[QUANTITY] The discussion so far

[David Berry]

Pierre,

> Waouh! Big work, impresive summary.

Agreed!

> Array-1D OK,  ND flattened in 1D (like pat proposal) OK.
> Otherwise we are no longer speaking of simple quantity but of
> complex dataset, data collections, images or whatever.
> I agree with doug email about hierachical data structure
> (http://www.ivoa.net/forum/dm/0219.htm) and may be we have to agree
> and try to define such a common (VO-DM) hierchy of data structure.

Is there really much difference between a 1-D array and an N-D array? Is
it not just a case of the number of indices you need to locate an
element within the array? If one is OK, why not the other? Allowing N-D
does not really add much complication in my opinion, and does allow the
structure (what ever we call it) to be of much wider use. It also removes
the need for a another layer of structure, which would be needed to
introduce N-D arrays if this basic structure was restricted to 1-D.


> I am in favor of having very simple class, firstly due to a general aspect,
> that very simple class are very general and can be reuse more oftenly
> without too much overhead. When you only need a value and his unit
> would you be inclined to use (in your code apllication or even in your
> appl definition) a class which contains a lot of other things you
> really don't care about? I feel that scientist (even computer ones) often use ,
> perhaps implictly, the Occam blade to throw away and suppress superfluity.

The sort of higher level N-D structure which I proposed ("DataContainer"
in Jonathan's message - but I'm not comitted to any particular name) can
be as simple as required since all components are optional. Ray also made
this point in his requirements list. Defining a class which includes
errors, quality, units, label, wcs, etc, is no burden in situations
where simple structures are more appropriate, since you can simply omit
the components which are not relevant. Let's have one structure which is
exandable rather than lots of little rigid structures.

In my view, the "basic building block" structure we are trying to define
should be a way of encapulating a set of homogenous related data values,
with sufficient metadata to allow it to be interpreted in isolation from
its parent structure (although obviously as Ray pointed out, when you
extract a structure from its parent you must expect to loose knowledge of
its significance and meaning within the higher level structure). All the
components are optional, so you can have something as simple as a UCD and
a Units description if necessary. The availability of the extra
components means that such a basic building block could be used within a
larger "observation" or "dataset" to hold an image, a spectrum, a spectral
bandpass, etc.



David