> Subject: Re: Coordinate mapping conventions... > From: Steve Emmerson <steve@unidata.ucar.edu> > To: netcdfgroup@unidata.ucar.edu > Date: Thu, 22 Oct 92 12:52:03 MDT > In the above message, Joe Sirott wrote: > > >It seems to me that support for components and data to coordinate mapping > >can be solved by adding a small extension to the netCDF language. Why not > >add support for something analogous to C structures. For instance, we might > >have the following definition: > > > > float u(lat,lon,depth); > > float v(lat,lon,depth); > > float w(lat,lon,depth); > > struct velocity { u,v,w }; > > When performing I/O, would the above structure be dimensioned > [lat,lon,depth,3] or [3,lat,lon,depth]? > > - -------- > Steve Emmerson <steve@unidata.ucar.edu> > The structure definition that I've described is really just an alias; it only contains pointers to its components. In other words, it's just a way of binding together a group of names, just as the previous proposals used attributes to accomplish the same thing. The difference is that all of the work in resolving the component names would be done by the netCDF package, rather than by an application. No data is actually associated with the "velocity" variable; the real data is contained in the variables "u", "v", and "w". Joe Sirott