wollok-doc

GitHub Discord Email

Interpreter

The Interpreter provides a set of tools to execute and debug Wollok code. It can be used to execute a Program, run a Test or evaluate any Sentence, either atomically or step-by-step.

The current implementation of the Interpreter is a Direct AST Interpreter, which means that the Nodes of the AST are interpreted directly, without first compiling them to any other intermediate representation (such as bytecode).

The current Interpreter is also effect-based, so every computation updates the current evaluation state in a destructive way. This might not be as fancy as a pure stateless implementation but it’s proven to be much faster.

Following is a summary of the main abstractions and concepts involved in the interpretation process.

Main Abstractions

Evaluation

An Evaluation State (or just Evaluation) for short)represents an isolated runtime environment. It is the main data structure used by the Interpreter to represent the complete state of an execution at any given moment. It contains the Frame Stack and references to all the Runtime Objects along with an interface that allows it to instantiate new objects and trigger the execution of AST Nodes.

The Evaluation is built from an Environment that provides it with all the static definitions and a set of natives that implement any native methods used.

Due to performance reasons, the evaluation and many of its sub-structures, are stateful mutable objects and most operations on it are destructive, so special care should be taken of making a copy of any instance which state is meant to be preserved.

Frame Stack

The Frame Stack consist of a stack of Contexts. Its main purpose is to keep track of the diferent lexical scopes during nested method calls.

Evaluations initialize their Frame Stack with a Root Context cointaining all global definitions.

Contexts

Contexts are hierarchical structures that represent the current stored named references in an evaluation scope. You can think of Contexts as Maps that relate each locally accessible reference name to the Runtime Object that is its current value.

Every Node is executed within a Context that provides values for References. When a Node needs to be executed in a separate lexical scope a new Context is created to allow it to maintain it’s own namespace.

Every Context except the Root Context has a parent Context. When a Context can’t find a reference it delegates the search in its parent.

Runtime Objects

Runtime Objects are special Contexts that, as their name implies, act as the runtime representation of Wollok objects. An Evaluation contains one RuntimeObject for every Singleton and Class instance used during an execution. These have an unique Id, its associated Module and, sometimes, an Inner Value used to store primitive data (such as TypeScript Numbers for instances of wollok.lang.Number).

The instance’s fields and self reference are stored as part of its Context mapping allowing Runtime Objects to be lexical scopes themselves.

Although most RuntimeObjects are generated as result of the explicit instantiation of a Class, some others need to be created as part of the Evaluation initialization (such as the named Singleton instances and special objects like null) and many are built on demand (like numbers, strings and other literals).

Executions as Typescript Generators

One of the Wollok Interpreter’s main goals is to provide an adequate support for Debugging and Inspection Tools. This implies that it should be possible to interrupt an execution at any point (for example, by placing a breakpoint) and resume it at will. Luckily for us this can be easily done using TypeScript’s Generator Functions.

For the purpose of our implementation, you can think of a Generator Function like a regular function that can yield other values before returning a result. Each time the function yields the control is returned to the caller, who can choose to resume the function if he chooses to.

We define the execution of a Node as a Generator Function that yields each Node it visits and ends up returning the result of the Node’s execution (usually a Runtime Object). So, every time we want to execute something we can just ask the execution function to keep yielding nodes until either it yields a Node at which we should stop or it returns the final result.

This implementation has some drawbacks, but it allow us to define the execution of each type of Node in a very clean way, with little boilerplate and nearly no execution overhead. The main difficulty that arises from this approach is the need to keep asking the generator for results and detecting when a final state has been reached. To do this we modeled one more abstraction: The Execution Director

Execution Directors

If an Execution is the iterable result of a Generator Function, an ExecutionDirector is an utility object that allow us to easily iterate the Execution.

It’s interface contains methods to set breakpoints, take different kinds of steps or keep executing until the Execution is over.

In most cases, a tool that wishes to run a Wollok expression should:

  1. Compile the code to obtain the static Environment.
  2. Create an Evaluation to contain the runtime entities.
  3. Find the desired Node in the environment and use one of the Evaluation methods such as exec to create an Execution.
  4. Provide the Execution to an Execution Director and either set a breakpoint and continue until it’s reached or finish the execution to obtain the final result.

Natives

Wollok methods defined as native require primitive implementations written in the host language. These implementations need to be provided to the interpreter in order to successfully evaluate most code.

In Wollok-TS, native implementations are modeled as Generator Functions that execute in the context of an Evaluation and have the following type:

export type NativeFunction = (this: Evaluation, self: RuntimeObject, ...args: RuntimeObject[]) => Execution<RuntimeValue>

These functions are expected to be provided within a TS object replicating the package structure of the owner. So, for example, a native method m for the class package.subpackage.C should be represented as the following structure:

{
  package: {
    subpackage: {
      C: {
        m: <place m primitive here>
      }
    }
  }
}

When a native method is called, the corresponding Native Function is evaluated with the receiver and argument objects. The Native Function is expected to return a generator that performs any transformation required satisfy the native behavior.

Native implementations for the Wollok Runtime Environment’s can be found in the /wre folder of the project sources.