[Uml-devel] Thoughts on Umbrello 1.3

[Andrew Sutton]

On Tuesday 20 January 2004 11:02 am, Jonathan Riddell wrote:

> > 1. I have about 3-4k lines of code of a "project design document
> > generator"
 for Umbrello (alias SoDA for Umbrello) into OpenOffice.org
> > writer format (KWord should use this in future too). Needs polishing and
> > merging with the current codebase and then it is going to be your choice
> > whether to include it or not.
>
> 
> Sounds good.

i agree. cool.

> > 3. Another thing to discuss is why do certain diagrams use only
> > UMLWidgets
 and why do others incorporate UMLObjects too. Quite confusing
> > without the ability of mindreading combined with average intelligence
> > (referring to myself).
>
> 
> It's a good question.  I'm not quite sure if there are theoretical
> reasons for states and activities not to be represented by UMLObject
> and be in the list view.  

it's a sign of premature aging. the original concept was lost, then the 
program was modified. now, nobody knows.

> > 4. When Umbrello2 was announced, a few times I heard something about
> > enabling Umbrello2 to use a database => enabling multiple users working
> > on
 the same project. Now that sounds very interesting to me (I tend to
> > have a liking in databases, especially PostgreSQL, because I can write so
> > nice stored procedures in it and have very thin clients) so I think I
> > would sit on that and do it. Not earlier than mid February, though.
>
> 
> It would be a big task.  As far as I know it hasn't been done before
> anywhere so chances are there would be numberous problems that you
> come accross.  And I don't know how useful it would be.  The database
> would have to be updated after every modification and any other
> Umbrello instances would have to reload it from the database.

i can speak on this since my work (the open modeling framework - omf) is based 
on the data engine of what was once umbrello2... here's a quick run down of 
how	 it works so far:

the omf implements a persistence layer. any objects you create or modify 
during the course of process execution also modify a DOM tree that sits 
underneath the "in-memory" model. i basically did this because it makes 
writing files about three lines of code. however, i'm not really too happy 
with how the design (it's hidden, but its heinous). what i'd like to do 
eventually is take the persistence layer and abstract it into an actual 
class. This would allow you to define new persistence layers and associate 
them with a model at run time. With this design, modifications to objects in 
the model wouldn't modify a dom tree, but potentially a database. see - 
persistence. it's still around :)

anyway, the reflective layer of the omf is designed perfectly for 
object-relational databases. one of my other sub-projects might be to build 
sql drivers for xmi files. it's also not that hard to do.
 
> When I last looked at this a long time ago KDevelop's code import
> didn't have a defined interface nor was the code in a separate library
> so it would have needed to copy each language individually (as
> happened with C++).  I suspect this may have changed but I havn't
> looked at it.
> 
> Then there is the question of adopting KDevelop's object model which
> would mean a lot of changes to Umbrello and probably to KDevelop so my
> hunch is that it's not worth it.

i think between the two groups, we could probably harmonize the interfaces to 
the parsers and use them interchangeably. it would be nice to make them (the 
parsers) completely independant of kdevelop or umbrello.

> > 6. ArgoUML has something they call "cognitive support" or something
> > similar.
 I think this a noble name for "error checking/validation" like
> > "there is no contructor in this class" and other goodies. I think it were
> > nice to have this feature with Umbrello, too.
>
> 
> I find this incredibly annoying and not very helpful at all (but maybe
> I find it not to be helpful just because the user interface on that
> program is horrible Swing).

i think its a load of crap - and annoying.

>
> > 8. Brainstorming goes on: a KPart for Konqueror to view the contents of
> > an
 .xmi file (but then this makes no great sense)
>
> 
> That would be fun but would people use it?  It would mean a read-only
> mode for Umbrello and then the problem is that .xmi files contain
> several diagrams and you'd need a way of deciding which one to show.

how about a kpart that took an xmi file and generated a complete doxygen-style 
web interface complete with diagrams, documentation, class listings, and 
other stuff. that would be f-ing cool.

andy