[Uml-devel] Re: karbon14 and uml

[Andrew Sutton]

On Tuesday 15 April 2003 9:41 am, Dirk Schönberger wrote:
> But I don't think you look at the _same_ thing from different sides. I
> think you build several represenations/documents/whatever from a subset of
> the complete model. Some parts of the model can be shared, while other are
> view/document specific. I think in UML the distinct parts are more
> important, unlike e.g. in Konqueror, where you see the same file entries,
> but one time as icons, one time as list, one time as tree and so on.

right and wrong. it's sort of the same thing from different sides. its all the 
same architectural description, but we have views corresponding to the 
interests or concerns of different stakeholders. a programmer isn't going to 
be as interested in the use case view as he would be in the logical view. 
but, yeah. you're basically right.

> These are relations, which should be visualized as some kind of (hyper-)
> link. I agree.

yup. i've got to do some thinking about this for my paper. it's actually kind 
of a non-trivial problem. i suspect that in part, we can construct these 
links via standard uml relationships, but we might also support 
user-definable links. ouch.

> a) I didn't had the idea, Andrew Sutton started the discussion.
> b) Neither, nor. What I would like to see is that "document management" is
> handled via a file system model. I.e. the different diagrams in their
> respective in a XMI document (I believe they are stored there?) can be
> accessed in a file system view like Konqueror. I.e. I can enter a URL like
> the following in Konqueror URL bar

the idea is that there's lots of room for improvement. i've basically been 
breaking down uml apps and trying to figure out what they do right and wrong. 
there is no REALLY good uml application. also, what do we mean when we say 
uml appliation? if its just an application that uses uml somewhere along the 
lines, that's cool. these ideas apply to that too because everythings based 
in a single xmi file. anyways, i deviate.

anyways, looking around, i see that there are a bunch of big weaknesses in 
most uml apps.
1. lack of architectural description/framework support. sure we have the use 
case view and the logical view, but who really knows what that means. the 
semantics have been lost and most user's don't really know where it comes 
from.
2. lack of good document management support. diagrams are documents, why treat 
them like model objects. they're different.
3. shitty layout algorithms. nobody does a really good job laying out 
diagrams. this is really a tough thing to do, and i don't want to do it, but 
i'm going to try.
4. ineffective use of the operating environment - basically, you should be 
able to maximize the usage of screen space to get the most out of the 
appliation. look at the gimp. windows everywhere.

my work right now is basically to address these issues in terms of 
visualizations - and i'm coming up with some interesting stuff. like using 
the fs browser to manage diagrams. the kio-slave is an extension of that, 
providing some kind of read-only access to xmi files. it would be better if 
it were dynamically generated web pages for an architecture browser. that 
would be pretty sweet.

andy