Re: case tools

[Colin_Paul_Gloster@ACM.org (Colin Paul Gloster)]

Martin Dowie said:
  
"There is a move within the AdaUK and led by Artisan
(www.artisansw.com) to come up with standard UML<->Ada mappings,
perhaps analogous to the mapping between IDL and Ada."

Perhaps HRT-HOOD would be more useful than UML, or given the lamented
prevalence of that fruit of Grady Booch's, Intecs HRT has been working on
upgrading realtime UML specifications to be close to HRT-HOOD. It is not
as if mapping between HOOD and Ada is hard!

Simon Wright wrote:
  
"I have probably said this before .. although I can see the value of an
out-of-the-box mapping, to get projects started, I don't believe that
the mapping appropriate to "my" project is necessarily appropriate to
"yours"."

Simon Wright may find some consolation in these statements from a
fairly recent document on UML and Ada entitled "What are Ada Mapping rules
& What is a Profile?" attributed to Derek Russell:

"It would be na�ve however to assume that the mapping rules would be
sufficient to cover all possible implementation requirements.  There will
be model specific mappings such as active class priorities, entry point
semantics, etc.  To accommodate such requirements, language specific
annotations should be added to the model to allow specifics to be
identified and enhance code generation.  These annotations shall be in the
form of stereotypes and tags and are known as the [bold bagan]profile[bold
ended] (or [bold began]Ada profile[bold ended] in the case of the Ada
community).  The profile should override the mapping rules if applied to
any modelling element."

Simon Wright wrote:
  
"I also think that the UML<-Ada part will prove challenging .. the Ada
package is quite a hard concept to express in UML, and the idea of a
generic signature is even more fun. To some extend this depends on the
tool concerned .. UML allows, I think, the concept of a parameterised
collaboration (generic package to you and me, I think) but Rose
doesn't."

From the document attributed to Derek Russell:

"The mapping from a UML design to an Ada implementation is not as obvious
as some would think. The size of the Ada language leads to many different
ways of implementing even a simple UML model.  In addition, the UML
notation (1.4) is often not rich enough to fully specify the designers
intentions.

[..]

[..] In addition, it should be an aim that standard UML notation be
applied wherever possible."
  
"I think it is highly questionable whether every Ada construct should
be representable in UML. For a start, it allows individual developers
a lot of licence as to what architectural features they use, and you
don't always want that; anyone who has seen the amazing code
generation property sheets available in Rose and, I think, Rhapsody
(don't know Artisan) will know what I mean."

To some extent this might be okay with the "language specific
annotations" mentioned earlier (but you do not want to spend forever
banning almost everything from the COTS inital setup!) and with the
dialects of Ada supported. From that document:

"To avoid excessive model annotations, organisations (or toolset
vendors) should define a set of rules for mapping UML modelling elements
to the target language (SPARK, Ada83, Ada95 etc).  These are termed
mapping rules; a set of mapping rules should exist for each flavour of the
target language.  These mapping rules should provide the general, and most
common, set of mappings."