GIGO: words unreadable aloud

First Class Attribute Grammars

Mostly finished reading the paper I mentioned the other day, about Aspect-Oriented Compilers. One of its co-authors, Ooge de Moor, is studying:

How do you decompose a compiler into logical components? To avoid having to specify passes, you can use lazy evaluation. But even then, you need to specify for each production in the context-free grammar what the semantics are, and all aspects (lexical level, type, etcetera) are specified in one place. Consequently the compiler writer cannot separate his concerns, and it is hard to re-use bits of compiler, or extend an existing one. I think that novel type systems for object-oriented programming are the answer.

He has links to (gzip'd postscript versions of) the paper about Aspect-Oriented Compilers and one that focuses on making Attribute Grammars be first-class entities.

My take on the approach is that yes, sometimes you want to look at a given aspect of the compiler, across all of the language's constructs (grammar productions); but sometimes you want to look at all of the aspects that pertain to a given construct. This leads me to think that there are big benefits to be had if you could cross-slice a program in multiple ways while developing it. Something similar to what the tangle/weave literate programs do, but with multiple ways of viewing and organizing the "real" code.

I guess I oughta read the paper first.
10:31:27 AM comment/