I had recently read a paper ‘Aspect-Oriented Programming’ by Gregor Kiczales and others, and found there the loop fusion example.
Here is a definition of the loop fusion from the paper
…the loop fusion composes by fusing the loops of those primitive
filters that have the same loop structure and that are direct
neighbors in the data flow graph. Each of these composition rules is
easy to understand when looking at its own appropriate picture. But
the two composition relation- ships cut each other so fundamentally
that each is very difficult to see in the other’s picture.
Later in the paper loop fusion mentioned to be implemented using metaobject protocol at Lisp:
…the loop fusion aspect … can be implemented, with some degree of
efficiency, using the method combination facility in the CLOS
metaobject protocol
What does loop fusion look like using any modern aspect-oriented frameworks in either static or dynamic typed languages? I’m looking for examples of aspect language and advices that could do loop fusion.
Status for mid-January’13
As long as there is no any other answer, I have marked my own answer as an answer. Anyway I’m still looking forward for examples from other languages/frameworks.
ps: loop fusion discusion at “AOP in .NET” book’s forum @manning forum
2
After initial paper released, Gregor Kiczales has started project AspectJ to implement an idea of Aspect-Oriented programming. Recently I found explanation why there are not “loop fusion” available at “AspectJ in Action” by Ramnivas Laddad:
In addition, the following are all join points: an object
construction, a conditional check, a comparison, an exception handler,
and even for, while, and do/while loops. Not all of the join
points in a system are available for your use. … In order to prevent
implementation-dependent or unstable crosscutting, AspectJ
deliberately exposes only a subset of all the possible join points in
the system. AspectJ, for example, does not expose for loops because
you can easily change a for loop to a while loop that functions in the
same manner. If such a change were to be made, all of the advice to
the join point for the for loop would no longer be valid since the
loop would no longer exist.
Another point I found at interview with Gael Fraiteur on PostSharp:
Some different people recommend AOP also for functional requirements
and I… so really invasive in its aspects. Using PostSharp, you are
forced to design aspects so that aspects don’t know their target
code. They don’t know look how variables… so pay attention not to
misuse PostSharp, not to misuse AOP, and respect also engineering
principles, respect separation of concerns, and in the Java eco-system
it is much, much longer concept and they have a lot of practice with
it and users come with feel about that it’s more difficult to program
with aspect because aspect has to know the target code. They have to
know the implementation of target method and so on. Target method has
to know they will be affected. I think this is a very bad design, and
when you plan to use AOP, I recommend to design your base code and
your AOP codes so that they don’t know about each other. It’s really
important not to get completely mad with a small change.
So it’s look like there is not “loop fusion” neither at AspectJ nor PostSharp, and Dynamic Proxy by it’s definition could not provide this kind of functionality also