This package provides an EDSL for building data flow graphs based on higher-order attribute grammars. The nodes of the graph are organized in a tree, and the data flow defined via functions (rules) between attributes of the nodes.
The formalism and the implementation in this package features:
- Define data flow in a hierarchically structured fashion.
- Incremental definitions: additional attributes and rules can be given, and existing rules can be refined.
- Implicit topdown and bottomup data flows; the latter case for monoids.
- The data flow is defined locally as functions between attributes of a node and those of its parent and children. A global data flow is derived automatically.
The DSL compiles into a monadic program representing the construction of the graph via explicit fork/join-style combinators. We provide a trivial Identity and an IO instance (called "Fiber"). The former provides an implicit representation of the graph based on lazy evaluation, and the latter emulates lazy evaluation. Additionally, we provide a more explicit representation with explicit batchable effects (called "Resumption").
See Why Attribute Grammars Matter (The Monad.Reader, Issue 4) by Wouter Swierstra for a general introduction to attribute grammars in relation to functional programming.
Known restrictions:
- Attribute types are monomorphic. There is no generialization/instantiation of the attribute types of nonterminals.
- The structure of the grammar is collected in a single step, thus intermediate representations are not accessible for defining the structure of the grammar itself (i.e. for generic programming).