Coordinate subscripting bug crashes NCL

From: Dave Allured (dave.allured AT noaa.gov)
Date: Thu Sep 22 2005 - 17:28:00 MDT


Hello. I am getting mysterious behavior from NCL. In a fairly simple
script, I read an input array from a netcdf file. I am subsetting the
array on input, reordering dimensions, and using coordinate
subscripting, all in the same statement:

  pcp = f->precip({lat|lat1:lat2},{time|:},{lon|lon1+360.:lon2+360.})

Note that the time subscript uses default range notation. According to
the documentation, "this should work".

The script crashes consistently on this statement, on two different
platforms: Solaris 5.8 on Sun4U, and Mac OS X version 10.3.9 on Mac G5.
Crash messages are only "Segmentation Fault" or "Bus error" (Mac). We
have the current NCL installed, version 4.2.0.a032.

Any of these are valid workarounds that allow the script to run:

  pcp = f->precip({lat|lat1:lat2},time|:,{lon|lon1+360.:lon2+360.})
  pcp = f->precip({lat|lat1:lat2},{time|0:364},{lon|lon1+360.:lon2+360.})
  pcp = f->precip({lat|lat1:lat2},{time|t1:t2},{lon|lon1+360.:lon2+360.})

Now here's the weird part.

1. This is described as a known bug that was fixed in the current
release notes:

   http://www.ncl.ucar.edu/new_release.shtml (for version 4.2.0.a032)

   * Another coordinate subscripting bug

      If you tried to use coordinate subscripting without actually using
      a subscript value, as in:

           buf2 = buf({lev|:},{lon|:},{lat|30:50})

      it would sometimes cause a core dump.

2. This script ran normally for me many times a couple months ago, with
the same version of NCL. Now, the identical script crashes consistently.

3. There was at least one other strange change in NCL behavior, having
to do with changes in the way that data types could be coerced at
certain places in the same script (I think).

This makes me suspect that my NCL installation may have somehow
changed. But a check shows that we have apparently been using the same
NCL executable and the same dynamic libraries for the entire time.

It is possible that somehow the last bug fix did not completely correct
the subscripting problem, and that it is still intermittent in nature.

Has anyone else seen this behavior? Do you know of a possible
installation or environment issue that could cause this sort of shift in
behavior?

--Dave A.
CDC/NOAA/CIRES

_______________________________________________
ncl-talk mailing list
ncl-talk@ucar.edu
http://mailman.ucar.edu/mailman/listinfo/ncl-talk



This archive was generated by hypermail 2b29 : Fri Sep 23 2005 - 16:13:29 MDT