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

[netCDF #FFY-177157]: NetCDF-4 and 64 bit dimensions

Rob,

> The only sticky area is when a 64 bit machine (AMD opteron, intel
> core2) creates a NC_NETCDF4 file with big dimensions, then sends that
> file to a collaborator with a 32 bit machine. In that scenario, you
> just give an error, essentially saying "try again on a 64 bit
> machine", right?

This uncovered a bug in our netCDF-4 code, which Ed just fixed in the
snapshot, so now dimension sizes can be full 64-bit integers in
netCDF-4/HDF5 files.  However, it also uncovered a gap in our handling
of big dimensions.  Specifically, it's currently not possible to 
detect a big dimension on a 32-bit platform, instead when you try to
inquire about the dimension size with nc_inq_dimlen(), you just get
the dimension length truncated to the 32-bit size_t.  We obviously 
have to do something about this, and I have proposed returning the 
error code NC_EDIMSIZE, which until now has only been returned when 
defining a dimension with an invalid size, such as a negative number.

We have to determine the details of how HDF5 handles big dimensions 
on 32-bit platforms in order to make netCDF-4 do the right thing.

I'll let you know how we resolve this, but thanks making us aware of
the need for considering these issues.

--Russ

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



Ticket Details
===================
Ticket ID: FFY-177157
Department: Support netCDF
Priority: Normal
Status: Closed