Proposal to split out the "members" from coreQ (Was: Re: Philosophy of basic Q)

[Pierre Didelon]

Hi David,

David Berry wrote:
> Pierre,
> 
> 
>>>I always thought we were way ahead of ourselves in diving into
>>>serialisations before we had firm agreement on the data model which we
>>>were supposed to be serialising!
>>>
>>
>>Well, speaking of data structures (data member inside class) certainly,
>>but concerning interfaces your point is not so straigthforward,
>>ans serialisation is important beacause it is, then, the only way (?sure)
>>to exchange "object" (quantity) as a whole, in one "transaction".
> 
> 
> I expect that there will be more than one way to serialise a Quantity
> object. The obvious example is FITS - a FITS image is a serialisation
> of a Quantity, but it has no relation at all to the XML serialisation in
> the doc. VOTABLE is another possible example - we presumably want to
> see a VOTABLE as a way of serialising one or more Quantities. And there
> could be other forms of serialisation. 

Well! As you can guess from my previous mail, I think
(but certainly it is only _my_ opinion) that these kind of quantity usage
is not interesting, or at least not as a first step/trial/....
I would feel really better if discussion could be limited
to very simpl quantities example usage : exchange of one or fews parameters.

> The only point in us including any
> suggested serialisation in the doc, is to ensure that we have at least one
> serialisation which is loss-less. But it willbe easier for everyone if
> this suggested serialisation has a *clear* and *one-to-one* connection to
> the interface.

agree
> 
> What I hoped is that the community would argue about the actual *data
> model* as described in the interface, until we have agreed on the data
> model, and *then* we would decide on a preferred loss-less serialisation.
> 
> I believe an interface is the correct way to describe a data model, rather
> than a serialisation. 

more than "correct", I hoped (before the mail storm) that it would be
the easiest path to reach agreement. ;-)

>An interface says "in order for an object to be
> considered an instance of this data model, it should be possible to do
> this and this and this with it - how it is stored is immaterial". With
> this information, we can write code to process Quantities, no matter how
> they are serialised.

But at least you need in the interface a method
public string serialise()
or
public String toString()
or whatever name method, but a one which can propose a way to exchange
a whole object in once.


-- 
Pierre 
--------------------------------------------------------------------------
DIDELON :@: pdidelon_at_cea.fr        Phone : 33 (0)1 69 08 58 89
CEA SACLAY - Service d'Astrophysique  91191 Gif-Sur-Yvette Cedex
--------------------------------------------------------------------------