Lambda the Ultimate

I parse some language (let's say MS Excel formulas), and I got a concrete parse tree. Are there any specific algorithms to convert it to AST? Are any methods or technics around? Or it is better (faster? simplier?) to build AST directly? Any information is highly appreciated.
Thanks.

n/a

I was just doing a search for the Fortress language specification that Sun's working on, and found that they released a new version of the specification (version 0.707) a couple of days ago.

I haven't read over it in much detail, but I've noticed that they made some changes to their interesting "do what I mean" unit manipulation fixups. For example, one of their samples that I questioned before now reads:

w: radian/s := 2 pi radian / 226 million year in s

Of course, parsing this the way they intend breaks a lot of normal mathematical precedence rules, and violates the recommendations of the SI. I'd sure like for them to list the rules that they use to "do what I mean" and perform "context-sensitive disambiguation." In my opinion, such things are equivalent to reading minds. And can "in" also mean "inches"? What's "3 in in in?" Or "1 foot^2 in in in?" Or, equivalently, "144 in in in in in"? Hmmm... It's still my bet that the final language will have to follow normal precedence rules more closely, because these "fixups" tend to paint you into an indefensible, surprising, and ugly corner, and they punish those who understand and write normal mathematical notation perfectly in favor of the dumber people who might make mistakes. I've taken the tactic of "err on the side of being pedantic, and try to teach correct, unambiguous usage" in my Frink language.

As an example, a friend just pointed me to the following scary differences in the Google calculator, which gives results not even having the same dimensions:

1/2 seconds
-1/2 seconds

Try these. Gah! Also see the Frink FAQ about this issue.

n/a

Programming Paradigms of the Andorra Kernel Language

The Andorra Kernel Language (AKL) is introduced. It is shown how AKL provides the programming paradigms of both Prolog and GHC. This is the original goal of the design. However, it has also been possible to provide capabilities beyond that of Prolog and GHC. There are means to structure search, more powerful than plain backtracking. It is possible to encapsulate search in concurrent reactive processes. It is also possible to write a multi-way merger with constant delay. In these respects AKL is quite original. Although AKL is an instance of our previously introduced Kernel Andorra Prolog framework, this exposition contains important extensions, and a considerable amount of unnecessary formal overhead has been stripped away.

That's the AKL that is a predecessor of Oz (and was codeveloped by a coauthor of CTM, Seif Haridi).

While not the latest word in its area (published in 1991), it is well worth reading (and I don't remember seeing links to it on LtU).