We don't currently have tests that check for bugs in our file ID allocation code.

The particular steps we are interested in recreating are:

• Create a table (t1).
• Do a clean restart.
• Create another table (t2).
• Insert data into t2.
• Make recovery run.

If we aren't tracking file IDs properly, it's possible that we'd end up apply the log records for t2 to table t1.

This sequence should be possible using our Python test suite, but would be different to existing test cases.