LtU Forum

lua book available online.

Fantazies on "final"

Yesterday I was playing with unification algorithms and decided to implement one for general Java "terms" (object instances). This naive attempt stopped after 10 minutes when I understood that to make logical variables transparent to user I have to be able to create proxies for objects of any type. Even if I outlaw unboxed types, unfortunately this is not possible also for final classes... Which are expected to form the bulk of typical business objects (they are strings, numbers, ...).

This failure caused me to exercise my English in not very creative ways - I understand the original considerations of the JDK standard libraries, but history aside - is it reasonable to even wish to create a dynamic (or for that matter even static) proxy for an instance of a final class? Does not it contradict the semantic of final (that is - not inheritable)? Should not ability to intercept be separated from implementation reuse?

PS: lack of proxies for any class is not the only problem for this micro-project - mutability by default instead of having explicit cells kills any chances for transparency of such manipulations in Java - one has to proxy all objects, not just cells and transients. Ironically, both problems are related to word "final" - the first for classes and the second for variables.

New CLR Language: Boo

This project is very interesting: a CLR-based language with Python-esque syntax, optional type declarations, open compilation pipeline, and some interesting ideas about generators and class attributes. There are some unique ideas about running a dynamic language atop the CLR, as well, like syntax to declare "duck" typing for individual objects, which lets you use them as dynamic proxies for COM objects, web services, etc.

It's definitely a bit green, but if even a significant fraction of the features discussed in the "manifesto" are there, it looks like an impressive effort, regardless.

Cω Preview Available

Cω (pronounced C-Omega) is an extention to C# in two areas: asynchronous concurrency (formerly known as Polyphonic C#) and XML data types (Xen). A preview compiler is now downloadable from MSR:

AOP Without The Buzzwords

I found this attempt at a buzzword free explanation of aspect oriented programming from Adrian Colyer's blog in the June archives of the ll-discuss maillist just posted. I think I finally have some idea what the aspect oriented hoohah is about, at least I was able to read the Introduction to AspectJ and make some sense of it for the first time. Maybe I'm just dense but until now I just didn't get it. This comment from Pascal Constanza in the ensuing thread shed a little more light on it at least for me coming from a functional perspective.

ll-discuss's new home

For all those (like me) who didn't get the memo, I just found out that the seemingly defunct ll1-discuss mailing list has in fact found a new home at:

Full abstraction is not very abstract (via comp.lang.scheme)

There's been a raging battle on comp.lang.scheme for two months and 300+ posts about interpreters, denotational semantics and compositionality. I've skipped most of it, but this post from Will Clinger was instructive to me, particularly the part on full abstraction:

In other words, one of the most un-abstract semantics that could be imagined is fully abstract. Examples like this are part of the reason full abstraction is no longer regarded as a terribly useful concept.

It turns out that full abstraction is more easily achieved by making the operational semantics *less* abstract than by making the denotational semantics *more* abstract. Hence a fully abstract pair of semantics is likely to be *less* abstract than a pair of semantics that are *not* fully abstract.

Why type systems are interesting

In the Explaining Monads thread, Luke Gorrie wrote:
If someone could next show what makes type systems so interesting in terms that I'll understand then I'll be well on my way into the modern functional programming world.. :-)

When you write code, even in the most dynamic language, you always know something about the types of the values you're operating on. You couldn't write useful code otherwise - if you doubt that, try coming up with some examples of operations you can perform on a value without knowing something about the type of that value.

Type systems provide a way to encode, manipulate, communicate and check this type knowledge. In an earlier thread, you mentioned that you write the types of functions in the comments. That's an informal type system! Why do you do use a type system? Because type systems are useful, important, and as a result, interesting.

The question didn't mention the word "static", but I'll raise it anyway. The type knowledge I'm talking about is necessarily mostly static - you know it when you're writing the program, and it's knowable when you're reading the program. Since we can't write useful code without it, it's also very important knowledge. That raises the question, if this important static knowledge is present in our programs, shouldn't we make it explicit somehow?

Further, if possible, since we exploit and depend on this knowledge on almost every line of code we write, shouldn't we check it for validity as early as we can, even while we're writing the programs (e.g. within an IDE), perhaps before it's possible to run unit tests on them? Given this ability, why wouldn't you take advantage of it?

Of course, the answer to that is that encoding a program so that its every term can have its type statically checked, introduces some constraints. But as I've argued before, the alternative of avoiding all static encoding of type information (other than in comments) is mostly a reflection of weaknesses in the state of the art.

Summary of techniques / approaches / models / languages for parallel computation


Can anyone point me to a survey paper for approaches to parallel computation?

Googling, I've found Models and Languages for Parallel Computation (Skillicorn and Talia) which is a pretty good example of what I'm looking for (from an initial skim through). However, that's from 1996 - I'm wondering if there's anything newer (or just complementary).


Looking for the source of a quote

I know I should be able to find it on Google, but I can't: Who was it who said something like, "Programming languages should be designed primarily for humans to read and only secondarily for machines to execute."?

XML feed