For a strict DBMS, the better approach is typically to attempt to generate expressions of the expected type. For PostgreSQL, the expression generator thus expects an additional type argument (see [PostgreSQLExpressionGenerator](https://github.com/sqlancer/sqlancer/blob/86647df8aa2dd8d167b5c3ce3297290f5b0b2bcd/src/sqlancer/postgres/gen/PostgresExpressionGenerator.java#L251)). This type is propagated recursively. For example, if we require a predicate for the `WHERE` clause, we pass boolean as a type. The expression generator then calls a method `generateBooleanExpression` that attempts to produce a boolean expression, by, for example, generating a comparison (e.g., `<=`). For the comparison's operands, a random type is then selected and propagated. For example, if an integer type is selected, then `generateExpression` is called with this type once for the left operand, and once for the right operand. Note that this process does not guarantee that the expression will indeed have the expected type. It might happen, for example, that the expression generator attempts to produce an integer value, but that it produces a double value instead, namely when an integer overflow occurs, which, depending on the DBMS, implicitly converts the result to a floating-point value.

SQLancer uses [JCommander](https://jcommander.org/) for handling options. The `MainOptions` class contains options that are expected to be supported by all DBMS-testing implementations. Furthermore, each `*Provider` class provides a method to return an additional set of supported options.

For up-to-date testing commands, check out the `.travis.yml` file.

Each class specific to a DBMS is prefixed by the DBMS name. For example, each class specific to SQLite is prefixed by `SQLite3`. The naming convention is [automatically checked](src/check_names.py).