Instead of putting the new decaf stuff up on SourceForge -- to
which I don't yet have access -- I've branched the CVS on jos.org. To
access it, either 'cvs co -r new_decaf JJOS' (with the appropriate -d
information as leache from CVS/Root) or 'cvs update -r new_decaf'; once
you've use the new_decaf tag, your revisions will be to the new_decaf
branch of the tree.
This is not the complete port of new_decaf; it's a distribution
mechanism for this mailing list, so that you can take a look at what I've
been doing for the past few months, off and on. In particular, a design
review would be very much appreciated before I get started in on 'porting'
interp.cc & jvmbuiltins.cc to the new design. Generally, look for places
where the design isn't clean because of 'home-stretch' hackery --
functions in the wrong class, or failure to make an abstraction
(e.g. failure recognize duplication of functionality, or failure to
properly enscapulate, etc.). Thanks in advance.
The major non-interpreter/JVM-builtin thing that's missing is the
'isInstanceOf()' function for JavaClassInstance. After completing the
interpreter, I'll begin work on the built-ins, separating JVM/JOS lib code
(h/w access) from the classlib code; it should be relatively soon after
completing this that I cleanly integrate it with a class
library (perhaps Kore), add multiple processes, and implement BCNI. At
some point I would like to work on per-JVM 'intern's to reduce garbage.
The current outstanding problem is the use of dynamic_cast<>(),
which is necessary in a few locations because of the way I designed the
class inheritance tree. My understanding is that the implementation of
dynamic_cast, once you understand it, is relatively straightforward but
compiler-specific, which is very unfortunate. If anyone knows more about
this, let me know. (This is only a problem for the i386 build; if there's
someway to get dynamic_cast<>() from the compiler/statically linked from a
library, it would very nice...)
For the numerically-minded, there are 5,845 lines of
non-interpreter code in common/decaf, compared to 11,032 in the old
common/decaf non-interpreter code. About 800 lines of this savings will
disappear when I put the license headers back in. (Some of it is also
old debug or #ifdef OBSOLETE_OPTION code.) (This count may also be old by
the time I actually send this message.)
-_Quinn
PS: I just checked and the main branch is OK, so that give me hope that I
did the branch correctly; let me know about whatever problems might show
up when you try a compile.
_______________________________________________
Kernel maillist - [EMAIL PROTECTED]
http://jos.org/mailman/listinfo/kernel
