[sc34wg3] N0391-0394: New SAM/XTM documents

[Robert Barta]

On Mon, Apr 21, 2003 at 05:09:51PM +0200, Jan Algermissen wrote:
> > > I have developed a query language that is expressed solely in terms of the
> > > TMM (meaning that it works for any TMA) and I'll try to think what I can
> > > do with it regarding your sentence above.
> > 
> > Interesting, can we have a look at it?
> 
> http://www.gooseworks.org/stmql.html

Ah, yes. Think I have seen that long ago, will take a fresh look, thx!

> > Well, a thing is 'perfect' if you cannot remove anything from it
> > without destroying it. The number of constraints in TMM is impressive,
> > though. 
> 
> I have tried a lot to remove some, but (at least for me) it turned out
> that all of them are neccessary.

This could well be. As you surely know, there can be several reasons
why a """"model"""" is too complex:

  - the original idea is not orthogonalized enough:

    TMs have these problems. For instance, characteristics can have
    more than one scope (like 'english' and 'beginner'). It is
    obvious, that that can ALWAYS be emulated with a SINGLE combined
    topic 'english-beginner'. It unnecessarily makes everything more
    complicated. And why did not have basenames types but instead
    variants introduced? <Long list of rant suppressed :>

  - the right abstraction level was not found, or, the model is living
    on more than one abstraction level which have to be linked against:

    This is the hardest to detect, but breaking the """"models"""" up
    into separate, well-balanced submodels can _significantly_ reduce
    the complexity.

  - the current """model""" makes use of none or the wrong notation:

    Maths can help to convey (or obfuscate) a lot of meaning.

There many more reasons, I guess.

> From my experience of implementing TMM, I even know that there are
> propably a few more (regarding value types for example).

That is also my experience, yes. In fact, my languages (+ the model
behind) changed radically with the first prototypes.

> > 'Simplicity' is not the first word which would come to my mind
> > :-) TMM is almost as long as the core specs of all my AsTMa languages
> > (not including the update) together.
> 
> But the length of the document and it's simplicity are not neccessarily
> proportional...

No, but it is the first what a (potential) implementor looks at.

\rho