Extension Types

Introduction

As well as creating normal user-defined classes with the Python class statement, Cython also lets you create new built-in Python types, known as extension types. You define an extension type using the cdef class statement. Here’s an example:

from __future__ import print_function

cdef class Shrubbery:
    cdef int width, height

    def __init__(self, w, h):
        self.width = w
        self.height = h

    def describe(self):
        print("This shrubbery is", self.width,
              "by", self.height, "cubits.")

As you can see, a Cython extension type definition looks a lot like a Python class definition. Within it, you use the def statement to define methods that can be called from Python code. You can even define many of the special methods such as __init__() as you would in Python.

The main difference is that you can use the cdef statement to define attributes. The attributes may be Python objects (either generic or of a particular extension type), or they may be of any C data type. So you can use extension types to wrap arbitrary C data structures and provide a Python-like interface to them.

Static Attributes

Attributes of an extension type are stored directly in the object’s C struct. The set of attributes is fixed at compile time; you can’t add attributes to an extension type instance at run time simply by assigning to them, as you could with a Python class instance. However, you can explicitly enable support for dynamically assigned attributes, or subclass the extension type with a normal Python class, which then supports arbitrary attribute assignments. See Dynamic Attributes.

There are two ways that attributes of an extension type can be accessed: by Python attribute lookup, or by direct access to the C struct from Cython code. Python code is only able to access attributes of an extension type by the first method, but Cython code can use either method.

By default, extension type attributes are only accessible by direct access, not Python access, which means that they are not accessible from Python code. To make them accessible from Python code, you need to declare them as public or readonly. For example:

cdef class Shrubbery:
    cdef public int width, height
    cdef readonly float depth

makes the width and height attributes readable and writable from Python code, and the depth attribute readable but not writable.

Note

You can only expose simple C types, such as ints, floats, and strings, for Python access. You can also expose Python-valued attributes.

Note

Also the public and readonly options apply only to Python access, not direct access. All the attributes of an extension type are always readable and writable by C-level access.

Dynamic Attributes

It is not possible to add attributes to an extension type at runtime by default. You have two ways of avoiding this limitation, both add an overhead when a method is called from Python code. Especially when calling cpdef methods.

The first approach is to create a Python subclass.:

cdef class Animal:

    cdef int number_of_legs

    def __cinit__(self, int number_of_legs):
        self.number_of_legs = number_of_legs


class ExtendableAnimal(Animal):  # Note that we use class, not cdef class
    pass


dog = ExtendableAnimal(4)
dog.has_tail = True

Declaring a __dict__ attribute is the second way of enabling dynamic attributes.:

cdef class Animal:

    cdef int number_of_legs
    cdef dict __dict__

    def __cinit__(self, int number_of_legs):
        self.number_of_legs = number_of_legs


dog = Animal(4)
dog.has_tail = True

Type declarations

Before you can directly access the attributes of an extension type, the Cython compiler must know that you have an instance of that type, and not just a generic Python object. It knows this already in the case of the self parameter of the methods of that type, but in other cases you will have to use a type declaration.

For example, in the following function:

cdef widen_shrubbery(sh, extra_width): # BAD
    sh.width = sh.width + extra_width

because the sh parameter hasn’t been given a type, the width attribute will be accessed by a Python attribute lookup. If the attribute has been declared public or readonly then this will work, but it will be very inefficient. If the attribute is private, it will not work at all – the code will compile, but an attribute error will be raised at run time.

The solution is to declare sh as being of type Shrubbery, as follows:

from my_module cimport Shrubbery

cdef widen_shrubbery(Shrubbery sh, extra_width):
    sh.width = sh.width + extra_width

Now the Cython compiler knows that sh has a C attribute called width and will generate code to access it directly and efficiently. The same consideration applies to local variables, for example:

from my_module cimport Shrubbery

cdef Shrubbery another_shrubbery(Shrubbery sh1):
    cdef Shrubbery sh2
    sh2 = Shrubbery()
    sh2.width = sh1.width
    sh2.height = sh1.height
    return sh2

Note

We here cimport the class Shrubbery, and this is necessary to declare the type at compile time. To be able to cimport an extension type, we split the class definition into two parts, one in a definition file and the other in the corresponding implementation file. You should read Sharing Extension Types to learn to do that.

Type Testing and Casting

Suppose I have a method quest() which returns an object of type Shrubbery. To access it’s width I could write:

cdef Shrubbery sh = quest()
print(sh.width)

which requires the use of a local variable and performs a type test on assignment. If you know the return value of quest() will be of type Shrubbery you can use a cast to write:

print( (<Shrubbery>quest()).width )

This may be dangerous if quest() is not actually a Shrubbery, as it will try to access width as a C struct member which may not exist. At the C level, rather than raising an AttributeError, either an nonsensical result will be returned (interpreting whatever data is at that address as an int) or a segfault may result from trying to access invalid memory. Instead, one can write:

print( (<Shrubbery?>quest()).width )

which performs a type check (possibly raising a TypeError) before making the cast and allowing the code to proceed.

To explicitly test the type of an object, use the isinstance() builtin function. For known builtin or extension types, Cython translates these into a fast and safe type check that ignores changes to the object’s __class__ attribute etc., so that after a successful isinstance() test, code can rely on the expected C structure of the extension type and its cdef attributes and methods.

Extension types and None

When you declare a parameter or C variable as being of an extension type, Cython will allow it to take on the value None as well as values of its declared type. This is analogous to the way a C pointer can take on the value NULL, and you need to exercise the same caution because of it. There is no problem as long as you are performing Python operations on it, because full dynamic type checking will be applied. However, when you access C attributes of an extension type (as in the widen_shrubbery function above), it’s up to you to make sure the reference you’re using is not None – in the interests of efficiency, Cython does not check this.

You need to be particularly careful when exposing Python functions which take extension types as arguments. If we wanted to make widen_shrubbery() a Python function, for example, if we simply wrote:

def widen_shrubbery(Shrubbery sh, extra_width): # This is
    sh.width = sh.width + extra_width           # dangerous!

then users of our module could crash it by passing None for the sh parameter.

One way to fix this would be:

def widen_shrubbery(Shrubbery sh, extra_width):
    if sh is None:
        raise TypeError
    sh.width = sh.width + extra_width

but since this is anticipated to be such a frequent requirement, Cython provides a more convenient way. Parameters of a Python function declared as an extension type can have a not None clause:

def widen_shrubbery(Shrubbery sh not None, extra_width):
    sh.width = sh.width + extra_width

Now the function will automatically check that sh is not None along with checking that it has the right type.

Note

not None clause can only be used in Python functions (defined with def) and not C functions (defined with cdef). If you need to check whether a parameter to a C function is None, you will need to do it yourself.

Note

Some more things:

  • The self parameter of a method of an extension type is guaranteed never to be None.

  • When comparing a value with None, keep in mind that, if x is a Python object, x is None and x is not None are very efficient because they translate directly to C pointer comparisons, whereas x == None and x != None, or simply using x as a boolean value (as in if x: ...) will invoke Python operations and therefore be much slower.

Special methods

Although the principles are similar, there are substantial differences between many of the __xxx__() special methods of extension types and their Python counterparts. There is a separate page devoted to this subject, and you should read it carefully before attempting to use any special methods in your extension types.

Properties

You can declare properties in an extension class using the same syntax as in ordinary Python code:

cdef class Spam:

    @property
    def cheese(self):
        # This is called when the property is read.
        ...

    @cheese.setter
    def cheese(self, value):
            # This is called when the property is written.
            ...

    @cheese.deleter
    def cheese(self):
        # This is called when the property is deleted.

There is also a special (deprecated) legacy syntax for defining properties in an extension class:

cdef class Spam:

    property cheese:

        "A doc string can go here."

        def __get__(self):
            # This is called when the property is read.
            ...

        def __set__(self, value):
            # This is called when the property is written.
            ...

        def __del__(self):
            # This is called when the property is deleted.

The __get__(), __set__() and __del__() methods are all optional; if they are omitted, an exception will be raised when the corresponding operation is attempted.

Here’s a complete example. It defines a property which adds to a list each time it is written to, returns the list when it is read, and empties the list when it is deleted.:

# cheesy.pyx
cdef class CheeseShop:

    cdef object cheeses

    def __cinit__(self):
        self.cheeses = []

    @property
    def cheese(self):
        return "We don't have: %s" % self.cheeses

    @cheese.setter
    def cheese(self, value):
        self.cheeses.append(value)

    @cheese.deleter
    def cheese(self):
        del self.cheeses[:]

# Test input
from cheesy import CheeseShop

shop = CheeseShop()
print(shop.cheese)

shop.cheese = "camembert"
print(shop.cheese)

shop.cheese = "cheddar"
print(shop.cheese)

del shop.cheese
print(shop.cheese)
# Test output
We don't have: []
We don't have: ['camembert']
We don't have: ['camembert', 'cheddar']
We don't have: []

Subclassing

An extension type may inherit from a built-in type or another extension type:

cdef class Parrot:
    ...

cdef class Norwegian(Parrot):
    ...

A complete definition of the base type must be available to Cython, so if the base type is a built-in type, it must have been previously declared as an extern extension type. If the base type is defined in another Cython module, it must either be declared as an extern extension type or imported using the cimport statement.

An extension type can only have one base class (no multiple inheritance).

Cython extension types can also be subclassed in Python. A Python class can inherit from multiple extension types provided that the usual Python rules for multiple inheritance are followed (i.e. the C layouts of all the base classes must be compatible).

There is a way to prevent extension types from being subtyped in Python. This is done via the final directive, usually set on an extension type using a decorator:

cimport cython

@cython.final
cdef class Parrot:
   def done(self): pass

Trying to create a Python subclass from this type will raise a TypeError at runtime. Cython will also prevent subtyping a final type inside of the same module, i.e. creating an extension type that uses a final type as its base type will fail at compile time. Note, however, that this restriction does not currently propagate to other extension modules, so even final extension types can still be subtyped at the C level by foreign code.

C methods

Extension types can have C methods as well as Python methods. Like C functions, C methods are declared using cdef or cpdef instead of def. C methods are “virtual”, and may be overridden in derived extension types. In addition, cpdef methods can even be overridden by python methods when called as C method. This adds a little to their calling overhead compared to a cdef method:

# pets.pyx
cdef class Parrot:

    cdef void describe(self):
        print("This parrot is resting.")

cdef class Norwegian(Parrot):

    cdef void describe(self):
        Parrot.describe(self)
        print("Lovely plumage!")


cdef Parrot p1, p2
p1 = Parrot()
p2 = Norwegian()
print("p1:")
p1.describe()
print("p2:")
p2.describe()
# Output
p1:
This parrot is resting.
p2:
This parrot is resting.
Lovely plumage!

The above example also illustrates that a C method can call an inherited C method using the usual Python technique, i.e.:

Parrot.describe(self)

cdef methods can be declared static by using the @staticmethod decorator. This can be especially useful for constructing classes that take non-Python compatible types.:

cdef class OwnedPointer:
    cdef void* ptr

    def __dealloc__(self):
        if self.ptr is not NULL:
            free(self.ptr)

    @staticmethod
    cdef create(void* ptr):
        p = OwnedPointer()
        p.ptr = ptr
        return p

Forward-declaring extension types

Extension types can be forward-declared, like struct and union types. This is usually not necessary and violates the DRY principle (Don’t Repeat Yourself).

If you are forward-declaring an extension type that has a base class, you must specify the base class in both the forward declaration and its subsequent definition, for example,:

cdef class A(B)

...

cdef class A(B):
    # attributes and methods

Fast instantiation

Cython provides two ways to speed up the instantiation of extension types. The first one is a direct call to the __new__() special static method, as known from Python. For an extension type Penguin, you could use the following code:

cdef class Penguin:
    cdef object food

    def __cinit__(self, food):
        self.food = food

    def __init__(self, food):
        print("eating!")

normal_penguin = Penguin('fish')
fast_penguin = Penguin.__new__(Penguin, 'wheat')  # note: not calling __init__() !

Note that the path through __new__() will not call the type’s __init__() method (again, as known from Python). Thus, in the example above, the first instantiation will print eating!, but the second will not. This is only one of the reasons why the __cinit__() method is safer and preferable over the normal __init__() method for extension types.

The second performance improvement applies to types that are often created and deleted in a row, so that they can benefit from a freelist. Cython provides the decorator @cython.freelist(N) for this, which creates a statically sized freelist of N instances for a given type. Example:

cimport cython

@cython.freelist(8)
cdef class Penguin:
    cdef object food
    def __cinit__(self, food):
        self.food = food

penguin = Penguin('fish 1')
penguin = None
penguin = Penguin('fish 2')  # does not need to allocate memory!

Instantiation from existing C/C++ pointers

It is quite common to want to instantiate an extension class from an existing (pointer to a) data structure, often as returned by external C/C++ functions.

As extension classes can only accept Python objects as arguments in their contructors, this necessitates the use of factory functions. For example,

from libc.stdlib cimport malloc, free

# Example C struct
ctypedef struct my_c_struct:
    int a
    int b


cdef class WrapperClass:
    """A wrapper class for a C/C++ data structure"""
    cdef my_c_struct *_ptr
    cdef bint ptr_owner

    def __cinit__(self):
        self.ptr_owner = False

    def __dealloc__(self):
        # De-allocate if not null and flag is set
        if self._ptr is not NULL and self.ptr_owner is True:
            free(self._ptr)
            self._ptr = NULL

    # Extension class properties
    @property
    def a(self):
        return self._ptr.a if self._ptr is not NULL else None

    @property
    def b(self):
        return self._ptr.b if self._ptr is not NULL else None

    @staticmethod
    cdef WrapperClass from_ptr(my_c_struct *_ptr, bint owner=False):
        """Factory function to create WrapperClass objects from
        given my_c_struct pointer.

        Setting ``owner`` flag to ``True`` causes
        the extension type to ``free`` the structure pointed to by ``_ptr``
        when the wrapper object is deallocated."""
        # Call to __new__ bypasses __init__ constructor
        cdef WrapperClass wrapper = WrapperClass.__new__(WrapperClass)
        wrapper._ptr = _ptr
        wrapper.ptr_owner = owner
        return wrapper

    @staticmethod
    cdef WrapperClass new_struct():
        """Factory function to create WrapperClass objects with
        newly allocated my_c_struct"""
        cdef my_c_struct *_ptr = <my_c_struct *>malloc(sizeof(my_c_struct))
        if _ptr is NULL:
            raise MemoryError
        _ptr.a = 0
        _ptr.b = 0
        return WrapperClass.from_ptr(_ptr, owner=True)

To then create a WrapperClass object from an existing my_c_struct pointer, WrapperClass.from_ptr(ptr) can be used in Cython code. To allocate a new structure and wrap it at the same time, WrapperClass.new_struct can be used instead.

It is possible to create multiple Python objects all from the same pointer which point to the same in-memory data, if that is wanted, though care must be taken when de-allocating as can be seen above. Additionally, the ptr_owner flag can be used to control which WrapperClass object owns the pointer and is responsible for de-allocation - this is set to False by default in the example and can be enabled by calling from_ptr(ptr, owner=True).

The GIL must not be released in __dealloc__ either, or another lock used if it is, in such cases or race conditions can occur with multiple de-allocations.

Being a part of the object constructor, the __cinit__ method has a Python signature, which makes it unable to accept a my_c_struct pointer as an argument.

Attempts to use pointers in a Python signature will result in errors like:

Cannot convert 'my_c_struct *' to Python object

This is because Cython cannot automatically convert a pointer to a Python object, unlike with native types like int.

Note that for native types, Cython will copy the value and create a new Python object while in the above case, data is not copied and deallocating memory is a responsibility of the extension class.

Making extension types weak-referenceable

By default, extension types do not support having weak references made to them. You can enable weak referencing by declaring a C attribute of type object called __weakref__. For example,:

cdef class ExplodingAnimal:
    """This animal will self-destruct when it is
    no longer strongly referenced."""

    cdef object __weakref__

Controlling cyclic garbage collection in CPython

By default each extension type will support the cyclic garbage collector of CPython. If any Python objects can be referenced, Cython will automatically generate the tp_traverse and tp_clear slots. This is usually what you want.

There is at least one reason why this might not be what you want: If you need to cleanup some external resources in the __dealloc__ special function and your object happened to be in a reference cycle, the garbage collector may have triggered a call to tp_clear to drop references. This is the way that reference cycles are broken so that the garbage can actually be reclaimed.

In that case any object references have vanished by the time when __dealloc__ is called. Now your cleanup code lost access to the objects it has to clean up. In that case you can disable the cycle breaker tp_clear by using the no_gc_clear decorator

@cython.no_gc_clear
cdef class DBCursor:
    cdef DBConnection conn
    cdef DBAPI_Cursor *raw_cursor
    # ...
    def __dealloc__(self):
        DBAPI_close_cursor(self.conn.raw_conn, self.raw_cursor)

This example tries to close a cursor via a database connection when the Python object is destroyed. The DBConnection object is kept alive by the reference from DBCursor. But if a cursor happens to be in a reference cycle, the garbage collector may effectively “steal” the database connection reference, which makes it impossible to clean up the cursor.

Using the no_gc_clear decorator this can not happen anymore because the references of a cursor object will not be cleared anymore.

In rare cases, extension types can be guaranteed not to participate in cycles, but the compiler won’t be able to prove this. This would be the case if the class can never reference itself, even indirectly. In that case, you can manually disable cycle collection by using the no_gc decorator, but beware that doing so when in fact the extension type can participate in cycles could cause memory leaks

@cython.no_gc
cdef class UserInfo:
    cdef str name
    cdef tuple addresses

If you can be sure addresses will contain only references to strings, the above would be safe, and it may yield a significant speedup, depending on your usage pattern.

Controlling pickling

By default, Cython will generate a __reduce__() method to allow pickling an extension type if and only if each of its members are convertible to Python and it has no __cinit__ method. To require this behavior (i.e. throw an error at compile time if a class cannot be pickled) decorate the class with @cython.auto_pickle(True). One can also annotate with @cython.auto_pickle(False) to get the old behavior of not generating a __reduce__ method in any case.

Manually implementing a __reduce__ or __reduce_ex__ method will also disable this auto-generation and can be used to support pickling of more complicated types.

Public and external extension types

Extension types can be declared extern or public. An extern extension type declaration makes an extension type defined in external C code available to a Cython module. A public extension type declaration makes an extension type defined in a Cython module available to external C code.

External extension types

An extern extension type allows you to gain access to the internals of Python objects defined in the Python core or in a non-Cython extension module.

Note

In previous versions of Pyrex, extern extension types were also used to reference extension types defined in another Pyrex module. While you can still do that, Cython provides a better mechanism for this. See Sharing Declarations Between Cython Modules.

Here is an example which will let you get at the C-level members of the built-in complex object.:

from __future__ import print_function

cdef extern from "complexobject.h":

    struct Py_complex:
        double real
        double imag

    ctypedef class __builtin__.complex [object PyComplexObject]:
        cdef Py_complex cval

# A function which uses the above type
def spam(complex c):
    print("Real:", c.cval.real)
    print("Imag:", c.cval.imag)

Note

Some important things:

  1. In this example, ctypedef class has been used. This is because, in the Python header files, the PyComplexObject struct is declared with:

    typedef struct {
        ...
    } PyComplexObject;
    

    At runtime, a check will be performed when importing the Cython c-extension module that __builtin__.complex’s tp_basicsize matches sizeof(`PyComplexObject). This check can fail if the Cython c-extension module was compiled with one version of the complexobject.h header but imported into a Python with a changed header. This check can be tweaked by using check_size in the name specification clause.

  2. As well as the name of the extension type, the module in which its type object can be found is also specified. See the implicit importing section below.

  3. When declaring an external extension type, you don’t declare any methods. Declaration of methods is not required in order to call them, because the calls are Python method calls. Also, as with struct and union, if your extension class declaration is inside a cdef extern from block, you only need to declare those C members which you wish to access.

Name specification clause

The part of the class declaration in square brackets is a special feature only available for extern or public extension types. The full form of this clause is:

[object object_struct_name, type type_object_name, check_size cs_option]

Where:

  • object_struct_name is the name to assume for the type’s C struct.

  • type_object_name is the name to assume for the type’s statically declared type object.

  • cs_option is warn (the default), error, or ignore and is only used for external extension types. If error, the sizeof(object_struct) that was found at compile time must match the type’s runtime tp_basicsize exactly, otherwise the module import will fail with an error. If warn or ignore, the object_struct is allowed to be smaller than the type’s tp_basicsize, which indicates the runtime type may be part of an updated module, and that the external module’s developers extended the object in a backward-compatible fashion (only adding new fields to the end of the object). If warn, a warning will be emitted in this case.

The clauses can be written in any order.

If the extension type declaration is inside a cdef extern from block, the object clause is required, because Cython must be able to generate code that is compatible with the declarations in the header file. Otherwise, for extern extension types, the object clause is optional.

For public extension types, the object and type clauses are both required, because Cython must be able to generate code that is compatible with external C code.

Attribute name matching and aliasing

Sometimes the type’s C struct as specified in object_struct_name may use different labels for the fields than those in the PyTypeObject. This can easily happen in hand-coded C extensions where the PyTypeObject_Foo has a getter method, but the name does not match the name in the PyFooObject. In NumPy, for instance, python-level dtype.itemsize is a getter for the C struct field elsize. Cython supports aliasing field names so that one can write dtype.itemsize in Cython code which will be compiled into direct access of the C struct field, without going through a C-API equivalent of dtype.__getattr__('itemsize').

For example we may have an extension module foo_extension:

cdef class Foo:
    cdef public int field0, field1, field2;

    def __init__(self, f0, f1, f2):
        self.field0 = f0
        self.field1 = f1
        self.field2 = f2

but a C struct in a file foo_nominal.h:

typedef struct {
     PyObject_HEAD
     int f0;
     int f1;
     int f2;
 } FooStructNominal;

Note that the struct uses f0, f1, f2 but they are field0, field1, and field2 in Foo. We are given this situation, including a header file with that struct, and we wish to write a function to sum the values. If we write an extension module wrapper:

cdef extern from "foo_nominal.h":

    ctypedef class foo_extension.Foo [object FooStructNominal]:
        cdef:
            int field0
            int field1
            int feild2

def sum(Foo f):
    return f.field0 + f.field1 + f.field2

then wrapper.sum(f) (where f = foo_extension.Foo(1, 2, 3)) will still use the C-API equivalent of:

return f.__getattr__('field0') +
       f.__getattr__('field1') +
       f.__getattr__('field1')

instead of the desired C equivalent of return f->f0 + f->f1 + f->f2. We can alias the fields by using:

cdef extern from "foo_nominal.h":

    ctypedef class foo_extension.Foo [object FooStructNominal]:
        cdef:
            int field0 "f0"
            int field1 "f1"
            int field2 "f2"

def sum(Foo f) except -1:
    return f.field0 + f.field1 + f.field2

and now Cython will replace the slow __getattr__ with direct C access to the FooStructNominal fields. This is useful when directly processing Python code. No changes to Python need be made to achieve significant speedups, even though the field names in Python and C are different. Of course, one should make sure the fields are equivalent.

Implicit importing

Cython requires you to include a module name in an extern extension class declaration, for example,:

cdef extern class MyModule.Spam:
    ...

The type object will be implicitly imported from the specified module and bound to the corresponding name in this module. In other words, in this example an implicit:

from MyModule import Spam

statement will be executed at module load time.

The module name can be a dotted name to refer to a module inside a package hierarchy, for example,:

cdef extern class My.Nested.Package.Spam:
    ...

You can also specify an alternative name under which to import the type using an as clause, for example,:

cdef extern class My.Nested.Package.Spam as Yummy:
   ...

which corresponds to the implicit import statement:

from My.Nested.Package import Spam as Yummy

Type names vs. constructor names

Inside a Cython module, the name of an extension type serves two distinct purposes. When used in an expression, it refers to a module-level global variable holding the type’s constructor (i.e. its type-object). However, it can also be used as a C type name to declare variables, arguments and return values of that type.

When you declare:

cdef extern class MyModule.Spam:
    ...

the name Spam serves both these roles. There may be other names by which you can refer to the constructor, but only Spam can be used as a type name. For example, if you were to explicitly import MyModule, you could use MyModule.Spam() to create a Spam instance, but you wouldn’t be able to use MyModule.Spam as a type name.

When an as clause is used, the name specified in the as clause also takes over both roles. So if you declare:

cdef extern class MyModule.Spam as Yummy:
    ...

then Yummy becomes both the type name and a name for the constructor. Again, there are other ways that you could get hold of the constructor, but only Yummy is usable as a type name.

Public extension types

An extension type can be declared public, in which case a .h file is generated containing declarations for its object struct and type object. By including the .h file in external C code that you write, that code can access the attributes of the extension type.