Source

pyobjc / pyobjc-core / Doc / blocks.rst

Full commit

PyObjC support for "blocks"

Introduction

Objective-C has the concept of "blocks", which are basically anonymous inline functions. The syntax for them is like this:

^{ printf("x is %d\n", 42); }

This is a literal for a block that takes no arguments and prints a value when called.

Blocks are only suppored when PyObjC is compiled using an Objective-C compiler that also supports blocks.

Calling blocks from Python

The Python representation for a block is a callable object, that is you can call the block just like you call any other function object.

PyObjC manages the memory for blocks, it is not necessary to manage the reference counts of blocks in your code.

Limitations

It is not possible to call arbitrary blocks because PyObjC needs to store some additional metadata for a block. This means it is only possible to call blocks where the bridge knows the call signature, which means:

  • Block was returned from a method for which we know the signature of returned blocks. PyObjC ships with metadata that covers all of Cocoa.
  • When a block is stored in a Cocoa datastructure, such as an NSArray, and that is the only reference to the block PyObjC will loose the additional information that is needed to call the block.

It is possible to retrieve and set the call signature of a block using the __block_signature__ attribute on blocks.

Implementing blocks in Python

It is very easy to use Objective-C methods that have a block as one of their arguments: just pass an arbitrary callable. PyObjC will automaticly wrap your callable in the right low-level datastructure.

One of the side-effects of this is that the variour storage classes that are defined for block-related variables are not relevant for Python users. Blocks behave just like regular functions.

Metadata for blocks

The current implementation of blocks doesn't allow for full introspection, which means that PyObjC must be taught about the signatures of blocks.