sablevm-developer

[Sablevm-developer] new debian packages of sablevm and friends

[Grzegorz P. <ga...@de...>]

Hello!

So I divided the packages as we agreed. If you had them installed -
you'll need to remove all *sablevm* packages before installing new
ones. I have not provided easy upgrade path. Remove&install it the way.

I have added /usr/lib/sablevm/bin/java wrapper script - I have only
done little testing. Seems to work. Please check if you can run any
app using -jar parameter. _Should_ work. Please report problems
and ideas of improvement (fixes are welcome also - of course).

I don't know how to run javadoc. The classes are in place but
which one is the starting one? I'd like to fix javadoc and add
it to alternatives. (try to run just /usr/lib/sablevm/bin/javadoc)

As a test - I tried to build ant - it has builded with exception
of javadoc. But I belive it's that we're just one string away from
solving this problem (initial method for javadoc)

I think it's enough for today. It's 1.30AM already, so expect me to
read and reply to your mail in about 8-9 hours from now.

I'd like to upload the packages in few days, but would like to have:
a) tested java wrapper (-jar and other basic stuff working)
b) javadoc working (if it's possible with today's gnu classpath)
c) proven structure of the packages (so that I didn't need to add
    or remove resulting binary packages, change names etc.)

I am done for today.
Best regards

					Grzegorz B. Prokopski

PS: I am Cc:ing ant maintainer in the hope he's still interested
in having ant moved from contrib to main.

Stefan - you may want to take a look at all of our conversation here:
http://www.geocrawler.com/lists/3/SourceForge/4435/0/

A word about ant compilation - it _compiles_ w/o problems because
classlib contains _a lot_ of stub classes implementations. It causes
that ant detects that those classes are available and thus compiles
a lot of features. There's nothing really bad with it I belive, as
this would allow us to compile all(?) the features with free
compiler/jvm/classlib thus avoiding split of the package (hmm.. are we
really sure it is not really needed to move unusable but optional parts
to contrib?)
However there's some problem with "reflection" code (the same as with
a bit older kaffe) when running ant - Etienne offerred himself to fix
them for us later.

What do you think about all this stuff?

Re: [Sablevm-developer] new debian packages of sablevm and friends

[Etienne M. G. <eti...@uq...>]

On Mon, Aug 12, 2002 at 01:47:49AM +0200, Grzegorz Prokopski wrote:
> Hello!
> 
> So I divided the packages as we agreed. If you had them installed -
> you'll need to remove all *sablevm* packages before installing new
> ones. I have not provided easy upgrade path. Remove&install it the way.
> 
> I have added /usr/lib/sablevm/bin/java wrapper script - I have only
> done little testing. Seems to work. Please check if you can run any
> app using -jar parameter. _Should_ work. Please report problems
> and ideas of improvement (fixes are welcome also - of course).

I tested it very quickly.

The java wrapper didn't really work. I had to run update-alternatives
by hand, and it linked to /usr/bin/sablevm instead of
/usr/lib/sablevm/bin/java.

Then I tried to run HelloWorld.jar but it failed, with something like
classnotfound.  (I am too tired to remember at this point).

> I don't know how to run javadoc. The classes are in place but
> which one is the starting one? I'd like to fix javadoc and add
> it to alternatives. (try to run just /usr/lib/sablevm/bin/javadoc)
> 
> As a test - I tried to build ant - it has builded with exception
> of javadoc. But I belive it's that we're just one string away from
> solving this problem (initial method for javadoc)
>

On the "running ant" side, I got Method.invoke() working.  Now it asks
for Class.getMethods() and Method.getArguments().  These are not
difficult to implement. per se (as SableVM already has all this
information in its internal data structures), But it is painfully long
to write, as you have to take care of array, class, and interface
types, as well as static, virtual and interface methods.  So this will
get done *after* my thesis, have I decided.  (Unless somebody else
wants to do it).

> I think it's enough for today. It's 1.30AM already, so expect me to
> read and reply to your mail in about 8-9 hours from now.


It's 1:22AM here.  I'm going to bed too.

Thanks for all your work.

Etienne

-- 
Etienne M. Gagnon                    http://www.info.uqam.ca/~egagnon/
SableVM:                                       http://www.sablevm.org/
SableCC:                                       http://www.sablecc.org/

Re: [Sablevm-developer] new debian packages of sablevm and friends

[Grzegorz P. <ga...@de...>]

W li=B6cie z pon, 12-08-2002, godz. 07:23, Etienne M. Gagnon pisze:=20
> On Mon, Aug 12, 2002 at 01:47:49AM +0200, Grzegorz Prokopski wrote:
> > I have added /usr/lib/sablevm/bin/java wrapper script - I have only
> The java wrapper didn't really work. I had to run update-alternatives
> by hand, and it linked to /usr/bin/sablevm instead of
> /usr/lib/sablevm/bin/java.
sablevm 1.0.1-3 is uploaded to debian.sente.pl - I fixed it there
I also added manual page java.sablevm.1.gz which is linke thru
alternatives to java.1.gz when you have sablevm 'java' wrapper choosen.

> Then I tried to run HelloWorld.jar but it failed, with something like
> classnotfound.  (I am too tired to remember at this point).
Not sure if that's really script's fault.
with jikes I compiled HelloWorld.java to .class and when I try to
run it using sablevm, I am getting:

greg@greg:~$ sablevm -Y ./HelloWorld.class=20
java.lang.ClassNotFoundException: ./HelloWorld.class
	at gnu.java.lang.SystemClassLoader.findClass(SystemClassLoader.java:73)
	at java.lang.ClassLoader.loadClass(ClassLoader.java:308)
	at java.lang.ClassLoader.loadClass(ClassLoader.java:259)
	at java.lang.VirtualMachine.main(VirtualMachine.java:79)

I think I am doing sth. wrong. Could you please send me your
HelloWorld.jar file (to private addr)?
How should I run it using sablevm (not java wrapper)?
If it runs with sablevm from debs and you can tell me how to run it -
then fixing the wrapper should be trivial.

> > I don't know how to run javadoc. The classes are in place but
> > which one is the starting one? I'd like to fix javadoc and add
> > it to alternatives. (try to run just /usr/lib/sablevm/bin/javadoc)
OK, I found cp-tools project http://savannah.gnu.org/projects/cp-tools/
I'll probably package some of these tools too, as they will give us
ability to have wrappers for:
javadoc, javah, javap, serialver

(com.sun.javadoc _really_ should be moved there, to gjdoc source tree,
as it seems to be kinda unusable in classpath alone)

However - those files (in cp-tools) were checked into cvs only once,
9 months ago and weren't touched since then. oh - if they work - then
no problem...

> On the "running ant" side, I got Method.invoke() working.  Now it asks
> for Class.getMethods() and Method.getArguments().  These are not
> difficult to implement. per se (as SableVM already has all this
> information in its internal data structures), But it is painfully long
> to write, as you have to take care of array, class, and interface
> types, as well as static, virtual and interface methods.  So this will
> get done *after* my thesis, have I decided.  (Unless somebody else
> wants to do it).
When do you plan to finish them and get the time to fix this?
(no pressure, just would like to know)

> Thanks for all your work.
And vice versa.

Best regards

						Grzegorz B. Prokopski

PS: Looking at all gnu-java related CVS trees - I am getting really worried=
,
as it looks like there's no real development being made. Times 'since last
update' are weeks at best - usually months and sometimes years. A lot of
features has only it's stub files put into CVS with no further action.
And where's awt/swing development? I hardly see any.

Can we really count on that GNU Classpath will be THE Classpath for
free software java projects? What do you think?

Re: [Sablevm-developer] new debian packages of sablevm and friends

[Etienne M. G. <eti...@uq...>]

On Mon, Aug 12, 2002 at 01:48:56PM +0200, Grzegorz Prokopski wrote:
> sablevm 1.0.1-3 is uploaded to debian.sente.pl - I fixed it there
> I also added manual page java.sablevm.1.gz which is linke thru
> alternatives to java.1.gz when you have sablevm 'java' wrapper choosen.

OK, I'll try it.

> 
> > Then I tried to run HelloWorld.jar but it failed, with something like
> > classnotfound.  (I am too tired to remember at this point).
> Not sure if that's really script's fault.
> with jikes I compiled HelloWorld.java to .class and when I try to
> run it using sablevm, I am getting:
> 
> greg@greg:~$ sablevm -Y ./HelloWorld.class 
> java.lang.ClassNotFoundException: ./HelloWorld.class
> 	at gnu.java.lang.SystemClassLoader.findClass(SystemClassLoader.java:73)
> 	at java.lang.ClassLoader.loadClass(ClassLoader.java:308)
> 	at java.lang.ClassLoader.loadClass(ClassLoader.java:259)
> 	at java.lang.VirtualMachine.main(VirtualMachine.java:79)

Of course...  Try "sablevm -Y HelloWorld" (you have to give the class
name, not the file name).

> 
> I think I am doing sth. wrong. Could you please send me your
> HelloWorld.jar file (to private addr)?
> How should I run it using sablevm (not java wrapper)?
> If it runs with sablevm from debs and you can tell me how to run it -
> then fixing the wrapper should be trivial.

I'll send you the .jar, but remember that you have to type "sablevm -c
HelloWorld.jar HelloWorld".  I've seen that you try to extract the
Main class name in your wrapper; nice.

> > On the "running ant" side, I got Method.invoke() working.  Now it asks
> > for Class.getMethods() and Method.getArguments().  These are not
> > difficult to implement. per se (as SableVM already has all this
> > information in its internal data structures), But it is painfully long
> > to write, as you have to take care of array, class, and interface
> > types, as well as static, virtual and interface methods.  So this will
> > get done *after* my thesis, have I decided.  (Unless somebody else
> > wants to do it).
> When do you plan to finish them and get the time to fix this?
> (no pressure, just would like to know)

Would 3 weeks from now be acceptable to you?

> PS: Looking at all gnu-java related CVS trees - I am getting really worried,
> as it looks like there's no real development being made. Times 'since last
> update' are weeks at best - usually months and sometimes years. A lot of
> features has only it's stub files put into CVS with no further action.
> And where's awt/swing development? I hardly see any.
> 
> Can we really count on that GNU Classpath will be THE Classpath for
> free software java projects? What do you think?

I think that one of Classpath's current problems was that it lacked an
appropriate VM for doing development.  There's ORP, but this is a JIT,
and it has some problems, and locating problems in a JIT is much more
difficult.  In fact, you can easily trace SableVM's execution with a
debugger, add debugging traces on every exception thrown, (etc.) very
easily.  Also, SableVM has a solid implementation of JNI (with some
minor holes, easily fillable), which should help.

There is actually some development on the AWT (Swing?) side.  It could
be faster, but it's there.  You should look at the classpath cvs
commits ML.  You'll see things like
http://mail.gnu.org/pipermail/commit-classpath/2002-July/000916.html
which is pretty recent.

GNU Classpath is a little like Debian; it is not backed off by a
companyy, so you have to give it time.  Eventually, robust things come
out of it.

Etienne

-- 
Etienne M. Gagnon                    http://www.info.uqam.ca/~egagnon/
SableVM:                                       http://www.sablevm.org/
SableCC:                                       http://www.sablecc.org/

Re: [Sablevm-developer] new debian packages of sablevm and friends

[John L. <je...@pi...>]

Grzegorz

> > PS: Looking at all gnu-java related CVS trees - I am getting really worried,
> > as it looks like there's no real development being made. Times 'since last
> > update' are weeks at best - usually months and sometimes years. A lot of
> > features has only it's stub files put into CVS with no further action.
> > And where's awt/swing development? I hardly see any.

Are you reading the Changelogs?
 
> > Can we really count on that GNU Classpath will be THE Classpath for
> > free software java projects? What do you think?

Well, do you see any competition?

> I think that one of Classpath's current problems was that it lacked an
> appropriate VM for doing development.  There's ORP, but this is a JIT,
> and it has some problems, and locating problems in a JIT is much more
> difficult.  In fact, you can easily trace SableVM's execution with a
> debugger, add debugging traces on every exception thrown, (etc.) very
> easily.  Also, SableVM has a solid implementation of JNI (with some
> minor holes, easily fillable), which should help.

Yeah, it's a cyclical process where the VMs gets better, Classpath gets
better, the VMs get better ...

The number of fairly stable VMs that work with classpath are increasing
too. I think the whole system is close to achieving critical mass if
enough people start using and testing the software. (And this is where
packages like debian packages are critical)
 
> There is actually some development on the AWT (Swing?) side.  It could
> be faster, but it's there.  You should look at the classpath cvs
> commits ML.  You'll see things like
> http://mail.gnu.org/pipermail/commit-classpath/2002-July/000916.html
> which is pretty recent.

AWT has traditionally not been the focus of classpath and free java
projects. But it's coming along,  I don't think there is too much work
left before AWT becomes useful.

> GNU Classpath is a little like Debian; it is not backed off by a
> companyy, so you have to give it time.  Eventually, robust things come
> out of it.

Indeed.

John Leuner