Documentation article
Traversal explained
Traversal centralizes tree-walking so generators and analyzers can stay focused on their own logic.
Traversal is the reusable mechanism for walking Shape IR consistently, regardless of whether you are generating output, collecting diagnostics, or adding analysis later.
Once Shape IR is the contract, traversal becomes the leverage point that prevents duplicated tree-walking logic across the codebase.
Without a shared traversal layer, every parser, generator, and analyzer ends up re-implementing slightly different recursion rules.
That duplication usually hides subtle inconsistencies. A common traversal model reduces those inconsistencies and makes extension work more predictable.
These rules make traversal useful across both current generators and future higher-level tooling.
The exact APIs can evolve, but the important part is that one traversal contract can serve many downstream behaviors.
A simplified traversal sketch
function visitShape(shape: Shape, visitor: Visitor) {
visitor.enter?.(shape);
if (shape.kind === "array") {
visitShape(shape.element, visitor);
}
if (shape.kind === "object") {
for (const field of Object.values(shape.fields)) {
visitShape(field, visitor);
}
}
visitor.exit?.(shape);
}