Implement a "loose mode" to parse an application

[marcoroth]

Right now when you parse an application with the Stimulus parser it wants to fully figure out and trace all registered controllers.

Meaning that if can't figure out or can't parse your setup it will not have good support from the parser/LSP - so it's not really going to be super helpful in those cases, it's even more annoying than helping.

In that case, parsing the controllers is and works fine, meaning we still have the parsed results from the controllers - it's just not able to figure out the if a certain controllers was registered.

Which is why it might make sense to implement an option so a user can opt into a "loose parsing mode" that doesn't try to trace the registering of the controllers.

This would mean that you don't get as accurate results, but would mean that you can still use the LSP with somewhat useful tooling instead of it just an LSP that's complaining about all the unregistered controllers.

[tvongaza]

This would be great to see! Our setup looks something like follows where we are using vite meta glob imports to include all stimulus controllers in a given directory. Stimulus parser/lsp sees all our controllers as unregistered, despite them being registered this way. Any ideas of temporary work arounds?

  const controllers: Record<string, any> = import.meta.glob('@/controllers/**/*_controller.ts')
  // For each of our controllers found in the above glob
  // ensure the controller import is loaded and then register it
  for (const [name, controllerModule] of Object.entries(controllers)) {
    controllers[name] = await controllerModule();
  }
  registerControllers(application, controllers);