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

[netCDF #MJR-379092]: more information about segmentation fault in ncdump when dumping HDF5 string

Kent,

> Fixed-size HDF5 string is very commonly used inside HDF5 files that users
> are interested in making netCDF-4 to access.
> I happened to find an old  report regarding this issue. The student did
> some investigations and provided a patch for the netcdf-4 package.
> Hopefully this can help you guys fix the bug.
> 
> Check
> ftp://ftp.hdfgroup.uiuc.edu/pub/outgoing/ymuqun/temp/netcdf_seg_fault/old-report/
> 
> 
> especially README can help you get started.

Yes, that's very helpful, although the fix has become more complex since we 
added a 
dispatch layer.  Nevertheless, I'll commit to getting the fix into the 4.2.1 
release,
which we're aiming to make available in the next two or three weeks.

Another complication is our inability to include a simple test that just uses 
the
netCDF-4 API, because our data model doesn't support HDF5's fixed-length string 
type.  Nevertheless, we may be able to just include a simple HDF5 file that 
uses 
it to make sure we can read it.

Thanks!

--Russ

> address@hidden> wrote:
> 
> > Hi Kent,
> >
> > Thanks for all your work on isolating the bug.
> >
> > I think the problem may be that netCDF-4 has no such type
> > as a "fixed length string", it only has the "string" type,
> > which is always variable length, and an "array of char"
> > type, which was in the netCDF-3 classic model, and for
> > which a dimension must be defined and named to identify
> > the langth of the char array.
> >
> > When Ed added support for reading HDF5 files that were not
> > netCDF-4 files, he may have not realized fixed-length chars
> > were a new type that the netCDF-4 data model does not support.
> > They either have to be converted to variable length strings
> > (perhaps with a hidden attribute indicating they are fixed
> > length) or as array of char, with a new dimension defined
> > that does not correspond to an associated HDF5 dataspace.
> >
> > Or, we have to add support for a new type, which is a lot of
> > work.
> >
> > The netCDF-4 data model supports fixed-size arrays as members
> > of structs, without having to define associated named dimensions.
> > But as far as I understand, it does not support fixed-size arrays
> > that have no associated, named dimensions.
> >
> > I'll have to look at this a little more carefully to verify that
> > the problem is in the netCDF-4 library layer rather than ncdump,
> > but if so, the fix will have to be in the library.
> >
> > --Russ
> >
> > Russ Rew                                         UCAR Unidata Program
> > address@hidden                      http://www.unidata.ucar.edu
> >
> >
> >
> > Ticket Details
> > ===================
> > Ticket ID: MJR-379092
> > Department: Support netCDF
> > Priority: Normal
> > Status: Closed
> >
> >
> 
> 
> --
> Kent Yang The HDF Group
> 1800 So. Oak St., Suite 203, Champaign, IL 61820
> 217.531.6107
> http://hdfgroup.org http://hdfeos.org
> 
> 

Russ Rew                                         UCAR Unidata Program
address@hidden                      http://www.unidata.ucar.edu



Ticket Details
===================
Ticket ID: MJR-379092
Department: Support netCDF
Priority: High
Status: Closed