intermediate rep

From: David R. Karger <karger_at_mit.edu>
Date: Fri, 20 May 2005 01:28:10 -0400

Here is a stefano snippet:


  The workflow I have in mind is something like this:

    1) get a model
    2) apply fresnel selection and get a submodel
    3) apply fresnel graph->tree operations (stuff like contentBefore,
    etc.., the 'non-presentational' part of our current styles)
    4) obtain the tree
    [if you don't care about view portability stop here]
    5) apply fresnel styles to the tree
    6) interpret the results

What is "non-presentational" about the graph-tree step 3? Isn't the
tree encoding presentation knowledge, in that it is fixing various
sibling relationships that will affect the way info is presented?

Continuing the exploration of "selection" versus "style", perhaps the
implication here is that there is something display independent about
the notion of a group? ie these property/values should be grouped?
With recursion---ie these properties should be grouped, and the lens
that selects information about the _value_ of the property will itself
do some grouping, that should not be broken up as we display the
information?

How does this relate to the "box model"? Generally we think of the
box model as presentational, but perhaps we should instead be thinking
about a "tree model" that is not presentational at all. Rather, we
recognize that many presentations of information are actually
presentations showing a number of different objects in their relation
to each other, and that one particular common way to present this
information is hierarchically, in that the items to which I am related
show up as children of me in whatever presentation is crafted. This
still says nothing about how the tree is presented, but could be
represented in xml because it is hierarchical. It is very natural,
from this perspective, for each node in the tree to be labelled with
which resource it represents.

Next question is, if we construct this group tree as above, is it now
too late to apply styles to the group tree itself? Does the styling
have to be "interleaved" with the construction of the group tree? I
am not sure there is enough local info in the group tree to let the
styles know what to do at various points in it. But perhaps the way
around this is to say that the lenses are asserting a style on each
object and/or property that describes the _logical roles_ of that
property. e.g. we might have a style like "elaboration" that says
that "the information in this node/subtree is an elaboration of
information found elsewhere in the submodel". This could be
interpreted by certain displays with limited space to leave that
subtree hidden from display unless the user clicked on some kind of
expander button to show it.
Received on Fri May 20 2005 - 05:26:42 EDT