In the new test I wrote for
WT-3257, I discovered a bug that if I run with log=(recover=error) on an unclean directory (properly getting the error that recovery must be run) and then, on the same unclean database, I run log=(recover=on) it does not properly recover all the data. The error run the first time moves a checkpoint LSN forward and we skip the earlier data.
- is related to
WT-3257 Need unit test for log recover setting