seg fault on startup

[David Daney]

Andrew Haley wrote:
> Ben Tatham writes:
>  > More fuel for the investigation.
>  > 
>  > I have now tried to build our application with GCC-4.2-20070307, with 
>  > the much simpler -static-libgcj option...
>  > 
>  > %.o : %.jar
>  >         $(gccdir)/bin/$(platform)-gcj $(includes) -c -g -O $<
>  > 
>  > apollo : $(objects)
>  >         $(gccdir)/bin/$(platform)-gcj -g -O -static-libgcj -o 
>  > apollo.debug --main=ca.nanometrics.apollo.taurus.MissionControlTaurus 
>  > $(objects)
>  >         $(gccdir)/bin/$(platform)-strip --strip-debug apollo.debug -o apollo
>  > 
>  > Unfortunately, it still seg faults right at startup.  This time, I get a 
>  > backtrace though.  
> 
> Ah, excellent.
> 
>  > Does this mean anything to anyone?  Why does it say that the
>  > language is c++ now though?
> 
> Because it is.
> 
>  > Program terminated with signal 11, Segmentation fault.
>  > #0  _Jv_RegisterClasses (classes=0x10f80074)
>  >     at 
>  > /disk1/xtool/crosstool-0.43/build/powerpc-860-linux-gnu/gcc-4.2-20070307-glibc-2.3.3/gcc-4.2-20070307/libjava/java/lang/natClassLoader.cc:189
>  > 189     _Jv_RegisterClasses (const jclass *classes)
> 
> classes is a pointer to an array of pointers, terminated by NULL.
> Every one of those pointers should pointo t a valid class structure
> inside your app.  Print them out.

As an aside, I think there were bugs in some part of the mipsel-linux 
toolchain (I forget exactly where, but it was either the linker or 
glibc) that caused problems like this for me.  I think you are on the 
right track here.  It may be useful to compile libgcj with -O0 -g (edit 
the Makefile to change -O2 to -O0 and make clean; make).

David Daney