OO.org2

[Tom Tromey]

>>>>> "Dan" == Dan Williams <dcbw at redhat.com> writes:

Dan> That would be 'gij' at this time.

Ok.

Dan> /usr/bin/gij -Dgnu.gcj.runtime.VMClassLoader.library_control=never -
[ .. evil command line .. ]
Dan> com.sun.star.help.HelpLinker @/tmp/mkfRPrRw

Dan> This particular step of the process is done quite a few times,
Dan> and even just this one invocation of gij took 5 minutes in my
Dan> observation last night on bugs.build (900Mhz Xeon).  Caolan has
Dan> more details, he was going to look at compiling a native binary
Dan> of HelpLinker soon.

The simplest way to do this with gcj 4 is to set up a .db that maps
classes to .so files.  Then compile each .jar to a .so (with
-findirect-dispatch) and add each to the database.  Finally, tell gij
about this database when running.

This requires no application changes; it is how we have compiled
eclipse, jonas, derby, etc.

Here's an overview on how to set this up:

http://gcc.gnu.org/wiki/How%20to%20BC%20compile%20with%20GCJ


We should probably move this discussion to
fedora-devel-java-list...  Internally we've talked a bit about things
like having a default systemwide .db (so no special 'gij' invocation
needed) and having boilerplate scripts (or rpm macros) to compile
.jars and load them into a .db.

Tom