Conference turris::languages

253.0. "Are simplest data objects best?" by SDEVAX::LOWRY () Thu Oct 26 1989 13:21

I would appreciate help in confirming or refuting some hypotheses 
about optimum design for the underlying parts of a formal language.
The hypotheses are described in
	KEEPER::CHIP:[LOWRY.PUBLIC]DATA_MODEL_CHARTS.
with more detail in
	KEEPER::CHIP:[LOWRY.PUBLIC]OPTIMUM_DATA_MODEL.

The ideas may be summarized as follows:

    - Providing both accurate representation of rich structure 
    	and powerful function in a language are almost conflicting.

    - Providing both requires a very constrained data model design.
	Powerful function requires a very small set of primitive
	data structures (as in APL or Relational DB).
	Good representation of rich structure requires that those
	few primitives be extremely simple.

    - The simplest possible primitive object type gives greatest
    	simplicity of expression -- in non-simple problem domains.
	The simplest objects that work are typed directed arcs
	(some of which point to themselves).

	       \					 /
		\					/
		 \				       /
		  \<-----.	       	       .----->/
		   \	  \	     	      /	     /
		    \      )  ------------>  (	    /
		     \____/	      	      \____/

    - The engineering optimum results from reduction to an irreducible 
	structure, not a tradeoff.

    - There are a few dozen other irreducible structure optima:
    	round wheels, tubular pipes, binary memory elements, 
    	vertical pillars, flat personal mirrors, etc. Few centuries 
	produce more than one.

    - Almost all irreducible structure optima:
	- endure indefinitely.
	- get near universal acceptance. 
	- are major factors supporting entire industries.
	- are easily understood.
	- are very robust. Competing engineering values do not
	  significantly compromise the irreducible structure over 
	  a wide range of conditions.
      They contrast sharply with tradeoff optima such as: operating
	  temperatures, rotation rates, recording densities, etc. 

    - I could be wrong, but the evidence is now 100% supportive
	that there is a new one. Many capable people have had 
	opportunities and offered no refutations to the hypotheses.

    - The constraints on choice of data primitive leave little room
	for specialization. As a result the advantages of a common 
	data model and common semantics for simple substructures are 
	much greater than potential advantages of specialization!

    - Potential applications:
    	- Improved user interface to computers,
    	    comprehensive, simple, durable, easily used.
	- Common data model for global information network.
    	- Framework for technical education.
    	- Universal language for technical literacy.
	- Powerful intellectual tool.
    	- Representation of complex knowledge (social concepts).

I would like to evaluate the soundness of the concepts and assess their
business and social value. Responses to the following would be helpful:

    What problem domains exist which have non-trivial evolving 
    structural richness and for which some choice other than the 
    directed arc allows greater simplicity of expression?

    What engineering values, if any, make it desirable to compromise
    on simplicity of expression in a way that favors departure from 
    directed arc data primitives?

    Is there any good reason to aim for less generality than
    a universal language for technical literacy when designing
    a new language which is not very specialized?

    What significant flaws exist in the 14 hypotheses in the 
    referenced charts?

The DAWN (or DIN) language can serve to illustrate concepts and 
provide a basis for comparisons. See
	KEEPER::CHIP:[LOWRY.PUBLIC]DBA_CH8.MEM 

When someone makes even a limited effort to to consider these 
questions, I would welcome a description of the result or non-result 
or questions that arise. Public or private discussion is encouraged. 
Suggestions about people who might contribute to or be interested in 
such discussion would be helpful. These views do not represent any 
organization.

Ed Lowry  
MRO1-2/M20
297-4065
KEEPER::LOWRY