Pure Python Mode¶
In some cases, it’s desirable to speed up Python code without losing the ability to run it with the Python interpreter. While pure Python scripts can be compiled with Cython, it usually results only in a speed gain of about 20%-50%.
To go beyond that, Cython provides language constructs to add static typing
and cythonic functionalities to a Python module to make it run much faster
when compiled, while still allowing it to be interpreted.
This is accomplished via an augmenting
.pxd file, via Python
type PEP-484 type annotations (following
PEP 484 and
PEP 526), and/or
via special functions and decorators available after importing the magic
cython module. All three ways can be combined at need, although
projects would commonly decide on a specific way to keep the static type
information easy to manage.
Although it is not typically recommended over writing straight Cython code
.pyx file, there are legitimate reasons to do this - easier
testing and debugging, collaboration with pure Python developers, etc.
In pure mode, you are more or less restricted to code that can be expressed
(or at least emulated) in Python, plus static type declarations. Anything
beyond that can only be done in .pyx files with extended language syntax,
because it depends on features of the Cython compiler.
Using an augmenting
.pxd allows to let the original
completely untouched. On the other hand, one needs to maintain both the
.pxd and the
.py to keep them in sync.
While declarations in a
.pyx file must correspond exactly with those
.pxd file with the same name (and any contradiction results in
a compile time error, see pxd files), the untyped definitions in a
.py file can be overridden and augmented with static types by the more
specific ones present in a
.pxd file is found with the same name as the
being compiled, it will be searched for
cdef classes and
cpdef functions and methods. The compiler will
then convert the corresponding classes/functions/methods in the
file to be of the declared type. Thus if one has a file
def myfunction(x, y=2): a = x - y return a + x * y def _helper(a): return a + 1 class A: def __init__(self, b=0): self.a = 3 self.b = b def foo(self, x): print(x + _helper(1.0))
cpdef int myfunction(int x, int y=*) cdef double _helper(double a) cdef class A: cdef public int a, b cpdef foo(self, double x)
then Cython will compile the
A.py as if it had been written as follows:
cpdef int myfunction(int x, int y=2): a = x - y return a + x * y cdef double _helper(double a): return a + 1 cdef class A: cdef public int a, b def __init__(self, b=0): self.a = 3 self.b = b cpdef foo(self, double x): print(x + _helper(1.0))
Notice how in order to provide the Python wrappers to the definitions
.pxd, that is, to be accessible from Python,
Python visible function signatures must be declared as cpdef (with default arguments replaced by a * to avoid repetition):
cpdef int myfunction(int x, int y=*)
C function signatures of internal functions can be declared as cdef:
cdef double _helper(double a)
cdef classes (extension types) are declared as cdef class;
cdef class attributes must be declared as cdef public if read/write Python access is needed, cdef readonly for read-only Python access, or plain cdef for internal C level attributes;
cdef class methods must be declared as cpdef for Python visible methods or cdef for internal C methods.
In the example above, the type of the local variable a in myfunction()
is not fixed and will thus be a Python object. To statically type it, one
can use Cython’s
@cython.locals decorator (see Magic Attributes,
and Magic Attributes within the .pxd).
Normal Python (
def) functions cannot be declared in
files. It is therefore currently impossible to override the types of plain
Python functions in
.pxd files, e.g. to override types of their local
variables. In most cases, declaring them as cpdef will work as expected.
Special decorators are available from the magic
cython module that can
be used to add static typing within the Python file, while being ignored
by the interpreter.
This option adds the
cython module dependency to the original code, but
does not require to maintain a supplementary
.pxd file. Cython
provides a fake version of this module as Cython.Shadow, which is available
as cython.py when Cython is installed, but can be copied to be used by other
modules when Cython is not installed.
compiledis a special variable which is set to
Truewhen the compiler runs, and
Falsein the interpreter. Thus, the code
import cython if cython.compiled: print("Yep, I'm compiled.") else: print("Just a lowly interpreted script.")
will behave differently depending on whether or not the code is executed as a compiled extension (
.pyd) module or a plain
cython.declaredeclares a typed variable in the current scope, which can be used in place of the
cdef type var [= value]construct. This has two forms, the first as an assignment (useful as it creates a declaration in interpreted mode as well):
import cython x = cython.declare(cython.int) # cdef int x y = cython.declare(cython.double, 0.57721) # cdef double y = 0.57721
and the second mode as a simple function call:
import cython cython.declare(x=cython.int, y=cython.double) # cdef int x; cdef double y
It can also be used to define extension type private, readonly and public attributes:
import cython @cython.cclass class A: cython.declare(a=cython.int, b=cython.int) c = cython.declare(cython.int, visibility='public') d = cython.declare(cython.int) # private by default. e = cython.declare(cython.int, visibility='readonly') def __init__(self, a, b, c, d=5, e=3): self.a = a self.b = b self.c = c self.d = d self.e = e
@cython.localsis a decorator that is used to specify the types of local variables in the function body (including the arguments):
import cython @cython.locals(a=cython.long, b=cython.long, n=cython.longlong) def foo(a, b, x, y): n = a * b # ...
@cython.returns(<type>)specifies the function’s return type.
@cython.exceptval(value=None, *, check=False)specifies the function’s exception return value and exception check semantics as follows:
@exceptval(-1) # cdef int func() except -1: @exceptval(-1, check=False) # cdef int func() except -1: @exceptval(check=True) # cdef int func() except *: @exceptval(-1, check=True) # cdef int func() except? -1: @exceptval(check=False) # no exception checking/propagation
If exception propagation is disabled, any Python exceptions that are raised inside of the function will be printed and ignored.
There are numerous types built into the Cython module. It provides all the
standard C types, namely
as well as their unsigned versions
ulonglong. The special
bint type is used for C boolean values and
Py_ssize_t for (signed) sizes of Python containers.
For each type, there are pointer types
pp_int, etc., up to
three levels deep in interpreted mode, and infinitely deep in compiled mode.
Further pointer types can be constructed with
and arrays as
cython.int. A limited attempt is made to emulate these
more complex types, but only so much can be done from the Python language.
The Python types int, long and bool are interpreted as C
bint respectively. Also, the Python builtin types
tuple, etc. may be used, as well as any user defined types.
Typed C-tuples can be declared as a tuple of C types.
Extension types and cdef functions¶
The class decorator
The function/method decorator
cpdeffunction, i.e. one that Cython code can call at the C level.
@cython.localsdeclares local variables (see above). It can also be used to declare types for arguments, i.e. the local variables that are used in the signature.
@cython.inlineis the equivalent of the C
@cython.finalterminates the inheritance chain by preventing a type from being used as a base class, or a method from being overridden in subtypes. This enables certain optimisations such as inlined method calls.
Here is an example of a
@cython.cfunc @cython.returns(cython.bint) @cython.locals(a=cython.int, b=cython.int) def c_compare(a,b): return a == b
cython.cimports package name gives access to cimports
in code that uses Python syntax. Note that this does not mean that C
libraries become available to Python code. It only means that you can
tell Cython what cimports you want to use, without requiring special
syntax. Running such code in plain Python will fail.
from cython.cimports.libc import math def use_libc_math(): return math.ceil(5.5)
Since such code must necessarily refer to the non-existing
cython.cimports ‘package’, the plain cimport form
cimport cython.cimports... is not available.
You must use the form
Further Cython functions and declarations¶
addressis used in place of the
cython.declare(x=cython.int, x_ptr=cython.p_int) x_ptr = cython.address(x)
sizeofemulates the sizeof operator. It can take both types and expressions.
cython.declare(n=cython.longlong) print(cython.sizeof(cython.longlong)) print(cython.sizeof(n))
structcan be used to create struct types.:
MyStruct = cython.struct(x=cython.int, y=cython.int, data=cython.double) a = cython.declare(MyStruct)
is equivalent to the code:
cdef struct MyStruct: int x int y double data cdef MyStruct a
unioncreates union types with exactly the same syntax as
typedefdefines a type under a given name:
T = cython.typedef(cython.p_int) # ctypedef int* T
castwill (unsafely) reinterpret an expression type.
cython.cast(T, t)is equivalent to
<T>t. The first attribute must be a type, the second is the expression to cast. Specifying the optional keyword argument
typecheck=Truehas the semantics of
t1 = cython.cast(T, t) t2 = cython.cast(T, t, typecheck=True)
Magic Attributes within the .pxd¶
The special cython module can also be imported and used within the augmenting
.pxd file. For example, the following Python file
def dostuff(n): t = 0 for i in range(n): t += i return t
can be augmented with the following
import cython @cython.locals(t=cython.int, i=cython.int) cpdef int dostuff(int n)
cython.declare() function can be used to specify types for global
variables in the augmenting
PEP-484 type annotations¶
Python type hints
can be used to declare argument types, as shown in the
following example. To avoid conflicts with other kinds of annotation
usages, this can be disabled with the directive
import cython def func(foo: dict, bar: cython.int) -> tuple: foo["hello world"] = 3 + bar return foo, 5
Note the use of
cython.int rather than
int - Cython does not translate
int annotation to a C integer by default since the behaviour can be
quite different with respect to overflow and division.
Annotations can be combined with the
@cython.exceptval() decorator for non-Python
import cython @cython.exceptval(-1) def func(x: cython.int) -> cython.int: if x < 0: raise ValueError("need integer >= 0") return x + 1
Note that the default exception handling behaviour when returning C numeric types
is to check for
-1, and if that was returned, check Python’s error indicator
for an exception. This means, if no
@exceptval decorator is provided, and the
return type is a numeric type, then the default with type annotations is
@exceptval(-1, check=True), in order to make sure that exceptions are correctly
and efficiently reported to the caller. Exception propagation can be disabled
@exceptval(check=False), in which case any Python exceptions
raised inside of the function will be printed and ignored.
Since version 0.27, Cython also supports the variable annotations defined in PEP 526. This allows to declare types of variables in a Python 3.6 compatible way as follows:
import cython def func(): # Cython types are evaluated as for cdef declarations x: cython.int # cdef int x y: cython.double = 0.57721 # cdef double y = 0.57721 z: cython.float = 0.57721 # cdef float z = 0.57721 # Python types shadow Cython types for compatibility reasons a: float = 0.54321 # cdef double a = 0.54321 b: int = 5 # cdef object b = 5 c: long = 6 # cdef object c = 6 pass @cython.cclass class A: a: cython.int b: cython.int def __init__(self, b=0): self.a = 3 self.b = b
There is currently no way to express the visibility of object attributes.
Cython does not support the full range of annotations described by PEP-484.
For example it does not currently understand features from the
Optional or typed containers such as
List[str]. This is partly
because some of these type hints are not relevant for the compilation to
efficient C code. In other cases, however, where the generated C code could
benefit from these type hints but does not currently, help is welcome to
improve the type analysis in Cython.
Tips and Tricks¶
Calling C functions¶
Normally, it isn’t possible to call C functions in pure Python mode as there is no general way to support it in normal (uncompiled) Python. However, in cases where an equivalent Python function exists, this can be achieved by combining C function coercion with a conditional import as follows:
# mymodule.pxd # declare a C function as "cpdef" to export it to the module cdef extern from "math.h": cpdef double sin(double x)
# mymodule.py import cython # override with Python import if not in compiled code if not cython.compiled: from math import sin # calls sin() from math.h when compiled with Cython and math.sin() in Python print(sin(0))
Note that the “sin” function will show up in the module namespace of “mymodule”
here (i.e. there will be a
mymodule.sin() function). You can mark it as an
internal name according to Python conventions by renaming it to “_sin” in the
.pxd file as follows:
cdef extern from "math.h": cpdef double _sin "sin" (double x)
You would then also change the Python import to
from math import sin as _sin
to make the names match again.
Using C arrays for fixed size lists¶
C arrays can automatically coerce to Python lists or tuples. This can be exploited to replace fixed size Python lists in Python code by C arrays when compiled. An example:
import cython @cython.locals(counts=cython.int, digit=cython.int) def count_digits(digits): """ >>> digits = '01112222333334445667788899' >>> count_digits(map(int, digits)) [1, 3, 4, 5, 3, 1, 2, 2, 3, 2] """ counts =  * 10 for digit in digits: assert 0 <= digit <= 9 counts[digit] += 1 return counts
In normal Python, this will use a Python list to collect the counts, whereas Cython will generate C code that uses a C array of C ints.