cexi aims to facilitate developing of C extension for CPython. In particular interfacing between python and C. In general building C extensions is a tedious process, especially when it comes to interfacing between two languages. cexi aims to make it easier by providing a simple interface to write C extensions in a pythonic way. It's not a replacement for Cython, CFFI, SWIG, Boost.Python or PyBind11, but rather a tool to make writing C extensions easier. It's main features are:
- pythonic interface to write C extensions
- automatic re/compilation of C extensions
- automatic loading of C extensions
from cexi import Module, s
class Foo(Module):
class options:
flags = ['-O3']
"""
#include "stdio.h"
"""
@s.cee
def foo(x: int) -> int:
"""
return x + 1;
"""
@s.share
def bar(self, x: int) -> int:
return x ** 2
@s.py
def baz(x: int, y: int) -> (int, int):
"""
printf("baz\\n");
int ret = x + y;
bar(ret, &ret);
return(foo(x + y), ret);
"""
def wrapped_baz(self, x, y):
print(f'calling {self.baz}')
return self.baz(x, y)
Foo().foo(1) # => 2
ret = Foo().wrapped_baz(3, 4) # => prints
# calling <cexi.proxy.Proxy object at 0x7f4bae56cf10>
# baz
print(ret) # => prints (8, 49)
class Foo(Module, near=__file__):
pass
class Foo(Module):
directory = Path('/some/path')
Each extension begins with a class definition that inherits from cexi.Module
. User can define flags
that will be passed to compiler via options
class attribute.
class Foo(Module):
class options:
flags = ['-O3']
Module docstring is directly inserted at the beginning of cee extension. It's useful to include headers that will be used in extension's code.
class Foo(Module):
"""
#include "stdio.h"
"""
Then user can define functions that will be compiled as extension's functions. There are three types of functions:
@s.cee
- plain cee function, it's useful to organize plain cee routines..cee
functions aren't available for use in python, they're simply excluded from final class namespace@s.share
- .share is a plain python function that is added to module to be used in cee code it's signature is transformed as such:func([input parameters]) -> <output parameters>
=>int func([input parameters], [pointers to output parameters])
e.g. bar here becomesint bar(int x, int *a)
, it returns non-zero value on failure & modify pointer to the return value@s.py
- .py is a cee extension function. it's available from python & can use both .cee & .share functions at will Untagged functions are left untouched, i.e. they're just instance's methods. cexi.Module children act as singletones - on instantiation they're get loaded/compiled by default they are built inside temporaty directories, but users are able to make modules persistent. There are two ways to do it:near
parameter - if it's set to__file__
module will be built in directory at__file__/../<module name>_cexi_module
directory
class attribute - if it's set module will be built in specified directory
Persistent extensions compiled on-demand. If extension is persistent cexi first loads it's module. Then it compares revisions, which are hashed module's contents. Thus if class definition is changed since last compilation it will be re-compiled and re-loaded.
You can also compile persistent extensions with command
python -mcexi <module path>:<class name>
pip install cexi