Hi Ed,
> > Ed,
> >
> > Bear in mind that the "names" are on links, not on the dataset.
>
> Yes, I do understand this.
>
> However, if I don't use references, then I can treat the name of
> objects as absolute, correct?
>
> Let me post a more complex question.
>
> Suppose I create a dataset called "Bob" in the root group, and then create a
> reference to it called "Joe".
>
> Then I if I use H5Gmove to rename "Bob" to "Sally", I still have a
> reference named "Joe" to the original dataset.
>
> Is that correct?
Sorta. :-) The reference will still point to the correct dataset if you
perform the operations above. However, references don't have names, they are
"anonymous" links to objects.
> I believe that Quincey has some references in the dimscales
> attributes, which are named for the dataset they point to. If I rename
> the dimension scale, then I would not expect the names of the
> dimensions in the DIMENSION_LIST attribute to change.
>
> HDF5 "tst_h_dimscales.h5" {
> GROUP "/" {
> GROUP "Adam" {
> GROUP "Able" {
> DATASET "Billy_goat_gruff" {
> DATATYPE H5T_IEEE_F32LE
> DATASPACE SIMPLE { ( 0, 0, 3 ) / ( H5S_UNLIMITED, H5S_UNLIMITED,
> 3 ) }
> DATA {
> }
> ATTRIBUTE "DIMENSION_LIST" {
> DATATYPE H5T_VLEN { H5T_REFERENCE}
> DATASPACE SIMPLE { ( 3 ) / ( 3 ) }
> DATA {
> (0): (DATASET 3576 /Adam/Time ),
> (1): (DATASET 3848 /Adam/Smelliness ),
> (2): (DATASET 4120 /Adam/Distance )
> }
> }
> }
> }
>
>
> But when I use dimension scales, I use the fileid and objid, not the
> name, so netcdf won't be confused.
Perhaps Bob can cross-check me here, but I think this should be fine.
Quincey
