ctools

I am a bit clueless about the origin of this problem. It might as well correspond to a gammalib issue rather than ctools.

Stange, on my Mac this runs without any problem, and monitoring with top does not reveal any memory eater.

What branch are you using?

I get another interesting problem:

1278
Traceback (most recent call last):
  File "./test.py", line 15, in <module>
  File "/usr/local/gamma/lib/python2.6/site-packages/ctools.py", line 422, in __init__
RuntimeError: *** ERROR in GApplicationPars::read(std::string&): Unable to open parameter file "/Users/jurgen/pfiles/ctselect.par".

By the way, your problem is related to temporary files that are created by ctselect, see http://www.gnu.org/software/libc/manual/html_node/Temporary-Files.html for some information. I recognized that the files are eventually not closed properly in ctselect, hence this may explain the problem. I’ll do some check here and let you know. However, as I cannot reproduce the problem I’ll send you some patch to check this on your side.

You may put the following code in ctselect.cpp from line 631 on:

        // Get temporary file name
        //std::string tmpname = std::tmpnam(NULL);
        char tpl[]  = "ctselectXXXXXX";
        int  fileid = mkstemp(tpl);
        std::string tmpname(tpl);

        // Save FITS file to temporary file
        file.saveto(tmpname, true);

        // Load observation from temporary file
        obs->load_unbinned(tmpname);

        // Close temporary file
        close(fileid);

        // Remove temporary file
        std::remove(tmpname.c_str());

Thanks for your quick reply and your efforts.

What branch are you using?

I am on devel

You may put the following code in ctselect.cpp from line 631 on:

my mac compiler did not know the close-function. On linux, I still have the same problem after introducing the close-command.

I also got the problem in GApplicationPars::read() at some points during my tests. So you are right, it must be about to many opened files at the same time.

I will try with checking out a new, clean version of gammalib and ctools to ensure that I don’t use some obsolete library files.

Continue to track this down ...

thanks a lot

Indeed, I recognized that this does not fix the problem, just moved it twice as much iterations later on my system. I probably should not use mkstemp which opens a file ... I’ll keep you posted.

Mayer Michael wrote:

I will try with checking out a new, clean version of gammalib and ctools to ensure that I don’t use some obsolete library files.

just checked with a new version - the problem remains.
However, using std::tmpnam instead of mkstemp, the problem disappears all of a sudden.

Should we change back to std::tmpnam or is it considered unsafe compared to mkstemp?