To ensure Review Bot works as expected, we need to write tests that
simulate a call to an external tool, and tests that actually execute an
external tool. A few of these have already been written, but maintenance
was already becoming problematic, as we want to ensure consistency
in both the data going into these tests and the data we assert.
This adds a new BaseToolTestCase class for tool tests, along with an
accompanying ToolTestCaseMetaclass and @integration_test and
@simulation_test decorators.
The decorators mark a test as being capable of integration testing and
simulation testing, respectively. They're intended for
BaseTool.execute() unit tests. The metaclass will scan for these and
split any test function into separate simulation/integration tests (if
the appropriate decorators are applied).
This keeps things very easy to maintain, but with the expense of a quirk
with test names. A test_foo() defined on the class will no longer be
available to the test runner. Instead, separate
test_simulation_foo() and test_integration_foo() test functions will
be available. This isn't immediately obvious, but is a quirk we can
learn to live with.
