Question: does a custom linter's `visitFooNode` method need to call `super.visitFooNode`?

[asilano]

We have developed a number of customer linters, all of which follow the pattern of having a visitERBContentNode (or similar) override, which performs the required check and the calls super.visitERBContentNode

e.g.

class ERBNoRemoteForLinkToVisitor extends BaseRuleVisitor {
  visitERBContentNode(node) {
    this.checkLinkToTypeConstruct(node);

    super.visitERBContentNode(node);
  }

We do this because it's the pattern established in the sample custom linter that existed in the Herb repo when we first wrote a custom linter.

However, I notice that the linter docs now have two custom linters; .herb/rules/no-inline-styles.mjs and .herb/rules/no-div-tags.mjs. no-inline-styles ends its visitor with the expected call to super.visitHTMLOpenTagNode; but no-div-tags does not call the superclass's method at any point.

Does a custom linter, inheriting from BaseRuleVisitor, need to call super.visitFooNode as part of its visitFooNode overrides?

[marcoroth]

Hey @asilano, great question!

I guess we should update the no-div-tags sample linter rule to call super so we are consistent. Whether you have to call super or not depends on the logic of the linter rule itself. But I'd argue that calling super is what you want, most of the time.

Each visitFooNode method in the base Visitor class does up to three things:

1. Calls this.visitNode(node), a generic hook for all node types
2. Calls this.visitERBNode(node), only for ERB node types (a shared hook across all ERB nodes)
3. Calls this.visitChildNodes(node), traverses the node's children

For example, visitHTMLOpenTagNode calls visitNode + visitChildNodes, while visitERBContentNode calls visitNode + visitERBNode + visitChildNodes.

When you override a visitFooNode method and don't call super, you skip all three of these steps. Most importantly, visitChildNodes won't be called, which means the visitor stops traversing deeper into the tree from that point.

For leaf nodes like ERBContentNode that don't have children, skipping super doesn't affect traversal. But for nodes with children, like HTMLElementNode, HTMLOpenTagNode, or ERBBlockNode, not calling super means their child nodes will never be visited, and your rule may miss offenses deeper in the tree.

As a general rule: always call super.visitFooNode(node) unless you intentionally want to control traversal yourself. If you do want more control, you have a couple of alternatives:

• Call this.visitChildNodes(node) directly instead of super, this traverses all children but skips the visitNode/visitERBNode hooks
• Call this.visit(node.someChild) to only visit specific children rather than all of them

[asilano]

Thanks, that's a great comprehensive answer.

I'm happy for this issue to be closed, unless you want it to track updating the no-div-tags linter example.