Re: 1.8 file format

[Quincey Koziol <koziol@xxxxxxxxxxxx>]

Hi John,

On Jul 21, 2007, at 11:33 AM, John Caron wrote:

> Hi Quincey:
>
> I am working my way through various tests of HDF 1.6 features,  
> hoping that 1.8 docs will be ready (or ready to work on) when Im  
> done. ;^)

	Sure.  Vailin's finished up the major changes to the h5stat tool and  
is wrapping up some directory rearrangement and extra testing this  
week.  Then she'll be starting on updating the file format  
specification for the 1.8 changes (with me).

> At the moment, im seeing a chunked data variable whose btree has 0  
> entries. I assume that it was never written, and I should return  
> fill values? It may or may not be notable thats its chunked size is  
> equal to its dataspace size, and its in an Aura file.

	Yes, any missing chunks in a B-tree (including empty B-trees) were  
never written and should return fill values.  If there is no fill  
value message, then all zero bytes for the datatype elements are  
assumed.

	Quincey


> Quincey Koziol wrote:
> > Hi John,
> > On Jul 13, 2007, at 11:43 AM, John Caron wrote:
> > > Hi Quincey:
> > >
> > > I'm just starting to look at the H5Fsuper.c (its been a while  
> > > since i coded in C).
> > >
> > > Meanwhile i have another question on a different topic:
> > >
> > > The B-tree nodes I assume are sorted on the keys. Is this a  
> > > simple sort on the byte values, or something else? For example,  
> > > are nodes of chunked data sorted on chunk size, then filter mask,  
> > > then chunk offset? Or do you just use the chunk offset? When  
> > > sorting the chunk offset, can i assume you sort on offset for  
> > > slowest dimension (0), then offset for dimension 1, etc
> >     Yes, the B-trees are sorted on the keys.  It's not a simple  
> > sort on the byte values though, each "class" of B-tree has a  
> > comparison function.  For chunks we sort only on chunk offset and  
> > it's as you describe above: first on offset in the slowest  
> > dimension, then toward the fastest dimension.
> >     Quincey
>
> ====================================================================== 
> =========
> To unsubscribe netcdf-hdf, visit:
> http://www.unidata.ucar.edu/mailing-list-delete-form.html
> ====================================================================== 
> =========

Attachment: smime.p7s
Description: S/MIME cryptographic signature