From: John L. <je...@pi...> - 2002-10-22 11:47:10
|
Hello Julian =20 > I have in fact been working on integrating the Classpath port into our > CVS tree so that stuff simply works > directly (with the possible goal of ultimately abadoning our current > libraries). Mostly, it is the patches that you > provided and have been incorporated into our tree. In some cases, I > modified Jikes RVm to make it easier to > run with both libraries (i.e. I changed the class loaders to use generic > java.net.URLClassLoader to avoid ifdef'ing > based on the library being used). This is excellent. =20 > The result is that the GNUClasspathSupport directory in our current CVS > head is an implementation of the files > that Classpath expects to be VM-supplied. While this code obviously uses > the interface required by classpath, it > consists mostly of calls into our own libSupport packages. I do not thin= k > that should cause any licensing issues. So I assume you're happy shipping some code that has the Classpath license as part of JikesRVM? > With this code added to our VM, I have been running our suite of Jikes > RVM sanity tests: the full range > of boot images and tests. The only gotcha is that I have made one or two > structural changes to the Classpath > code (I keep a check-out of Classpath'sCVS), involving moving some stuff > out of the core vm classes into the > classes like VMSystem. In addition,in my CVS version of classpath, there > are various bug fixes to the classpath > code that are required for Jikes RVM to pass our full suite of sanity > tests. Could you send me a diff from your Classpath tree? =20 > As for support in the Classpath build process, the code that I have in > GNUClasspathSupport directory allows me > to build and run JikesRVM against the Classpath files without any special > preprocessing of them. We were only planning on preprocessing to handle changes such as the ones you have in your Classpath CVS tree. If that can be avoided, it makes things easier for everyone. John Leuner > Brian Jones = =20 > <cb...@gn...> To: John Leun= er <je...@pi...> =20 > Sent by: cc: jikesrvm-= co...@os..., =20 > jikesrvm-core-admin@www- cla...@gn... = =20 > 124.ibm.com Subject: [Jikesrvm= -core] Re: Updates to GNU Classpath=20 > port of JRVM = =20 > = =20 > 10/18/02 05:55 PM = =20 > Please respond to = =20 > jikesrvm-core = =20 > = =20 > = =20 >=20 >=20 >=20 > John Leuner <je...@pi...> writes: >=20 > > > > I've updated the files necessary to get a working bootimage out of > JRVM > > > > 2.1.1 (latest CVS) and GNU Classpath (latest CVS) at: > > > > > > > > http://people.debian.org/~jewel//jikes/jikesport/2.1.1CVS/port.html > > > > > > John, what prevents us from incorporating the necessary changes into > > > Classpath CVS? I'll assume we're talking a compilation/packaging > > > issue until I get more time to look at this. > > > > There are two parts: > > > > 1. The VM-specific files. Ideally these would go into the JRVM CVS tree= , > > but I fear there are licensing issues. >=20 > Yes, but we can include them if need be. Certainly if jikesrvm want to > keep this working they would I think want or demand to keep them... >=20 > > 2. The modified Classpath files. I believe that to include these we > > would have to do some kind of preprocessing of the .java files. > > > > The JRVM people have written their own preprocessor to handle this task > > (see rvm/src/tools/preprocessor/preprocessModifiedFiles.C). I think we > > could use the same tool. > > > > It understands these directives: > > > > // The following preprocessor directives are recognized in source files= : > > // //-#if <name> > > // //-#elif <name> > > // //-#else <optional-comment> > > // //-#endif <optional-comment> > > > > I suppose it's also possible to use cpp, I don't know if it would choke > > on anything in the .java source. >=20 > cpp works just fine on .java source as far as I can tell. I've seen > it used elsewhere for such things. As the build already looks into > builddir for source (produced from srcdir via configure for instance) > then adding this shouldn't be hard, probably change the order of > compilation to look in 'lib' after going through all the .java source > directories, 'java', 'javax', and 'gnu'. In those add the cpp rules > to Makefile.am. >=20 >=20 > > I don't think there are any other obstacles to incorporating this into > > CVS. > > > > The JRVM code base is still under development and although there is a > > 'library' interface between Classpath and JRVM, there might still be > > maintenance work to be done to keep the two CVS trees in sync. >=20 > There is maintenance work with every JVM... any time we change the JVM > interface we introduce that work ourselves. >=20 > > > > I'm also looking for someone to host an 'autobuilder' and perhaps r= un > > > > nightly regression tests (starting with mauve and maybe others). > > > > > > I'm looking to do the same for Kissme, maybe others if they work with > > > Classpath CVS. I'll let the list know when I've done Kissme and > > > getting the other one in there shouldn't be difficult. > > > > Great. Do you have a machine where you can have this run every night an= d > > produce binaries? >=20 > I can run it on my machine for now... I don't have anything suitable > for distribution of binaries at the moment, so producing them is > somewhat useless except to perform tests and product test results. >=20 > > For JRVM an autobuilder would have to be a beefy machine. Building the > > basebase image (which is what I mostly do with the port) needs a machin= e > > with 512MB ram and will take roughly 5 minutes of CPU time. >=20 > Shouldn't be a problem. I don't use my machine when I'm sleeping. >=20 > > Building the image with the optimising compiler, (which we'll do at a > > later stage), will take a substantial amount of time and lots of RAM. >=20 > Maybe I'll get more RAM if it becomes a problem. >=20 > Brian > -- > Brian Jones <cb...@gn...> > _______________________________________________ > Jikesrvm-core mailing list > Jik...@ww... > http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jikesrvm-core >=20 >=20 >=20 > _______________________________________________ > Jikesrvm-core mailing list > Jik...@ww... > http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jikesrvm-core |