classpath in 2011.03-maintenance

classpath in 2011.03-maintenance

[George C. Huntington, III]

who are the people I should talk to about problems building classpath and
jamvm on the  2011.03-maintenance branch?

Re: classpath in 2011.03-maintenance

[Tom Rini]

On 07/07/2011 06:39 AM, George C. Huntington, III wrote:
> who are the people I should talk to about problems building classpath and
> jamvm on the  2011.03-maintenance branch?

Start by posting your problems (and your config) to the ML.

-- 
Tom Rini
Mentor Graphics Corporation

Re: classpath in 2011.03-maintenance

[Jesse Gilles]

[-- Attachment #1: Type: text/plain, Size: 1073 bytes --]

> On 07/07/2011 06:39 AM, George C. Huntington, III wrote:
> > who are the people I should talk to about problems building classpath and
> > jamvm on the  2011.03-maintenance branch?
> 
> Start by posting your problems (and your config) to the ML.
> 

I am also having issues with classpath 0.98 on the 2011.03-maintenance
branch.  The package builds successfully, but the class library zip
(/usr/share/classpath/glibj.zip) produced is way too small and is
missing most classes.  The compile log shows many (around 4500) java
compile errors, but the make doesn't error out so the bitbake task
succeeds.

current commit on my branch is 819822d3d32700dafbb3e33172450c34c6334f13
DISTRO = "angstrom-2010.x"
MACHINE = "at91sam9g20ek"

Build machine is Ubuntu 10.04 x86_64.

Lots of errors like this:

2. ERROR in ./external/asm/org/objectweb/asm/ClassVisitor.java (at line
65)
    String name,
    ^^^^^^
String cannot be resolved to a type

I attached the entire compile log for more detail.

Is this perhaps an issue with ecj?

Thanks,
Jesse

[-- Attachment #2: log.do_compile.1658.gz --]
[-- Type: application/x-gzip, Size: 764782 bytes --]

Re: classpath in 2011.03-maintenance

[George C. Huntington, III]

classpath is not actually building correctly, a bug in the shell script
wrapper for calling javac is eating the error code from javac, and making it
look like things built ok when they did not.

On Fri, Aug 5, 2011 at 10:16 AM, Jesse Gilles <jgilles@multitech.com> wrote:

> > On 07/07/2011 06:39 AM, George C. Huntington, III wrote:
> > > who are the people I should talk to about problems building classpath
> and
> > > jamvm on the  2011.03-maintenance branch?
> >
> > Start by posting your problems (and your config) to the ML.
> >
>
> I am also having issues with classpath 0.98 on the 2011.03-maintenance
> branch.  The package builds successfully, but the class library zip
> (/usr/share/classpath/glibj.zip) produced is way too small and is
> missing most classes.  The compile log shows many (around 4500) java
> compile errors, but the make doesn't error out so the bitbake task
> succeeds.
>
> current commit on my branch is 819822d3d32700dafbb3e33172450c34c6334f13
> DISTRO = "angstrom-2010.x"
> MACHINE = "at91sam9g20ek"
>
> Build machine is Ubuntu 10.04 x86_64.
>
> Lots of errors like this:
>
> 2. ERROR in ./external/asm/org/objectweb/asm/ClassVisitor.java (at line
> 65)
>    String name,
>    ^^^^^^
> String cannot be resolved to a type
>
> I attached the entire compile log for more detail.
>
> Is this perhaps an issue with ecj?
>
> Thanks,
> Jesse
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
>
>

Re: classpath in 2011.03-maintenance

[Henning Heinold]

Hi,

we made some hacks bescause for some java packages javac fails at random and
we just try to recompile it for some rounds and have a higher chance to
get it done. To admit the ecj script is now diffrent from the one I wrote original,
maybee Stefan Schmidt made some more tweaks.

As side note we did not test classpath compiling for target architecture,
we only made sure classpath-native works, so we can compile openjdk
for the target.

Bye Henning

Re: classpath in 2011.03-maintenance

[George C. Huntington, III]

I posted a patch that makes the wrapper script return the error code if it
fails after 5 retries.  it is on patchwork.
http://patches.openembedded.org/patch/7249/

I don't know if anyone has reviewed it, but it worked for me.


On Fri, Aug 5, 2011 at 11:17 AM, Henning Heinold
<heinold@inf.fu-berlin.de>wrote:

> Hi,
>
> we made some hacks bescause for some java packages javac fails at random
> and
> we just try to recompile it for some rounds and have a higher chance to
> get it done. To admit the ecj script is now diffrent from the one I wrote
> original,
> maybee Stefan Schmidt made some more tweaks.
>
> As side note we did not test classpath compiling for target architecture,
> we only made sure classpath-native works, so we can compile openjdk
> for the target.
>
> Bye Henning
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
>

Re: classpath in 2011.03-maintenance

[Henning Heinold]

On Fri, Aug 05, 2011 at 11:38:22AM -0400, George C. Huntington, III wrote:
> I posted a patch that makes the wrapper script return the error code if it
> fails after 5 retries.  it is on patchwork.
> http://patches.openembedded.org/patch/7249/

Did you try to compile openjdk with it serval times under
32bit and 64bit?

Bye Henning

Re: classpath in 2011.03-maintenance

[George C. Huntington, III]

only 2 times under 64 bit


On Fri, Aug 5, 2011 at 11:43 AM, Henning Heinold
<heinold@inf.fu-berlin.de>wrote:

> On Fri, Aug 05, 2011 at 11:38:22AM -0400, George C. Huntington, III wrote:
> > I posted a patch that makes the wrapper script return the error code if
> it
> > fails after 5 retries.  it is on patchwork.
> > http://patches.openembedded.org/patch/7249/
>
> Did you try to compile openjdk with it serval times under
> 32bit and 64bit?
>
> Bye Henning
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
>

Re: classpath in 2011.03-maintenance

[Jesse Gilles]

On Fri, 2011-08-05 at 11:38 -0400, George C. Huntington, III wrote:
> I posted a patch that makes the wrapper script return the error code
> if it
> fails after 5 retries.  it is on patchwork.
> http://patches.openembedded.org/patch/7249/
> 
> I don't know if anyone has reviewed it, but it worked for me.

I gave it a try and now the classpath build errors out earlier:

| Adding java source files from srcdir '..'.
| Adding java source files from VM directory ../vm/reference
| javac -Xlint:unchecked,cast,divzero,empty,finally,overrides  -J-Xmx768M -source 1.5 -target 1.5 -bootclasspath '' -classpath ../vm/reference:..:../external/w3c_dom:../external/sax:../external/relaxngDatatype:../external/jsr166:.:: -d . @classes
| Unrecognized option : ../vm/reference:..:../external/w3c_dom:../external/sax:../external/relaxngDatatype:../external/jsr166:.::
| make[1]: *** [compile-classes] Error 255
| make[1]: *** Waiting for unfinished jobs....
| touch resources
| make[1]: Leaving directory `/home/jjg/projects/angstrom-oe-2011.03/build/tmp/work/armv5te-angstrom-linux-gnueabi/classpath-0.98-r4/classpath-0.98/lib'
| make: *** [all-recursive] Error 1
| + die 'oe_runmake failed'
| + oefatal 'oe_runmake failed'
| + echo FATAL: 'oe_runmake failed'
| FATAL: oe_runmake failed
| + exit 1
NOTE: package classpath-0.98-r4: task do_compile: Failed

Jesse

Re: classpath in 2011.03-maintenance

[George C. Huntington, III]

the -bootclasspath ''
part where it uses the single quotes next to eack other, if you replace that
with ::
two colons it will magically work
I think for some reason it doesn't like '' for an empty path but two
path separators back to back ::
will work great.



On Fri, Aug 5, 2011 at 4:44 PM, Jesse Gilles <jgilles@multitech.com> wrote:

> ''

Re: classpath in 2011.03-maintenance

[Jesse Gilles]

On Fri, 2011-08-05 at 17:24 -0400, George C. Huntington, III wrote:
> the -bootclasspath ''
> part where it uses the single quotes next to eack other, if you
> replace that
> with ::
> two colons it will magically work
> I think for some reason it doesn't like '' for an empty path but two
> path separators back to back ::
> will work great.

Yes, I had just figured out the same thing :)

Re: classpath in 2011.03-maintenance

[Stefan Schmidt]

Hello.

On Fri, 2011-08-05 at 11:38, George C. Huntington, III wrote:
> I posted a patch that makes the wrapper script return the error code if it
> fails after 5 retries.  it is on patchwork.
> http://patches.openembedded.org/patch/7249/
> 
> I don't know if anyone has reviewed it, but it worked for me.

It was on my list but I dropped it once the list got to big. ;) Sorry
for that.

I just started a build with our full image (openjdk for the target and
many java apps) and we should know more when I'm up again tomorrow.
Will let you know.

regards
Stefan Schmidt

Re: classpath in 2011.03-maintenance

[Jesse Gilles]

On Fri, 2011-08-05 at 11:38 -0400, George C. Huntington, III wrote:
> I posted a patch that makes the wrapper script return the error code
> if it
> fails after 5 retries.  it is on patchwork.
> http://patches.openembedded.org/patch/7249/

It seems like saving the javac script arguments into the ARGS variable
is what broke the old behavior where -bootclasspath '' worked for
building classpath.  Putting that part of the script back to the way it
used to be fixes it.

Do you want to merge this in with your error code return patch?

--- a/recipes/ecj/files/ecj.in
+++ b/recipes/ecj/files/ecj.in
@@ -1,13 +1,16 @@
-ARGS=${1+"$@"}
 COUNT=5
 END=0
+RETURNCODE=0

 while test "${COUNT}" -gt "${END}"
 do
- ${RUNTIME} -Xmx1024m -cp ${ECJ_JAR} org.eclipse.jdt.internal.compiler.batch.Main ${ARGS}
+ ${RUNTIME} -Xmx1024m -cp ${ECJ_JAR} org.eclipse.jdt.internal.compiler.batch.Main ${1+"$@"}
+ RETURNCODE=${?}
  if test "${?}" -eq "${END}"
  then
     break
  fi
  COUNT=$(($COUNT-1))
 done
+
+exit ${RETURNCODE}

Otherwise, I can submit it as another patch, but it will depend on yours...

Thanks,
Jesse

Re: classpath in 2011.03-maintenance

[George C. Huntington, III]

I have no problem with it.  whatever is easier for review and testing.

On Mon, Aug 8, 2011 at 3:08 PM, Jesse Gilles <jgilles@multitech.com> wrote:

> On Fri, 2011-08-05 at 11:38 -0400, George C. Huntington, III wrote:
> > I posted a patch that makes the wrapper script return the error code
> > if it
> > fails after 5 retries.  it is on patchwork.
> > http://patches.openembedded.org/patch/7249/
>
> It seems like saving the javac script arguments into the ARGS variable
> is what broke the old behavior where -bootclasspath '' worked for
> building classpath.  Putting that part of the script back to the way it
> used to be fixes it.
>
> Do you want to merge this in with your error code return patch?
>
> --- a/recipes/ecj/files/ecj.in
> +++ b/recipes/ecj/files/ecj.in
> @@ -1,13 +1,16 @@
> -ARGS=${1+"$@"}
>  COUNT=5
>  END=0
> +RETURNCODE=0
>
>  while test "${COUNT}" -gt "${END}"
>  do
> - ${RUNTIME} -Xmx1024m -cp ${ECJ_JAR}
> org.eclipse.jdt.internal.compiler.batch.Main ${ARGS}
> + ${RUNTIME} -Xmx1024m -cp ${ECJ_JAR}
> org.eclipse.jdt.internal.compiler.batch.Main ${1+"$@"}
> + RETURNCODE=${?}
>  if test "${?}" -eq "${END}"
>  then
>     break
>  fi
>  COUNT=$(($COUNT-1))
>  done
> +
> +exit ${RETURNCODE}
>
> Otherwise, I can submit it as another patch, but it will depend on yours...
>
> Thanks,
> Jesse
>
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
>