Currently we give the user the option of exporting visualizations and then, on import, the user can create a new index pattern or choose an existing pattern with which to associate those visualizations. If we assume association based on the index pattern rather than the id, we run the risk of attempting to associate a visualization with an entirely different index that just happens to be found/identified with the same pattern. The user is of welcome to make that mistake, we just don't want to automate it!
That's fair. The problem comes up when we have a *lot* of visualizations and they're all broken when an index is re-created. I don't see an option to export a broken visualization, and when bringing it in manually with json it keeps the invalid UID. It doesn't help that there's no correlation to what the old UID belonged to index-wise, and no drop down to choose the new index to represent, or if the UID is listed several times in the json it needs to be fixed each time..
Basically we're stuck digging around and manually re-building visualizations.. hundreds of times.