[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Problem with netcdf library on IRIX 4.0.5F

> Organization: Scripps Institute of Oceanography
> Keywords: 199408230054.AA18398 netCDF SGI IRIX

Hi Adam,

> Hello. I have what might possibly be a bug in netcdf 2.3.
> I am using the library under IRIX 4.0.5F
> 
> I have had a consistent problem where at odd intervals certain nc
> operations will crash during a call to malloc.  This occurs most often
> in ncopen. This problem crops up unexpectedly and seemingly at 
> random(except for when we have site reviews!).  Neither I nor
> my co-workers have been able to discover why this happens.
> One typical instance where this can happen is when a driver program that
> calls libraries which in turn use netcdf is modified.  Although these
> modifications do not use netcdf functions or seem to have any memory
> corrupting properties, the malloc problem sometimes appears.
> 
> I would very much appreciate it if you can shed any light on this
> problem.

Sorry, but we have never seen the symptoms you describe and have gotten no
reports of similar problems from other users, either on SGI/IRIX or other
platforms.  It sounds like memory heap corruption due to either a program
writing beyond its malloc'ed space, or due to freeing memory space that was
never malloced.  Since the malloc heap is shared by functions that call
netCDF as well as the netCDF library, such a problem might well show up as a
crash when malloc is called from within the netCDF library even though the
actual problem occurs earlier outside of the netCDF library when the heap
is first corrupted.

One way to track down such problems is to use development tools such as
ObjectCenter or Purify that check at run-time all malloc and free calls and
all memory references.  If you have no access to such tools, you may have to
try using a freely-available debugging malloc package such as Brandon
Allbery's malloc-debug, available from comp.sources.misc archives.

We can't help diagnose this problem without a small test case that
demonstrates the problem.  We do have access to ObjectCenter and Purify (we
use them in developing and testing netCDF), so if you can develop a small
test case that reproduces the symptoms, we could use those tools to locate
the problem.

--
Russ Rew                                                   UCAR Unidata Program
address@hidden                                             P.O. Box 3000
http://www.unidata.ucar.edu/                             Boulder, CO 80307-3000