Too many ast visitor in ecma minifier

[bvanjoi]

We can merge the unrelated visitors to improve performance.

[bvanjoi]

I've observed that the current ecma_minifier implementation contains an excessive number of visitors, which introduces significant unnecessary overhead. My goal is to reduce this inefficiency.

There are two potential approaches:

• ​Fundamental Solution: Create an IR that's both visitor-friendly and easily parallelizable. However, this would require substantial architectural changes that may not be immediately practical.
• ​Incremental Improvement: Implement a more efficient visitor pattern similar to Webpack's hook system. For example, we currently have two rename operations:

struct LabelRenamer {}
impl VisitorMut for LabelRenamer {
    fn visit_label() {
        rename_label()
    }
}

struct PrivateRenamer {}
impl VisitorMut for PrivateRenamer {
   fn visit_private_ident() {
        rename_private_ident()
   }
}

fn rename() {
   node.visit_mut_with(LabelRenamer);
   node.visit_mut_with(PrivateRenamer);
//~^ It means we should traverses twice.
}

If we implement a hook-based approach, the minifier would only need to traverse the AST once. This single traversal could then trigger all necessary transformations through registered hooks:

struct WebpackHooksLikeVisitor {
    plugins: Vec<Box<dyn NodeMutablePlugin>>
}

impl WebpackHooksLikeVisitor {
   fn visit_private_ident() {
        self.plugins.run(private_ident_node)
   }
    fn visit_label() {
        self.plugins.run(label_node)
    }
}

struct LabelRenamer;
impl NodeMutablePlugin for LabelRenamer {
   fn visit_label() {
      rename_label();
      true // means continue visit
   }
}

struct PrivateRenamer;
impl NodeMutablePlugin for PrivateRename {
   fn visit_private_ident() {
        rename_private_ident()
        true // means continue visit
   }
}

fn rename() {
  let plugins = vec![Box(LabelRenamer), Box(PrivateRenamer)]
  let visitor = WebpackHooksLikeVisitor::new(root_node);
  visitor.visit_root_node();
}