Comment 6 for bug 860159

I'm down to three test failures in the above branch. There seems to be two actual bugs going on here:

1) Even though I fixed the SQL type for digest_size_threshold to REAL for Postgres (and FLOAT for SQLite, but it doesn't care), it still seems that when we store 10.5 into it in configuration.rst, 11.0 comes out.

2) It looks to me like Postgres doesn't reset the primary key id to 1 for deleted tables. Two of the failing tests print the Pended dictionary, which contains its id, i.e. its primary key. For SQLite, this always gets reset to 1 for subsequent tests (because we delete all tables between every test). For Postgres though, deleting the table does not seem to reset the primary key back to 1.

This might be a case where the tests need to be fixed to not assume the id is already reset at the beginning of the test, unless there's a good way to guarantee this for Postgres.