Migrate assists to SyntaxEditor

[darichey]

We want to migrate all assists to use the new SyntaxEditor/SyntaxFactory abstraction. This will unblock moving to immutable syntax trees (#15710). In doing so, we can choose a flat, contiguous representation that should yield significant performance wins (#17491).

@DropDemBits has outlined the process for migrating most assists:

• Create a SyntaxFactory, optionally named make to ease migration
• Create a SyntaxEditor using builder.make_editor with a node from the file
• Replace all make:: with make
  • Add any missing methods to syntax::ast::syntax_factory::constructors as needed
• Replace all ted:: with editor.
• After all changes are complete:
  • editor.add_mappings(make.finish_with_mappings()) to add the generated mappings
  • builder.add_file_edits(ctx.file_id(), editor) to add the edits
    For assists that use snippets, check the snippet capability, and then use editor.add_annotation to associate the snippets with the elements.

They have also already migrated the extract_variable assist in #18196 which should be a useful reference.

The assists are located in crates/ide-assists/src/handlers. They should all have tests which you can use to verify that your change works properly by running:

cargo test --package ide-assists -- handlers::<ASSIST>::tests  

replacing <ASSIST> with the assist name, e.g., add_braces.

[darichey]

We have two MLH fellows, @tareknaser, @davidkurilla, who are interested in working on this.

[Giga-Bowser]

I'm not sure of the precedent on this sort of thing, but I've got two open PRs, #18458 and #18459, that both have to do with assists, and if it would help I'd gladly migrate those assists as part of those PRs.

Specifically I would be migrating:

• remove_unnecessary_wrapper which is added in feat: Add diagnostic fix to remove unnecessary wrapper in type mismatch #18458
• add_missing_match_arms which is modified in feat: Use snippet placeholders for generated match arms #18459

[Giga-Bowser]

While migrating those PRs I ran into make::ext::expr_todo(), is there any preference as far as how to migrate this? I could define an expr_todo() method on SyntaxFactory, but if there's a desire to preserve this split between make and make::ext then I could also define it as ext_expr_todo() in a separate impl block, or even do some sort of type shenanigans with an ext() method on SyntaxFactory that returns a SyntaxFactoryExt with its own methods. Let me know what is preferred.

[darichey]

Not sure if anyone has a strong opinion, but just putting it directly in SyntaxFactory for now seems fine to me.

[Giga-Bowser]

One observation while adding methods to syntax::ast::syntax_factory::constructors: I often found that the equivalent make:: functions would return more general AST types than necessary, which I think is a mistake we shouldn't repeat. expr_return should return an ast::ReturnExpr, not an ast::Expr, especially since casting an ast::ReturnExpr into an ast::Expr is infallible and the other way around is not.