Rick Hillegas <[EMAIL PROTECTED]> writes: > This ought to work even if you decide to sand down the catalog support > for 10.5. You'll just have to supply 10.4->10.5 upgrade logic to morph > the catalogs into their final form.
I agree. > > Then there's the issue of the 10.4 upgrade tests for the roles > catalogs. I think it's good to have some logic in there which verifies > that the 10.4 catalogs look right. You can verify their shape I > suppose. But, having short-circuited the roles syntax, you won't be > able to test-populate the catalogs. Correct, I have disabled the roles fixtures in Changes10_4 for roles in the "mask roles feature" patch I am readying. The shape is also being tested for new (10.4) databases in SystemCatalogTest, no need to roll that back I think. But it may be good to add such tests to Changes10_4, as well, although I am not sure it yields more than verifying that hard upgrade actually adds SYSROLES (shape is tested by SystemCatalogTest and generic code (upgradeMakeCatalog) is used to create SYSROLES). Dag
