>And you don't know how it got into the state where this optimization >is applicable?
I believe that Arne already has answered this question; the patch caused the minimum hashtable size to shrink to 4, which means that the realloc size (old_sz * 2 + 2) matched the hashtable size for the case where the old size was 1 and the hashtable size was 4 and the number of mapping_data references were > 1.
