#17317 reclaimed much of the performance regression from #17286, but it appears as though we have still lost a fair bit of performance over the last two days, much of it due to work on * imports. Here's an annotated graph of how the Codspeed red_knot_check_file[cold] benchmark has changed over the last two days:

We may still be doing more work than necessary when it comes to applying and resolving visibility constraints for * imports. There are some suggestions in this thread for further optimisations that we could make: #17317 (comment)