Why is a helper needed for replacement?

Why is a helper needed for replacement?

As I mentioned above cases, we need to replace a statement with multiple statements. If just one-to-one replacement, we can mutate the original statement directly

replace -> splice

Isn't "replace 1 statement with 3 statements" equivalent to "mutate existing statement + insert 2 new statements after it"? (like in your example)

So then do we need the "replace" action? Is there a reason why that's unergonomic?

2nd point: "Delete" action doesn't necessarily need an Address. It could be a single boolean flag per Vec<Statement> - probably stored as a NonEmptyStack<bool>.

• Helper contains a flag for "statement block needs some statements deleted".
• During transform of the statement, replace it with ctx.ast.move_statement() which creates a dummy EmptyStatement with Span { start: 0, end: 0 }.
• Call method on helper to indicate that some statements need to be deleted, but you don't need to tell it which.
• Helper checks the "does it need statements deleted" flag in it's exit_statements visitor, and if that flag says "yes", it deletes all EmptyStatements with 0 span.

I can't imagine a situation where a transform actually needs to insert a real EmptyStatement and that it must not be deleted.

Note: Ideal pattern if you're going to delete statements is to do it as early as possible, so then other transforms don't spend time traversing and transforming code which is going to be deleted anyway.

Note2: NonEmptyStack is really cheap. Because the stack is never allowed to be empty, stack.last() is infallible (returns T not Option<T>) and is just a single assembly instruction. This is much cheaper than Vec::last, which has to do bounds checks.