Wes Biggs <[EMAIL PROTECTED]> writes:
> Is it not possible to have makefiles for both javac and guavac? Aaron,
> what in your code won't work with javac?
Aaron's problem is that javac is trying to use his java.io
implementation since javac uses $CLASSPATH to both look for files to
compile *and* to determine which classes it should use to run.
> >From the brief notes on the guavac page it looks like Cygnus might be
> working toward a gcc-based java compiler. Does anyone have any more
> details on this?
There is currently a front-end that compiles java byte code to native
code. There is not (as far as I know) a java-source front-end
(although it is mentioned that work is being done on it). Also,
runtime support has to be added to basically do the things the VM
normally does.
> On second thought: Should classpath be providing a written-in-java javac
> implementation as well (OK, not right now, but at some point down the
> line)? That would make sense if the Japhar/Classpath master plan is to
> fully replace dependencies on Javasoft products. (IMO guavac is good, but
> it does suffer from platform constraints more than a pure-java
> implementation like javac does.)
Glad you asked; I am working on a java compiler written in java.
About half the work is done so far (it now generates ASTs), and
progress is coming along nicely (its actually for a course, so I
*have* to work on it). I hope to have a preliminary release in about
a months time. A plan to solve Aaron's problem by separating the
double meaning of the -classpath switch into two switches.
> Since guavac is already C++, would there be any benefit in trying to
> convert it to java?
I've submitted a few changes to guavac to support 1.1 features, but in
the end it was just not worth working in C++ (for me).
--
Geoff Berry
