I have the same problem:

PHP 4.1.1

Debian Linux 2.2 with all updates as of 1st Feb. 2002

Apache 1.3.23 (or Apache delivered with debian, doesn't matter)

Java(TM) 2 Runtime Environment, Standard Edition (build 1.3.1_02-b02)
Java HotSpot(TM) Client VM (build 1.3.1_02-b02, mixed mode)
(or JDK 1.3.1_01, doesnt' matter)

Sablotron 0.71 or 0.82 (doesn't matter)

./configure \
        --with-tsrm-pthreads \
        --prefix=/opt/apache-1.3.23/php-4.1.1 \
        --with-apxs=/opt/apache-1.3.23/bin/apxs \
        --enable-xslt \
        --with-xslt-sablot=/opt/local \
        --enable-sablot-errors-descriptive

Adding 50 other directives doesn't matter. As soon
as I remove --enable-xslt and --with-xslt-sablot
"new Java(...)" works.

Please test with  PHP 4.1.1+JDK 1.2 and report the result back 
Please do not forget  updating PHP version. Thanks.

It always worked with JDK 1.2... 

And i don't think, there was much of a change in ext/java or ext/xslt from 2001-09-1 till today (in the 4.1 branch, don't know about HEAD)

anyway, i use at the moment JDK 1.2, so this is not that important for me, someone else has to verify it (maybe giving JDK 1.4 a try)

The system:
PIII 1GHz 512Mb RAM
Red Hat 7.1 (kernel 2.4.2-2)
Sun JDK 1.4.0 RC
Apache 1.3.20 (compiled with LDFLAGS...)
PHP 4.0.6 (--with-apxs...)
Sablotron 0.90
Expat 1.95.2

The problem:
PHP compiles fine with Apache (DSO) and Sablotron/Expat.
When compiling with JDK, pure PHP+XML/XSLT code executes fine either via browser or via command line. Any java code inserted and core dumps and segmentation faults appear.
If compiling Apache, PHP and JDK, simple PHP+Java code resumes fine in both modes (command line/browser).
I noticed that related problems had been already reported with different versions, but they'd never solved.
I also read someone arguing why is that someone would use PHP and Java and I got twisted.
The issue:
There's an application project I'm working with where we use PHP code to execute a database query (via WWWISIS), returning a XML content that is transformed with XSLT. This is already done with PHP/Sablot/Expat. But, with java, we would execute multiple threads, each one collecting a piece of XML data, which bundled together would be then returned to PHP and then transformed.
I appreciate any suggestions.

It's strange from the point of view of Sablotron.

First... Sablotron never performs some thread locking, since the code (if API is used in proper way) is thread safe. It's not absolutely bulletproof, but should work undem most conditions.

What is very suspicious is, that the Arena::~Arena in the backtrace above, is called from some other library, since it should be hidden form other libs and used internally (and called from other Sablotron destructors as a result of this).

Has somebody a clue, what library defines the ciEnv class? It looks like this is a linking problem (and is it even possible?) Could somebody rename Arena to SabArena in the Sablotron source (there are just few occurences) and test it?

can you please try the latest CVS?  I believe I may have fixed this for you... but I could be wrong.  

Just tested with latest CVS and jdk 1.4 and sablotron 0.96. It still segfaults... with the same bt:

#0  0x40109b00 in free () from /lib/libc.so.6
#1  0x40109aa3 in free () from /lib/libc.so.6
#2  0x403e303f in Arena::dispose () from /usr/lib/libsablot.so.0
#3  0x403e308c in Arena::~Arena () from /usr/lib/libsablot.so.0
#4  0x40c8224c in Compilation::~Compilation () from /usr/lib/java/jre/lib/i386/client/libjvm.so
#5  0x40c48c8d in Compiler::compile_method () from /usr/lib/java/jre/lib/i386/client/libjvm.so
#6  0x40c352d9 in CompileBroker::invoke_compiler_on_method ()
   from /usr/lib/java/jre/lib/i386/client/libjvm.so
#7  0x40c34db9 in CompileBroker::compiler_thread_loop () from /usr/lib/java/jre/lib/i386/client/libjvm.so
#8  0x40c06c36 in compiler_thread_entry () from /usr/lib/java/jre/lib/i386/client/libjvm.so
#9  0x40c0762a in JavaThread::thread_main_inner () from /usr/lib/java/jre/lib/i386/client/libjvm.so
#10 0x40c03b90 in JavaThread::run () from /usr/lib/java/jre/lib/i386/client/libjvm.so
#11 0x40bcd10a in _start () from /usr/lib/java/jre/lib/i386/client/libjvm.so
#12 0x40e530ba in pthread_start_thread () from /lib/libpthread.so.0
#13 0x40e53101 in pthread_start_thread_event () from /lib/libpthread.so.0

It's maybe clear for everyone, but both (sablotron and libjvm) have obviously a class called Arena in the global namespace... This class is not used in the php-source-code, but libjvm.so calls it on destruction of some compiling stuff. 

I'm by no way the compiling, resp c++ expert to resolve this problem, but maybe someone else has an idea. Would be great :)

chregu

Hi

I found the solution (as suggested by pavel)

I did a quick search&replace (Arena to SabArena) in Sablot/src/engine and now it doesn't segfault anymore...

I will check with the Sablot guys, what they think of it.

chregu

Good news from Sablotron.

They will change this in the next release (0.97?) and therefore we can close now this almost one year old bug. 

chregu