Re: Getting gcj to compile what it finds when scanning

[Tom Tromey <tromey at redhat dot com>]

>>>>> "Andrew" == Andrew Cowie <andrew@operationaldynamics.com> writes:

Andrew> [did I hear that kaffe is [considering] using GCJ as a runtime inline
Andrew> compiler?]

It has definitely been discussed.  Last year at FOSDEM we talked about
the possibility of coming up with some kind of "free java jit"
specification that would let us mix-and-match execution engines among
the free VMs.  There are still some technical hurdles to get over.

Andrew> I see, again and again, people trying to use `gcj -C` as an
Andrew> inline replacement for javac in in whatever context (Makefile,
Andrew> ANT build.xml, whatever) that just ends up making a mess
Andrew> because its not recursive in the way that javac is.

Yeah, this ought to be fixed.  gcjx fixes it :-)

Andrew> I go one further, and wish that `gcj -c` was a drop in
Andrew> replacement, more or less (ok, so we need a custom linking
Andrew> step after. That's the cool part anyway).

For already-existing java applications, the approach we've been taking
is to just compile the app (or library) to bytecode as always, and
then just compile the whole .jar to a .so in one go.  This is pretty
simple, is nearly as good as compiling from source (and at the moment
BC doesn't work from source anyway), and avoids complicated hacking on
the gcc driver :-)

Tom