Shex/Obsolete/Full discussion about semantics and shapes

Obsolete - please see ShEx Semantics

Here is the discussion about the separation of the shape definitions and the semantics.

To make a few things clear first 2 comparisons.

When comparing XML documents to RDF there are 3 main differences.

1. XML documents always contains ordered lists of childs whereas RDF documents are unordered.
2. Parents to child relationships are always 1 to 1..N, whereas RDF objects can references each order 0..N to 1..N. XML documents can always be presented as tree.
3. Childs are not referenced via a property, whereas in RDF they are. Note that this very important difference, which has many implications.(This is why you need a schema to interpret a XML document)

When comparing the RELAX NG, Java class defenitions and C structure definitions give the following comparison table.

	SHEX	RELAX NG	Java	C
1) Name	resourceshape	element	class	struct/union
2) Define 'rules/groups'	yes	yes	no	no, but there is option to use unions
3) Reference rule by name	yes	yes	-	you can reference unions
4) Have sequence of groups	yes	yes ->1?+*	-	special construct (need to define size field before)
5) Extending classes	currently not	no	yes	you can inline a base structure
6) Inlining	only rules	only rules	'no' only by extending	you can choice by reference(pointer) or to inline
7) Reference	only resource shape	only elements	default	you can choice by reference(pointer) or to inline
8) Parent child relation ship	0..N -> 1..N	1 -> 1..N	0..N -> 1..N	0..N -> 1..N (inlining possible)
9) Properties/childs ordered	no	yes	yes	yes

1) Term used for the concept 'class'. 
2) Can we define rules that describe the shape to which a instance of a 'class' must comply. 
3) Can we reference use rule by name, so we can reuse and 'extend' them
4) Can we tell that we expect a not yet predefined amount of occurrences of a certain rule within a shape. For example
 ( ex:reproducedBy @<EmployeeShape>,
   ex:reproducedOn xsd:dateTime )*
  In RELAX NG its possible to choice between exactly 1(1),0 or 1(?), 1 or more(+) and 0 or more(*). 
5) Can we make an extention on an existing class.
6) Can we tell the system to inline a certain group/rule, without making a separate object and referencing to it.
7) Can you tell the system to make a new object of the rule/group and reference to it.
8) What kind of relationships type is there from the parent to the child. In xml child elements are always contained in only one parent, so you always will have 
1 -> 1..N relationships.
9) Are the child properties and references in some way ordered. When SHEX described an RDF database then they are not ordered.

When comparing SHEX and RELAX NG to Java there is the following main distinction SHEX and RELAX NG have definitions of rule and rule group whereas Java and C do not, but Java has clear definition of OO class semantics. C sits somewhere in the middle.

So looking at this, I can make a distinction between the definitions of the shapes/rules and that of the class semantics. Shapes define the structure of a graph in the same way as a database scheme defines the structure of a SQL database. Where as the class semantics give meaning to what the object represent. I think is good to keep this 2 things seperated from each other, so we can define shapes, without having to define class with a full semantic meaning attached to it.

When understanding a unknown RDF database, you must have the shape structure to be able to query, whereas the semantics is handy to have. When integrating it with another source then both the shape structure and the semantics become a must have.

The shape definitions are needed for:

1. Understand the structure of a RDF resource, so that it become possible to create queries for an unknown resource.
2. Define rules to which a RDF resource should comply to when given as an input into an application.

The semantics are need for:

1. Find the meaning of the properties and classes used within a RDF resource
2. (Automatically) integrating two different sources which each other
3. Smart combining/integration of 2 or more documents that contain the same semantic information but are encoded in different structures.