blob: 5faff8c197b0670346da2d08b3256066379112c3 [file] [log] [blame]
TODO NGPM Metamodel:
- Class extends EventProducer, EventProducer has :1-assoc to
TypeDefinition for notification type: this won't fly because a class
can emit / produce more than one event type. How should this be
- Model pre/post conditions and play with it as a S&AM
replacement. Thoughts: This has an impact on extensibility and
grouping of conditions. The S&AM approach has this notion of a schema
as a grouping entity, spanning multiple actions and multiple status
variables / values. A pre/post condition centers around the individual
action. This makes recognizing the grouping provided by a schema much
harder. Extensions have to happen within the expressions on the
action. The benefit of this, however, is full visibility of the sum of
all applicable conditions for the action which may be harder with a
set of overlapping schemas affecting a single action. A pre/post
condition that purely consists of a boolean expression over equality
checks of enumeration-typed features against values of these
enumerations can be depicted as a S&AM schema.
The next fundamental question is whether postconditions should be
expressed primarily as separate (redundant) constraints or be derived
from the implementation of the action. The action's implementation may
effect the values of one or more attributes. For those where the
attribute has an enumeration type, the connection from the action to
the target value would be part of the "derived" S&AM diagram. At
least, there should be a relation between the existance of a S&AM
transition from an action to a value of an attribute and a statement
reachable from the action's code on a reachable execution path that
sets that attribute to the respective value.
- Investigate enums / code lists more closely. How to do value help and
content assist for those? Is an enum / value set just another class
declaration where the "literals" are the names of the association ends
pointing to the class itself for the literals / code list items?
- What about handling of VOID expressions such as calling a block
that does not define an output type?
- Take care of nested TypeDefinitions, e.g., in order to allow for
types such as nested "collections" that correspond, e.g., with
Set<List<SalesOrder>>. This entails respective modifications to the
conformsTo definition on TypeDefinition.
- Allow for Block instances as values whose type is defined by the
block's Signature. Thoughts: a Signature then acts similar to a
Class. It defines the type of a value. SignatureTypeDefinitions define
value types, i.e., its objects are immutable and always passed by
class A {
int->A m;
B[ordered, unique] theBs;
C[ordered, unique][unique] setOfOrderedSetOfCs;
class B {
class C {