QVTo has one form of mapping overloading provided by disjuncts. This is poorly specified.

It probably has another provided by name collision. This is unspecified.

What are the QVTo mapping overloading semantics?

QVTo has one form of mapping overloading provided by disjuncts. This is poorly specified.

It probably has another provided by name collision. This is unspecified.

The <rulerefs> between mappings has no defined semantics.

Presumably:

<package-name>*::<transformation-name>::<context-type>::<mapping-name>

allowing <package-name>*::<transformation-name> to be omitted defauklting to the current transdformation, and, preserving the current example semantics, resolving an omitted <context-type> by serching all mapping names in the transformation.

Undefined rulerefs

The <rulerefs> between mappings has no defined semantics.

Presumably:

<package-name>*::<transformation-name>::<context-type>::<mapping-name>

allowing <package-name>*::<transformation-name> to be omitted defauklting to the current transdformation, and, preserving the current example semantics, resolving an omitted <context-type> by serching all mapping names in the transformation.

Issue Links

Discussion

Since <mapping_id> needs definition, we can take the opportunity to rename the strange <rulerefs>.

8.2.2.10 VariableInitExp says

A variable may not declare an initialization value.

when surely it means

A variable may omit an initialization value.

Unfortunate VariableInitExp wording for missing initializer

8.2.2.10 VariableInitExp says

A variable may not declare an initialization value.

when surely it means

A variable may omit an initialization value.

The return type of xselectOne, xcollectselectOne returns a SEquence rather than an optional value.

xselectOne, xcollectselectOne is Sequence of One not One

The return type of xselectOne, xcollectselectOne returns a Sequence rather than an optional value.

Discussion

QVT13-34 revealed numerous errors in the pseudocode. This error is just one of many fixed by its resolution.

"A compound assignment is equivalent to perform as much simple assignments as there are expressions. Null values are skipped."

This is poor English and it conflicts with the OCL specification.

Pedantically it only skips "Null" values so "null values" are allowed anyway.

"An assignment may receive a future variable produced by a deferred resolve expression (see ResolveExp). The effect is equivalent to receive a null value except that a side-effect action occurs in order to allow re-executing the assignment at the end of the transformation.
An assignment expression returns the assigned value unless it is a future value, in which case null is returned."

This is barely intelligible and presumably related to the SmartQVT implementation. It certainly has confusion about nulls.

I recommend following OCL and allowing null values. ->excluding(null) is easily added.

Conflicting exclusion of null during assign

"A compound assignment is equivalent to perform as much simple assignments as there are expressions. Null values are skipped."

This is poor English and it conflicts with the OCL specification.

Pedantically it only skips "Null" values so "null values" are allowed anyway.

"An assignment may receive a future variable produced by a deferred resolve expression (see ResolveExp). The effect is equivalent to receive a null value except that a side-effect action occurs in order to allow re-executing the assignment at the end of the transformation.
An assignment expression returns the assigned value unless it is a future value, in which case null is returned."

This is barely intelligible and presumably related to the SmartQVT implementation. It certainly has confusion about nulls.

I recommend following OCL and allowing null values. ->excluding(null) is easily added.

Issue 19238

AssignExp to an Collection should make clear that the results is

target->including(source)->excluding(null)

Issue 19687

The two examples

mymultivaluedproperty += object Node

; // the list is reset and re-assigned

are simple assignments from single valued RHS. These are not covered in the preceding description.

I consider both to be errors since the RHS is unsuitable for assignment to the LHS. In both cases an asSet() or whatever should be appended.

More generally, the library operation that exhibits the "additive semantics" should be clearly identified.

Since addition to OCL collections is not permissible, it may be necessary to define constructors, so that

var c : Set(String) := Set

;

invokes the Set(T)(Set(T)) constructor.

This would allow initialization for a Sequence from a List and vice-versa that is otherwise magic.

which should exploit an OCL definition of OrderedSet::including to be add at end if necessary.

Discussion - nulls

UML does not allow nulls in multi-values, therefore nulls must be excluded during assignments to properties.

OCL does allow nulls, and even though this may appear to be a mistake, QVTo must support nulls in OCL collections until OCL is changed.

Discussion - multiple values

The current wording leaves open the design choice between:

a) perfect fidelity of nested types, i.e. addition of a Collection of Collections adds the one Collection of Collection element

b) perfect flattening, i.e. addition of a Collection of Collections adds all the elements in the nested collections.

c) something in between, perhaps behaving differently depending whether the LHS is a nested collection or not,

OCL unfortunately started without nested collections so that imploicit-collect flattens requiring the explicit collectNested for perfect Collection type fidelity. a) is not really an option. c) risks a confusing anarchy. b) is what is often wanted. If users want nested collections they can use more explicit operations.

Discussion - future values

Premature access to yet-to-be initialized values is obviously bad. Making this an error is an option, but it would be a breaking change; there may be users who poll the variable awaiting initialization. Simplest to just clarify the words to indicate that the value of any yet-to-be-initialized value is null, even when null is not assignable to the value. It is not initialized to null; it is null until initialized.

For imported compilation units, the specification should clearly state what happens if there is no corresponding ModuleImport for any of the modules included inside the imported unit.

The Eclipse QVTo implementation applies extends semantics by default. A very bad idea, because it leads to undesired extensions and overridings (especially in case of multiple modules per unit).

Using access as a default is a much better idea, but maybe clutters the namespaces at AST level with needless module imports.

Therefore applying neither access nor extends could be another good idea, but lacks a helpful default behavior.

Unspecified handling of imports without access/extends

For imported compilation units, the specification should clearly state what happens if there is no corresponding ModuleImport for any of the modules included inside the imported unit.

The Eclipse QVTo implementation applies extends semantics by default. A very bad idea, because it leads to undesired extensions and overridings (especially in case of multiple modules per unit).

Using access as a default is a much better idea, but maybe clutters the namespaces at AST level with needless module imports.

Therefore applying neither access nor extends could be another good idea, but lacks a helpful default behavior.

Discussion

Following some off-JIRA correspondence, the original issue is clearer and simpler.

The 8.2.1.4 words say that ImportKind has only two possibilities. But the words and model omit a [1] multiplicity or defaultValue so that null is a third possibility. The grammar also permits a missing ImportKind.

So as suggested we could introduce a third null semantics, or define one of access/extends as the default.

I see no benefit in a third semantics.

"extends" is a very serious mutating activity, not a default.

"access" is harmless and non-mutating. Hardly AST clutter. If the user didn't want the names the user should not have specified an import.

Let "access" be the default.

Clause 8.1.14 uses "mylist.add(5);" rather than "mylist->add(5);

and "mydict.put"

Incorrect navigation operator for LIst operations

Clause 8.1.14 uses "mylist.add(5);" rather than "mylist->add(5); and "mydict.put"

The objectsOf...(T) family of operations (e.g. Element::subobjectsOfKind(T), or Element::allSubobjectsOfType(T), or Model::objectsOfType(T)) returns a set of Elements.

It should be possible to infer the precise return type from the given T, i.e., return a Set(T) instead of Set(Element).

BTW, why is there no Model::objectsOfKind(T) ?

Define return type for objectsOf...(T) operations more precisely

The objectsOf...(T) family of operations (e.g. Element::subobjectsOfKind(T), or Element::allSubobjectsOfType(T), or Model::objectsOfType(T)) returns a set of Elements.

It should be possible to infer the precise return type from the given T, i.e., return a Set(T) instead of Set(Element).

BTW, why is there no Model::objectsOfKind(T) ?

Discussion

We can make the declaraions using the same approximation to a rigorous declaration as used by selectByKind in OCL 2.4.

Omission of objectsOfKInd is an easily corrected oversight.

The incorrect objectsOfKind usage by the current objectsOfType is a confusing error that gets worse when the correct operation is provided. Better to correct it.

clone and deepClone from QVT13-12 can similarly exploit a T.

The confusing wording of Model::createEmptyModel can be clarified,

The prohibition on creation or update of instances in helpers has no justification. Why?

Allow object creation in helpers

The prohibition on creation or update of instances in helpers has no justification. Why?

Discussion

This came up while clarifying wording for QVT13-33.

The prohibition is presumably intended to ensure that all object creation has a corresponding trace record. A good idea but beyond the pragmatic integrity of QVTo. Untraced objects can be created in mappings anyway, so prohibiting their creation in helpers is just an unhelpful restriction on programming flexibility.

There is currently no way for an overriding operation to call its overridden super implementation. ImperativeOCL should provide something similar to the 'super' reference which is available in Java.

Calling the super implementation of an overriding operation

There is currently no way for an overriding operation to call its overridden super implementation. ImperativeOCL should provide something similar to the 'super' reference which is available in Java.

Discussion

QVT13-93 clarifies the concept of a mapping identifier. An arbitrary mapping may therefore be called:

self.map My::Tx::Your::Type::doIt(withSomething);

The following helper from the specification is a NOP, because List::append has no side effect in QVT 1.2.

helper Package::computeCandidates(inout list:List) : List

Therefore the snippet should rather use List::add.

Incorrect use of side-effect free append in example

The following helper from the specification is a NOP, because List::append has no side effect in QVT 1.2.

helper Package::computeCandidates(inout list:List) : List

Therefore the snippet should rather use List::add.

Discussion

Yes

The "import" feature of Relations Language is not yet explained. And there is no example for it, too. For instance what does the "unit" in "import <unit>;" mean ?

Section: 7.13.5

The "import" feature of Relations Language is not yet explained. And there is no example for it, too. For instance what does the "unit" in "import <unit>;" mean ?

OCL's allInstances() applies to the one and only immutable model extent.

The meaning of this in QVT with multiple, some mutable, extents/domains/typedModels is unclear.

Suggest that allInstances applies to all the input and original inout extents only.

Obtaining instances from other domains is available through alternate library operations such as objectsOfType in QVTo.

allInstances() needs clarification

OCL's allInstances() applies to the one and only immutable model extent.

The meaning of this in QVT with multiple, some mutable, extents/domains/typedModels is unclear.

Suggest that allInstances applies to all the input and original inout extents only.

Obtaining instances from other domains is available through alternate library operations such as objectsOfType in QVTo.

Discussion - QVTo

allInstances() could be defined to apply to the initial, prevailing or final state. Since a QVTo execution proceeds through a predictable sequence of states limiting allInstances() to the initial state is unhelpful. A limitation to the final state is almost unuseable and may require a tool to introduce another form of late processing. Therefore the prevailing state is most appropriate for a QVTo allInstances().The returned instances are the mutable instances since any form of caching would impose an unjustified overhead.

Discussion - QVTr/QVTc

allInstances() could be defined to apply to the initial, prevailing or final state. The initial state is easy. The prevailing state does not exist declaratively. The final state is harder but may be useful for computing derived output values based on the outputs.

However consider a cascade of endogeneous transformations A2B, B2C, C2D in which B2C performs an allInstances(). If we now compose the trio into A2B2C2D, the functionality of the allInstances() must not change. Therefore neither initial nor final state will do, since the initial state changes from B to A and the final state from C to D. We must use a prevailing state, which we can define as the final state of the domain in which the invocation of allInstances() occurs. For our example this is one of B or C and does not change as a consequence of the cascade.

a production rule seems to be missing: <module_element> ::= <modeltype> Since, a transformation file can define several transformations and libraries (modules), it is desirable having the possibility of defining modeltypes exclusively to a module. These "local" modelTypes should belong to the scope of a module, and it shouldn't be accessible to the remaining defined modules (unless the use of extension mechanisms is specified).

Section 8.7.1a production rule seems to be missing

a production rule seems to be missing: <module_element> ::= <modeltype> Since, a transformation file can define several transformations and libraries (modules), it is desirable having the possibility of defining modeltypes exclusively to a module. These "local" modelTypes should belong to the scope of a module, and it shouldn't be accessible to the remaining defined modules (unless the use of extension mechanisms is specified).

Discussion

A transformation is a major programming element so multiple transformations per file seems as unusual as multiple Java classes per file.

But we do already support

transformation X(...) {...}
transformation Y(...) {...}

so perhaps we need per transformation modeltypes.

But if we provided them as suggested: we would have

transformation X(in q : Q) {
modeltype Q uses ....;
}

allowing an outer declaration to depend on an inner.

If a user really wants many transformations per file with many distinct conflicting modeltypes, then the user has the option to use distinctive names for each modeltype.

Much more sensible to use multiple files in the first place.

Similarly we don't want or support per-transformation imports.

The EBNF:

<modeltype> ::= 'modeltype' <identifier> <compliance_kind>?
'uses' <packageref_list> <modeltype_where>? ';'
<modeltype_where> ::= 'where' <expression_block>

allows only a single where which may be a BlockExp, so ModelType.additionalCondition should be [0..1] rather than [*].

Incorrect ModelType.additionalCondition multiplicity

The EBNF:

<modeltype> ::= 'modeltype' <identifier> <compliance_kind>?
'uses' <packageref_list> <modeltype_where>? ';'
<modeltype_where> ::= 'where' <expression_block>

allows only a single where which may be a BlockExp, so ModelType.additionalCondition should be [0..1] rather than [*].

Discussion

Yes, but

<expression_block> ::= '{' <expression_list>? '}'

which allows for the multiple expressions.

The multiple expressions could be shown more clearly by flattening as:

<modeltype> ::= 'modeltype' <identifier> <compliance_kind>?
'uses' <packageref_list> ('where' '{' <expression_list>? '}')? ';'

But in the absence of a CS2AS mapping, implementations are not obliged to create the inappropriate BlockExp.

Initializing variables at the point of declaration is generally recognized as good practice. QVTc does not support it.

Suggest extend the (Realized) Variable syntax to support an initializer.

Allow QVTc to use initialized variables

Initializing variables at the point of declaration is generally recognized as good practice. QVTc does not support it.

Suggest extend the (Realized) Variable syntax to support an initializer.

Discussion

Initializing RealizedVariables makes no sense, since a RealizedVariable is initialized by the new Type().

Initializing (Unrealized)Variables is however useful and supported by the Abstract Syntax already; just need some Concrete Syntax re-using the := of an Assignment.

Issue 19146 removed most of the erroneous matchXXX arguments but omitted to do so for String:: matchIndentifier

Redundant argument on matchIdentifier

Issue 19146 removed most of the erroneous matchXXX arguments but omitted to do so for String:: matchIdentifier

Discussion

Yes

AssignExp to an Collection should make clear that the results is

target->including(source)->excluding(null)

which should exploit an OCL definition of OrderedSet::including to be add at end if necessary.

AssignExp unclear for OrderedSet

AssignExp to an Collection should make clear that the results is

target->including(source)->excluding(null)

which should exploit an OCL definition of OrderedSet::including to be add at end if necessary.

The Issue 19146 resolution is not sufficiently clear for Collection::deepclone.

Any nested mutables must be deepcloned, so the specification should require a recursive deepclone that the return decides between self or the clone according to whether any elements need to change

Clarify deepclone on Collections

The Issue 19146 resolution is not sufficiently clear for Collection::deepclone.

Any nested mutables must be deepcloned, so the specification should require a recursive deepclone that the return decides between self or the clone according to whether any elements need to change

Discussion

Yes

There is no mention of extent in Section 8.1 where one might hope to find an overview of the utility.

The proposed resolution of Issue 13103 clarifies the ability of ObjectExp to (re-)define the extent residence of a pre-existing object.

MappingParameter has some contradictory indications that an extent is bound by the callee rather than the caller.

Is this a consequence of the need for a MappingOperation to have a 1:1 correspondence to a QVTr Relation requiring the domain to be determined solely from the declaration? How is the inability of a Relation to have more than one parameter per domain accommodated when there can be more than one parameter per extent?

Specify the utility of an extent

There is no mention of extent in Section 8.1 where one might hope to find an overview of the utility.

The proposed resolution of Issue 13103 clarifies the ability of ObjectExp to (re-)define the extent residence of a pre-existing object.

MappingParameter has some contradictory indications that an extent is bound by the callee rather than the caller.

Is this a consequence of the need for a MappingOperation to have a 1:1 correspondence to a QVTr Relation requiring the domain to be determined solely from the declaration? How is the inability of a Relation to have more than one parameter per domain accommodated when there can be more than one parameter per extent?

Discussion

A section on Extents is sadly lacking; write it.

The confusing fiction that there is a correspondence between QVTo Model Parameters and QVTr Domains is best eliminated.

QVTr does not allow variables to be initialized at the point of declaration. This is a rather prehistoric limitation that significantly degrades readability.

a) it requires separate declaration and assignment lines
b) it inhibits inference of the variable type from the initializer
c) it obfuscates the variable assignment

The latter is a particular problem in QVTr where the assignment may a consequence of a pattern match;

Suggest allowing an initializer on a VarDeclarationCS and allowing the type to be inferred from the initializer.

QVTr Variable Initialization

QVTr does not allow variables to be initialized at the point of declaration. This is a rather prehistoric limitation that significantly degrades readability.

a) it requires separate declaration and assignment lines
b) it inhibits inference of the variable type from the initializer
c) it obfuscates the variable assignment

The latter is a particular problem in QVTr where the assignment may a consequence of a pattern match;

Suggest allowing an initializer on a VarDeclarationCS and allowing the type to be inferred from the initializer.

Discussion

The underlying Abstract Syntax supports an initializer.

Prototyping in Eclipse QVTr demonstrated a useful reduction in line count and increase in clarity of the RelToCore transformation.

The QVTr AS allows CollectionTemplateExp::rest to be null corresponding to a precisely enumerated list of collection members to be matched.

The QVTr CS does not support this.

Suggest changing the

'++' (restIdentifier=[pivot::Variable|UnrestrictedName] | '_')

sub-parsing rule to

('++' (restIdentifier=[pivot::Variable|UnrestrictedName] | '_'))?

Exact QVTr Collection match

The QVTr AS allows CollectionTemplateExp::rest to be null corresponding to a precisely enumerated list of collection members to be matched.

The QVTr CS does not support this.

Suggest changing the

'++' (restIdentifier=[pivot::Variable|UnrestrictedName] | '_')

sub-parsing rule to

('++' (restIdentifier=[pivot::Variable|UnrestrictedName] | '_'))?

When mapping/relation refinement is used, it should be possible to specify via the abstract keyword that a mapping/relation is only intended for use in a refined form.

Thus in A.3.1, declaring classAttributes abstract could determine whether the execution needs to consider matches not covered by the classPrimitiveAttributes and classComplexAttributes refinements.

Support abstract mappings/relations

When mapping/relation refinement is used, it should be possible to specify via the abstract keyword that a mapping/relation is only intended for use in a refined form.

Thus in A.3.1, declaring classAttributes abstract could determine whether the execution needs to consider matches not covered by the classPrimitiveAttributes and classComplexAttributes refinements.

Discussion

A simple enhancement

The two examples

mymultivaluedproperty += object Node

; // the list is reset and re-assigned

are simple assignments from single valued RHS. These are not covered in the preceding description.

I consider both to be errors since the RHS is unsuitable for assignment to the LHS. In both cases an asSet() or whatever should be appended.

More generally, the library operation that exhibits the "additive semantics" should be clearly identified.

Since addition to OCL collections is not permissible, it may be necessary to define constructors, so that

var c : Set(String) := Set

;

invokes the Set(T)(Set(T)) constructor.

This would allow initialization for a Sequence from a List and vice-versa that is otherwise magic.

Unclear multi-valued initialisation from single value

The two examples

mymultivaluedproperty += object Node

; // the list is reset and re-assigned

are simple assignments from single valued RHS. These are not covered in the preceding description.

I consider both to be errors since the RHS is unsuitable for assignment to the LHS. In both cases an asSet() or whatever should be appended.

More generally, the library operation that exhibits the "additive semantics" should be clearly identified.

Since addition to OCL collections is not permissible, it may be necessary to define constructors, so that

var c : Set(String) := Set

;

invokes the Set(T)(Set(T)) constructor.

This would allow initialization for a Sequence from a List and vice-versa that is otherwise magic.

Issue Links

The second example contains

"if result then return;"

which has a non-boolean condition expression and a missing endif.

In the first example, it is not clear that the revisit adds rather than overwrites.

In the third and fourth examples it is not clear why the second pass reuses the context for the first rather than creates new objects.

8.1.10 Errors in Examples

The second example contains

"if result then return;"

which has a non-boolean condition expression and a missing endif.

In the first example, it is not clear that the revisit adds rather than overwrites.

In the third and fourth examples it is not clear why the second pass reuses the context for the first rather than creates new objects.

Discussion

The examples are much worse than suggested; verging on the diabolical.

The resolution for Issue 12375 specifies that ListLiteralExp has
CollectionLiteralExp
as a superclass.

This leaves the behaviour of the inherited kind and part attributes
unspecified and
rather embarrassing.

ListLiteralExp (like DictLiteralExp) should inherit directly from
LiteralExp.

Inappropriate ListLiteralExp inheritance (Correction to issue 12375 resolution)

The resolution for Issue 12375 specifies that ListLiteralExp has CollectionLiteralExp as a superclass.

This leaves the behaviour of the inherited kind and part attributes unspecified and rather embarrassing.

ListLiteralExp (like DictLiteralExp) should inherit directly from LiteralExp.

Discussion

The lack of specification of ListLiteralExp::kind and ListLiteralExp::part is indeed a problem.

CollectionLiteralExp::kind is inextensible bloat that OCL could usefully eliminate, so no ListLiteralExp::kind is needed in QVTo.

ListLiteralExp::part is needed if QVTo is truly an extension of OCL. Unfortunately the QVTo BNF omits the ".." operator and support for CollectionRanges.

Therefore we need to introduce the missing CollectionRange parsing. ListLiteralExp can inherit LiteralExp adding CollectionLiteralExp::part but not CollectionLiteralExp::kind. ListLiteralExp::element is misguided; ListLiteralExp::part replaces it.

As interpretated from the especification (pag 104), the way of adding new operations to OCL predefined types is creating new Typedef instances
which must have the OCL predefined type as the base type. The new operations are added to this new typedef. However there are several problems:
1. The specification doesn't provide any name for these typedefs.
2. The specification doesn't specify which type (QVT typedef or OCL predefined type) should be used when referencing such OCL predefined types in a QVTo transformation.

Solution for 1).
Suggestion a: Name the typedef with the same name of the base type. This provokes name's clash with the predefined type's name, due to there are two different types from two standard libraries
which have the same name. A possible solution, would be expliciting that typedefs (aliases) will never clash its name with its base type.

Suggestion b: Name the tpyedef with a different name, such as QVToXXXXXX or XXXX_Alias.

Solution for 2).
Suggestion a: Taking the typedef as the referenced type in QVTo transformations.
Suggestion b: Taking the OCL predefined type as the referenced type in QVTo transformations.
Suggestion c: Considering resolution of issue 13168, so that only OCL predefined exists, and therefore, the only type which can be referenced.

It's a little bit weird having 2 different types (a type, and its alias typedef) which represent just the same type, specially when they are related by a reference.
My solution's preference in order are:
Suggestion c: Just having one type to refer.
Suggestion a: Since the typedef "extends" the behaviour of the predefined type (adding new operations), the former must be the referred one.
Suggestion b: The OCL predefined type is referenced, but we must take into account that the operations added to the typedef are also available.

QVTo Standard Library and typedefs Issue. Extending OCL predefined types

As interpretated from the especification (pag 104), the way of adding new operations to OCL predefined types is creating new Typedef instances
which must have the OCL predefined type as the base type. The new operations are added to this new typedef. However there are several problems:
1. The specification doesn't provide any name for these typedefs.
2. The specification doesn't specify which type (QVT typedef or OCL predefined type) should be used when referencing such OCL predefined types in a QVTo transformation.

Solution for 1).
Suggestion a: Name the typedef with the same name of the base type. This provokes name's clash with the predefined type's name, due to there are two different types from two standard libraries
which have the same name. A possible solution, would be expliciting that typedefs (aliases) will never clash its name with its base type.

Suggestion b: Name the tpyedef with a different name, such as QVToXXXXXX or XXXX_Alias.

Solution for 2).
Suggestion a: Taking the typedef as the referenced type in QVTo transformations.
Suggestion b: Taking the OCL predefined type as the referenced type in QVTo transformations.
Suggestion c: Considering resolution of issue 13168, so that only OCL predefined exists, and therefore, the only type which can be referenced.

It's a little bit weird having 2 different types (a type, and its alias typedef) which represent just the same type, specially when they are related by a reference.
My solution's preference in order are:
Suggestion c: Just having one type to refer.
Suggestion a: Since the typedef "extends" the behaviour of the predefined type (adding new operations), the former must be the referred one.
Suggestion b: The OCL predefined type is referenced, but we must take into account that the operations added to the typedef are also available.

Discussion

The offending sentence is:

This is specifically used in the QVT standard library to attach operations to pre-defined primitive types, like String and Collections.

This facility is not used anywhere in the specification and there is no grammar that facilitates it. No model of the QVTo library is provided that exploits it. The sentence is therefore just an observation on a particular implementation approach.

The approach takes no account of the very similar requirement for Complete OCL to extend types. This is being resolved by work towards OCL 2.5 / 3.0 with which QVTo will almost certainly have to comply.

Therefore, tomorrow this sentence will be embarassing and incompatible. Today it is just irrelevant and confusing.

In operational QVT, it is not clear what happens when the "when" or "where" clause of an inherited mapping does not hold.

Suggestion:
Considering inheritance is a type (or, maybe, a synonym) of generalization, it would be expected that the semantics of inheritance mimics the semantics of generalization in MOF. The UML, which defines generalization used by CMOF and EMOF, states: "... features specified for instances of the general classifier are implicitly specified for instances of the specific classifier. Any constraint applying to instances of the general classifier also applies to instances of the specific classifier." (UML Infrastructure 2.4.1, p.51). If the "when" and "where" clauses are considered as features of a mapping, the clauses of the inherited mapping should be implicitly specified. Similarly, if they are considered as constraints applying to a mapping, the clauses defined in the inherited mapping should apply to the inheriting mapping. Therefore, a possible solution in both situations is to consider that the "when" and "where" clauses must hold in the inheriting mapping.

Commentary:
An interesting discussion is if something similar to the Liskov substitution principle should be applied in this situation.

Undefined semantics for unsatisfied "when" and "where" in inherited mapping

In operational QVT, it is not clear what happens when the "when" or "where" clause of an inherited mapping does not hold.

Suggestion:
Considering inheritance is a type (or, maybe, a synonym) of generalization, it would be expected that the semantics of inheritance mimics the semantics of generalization in MOF. The UML, which defines generalization used by CMOF and EMOF, states: "... features specified for instances of the general classifier are implicitly specified for instances of the specific classifier. Any constraint applying to instances of the general classifier also applies to instances of the specific classifier." (UML Infrastructure 2.4.1, p.51). If the "when" and "where" clauses are considered as features of a mapping, the clauses of the inherited mapping should be implicitly specified. Similarly, if they are considered as constraints applying to a mapping, the clauses defined in the inherited mapping should apply to the inheriting mapping. Therefore, a possible solution in both situations is to consider that the "when" and "where" clauses must hold in the inheriting mapping.

Commentary:
An interesting discussion is if something similar to the Liskov substitution principle should be applied in this situation.

Discussion

The suggestions regarding constraint inheritance and LSP are appropriate for a declarative language; they are not appropriate for a pragmatic imperative language such as QVTo. Each mapping can be independently specified.

It can be made clearer that pre and post conditions are assertions that fail accordingly.

Comparison of QVTr/QVTo when and where highlights the fallacy that a QVTo mapping is a refinement of QVTr relation. In QVTo, failure of a where post-condition is a catastrophic assertion failure, whereas a QVTr where clause is a corrollary to be attempted afterwards and ignored if it fails/fails to match. The claim in 8.1.4 that the QVTo when and where clauses are owned by QVTr is untrue; QVTo owns its own when and where clauses and works very well without any QVTr objects.

Trace data creation is specified to happen "at the end of the initialization section".

For disjuncting mappings, the initialization section is never executed, which prevents any trace data from being stored.

As a consequence, no resolution via resolve-in-expressions is possible on the disjuncting mapping, due to the missing trace record. This is problematic, since disjunction should be transparent from a resolver's point of view, i.e. it should not make a difference for resolution whether a mapping disjuncts or not.

Hence, some clarification is required whether trace records are deliberately avoided for disjuncting mappings (for whatever reason), or whether trace data must be created in another place than the init section in case of a disjuncting mapping.

No trace data for disjuncting mapping

Trace data creation is specified to happen "at the end of the initialization section".

For disjuncting mappings, the initialization section is never executed, which prevents any trace data from being stored.

As a consequence, no resolution via resolve-in-expressions is possible on the disjuncting mapping, due to the missing trace record. This is problematic, since disjunction should be transparent from a resolver's point of view, i.e. it should not make a difference for resolution whether a mapping disjuncts or not.

Hence, some clarification is required whether trace records are deliberately avoided for disjuncting mappings (for whatever reason), or whether trace data must be created in another place than the init section in case of a disjuncting mapping.

Discussion

It seems the 8.1.x section on Tracing never got written. Time for a thorough review and clarification.

Trace lookups by source objects treat context and all in/inout parameters as source objects.

Trace lookups by target objects treat all out/inout and result parameters as target objects.

Trace lookups by mapping allow either disjuncting mapping or candidate mapping to be used as the mapping selection.

If re-execution is to be completely suppressed and fully re-use a previous execution, the output trace should be the final outputs, but the existing text is clear that it is the preliminary outputs at the end of the initialization section. The earlier trace gives a slightly different semantics that requires the population section to modify the created object after it is traced and allows the population section to use the trace. This is different but fine when explained clearly.

References to 8.1.12 refer to the new text from the QVT13-99 resolution.

The spec should clarify the interaction of
1.) explicit instantiation + execution of 'accessed' transformations, and
2.) trace records / resolving.

The following questions are of interest:

How does the initial trace for an 'accessed' transformation look like? Does it reuse the records previously created by the 'accessing' transformation, or does an 'accessed' transformation always start on an empty trace?

How does an 'accessed' transformation affect the trace? Are the trace records post-visible that were created during execution of the 'accessed' transformation, or is the trace of the 'accessing' transformation unchanged?

Both issues heavily affect the results of resolve-operations. Resolving object references after executing an 'accessed' transformation would be very practical.

Trace data for an 'accessed' transformation

The spec should clarify the interaction of
1.) explicit instantiation + execution of 'accessed' transformations, and
2.) trace records / resolving.

The following questions are of interest:

How does the initial trace for an 'accessed' transformation look like? Does it reuse the records previously created by the 'accessing' transformation, or does an 'accessed' transformation always start on an empty trace?

How does an 'accessed' transformation affect the trace? Are the trace records post-visible that were created during execution of the 'accessed' transformation, or is the trace of the 'accessing' transformation unchanged?

Both issues heavily affect the results of resolve-operations. Resolving object references after executing an 'accessed' transformation would be very practical.

Discussion

The resolution of QVT13-23 makes clear that the trace data contains a trace record for every mapping.

Concerning to Relations Language, how will the metamodels get into a transformation script ? Is there a technique similar to Operational Mappings using metamodel declaration and modeltypes ? The RL sample transformation script in annex A.1 (pp 164) doesn't declare the metamodels. The OM sample (A.2.3) does. There is some syntax for declaring and using metamodels and modeltypes in OM (pp118), there isn't for RL (pp38).

Relations Language: how will metamodels get into a transformation scrip

Concerning to Relations Language, how will the metamodels get into a transformation script ? Is there a technique similar to Operational Mappings using metamodel declaration and modeltypes ? The RL sample transformation script in annex A.1 (pp 164) doesn't declare the metamodels. The OM sample (A.2.3) does. There is some syntax for declaring and using metamodels and modeltypes in OM (pp118), there isn't for RL (pp38).

Discussion

We needs to support definition of e.g. UML in

transformation umlToRdbms(uml:UML, rdbms:RDBMS)

But we actually may want to reference a document URI or namespace URI

http://www.omg.org/spec/UML/20131001/UML.xmi
http://www.omg.org/spec/UML/20131001

But the current syntax is essentially a Java import without the trailing * option.

import my.model.package;

It cannot reference URIs and it has the wrong separator.

Something similar to QVTo's ModelType is needed:

modeltype UML uses SimpleUml ("http://omg.qvt-examples.SimpleUml");

However ModelType has many problems that we do not need to replicate:

• it is a Class; absolutely not
• it supports strict/effective inside rather than outside the program
• it has undefined URI usage

  Conversely Complrete OCL has a requirement for an import too, and here ModelType's are not appropriate at all.

Let's fix up import.

top relation AssocToFKey { pKey : Key; ... when

... } In my opinion that doesn't work. pKey is of type Key, destTbl.key is of type OrderedSet key, isn't it ?

Section: A1.1.1

top relation AssocToFKey { pKey : Key; ... when

... } In my opinion that doesn't work. pKey is of type Key, destTbl.key is of type OrderedSet key, isn't it ?

Discussion

Yes. Well at least it's a Collection; the diagram leaves an ambiguity.

Comparison with the QVTc solution shows that the 'primary' key should be selected, and initialized on creation.

Loading the example in the Eclipse QVTr editor reveals many more problems.

QVTr already has queries but they are much less user friendly than e.g. MOFM2T's equivelent for which the first parameter is a hidden self, or indeed QVTo.

Perhaps something closer to Complete OCL would do, allowing def'd attributes or operations.

QVTr already has queries but they are much less user friendly than e.g. MOFM2T's equivelent

QVTr already has queries but they are much less user friendly than e.g. MOFM2T's equivelent for which the first parameter is a hidden self, or indeed QVTo.

Perhaps something closer to Complete OCL would do, allowing def'd attributes or operations.

Discussion

QVTo has contextual helpers.

Syntactically it is trivial for QVTc and QVTo to have contextual queries just by allowing a scoped rather than simple name for the query.

Semantically to avoid duuplicating QVTo AS irregularities we can use the synonym package merges for Complete OCL to make the implementation easy.

This is a suggestion:

A developer may need to take a set of in models that may vary in size each time he runs the transformation.

So, instead of hard-coding the input/out put models when defining the transformation like this:

transformation Uml2Rdbms(in uml:UML,out rdbms:RDBMS)

The developer should be able to define a dynamic set of input models. So that he will not be obligated to change the transformation definition if the number of input models change every time he runs the transformation.

For example, I defined a transformation that takes multiple input models (conforming to the same meta-model) to merge them. Now, I got different number of in models every time I run the transformation. To make this more dynamic, I need to take an array of models as an input to my transformation without specifying their numbers in the transformation definition each time I run the transformation.

Dynamic set of input models

This is a suggestion:

A developer may need to take a set of in models that may vary in size each time he runs the transformation.

So, instead of hard-coding the input/out put models when defining the transformation like this:

transformation Uml2Rdbms(in uml:UML,out rdbms:RDBMS)

The developer should be able to define a dynamic set of input models. So that he will not be obligated to change the transformation definition if the number of input models change every time he runs the transformation.

For example, I defined a transformation that takes multiple input models (conforming to the same meta-model) to merge them. Now, I got different number of in models every time I run the transformation. To make this more dynamic, I need to take an array of models as an input to my transformation without specifying their numbers in the transformation definition each time I run the transformation.

Discussion - imperative

This is already supported by the QVTo grammar. Just define a Collection of models as input or output.

transformation Uml2Rdbms(in uml:Sequence(UML),out rdbms:Sequence(RDBMS))

and use collection operations to access the individual models.

Discussion - declarative

The syntax of identifiers is undefined.

The syntax for mapping clearly prohibits the use of a direction named 'where'.

Suggest: identifier is an OCL simpleName, less the new reserved words (check default enforce
imports map realize refines transformation uses where)

Suggest: a string-literal may be used as an identifier to support awkward identifiers such as 'where'.

Identifiers

The syntax of identifiers is undefined.

The syntax for mapping clearly prohibits the use of a direction named 'where'.

Suggest: identifier is an OCL simpleName, less the new reserved words (check default enforce
imports map realize refines transformation uses where)

Suggest: a string-literal may be used as an identifier to support awkward identifiers such as 'where'.

Discussion

OCL 2.3/2.4 clarified identifiers and strings so we can just exploit the OCL productions including it's underscore-prefixed-string identifier.

OCL 2.3 also introduced String::_'+' so no special description is needed for a QVTr extension.

xcolect is specified to like collect, which flattens, however the xcollect pseudocode does not flatten.

xcollect is ambiguously flattened

xcollect is specified to like collect, which flattens, however the xcollect pseudocode does not flatten.

Discussion

Ditto xcollectselect, xcollectselectOne

Once an attempt is made to put the pseudo-code through Eclipse QVTo, it is quite amazing how many different errors there are.

• spurious source/iterator arguments
• missing compute open {
• increarect non-collection returns
• incorrect accommodation of ordered collections
• no flattening
• no single values
• use of mutable Sequence rather than List

Intuitively List and Dict are objects, so if you pass them to a mapping/query/helper as an inout parameter, the object in the caller may be updated by the call.

This is the behaviour of a Class Instance not a DataType Value.

Please use the open https://bugs.eclipse.org/bugs/show_bug.cgi?id=420150 to discuss this topic.

List and Dict Class/DataType confusion

Intuitively List and Dict are objects, so if you pass them to a mapping/query/helper as an inout parameter, the object in the caller may be updated by the call.

This is the behaviour of a Class Instance not a DataType Value.

Discussion

Equally intuitively List (and Dict) are values. Two distinct 'occurrences' may be equal. This is the behavior of a DataType not a Class.

List and Dict are specified as CollectionType (DataType) derivations so we just need to fix up indications that List and Dict are Classes.

While clarifying wording, QVT13-120 is merged to remove the prohibition on object creation/update in helpers.

The Helper class in the AS syntax supports both helpers and queries.

These two concepts have important distinctions that are poorly differentiated in 8.2.1.12, where as a minimum the two italicized terms 'helper' and 'query' should be used when defining the semantics of each.

Presumably a 'query' cannot modify anything (other than the log file) anywhere.

Presumably a 'helper' can modify anything anywhere, except of course sets and tuples that are always immutable.

The current wording suggests that a 'helper' can modify sets and tuples but cannot modify objects or create objects. Surely a 'helper' cannot modify sets or tuples but can create or update objects?

Inadequate helper/query distinction

The Helper class in the AS syntax supports both helpers and queries.

These two concepts have important distinctions that are poorly differentiated in 8.2.1.12, where as a minimum the two italicized terms 'helper' and 'query' should be used when defining the semantics of each.

Presumably a 'query' cannot modify anything (other than the log file) anywhere.

Presumably a 'helper' can modify anything anywhere, except of course sets and tuples that are always immutable.

The current wording suggests that a 'helper' can modify sets and tuples but cannot modify objects or create objects. Surely a 'helper' cannot modify sets or tuples but can create or update objects?

OCL defines invalid as the result of a failed evaluation. The QVT specification is almost totally silent on invalid, except for a few library operations that may generate it.

Presumably an invalid not 'caught' by oclIsInvalid/oclIsUndefined in QVTo causes a to-be-specifued exception to be raised in the problem statement.

Presumably an invalid not 'caught' by oclIsInvalid/oclIsUndefined in QVTc/QVTr inhibits the detection of matches.

invalid in QVT

OCL defines invalid as the result of a failed evaluation. The QVT specification is almost totally silent on invalid, except for a few library operations that may generate it.

Presumably an invalid not 'caught' by oclIsInvalid/oclIsUndefined in QVTo causes a to-be-specifued exception to be raised in the problem statement.

Presumably an invalid not 'caught' by oclIsInvalid/oclIsUndefined in QVTc/QVTr inhibits the detection of matches.

Discussion

But invalid may be temporarily caught by a let variable of QVTo var. It is the use that is a problem.

The current phrasing

Within a contextual operation <it>result</it> represents the unique result parameter (if there is a unique declared result) or the
tuple containing the list of declared result parameters.

suggests that the synthetic Tuple return is accessible as result. This is not necessary and imposes considerable implementation difficulties since a new Tuple would need creation with each mutation.

Sucggest the new wording.

Within a contextual operation for which no name is specified for a single return parameter, <it>result</it> is the implicit name of the parameter. If the name is specified explicitly, as is necessary for multiple returns, no pre-defined <it>result</it> variable exists.

Clarify availability of 'result' for Tuple returns

The current phrasing

Within a contextual operation result represents the unique result parameter (if there is a unique declared result) or the
tuple containing the list of declared result parameters.

suggests that the synthetic Tuple return is accessible as result. This is not necessary and imposes considerable implementation difficulties since a new Tuple would need creation with each mutation.

Suggest the new wording.

Within a contextual operation for which no name is specified for a single return parameter, result is the implicit name of the parameter. If the name is specified explicitly, as is necessary for multiple returns, no pre-defined result variable exists.

Discussion

Yes

The QVTr AS and CS permit only a single pattern to be matched in the input (and output) domains. QVTc permits multi-matches.

This can be worked around with nested collections as demonstrated by RDomainToMBottomPredicateForEnforcement in RelToCore

remainingUnBoundDomainVars: Set(essentialocl::Variable);
predicatesWithVarBindings:Set(qvtbase::Predicate);

domain relations rdtVarsSeq:Sequence(Set(Element)) {
rdtSet:Set(Element) {
r:Relation{},
rd:RelationDomain{},
te:ObjectTemplateExp {bindsTo = v:Variable {}}
++ _
}
++ _
};

rdtVarsSeq->at(2) = predicatesWithoutVarBindings;
rdtVarsSeq->at(3) = unboundDomainVars;

The above is hard to read and uses nested collections in ways that are claimed to be unsupported. Much simpler to just write:

domain relations
r:Relation{},
rd:RelationDomain{},
te:ObjectTemplateExp {bindsTo = v:Variable {}},
predicatesWithVarBindings:Set(qvtbase::Predicate) {},
remainingUnBoundDomainVars: Set(essentialocl::Variable) {}
;

Suggest in the AS, RelationDomain.pattern changes to RelationDomain.patterns, RelationDomain.rootVariable moves to DomainPattern.rootVariable.

Suggest in the DomainCS CS, "pattern=DomainPatternCS" changes to "patterns+=DomainPatternCS (',' patterns+=DomainPatternCS)*"

As a consequence additional variables may be passed in a RelationCallExp. These can still be in rootVariable order.

QVTr Multi-root match

The QVTr AS and CS permit only a single pattern to be matched in the input (and output) domains. QVTc permits multi-matches.

This can be worked around with nested collections as demonstrated by RDomainToMBottomPredicateForEnforcement in RelToCore

remainingUnBoundDomainVars: Set(essentialocl::Variable);
predicatesWithVarBindings:Set(qvtbase::Predicate);

domain relations rdtVarsSeq:Sequence(Set(Element)) {
rdtSet:Set(Element) {
r:Relation{},
rd:RelationDomain{},
te:ObjectTemplateExp {bindsTo = v:Variable {}}
++ _
}
++ _
};

rdtVarsSeq->at(2) = predicatesWithoutVarBindings;
rdtVarsSeq->at(3) = unboundDomainVars;

The above is hard to read and uses nested collections in ways that are claimed to be unsupported. Much simpler to just write:

domain relations
r:Relation{},
rd:RelationDomain{},
te:ObjectTemplateExp {bindsTo = v:Variable {}},
predicatesWithVarBindings:Set(qvtbase::Predicate) {},
remainingUnBoundDomainVars: Set(essentialocl::Variable) {}
;

Suggest in the AS, RelationDomain.pattern changes to RelationDomain.patterns, RelationDomain.rootVariable moves to DomainPattern.rootVariable.

Suggest in the DomainCS CS, "pattern=DomainPatternCS" changes to "patterns+=DomainPatternCS (',' patterns+=DomainPatternCS)*"

As a consequence additional variables may be passed in a RelationCallExp. These can still be in rootVariable order.

Discussion

This seems to be a completely unnecessary restriction that caused the author of relToCore significant difficulties.

The only downside seems to be a slight increase in complexity of a RelationCallExp. No longer exactly one argument per domain, but linearizing the roots per domain is not exactly hard.

(The spelling in the discussion is that of the Eclipse QVTr prototype. Changing pattern to patterns is not appropriate for QVT1.x where property names tend to be simple-singulars. Perhaps a change to owned-plurals and so ownedPatterns may be appropriate for QVT 2.0.)

Exploiting this in relToCore should wait until an executable version has been debugged.

The full names of the QVT languages are a bit cumbersome and so QVTc/QVTo/QVTr are often found in non-OMG usage.

Suggest that the QVT specification at least define them and perhaps use them a little too.

QVTc/QVTo/QVTr acronyms

The full names of the QVT languages are a bit cumbersome and so QVTc/QVTo/QVTr are often found in non-OMG usage.

Suggest that the QVT specification at least define them and perhaps use them a little too.

8.1.4 states "a mapping operation is always a refinement of an implicit relation" but 8.2.15 defines "refinedRelation: Relation [0..1] The refined relation, if any." Clearly a contradiction.

8.1.4. provides the REL_PackageToSchema example of how an implicit relation might be defined, but no other examples or synthesis rules are supplied.

enforce and checkonly appear in the REL_PackageToSchema example indicating that these define execution mode of a QVTo trnasformation, but there does not appear to be any description of how a transformation might be executed to for instance update an output model. Perhaps the 'output' parameter is 'inout' for create/update but 'out' for create/overwrite.

8 Unclear mapping operation characteristics

8.1.4 states "a mapping operation is always a refinement of an implicit relation" but 8.2.15 defines "refinedRelation: Relation [0..1] The refined relation, if any." Clearly a contradiction.

8.1.4. provides the REL_PackageToSchema example of how an implicit relation might be defined, but no other examples or synthesis rules are supplied.

enforce and checkonly appear in the REL_PackageToSchema example indicating that these define execution mode of a QVTo trnasformation, but there does not appear to be any description of how a transformation might be executed to for instance update an output model. Perhaps the 'output' parameter is 'inout' for create/update but 'out' for create/overwrite.

Discussion

The aspiration that every QVTo mapping has an underlying QVTr relation is fairly dubious. But there is no point respecifying until some prototyping feedback is available.

The numerous problems identified in http://www.omg.org/archives/qvt-rtf/msg00094.html need to be addressed. Apologies too for the long email. This is a very difficult area where precision is important. Provision of
this resolution demonstrates the need for the resolution, but unfortunately the resolution has an
erroneous precision that will make QVTr 1.1 unimplementable whereas QVTr 1.0 is just mildly
ambiguous and conflicting.
Please do not include the resolution in QVT 1.1 without significantrework.
I found the definition of the "checkonly"/"enforce" to isCheckable/isEnforceable helpful, although
different to three of my own intuitive guesses based on attempts to avoid errors in ModelMorf and
Rel2Core examples.
The problems identified below are the result of a local review of the resolution. In the absence of a
coherent Environment semantics it has not been possible to perform a global review. In particular, I
was unable to review the specification for the arguably redundant bindsTo.
[Incidentally, the same resolution approach is needed for QVTc and QVTo].
Disambiguating rules
--------------------
The resolution uses a similar approach to the OCL 2.0 specification, but neglects to provide any
disambiguating rules although many are needed.
Environments
------------
OCL and the resolution employ environments to carry definitions specific to particular CS constructs,
so that a CS reference may be resolved with the aid of an environment or the AST.
In EssentialOCL, all definitions are resolvable with in the immutable AST with the exception of let and
iterator expressions for which a nestedEnvironment() is used to carry the extra variable declaration
with the aid of addElement(). The nestedEnvironment supports name occlusion from outer scopes.
In CompleteOCL, further definitions may be introduced by a definition constraint. The OCL
specification provides no precise insight into how an environment changes to accommodate the
definition. Should the name be added to the AST or to the environment? Is the name available for
forward reference?
Environment::addElement is defined as a query operation thereby inhibiting side effects. Consequently
usage such as
XX.env = YY.env.addElement(Z.name, Z.ast, false)
leaves the environment of YY unchanged and creates an extended environment for XX.
A series of such usages creates a series of progressively elaborated environments that support
backward but not forward referencing. This was not a clear requirement of the QVT 1.0 specification
and it prevents any variable declarations being introduced in an object template tree being resolved
through environments in other object template trees or predicate expressions.
QVT 1.2 RTF Report Page 173 of 190
Object Management Group
RTF/FTF Report
Imposition of no forward referencing seems very undesirable, particularly since the order of domains is
imposed by the model parameter order. Imagine a Java program in which all methods had to be
defined in a no-forward reference order.
As noted above, CompleteOCL neglected to define how AST definitions were added to the AST. QVTr
must solve this problem since QVTr defines a hierarchically scoped AST rather than an annotation of
a pre-existing AST.
I recommend a two stage approach. The inherited attributes section should first compute an
environment from the pushed-down parent environment augmented by pull-ups from child constructs
so that a complete immutable and consequently unambiguous environment is associated with each
construct and then pushed-down to the children. During the pull-up the environment acquires a
mapping from name to future-AST. During the push-down residual future-AST attributes are populated
to give a valid AST.
Reference resolution
--------------------
OCL uses lookup functions to resolve variable references. It is necessary either to overload the lookup
functions so that the the distinct QVTr variable definition sites can be located in the AST, or to use
some form of Environment::addElement where each definition is defined so that resolution in the
environment is possible.
Details
=======
Throughout, many disambiguating rules are needed to define illegal semantics. For instance "any" is
often used to select a syntactically valid value. Corresponding usage in the OCL specification has a
disambiguating rule to clarify what the consequence of not "one" is.
My current best attempt at Disambiguating Rules is attached.
Environment::addElement takes three arguments, the third being a mayBeImplicit argument. This has
been omitted throughout without explanation.
identifierCS
------------
OCL defines SimpleNameCS. A degenerate mapping from identifierCS to SimplenameCS is required.
topLevelCS
----------
The 'imported transformation' environment element is later referenced as 'imported transformations'.
Typo: TransformationListCS for transformationListCS in Synthesized attributes.
importListCS
------------
Semantics of import conflicts must be defined.
unitCS
------
Typo: ast is not a Set.
Surely the import is of packages (enumerations or operations)or at least transformations (QVTo
implementations) rather than necessarily relational-transformations?
transformationCS
----------------
QVT 1.2 RTF Report Page 174 of 190
Object Management Group
RTF/FTF Report
ownedTag is not synthesized.
keyDeclListCS
-------------
Typo: wrong font in synthesized attributes
modelDeclCS
-----------
The [B] grammar:
modelDeclCS ::= modelIdCS ':' '

'
is missing.
keyDeclCS
---------
Synthesized attributes appear to have experienced a copy and paste error while providing distinct part
and oppositePart left hand sides.
keyPropertyCS
-------------
The synthesized attributes poke the parent.
Suggest: it would be clearer for the parent to gather and distribute children similar to the relation/query
allocation by transformationCS.
relationCS
----------
Transformation.extends does not appear to be transitive.
topQualifierCS
--------------
Suggest: a boolean or enumerated value rather than a string.
domainListCS
------------
Typo: missing indentation.
primitiveTypeDomainCS
---------------------
isCheckable, isEnforceable not synthesized.
objectTemplateCS
----------------
Variable needs to be added to relation to provide a container.
Variable needs to be added to relation environment to provide visibility.
collectionTemplateCS
--------------------
Variable needs to be added to relation to provide a container.
Variable needs to be added to relation environment to provide visibility.
Suggest: last two if guards are redundant.
QVT 1.2 RTF Report Page 175 of 190
Object Management Group
RTF/FTF Report
restCS
------
Variable needs to be added to relation to provide a container.
Non-_ named variable needs to be added to relation environment to provide visibility.
memberCS
--------
Variable needs to be added to relation to provide a container.
Non-_ named variable needs to be added to relation environment to provide visibility.
whenCS
------
predicateListCS should be optional.
whereCS
-------
predicateListCS should be optional.
ExtOclExpressionCS
------------------
This is not present in the QVTr or OCL grammar.
Presumably it represents the QVTr extension to OCL's OclExpressionCS.
However it is an extension, since at least RelationCallExpCS can be used in an ordinary
OclExpressionCS using "not" or "and".
[A], [B], [C] should therefore follow on from OCL's
[A], [B], [C]..., [I].
RelationCallExpressionCS
------------------------
How is a RelationCallExpressionCS distinguished from an OperationCallExpCS?

QVTr syntax mapping (correction to Issue 10646 resolution)

The numerous problems identified in http://www.omg.org/archives/qvt-rtf/msg00094.html need to be addressed. Apologies too for the long email. This is a very difficult area where precision is important. Provision of
this resolution demonstrates the need for the resolution, but unfortunately the resolution has an
erroneous precision that will make QVTr 1.1 unimplementable whereas QVTr 1.0 is just mildly
ambiguous and conflicting.
Please do not include the resolution in QVT 1.1 without significantrework.
I found the definition of the "checkonly"/"enforce" to isCheckable/isEnforceable helpful, although
different to three of my own intuitive guesses based on attempts to avoid errors in ModelMorf and
Rel2Core examples.
The problems identified below are the result of a local review of the resolution. In the absence of a
coherent Environment semantics it has not been possible to perform a global review. In particular, I
was unable to review the specification for the arguably redundant bindsTo.
[Incidentally, the same resolution approach is needed for QVTc and QVTo].
Disambiguating rules
--------------------
The resolution uses a similar approach to the OCL 2.0 specification, but neglects to provide any
disambiguating rules although many are needed.
Environments
------------
OCL and the resolution employ environments to carry definitions specific to particular CS constructs,
so that a CS reference may be resolved with the aid of an environment or the AST.
In EssentialOCL, all definitions are resolvable with in the immutable AST with the exception of let and
iterator expressions for which a nestedEnvironment() is used to carry the extra variable declaration
with the aid of addElement(). The nestedEnvironment supports name occlusion from outer scopes.
In CompleteOCL, further definitions may be introduced by a definition constraint. The OCL
specification provides no precise insight into how an environment changes to accommodate the
definition. Should the name be added to the AST or to the environment? Is the name available for
forward reference?
Environment::addElement is defined as a query operation thereby inhibiting side effects. Consequently
usage such as
XX.env = YY.env.addElement(Z.name, Z.ast, false)
leaves the environment of YY unchanged and creates an extended environment for XX.
A series of such usages creates a series of progressively elaborated environments that support
backward but not forward referencing. This was not a clear requirement of the QVT 1.0 specification
and it prevents any variable declarations being introduced in an object template tree being resolved
through environments in other object template trees or predicate expressions.
QVT 1.2 RTF Report Page 173 of 190
Object Management Group
RTF/FTF Report
Imposition of no forward referencing seems very undesirable, particularly since the order of domains is
imposed by the model parameter order. Imagine a Java program in which all methods had to be
defined in a no-forward reference order.
As noted above, CompleteOCL neglected to define how AST definitions were added to the AST. QVTr
must solve this problem since QVTr defines a hierarchically scoped AST rather than an annotation of
a pre-existing AST.
I recommend a two stage approach. The inherited attributes section should first compute an
environment from the pushed-down parent environment augmented by pull-ups from child constructs
so that a complete immutable and consequently unambiguous environment is associated with each
construct and then pushed-down to the children. During the pull-up the environment acquires a
mapping from name to future-AST. During the push-down residual future-AST attributes are populated
to give a valid AST.
Reference resolution
--------------------
OCL uses lookup functions to resolve variable references. It is necessary either to overload the lookup
functions so that the the distinct QVTr variable definition sites can be located in the AST, or to use
some form of Environment::addElement where each definition is defined so that resolution in the
environment is possible.
Details
=======
Throughout, many disambiguating rules are needed to define illegal semantics. For instance "any" is
often used to select a syntactically valid value. Corresponding usage in the OCL specification has a
disambiguating rule to clarify what the consequence of not "one" is.
My current best attempt at Disambiguating Rules is attached.
Environment::addElement takes three arguments, the third being a mayBeImplicit argument. This has
been omitted throughout without explanation.
identifierCS
------------
OCL defines SimpleNameCS. A degenerate mapping from identifierCS to SimplenameCS is required.
topLevelCS
----------
The 'imported transformation' environment element is later referenced as 'imported transformations'.
Typo: TransformationListCS for transformationListCS in Synthesized attributes.
importListCS
------------
Semantics of import conflicts must be defined.
unitCS
------
Typo: ast is not a Set.
Surely the import is of packages (enumerations or operations)or at least transformations (QVTo
implementations) rather than necessarily relational-transformations?
transformationCS
----------------
QVT 1.2 RTF Report Page 174 of 190
Object Management Group
RTF/FTF Report
ownedTag is not synthesized.
keyDeclListCS
-------------
Typo: wrong font in synthesized attributes
modelDeclCS
-----------
The [B] grammar:
modelDeclCS ::= modelIdCS ':' '

'
is missing.
keyDeclCS
---------
Synthesized attributes appear to have experienced a copy and paste error while providing distinct part
and oppositePart left hand sides.
keyPropertyCS
-------------
The synthesized attributes poke the parent.
Suggest: it would be clearer for the parent to gather and distribute children similar to the relation/query
allocation by transformationCS.
relationCS
----------
Transformation.extends does not appear to be transitive.
topQualifierCS
--------------
Suggest: a boolean or enumerated value rather than a string.
domainListCS
------------
Typo: missing indentation.
primitiveTypeDomainCS
---------------------
isCheckable, isEnforceable not synthesized.
objectTemplateCS
----------------
Variable needs to be added to relation to provide a container.
Variable needs to be added to relation environment to provide visibility.
collectionTemplateCS
--------------------
Variable needs to be added to relation to provide a container.
Variable needs to be added to relation environment to provide visibility.
Suggest: last two if guards are redundant.
QVT 1.2 RTF Report Page 175 of 190
Object Management Group
RTF/FTF Report
restCS
------
Variable needs to be added to relation to provide a container.
Non-_ named variable needs to be added to relation environment to provide visibility.
memberCS
--------
Variable needs to be added to relation to provide a container.
Non-_ named variable needs to be added to relation environment to provide visibility.
whenCS
------
predicateListCS should be optional.
whereCS
-------
predicateListCS should be optional.
ExtOclExpressionCS
------------------
This is not present in the QVTr or OCL grammar.
Presumably it represents the QVTr extension to OCL's OclExpressionCS.
However it is an extension, since at least RelationCallExpCS can be used in an ordinary
OclExpressionCS using "not" or "and".
[A], [B], [C] should therefore follow on from OCL's
[A], [B], [C]..., [I].
RelationCallExpressionCS
------------------------
How is a RelationCallExpressionCS distinguished from an OperationCallExpCS?

Discussion

These issues cannot be sensibly addressed until the auto-generation being pioneered for OCL can be re-used.

• • QVTo Standard Library. Some operation's returned values would better return the TemplateParameterType **
    Several stdlib operations would be better changed to avoid doing unnecessary castings on the returned value of the operation.

For AnyType::oclAsType(oclType) operation I could write:

var aClass : Class := anotherVar.oclAsType(Class);

However, for Model::objectsOfType(OclType) operation I can't do:

var aClassSet : Set(Class) := aModel.objectsOfType(Class);

I have to do the following, instead:

var aClassSet : Set(Class) := aModel.objectsOfType(Class).oclAsType(Set(Class));

Therefore, for end-user usability, I propose exploiting TemplateParameterType and changing some QVTo Standard Library operations

Element::subobjectsOfType(OclType) : Set(T)
Element::allSubobjects(OclType) : Set(T)
Element::subobjectsOfKind(OclType) : Set(T)
Element::allSubobjectsOfKind(OclType) : Set(T)
Element::clone() : T
Element::deepclone() : T
Model::objectsOfType(OclType) : Set(T)
Model::copy() : T
Model::createEmptyModel(): T

Note: this approach is made in the Object::asOrderedTuple() : OrderedTuple(T) operation.

8.3.5.7 createEmptyModel static Model::createEmptyModel() : Model Creates and initializes a model of the given type. This operation is useful when creating intermediate models within a transformation.

QVTo Standard Library. Some operation's returned values would better return the TemplateParameterType **
Several stdlib operations would be better changed to avoid doing unnecessary castings on the returned value of the operation.

For AnyType::oclAsType(oclType) operation I could write:

var aClass : Class := anotherVar.oclAsType(Class);

However, for Model::objectsOfType(OclType) operation I can't do:

var aClassSet : Set(Class) := aModel.objectsOfType(Class);

I have to do the following, instead:

var aClassSet : Set(Class) := aModel.objectsOfType(Class).oclAsType(Set(Class));

Therefore, for end-user usability, I propose exploiting TemplateParameterType and changing some QVTo Standard Library operations

Element::subobjectsOfType(OclType) : Set(T)
Element::allSubobjects(OclType) : Set(T)
Element::subobjectsOfKind(OclType) : Set(T)
Element::allSubobjectsOfKind(OclType) : Set(T)
Element::clone() : T
Element::deepclone() : T
Model::objectsOfType(OclType) : Set(T)
Model::copy() : T
Model::createEmptyModel(): T

Note: this approach is made in the Object::asOrderedTuple() : OrderedTuple(T) operation.

Discussion

subobjectsOfType etc duplicate QVT13-53

clone/deepclone merged into QVT13-53

Model::copy/createEmptyModel - NO. Model is the correct return type.

his section (8.3.2) is very confusing for the reader. What does synonym means ? - Is just like a shorthand of the concrete syntax ? - Is just a new operation included in the QVTo Standard Library semantically equivalent?. Ideally, just one type/operation must exist (the OCL predefined type and the operation of an OCL predefined type), so that, writing Void (a type) or asType (operation) should produce the same AST as if I write OclVoid or oclAsType. With this idea, I'm really puzzled with the third paragraph of the section. Please revise this section so that it is less confusing.

section (8.3.2) is very confusing for the reader

This section (8.3.2) is very confusing for the reader. What does synonym means ? - Is just like a shorthand of the concrete syntax ? - Is just a new operation included in the QVTo Standard Library semantically equivalent?. Ideally, just one type/operation must exist (the OCL predefined type and the operation of an OCL predefined type), so that, writing Void (a type) or asType (operation) should produce the same AST as if I write OclVoid or oclAsType. With this idea, I'm really puzzled with the third paragraph of the section. Please revise this section so that it is less confusing.

Discussion - Ocl type prefixes

As an OCL user, I find the attempt and recommendation to hide the underlying OCL functionality very confusing, particularly when it introduces potential collisions whose prevention is inaccurately described.

Since the description of the M1 types does not enumerate them, I am not clear which types are intended. If Eclipse QVTo is any guide, it would be OclAny/Any, OclVoid/Void, but not OclElement, OclInvalid, OclMessage, OclState, OclType.

If this policy is to remain recommended, it must also apply to all of the above. I find the injection of Element and Type into the user's namespace particularly offensive.

I recommend that the recommendation be reversed, so that we just observe that QVTo implementations offering backward compatibility may permit the seven enumerated synonyms to be used as shortforms of the full spelling.

Discussion - ocl operation prefixes

The references to the "Ocl" prefix are inaccurate for operations where it is "ocl".

As an OCL user, I find the attempt and recommendation to hide the underlying OCL functionality a bit confusing, particularly since the abbreviated forms are used very little used in the editorial text and have no definition.

I recommend eliminating their editorial usage and defining them as synonyms that a QVTo implementation offering backward compatibility may support as an alternate spelling that is mapped to the full spelling when used within the abstract syntax

Discussion - Operation name synonyms

This section was written for OCL 2.0 which has no String::+ operator. String::+ is available in OCL 2.4 so there is no need for QVT 1.2 to define it as a synonym.The other mathematical operators have always been part of OCL, so the purpose of enumerating them is unclear and indeed confusing. Is there actually an operation with the spelling "operator+"? I see no need for one. In the concrete syntax, "+" is fully available. In UML and OCL "+" is a permissible internal spelling for an operation name. The _'+' escape enables its use in arbitrary name contexts.

The only question is whether a particular implementation might choose to use "plus" as a spel;ling to avoid problems in languages that do not allow arbitrary spellings. But that is an implementation language-specific choice that does not affect QVT or OCL; perhaps relevant to the Java-binding for OCL specification if written. FWIW, Eclipse QVTo has no "operator+" operation.

Conversely, is the specification actually saying that "plus" may be used rather than "+"? This might be a way of elevating an implementation language workaround into the source, which seems rather perverse. FWIW, Eclipse QVTo provides no "divide" operation.

The operation synonym sentences can be eliminated without any loss.

Creating aliases seems to be a good idea specially when dealing with complex types. However, for that purpose, it is not clear to me that we need to create a (useless) typedef just to represent an alias in the concrete syntax. In practice aliases are very useful when writing transformations (in concrete syntax), but they are useless in the abstract syntax (extra unnecessary classes in modules). One may still think that a typedef as an alias (no extra condition) may be needed to add operations to existing types, as specification suggests doing in order to include new operations for the OCL standard library predefined types. However, this still can be done by means of helpers. Suggestions: - Remove the statements related to aliases in the section 8.2.2.24 - Make Typedef.condition reference be mandatory in Figure 8.7 since, now, Typedef in the abstract syntax will be used to constrain existing types. - Add statements in the concrete syntax section, to clarify the use of aliases to rename existing types. Maybe, a new keyword (like alias) could be included to avoid confusions with the typedef one. - Change the grammar to adapt it to these suggestions. - Clarify in the Standard Library section that helpers will be used to add new operations to the OCL Std Lib predefined type.

Typedef aliases issue

Creating aliases seems to be a good idea specially when dealing with complex types. However, for that purpose, it is not clear to me that we need to create a (useless) typedef just to represent an alias in the concrete syntax. In practice aliases are very useful when writing transformations (in concrete syntax), but they are useless in the abstract syntax (extra unnecessary classes in modules). One may still think that a typedef as an alias (no extra condition) may be needed to add operations to existing types, as specification suggests doing in order to include new operations for the OCL standard library predefined types. However, this still can be done by means of helpers. Suggestions: - Remove the statements related to aliases in the section 8.2.2.24 - Make Typedef.condition reference be mandatory in Figure 8.7 since, now, Typedef in the abstract syntax will be used to constrain existing types. - Add statements in the concrete syntax section, to clarify the use of aliases to rename existing types. Maybe, a new keyword (like alias) could be included to avoid confusions with the typedef one. - Change the grammar to adapt it to these suggestions. - Clarify in the Standard Library section that helpers will be used to add new operations to the OCL Std Lib predefined type.

Discussion

Ignoring the QVT Standard Library usage eliminated by the resolution of Issue 13252, http://solitaire.omg.org/browse/QVT13-13, there are two distinct usages of typedef.

a) to create an alias
b) to create a new constrained type

Discussion - alias

If an alias introduces a new type as the inheritance of Typedef from Type would suggest, we have semantic nightmares. Which type conforms to which? How do we ensure that all functionality understands the typedef equivalence? As observed, a typedef is a concrete syntax text macro that should be eliminated during conversion to the abstract syntax to avoid ambiguities. However an AS concept is needed to persist the macro. A Tag is sufficient.

Discussion - constrained type

The typedef syntax with a constraint introduces a new type. However all the important semantics of this type are unspecified. What are the superclasses of this type? What does it conform to? How do its operations and properties interact with the unconstrained type? Can a Constrained Integer be used as a Sequence index?

The only example of a constrained type is:

typedef TopLevelPackage = Package [container()=null];

suggesting that perhaps self is to be interpreted as the type instance. If container() is the proposed oclContainer() then the example is particularly interesting. If a TopLevelPackage is assigned to a container, how does the type change? Are all assignments supposed to have a predicate evaluation to ensure that typedef constraints are not violated? Are assigned types to be changed by an assignment? Does an assignment that violates a constrained type constraint render a transformation invalid and require some form of failure?

Look at it another way. Suppose we used bland UML. We can add an invariant to a class, which can inherit/aggregate another type to re-use it. We and general purpose tooling understand these semantics. What does a typedef constrained type offer that is new? just magic undefined semantics that cannot be used with any non-QVTo tool.

The constrained typedef appears to be a clear case of misguided syntax sugar. If metamodeling syntax sugar is really needed, let it have a clear relationship to UML.

Conclusion

The alias usage is a concrete syntax synonym that has no need of a Typedef class; a Tag will do.

The constrained type is underspecified syntax sugar. The Typedef class can be deprecated.

FWIW: neither of these forms of typedef is implemented by Eclipse QVTo.

Suggestion: In the Operational Mappings language, element creation and element attachment/detachment to/from an extent should be seen and treated as two different and independent activities. Once an element is created via an ObjectExp, this element is usually attached to another element immediately afterwards, which becomes its container, so the ObjectExp semantics of assigning it to an explicit or inferred extent becomes an unnecessary overhead. And also there are other times when we need to assign to an extent some model elements that may have been created at an unrelated time. They could even exist prior to the transformation execution. A case where this is relevant is with the result of a 'Element::clone()' or 'Element::deepclone()' operation execution. Is it expected that these model elements must belong by default to the same model as the original? How to clone parts of an extent with direction kind == 'in'? How to make them become part of the Set of root elements of another, different extent?

element creation and element attachment/detachment to/from an extent

Suggestion: In the Operational Mappings language, element creation and element attachment/detachment to/from an extent should be seen and treated as two different and independent activities. Once an element is created via an ObjectExp, this element is usually attached to another element immediately afterwards, which becomes its container, so the ObjectExp semantics of assigning it to an explicit or inferred extent becomes an unnecessary overhead. And also there are other times when we need to assign to an extent some model elements that may have been created at an unrelated time. They could even exist prior to the transformation execution. A case where this is relevant is with the result of a 'Element::clone()' or 'Element::deepclone()' operation execution. Is it expected that these model elements must belong by default to the same model as the original? How to clone parts of an extent with direction kind == 'in'? How to make them become part of the Set of root elements of another, different extent?

Discussion

This seems to be a misunderstanding. ObjectExp does not require a created object to be attached to an inferred extent.
This functionality seems to be present already.
Creation without attachment requires a null or null-valued extent during creation.
Attachment after creation can occur through an ObjectExp update to the required extent.
Re-attachment presumably occurs through an ObjectExp update to the new extent.
Detachment then naturally occurs through an ObjectExp update to a null-valued extent.

This issue inspired the provision of a better desciption of extents. The resolution is therefore merged with that of QVT-36.

Major Problem:
(1) The current abstract syntax of ImperativeOCL introduces a couple of
unclear situations. This may lead to incompatible QVT implementations.
Further Problems:
(2) Control flow constructs introduced by ImperativeOCL are redundant
compared with existing conventional OCL constructs.
(3) Several OCL equivalence rules break when ImperativeOCL is present.
Detailed problem description:
(1) The current abstract syntax of ImperativeOCL introduces a couple of
unclear situations. This may lead to incompatible QVT implementations.
In the abstract syntax, ImperativeOCL expressions / statements are
inherited from OclExpression. Therefore, conventional OCL
expressions may (and will) contain sub-expressions that are
actually ImperativeOCL expressions.
In conventional OCL, the interpretation of an expression under a
given environment is a value. In ImperativeOCL, the interpretation
of an expression is a value and a new environment
(state,variables). This extended interpretation is not given for
conventional OCL expressions, leading to undefined operational
semantics of those expressions.
Example: Given the following compute expression:
compute(z:Boolean) {
var x : Boolean := true
var y : Boolean := true
if ((x:=false) and (y:=false))

z := y
}
What is the value of this expression: is it true or false (It
depends on whether the 'and' operator is evaluated short-circuit
or strict). The situation is similar for the evaluation of the
other logical connectives, forAll, and exists when these
expressions contain imperative sub-expressions.
(2) Control flow constructs introduced by ImperativeOCL are redundant
compared with existing conventional OCL constructs.
Some of the new language features in ImperativeOCL such as forEach
and the imperative conditional are not really necessary. Their
effect can already be achieved using conventional OCL expressions:
For example:
company.employees->forEach(c)

has the same effect as
company.employees->iterate(c; r:OclAny=Undefined |
c.salary := c.salary * 1.1
)
and
if ( x < 0 )

else

endif is the same as
if x < 0 then x := 0 else x := 1 endif
(3) Several OCL equivalence rules break when ImperativeOCL is present.
In conventional OCL, several equivalence rules well known from
logic hold. Allowing OCL expression to contain imperative
sub-expressions breaks these equivalence rules.
Examples:
let x=e1 in e2 equiv. e2

e1 and e2 equiv. e2 and e1
These equivalences do not necessarily hold if e1 or e2 are allowed
to have side-effects.
Proposed solution:
(A) - (The cheap solution.) State textually that conventional OCL
expressions (as described in the OMG OCL spec.) are not
allowed to have side effects unless used as part of a top
level ImperativeOCL expression. Therefore, even in a system
supporting ImperativeOCL, class invariants, and pre- and
postconditions (e.g.) will not be allowed to contain
ImperativeOCL sub-expressions.
State explicitly that the redundant flow control statements
have been introduced (solely) to write concise imperative
programs and that the side-effect free forms of conditional
evaluation ( 'if-then-else-endif' and 'iterate' ) shall not be
used to program side-effects (instead, the ImperativeOCL forms
shall be used).
(B) - (Major rework.) Rework the abstract syntax to reuse OCL
expressions by composition rather than by inheritance.
Imperative expressions ( => rename to 'statements' ) then may
contain sub-statements and OCL expressions; OCL expressions
are reused unchanged from the OCL spec (no imperative
sub-expressions, no side-effects).
These issues have been discussed on the MoDELS 2008 OCL Workshop,
more details can be found at
http://www.fots.ua.ac.be/events/ocl2008/PDF/OCL2008_9.pdf

current abstract syntax of ImperativeOCL introduces a couple of unclear situations

Major Problem:
(1) The current abstract syntax of ImperativeOCL introduces a couple of unclear situations. This may lead to incompatible QVT implementations.
Further Problems:
(2) Control flow constructs introduced by ImperativeOCL are redundant compared with existing conventional OCL constructs.
(3) Several OCL equivalence rules break when ImperativeOCL is present.
Detailed problem description:
(1) The current abstract syntax of ImperativeOCL introduces a couple of unclear situations. This may lead to incompatible QVT implementations.
In the abstract syntax, ImperativeOCL expressions / statements are inherited from OclExpression. Therefore, conventional OCL expressions may (and will) contain sub-expressions that are actually ImperativeOCL expressions. In conventional OCL, the interpretation of an expression under a given environment is a value. In ImperativeOCL, the interpretation of an expression is a value and a new environment (state,variables). This extended interpretation is not given for conventional OCL expressions, leading to undefined operational semantics of those expressions.
Example: Given the following compute expression:
compute(z:Boolean) {
var x : Boolean := true
var y : Boolean := true
if ((x:=false) and (y:=false))

z := y
}
What is the value of this expression: is it true or false (It depends on whether the 'and' operator is evaluated short-circuit or strict). The situation is similar for the evaluation of the other logical connectives, forAll, and exists when these expressions contain imperative sub-expressions.

(2) Control flow constructs introduced by ImperativeOCL are redundant compared with existing conventional OCL constructs.
Some of the new language features in ImperativeOCL such as forEach and the imperative conditional are not really necessary. Their effect can already be achieved using conventional OCL expressions:
For example:
company.employees->forEach(c)

has the same effect as
company.employees->iterate(c; r:OclAny=Undefined |
c.salary := c.salary * 1.1
)
and
if ( x < 0 )

else

endif is the same as
if x < 0 then x := 0 else x := 1 endif

(3) Several OCL equivalence rules break when ImperativeOCL is present.
In conventional OCL, several equivalence rules well known from logic hold. Allowing OCL expression to contain imperative sub-expressions breaks these equivalence rules.
Examples:
let x=e1 in e2 equiv. e2

e1 and e2 equiv. e2 and e1
These equivalences do not necessarily hold if e1 or e2 are allowed to have side-effects.
Proposed solution:
(A) - (The cheap solution.) State textually that conventional OCL expressions (as described in the OMG OCL spec.) are not allowed to have side effects unless used as part of a top level ImperativeOCL expression. Therefore, even in a system supporting ImperativeOCL, class invariants, and pre- and postconditions (e.g.) will not be allowed to contain ImperativeOCL sub-expressions.
State explicitly that the redundant flow control statements have been introduced (solely) to write concise imperative programs and that the side-effect free forms of conditional evaluation ( 'if-then-else-endif' and 'iterate' ) shall not be used to program side-effects (instead, the ImperativeOCL forms shall be used).
(B) - (Major rework.) Rework the abstract syntax to reuse OCL expressions by composition rather than by inheritance.
Imperative expressions ( => rename to 'statements' ) then may contain sub-statements and OCL expressions; OCL expressions are reused unchanged from the OCL spec (no imperative sub-expressions, no side-effects).
These issues have been discussed on the MoDELS 2008 OCL Workshop, more details can be found at http://www.fots.ua.ac.be/events/ocl2008/PDF/OCL2008_9.pdf

Discussion - 1 short circuit semantics

It is possible to weasel out of this narrowly described problem by observing that OCL only supports short-circuit semantics once it is proven that no invalid lurks beyond the short-circuit. An extension to this could inhibit short-circuit semantics if any imperative expression is involved.

Discussion - (2) redundant control flow

The first suggestion is contradictory. The replacement for forEach is noticeably longer and uses a side effect within an iterate, which the author argues against elsewhere.

The second suggestion incorrectly assumes that an else clause is present. The imperative if (SwitchExp) allows one-sided functionality including selective control flow adjustment.

Discussion - (3) broken equivalence

The example is unreadable, but the point is unarguable; imperative semantics within functional semantics subvert the functional.

(A) - (The cheap solution.) is simple and adopted here.

(B) - (Major rework.) is a breaking structural change and so forked off to http://solitaire.omg.org/browse/QVT13-87 for consideration by QVT 2.0.

In 7.11.3.6 (and 7.8) the ordering of parameters is implied but not explicit. Presumably the first is the output (enforced direction) so:

package QVTBase

context TypedModel
def: allUsedPackage() : Set(EMOF::Package)
= self.dependsOn.allUsedPackage()>asSet()>union(self.usedPackage)

endpackage

package QVTRelation

context RelationImplementation
inv RootNodeIsBoundToRootVariable : self.inDirectionOf.allUsedPackage()>includes(self.impl.ownedParameter>at(1).type._package)

endpackage

This is not satisfied by almost the last line of RelToCore in 10.3:

enforce domain core me:OclExpression{} implementedby CopyOclExpession(re, me);

which seems to have output second.

-----------------------------------------------------

More significantly CopyOclExpession(re, me) is not meaningful as a black box signature, since it is a static function and EMOF has no way to define static functions. Perhaps it is a query, for which the declaration was omitted from the example. If this is the case, it should be made clear that black box operations must be declared internally as queries and bound externally to static operations of the transformation.

This semantic clumsiness could be resolved, if, within a transformation,
relation, domain or query, self is bound to a transformation instance. Queries would then be normal non-static operations of the transformation (class) and the implementedBy operation call would be a normal implicit call via self of a non-static transformation operation or query. (A RelationCallExp would also deviate less from OCL than it currently does.) This would also allow a transformation to extend a Utility class that provided the required black box operations. Since queries and relations are declarative, it is not obvious that there need be any prohibition on the content of an extended Class; if the Class has properties, these cannot mutate during a query or relation match, so the properties are ok; they might even permit useful behavioural tailoring. For instance an 'inherited' Boolean mangledNames property could influence the mapping of names between input and output.

The RelToCore example can then be mended by declaring that:

RelToCore(...) extends utils::CopyUtilities

and externally binding the utils model name to a package that has a CopyUtilities class with suitable a CopyOclExpession operation.

7.11.3.6 (and 7.11.1.1) BlackBox operation signature difficulties

In 7.11.3.6 (and 7.8) the ordering of parameters is implied but not explicit. Presumably the first is the output (enforced direction) so:

package QVTBase

context TypedModel
def: allUsedPackage() : Set(EMOF::Package)
= self.dependsOn.allUsedPackage()>asSet()>union(self.usedPackage)

endpackage

package QVTRelation

context RelationImplementation
inv RootNodeIsBoundToRootVariable : self.inDirectionOf.allUsedPackage()>includes(self.impl.ownedParameter>at(1).type._package)

endpackage

This is not satisfied by almost the last line of RelToCore in 10.3:

enforce domain core me:OclExpression{} implementedby CopyOclExpession(re, me);

which seems to have output second.

-----------------------------------------------------

More significantly CopyOclExpession(re, me) is not meaningful as a black box signature, since it is a static function and EMOF has no way to define static functions. Perhaps it is a query, for which the declaration was omitted from the example. If this is the case, it should be made clear that black box operations must be declared internally as queries and bound externally to static operations of the transformation.

This semantic clumsiness could be resolved, if, within a transformation,relation, domain or query, self is bound to a transformation instance. Queries would then be normal non-static operations of the transformation (class) and the implementedBy operation call would be a normal implicit call via self of a non-static transformation operation or query. (A RelationCallExp would also deviate less from OCL than it currently does.) This would also allow a transformation to extend a Utility class that provided the required black box operations. Since queries and relations are declarative, it is not obvious that there need be any prohibition on the content of an extended Class; if the Class has properties, these cannot mutate during a query or relation match, so the properties are ok; they might even permit useful behavioural tailoring. For instance an 'inherited' Boolean mangledNames property could influence the mapping of names between input and output.

The RelToCore example can then be mended by declaring that:

RelToCore(...) extends utils::CopyUtilities

and externally binding the utils model name to a package that has a CopyUtilities class with suitable a CopyOclExpession operation.

Discussion

Binding self to a Transformation instance may conflict with self as the context in a query. Much safer to bind this; just like QVTo.

Using this as the source of an OperationCallExp of a query eliminates an incompatibility with OCL. But there is no need bloat every RelationCallExp with an extra VariableExp for this.

Any enhancement here needs prototyping.

The whole Concrete Syntax section deserves a much more substantial description.

In particular...

The mechanism by which a package name is located is unresolved, perhaps deliberately,
but the omission should be explicit.

What constraints exist on forward referencing of names?

Transformations and mappings could be ordered so that forward references are avoided,
but large modules benefit from an alphabetical ordering of elements, so requiring
a parser friendly order is not user friendly.

9.18 Undefined semantics

The whole Concrete Syntax section deserves a much more substantial description.

In particular...

The mechanism by which a package name is located is unresolved, perhaps deliberately,
but the omission should be explicit.

What constraints exist on forward referencing of names?

Transformations and mappings could be ordered so that forward references are avoided,
but large modules benefit from an alphabetical ordering of elements, so requiring
a parser friendly order is not user friendly.

Discussion

This isn't going to happen until a QVT 2.0 rewrite adopts the autogeneration approaches underway for OCL.

I’m having trouble with the semantics of DELETE on p. 189 of the QVT Specification (v1.1). It reads in part:

FORALL OBJVAR IN MAKESET(<DOMAIN_K_VARIABLE_SET>) (
…
AND BELONGSTO(OBJVAR, MAKESET(<DOMAIN_K_VARIABLE_SET>))

I guess I don’t understand MAKESET and BELONGSTO. First of all, <DOMAIN_K_VARIABLE_SET> is already a set, so what’s the MAKESET function do? Second, the FORALL iterates OBJVAR over the results of the same MAKESET that BELONGSTO tests. So how can BELONGSTO be false? That is, I would assume BELONGSTO is defined as follows:

BELONGSTO(e, S) º e ÎS

except that under this definition the expression above is always satisfied.

Any and all help appreciated. Thank you very much.

Specification of deletion semantics

I’m having trouble with the semantics of DELETE on p. 189 of the QVT Specification (v1.1). It reads in part:

FORALL OBJVAR IN MAKESET(<DOMAIN_K_VARIABLE_SET>) (
…
AND BELONGSTO(OBJVAR, MAKESET(<DOMAIN_K_VARIABLE_SET>))

I guess I don’t understand MAKESET and BELONGSTO. First of all, <DOMAIN_K_VARIABLE_SET> is already a set, so what’s the MAKESET function do? Second, the FORALL iterates OBJVAR over the results of the same MAKESET that BELONGSTO tests. So how can BELONGSTO be false? That is, I would assume BELONGSTO is defined as follows:

BELONGSTO(e, S) º e ÎS

except that under this definition the expression above is always satisfied.

Any and all help appreciated. Thank you very much.

Discussion

Well this is very embarrassing. I don't understand it all either, and I certainly do not understand why it is not written using OCL.

I'm not convinced that QVTr has any delete semantics since declaratively delete occurs as an absence of creation.

Must await the Eclipse QVTr prototype.

The abstract syntax of QVTr allows a rule to be an override of another rule.

Rule::overrides: Rule [0..1]
The rule that this rule overrides.

The concrete syntax of QVT allows it too:

<relation> ::= ['top'] 'relation' <identifier>
['overrides' <identifier>]
'

'

However, the only semantics I can see for 'overrides' is in clause 7.11.1.4 that says:

"A rule may conditionally override another rule. The overriding rule is executed in place of the overridden rule when the overriding conditions are satisfied. The exact semantics of overriding are subclass specific. "

Questions:

1- Whtat are the overriding conditions? are they implied or specified and if so how?

2- I have not seen any other discussion of overrding in a subclass or Rule so not sure what is meant by "The exact semantics of overriding are subclass specific"?

3- I have not seen any example of using 'overrides' what so ever in the spec, shouldn't there be one?

4 - What is the semantics of overriding? is it related to inheritance in the OO sense ? I think QVTr needs a good "inheritance" model where you can relations can be called polymorphically.

Rule Overriding in QVTr

The abstract syntax of QVTr allows a rule to be an override of another rule.

Rule::overrides: Rule [0..1]
The rule that this rule overrides.

The concrete syntax of QVT allows it too:

<relation> ::= ['top'] 'relation' <identifier>
['overrides' <identifier>]
'

'

However, the only semantics I can see for 'overrides' is in clause 7.11.1.4 that says:

"A rule may conditionally override another rule. The overriding rule is executed in place of the overridden rule when the overriding conditions are satisfied. The exact semantics of overriding are subclass specific. "

Questions:

1- What are the overriding conditions? are they implied or specified and if so how?

2- I have not seen any other discussion of overriding in a subclass or Rule so not sure what is meant by "The exact semantics of overriding are subclass specific"?

3- I have not seen any example of using 'overrides' what so ever in the spec, shouldn't there be one?

4 - What is the semantics of overriding? is it related to inheritance in the OO sense ? I think QVTr needs a good "inheritance" model where you can relations can be called polymorphically.

Discussion

This now a much better understood field:

M. Wimmer, G. Kappel, A. Kusel, W. Retschitzegger, J. Schönböck, W. Schwinger, D. Kolovos, R. Paige, M. Lauder, A. Schürr, D. Wagelaar. Surveying Rule Inheritance in Model-to-Model Transformation Languages. In Journal of Object Technology, vol. 11, no. 2, 2012, pages 3:1–46. doi:10.5381/jot.2012.11.2.a3

QVTBase currently imposes single overriding which is unnecessary, and causes a problem for an application that needs to extend when B1 overrides A and B2 overrides A. The extending mapping needs C to override both B1 and B2 to ensure that C occludes both B1 and B2. Therefore we need multi-overrides.

Oops. A QVTc Mapping::refines is an alternative concept that competes without comment with the inherited Rule::overrides. relToCore completely ignores the semantics of overrides translation. Relation::overrides should probably map across unchanged to Mapping::overrides. QVTc and QVTr semantics are unchanged.

Fixing this textually without a prototype is too much of change.

"The starting point is a major flaw in all three QVT languages at present and enabled Perdita Stevens to correctly conclude in http://www.springerlink.com/content/9x368617317l3q87/ that QVTr and QVTc are incompatible. QVT currently provides no way to distinguish whether for instance a check-mode transformation is a query of whether a transformed input pattern can be discovered in the output (e.g. a database lookup), or a validation that the transformed input exactly matches the output (e.g. an already transformed check). Both facilities are useful and so when a QVT transformation is invoked the invoker needs to specify what I call the 'rooting' condition in addition to the direction

Unclear transformation rooting condition

"The starting point is a major flaw in all three QVT languages at present and enabled Perdita Stevens to correctly conclude in http://www.springerlink.com/content/9x368617317l3q87/ that QVTr and QVTc are incompatible. QVT currently provides no way to distinguish whether for instance a check-mode transformation is a query of whether a transformed input pattern can be discovered in the output (e.g. a database lookup), or a validation that the transformed input exactly matches the output (e.g. an already transformed check). Both facilities are useful and so when a QVT transformation is invoked the invoker needs to specify what I call the 'rooting' condition in addition to the direction

Discussion

We need a much simpler principled exposition of the intent of QVTc/QVTr. Since Perdita has correctly come to such unwelcome conclusions, we clearly have a problem. The new statement should be uncontroversial and therefore provide a reference against which subsequent text can be assessed.

The following is perhaps what is needed, but needs more discussion with QVTr users and prototyping of its implications.

In Section 6.3 replace

The semantics of the Core language (and hence the Relations language) allow for the following execution scenarios:
• Check-only transformations to verify that models are related in a specified way.
• Single direction transformations.
• Bi-directional transformations. (In fact more than two directions are possible, but two is the most common case.)
• The ability to establish relationships between pre-existing models, whether developed manually, or through some other
tool or mechanism.
• Incremental updates (in any direction) when one related model is changed after an initial execution.
• The ability to create as well as delete objects and values, while also being able to specify which objects and values must
not be modified.

by

A Declarative QVT transformation capability establishes a correspondence between source objects in one or more source models and transformed objects in one or more transformed models. The correspondence may be used to check or enforce a view of, or a transformation to, target objects in one or more target models.

The correspondence takes the form of trace data comprising a set of trace records. Each trace record identifies a relationship to one or more transformed objects from, one or more, source or transformed, objects or values. The production of each transformed object is traced by exactly one trace record. The trace data includes relationships for all satisfied constraints imposed by the language constructs that express the transformation program. No relationships for satisfiable constraints are omitted from the trace data.

In practice a Core language (and hence the Relations language) transformation is executed after a form, mode and direction have been selected.

There are two main forms of QVT execution:

• For a Transformation execution, there is an exactly 1:1 mapping between the transformed objects and the target objects.

• For a View execution, there is an exactly 1:1 mapping between the transformed objects and the target objects in the view. There may be further target objects outside the view that are unaffected by the execution.

Additionally, for a Query execution, an OCL query is executed on the trace data. No actual target models are required.

There are three modes of execution:

• An enforced View or Transformation execution coerces the target objects to correspond to the transformed objects. (This is most simply achieved by model replacement, but may be realized by executing a series of changes so that desirable extra-model considerations such as xmi:id preservation or minimal database activity are accommodated.)

• A check execution is a degenerate Query execution that just returns true or false according to the existence of the trace data. (An optimized execution capability may bypass the creation of the transformed objects and look for correspondence between source and target objects directly.)

• An update Query, View or Transformation execution compares an old correspondence between source objects and transformed objects with a new correspondence between changed source objects and correspondingly changed transformed objects. The correspondence changes may be used to enforce or check for corresponding creations, deletions and updates in the target objects.

Additionally an in-place update View or Transformation execution shares source and target models and so each source object is coerced to correspond to its transformed equivalent. (This can be naively achieved by model copy. An optimized execution may bypass the creation of the transformed objects by performing each source read before any corrupting write.)

The declarative transformation languages do not have fixed sources and targets, rather they have multiple domains some of which may be designated sources and others as targets. The direction of the transformation is chosen by selecting the domain to be used as a target.

• A unidirectional transformation has just a single choice of direction since only one domain is able to be used as the target

• For a bi-directional, and more generally a multi-directional transformation, the domain used as the target may be selected

In practice it is often difficult to satisfy the bijective requirements of bidirectional execution and so many transformations are just unidirectional.

9.18 Undefined syntax
---------------------

The syntax for TransformationName, DirectionName, MappingName, PropertyName and VariableName
is undefined. Presumably each of these is an Identifier (below) when defining, but a PathNameCS
when referencing.

The syntax for PackageName is undefined. Presumably it is an OCL PathNameCS.

The syntax for ValueOCLExpr is undefined. This is presumably an OclExpressionCS.

The syntax for BooleanOCLExpr is undefined. This could be an OclExpressionCS of Boolean type but ...

The syntax for SlotOwnerOCLExpr is undefined. This could be an OclExpressionCS of Class type but ...

If BooleanOCLExpr and SlotOwnerOCLExpr are parsed as OclExpressionCS the 'default' prefix
causes a major ambiguity for 'default(...).xx' as a parenthesised slot owner or function call.
It is necessary to make 'default' a reserved word within OCL expressions.

Suggest: define BooleanOCLExpr and SlotOwnerOCLExpr very narrowly.

Predicate ::= SimpleNameCS ("." SimpleNameCS)* "=" OclExpressionCS
Assignment ::= ["default"] SimpleNameCS ("." SimpleNameCS)* ":=" OclExpressionCS

Undefined syntax

9.18 Undefined syntax
---------------------

The syntax for TransformationName, DirectionName, MappingName, PropertyName and VariableName
is undefined. Presumably each of these is an Identifier (below) when defining, but a PathNameCS
when referencing.

The syntax for PackageName is undefined. Presumably it is an OCL PathNameCS.

The syntax for ValueOCLExpr is undefined. This is presumably an OclExpressionCS.

The syntax for BooleanOCLExpr is undefined. This could be an OclExpressionCS of Boolean type but ...

The syntax for SlotOwnerOCLExpr is undefined. This could be an OclExpressionCS of Class type but ...

If BooleanOCLExpr and SlotOwnerOCLExpr are parsed as OclExpressionCS the 'default' prefix
causes a major ambiguity for 'default(...).xx' as a parenthesised slot owner or function call.
It is necessary to make 'default' a reserved word within OCL expressions.

Suggest: define BooleanOCLExpr and SlotOwnerOCLExpr very narrowly.

Predicate ::= SimpleNameCS ("." SimpleNameCS)* "=" OclExpressionCS
Assignment ::= ["default"] SimpleNameCS ("." SimpleNameCS)* ":=" OclExpressionCS

Discussion

Merged with the inadequate definition of identifiers.

The pragmatic decision to define Module/Transformation as inheriting both Package and Class violates UML inheritance.

For a Package: self.nestingPackage.nestedPackages->includes(self)

For a Class:
self.package.ownedTypes->includes(self)

But self cannot have two containers.

The problem is easily resolved by extending only Package and adding those Class features that are actually required.

Inconsistent multiple inheritance

The pragmatic decision to define Module/Transformation as inheriting both Package and Class violates UML inheritance.

For a Package: self.nestingPackage.nestedPackages->includes(self)

For a Class:
self.package.ownedTypes->includes(self)

But self cannot have two containers.

The problem is easily resolved by extending only Package and adding those Class features that are actually required.

Discussion

Transformation is mostly a Class that may have arbitrary Package scoping. This has been successfully prototyped by the Eclipse QVT projects, but not in sufficient detail to justify this slightly breaking change.

QVT Relations and working with stereotypes: Is there something like a QVT-R standard library with methods on stereotypes in it? There is none in the specification; compare QVT-R, there are some methods. Are there some else options for accessing stereotypes with QVT-R?

QVT Relations and working with stereotypes:

QVT Relations and working with stereotypes: Is there something like a QVT-R standard library with methods on stereotypes in it? There is none in the specification; compare QVT-R, there are some methods. Are there some else options for accessing stereotypes with QVT-R?

Discussion

Good question. Recent progress on ensuring that typesafe OCL navigation works for stereotypes may render the answer trivial, but until such time as a stereotyped QVTr transformation has been executed it is too soon to respond.

The author of Issue 9379 misunderstood the poor description of Pattern::bindsTo and consequently removed the

.

There was actually nothing wrong with bindsTo that could not have been fixed by clearer wording to the effect that bindsTo was a dustbin container for variables not contained elsewhere.

The above kind of emphasises the benefits of an ownedPlurals naming policy. Therefore suggest that:

a) the AST bloat from the redundant bindsTo be replaced by definition of a suitable derived property that satisfies the intent of all variables.

b) that variables be renamed ownedVariables and used throughout the QVTr to QVTc transformation

Rewind Issues 9379, 10938

The author of Issue 9379 misunderstood the poor description of Pattern::bindsTo and consequently removed the

.

There was actually nothing wrong with bindsTo that could not have been fixed by clearer wording to the effect that bindsTo was a dustbin container for variables not contained elsewhere.

The above kind of emphasises the benefits of an ownedPlurals naming policy. Therefore suggest that:

a) the AST bloat from the redundant bindsTo be replaced by definition of a suitable derived property that satisfies the intent of all variables.

b) that variables be renamed ownedVariables and used throughout the QVTr to QVTc transformation

Discussion

The problem is genuine, but further Eclipse QVTr prototyping reveals two distinct cases, one requiring containment, another not. Further work is needed to specify these accurately.