sablevm-developer — SableVM developer mailing list

[Sablevm-developer] Re: Running jetty

[Laurent M. <la...@ao...>]

Thanks for your help.

-- 
Laurent Martelli
la...@ao...                                Java Aspect Components
http://www.aopsys.com/                          http://jac.objectweb.org

Re: [Sablevm-developer] Re: Running jetty

[Etienne G. <gag...@uq...>]

OK.  I found the bug.

SableVM does depend on classes being prepared before you can use them
in operations such as "instanceof".  To this end, I have modified SableVM
to be a little more eager in linking classes (this is *NOT* initialization,
if you care about details such as Chapter 5 of the JVM spec.).

The problem was that the System class loader (gnu.java.lang.SystemClassLoader)
called java.lang.ClassLoader.defineClass() but not resolveClass().  I could
have modified the System class loader to also call resolveClass, but this would
have not been a robust fix.

defineClass() documentation say:

   Converts an array of bytes into an instance of class Class. Before the
   Class can be used it must be resolved.

But, I think this can make a VM quite fragile.  As the JVM spec allows for it,
I have modified SableVM "link" the class in defineClass.  I also made sure
that no reference to java.lang.Class instances escape to java code before the
related class is linked.  This should make the VM more robust in face of
weaker java code.

Laurent, you can try the staging snapshot (staging+2298) in:
http://www.info.uqam.ca/~egagnon/staging

and report your findings.  :-)

Thanks a lot for the bug report!

Etienne

Etienne Gagnon wrote:
> Hmmm...
> 
> Laurent Martelli wrote:
> 
>> It's not really a small test case since it relies on Jetty, but it
>> triggers ... 
> 
> 
> Somehow, the isAssignableFrom is called at a time when OutputStreamLogSink
> is not yet *prepared* (this is even before initialization!).  As 
> preparation
> is where the appropriate structures are created for doing constant-time
> instanceof checks, it explains why the check fails (as the data used for
> the check is invalid).
> 
> So, some code lets a Class reference escape prior to preparation!  This 
> must
> be fixed...  This will require some hunting.
> 
> Feel free to help.  To do so, reconfigure and recompile sablevm:
> 
> $ cd sablevm-staging-...
> $ ./configure --enable-debugging-features # then any other option use 
> previously used
> $ make clean
> $ make
> $ make install
> 
> then, I recommend that you use ddd to debug sablevm:
> 
> $ ddd path-to../bin/sablevm &
> 
> [on Debian, you simply:
> $ apt-get install ddd
> ]
> 
> Have fun!
> 
> Etienne
> 

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

Re: [Sablevm-developer] Re: Reflection and interfaces

[David <db...@cs...>]

On Fri, May 21, 2004 at 01:36:07PM -0400, Etienne Gagnon wrote:
> I wonder...
>=20
> David B=E9langer wrote:
> >>-    if (isPrimitive() || isArray() || isInterface())
> >>+    if (isPrimitive() || isArray())
> >
> >This is the correct fix and will make it in SableVM/staging version.
>=20
> Wouldn't the following be the correct fix?
>=20
> -    if (isPrimitive() || isArray() || isInterface())
> +    if (isPrimitive())
>=20
> Arrays inherit from java.lang.Object, so they have methods, isn't it?
>=20

Yes, but getDeclaredMethods returns only methods declared by the type,
not all methods inherited.
Arrays do not declare them and the spec says it should return an empty
array of Methods if it is an array type.

If getMethods() is called (for all methods), it should correctly figure
out that the superclass of the array type is Object and collect them.


David

---

David B=E9langer
Graduate Student
School of Computer Science
McGill University
Office: MC226

Web page:   http://www.cs.mcgill.ca/~dbelan2/
Public key: http://www.cs.mcgill.ca/~dbelan2/public_key.txt

Re: [Sablevm-developer] Re: Running jetty

[Etienne G. <gag...@uq...>]

Hmmm...

Laurent Martelli wrote:
> It's not really a small test case since it relies on Jetty, but it
> triggers 
>... 

Somehow, the isAssignableFrom is called at a time when OutputStreamLogSink
is not yet *prepared* (this is even before initialization!).  As preparation
is where the appropriate structures are created for doing constant-time
instanceof checks, it explains why the check fails (as the data used for
the check is invalid).

So, some code lets a Class reference escape prior to preparation!  This must
be fixed...  This will require some hunting.

Feel free to help.  To do so, reconfigure and recompile sablevm:

$ cd sablevm-staging-...
$ ./configure --enable-debugging-features # then any other option use previously used
$ make clean
$ make
$ make install

then, I recommend that you use ddd to debug sablevm:

$ ddd path-to../bin/sablevm &

[on Debian, you simply:
$ apt-get install ddd
]

Have fun!

Etienne

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

[Sablevm-developer] Re: Running jetty

[Laurent M. <la...@ao...>]

>>>>> "Laurent" == Laurent Martelli <la...@ao...> writes:

[...]

  Laurent> I suspect it's classloader related because this small test
  Laurent> is passed

Or maybe not. I've put some traces in Jetty and it turns that the
class loader of both classes is gnu.java.lang.SystemClassLoader@14944.

Anyway, the location of the problem in Jetty is org.mortbay..util.Log
line 96.

-- 
Laurent Martelli
la...@ao...                                Java Aspect Components
http://www.aopsys.com/                          http://jac.objectweb.org

[Sablevm-developer] Re: Running jetty

[Laurent M. <la...@ao...>]

I suspect it's classloader related because this small test is passed
(it's a copy of the jetty code that triggers the above bug): 

Re: [Sablevm-developer] Re: Reflection and interfaces

[Etienne G. <gag...@uq...>]

I wonder...

David B=E9langer wrote:
>>-    if (isPrimitive() || isArray() || isInterface())
>>+    if (isPrimitive() || isArray())
>=20
> This is the correct fix and will make it in SableVM/staging version.

Wouldn't the following be the correct fix?

-    if (isPrimitive() || isArray() || isInterface())
+    if (isPrimitive())

Arrays inherit from java.lang.Object, so they have methods, isn't it?

Etienne

--=20
Etienne M. Gagnon, Ph.D.             http://www.info.uqam.ca/~egagnon/
SableVM:                                       http://www.sablevm.org/
SableCC:                                       http://www.sablecc.org/

[Sablevm-developer] Re: SableVM 1.1.4 Released

[Laurent M. <la...@ao...>]

laurent@stan:~/tmp$ sablevm -Y Loader
java/lang/Integer.class -> null
laurent@stan:~/tmp$ java Loader
java/lang/Integer.class -> java.util.zip.ZipFile$ZipFileInputStream@14fe5c

-- 
Laurent Martelli
la...@ao...                                Java Aspect Components
http://www.aopsys.com/                          http://jac.objectweb.org

Re: [Sablevm-developer] Re: Reflection and interfaces

[Etienne G. <gag...@uq...>]

Laurent Martelli wrote:
> I think java.lang.Class.getDeclaredMethods() should rather look like this:
> 
>   Method[] getDeclaredMethods (boolean publicOnly)
>   {
> -    if (isPrimitive() || isArray() || isInterface())
> +    if (isPrimitive() || isArray())
>     {
>       return new Method[] { };
>     }
>     else
>     {
>       return VMClass.getDeclaredMethods(this, publicOnly);
>     }
>   }

This makes sense.  I have checked that code in staging, but it would be nice if
you could actually test it.

I have put packages for you to test on:
http://www.info.uqam.ca/~egagnon/sablevm/

Please let us know it it works.

Thanks!

Etienne

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

Re: [Sablevm-developer] Re: Reflection and interfaces

[David <db...@cs...>]

On Fri, May 21, 2004 at 12:21:25PM +0200, Laurent Martelli wrote:
> >>>>> "Grzegorz" =3D=3D Grzegorz B Prokopski <ga...@de...> writes:
>=20
>   Grzegorz> On (19/05/04 14:19), Laurent Martelli wrote:
>   >> Hello,
>   >>=20
>   >> java.lang.Class.getMethods() returns an empty for interfaces with
>   >> sablevm 1.1.4. Is it a known bug/limitation ?
>=20
>   Grzegorz> I have looked at the code of Classpath and functions
>   Grzegorz> called into the VM and it seems that this should be
>   Grzegorz> handled.  I presume it's a bug then.
>=20
>   Grzegorz> Could somebody more insightful take a look at that?
>=20
> I think java.lang.Class.getDeclaredMethods() should rather look like th=
is:
>=20
>   Method[] getDeclaredMethods (boolean publicOnly)
>   {
> -    if (isPrimitive() || isArray() || isInterface())
> +    if (isPrimitive() || isArray())
>     {
>       return new Method[] { };
>     }
>     else
>     {
>       return VMClass.getDeclaredMethods(this, publicOnly);
>     }
>   }

Hi Laurent,

This is the correct fix and will make it in SableVM/staging version.

Thanks,
David

---

David B=E9langer
Graduate Student
School of Computer Science
McGill University
Office: MC226

Web page:   http://www.cs.mcgill.ca/~dbelan2/
Public key: http://www.cs.mcgill.ca/~dbelan2/public_key.txt

Re: [Sablevm-developer] primitive types and isAssignableFrom

[Etienne G. <gag...@UQ...>]

Hi Laurent,

Laurent Martelli wrote:
> I think the following implementation of
> java.lang.isAssignableFrom(Class) should be OK for primitive types too:
> 
>   public boolean isAssignableFrom(Class c)
>   {
>     if (c == null) {
>       throw new NullPointerException();
>     } else if (this.isPrimitive() || c.isPrimitive()) {
>         return this == c;
>     } else {
>       return VMClass.isAssignableFrom (this, c);
>     }
>   }

Applied in staging.  Thanks!

Little request for future patches...

Usually, the preferred form of patches is what is called "unified diffs".
You get such a thing by running diff with the -u option, e.g.:

diff -u original_file modified_file

Thanks,

Etienne

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

Re: [Sablevm-developer] Re: SableVM 1.1.4 Released

[Etienne G. <gag...@UQ...>]

Hi Laurent,

Laurent Martelli wrote:
> I've just tried it and have a problem:
> java.lang.ClassLoader.getResourceAsStream("gnu/java/io/decode/Decoder8859_1.class")
> returns null. I encounter this because my application has custom class
> loader. I can work around it, but I think it's a bug anyway.

Again, could you provide a simple test program that exhibits the bug.  (e.g.
a Test.java file that can be compiled and run exhibiting the bug).

Thanks,

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

Re: [Sablevm-developer] Running jetty

[Etienne G. <gag...@uq...>]

Laurent Martelli wrote:
> I have a framework (http://jac.objectweb.org/) which uses Jetty to
> provide a web frontend for applications (via servlets). I almost have
> this running with sablevm but I have these exception:
>... 
> Our framework use a custom classloader, so my guess is that problems
> may arise from this in java.lang.Class.isAssignableFrom().

Could you provide a small test case that fails, to help us identify the bug?

Thanks,

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

[Sablevm-developer] Re: Reflection and interfaces

[Laurent M. <la...@ao...>]

>>>>> "Chris" == Chris Pickett <chr...@ma...> writes:

  Chris> Chris Pickett wrote:
  >> Laurent Martelli wrote:
  >>>  I think java.lang.Class.getDeclaredMethods() should rather look
  >>> like this:
  >>> 
  >>> Method[] getDeclaredMethods (boolean publicOnly) { - if
  >>> (isPrimitive() || isArray() || isInterface()) + if
  >>> (isPrimitive() || isArray()) { return new Method[] { };
  >>> }
  >>> else { return VMClass.getDeclaredMethods(this, publicOnly);
  >>> }
  >>> }
  >> In the meantime, for the above, you can take a look at
  >> VMClass.getDeclaredMethods() in
  >> sablevm/src/libsablevm/java_lang_VMClass.c and maybe you'll find
  >> the problem.

  Chris> That's a silly answer from me, because you're suggesting the
  Chris> problem is with GNU Classpath, not SableVM.  Does it work for
  Chris> you if you make the above change to Classpath?

Yes it does. 

-- 
Laurent Martelli
la...@ao...                                Java Aspect Components
http://www.aopsys.com/                          http://jac.objectweb.org

Re: [Sablevm-developer] Re: Reflection and interfaces

[Chris P. <chr...@ma...>]

Chris Pickett wrote:
> Laurent Martelli wrote:
>>
>>I think java.lang.Class.getDeclaredMethods() should rather look like this:
>>
>>  Method[] getDeclaredMethods (boolean publicOnly)
>>  {
>>-    if (isPrimitive() || isArray() || isInterface())
>>+    if (isPrimitive() || isArray())
>>    {
>>      return new Method[] { };
>>    }
>>    else
>>    {
>>      return VMClass.getDeclaredMethods(this, publicOnly);
>>    }
>>  }
> 
> In the meantime, for the above, you can take a look at 
> VMClass.getDeclaredMethods() in 
> sablevm/src/libsablevm/java_lang_VMClass.c and maybe you'll find the 
> problem.

That's a silly answer from me, because you're suggesting the problem is 
with GNU Classpath, not SableVM.  Does it work for you if you make the 
above change to Classpath?

Cheers,
Chris

Re: [Sablevm-developer] Re: Reflection and interfaces

[Chris P. <chr...@ma...>]

Laurent Martelli wrote:
>>>>>>"Grzegorz" == Grzegorz B Prokopski <ga...@de...> writes:
> 
> 
>   Grzegorz> On (19/05/04 14:19), Laurent Martelli wrote:
>   >> Hello,
>   >> 
>   >> java.lang.Class.getMethods() returns an empty for interfaces with
>   >> sablevm 1.1.4. Is it a known bug/limitation ?
> 
>   Grzegorz> I have looked at the code of Classpath and functions
>   Grzegorz> called into the VM and it seems that this should be
>   Grzegorz> handled.  I presume it's a bug then.
> 
>   Grzegorz> Could somebody more insightful take a look at that?
> 
> I think java.lang.Class.getDeclaredMethods() should rather look like this:
> 
>   Method[] getDeclaredMethods (boolean publicOnly)
>   {
> -    if (isPrimitive() || isArray() || isInterface())
> +    if (isPrimitive() || isArray())
>     {
>       return new Method[] { };
>     }
>     else
>     {
>       return VMClass.getDeclaredMethods(this, publicOnly);
>     }
>   }
> 
> I though I had already sen this to the list, but I've been using gmane
> and there seems to be a delay.

Hi Laurent,

I never got this java.lang.getDeclaredMethods() patch before and I'm 
subscribed to sab...@li....  I did however 
get a suggestion for java.lang.isAssignableFrom().

I'm not a Classpath hacker, so you'll need to wait for a response from 
Etienne or David about that.

In the meantime, for the above, you can take a look at 
VMClass.getDeclaredMethods() in 
sablevm/src/libsablevm/java_lang_VMClass.c and maybe you'll find the 
problem.

Also, if you want to get a recent "staging" branch of both sablevm and 
sablevm-classpath (this is the first-level shared developer branch), you 
can follow the instructions here:

http://devel.sablevm.org/wiki/Getting

Cheers,
Chris

[Sablevm-developer] Re: Reflection and interfaces

[Laurent M. <la...@ao...>]

>>>>> "Grzegorz" == Grzegorz B Prokopski <ga...@de...> writes:

  Grzegorz> On (19/05/04 14:19), Laurent Martelli wrote:
  >> Hello,
  >> 
  >> java.lang.Class.getMethods() returns an empty for interfaces with
  >> sablevm 1.1.4. Is it a known bug/limitation ?

  Grzegorz> I have looked at the code of Classpath and functions
  Grzegorz> called into the VM and it seems that this should be
  Grzegorz> handled.  I presume it's a bug then.

  Grzegorz> Could somebody more insightful take a look at that?

I think java.lang.Class.getDeclaredMethods() should rather look like this:

  Method[] getDeclaredMethods (boolean publicOnly)
  {
-    if (isPrimitive() || isArray() || isInterface())
+    if (isPrimitive() || isArray())
    {
      return new Method[] { };
    }
    else
    {
      return VMClass.getDeclaredMethods(this, publicOnly);
    }
  }

I though I had already sen this to the list, but I've been using gmane
and there seems to be a delay.

-- 
Laurent Martelli
la...@ao...                                Java Aspect Components
http://www.aopsys.com/                          http://jac.objectweb.org

Re: [Sablevm-developer] Reflection and interfaces

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

On (19/05/04 14:19), Laurent Martelli wrote:
> Hello,
> 
> java.lang.Class.getMethods() returns an empty for interfaces with
> sablevm 1.1.4. Is it a known bug/limitation ?

I have looked at the code of Classpath and functions called into the VM
and it seems that this should be handled.  I presume it's a bug then.

Could somebody more insightful take a look at that?

				GBP

PS: I am not of much help these days, but it'll change RSN.

-- 
Grzegorz B. Prokopski <ga...@de...>
Debian GNU/Linux      http://www.debian.org
SableVM - LGPLed JVM  http://www.sablevm.org
Why SableVM ?!?       http://devel.sablevm.org/wiki/WhySableVM

[Sablevm-developer] sandbox types and utility functions

[Chris P. <chr...@ma...>]

Hi,

Currently I have all of my types declared inside #ifdef ... #endif 
pairs.  However, it would be nice if I could use the m4 macros to manage 
memory and splay trees (and possibly others too).  What is the best way 
to do this?  Create my own versions of the necessary files?  Or remove 
the restriction that everything belonging to me be inside an #ifdef ... 
#endif pair, as long as VM functionality does not change?

Cheers,
Chris

[Sablevm-developer] Running jetty

[Laurent M. <la...@ao...>]

I have a framework (http://jac.objectweb.org/) which uses Jetty to
provide a web frontend for applications (via servlets). I almost have
this running with sablevm but I have these exception:

class org.mortbay.util.OutputStreamLogSink is not a org.mortbay.util.LogSink
1084973010357: EVENT  ,Starting Jetty/4.2.x,[Main-1]org.mortbay.util.Log.event(Log.java:248)
1084973010512: EVENT  ,Started HttpContext[/jac/resources],[Main-1]org.mortbay.util.Log.event(Log.java:248)
1084973010542: EVENT  ,Started SocketListener on 0.0.0.0:8088,[Main-1]org.mortbay.util.Log.event(Log.java:248)
2004-05-19 13:23:30,554 ERROR gui.display:  Failed to start webgui : org.mortbay.util.MultiException[java.lang.IllegalStateException: Servlet class org.objectweb.jac.aspects.gui.web.JacLocalServlet is not a javax.servlet.Servlet]
2004-05-19 13:23:30,562 ERROR gui.display:  Check that another server is not running on port 8088

Our framework use a custom classloader, so my guess is that problems
may arise from this in java.lang.Class.isAssignableFrom().

-- 
Laurent Martelli
la...@ao...                                Java Aspect Components
http://www.aopsys.com/                          http://jac.objectweb.org

[Sablevm-developer] primitive types and isAssignableFrom

[Laurent M. <la...@ao...>]

Hi,

I think the following implementation of
java.lang.isAssignableFrom(Class) should be OK for primitive types too:

  public boolean isAssignableFrom(Class c)
  {
    if (c == null) {
      throw new NullPointerException();
    } else if (this.isPrimitive() || c.isPrimitive()) {
        return this == c;
    } else {
      return VMClass.isAssignableFrom (this, c);
    }
  }



-- 
Laurent Martelli
la...@ao...                                Java Aspect Components
http://www.aopsys.com/                          http://jac.objectweb.org

[Sablevm-developer] Reflection and interfaces

[Laurent M. <la...@ao...>]

Hello,

java.lang.Class.getMethods() returns an empty for interfaces with
sablevm 1.1.4. Is it a known bug/limitation ?

-- 
Laurent Martelli
la...@ao...                                Java Aspect Components
http://www.aopsys.com/                          http://jac.objectweb.org

[Sablevm-developer] Re: SableVM 1.1.4 Released

[Laurent M. <la...@ao...>]

>>>>> "Etienne" == Etienne Gagnon <gag...@uq...> writes:

  Etienne> The developers of the SableVM Project are proud to announce
  Etienne> the official release of SableVM 1.1.4.  SableVM is a
  Etienne> liberally licensed Free Java Virtual Machine.  See the
  Etienne> "About SableVM" section below for more informations about
  Etienne> SableVM.

I've just tried it and have a problem:
java.lang.ClassLoader.getResourceAsStream("gnu/java/io/decode/Decoder8859_1.class")
returns null. I encounter this because my application has custom class
loader. I can work around it, but I think it's a bug anyway.

Regards,
Laurent

-- 
Laurent Martelli
la...@ao...                                Java Aspect Components
http://www.aopsys.com/                          http://jac.objectweb.org

Re: [Sablevm-developer] staging CP does not build java classes

[Etienne G. <gag...@uq...>]

David B=E9langer wrote:
> The recent CP --disable-java "fix" prevents the installation of java
> classes as both if condition in the lib/Makefile.am are not both true
> for SableVM.

So, I screwed up...

> Could not really figure out how the BUILD_GLIBJ_ZIP is used to disable
> java, it is unused in that Makefile.am.

This variable makes no sense; it should probably be INSTALL_GLIB_ZIP.

Anyway, as sablevm does not yet support zip files, I simple disabled
CLASSPATH_WITH_GLIBJ, and directly set INSTALL_CLASS_FILES appropriately.=


Please let me know if something's still wrong.

Etienne

--=20
Etienne M. Gagnon, Ph.D.             http://www.info.uqam.ca/~egagnon/
SableVM:                                       http://www.sablevm.org/
SableCC:                                       http://www.sablecc.org/

[Sablevm-developer] staging CP does not build java classes

[David <db...@cs...>]

Hi,

The recent CP --disable-java "fix" prevents the installation of java
classes as both if condition in the lib/Makefile.am are not both true
for SableVM.

Could not really figure out how the BUILD_GLIBJ_ZIP is used to disable
java, it is unused in that Makefile.am.

David

---

David B=E9langer
Graduate Student
School of Computer Science
McGill University
Office: MC226

Web page:   http://www.cs.mcgill.ca/~dbelan2/
Public key: http://www.cs.mcgill.ca/~dbelan2/public_key.txt