Re: Some OOP questions

[Judd Storrs]

On Thu, Feb 9, 2012 at 12:19 PM, CdeMills <address@hidden> wrote:
> This is exactly the purpose of the dataframe package :-)

I guess I think of dataframe as more like R's data.frame for solving
the problem where you have heterogeneous data that's structured
somewhat as rows and columns.

What I'm dealing with are 3D and 4D complex/real images. I just want
to tack some metadata on them to reduce the amount of
housekeeping/boilerplate code that gets spread everywhere (things like
real world positions, dimensions, sample identifiers, timestamps).
I've usually done it two ways:

    [image,info] = process(image,info,args);

But too often you start out implementing a processing function (or you
get it from someone else) that doesn't use meta data and you start a
function like this:

    image = process(image);

and you get a few calls deep and you want to extend things but you
need to peek at some metadata and then you have to go update all the
function interfaces to add the meta data (and don't even think about
passing function handles around). So at some point I started doing it
this way:

    image.data = ... ;
    image.meta = struct(...) ;
    image = process(image);

but then you have to rewrite your functions to use image.data instead
of image (really not too huge a problem, but still annoying). It's
just so tempting to somehow make use of the unused "." indexing on the
numeric types somehow, because that syntax currently does absolutely
nothing. And the more I thought about it, I kept thinking of different
ways to use meta data.

In my mind the two approaches are (1) add it as a core functionality
for all numeric types or (2) fix OOP subclassing of builtin types so a
subclass can easily add it. I'm leaning towards (2) as a better
approach. I think it can be mocked up using @class style OOP. Anyway,
I think it's a little different from dataframe.


--judd