On Thu, 2 Jul 2009, Mark Taylor wrote:
> On Wed, 1 Jul 2009, Mark Taylor wrote:
>
>> I've committed a JNIAST fix which is an improvement - the unit tests
>> now run without a core dump. However, I am still working on some
>> related issues.
>
> I've done what I'm going to do to JNIAST. It builds against the
> current git HEAD (d2434b5da60a), and the unit tests, including a
> few new ones, pass.
>
> The FitsChan problem was somewhat related to threading; there may
> be a small number of other similar ones lurking in JNIAST, but
> I don't have time to track them down exhaustively. The unit tests
> are fairly comprehensive, so my expectation is that these things
> won't reveal themselves in use, though it's always a possiblity;
> in particular they could be triggered by changes to AST's internal
> behaviour (as happened with FitsChan). Certainly if anyone spots
> a core dump which may be related to JNIAST they should shout.
Mark,
just for reference, I've built JNIAST and tested it with SPLAT and that
didn't uncover any problems.
> The fix I've made to FitsChan is not entirely satisfactory - there's
> a memory leak in that FitsChan objects will never be garbage collected,
> which is untidy but very unlikely to cause serious problems in
> practice. I made an attempt to sort this out, but I couldn't get it
> to work for reasons I don't fully understand. No doubt I could
> resolve it with enough effort, but I'm going to leave it as it is.
OK, thanks for the help, no doubt if this does become an issue you'll
hear.
Cheers,
Peter.
|