Re: FYI: submitted ecj patch upstream

[Andrew Haley <aph at redhat dot com>]

David Daney writes:
 > Tom Tromey wrote:
 > > Andrew> So now we have somewhere between 1-3 plus driver scripts plus
 > > Andrew> the JVM which can be a pain in the butt to install on say a
 > > Andrew> powerpc Linux box because there is no JVM that exists to begin
 > > Andrew> without gcj.
 > > 
 > > Tom> There's kaffe and jamvm, and probably others.
 > > 
 > > Andrew Haley also floated the idea of checking in all the .class files
 > > and CNI headers, and simply disabling the java->class compilation for
 > > most developers.
 > > 
 > > This would solve one problem but not the one at hand... we would still
 > > need ecj in order to run 'make check', or for the installed gcj to be
 > > useful.

I'm not sure in most cases we'd even need a source for 'make check'.
'make check' is supposed to check what we just built: we don't need to
test ecj, but the JVM we just built and the .class -> .o compiler.

 > > Folks, please comment on this.
 > 
 > I like the idea of checking in the .class files (perhaps in the
 > form a a jar file)

Yes, that's what I fancied doing.

 > and CNI headers.
 > 
 > The idea of having to have a working JVM to build is not at all
 > appealing to me.  Having to wget ecj.jar is one thing, but having
 > to build/port jamvm and classpath to my build platform before I can
 > build GCC does not sound like fun.  Right now I can bootstrap GCC
 > with libgcj on my little MIPS board.  If I need a functioning JVM
 > in order to build I don't think I will be able to bootstrap
 > anymore.  That would be unfortunate.

Right, this is something that we must do.  Move the need for a working
JVM to the runtime, not the bootstrap time.  Once bootstrap is done,
we have a working JVM, and we can use that to run ecj1.

Andrew.