Threaded Mode | Linear Mode

***Emmanuel Bertin*** · 06-29-2007, 15:26

Thanks Jan for your efforts to understand the problem. And sorry for my lack of communication during the past weeks. I just could not find the time.
Now if you PM me a URL from where I could copy this dataset I can investigate on my side.
- Emmanuel.

Joerg Dietrich · 07-09-2007, 10:36

jankoh Wrote:After looking, at which files are left over after cleaning, I found out, that its the 63th and 127th file, which is not removed correctly, but for which some tmp-files are deleted. It seems to be directly links to the NFILES constant in fitscleanup.c.

Sorry for the delay; I was travelling. Here it is also the 63rd file for which a tmp file is removed, the 127th is the invalid pointer that causes the crash. Slight difference to your case, but segmentation violations can always lead to non-reproducible results on different machines. If you haven't done yet, I'll make the files available to Emmanuel.

Jan Kohnert · 07-09-2007, 11:57

I was traveling, too, so I did not make them available yet.

As far as I could test yet, the problem seems to be in remove_cleanupfilename(), it gets confused, when getting over/under the NFILES border. But I had no debugger available, so I'm not completely sure about that.

Just to mention: If you set NFILES to maybe 5 or 6 temporarily, you can get the crash with a much less files. Smile

Best regards Jan

***Emmanuel Bertin*** · 07-10-2007, 10:53

OK thanks Joerg, I am downloading the 24GB thing... Big Grin

- Emmanuel.

***Emmanuel Bertin*** · (This post was last modified: 07-16-2007 18:51 by Emmanuel Bertin.)

I believe I fixed the bug. The code in remove_cleanupfilename() was actually complete rubbish Big Grin

, and would lead to bugs whenever the memory mapping feature was triggered in SWarp.
A new release will occur very soon. Meanwhile, you may download the code at
http://terapix.iap.fr/wsvn/index/public/...=0&isdir=1
Would one of you be willing to test it on your data and machines to confirm that the problem is solved?
Thanks a lot.
- Emmanuel.

Jan Kohnert · 07-17-2007, 15:22

Emmanuel Bertin Wrote:Would one of you be willing to test it on your data and machines to confirm that the problem is solved?

Confirmed for my multithreaded 64bit machine as well as the singlethreaded 32bit machine. Cool

Best regards Jan

Joerg Dietrich · 07-17-2007, 15:40

Emmanuel Bertin Wrote:I believe I fixed the bug. The code in remove_cleanupfilename() was actually complete rubbish

Ah, that's why I didn't understand how it was supposed to work :-)

Emmanuel Bertin Wrote:Would one of you be willing to test it on your data and machines to confirm that the problem is solved?

Yes, the problem appears to be solved now.

-- Joerg