The regression was probably due to a bad interaction between Implicitly Opened Existentials (SE-0352) and Opaque Parameter Declarations (SE-0341) which GRDB 6 now uses in a lot of places.
Considering the sample code below:
try db.create(table: ...) { t in
t.column(...).check { $0 != Column(...) }
}
- The `check` method accepts a closure that returns `any SQLExpressible` (the same as GRDB 5)
- The `!=` SQL comparison operator is `(some SQLSpecificExpressible, some SQLSpecificExpressible) -> SQLExpression`, when it was `(SQLSpecificExpressible, SQLSpecificExpressible) -> SQLExpression` in GRDB 5
The compiler has started preferring interpreting the != operator as Column's Equatable implementation, instead of the `!=` SQL comparison operator
We force the compiler to use the SQL comparison operator by removing the Equatable conformance from Column. It was introduced in a70b42c, as a support for a feature that has been removed since then.
This is a breaking change, but this also is an improvement. Now `column1 != column2` has a single meaning, and the library is less prone to similar regressions in the future.