TSLspImportAll: implement priority

[jose-elias-alvarez]

The idea came up in #39 to prioritize importing from open buffers. Potential implementations, in order of increasing complexity:

• Import only from open buffers (meaning implement this as a separate function)
• Prioritize importing from open buffers, fall back to any other source (as a config option, which should probably be enabled by default)
• Separately configure priorities for open buffers, project files, and modules, using simple (but imperfect) patterns to discriminate modules from project files

The last option is the best but might not be worth the effort, since I don't know how many users would be interested in that level of control.

[non25]

The only thing I'd like to add is that it should look for existing imports too in the open buffers.

I think prioritizing project files is really useful. If you don't have buffers opened, e.g. when you just opened one file, but you have third party abstractions wrapped in yours, or you are using exports with the same name across the project, then I imagine you would want to autoimport yours 100% of time.

Don't know if configurability is needed though.

Also would be cool to do this interactively. Like if you have >2 options to import it would prompt you in that case.

[jose-elias-alvarez]

Could you clarify what you mean by "existing imports" from open buffers?

Interactive choice would be great, and it shouldn't be too hard to implement, either. For priority, I'm leaning towards open buffers > project files > external modules, with a simple config option to disable priority for maximum speed.

[non25]

Sure,

// other opened buffer
// react material has this too, prefer auto-importing this EditIcon because it is imported in the other buffer
import { EditIcon } from 'common/icons';

function someFunction () {
  // some code
}

export { someFunction };

[jose-elias-alvarez]

Ah, okay, I see. That's a good idea – it's a simple heuristic but I imagine it would be very accurate in predicting the import the user wants. As far as implementation, it's definitely possible but I'd want to test performance, especially with a large number of buffers open (though we could put in a limit).