Separate test loading and introduce test state
This commit implements two very different things, but still very
much connected:
1) Separate test loading to the process that will be actually
running the test. This leads to a better separation of duties of
the process interacting with the user and the process that is
actually doing the (usually) heavy work.
2) Remove the pickling of the test instance and sending it back
and forth.
As an added bonus, it was noted during development, that both
processes load all necessary Python modules (including the test
module) when unmarshalling the pickled instance, resulting in
unecessary work. Actual data on performance improvements were not
collected though.
It has one change in behaviour (regression) though. Test timeout
now comes only from the test parameters. The reasoning is that
since there's now a better separation of roles and the fact that
the first process is not loading the test instance, it can not
look at the timeout defined there.
IMHO this is a sign of good design, since other types of test
(think of shell scripts) can not communicate their intended timeout.
Maybe a 'test.cfg' or something like that should be the solution to
that.
Signed-off-by: NCleber Rosa <crosa@redhat.com>
Showing
想要评论请 注册 或 登录