ByteLabs

My diploma thesis has been published as book.

Amongst many other bookstores I found these quite cool:

• Buecher.de has a nice page about it.

• Amazon.com and Amazon.de has it as well.

Last weekend I did a huge rewrite of HBURG's backend. Besides many simplifications and optimizations I finally found some time to replace code that passed around state explicitly with Haskell Monad Transformers. The following is an incomplete list of other fixes and improvements:

• Rewrite rule costs can now include arbitrary expressions
• Context sensitive analysis is fixed now since it was too restrictive and rejecting even correctly defined specifications
• The Gen.Tiling module has been complete removed and replaced by much simpler code
• ...

You can either get the new version from The Hackage or from HBURG's home.

My code generator which is still in the prototyping and experimentation phase does a lot of recursive tree traversals which is quite obvious since it needs to match tree patterns and emit code for abstract syntax trees (AST).

Especially during the code emission phase which does a top down left to right tree traversal it would be nice to recursively evaluate child nodes of an AST in parallel if the evaluation of the child nodes does not yield any side effects. The following paper nicely describes how to transform standard multi-way recursive methods into parallel mutli-way recursive methods in Java.

The transformation described is quite simple and works well for direct recursion. However, due to the attribute type grammar I want to use for my tree pattern matching code generator and the way most tree patterns are augmented with semantic actions, it seems that it would cause a bit of a mess, or at least require a lot of careful thinking how to integrate such parallelism into the tree pattern matching grammar and the code generator which is produced by the grammar specification.

Unfortunately right now this idea can not be assigned a high priority and must be post-poned…

After signing the official papers my diploma thesis is official as of now. Even though it is going to be a lot of work to provide a clean, functional and good implementation, I am really looking forward to it. Especially after having messed around with and read about various compiler back end strategies.

Anyway, I am not dead but unfortunately I have almost no more time to blog. There are also some people I really really really have to send an E-Mail to and this right here is a promise that it will happen really soon!!!!