On Thu, 2 Feb 2006, Tim Jenness wrote:
> On Fri, 3 Feb 2006, Starlink Software wrote:
>
>> Module: starjava15
>> Status: failed
>> Build URL:
>> http://dev.starlink.ac.uk/build/RHEL-WS3-3_i386/logs/starjava15.1138927243.log
>>
>
> Hmm. Looks like its still broken. Don't suppose anyone has any hints on how
> to approach debugging this without using printf (I haven't even got a
> starjava build working at the moment)?
Hi Tim,
I think I've tracked this down to a realloc call in dauloc.c. Your number
of bytes was off by factor "*sizeof(struct LCP*)". I guess this means that
the realloc call actually freed some memory (changes amount to min of
existing and requested size), and this fact isn't tested by HDS or JNIHDS,
as neither need more than NBINS_INC.
As for hints. The best you can do is recompile the libraries (HDS in this
case) using -g, rebuild the JNI native one in question and then look at
any core dumps in gdb. This took me straight to your new code, so I peered
harder until the problem became clear. If that doesn't help then you're
stuck, as a C-based equivalent of the test case is needed.
Cheers,
Peter.
|