UML StringExpressions, Operands and SubExpressions
In the latest UML specification for 2.5.1, the implementation in the OCL language of the overridden StringExpression::stringValue() is given at section 8.6.17.5 "Operations" on page 130:
body: if subExpression->notEmpty() then subExpression->iterate(se; stringValue: String = '' | stringValue.concat(se.stringValue())) else operand->iterate(op; stringValue: String = '' | stringValue.concat(op.stringValue())) endif
The intent of this is perfectly clear; however, for example in specifying an ordered list of function parameters, there is always some kind of separator token (usually the comma) to delimit the individual arguments passed to a function. AFAICT the OCL 'concat' operation does not supply any separator token.
Since theoretically a StringExpression's subexpressions might return their own concatenated subexpressions, ending up with some kind of nesting of operands, it seems that no separator at all would quickly lead to confusion and errors.
It doesn't seem to be specified anywhere as to whether the concatenated operands or subexpressions need to be separated by commas or perhaps some other token? There might be languages in use where something other than the comma is used (perhaps a space?), so it is conceivable that the behavior could be overridden by a Profile in some form or fashion.
Is this a defect in the specification?
OCL concat doesn't add any separators, but this no problem for stringValue since it doesn't need them. As you can see in figure 7.6 StringExpressions are mainly used to name elements that are created by binding a TemplateParameter to a LiteralString. If you want to include a separator, you can do it in the specification of the nameExpression (e.g. $foo, <Resource>, <ResourceKind>, ba$).
By the way, I tried to replicate the diagram in figure 7.6, but it seems that my tool doesn't completely support templates and name expressions. It's very likely that other tools will also fail. Seems this part of the specification has been neglected by the tool vendors.
- Does C++11 allow dollar signs in identifiers?
- What are the prohibitions against certain kinds of aliasing through unions?
- Does calling non-const mem. fun. on const object via non-const ref obtained via const_cast invoke UB only if the fun actually modifies the object?
- Common initial sequence with nested struct
- Is there a way to assert something about an argument to a consteval function, with a meaningful error message on failure
- C++14 constexpr function requirements on cppreference
- Error when binding a constexpr reference to variable of different type
- Is there inlinable function for pow2?
- Can & (address-of-operator) ever return nullptr?
- Why can’t I use a unary operator in a chained assignment, but only in separate assignment statements?
- Changing active member in a union with copying using std::construct_at
- Can std::ranges::enumerate enumerate any possible valid C++ array?
- Is definition of variable length array (VLA) / known constant size recursive?
- Why global-scope function declaration with built-in type arg must be visible before unqualified call to that name with argument of template type?
- UML CollaborationUse ownership
- Is using if (0) to skip a case in a switch supposed to work?
- std::ranges::sort not working with non-default operator<=>?
- How to initialize a union containing const data members in C++11
- is optional<void> type name valid?
- Does direct initialization of class type consider user defined conversion function?
- Expected behaviour when expanding a template parameter pack into an inner template declaration
- Do I need `std::launder` when working with unions?
- No concept subsumption with template parameter pack?
- UML Collaboration and Structured Classifier
- Scope of struct definition defined in a closure
- Why can't I static_cast a void* to a pointer-to-function?
- Is shifting 0 to the left considered undefined behavior in C?
- Treatment of backslash character in the bracket expression
- Structured binding declaration cannot be `constinit`
- Do C and C++ differ on the legality of aligning pointers at runtime?