[Uml-devel] JavaCodeGenerator

[Brian Thomas]

On Friday 28 February 2003 07:23 pm, JR wrote:
> >       - multiplicity should be limited to a carrier class, not
> > transported about as a String that has to
> >       be interpreted. IF someone uses older standards to specify
> > multiplicity, the code generator could
> >       fail.
>
> A good idea.  What older standards?

	Well, I know that at one point multiplicity could be specified using plus sign
	(argh.. I might be showing my age here, I think that happend in the pre-UML 
	days of Booch diagramming method, way back in the early 90's/late 80's).
	So I guess the greatest danger these days is that the user specifies the multiplicity
	incorrectly and the code cant recover the correct meaning.

> > that a given association obeys an
> >       interface or class. This is in part because umbrello wont let
> > you diagram these type of associations.
> 
> That would be a nice feature but quite tricky to do.

	I agree, but ultimately its necessary to do any kind of semi-sophisticated code generation correctly
	(at least in the Java/C++ world). I also wonder how to incorporate correct generation for
	a number of other language-specific stuff like C++'s virtual methods, when to/when not to
	make a method inline, etc. But some of this stuff isnt specifiable in UML anyways and would
	have to be toggled on/off within an C++ extension of the code generation interface. Perhaps interfaces
	for associations can be temporarily treated this way for now.

> >       - description (comments) on association roles not carried
> > through. This is in part a failing of the association
> >       model in umbrello.
> 
> But shouldn't be too hard to add should it?

	Agreed. I could add this when you tell me you are done hacking on the association class.


> >        startline = "\n"+indent; // using UNIX newLine standard.. bad

> Why is that bad, \n is the one true line ending.

	Heh. Yeah, I know you have tongue in cheek here, but it's still true. From what I understand, Qt
	(and thus Umbrello) should theoretically compile on Windoze platforms (as well as Apple) so we
	dont want to create just  *NIX-specific code with the tool. There _should be_ a facility within Qt to 
	specify the correct platform end-line string (e.g. "/n" on *NIX, "/r/n" on Win and "/r" on Apple platforms), 
	but I couldnt find it. Any Qt experts out there with this knowledge??


> >        // do people REALLY want these comments? Hmm.
> 
> That being the comments marking public operations, private operations etc.
> I'm not honestly sure if they do.

	An older programmer once told me that code comments are like sex...its better to have the bad than to 
	have none at all :)


													-b.t.