These pages written, maintained, and © by Atze Dijkstra.

Small(er) projects/problems

• The default chaining policy is left-to-right. Allow different chaining strategies, e.g. right-to-left. Allow explicit turning on/off the copy rules.
• Naming a leftmost/rightmost occurence of a child attribute anonymously, that is, whithout knowing the specific leftmost/rightmost child.
• Pretty printing of AG code.
• Overriding an existing attribute definition and/or allowing reuse of an existing definition inside a new definition.
• An attribute may have a default value for a certain scope (everything, view, node, ...)
• Usage of 'self' as mechanism to refer (from an attribute definition) to a node (where the definition takes place) itself.
• Wrappers around a AG tree allowing the inherited attributes to be passed as a (extendable) record, and, vice-versa, the synthesized attributes to be returned.
• FFI (Foreign function interface), allowing the semantic function be bound to a Haskell function.

Larger projects/problems

• Plugin architecture for extending AG functionality. An example of this is the specific usage of a chained unique counter (for creating unique identifiers). The plugin should cater for inclusion of special generated (Haskell) code taking care of this counter.
• Grouping attribute definitions, called a view. Actually this covers more/less tightly coupled issues.
  • Explicit hiding/exporting/importing attributes
  • Group has a name used for (e.g.) importing, redefinition
  • Parameterizing the name of generated code (Haskell semantic) functions in order to avoid name clashes
  • Spreading AG code over different files, combining by importing (logically as a view) and/or including (textually) them.
• A mechanism for choosing between different sets of attribute definitions. For example, pretty printing may have different attribute definitions yes/no including error messages. The attributes have the same name. In other words, given an attribute interface allow different realisations/implementations. This may or may not also be done runtime.
• Usage of AG combinators (Oege de Moor), generic policies, rule based.
• Incorporation of a typechecked language for attribute definitions (SL, ...)