Problems with usePrevious

[MatthewWhiting]

Courtesy of Mary Putman:

Hi Matt,

We've been trying to install the latest and greatest version of Duchamp on my computer that is now at Columbia University and running into problems (see below). 
Do you have any ideas on what we can do to make it work?

Thanks very much,

Mary




Begin forwarded message:
From: David Secrest <secrest@astro.columbia.edu>
Date: July 9, 2009 9:33:21 AM GMT-04:00
To: Mary Putman <mputman@astro.columbia.edu>
Cc: wen-Hsin hsu <wenhsin@umich.edu>
Subject: duchamp, still troubles

Hi Mary,

I fixed all the library problems and recompiled duchamp. Everything
compiled fine, and the VerifyDuchamp.sh script reported no errors.

However, I get the same problem that Tina had the first time around (a
segmentation fault when running with usePrevious=true), so maybe we
should contact the author.

I attached the config.log file and Tina's Input_W file.

To see the error, I ran:

Duchamp -p Input_W

where the Input_W file had

usePrevious     false

That worked fine. But then, running the same command after changing
Input_W to have

usePrevious     true

caused a segmentation fault. The tail end of the output in that case was:

Reading detections from existing log file...
Reading from logfile : M33_Log.txt
Detection threshold used was 0.090884
Final object count = 19
Done.
Segmentation fault

I didn't attach the fits file because it was so large.

If he needs more information, we can let him log into bear directly.
Either he could send me his ssh key, or I could create an account for him.

David

Have attached the config.log and the input file separately.

[MatthewWhiting]

Problem lies with the calculation of the parameters in the getPreviousDetections() function. For some reason calcParams() is being called for objects of zero size.

A temporary fix for Mary et al is the patch file attached (which I've sent them), which only calls the calcParams() when the size is >0. I'll need to investigate a bit more closely why this was happening in the first place...

Leaving ticket open until that is resolved.

[MatthewWhiting]

Still problems reported. Seems to be that we were not picking up all the non-pixel lines in the log file.

I have added in extra tests to ignore the lines with "Detection #X", and made a few other getline statements etc more robust.

New patch file is attached.

[MatthewWhiting]

New version of the patch file solving segfault issue in existingDetections

[MatthewWhiting]

Changes committed to trunk in [583]. Will wait to make sure it all works before closing.

[MatthewWhiting]

Still not working, with segfaults happening.

The detections appear to be getting read in, but then a segfault is triggered when the WCSparams are calculated later on. Perhaps not everything is getting written? Or perhaps it's a problem with subsectioning?

[MatthewWhiting]

This was solved with the release of 1.1.8, so I think I'll close this ticket. (Problem lay with the use of subsections, but no problems have been reported since 1.1.8 was released).