Where the discussion in#28115 got stalled is over using difflib or vendoring a copy of Levenstien distance with a question of how different the results are (and then I did not have time to answer that question). Do we want to resume that discussion or just usedifflib and move on?

I think either way getting close matches is better than the current state of things and contra my previous comments I'm in favor of using difflib and changing the algorithm later if it is a problem (I'm not going to consider the guessed "did you mean..." list to be something we need to maintain API stability on!).