[Uml-devel] Refactoring the code base -- what should be done?

[Andrew Sutton]

On Thursday 23 September 2004 02:54 am, Pekka Jääskeläinen wrote:
> > - Look at Andy's OMF2 implementation
> >   -  He doesn't use Qt, just the C++ STL classes (std::string etc.)
> >      What shall we do? Change Umbrello or change the OMF?
>
> I have no idea what the OMF actually does. Andy, do you have some
> documentation for it? For example, Doxygen API docs? Anyways, if it
> doesn't handle anything about visualization, then it's purely model
> stuff and I think we shouldn't even have a smell of Qt in that part.

the omf is a library that implements (or at least attempts to) the full OMF 
modeling stack. basically the MOF + UML and some way of interconnecting the 
two (reflection). unless you consider XMI to be visual, it doesn't do 
anything graphical. a previous version used Qt. the current version does not 
for portability reasons.

i'm slowly working on a programmer's guide, but it's not coming along very 
quickly. there's doxygen docs on the web site:

http://www.sdml.info/projects/omf

> Would it make sense to go the other way around with this? I mean not
> trying to think how Umbrello could be changed, but think about how the
> high level architecture *should* look like, and then think which parts
> of the current Umbrello code base could be used in the new design?
>
> Basically it could mean that the high level architecture is modeled
> first and everyone could introduce their ideas. After most people are
> satisifed with it, proceed with implementation (which can be partly
> copy-pasted from Umbrello).

we sort of tried this... umbrello2. there just wasn't enough interest to 
actually rebuild the whole thing.

> Andrew Sutton, do you have some kind of architecture design
> documentation (even a diagram) of the Umbrello 2 you seemed to
> have started some time ago (but now halted)? Could the work on
> that be continued?

there's an xmi file somewhere on the website. of course, it was built using an 
old version of umbrello's xmi format, so it might not be viewable. i would 
seriously steer clear of that code base. it was a good experiment, but not a 
good starting point.

andrew sutton
asutton at cs.kent.edu