Re: Running Eclipse on Jikes RVM and Classpath (Prize money inside!)

[Mark Wielaard]

Hi,

On Mon, 2003-04-07 at 18:55, Michael Hind wrote:
> As Julian's recent post mentioned, we just put a release of Jikes RVM
> out that relies totally on Classpath and runs a significant subset of
> Eclipse.   A few months back, someone mentioned that getting Eclipse
> started on an open source JVM with Classpath would be a good benchmark
> and suggested that getting it to run in under a minute would be a nice
> goal.
> 
> Using a FastAdaptiveSemiSpace image with a 200 Mb heap, we got to the
> sign on screen in approximately 12 secs on a Linux/IA32 box.

That is really amazing. Very good work!

I don't know if you can claim the prize yet. Although I really hope you
can! Since I never publicly announced the rules. Here they are:

The challenge is to start up a fresh Eclipse (*) and being able to quit
the application through the File menu just after the complete welcome
text is shown under one minute. This must be possible on a some standard
PC (say 1Ghz Athlon/Pentium with 512MB) with only Free Software
installed. It should also be possible to run the rest of the Eclipse IDE
reasonably well (but it doesn't have to be perfect yet).

(*) Note that the first time you startup eclipse it first does some
installion and creates your workspace before really starting up. This
does take some extra time, compared to starting it up the second time
which is noticably faster. (Fresh startup gives two splash screens.)

Now that Eclipse 2.1 is officially out we could say that this should be
done with the official 2.1 Linux (x86/GTK 2) release. And that starting
should be as simple as unzipping eclipse-SDK-2.1-linux-gtk.zip and
doing: cd eclipse; ./eclipse -vm yourvm.

Note that there is no requirement to use GNU Classpath as core class
library, any free implementation of the standard classes is acceptable
so Kaffe or Wonka can also try to claim the prize. Basically if the
entry makes it possible to include all software necessary to run Eclipse
and Eclipse itself into the main archive of Debian GNU/Linux (which only
contains Free Software) it qualifies.

The Prize! I will make the winner an Associate Member of FSF for one
year ($120,-) <http://member.fsf.org/benefits.html> or (if requested)
buy them something (books or gear) from GNU Press (for the same amount
of money) <http://www.gnupress.org/>.

I am not sure what to do if the entry is clearly a group effort. Maybe
we can convince the FSF to give a honorably group membership. Or just
buy all group members (max 6) a GNU Cap
<http://www.gnu.org/gear/baseball-cap.html>.

Since I will pay for the prize I will be the judge.
We could have a little vote on the Classpath mailinglist,
but I can veto and my decision will be final ;-)
Just email me or the classpath mailinglist (a pointer to) the
instructions to build your VM and run Eclipse on a free platform.
(Please allow a week to recreate the environment and check your entry.)

This is the status of some of the VMs as far as I am aware:

- IKVM.NET: Can run eclipse and starts just under one minute. But uses
proprietary MS .NET implementation to do it. We need to get it running
on Mono (or dotGNU) on a free platform (GNU/Linux, FreeBSD, ...)

- gcj (3.3 CVS): Able to run eclipse when the verifier is disabled. But
it still takes several minutes. Class loading and exception handling
seem to be the bottlenecks. It might be possible to precompile some of
the Eclipse libraries to native code which would speedup the program
alot. (I am willing to not count the disabling of the byte code verifier
against gij/gcj since non of the other free VMs seem to have one.)

- Jikes RVM (2.1.1): Starts Eclipse amazingly fast. Only roadblock seems
to be that it is not yet selfhosting. (It would be OK to use a non-free,
proprietary VM to start the bootstrap, but after this it should be able
to bootstrap itself.)

I am really excited that we are so close to this important milestone. I
certainly had not expected to get this far (and fast!) just 4 months
ago.

Cheers,

Mark

P.S. I know that it is unfair to give a prize to the someone that "just"
gives that last missing link while alot of people have spend years
creating all the supporting code. But I really don't have the means to
reward all the people involved appropriatly. I just thought it would be
a fun thing (for everybody) to make it into a challenge with a small
(but real!) prize :)