Pyroot 0 0 2_Pythonization Decorator

This tutorial shows how to use the @pythonization decorator to add extra behaviour to C++ user classes that are used from Python via PyROOT.

Author: Enric Tejedor
This notebook tutorial was automatically generated with ROOTBOOK-izer from the macro found in the ROOT repository on Tuesday, May 24, 2022 at 04:26 PM.

In [1]:
import ROOT
from ROOT import pythonization
Welcome to JupyROOT 6.27/01

Let's first define a new C++ class. In this tutorial, we will see how we can "pythonize" this class, i.e. how we can add some extra behaviour to it to make it more pythonic or easier to use from Python.

Note: In this example, the class is defined dynamically for demonstration purposes, but it could also be a C++ class defined in some library or header. For more information about loading C++ user code to be used from Python with PyROOT, please see:

In [2]:
class MyClass {};

Next, we define a pythonizor function: the function that will be responsible for injecting new behaviour in our C++ class MyClass.

To convert a given Python function into a pythonizor, we need to decorate it with the @pythonization decorator. Such decorator allows us to define which which class we want to pythonize by providing its class name and its namespace (if the latter is not specified, it defaults to the global namespace, i.e. '::').

The decorated function - the pythonizor - must accept either one or two parameters:

  1. The class to be pythonized (proxy object where new behaviour can be injected)
  2. The fully-qualified name of that class (optional).

Let's see all this with a simple example. Suppose I would like to define how MyClass objects are represented as a string in Python (i.e. what would be shown when I print that object). For that purpose, I can define the following pythonizor function. There are two important things to be noted here:

  • The @pythonization decorator has one argument that specifies our target class is MyClass.
  • The pythonizor function pythonizor_of_myclass provides and injects a new implementation for __str__, the mechanism that Python provides to define how to represent objects as strings. This new implementation always returns the string "This is a MyClass object".
In [3]:
def pythonizor_of_myclass(klass):
    klass.__str__ = lambda o : 'This is a MyClass object'

Once we have defined our pythonizor function, let's see it in action. We will now use the MyClass class for the first time from Python: we will create a new instance of that class. At this moment, the pythonizor will execute and modify the class - pythonizors are always lazily run when a given class is used for the first time from a Python script.

In [4]:
my_object = ROOT.MyClass()

Since the pythonizor already executed, we should now see the new behaviour. For that purpose, let's print my_object (should show "This is a MyClass object").

In [5]:
This is a MyClass object

The previous example is just a simple one, but there are many ways in which a class can be pythonized. Typical examples are the redefinition of dunder methods (e.g. __iter__ and __next__ to make your objects iterable from Python). If you need some inspiration, many ROOT classes are pythonized in the way we just saw; their pythonizations can be seen at:

The @pythonization decorator offers a few more options when it comes to matching classes that you want to pythonize. We saw that we can match a single class, but we can also specify a list of classes to pythonize.

The following code defines a couple of new classes:

In [6]:
namespace NS {
    class Class1 {};
    class Class2 {};

Note that these classes belong to the NS namespace. As mentioned above, the @pythonization decorator accepts a parameter with the namespace of the class or classes to be pythonized. Therefore, a pythonizor that matches both classes would look like this:

In [7]:
@pythonization(['Class1', 'Class2'], ns='NS')
def pythonize_two_classes(klass):
    klass.new_attribute = 1

Both classes will have the new attribute:

In [8]:
o1 = ROOT.NS.Class1()
o2 = ROOT.NS.Class2()
print("Printing new attribute")
for o in o1, o2:
Printing new attribute

In addition, @pythonization also accepts prefixes of classes in a certain namespace in order to match multiple classes in that namespace. To signal that what we provide to @pythonization is a prefix, we need to set the is_prefix argument to True (default is False).

A common case where matching prefixes is useful is when we have a templated class and we want to pythonize all possible instantiations of that template. For example, we can pythonize the std::vector (templated) class like so:

In [9]:
@pythonization('vector<', ns='std', is_prefix=True)
def vector_pythonizor(klass):
    # first_elem returns the first element of the vector if it exists
    klass.first_elem = lambda v : v[0] if v else None

Since we defined a prefix to do the match, the pythonization will be applied both if we instantiate e.g. a vector of integers and a vector of doubles.

In [10]:
v_int = ROOT.std.vector['int']([1,2,3])
v_double = ROOT.std.vector['double']([4.,5.,6.])
print("First element of integer vector: {}".format(v_int.first_elem()))
print("First element of double vector: {}".format(v_double.first_elem()))
First element of integer vector: 1
First element of double vector: 4.0

Note that specifying a list of class name prefixes is also possible (similarly to what we saw with a list of class names). Again, is_prefix=True is required to signal that we are providing a list of prefixes.

These are some examples of combinations of prefixes and namespaces and the corresponding classes that they match:

  • '' : all classes in the global namespace.
  • '', ns='NS1::NS2' : all classes in the NS1::NS2 namespace.
  • 'Prefix' : classes whose name starts with Prefix in the global namespace.
  • 'Prefix', ns='NS' : classes whose name starts with Prefix in the NS namespace.

Moreover, a pythonizor function can have a second optional parameter that contains the fully-qualified name of the class being pythonized. This can be useful e.g. if we would like to do some more complex filtering of classes in our pythonizor, for instance using regular expressions.

In [11]:
@pythonization('pair<', ns='std', is_prefix=True)
def pair_pythonizor(klass, name):
    print('Pythonizing class ' + name)

The pythonizor above will be applied to any instantiation of std::pair - we can see this with the print we did inside the pythonizor. Note that we could use the name parameter to e.g. further filter which particular instantiations we would like to pythonize.

In [12]:
p1 = ROOT.std.pair['int','int'](1,2) # prints 'Pythonizing class std::pair<int,int>'
p2 = ROOT.std.pair['int','double'](1,2.) # prints 'Pythonizing class std::pair<int,double>'
Pythonizing class std::pair<int,int>
Pythonizing class std::pair<int,double>

Note that, to pythonize multiple classes in different namespaces, we can stack multiple @pythonization decorators. For example, if we define these classes:

In [13]:
class FirstClass {};
namespace NS {
    class SecondClass {};

We can pythonize both of them with a single pythonizor function like so:

In [14]:
@pythonization('SecondClass', ns='NS')
def pythonizor_for_first_and_second(klass, name):
    print('Executed for class ' + name)

If we now access both classes, we should see that the pythonizor runs twice.

In [15]:
f = ROOT.FirstClass()
s = ROOT.NS.SecondClass()
Executed for class FirstClass
Executed for class NS::SecondClass

So far we have seen how pythonizations can be registered for classes that have not been used yet. We have discussed how, in that case, the pythonizor functions are executed lazily when their target class/es are used for the first time in the application. However, it can also happen that our target class/es have already been accessed by the time we register a pythonization. In such a scenario, the pythonizor is applied immediately (at registration time) to the target class/es.

Let's see an example of what was just explained. We will define a new class and immediately create an object of that class. We can check how the object still does not have a new attribute pythonized that we are going to inject in the next step.

In [16]:
class MyClass2 {};
o = ROOT.MyClass2()
except AttributeError:
    print("Object has not been pythonized yet!")
Object has not been pythonized yet!

After that, we will register a pythonization for MyClass2. Since the class has already been used, the pythonization will happen right away.

In [17]:
def pythonizor_for_myclass2(klass):
    klass.pythonized = True

Now our object does have the pythonized attribute:

In [18]:
print(o.pythonized) # prints True