I asked on [EMAIL PROTECTED] about building
JRVM using itself. This is the answer I got from David P Grove (groved at
us.ibm.com):
Hi,
It should be possible for Jikes RVM to bootstrap itself, but it will
probably take a little bit of work to make it happen. I think the main issues
are going to be (1) setting up the classloaders in such a way that we can tell
apart the VM that is running from the VM that is building and (2) doing the
'right thing' with the subset of VM_Magic operations that need to have different
interpretations at bootstrapping time.
The classpath libraries should be sufficient, so it might be easier to
try with a different VM that uses the classpath libraries first. Bootstrapping
Jikes RVM is a fairly severe stress test for a JVM. It has a very good record
of finding bugs in various product JVMs. I know that Mark Wielaard made a very
quick attempt to bootstrap Jikes RVM on kaffe a month or two ago that caused
kaffe to crash and burn very early in our build process. In theory, one should
be able to use gcj/gij to bootstrap Jikes RVM as well. I suspect the issue
there is going to be that our build scripts assume something like a standard JDK
in terms of command lines, etc. Probably fixable, but might take more work.
If I were going to try to do this, I would probably spend some time
trying to bootstrap on kaffe or gcj/gij first just to try to work out any issues
with the classpath libraries. Then once that was working, try self hosting and
work out the issues I mentioned in the first paragraph.
If backed into a corner and forced to make a wild guess, I would size
this as a 1-2 week effort by someone who was fairly familiar with either Jikes
RVM or with kaffe/gcj to either (1) make it happen or (2) discover some hidden
nasty that I'm not thinking of off the top of my head.
--dave
Cheers //Johan
