On Mon, 31 Jan 2005, Peter W. Draper wrote:
> Yes, had some problems in the early days (mainly before I started using
> MINGW interfaces, rather than Cygwin), but it seems to have been rock
> solid for quite a while.
>
> > What's started to go wrong now? (not expecting to be able to help, just
> > curious).
>
> Well it just randomly crashes when loading in FITS table spectra.
>
> I'd swear it was some problem in the newer parts of AST (the units
> handling code), but cannot track anything down. Windows doesn't give you a
> lot to work from when things are going badly just some memory references,
> or offsets from the start of a routine if you're lucky. This is
> particularly bad because I'm building for Windows under Cygwin targeting
> MINGW, so there's a lot to get in the way. Then of course Java doesn't
> help. Checking native code for memory overruns or leaks as you'd naturally
> do when seeing this kind of behaviour is impossible as the JVM does so
> many clever things all the memory leaks/overrun checkers are completely
> stuffed.
Have you tried java -verbose:jni (and possibly -Xcheck:jni, not that
I really understand what that does)? If you can narrow it down to one
JNI call then I (or you) could take a Very Hard Look at said bit of
JNIAST/AST.
--
Mark Taylor Starlink Programmer Physics, Bristol University, UK
[log in to unmask] +44-117-928-8776 http://www.star.bris.ac.uk/~mbt/
|