Path: utzoo!attcan!uunet!samsung!brutus.cs.uiuc.edu!psuvax1!psuvm!uh2
From: UH2@psuvm.psu.edu (Lee Sailer)
Newsgroups: comp.software-eng
Subject: Re: CASE tools, esp. Software through Pictures.
Message-ID: <90043.114236UH2@PSUVM.BITNET>
Date: 12 Feb 90 16:42:36 GMT
References: <1990Feb6.215234.6561@is.uu.no> <1378@island.uu.net>
Organization: Penn State University
Lines: 47

In article <1378@island.uu.net>, dav@island.uu.net (David McClure) says:
>
>
>It seems that one of the most basic ideas in structured design and analysis
>is the ability to "push down" or "pop up" in order to encompass higher or
>lower levels of detail and complexity.  IMO, this is an extremely important
>concept that mirrors how the human brain thinks about design.  Implementation
>of this idea is currently very primitive: in the DFD's that I have seen,
>being able to group and push or pop a collection of objects to another level
>is done by cut and paste (if at all), and NOT integrated to an associated data
>dictionary.  This idea of "dynamic levelling" with automatic consistency
>updating is *essential* IMO, and would represent another order-of-magnitude
>leap in functionality.
>
 Yes, Yes, Yes.  I agree.  It looks pretty simple, too, though like all things
that simplicity will probably evaporate as soon as wee think about it awhile.
The idea, to say it a different way, is that sometimes during design it
would be useful to brainstorm a radical new rearrangement of all the
subsystems.  Suppose one has a fairly complete DFD that goes three or
four levels deep.  (After all, that's the idea, to hide complexity.)  Now,
I'd like to step way back and look at the whole thing from a distance, but
still grovel in the complexity.  So what I want is to "flatten" the
tree so that there is only one level, every single primitive process
and data store appears on that level, and then I want to be able to
move them around on the screen, and lasso new groups of them to create
a completely new tree.


>In ERD editors, this concept is even less matured.  I admit there is a void
>of consensus on implementation of abstract data types and entity clustering,
>yet I have seen no attempt to even begin forging into this uncharted
>territory.
>I read an excellent article in Communications of the ACM (Aug/89) discussing
>ER clustering that I would love to see expanded on and eventually automated.
>Personally, I find the Chen model both inadequate and misleading for design
>of the information model, especially where many-to-many and non-binary
>relationships are concerned.  Until more advanced techniques are developed,
>I find a simple mapping of foreign key references and expansion of more
>complex relationships to be more helpful than Chen notation.
>

Perhaps because I started in the data modeling world, I find there is
*more* agreement there, not less.  If one is willing to compile all the
functional dependencies, the tool ought to be able to do a better job
of helping build the right (gasp, dare I say that word?) objects.

                                                                 lee