On Mon, Oct 25, 2021 at 02:09:36PM -0500, Robert Pearson wrote: > Here (not related to the mcast issue) I am guilty of hiding my plans. > A little later I was going to move creating the metadata (RB tree and > index table) into the lock and then later after converting to xarrays > replace the lock with the xa_lock which you have to have anyway > because it's part of xa_alloc(). > As you point out it is the index/key tree change that is visible to > everyone. The races here are in trying to get the index or change the > RB tree. Perhaps I should merge this change with the later one where I > get rid of add/drop_index(). Or, rearrange things and convert to > xarrays and then worry about races. Right, it should be later Jason
