Hi,
I’ve just pushed PyObjC 3.0.1 to PyPI. This is a major update to the previously uploaded release and contains a rewrite of the core bridge itself. All tests pass without problems, but there could still be problems.
With this update the bridge is a lot lazier about when to add methods to the class proxy __dict__: methods are now only added when they are actually used. This should result in slightly better performance, and more importantly removes some very ugly code that tried very hard to keep the __dict__ up to date without there being support for detecting changes to a class in the Objective-C runtime.
Because of this it is more important that ever to have “from objc import super” at the top of modules that use super calls in subclasses of Cocoa classes. In previous releases this was already necessary to avoid a race condition, but with this release you will definitely run into problems when using builtin.super. I am working on a PEP that adds a new API to Python that would remove the need of a custom super class, but that will at best be present in Python 3.5 (and that’s assuming the PEP is accepted).
The metadata and framework wrappers have not yet been updated with support of OSX 10.9, let alone support of 10.10. That means that those wrappers work fine on OSX 10.9, but any special support for API’s introduced in OSX 10.9 is not present yet. Adding such support is on my todo list, but is a significant amount of work and sadly enough requires updating some tooling I used previously (which means yet more work).
Getting out this release took way to much time, I started work on this during a Dropbox hack week in Januari 2013 (!). It’s highly unlikely that this rewrite and cleanup would have happend without the uninterrupted week of hacking during that event, it what just not something I could have managed in the couple of (fragmented) hours a week I can spent on PyObjC at the moment.
Ronald
P.S. I’ve also done some more optimisation work with PyObjC 3.0 which has sped up method calls by removing malloc calls in a number of cases. That seems to have been more effective for speed enhancements than the core rewrite itself ;-)
