Re: ./.libs/libgcj.so: undefined reference to `__gxx_personality_v0'

["Zack Weinberg" <zackw at stanford dot edu>]

On Thu, May 17, 2001 at 09:25:34AM +0200, Andreas Jaeger wrote:
> 
> Bootstrap for i686-linux fails on the trunk with:
...
> ./.libs/libgcj.so: undefined reference to `__gxx_personality_v0'
> collect2: ld returned 1 exit status
> make[3]: *** [jv-convert] Error 1
> make[3]: Leaving directory `/builds/gcc/gcc-3.1-devel/i686-pc-linux-gnu/libjava'

I've got this too.  A bunch of libgcj files have suddenly started
generating EH data:

.libs/libgcj.a:natInput_EUCJIS.o:		U __gxx_personality_v0
.libs/libgcj.a:natInput_SJIS.o:			U __gxx_personality_v0
.libs/libgcj.a:natOutput_EUCJIS.o:		U __gxx_personality_v0
.libs/libgcj.a:natOutput_SJIS.o:		U __gxx_personality_v0
.libs/libgcj.a:shs.o:				U __gxx_personality_v0
.libs/libgcj.a:natMath.o:			U __gxx_personality_v0
.libs/libgcj.a:natPlainDatagramSocketImpl.o:	U __gxx_personality_v0
.libs/libgcj.a:natPlainSocketImpl.o:		U __gxx_personality_v0

On a casual skim, these files appear not to make any use of EH, but it
may be hiding in the headers.  They don't include <jvm.h> and
therefore are not picking up #pragma GCC java_exceptions.
Unfortunately, the only place that seems to be global enough is
config.h, and I would want to know that there is a good reason why
these files need an .eh_frame first, anyway.  I've attached natMath.s
and natMath.ii to this message in hopes someone else can divine a clue
from it.

-- 
zw  Patrick, being from a culture that has embraced scientific rationality
    _as a way to settle arguments_, naturally conflates 'true' and 'factual',
    where factual means 'measured; agreed on by means independent of any
    specific person'.  -- Graydon Saunders

natMath.ii.gz

natMath.s.gz