Re: [guppi-list] Guppi as a mime-type handler

[Magnus Hagdorn <Magnus Hagdorn glg ed ac uk>]

On Wed, 31 Jan 2001, Neil J Pilgrim wrote:

[licence stuff]
i was wondering about the netcdf licence, that was one reason why i wasn't
doing anything.


> [HDF5 as alternative file-format]
> 
> Having looked at the docs for both briefly, HDF5 looks much more complex
> and overkill for 'just' data, but there's no reason it couldn't be added
> too :)

right some random musings...

well, i think it doesn't really matter in terms of effort. hdf5 supports
netcdf data structures. actually i was thinking about moving my code from
netcdf to hdf5 since you get parallel i/o. the only bugger is that f90
interface is still in beta.

anyway, the thing about supporting either netcdf/hdf5 that it would
require a lot of gui work to make it useful. features should include
- selecting variables
- selecting hyperslabs
- selecting subsections of the data

actually, i seem to remember that there is a python interface to
netcdf. what i'd really like to have is some program with which i can read
netcdf/hdf5 files. do calculations on it and display the data. 

in terms of actually using the library, that is very easy. essentially all
you need to do is:
- open file
- read dimensions, variables, metadata
- read selected variables and ranges
- close file
there are library functions for all these tasks

i wee while back i tried a netcdf importer for avs, i can't remember what
it was called. the really nice feature of that was that it was network
aware, ie the data could reside on some big server, you select your
variables and ranges and request that data from the server to be displayed
on your local machine. i dunno, maybe that sort of thing is not required
anymore since your average workstation can probably handle quite a bit.

you can always do the conversion from netcdf/hdf5 to something simple,
like comma separated file. but then i really would want to avoid that
since i am talking quite big files of the order of 100s mb to gb. ascii
data files are just out of question. also you'd loose the flexibility to
easily select subsections/hyperslabs.

right, i think i will print out the hdf5 documentation to see how hard it
would be to support that (or at least the gridded data features) rather
than netcdf.

cheers
magi

-------------
Magnus Hagdorn
Department of Geology and Geophysics
The University of Edinburgh