Discoved while debugging in 5.1 that there was a bug where a certain crash could lead to two problems. 1) An additional share in memory that was allocated but did not have the correct use_count (so it would never be fulled deleted). Also discovered that a thread that called repair would write new rows, but would not see them. All other threads were ok, and the data was fine, but the thread doing the repair was unable to see the new rows.