[kaffe] Plans and progress

Guilhem Lavaux guilhem at kaffe.org
Tue Jul 13 01:58:01 PDT 2004


To the request of Jim Huang, I am posting my plans and progresses here:

* stack handling: We have added a new way to detect and set stack 
pointer/size which should eliminate many problems with the locking/gc 
systems. However it has raised many problems that were hidden previously 
on many architectures. According to Timothy Stack we need also to secure 
the exception handling when there is a stack overflow. This is yet to be 

* Boehm GC: I am currently merging and building an interface between 
kaffe and the boehm gc. Fortunately the garbage collector has already a 
modular interface in kaffe so it should be quite easy to adapt. There is 
principally three functions left to implement concerning walking and 
marking objects in memory using boehm-gc/gcj API.

* Solaris: We have managed to make jthreads and pthreads work on solaris 
with kaffe. Apparently there is still missing a modification I've not 
merged in the CVS as Riccardo has reported there is still a failure on 
solaris/pthreads. This is to be investigated.

* HP-UX/parisc (libffi): For the moment the debugging of this 
architecture is suspended while we investigate the possibility to 
incorporate libffi in kaffe. sysdepCallMethod has apparently some 
problems on this architecture so we'll try later with libffi to see if 
that solves the problems.

* engine separation: We all like to be able to separate engines from the 
core VM. There will be several steps to achieve this goal:
    - first we have to remove all ifdefs occuring in the source code to 
be sure the VM is neutral concerning the engine and that all engines use 
the pointers/reference in kaffe in the same way.
    - build a common interface to all engines. This interface will 
surely be OO.
    - finally we will be able to separate the engines from the core VM 
library. It will be possible to specify the engine to use at the run time.


More information about the kaffe mailing list