Peter,
> But, as I pointed out above. If these mappings do have a limited
> applicability (units, domain) then this should be discoverable/documented.
The applicability of a Mapping is described by the associated Frames. The
Frames says what units,domain, etc., are given to and returned by the
Mapping. A FrameSet combines the Mapping and the Frames to form a complete
description of the entire system. However, since I'm sure you already
knew all this, I think I must have mis-understood your point.
Whilst the document I've produced only talks about Mappings, and not
Frames, I am aware that there is a need for some equivalent to a Frame.
However, I'm leaving that to future discussion (as I mentioned, instances
of Arnold Rots schema are likely to serve the function of the Frame
class). I'll include some more in the document, though, about the need for
some equivalent to a Frame class, without going into the sort of detail I
do about Mappings.
> In AST the programmer reads the manual and uses a fixed set of Frames and
> Mappings to build a FrameSet, the correctness of this is established in
> the program. For a service based system it's not clear to me that you can
> so rigid.
Is there a problem with a service returning an equivalent to a FrameSet? I
do not understand why you call it "rigid". A simple Frame can be created
to describe a Cartesian coordinate system in any domain in any units with
any number of axes. Conventions are always needed, and no matter which
conventions are used, they only work if the data providers correctly
describe their data.
David
|