ast-codegen: support `#[visit(enter_before)]` attribute

[Dunqing]

We have discussed this at #4147 (comment). But we do need this at the moment.

In semantic builder, We need enter_scope before enter_node in visit_function and visit_class.

oxc/crates/oxc_semantic/src/builder.rs

Lines 1423 to 1450 in 2eb37d5

	fn visit_function(&mut self, func: &Function<'a>, flags: Option<ScopeFlags>) {
	let kind = AstKind::Function(self.alloc(func));
	self.enter_scope(
	{
	let mut flags = flags.unwrap_or(ScopeFlags::empty()) | ScopeFlags::Function;
	if func.is_strict() {
	flags |= ScopeFlags::StrictMode;
	}
	flags
	},
	&func.scope_id,
	);
	/* cfg */
	let (before_function_graph_ix, error_harness, function_graph_ix) =
	control_flow!(|self, cfg| {
	let before_function_graph_ix = cfg.current_node_ix;
	cfg.push_finalization_stack();
	let error_harness = cfg.attach_error_harness(ErrorEdgeKind::Implicit);
	let function_graph_ix = cfg.new_basic_block_function();
	cfg.ctx(None).new_function();
	(before_function_graph_ix, error_harness, function_graph_ix)
	});
	/* cfg */
	// We add a new basic block to the cfg before entering the node
	// so that the correct cfg_ix is associated with the ast node.
	self.enter_node(kind);

We want to treat FunctionExpression and FunctionDeclaration differently and the class is the same too.

FunctionExpression: create a binding for function id to the current scope
FunctionDeclaration: create a binding for function id to the parent scope

To add the binding to the parent scope, we must need enter_scope before enter_node.

I tried another way to get around it (Same as TypeScript). But I find it a bit confusing. See: #4195 (review)

[Dunqing]

Ah, Haha. This is also a bit confusing for visit orders. I thought about it again. We can defer binding FunctionExpression and ClassExpression and bind them when we visit BindingIdentifier.

[overlookmotel]

@Dunqing Just to throw in my 2 cents, I wonder if FunctionExpression and FunctionDeclaration should be separate types (and each type would have Function as a field within it). The 2 have various significant differences:

1. One is a Statement, the other an Expression.
2. FunctionExpression's name field should be an Option, whereas it's compulsory in FunctionDeclaration.
3. Scope should be entered before name for FunctionExpression and after name for FunctionDeclaration.

We could also have a separate type for ObjectMethod (whose name isn't a binding at all).

At present, it's possible to create non-sensical AST nodes (e.g. a FunctionDeclaration with no name) - it'd be good to make this impossible through the type system.

[Dunqing]

@Dunqing Just to throw in my 2 cents, I wonder if FunctionExpression and FunctionDeclaration should be separate types (and each type would have Function as a field within it). The 2 have various significant differences:

1. One is a Statement, the other an Expression.
2. FunctionExpression's name field should be an Option, whereas it's compulsory in FunctionDeclaration.
3. Scope should be entered before name for FunctionExpression and after name for FunctionDeclaration.

We could also have a separate type for ObjectMethod (whose name isn't a binding at all).

At present, it's possible to create non-sensical AST nodes (e.g. a FunctionDeclaration with no name) - it'd be good to make this impossible through the type system.

Yes, I talked with @Boshen about the above. Since they are not very different, they have been combined into one type. I agree we should have different types for FunctionExpression and FunctionDeclaration, and also class should be ClassExpression and ClassDeclaration

3. Scope should be entered before name for FunctionExpression and after name for FunctionDeclaration.

We bind in enter_node. So we have to switch the order of enter_node and enter_scope to bind it to the corresponding scope. Or do something like #4195 (comment).