This post is intended for users of old versions of the decorator module who want to know what's new and the reasons for the change. Version 3 is a major release and it breaks compatibility with the past in a minor way, but I expect 99.9% of my users to upgrade to the new version without seeing any difference. You can download the tarball here.

Here is a list of the most relevant changes/improvements.

• I have completed the move to PyPI. For a long time I have wanted to move the package from my site (which is hosted on the Pittsburgh University servers and completely out of my control) to PyPI. The first version to be hosted on PyPI was version 2.3.2, released two weeks ago. The impressive thing - to me, at least - is that I had 1008 downloads in thirteen days: incredible! I have no idea of how many downloads I had for the previous versions, so I cannot compare, but from now on I can have an idea of the popularity of the module. That's good. The move to PyPI was not complete however, since, the documentation for the module was still hosted on my site. With version 3.0, instead, everything is hosted on PyPI.
• The documentation now is extracted from a Python file containing lots of doctests, following an approach that I have called geek publishing and that I have described here: http://stacktrace.it/2008/01/geek-publishing/ (I intend to translate that article and to republish it here soon or later). That means that all the examples provided are doctested by using the standard Python module: I do not need to supplement a custom doctester script as I did in the past. The documentation is extracted by using a custom script, by I do not need to distribute it, I have just included in the tarball the generated .html and .pdf files.
• I have substantially modified the documentation. Various sections have been removed. In particular the examples redirecting_stdout and locked are no more there, since the functionality is better implemented by using the with statement instead of a decorator. At the time of the first version of the module, in early 2005, those examples were good examples, but Python has changed in the meanwhile. You can still find them in the old documentation of the module, which will stay on my personal site for the foreseeable future: http://www.phyast.pitt.edu/~micheles/python/documentation.html. The example of a permission system based on decorators has been removed because the documentation is already quite long (the PDF version takes 18 pages).
• Various examples have been substantially improved. For instance, the new implementation of the memoize decorator if much nicer than the old one, and it is easier to compare the approach suggested by the decorator module with the approach suggested by the standard library, i.e. functool.update_wrapper trick. The delayed decorator example has been removed: instead a much more interesting async decorator has been added, showing how to run a blocking function in a separated thread or process.
• The utility functions new_wrapper and get_info have been deprecated, since they were little used. Their job can be better performed by the new class FunctionMaker, which is the building block over which decorator is implemented. That allows you to define your own custom version of decorator, if you really want.
• There was a long standing issue with the decorator module, i.e. the problem of extracting the source code from a decorated function. Version 3.0 gives a workaround. It adds an attribute .undecorated to the decorated function, a reference to the original function, so that you can get the wanted source code with the trick inspect.getsource(func.undecorated). This is actually possible even in version 2.3, but it was (intentionally) left undocumented, since I hoped in a better solution such as a patch to the inspect module. However, the fix did not make it through Python 2.6 and 3.0, so I have decided to document the workaround for the time being.
• The decorator function has been extended and now it may take one or two arguments. decorator(caller) still returns a decorator, as before, but now you can also use the syntax decorator(caller, func) which returns directly the decorated function; decorator(caller, func) is akin to decorator(caller)(func) but more efficient. The new syntax made possible to simplify many examples and I could remove the explicit support for decorator factories which I added in haste in version 2.3 and I regretted pretty soon.

There are also a few considerations I would like to make.

From the start the decorator module was developed with the attitude of teach a man to fish: instead of providing a large API, I have provided a significant collections of examples and recipes. The idea is that you should be able to write your own decorators by yourselves. Version 3 of the module is going even more in that direction.

I have refactored the internals so that now you can not only write you decorators on your own, but you can also write your own decorator facility - the equivalent of decorator - by means of the FunctionMaker class. At the same time the rewriting makes the module more of a library and less of a framework. For instance, in past versions you were forced to write your decorators in terms of caller functions with the signature caller(f, *args, **kw); now you can write your own decorator framework and use the conventions you like. In the documentation I give the example of decorator_apply, which is able to convert third party decorators into signature preserving decorators without rewriting them.

I did not expect the decorator module to leave so long (it is nearly four years old already). In my original intentions, the module was intended to be provisional, a workaround that should have been dropped once better support for decorators entered in the standard library. Unfortunately that never happened. It is true that Python 2.5 added some support for decorators in the functools module, but that support is insufficient in my opinion. Also, I had great hopes for the Function Signature Object (PEP 362) but after more than two years nothing happened. I still hope it will become possible to change the signature of functions in future versions of Python: when that will happen, the decorator module will become obsolete and I will have less code to maintain.

Finally, I have a couple of questions for you, PyPI experts. Is there a simple way to remove the annoying excessive vertical space in the PyPI style-sheet? Look at http://pypi.python.org/pypi/decorator to see what I am referring to. I have uploaded the documentation simply by inserting raw XHTML into the long_description field of the setup.py script and running python setup.py register. It worked but it is kind of a hack. I see that there is the possibility to upload the documentation as a zip file, labeled as experimental feature. I tried it by hand and it works, but I would like to know if there is a way to perform the upload automatically, with some option in the setup.py script.