Static Python <--> Cocoa bridge through code generation
Python Objective-C
Latest commit 6db8ff4 Jan 10, 2016 @hsoft hsoft v1.3.2
Failed to load latest commit information.
demos Fix demos Dec 26, 2015
objp Fix bug in generate_python_proxy_code_from_clsspec() Dec 26, 2015
.gitignore Update gitignore Dec 26, 2015
CHANGES.rst v1.3.2 Jan 10, 2016
LICENSE v1.0.0 Jan 16, 2012 README --> README.rst Oct 19, 2013
README.rst README --> README.rst Oct 19, 2013 v1.3.2 Jan 10, 2016


ObjP's goal is to create a two-way bridge between Python and Objective-C. Unlike PyObjC, which uses dynamic calls to methods on runtime, ObjP generates static code. It generates either Objective-C interfaces to Python code or Python modules to interface Objective-C code.

The library is exceedingly simple and it's intended that way. Unlike PyObjC, there's no way ObjP could possibly wrap the whole Cocoa framework, there's way too many things to support. ObjP is made to allow you to bridge your own code.

Also note that ObjP works on Python 3.2 and up.

The best way to learn how to use ObjP is, I think, to look at an example. There are many of them in the 'demos' subfolder. These are built using waf (it's already included in here, no need to install it). For example, if you want to build the simple demo, do:

$ cd demos/simple
$ ./waf configure build
$ cd build
$ ./HelloWorld

That programs calls a simple Python script from Objective-C, and that python script itself calls an Objective-C class.


There are two types of bridge: Objective-C class wrapping a Python class (o2p) and Python class wrapping an Objective-C class (p2o).

To generate an o2p wrapper, you need a target class. Moreover, for this class' methods to be wrapped, you need to have its arguments and return value correctly annotated (You can browse the demos for good examples of how to do it). This is an example of a correctly annotated class:

class Foo:
    def hello_(self, name: str) -> str:
        return "Hello {}".format(name)

To wrap this class, you'll use objp.o2p.generate_objc_code() in this fashion:

import foo
import objp.o2p
objp.o2p.generate_objc_code(foo.Foo, 'destfolder')

This will generate "Foo.h|m" as well as "ObjP.h|m" in "destfolder". These source files directly use the Python API and have no other dependencies.

To generate a p2o wrapper, you either need an Objective-C header file containing an interface or protocol or a Python class describing that interface:

@interface Foo: NSObject {}
- (NSString *)hello:(NSString *)name;

To generate a python wrapper from this, you can do:

import objp.p2o
objp.p2o.generate_python_proxy_code(['Foo.h'], 'destfolder/Foo.m')

This will generate the code for a Python extension module wrapping Foo. The name of the extension module is determined by the name of the destination source file. You can wrap more than one class in the same unit:

objp.p2o.generate_python_proxy_code(['Foo.h', 'Bar.h'], 'destfolder/mywrappers.m')

Method name conversion

ObjP follows PyObjC's convention for converting method names. The ":" character being illegal in Python method names, they're replaced by underscores. Thus, a method - (BOOL)foo:(NSInteger)arg1 bar:(NSString *)arg2; is converted to def foo_bar_(self, arg1: int, arg2: str) -> bool: and vice versa.

Note that if your method's argument count doesn't correspond to the number of underscores in your method name, objp will issue a warning and ignore the method.

Argument Types

Only a few argument types are supported by ObjP, the goal being to keep the project simple.

  • int/NSInteger
  • float/CGFloat
  • str/NSString*
  • bool/BOOL
  • list/NSArray*
  • dict/NSDictionary*
  • nspoint/NSPoint
  • nssize/NSSize
  • nsrect/NSRect

ObjP also supports object which dynamically converts the argument depending on its type and returns an NSObject subclass (which means that int, float and bool convert to NSNumber instead of converting to NSInteger, CGFloat and BOOL). This type of conversion is used to convert the contents of list and dict (it's impossible to have an NSArray directly containing BOOL).

Another special argument type is pyref (which you must import from objp.util in your code) which simply passes the PyObject* instance around without converting it.

The structure arguments allow you to transform tuples to native objc structures and vice-versa. Python has no "native" structure for points, sizes and rects, so that's why we convert to/from tuples ((x, y), (w, h) and (x, y, w, h)). Like pyref, the ns* signature arguments have to be imported from objp.util.


objp.util contains the pyref and ns* argument types, but it also contains two useful method decorators: dontwrap and objcname. A method decorated with dontwrap will be ignored by the code generator, and a method decorated with @objcname('some:selector:') will use this name for generating objc code instead of the automatically generated name.

Constant conversion

When having code in two different languages, we sometimes have to share constants in between the two. To avoid having to manually maintain an Objective-C counterpart to your Python constants, objp offers a small utility, objp.const.generate_objc_code(module, dest). This takes all elements in module's namespace and convert them to an Objective-C's constant unit at dest.

int, float and str types are going to be converted to #define <name> <str(value)> (with @"" around str values). You can also have enum classes in your python constant module. A class is considered an enum if it has integer members. For example:

class Foo:
    Bar = 1
    Baz = 2

will be converted to:

typedef enum {
} Foo;

Because this function will choke on any value that it can't convert, it is recommended that you use it on modules specifically written for that, for example a that imports from your real const unit. Since constants in Objective-C often have prefixes, you can also add them in that unit. It could look like that:

from real_const import FOO as XZFOO, BAR as XZBAR, MyEnum as XZMyEnum