Metadata-Version: 2.1
Name: fullofeels
Version: 1.7.0
Summary: Creates C code skeleton for handcrafted Python extension modules
Home-page: https://hg.sr.ht/~hfisher/fullofeels
Author: H Fisher
Author-email: hugo.fisher@gmail.com
License: UNKNOWN
Platform: UNKNOWN
Classifier: Programming Language :: Python :: 3
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Requires-Python: >=2.7
Description-Content-Type: text/markdown


# Full of Eels


FullOfEels is an assistant program to generate the skeleton code for
a handcrafted artisanal Python module to be implemented in native C.

It is not a replacement for SWIG or SIP. If you have an existing C/C++
library and want to create a Python wrapper with the same API, either
of those will do just fine. Nor is it a replacement for ctypes.

FullOfEels works in the opposite direction. You start by specifying
the Python API that you want: module, classes, methods, and functions.
This program will create the initial .h/.c files for you, generating
empty C functions with argument parsing code, struct definitions for
the object and type, module init function ... all the the boring but
necessary bits. You then fill them out with working C code.

How do you specify the Python API you want? By writing it in Python.
You create a module that defines _what_ the functions and classes will
be, with all the arguments and methods, but none of the code that
defines _how_ it works - methods and functions should all just `pass`.
FullOfEels keeps docstrings and some globals, but actual statements
are discarded.

Why might you prefer this? Because you can design a more Pythonic API.
The great advantage of SWIG/SIP is that the API is already known and
(ideally) documentation exists. The disadvantage is that you're just
recreating a C/C++ API in Python, and that programmers will have to
be able to understand C/C++ to understand the documentation. There are
times when this is the best solution. Fullofeels is for when it isn't.

Written by Hugh(Hugo) Fisher, Canberra Australia, 2016-now. <br>
Released under MIT license <br>
All feedback welcome: hugo.fisher@gmail.com (Yes, hugo not hugh)


### Example

The Python "Extending Python with C or C++" web pages
start by creating a spam module with a single `system()` function.
With fullOfEels you start this by writing `spam.py` which is just

    def system(command):
        """Execute a shell command."""
        pass

Then you run

    python -m fullofeels -print spam.py

which will create

    spam.h
    spam.c
    setup.py

Look inside and you'll see all the skeleton code needed for the complete
Python extension module. Run

    python setup.py build_ext --inplace

and you will have a valid module. From your Python interpreter you can
import it and invoke the `spam.system("hello")` function, which will just
print out the function name and arguments (specified by the -print CLI option).

Because fullOfEeels is an assistant, not an automated solution, you still
need to fill in the function body yourself to make it do anything useful.


For the "Defining New Types" web example you would write custom.py:

    class Custom(object):
        """Custom objects"""

        def __new__(cls):
            pass

        def __init__(self, first, last, number: int):
            pass

        def __del__(self):
            pass

        def name(self) -> str:
            """Return the name, combining the first and last name"""
            pass

and again run fullOfEels on it. The generated C code will have the all
the code necessary to declare and initialize the Custom type structure.
As before you could build this module, import it, and even create Custom
objects. But without additional code they won't do anything and won't
have any attributes or properties.


### TODO

Heap allocated types

Look at new Python C API for 3.10+


### Tested on

    Python 2.7  Fedora Core 24, MacOS 10.14 XCode 11
    Python 3.5  Fedora Core 24
    Python 3.6  Windows 8.1 MSVC 2017
    Python 3.7  Fedora Core 31, Windows 10 MSVC 2019
    Python 3.9  Fedora Core 31


### Versions

1.7 Mar 2021        Type annotations recognised for parameters

1.6 Feb 2021        Automate testing on Windows, cleanup for PyPI release

1.5 Jan 2019        Replace separate program with -m fullofeels package

1.4 Nov 2017        Generate setup.py for module

1.3 Oct 2017        Properties on class become C fields in object struct

1.2 Feb 2017        Can specify argument types for PyArg_ParseTuple

1.1 Nov 2016        Special slots for collections, numbers, buffer interface

1.0 Apr 2016        Modules, objects, classes, methods


