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

[netCDF #CKA-677320]: NF_CREATE crash

This archive contains answers to questions sent to Unidata support through mid-2025. Note that the archive is no longer being updated. We provide the archive for reference; many of the answers presented here remain technically correct, even if somewhat outdated. For the most up-to-date information on the use of NSF Unidata software and data services, please consult the Software Documentation first.

  • Subject: [netCDF #CKA-677320]: NF_CREATE crash
  • Date: Wed, 23 Feb 2011 18:47:38 -0700
> Hi Ansley,
> 
> > Using the source from this URL,
> >
> > ftp://ftp.unidata.ucar.edu/pub/netcdf/netcdf-test.tar.gz
> >
> > I'm seeing a problem with NF_CREATE. The call is this, with
> > cdftype=NF_CLASSIC_MODEL and with clobber_mode set to either
> > NF_NOCLOBBER or NF_CLOBBER. The call is made by program ferret_c, and
> > its result is below:
> >
> > cdfstat = NF_CREATE(cdfname, OR(clobber_mode,cdftype), cdfid)
> >
> > ferret_c: nc4file.c:395: NC4_create: Assertion `0' failed.
> > Abort
> >
> >
> > Setting cdftype to NF_NETCDF4 and making the same call lets me
> > successfully create and write to the file.
> 
> I can't reproduce this problem using gfortran.  What happens when you
> use the function IOR() instead of OR() ?, as in:
> 

Howdy Russ and Ansley!

I can reproduce this in C.

The problem is that Ansley is using NC_CLASSIC_MODEL, but not using netCDF-4! 
It didn't occur to us that anyone would do that. ;-(

So this caused two problems. The dispatch layer thought that it should use the 
netCDF-4 dispatch table (because if you are not using netCDF-4, then why use 
CLASSIC_MODEL?), and then the netCDF-4 version of the create function got 
called, and asserted itself because it believed that it should only be called 
with NC_NETCDF4 set in the mode flag.

Here we see a lesson in why asserts can be much worse than simply returning an 
error code. By returning the generic NC_EINVAL the nc_create function would hit 
the nail on the head! The input parameters are invalid.

But, of course, we are handling things is a very sloppy manner and that can be 
fixed. I have changed the code so that we do not assume netCDF-4 dispatch table 
when CLASSIC_MODEL is used. In fact, we ignore CLASSIC_MODEL if a classic 
netCDF file is being opened. This is now tested in libsrc4/tst_files.c. So your 
code should work as is, even though there is no reason in the world why you 
should set NC_CLASSIC_MODEL when creating a classic format file (because what 
else can it be?).

But that's OK. We want our users to be happy, and if setting this flag brings 
even a tiny bit of happiness to you, Ansley, I am delighted. ;-)

I have checked in this code and regenerated the daily snapshot with it. This 
also includes the fix that Dennis sent you as a branch build a few days ago 
(the vars fix).

ftp://ftp.unidata.ucar.edu/pub/netcdf/snapshot/netcdf-4-daily.tar.gz

Thanks for finding this Ansley!

Ed

Ticket Details
===================
Ticket ID: CKA-677320
Department: Support netCDF
Priority: Normal
Status: Closed