a separate process, but it can easily be forgotten due to the blurring of the boundary.
  • Calling functions across the bridge is slow, also in Python, due to the serialization. But the execution speed of the functions themselves is not affected. This might be perfectly reasonable if there are only occasional calls between languages, and/or the calls themselves take a significant fraction of time.

Latest release notes

  • The API has been streamlined: make_bond() is now the primary method of constructing Bond objects, independently of the interpreter language. The old language constructors are still supported, but are deprecated and will be removed in a future release.
  • All functions/objects/methods are now documented with docstrings.
  • Bond initialization errors, especially errors related to missing dependencies, are now much easier to understand.
  • Serialization exceptions on the remote side have been renamed to __BOND_SerializationException for consistency with other languages.
  • JavaScript/Node.js support was previously limited to versions >= 0.10. Any version of Node.js starting with 0.6.12 is now supported.
  • PHP support was previously limited to versions >= 5.6. Any version of PHP starting with 5.3 is now supported.
  • A Perl dependency on IO::String was previously missing, and has now been correctly documented.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Filename, size & hash SHA256 hash help File type Python version Upload date
python-bond-1.0.tar.gz (48.0 kB) Copy SHA256 hash SHA256 Source None Aug 19, 2014

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN SignalFx SignalFx Supporter DigiCert DigiCert EV certificate StatusPage StatusPage Status page