Inline::Python(3) Write Perl subs and classes in Python.


print "9 + 16 = ", add(9, 16), "\n";
print "9 - 16 = ", subtract(9, 16), "\n";
use Inline Python => <<'END_OF_PYTHON_CODE';
def add(x,y):
return x + y
def subtract(x,y):
return x - y


The "Inline::Python" module allows you to put Python source code directly ``inline'' in a Perl script or module. It sets up an in-process Python interpreter, runs your code, and then examines Python's symbol table for things to bind to Perl. The process of interrogating the Python interpreter for globals only occurs the first time you run your Python code. The namespace is cached, and subsequent calls use the cached version.

This document describes "Inline::Python", the Perl package which gives you access to a Python interpreter. For lack of a better place to keep it, it also gives you instructions on how to use "perlmodule", the Python package which gives you access to the Perl interpreter.


Version 0.21 provides the ability to bind to 'new-style' classes (as defined by the python PEP's 252 and 253.) See ``New-Style Classes'' for details.

See the Changes file for new features in recent versions.

Using the Inline::Python Module

Using Inline::Python will seem very similar to using another Inline language, thanks to Inline's consistent look and feel.

This section will explain the different ways to use Inline::Python. For more details on "Inline", see 'perldoc Inline'.

Giving Your Source to Inline

The most basic form for using "Inline::Python" is this:

   use Inline Python => 'Python source code';

Of course, you can use Perl's ``here document'' style of quoting to make the code slightly easier to read:

   use Inline Python => <<'END';
     Python source code goes here.

The source code can also be specified as a filename, a subroutine reference (sub routine should return source code), or an array reference (array contains lines of source code). The recommended way of using Inline is this:

   use Inline Python;
   Python source code goes here.

This information is detailed in 'perldoc Inline'.

Importing Functions

Maybe you have a whole library written in Python that only needs one entry point. You'll want to import that function. It's as easy as this:

   use Inline Python;
   from mylibrary import doit

Inline::Python actually binds to every function in Python's ``global'' namespace (those of you in the know, know that namespace is called '__main__'). So if you had another function there, you'd get that too.

Importing Classes

If you've written a library in Python, you'll make it object-oriented. That's just something Python folks do. So you'll probably want to import a class, not a function. That's just as easy:

   use Inline Python;
   my $obj = new Myclass;
   from mylibrary import myclass as Myclass

New-Style Classes

As of python 2.2, the python internals have begun to change in a way which makes types 'look' more like classes. This means that your python code can now subclass builtin python types such as lists, tuples, integers, and etc. It also means that identifying python objects and creating Perl bindings for them has become a little trickier.

See Guido's write-up ( and the relevant Python Enhancement Proposals (PEP) numbers 252 and 253 for details about the python code. Also, see the mailing-list discussion ( for possible implications regarding C-language python extensions.

This change should not affect code which uses Inline::Python, except that it allows you to bind to python classes which have been written using these new features. In most cases, you will be importing an entire class from an external library as defined in the example above.

In other cases, you may be writing Inline::Python code as follows:

   use Inline Python => <<'END';
   class Foo(object):
      def __init__(self):
         print "new Foo object being created" = {}
      def get_data(self): return
      def set_data(self,dat):  = dat

Additional caveats may exist. Note that if the python class is subclassing one of the builtin types which would normally be accessible as a 'Perlish' translation, that the instance will be an opaque object accessible only through its class methods.

  # Class is defined as 'def Class(float):'
  my $obj = Class->new(4);
  print $$obj, "\n"; # will NOT print '4.0'


Boundary Conditions

What if you have a class that wasn't imported? Can you deal with instances of that class properly?

Of course you can! Check this out:

   use Inline Python => <<END;
   def Foo():
       class Bar:
           def __init__(self):
               print "new Bar()"
           def tank(self):
               return 10
       return Bar()
   my $o = Foo();
   print $o->tank(), "\n";

In this example, "Bar" isn't imported because it isn't a global --- it's hidden inside the function Foo(). But Foo() is imported into Perl, and it returns an instance of the "Bar" class. What happens then?

Whenever Inline::Python needs to return an instance of a class to Perl, it generates an instance of Inline::Python::Object, the base class for all Inline::Python objects. This base class knows how to do all the things you need: calling methods, in this case.


Exceptions thrown in Python code get translated to Perl exceptions which you can catch using eval.


Python supports a Boolean type and two constants False and True. If one of these is passed from Python to Perl, the value is represented by an Inline::Python::Boolean object that uses overload to behave like 1 or undef in boolean context in Perl. When this object is passed back to Python, it is translated back to the False or True constant it originated from.

To pass a Boolean value that originated from Perl to Python use the two constants $Inline::Python::Boolean::true and $Inline::Python::Boolean::false if it is important that the value is of type Boolean in Python.

Using Perl inside Python (inside Perl)

This section doesn't talk at all about "Inline::Python". It's about how to use "perl". "perl" is a Python module bundled with Inline::Python that gives you access to Perl from inside your Python code. In the future, it will be possible to compile Inline::Python to work the other way around --- to use Python as the main programming language, and jump into Perl when you want to.

The "perl" package exposes Perl packages and subs. It uses the same code as Inline::Python to automatically translate parameters and return values as needed. Packages and subs are represented as "PerlPkg" and "PerlSub", respectively.

Using the PerlPkg Type

The "perl" package is actually not a package at all. As soon as you import it, it replaces itself with an instance of the PerlPkg class, wrapping the Perl package ``main''. Perl's 'main' package is analogous to '__main__' in Python.

Here's what you can do with the 'main' PerlPkg:


   eval(source code)

Unlike Python, Perl has no exec() --- the eval() function always returns the result of the code it evaluated. eval() takes exactly one argument, the perl source code, and returns the result of the evaluation.

require() and use()

   require(module name)
   use(module name)

Use require() instead of "import". In Python, you'd say this:

   import md5

But using the perl module, you'd say this:


Of course, in Perl there's more than one way to do it (TM). require() doesn't run the package's import() function. If you want symbols exported, for instance, use use() instead of require().

Here is the functionality common to all PerlPkg instances:


Python's __getattr__() function allows the package to dynamically return something to satisfy the request. For instance, you can get at the subs in a perl package by using dir() (which is the same as "getattr(perl, '__methods__')".

Here's an example:

   perl.eval("sub f { 10 }")    # define main::f
   f = perl.f
   f(); f("hello")              # no argument checking
   if perl.f() != 10: 
       import sys; sys.exit(1)

Notice what happens. First we call eval() to define a sub 'f'. Then we say "perl.f", which goes into the __getattr__() method. We check the Perl namespace and see a function called f, which we return, wrapped in an instance of the PerlSub type.

Accessing a perl object's data

__getattr__ may also be used to access a Perl object's attributes, just like Python allows. The Perl object just has to implement a sub __getattr__ returning the requested attribute, which may even be calculated on the fly.

An example for the common hash based objects:

   sub __getattr__ {
       my ($self, $attr) = @_;
       return $self->{$attr}; 

This allows Python code to access the perl object's data like:

   print my_perl_object.field_name

named arguments

When a Perl sub is called with named arguments from Python code, Inline::Python follows the PyObject_Call protocol: positional arguments are given as array ref followed by named arguments as a hash ref. A Perl method supporting named arguments would therefore look like:

    sub supports_named_arguments {
        my ($self, $positional, $named) = @_;
        foreach (qw( named1 named2 )) {
            last unless @$positional;
            $named->{$_} = shift @$positional;

If this method is called using only positional arguments, they would just be pushed into @_ like in any other method, complicating it to:

    sub supports_named_arguments {
        my ($self, $positional, $named) = @_;
        if (@_ == 3 and $size and ref $size and ref $size eq 'ARRAY' and ref $useimage eq 'HASH') { # called using named parameters
            foreach (qw( named1 named2 ... )) {
                last unless @$positional;
                $named->{$_} = shift @$positional;
        else {
            $named = { named1 => $positional, named2 => $named, named3 => $_[3], ... };

As this adds a lot of boiler plate code to subroutines, it is better to just use Perl named arguments conventions (single hashref parameter) if possible.

Using the PerlSub Type

All Perl subs are wrapped in the PerlSub type, so that they can emulate Python subroutines. You can call them. It's all good. Here's what you can do with PerlSub objects:


PerlSub catches the call action and forwards the call to the real sub in Perl.

Set the evaluation flags

Perl has this notion of calling context. A subroutine can ask Perl what it is being used for. The idea is that if no one cares about your return value, you might be able to save time by not building it. By default, PerlSub objects evaluate in 'list' context with no extra flags turned on.

   perl.eval("sub f { 10 }")
   f = perl.f
   f.flags = f.flags | f.G_SCALAR
   x = f()

Here are the most common flags you'll need. For more details about these and other possible flags, see perlcall.


Calls the Perl subroutine in a void context. Guarantees that no results will be returned. If any are returned, Perl deletes them.


Calls the Perl subroutine in a scalar context. Ensures that only one element is returned from the sub. If the sub returns a list, only the last element is actually saved.


Calls the Perl subroutine in a list context. Ensures that any items returned from the subroutine are returned. This is the default for PerlSub objects.


If you are not interested in the return values, you can optimize slightly by telling Perl, and it will discard all returned values for you.


If you are not passing any arguments, you can optimize the call so that Perl doesn't bother setting up the stack for parameters.


It is possible for the Perl sub to fail, either by calling die() explicitly or by calling a non-existent sub. By default, the process will terminate immediately. To avoid this happening, you can trap the exception using the G_EVAL flag.

Under the Hood

When Inline::Python imports a class or function, it creates subs in Perl which delegate the action to some C functions I've written, which know how to call Python functions and methods.

   use Inline Python => <<'END';
   class Foo:
      def __init__(self):
         print "new Foo object being created" = {}
      def get_data(self): return
      def set_data(self,dat):  = dat

Inline::Python actually generates this code and eval()s it:

   package main::Foo;
   @main::Foo::ISA = qw(Inline::Python::Object);
   sub new {
     splice @_, 1, 0, "__main__", "Foo";
     return &Inline::Python::py_new_object;
   sub set_data {
     splice @_, 1, 0, "set_data";
     return &Inline::Python::py_call_method;
   sub get_data {
     splice @_, 1, 0, "get_data";
     return &Inline::Python::py_call_method;
   sub __init__ {
     splice @_, 1, 0, "__init__";
     return &Inline::Python::py_call_method;

More about those "py_*" functions, and how to generate this snippet of code yourself, in the next section.

The Do-it-yourselfer's Guide to Inline::Python

Sometimes you don't actually want to do things the Inline Way. Maybe you just want to use a Python class as-is, without ever treating it like a normal Perl class:

   use Inline::Python qw(py_eval);
   class MyClass:
       def __init__(self): = {}
       def put(self, key, value):[key] = value
       def get(self, key):
           try: return[key]
           except KeyError: return None
   my $o = Inline::Python::Object->new('__main__', 'MyClass');
   $o->put("candy", "yummy");
   die "Ooops" unless $o->get("candy") eq 'yummy';

Inline::Python provides a full suite of exportable functions you can use to manipulate Python objects and functions ``directly''.


   py_eval("python source code", [context])

The new py_eval() behaves a little like Perl's eval(). It evaluates the code or croaks on failure. The optional context argument can be used to place restrictions on the type of code allowed, as well as influence what happens to the result.

Accepts only expressions. Complete statements yield a syntax error. An expression is anything that can appear to the right of an '=' sign. Returns the value of the expression.
The default. Accepts arbitrarily long input, which may be any valid Python code. Always returns "undef".
Accepts exactly one statement, and prints the result to STDOUT. This is how Python works in interactive mode. Always returns "undef".


   py_call_function("package", "function", args...)

This function runs a Python function and returns the result. The ``package'' and ``function'' uniquely identify a function, and the remaining args are passed to the function.

Those who know Python well enough will know you can actually ``run'' a class and get an instance of that class back. But in case that's just too weird for you, I've given you a slightly higher-level wrapper around that common idiom.


   py_new_object("perl package", "python package", 
                 "python class", args...)

This function creates an instance of a Python class. The ``python class'' is the name of the class inside the ``python package''. The new object is blessed into the given ``perl package''. The remaining args are passed directly to the constructor.


   py_call_method(object, "method name", args...)

Given an instance of a Python class, this function can call a method on it. This is useful if you have an object which is blessed into a non-existent Perl package. Attempts to use Perl's object syntax would fail, because Perl wouldn't find any methods in that package. But py_call_method() can always perform method calls correctly since it unwraps the underlying Python object.


Unlike in previous releases of Inline::Python, eval_python() can now return the result of the code. As before, eval_python() is overloaded:
eval_python(code, [context])

Evaluate the code using py_eval().

eval_python(python package, function, args...)

Run the given function and return the results using py_call_function().

eval_python(object, method, args...)

Invoke the given method on the object using py_call_method() and return the results.


   py_bind_func("symbol name", "python package", "function")

This function imports a Python function (named ``function'') as the symbol named by ``perl symbol''. After this function has been called, the Python function can be called as if it were a Perl function in the given package.

   use Inline::Python qw(py_eval py_bind_func);
   def Foo():
      return 42
   # For the purposes of this example, so I know the package, I set it:
   py_bind_func("main::Bar", "__main__", "Foo");
   print "The meaning of life is: ", Bar(), "\n";

This call to py_bind_func() will generate this code and eval() it:

   sub main::Bar {
       unshift @_, "__main__", "Foo";
       return &Inline::Python::py_call_function;


   py_bind_class("perl package", "python package", "class", methods...)

This function imports a Python class (named ``class'') into the Perl package named by ``perl package''. After this function has been called, the Perl package will look just like a regular Perl class.

The example I showed earlier in the ``Under the Hood'' section shows the output of py_bind_class. Here's another look at it:

   use Inline::Python qw(py_eval py_bind_class);
   class Foo:
      def __init__(self):
         print "new Foo object being created" = {}
      def get_data(self): return
      def set_data(self,dat):  = dat
   py_bind_class("main::Foo", "__main__", "Foo", "set_data", "get_data");
   my $o = new Foo;

This call to py_bind_class() will generate this code and eval() it:

   package main::Foo;
   @main::Foo::ISA = qw(Inline::Python::Object);
   sub new {
     splice @_, 1, 0, "__main__", "Foo";
     return &Inline::Python::py_new_object;
   sub set_data {
     splice @_, 1, 0, "set_data";
     return &Inline::Python::py_call_method;
   sub get_data {
     splice @_, 1, 0, "get_data";
     return &Inline::Python::py_call_method;

Note that if you want methods to be created as I've shown, you must pass them to py_bind_class() yourself. It doesn't create anything except new() and the @ISA array. It doesn't need to, since the base class knows how to deal with any method call --- but it's also slower, since it has to walk up the inheritance tree to the AUTOLOAD method. I recommend binding to the functions you know about, especially if you're the one writing the code. If it's auto-generated, use py_study_package(), described below.



This function interrogates the Python interpreter about the given package (or '__main__' if you don't specify one). It returns a list of key/value pairs, so it should be used like this:

   py_eval('import pickle');
   my %namespace = py_study_package("pickle");

On my machine, %namespace looks something like this:

   $VAR1 = {
             'classes' => { ... },
             'functions' => [

Each result can be fed to py_bind_function() and py_bind_class(), which is exactly what Inline::Python itself does.


   my $array_ref = py_eval('(1, 2)')
   $is_tuple = py_is_tuple($array_ref)

This function can tell you if the array reference you got from calling some Python code was a tuple in Python or not (e.g. a normal array). This can be useful if an API requires a distinction between those cases. py_is_tuple works by looking for a magic marker put onto array refs by Py2Pl. Bear in mind that this marker may get lost when copying the array data.


This is a production quality release of Inline::Python. It is fairly feature complete and runs stable with no known crasher bugs or memory leaks. Further testing and expanded support for other operating systems and platforms will be a focus for future releases.

When reporting a bug, please do the following:

 - Put "use Inline REPORTBUG;" at the top of your code, or 
   use the command line option "perl -MInline=REPORTBUG ...".
 - Run your code.
 - Follow the printed instructions.

Here are some things to watch out for:

Note that the namespace imported into Perl is NOT recursively traversed. Only Python globals are imported into Perl --- subclasses, subfunctions, and other modules are not imported.


   use Inline Python => <<'END';
   import mymodule
   class A: 
       class B: pass

The namespace imported into perl is ONLY that related to "A". Nothing related to "mymodule" or "B" is imported, unless some Python code explicitly copies variables from the mymodule namespace into the global namespace before Perl binds to it.


Inline::Python has been tested on RedHat Linux 6.2 with a variety of different Perl and Python configurations. It also seems to be running pretty well on openSUSE at least from 10.3 to 13.1 and on Solaris. Previous versions of Inline::Python worked on Windows and Cygwin --- this version has never been tested there. I strongly suspect it will require patching. Please send me patches.

This version of Inline::Python has been tested with Python versions from 2.5 to 2.7 and from 3.1 to 3.4.


First of all, follow the Python guide from 2 to 3:

For Perl integration:

 - Non-utf8-flagged Perl strings will be Python bytes, utf8-flagged Perl strings will be Python string
 - __cmp__ is no more supported in Python 3 and has been replaced by "rich comparison" (i.e. __eq__, __le__, etc.).
Since booleans in Perl are integers, renaming __cmp__ to __eq__ is often enough while wrapping a Perl object in Python.
 - perl.require, perl.use and perl.eval accept either bytes or strings.


The Github repository for this project is at <>. Pull requests are welcome.


Neil Watkiss <[email protected]>

Brian Ingerson <[email protected]> is the author of Inline, Inline::C and Inline::CPR. He was responsible for much encouragement and many suggestions throughout the development of Inline::Python.

Eric Wilhelm provided support for 'new-style' classes in version 0.21. Many thanks, Eric!

Stefan Seifert <[email protected]> fixed some bugs and is current co-maintainer.


Copyright (c) 2001, Neil Watkiss.

All Rights Reserved. This module is free software. It may be used, redistributed and/or modified under the same terms as Perl itself.