Guidance for bugfix in swc's chain!() of ImportAnalyzer and CommonJs transformers

[cspotcode]

I'm attempting a bugfix for #1786. The README suggests that I should ask here for guidance understanding swc's codebase.

I see that ImportAnalyzer and CommonJs transformers both use scope.import_types to understand if a require() needs to be wrapped in interopRequireDefault or interopRequireWildcard. However, the transformers do not seem to share a Scope instance.

This causes problems when ImportAnalyzer marks an import as wildcard, but CommonJs does not know this and treats it as a default import.

Should transformers share a single scope instance, or should import_types be copied from one transformer to the next?

[kdy1]

Sharing scope is better, but it may require some strange tricks or Mutex (Possibly RwLock).

[cspotcode]

I was trying to understand that. If the transformers run sequentially, then two are never running at the same time, right?

If so, can we use RefCell to give a mutable borrow of the Scope to each transformer while it is running?

I think this will require augmenting the chain!() so that it owns the Scope. The transformers can be wrapped in some sort of factory which creates them when they execute and drops immediately after, instead of creating them all ahead-of-time.

Is this a smart idea?

[kdy1]

You can do it with Rc<RefCell>> and pass down the rc to both pass.

I think this will require augmenting the chain!() so that it owns the Scope. The transformers can be wrapped in some sort of factory which creates them when they execute and drops immediately after, instead of creating them all ahead-of-time.

It's possible, but it should allow injecting passes between analysis and import replacer.