[gdal-dev] JVM crash on making a call to the class/method OGRLineString::exportToWkt

[Even Rouault]

Brian,

This is indeed going be difficult to help you in an efficient way without a 
testable case...

First, check that GDAL is configured --with-threads

Then try the OGRFeature and OGRTestGC test applications that are generated in 
swig/java/build/apps

How ogrinfo and ogr2ogr work on the TIGER data ? (try with valgrind) That 
could be a bug in the TIGER driver if it doesn't occur with other format.

Your errors seem to indicate some native heap memory corruption. Perhaps a bug 
in GDAL or the Java bindings, or perhaps a misuse of the API that lead to 
that. (There are indeed some ways of causing corruption by misusing the API if 
you don't follow the do's / dont do's in the API doc, and potentially other 
undocumented ways...)

The best would really for you to extract the parts where you use OGR API and 
try having a minimum app that trigger the issue.

Best regards,

Even

Le jeudi 30 septembre 2010 15:18:48, Brian Kelly a écrit :
> Hi all,
> 
> Apologies in advance for not providing a testable case that can be fired
> up. I'm hoping the background info provided and investigation work shown
> below is enough to help trace what the issue is.
> 
> PROBLEM DEFINITION:
> I work with a custom java application that loads GIS data using GDAL
> (current TRUNK code in use though issue occurs with v1.7.2 also). We can
> load a variety of data types (MapInfo, DGN,etc) with no issues. A problem
> occurs when TIGER data is loaded though - the TIGER data loads and is
> viewable but after a few pans and zooms the JAVA server process crashes.
> This seems to happen when it tries to unload loaded features. Was using US
> Census data tiger files.
> 
> PROBLEM ANALYSIS:
> Two sources of debug information were looked at. A (1) general log file
> generated by java and (2) a core dump file. The outputs of these are below.
> 
> 1. LOG NUMBER ONE: The log file generated on crash indicates the source of
> the problem is the GDAL library when features are deleted.
> 
> Current thread (0x0000000041a4d800):  JavaThread
> "FeatureNativeObjectsCleaner" daemon [_thread_in_native, id=28311,
> stack(0x00007ff5f4cc8000,0x00007ff5f4dc9000)]
> siginfo:si_signo=SIGSEGV: si_errno=0, si_code=1 (SEGV_MAPERR),
> si_addr=0x000000000000003a
> Stack: [0x00007ff5f4cc8000,0x00007ff5f4dc9000],  sp=0x00007ff5f4dc77a8,
> free space=3fd0000000000000018k
> Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, C=native
> code)
> C  0x0000000042355af0
> j  org.gdal.ogr.FeatureNative.delete()V+23
> 
> 2. LOG NUMBER TWO: A linux core dump was captured when the JVM crashed.
> This was analysed for more precise details using the GDB debugger using
> the approach outlined at
> http://www.dialogic.com/support/helpweb/dxall/tn957.aspx
> 
> The output from the core dump file analysis was:
> Loaded symbols for /usr/lib/jvm/jrrt-4.0.1-1.6.0/jre/lib/amd64/libdcpr.so
> Core was generated by `/usr/lib/jvm/java-6-jrockit/bin/java -Xms2048m
> -Xmx2048m -Djava.util.logging.ma'.
> Program terminated with signal 6, Aborted.
> #0  0x00007f51ac719a75 in raise () from /lib/libc.so.6
> (gdb) bt
> #0  0x00007f51ac719a75 in raise () from /lib/libc.so.6
> #1  0x00007f51ac71d5c0 in abort () from /lib/libc.so.6
> #2  0x00007f51ac7534fb in ?? () from /lib/libc.so.6
> #3  0x00007f51ac75d5b6 in ?? () from /lib/libc.so.6
> #4  0x00007f51ac7616d8 in ?? () from /lib/libc.so.6
> #5  0x00007f51ac76258e in malloc () from /lib/libc.so.6
> #6  0x00007f515d06c262 in OGRLineString::exportToWkt (this=0x7f5154722bb0,
>     ppszDstText=0x7f5158804728) at ogrlinestring.cpp:1049
> #7  0x00007f515d51655f in OGRGeometryShadow_ExportToWkt__SWIG_1
> (jenv=0x5952bf0,
>     jcls=<value optimized out>, jarg1=6, jarg1_=0xffffffffffffffff)
>     at ogr_wrap.cpp:1792
> #8  Java_org_gdal_ogr_ogrJNI_Geometry_1ExportToWkt_1_1SWIG_11
> (jenv=0x5952bf0,
>     jcls=<value optimized out>, jarg1=6, jarg1_=0xffffffffffffffff)
>     at ogr_wrap.cpp:7166
> #9  0x00007f5166de2715 in ?? ()
> #10 0x00000000827520c0 in ?? ()
> 
> 
> 1. The core dump file indicates that an exception occurs in the file
> ogrlinestring.cpp at L1049. In the GDAL trunk source code that I have this
> corresponds to the line:
>        *ppszDstText = (char *) VSIMalloc( nMaxString );
> 2. The output from Log #1 indicates that there might be an issue with
> memory pointers. This is based on the SEGV_MAPERR error output and the
> opinion on the post here:
> http://www.linuxquestions.org/questions/programming-9/free-invalid-next-siz
> e-fast-c-error-729852/
> 
> It looks like the problem could be a memory pointer size issue related to
> ogrlinestring. If anyone could take a quick look and suggest a fix that
> would be appreciated. I'm a java programmer lost in a C++ world on this
> one.
> 
> Thanks,
> Brian