This chapter describes the scope rules, and most of the name lookup and flattening of Modelica.
Flattening is made in a context which consists of a modification environment (section 7.2.2) and an ordered set of enclosing classes.
The classes lexically enclosing an element form an ordered set of enclosing classes. A class defined inside another class definition (the enclosing class) precedes its enclosing class definition in this set.
Enclosing all class definitions is an unnamed enclosing class that contains all top-level class definitions, and not-yet read classes defined externally as described in section 13.2.2. The order of top-level class definitions in the unnamed enclosing class is undefined.
During flattening, the enclosing class of an element being flattened is a partially flattened class. [For example, this means that a declaration can refer to a name inherited through an extends-clause.]
The unnamed enclosing class of class definition C3 contains C1, C2, and C3 in arbitrary order. When flattening class definition C3, the set of enclosing classes of the declaration of x is the partially flattened class C3 followed by the unnamed enclosing class with C1, C2, and C3. The set of enclosing classes of z is C4, C3 and the unnamed enclosing class in that order.]
Names are looked up at class flattening to find names of base classes, component types, etc. Implicitly defined names of record constructor functions and enumeration type conversion functions are ignored during type name lookup [since a record and the implicitly created record constructor function, see section 12.6, , and an enumeration type and the implicitly created conversion function (section 184.108.40.206), have the same name]. Names of record classes and enumeration types are ignored during function name lookup.
When an element, equation, or section is flattened, any simple name [not composed using dot notation] is looked up sequentially in each member of the ordered set of instance scopes (see section 220.127.116.11) corresponding to lexically enclosing classes until a match is found or an enclosing class is encapsulated. In the latter case the lookup stops except for the predefined types, functions and operators defined in this specification.
Reference to variables successfully looked up in an enclosing class is only allowed for variables declared as constant. The values of modifiers are thus resolved in the instance scope of which the modifier appears; if the use is in a modifier on a short class definition,
This lookup in each instance scope is performed as follows
Among declared named elements (class-definition and component-declaration) of the class (including elements inherited from base-classes).
Among the import names of qualified import statements in the instance scope. The import name of import A.B.C; is C and the import name of import D=A.B.C; is D.
Among the public members of packages imported via unqualified import-statements in the instance scope. It is an error if this step produces matches from several unqualified imports.
Import statements defined in inherited classes are ignored for the lookup, i.e. import statements are not inherited.
For a composite name of the form A.B or A.B.C, etc. lookup is performed as follows:
The first identifier (A) is looked up as defined above.
If the first identifier denotes a component, the rest of the name (e.g., B or B.C) is looked up among the declared named component elements of the component.
If not found, and if the first identifier denotes a scalar component, or component[j] where component is an array of components and the indices j can be evaluated at translation time and component[j] is a scalar; and if the composite name is used as a function call, the lookup is also performed among the declared named class elements of the scalar component, and must find a non-operator function. All identifiers of the rest of the name (e.g., B and B.C] must be classes.
If the identifier denotes a class, that class is temporarily flattened (as if instantiating a component without modifiers of this class, see section 7.2.2) and using the enclosing classes of the denoted class. The rest of the name (e.g., B or B.C] is looked up among the declared named elements of the temporary flattened class. If the class does not satisfy the requirements for a package, the lookup is restricted to encapsulated elements only. The class we look inside may not be partial in a simulation model.
[The temporary class flattening performed for composite names follow the same rules as class flattening of the base class in an extends-clause, local classes and the type in a component clause, except that the environment is empty. See also MoistAir2 example in section 7.3 for further explanations regarding looking inside partial packages.]
For a name starting with dot, e.g.: .A [or .A.B, .A.B.C etc.] lookup is performed as follows:
The first identifier [A] is looked up in the global scope. This is possible even if the class is encapsulated and import statements are not used for this. If there does not exist a class A in global scope this is an error.
If the name is simple then the class A is the result of lookup.
If the name is a composite name then the class A is temporarily flattened with an empty environment (i.e. no modifiers, see section 7.2.2) and using the enclosing classes of the denoted class. The rest of the name [e.g., B or B.C] is looked up among the declared named elements of the temporary flattened class. If the class does not satisfy the requirements for a package, the lookup is restricted to encapsulated elements only. The class we look inside may not be partial.
[The package-restriction ensures that global name lookup of component references can only find global constants.]
See section 18.104.22.168.
An element declared with the prefix outer references an element instance with the same name but using the prefix inner which is nearest in the enclosing instance hierarchy of the outer element declaration.
An outer element reference in a simulation model requires that one corresponding inner element declaration exist or can be created in a unique way:
If there are two (or more) outer declarations with the same name, both lacking matching inner declarations, and the outer declarations are not of the same class it is in error.
If there is one (or more) outer declarations of a partial class it is an error.
In other cases, i.e. if a unique non-partial class is used for all outer declarations of the same name lacking a matching inner declaration, then an inner declaration of that class is automatically added at the top of the model and diagnostics is given.
The annotations defined in section 18.7 does not affect this process, other than that:
missingInnerMessage can be used for the diagnostic (and possibly error messages)
An outer element component may be of a partial class [but the referenced inner component must be of a non-partial class]. [inner/outer components may be used to model simple fields, where some physical quantities, such as gravity vector, environment temperature or environment pressure, are accessible from all components in a specific model hierarchy. Inner components are accessible throughout the model, if they are not “shadowed” by a corresponding inner declaration in a more deeply nested level of the model hierarchy.]
More complicated example:
The inner component shall be a subtype of the corresponding outer component. [If the two types are not identical, the type of the inner component defines the instance and the outer component references just part of the inner component].
[Inner declarations can be used to define field functions, such as position dependent gravity fields, e.g.:
An element declared with both the prefixes inner and outer conceptually introduces two declarations with the same name: one that follows the above rules for inner and another that follows the rules for outer. [Local references for elements with both the prefix inner and outer references the outer element. That in turn references the corresponding element in an enclosing scope with the prefix inner.]
Outer component declarations may only have modifications [including declaration equations] if they also have the inner prefix. Outer class declarations should be defined using short-class definitions which only may have modifications if they also have the inner prefix. For both cases those modifications are only applied to the inner declaration.
Intent of the following example: Propagate enabled through the hierarchy, and also be able to disable subsystems locally.
In order to guarantee that elements can be used before they are declared and that elements do not depend on the order of their declaration (section 4.3) in the enclosing class, the flattening proceeds in the following two major steps:
Generation of the flat equation system
The result is an equation system of all equations/algorithms, initial equations/algorithms and instances of referenced functions. Modifications of constants, parameters and variables are included in the form of equations.
The constants, parameters and variables are defined by globally unique identifiers and all references are resolved to the identifier of the referenced variable. No other transformations are performed.
The instantiation is performed in two steps. First a class tree is created and then from that an instance tree for a particular model is built up. This forms the basis for derivation of the flat equation system.
An implementation may delay and/or omit building parts of these trees, which means that the different steps can be interleaved. If an error occurs in a part of the tree that is not used for the model to be instantiated the corresponding diagnostics can be omitted (or be given). However, errors that should only be reported in a simulation model must be omitted there, since they are not part of the simulation model.
All necessary libraries including the model which is to be instantiated are loaded from e.g. file system and form a so called class tree. This tree represents the syntactic information from the class definitions. It contains also all modifications at their original locations in syntactic form. [The class tree is built up directly during parsing of the Modelica texts. For each class a local tree is created which is then merged into the one big tree, according to the location of the class in the class hierarchy. This tree can be seen as the abstract syntax tree (AST) of the loaded libraries.]. The builtin classes are put into the unnamed root of the class tree.
The output of the instantiation process is an instance tree. The instance tree consists of nodes representing the elements of a class definition from the class tree. For a component the subtree of a particular node is created using the information from the class of the component clause and a new modification environment as result of merging the current modification environment with the modifications from the current element declaration (see section 7.2.3).
The instance tree has the following properties:
It contains the instantiated elements of the class definitions, with redeclarations taken into account and merged modifications applied.
Each instance knows its source class definition from the class tree and its modification environment.
Each modification knows its instance scope.
The instance tree is used for lookup during instantiation. To be prepared for that, it has to be based on the structure of the class tree with respect to the class definitions. The builtin classes are instantiated and put in the unnamed root prior to the instantiation of the user classes, to be able to find them.
[The existence of the two separate trees (instance tree and class tree) is conceptual. Whether they really exist or are merged into only one tree or the needed information is held completely differently is an implementation detail. It is also a matter of implementation to have only these classes instantiated which are needed to instantiate the class of interest.]
A node in the instance tree is the instance scope for the modifiers and elements syntactically defined in the class it is instantiated from. The instance scope is the starting point for name lookup. [If the name is not found the lookup is continued in the instance scope corresponding to the lexically enclosing class. Extends clauses are treated as unnamed nodes in the instance tree – when searching for an element in an instance scope the search also recursively examines the elements of the extends clauses. Except that inherited import-statements are ignored.]
The instantiation is a recursive procedure with the following inputs:
the class to be instantiated (current class)
the modification environment with all applicable redeclarations and merged modifications (initially empty)
a reference to the node of the instance tree, which the new instance should go into (parent instance)
The instantiation starts with the class to be instantiated, an empty modification environment, and an unnamed root node as parent node.
During instantiation all lookup is performed using the instance tree, starting from the instance scope of the current element. References in modifications and equations are resolved later (during generation of flat equation system) using the same lookup.
A partially instantiated class or component is an element that is ready to be instantiated; a partially instantiated element (i.e. class or component) is comprised of a reference to the original element (from the class tree) and the modifiers for that element (including a possible redeclaration).
The possible redeclaration of the element itself takes effect.
The class of a partially instantiated component is found in the instance tree (using the redeclaration if any), modifiers merged to that class forming a new partially instantiated class that is instantiated as below.
For local classes and components in the current class, instance nodes are created and inserted into the current instance. Modifiers (including class redeclarations) are merged and associated with the instance and the element is partially instantiated. [The partially instantiated elements are used later for lookup during the generation of the flat equation system and are instantiated fully, if necessary, using the stored modification environment.]
Equations, algorithms, and annotations of the class and the component declaration are copied to the instance without merging. [The annotations can be relevant for simulations, e.g. annotations for code generation (section 18.3.), simulation experiments (section 18.4) or functions(section 12.7,section 12.8 and section 12.9).]
Extends clauses are not looked up, but empty extends clause nodes are created and inserted into the current instance – to be able to preserve the declaration order of components.
Classes of extends clauses of the current class are looked up in the instance tree, modifiers (including redeclarations) are merged, the contents of these classes are partially instantiated using the new modification environment, and are inserted into an extends clause node, which is an unnamed node in the current instance that only contains the inherited contents from that base-class.
The classes of extends-clauses are looked up before and after handling extends-clauses; and it is an error if those lookups generate different results.
At the end, the current instance is checked whether their children (including children of extends-clauses) with the same name are identical and only the first one of them is kept. [This is important for function arguments where the order matters.] It is an error if they are not identical.
Components (local and inherited) are recursively instantiated.
As an example consider:
To recursively instantiate M allowing the generation of flat equation system we have the following steps (not including checks):
Instantiate M: which partially instantiates B, b, C, D, E.
First find the class B in M (the partially instantiated elements have correct name allowing lookup)
instantiate the partially instantiated M.B with the modifier ”redeclare model A=D(p=1)”
partially instantiate M.b.a (no modifier), and M.b.A (with modifier ”=D(p=1)”)
First find the class A in M.b (the partially instantiated elements have correct name allowing lookup)
Instantiate the partially instantiated M.b.A with the modifier ”=D(p=1)”.
Find the base-class ”=D” from the modifier. This performs lookup for D in M, and finds the partially instantiated class D
Instantiate the base-class M.D with modifier p=1, and insert as unnamed node in M.b.A.
Partially instantiate the component p with modifier ”=1”
Find the base-class ”C” in M.D. Since there is no local element called ”C” the search is then continued in M and finds the partially instantiated class M.C
Instantiate the base-class M.C as below
Instantiate the base-class M.C inserting the result into unnamed node in M.b.a
Partially instantiate ”e”
Instantiate ”e” which requires finding ”E”. First looking for ”E” in the un-named node for extends ”M.C”, and, since there is no local element ”E” the search is then continued in ”M” (which lexically encloses M.C) and finds ”E” class inheriting from Real. The ”e” is then instantiated using class ”E” inheriting from ”Real”.
First the class ”E” in M.b.a finding ”E” class inheriting from Integer.
Instantiate the ”M.b.a.p” using the class ”E” inheriting from Integer with modifier ”=1”
Instantiate the base-class Integer with modifier ”=1”, and insert as unnamed node in ”M.b.a.p”.
An implementation can use different heuristics to be more efficient by re-using instantiated elements as long as the resulting flat equation system is identical.
Note that if ”D” was consistently replaced by ”A” in the example above the result would be identical (but harder to read due to two different classes called ”A”).
During this process, all references by name in conditional declarations, modifications, dimension definitions, annotations, equations and algorithms are resolved to the real instance to which they are referring to, and the names are replaced by the global unique identifier of the instance. [This identifier is normally constructed from the names of the instances along a path in the instance tree (and omitting the unnamed nodes of extends clauses), separated by dots. Either the referenced instance belongs to the model to be simulated the path starts at the model itself, or if not, it starts at the unnamed root of the instance tree, e.g. in case of a constant in a package.]
[To resolve the names, a name lookup using the instance tree is performed, starting at the instance scope (unless the name is fully qualified) of the modification, algorithm or equation. If it is not found locally the search is continued at the instance of the lexically enclosing class of the scope [this is normally not equal to the parent of the current instance], and then continued with their parents as described in section 5.3. If the found component is an outer declaration, the search is continued using the direct parents in the instance tree (see section 5.4). If the lookup has to look into a class which is not instantiated yet [or only partially instantiated], it is instantiated in place.]
The flat equation system consists of a list of variables with dimensions, flattened equations and algorithms, and a list of called functions which are flattened separately. A flattened function consists of algorithm or external clause and top-level variables (variables directly declared in the function or one of its base-classes) – which recursively can contain other variables; the list of non-top level variables is not needed.
The instance tree is recursively walked through as follows for elements of the class (if necessary a partially instantiated component is first instantiated):
At each visited component instance, the name is inserted into the variables list. Then the conditional declaration expression is evaluated if applicable.
The variable list is updated with the actual instance
The variability information and all other properties from the declaration are attached to this variable.
Dimension information from the declaration and all enclosing instances are resolved and attached to the variable to define their complete dimension.
If it is of record or simple type (Boolean, Integer, enumeration, Real, String, Clock, ExternalObject):
In the modifications of value attribute references are resolved using the instance scope of the modification. An equation is formed from a reference to the name of the instance and the resolved modification value of the instance, and included into the equation system. Except if the value for an element of a record is overridden by the value for an entire record; section 7.2.3.
If it is of simple type (Boolean, Integer, enumeration, Real, String, Clock, ExternalObject):
In the modifications of non-value attributes, e.g. start, fixed etc. references are resolved using the instance scope of the modification. An equation is formed from a reference to the name of the instance appended by a dot and the attribute name and the resolved modification value of the instance, and included into the equation system.
If it is of a non-simple type the instance is recursively handled.
If there are equation or algorithm sections in the class definition of the instance, references are resolved using the instance scope of the instance and are included in the equation system. Some references – in particular to non simple, non record objects like connectors in connect statements and states in transition statements are not resolved yet and handled afterwards.
Instances of local classes are ignored.
The unnamed nodes corresponding to extends-clauses are recursively handled.
If there are function calls encountered during this process, the call is filled up with default arguments as defined in section 12.4.1. These are built from the modifications of input arguments which are resolved using their instance scope. The called function itself is looked up in the instance tree. All used functions are flattened and put into the list of functions.
Conditional components with false condition are removed afterwards and they are not part of the simulation model. [Thus e.g. parameters don’t need values in them. However, type-error can be detected.]
Each reference is checked, whether it is a valid reference, e.g. the referenced object belongs to or is an instance, where all existing conditional declaration expressions evaluate to true or it is a constant in a package. [Conditional components can be used in connect-statements, and if the component is conditionally disabled the connect-statement is removed.]
This leads to a flattened equation system, except for connect and transition statements. These have to be transformed as described in chapter 9 and chapter 17. This may lead to further changes in the instance tree [e.g. from expandable connectors section 9.1.3)] and additional equations in the flattened equation system [e.g. connect equations (section 9.2), generated equations for state machine semantics (section 17.3.4)].
[After flattening, the resulting equation system is self contained and covers all information needed to transform it to a simulatable model, but the class and instance trees are still needed: in the transformation process, there might be the need to instantiate further functions, e.g. from derivative annotation or from inverse annotation etc., on demand.]