[Python-Dev] Ph.D. dissertation ideas?

[Phillip J. Eby]

At 02:36 PM 1/12/2006 -0800, Brett Cannon wrote:
>(I have a third, AOP for anything, but  I don't think AOP is a good match
>for Python and thus not considering it for Python work)

For what it's worth, I've been doing what I'll loosely refer to as 
"research" on implementing a non-traditional form of AOP in Python via the 
combination of extensible generic functions and contextual variables -- two 
concepts that already exist in Python but which can be extended far beyond 
what is in current use.

The combination can be used to implement an "explicit is better than 
implicit" form of AOP, that is usually handled by a "weaving language" in 
classic AOP implementations.  I don't know of any published AOP work that 
focuses on the use of these kinds of tools; the communities that talk about 
the respective technologies appear to be disjoint.

Anyway, extensible generic operations in Python like 'len()', 'copy()' and 
so on are fairly common, but it's rare that people define their own such 
functions, as opposed to using ones in the stdlib.  A typical workaround in 
larger frameworks is to use interfaces and adaptation, but extensible 
generic operations make it easier to separate concerns orthogonally.

So, although you might think that AOP isn't "a good match for Python", the 
truth is that there's actually a fair amount of AOP-like things being done 
with Python.  It's just that classic structuring of AOP into things like 
aspects and weavers and all that crud is overcomplication.  If you view it 
simply in terms of extensible operations, which can also make decisions 
based on context, then you can achieve all the same goals of "classic" AOP 
without using any of its concepts or terminology.

One definition of AOP claims that it allows asserting qualified statements 
over oblivious code.  Python has this basic ability natively -- 
monkeypatching suffices to meet that definition.  The interesting question 
is, how can you do it *without* monkeypatching?