Subject: Re: compiler investigation
noth@cs.washington.edu
Date: Thu Apr 19 2001 - 13:36:20 PDT
>From: Jonathan Aldrich <jonal@cs.washington.edu>
>Subject: compiler investigation
>
>I did some investigation of Java compilers that could be extended to
>support Java language extensions. The results (including all the
>semi-realistic choices I could find on the web) are summarized in a table
>at:
>
>http://www.cs.washington.edu/homes/jonal/archjava/compilers.html
>
>If we limit ourselves to compilers with the following characteristics:
> * open source
I'd clarify "open source" to mean "GPL'd" to differentiate it from
"open source" licensing terms that permit some other entity to
retain rights/control over the source (e.g., Apple's "Open Source"
license).
> * written in Java or perhaps GJ
> * target pure Java, not some tricky extension
>
>then we have only two choices: Kopi or Barat. Kopi is the larger project,
>thus hopefully better quality, and our connection with Clifton gives us
>some background for using it. However, we do have a certain amount of
>negative feedback saying that Kopi is not all one might hope for.
>
>We could also give up open source and choose javac or deal with extra
>typechecking hassle and choose pizza/mjc. I don't really want to
>implement anything in C code, though.
For my purposes, at least, I'd very much rather stay away from
anything other than Java as the implementation language.
>Another alternative is to "roll our own" extensible Java compiler. The
>advantage is that we could do the minimum work necessary: parse Java,
>build an AST, typecheck it minimally, and output it back to Java source
>for compilation by javac. Also, Michael and I could share the base work,
>make it available to the community, and then both extend the result with
>our favorite language extensions.
>
>Right now I think I'd lean towards modifying Kopi, but also checking out
>Barat and considering rolling my own if I am getting stuck down a rathole.
>What is your preference, Michael? Do others in the group have an interest
>in this issue? Know any undergrads who might want to write a Java
>compiler?
I've had reasonably good success thus far using JavaCC + JTB + some
simple handwritten tools to handle the first stages a compiler (parsing,
AST generation, rudimentary code generation for a pretty-printer). I'm
currently investigating Kopi (and learning ANTLR first, bleah) to get a
sense of how difficult it may be to do what I'd like with it. Kopi does
seem to suffer from one of the potential problems of using an OO style to
build a compiler, in that aspects such as typechecking and optimizations
are distributed across every class instead of being in one place.
JavaCC/JTB allows one to use the visitor style, so typechecking and some
forms of optimizations can easily be written as visitors with all
associated code in one place, which has some advantages in terms of
complexity, understantability, and extendability.
If we were to collaborate and roll our own compiler, it'd probably
be a good idea for all interested parties to meet soon and discuss
requirements and decide how to proceed.
In any case, it may well be possible to use Kopi's classfile
tools to handle the processing of binary .class files (e.g.,
extracting class and method signatures for typechecking), even if
we decide to roll our own compiler.
Michael Noth
noth@cs.washington.edu
_______________________________________________
Cecil mailing list
Cecil@cs.washington.edu
http://majordomo.cs.washington.edu/mailman/listinfo/cecil
This archive was generated by hypermail 2b25 : Thu Apr 19 2001 - 13:37:04 PDT