comp.lang.ada
 help / color / mirror / Atom feed
From: Leif Holmgren <nobody@nowhere.com>
Subject: Re: n versioning
Date: Mon, 01 May 2006 09:47:05 +0200
Date: 2006-05-01T09:47:05+02:00	[thread overview]
Message-ID: <4455BCF9.9030101@nowhere.com> (raw)
In-Reply-To: 1145851238.989726.218590@t31g2000cwb.googlegroups.com

Ananth the Boss wrote:
> thanks to all for ur interst in helping me.in my case the problem spec
> is the same and the problem is independantly developed by three
> different groups in three different languages. instead of coding the
> same problem thrice in three different programming languages; is there
> any approach like UML so that we can develop software models and from
> that the code can be generated automatically?

Yes.

I can recommend this approach, but do not think it will come cheap or 
out of the box. Costs will only be slightly smaller than with standard 
programming (but much lower than n-versioning anyway).

I suggest you call Mentor and ask them to come and show you BridgePoint. 
They have code generators that will give you C, C++, Java and Ada. If 
they claim not to have Java or Ada, ask them again.

However having out of the box code generator is not what you want, trust 
  me. It will give you a good starting point to learn how to make 
platform independent models. If you already know UML you will have to 
un-learn a lot a re-learn what object oriented analysis is really about.

For all full MDA projects the old saying "Uncertainty is not an option" 
is not to be taken mildly. What's not in your model will not be in your 
generated code.

After one or two years of (re)learning to model you will discover that 
you need a team maintaining the model compilers just as large as your 
modelling team. My experience is that by that time you will have a 
problem in convincing your management that this is true, after all you 
bought a model compiler so that you would have auto-generated code.

Another thing you will learn is that it's not the same people that 
should do the modelling as you may be used to. The modelling really 
should be done by the domain experts. So if your model is targeted at 
flight control it should be done by those who really know what flight 
control is about, not by your software staff. Other models you do may be 
targeted at communication and should therefore be done by experts in 
communication. Your software staff should possibly not be doing 
modelling at all. (See above)

I also suggest you hire a professional to get you started. The man my 
company has used is Leon Starr, but i'm sure there are alternatives. He 
is however a real pro at this.

The only drawback I have found using xtUML is it's weak support for 
mathematical stuff. Implementing some mathematical algorithms will be a 
real problem and may perhaps be better to do directly in the target 
language and bridge into your models.

If someone points you towards iLogix I would like to warn you a bit. 
Their tool does not support true platform independent modelling. You 
have to write action code in the target language, which in the end takes 
away the possibilities of simulating your code without first generating 
and it will allow your modellers to do stuff that should not be in your 
analysis models. And since your models then contains target specific 
stuff you will not have the possibility to change target language.

KennedyCarter also has a product of which I have no experience.

/Leif




      parent reply	other threads:[~2006-05-01  7:47 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2006-04-24  4:00 n versioning Ananth the Boss
2006-04-24  4:26 ` Jeffrey Creem
2006-04-24  7:36   ` Ananth the Boss
2006-05-01  7:47 ` Leif Holmgren [this message]
replies disabled

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox