-
Notifications
You must be signed in to change notification settings - Fork 7
Home
PIG is a compiler framework, domain modeling tool and code generator for tree data structures such as ASTs (Abstract Syntax Tree), database logical plans, database physical plans, and other intermediate representations. Using PIG, the developer concisely defines the structure of a tree by specifying named constraints for every node and its attributes. Every constraint is known as a "data type", a collection of data types is known as a "type domain" and a collection of type domains is known as a "type universe".
Every type domain has two representations:
- An Ion s-expression representation, allowing type domains to serve as a language and platform neutral wire protocol and compact serialization format.
- A strongly typed set of data types specific to a target language such as Kotlin (the only currently supported language target).
PIG also provides facilities that allow for manipulation and rewriting of trees for the purposes such as program optimization, query planning and code generation.
Query engines and other kinds of compilers require numerous tree-like representations of a program, starting with an AST. Query engines typically parse a query from text to an AST, and then apply multiple passes over the AST to incrementally transform it to a logical plan, then a physical plan, and possibly other intermediate representations. Compiler passes and related data structures of this sort are large, complex, and difficult to maintain. PIG's "permuted domains" feature increases the maintainability of such compiler passes and related data structures by allowing new type domains to be created by specifying only the differences to another type domain. This avoids having to duplicate the data type definitions that are common to both type domains, allowing more numerous, smaller, less complex and more maintainable compiler passes.
PIG's permuted domain feature has been heavily inspired by the Nanopass Framework.
Metas are arbitrary key/value pairs that can be associated with any node. Metas can be used to store metadata about a node, such as the location in the source text of the grammar element it represents, or the data type of the value returned by an expression.
PIG generates the following components in Kotlin (and may generate similar components in languages such as Rust in the future):
- Immutable, strongly typed classes representing each data type within each type domain.
- Abstract base classes for implementing compiler passes that:
- Transform from one type domain to another (the developer must only account for the differences between the domains!).
- Transform to a modified tree of the same domain.
- Functions to convert each tree between instances of the generated classes and its s-expression representation.
- Builder functions, for easily composing deeply nested instances of the strongly typed classes.
Details can be found in Code Generation.
This project is licensed under the Apache-2.0 License.