Mikhail Loenko wrote:
Well, I like things clean and simple too :)
We could develop our own implementations for unit tests, it would take
the same efforts as develop our own provider. It's a huge amount of work.
Now we use BouncyCastle provider for the purpose of unit testing:
it's open source, everyone can download it etc.
But our unit tests without any modification could serve as 'system' tests
as Anton called them: Different providers might be installed to the system
and the tests could be run in that configuration.
Does it mean that we are going to copy unit tests to a parallel 'system' test
suite?
No - if you reuse something, that's cool.
What I'm trying to avoid is having a big bag of tests that require extra
things to be maintained in order for simple, common things to happen.
Case in point would be our current security2 testbed, where there's no
naming convention (yet) and there has to be exclusion lists of tests by
package in order for non-test code that is in the same packages to not
be run.
(the solution is a *Test.java naming convention that gives full freedom
for co-located support resources w/o the need for unncessary maintenance...)
Similarly, for things like system tests where there's going to be
special setup and configuration anyway (like modifying the classpath,
setting system properties, etc) then defining test cases from wherever
is less of a burden.
See what I'm getting at?
geir
Thanks,
Mikhail
On 1/27/06, Geir Magnusson Jr <[EMAIL PROTECTED]> wrote:
Mikhail Loenko wrote:
On 1/27/06, George Harley1 <[EMAIL PROTECTED]> wrote:
But because we live in a less than ideal world there will, no doubt, be
some tests that will demand an
environment that is impossible or at the very least difficult to mock up
for the majority of developers/testers.
I'm absolutely agree that we are neither living in the ideal world nor trying
to make it ideal :)
So until we got a 'system' test suite why should we weaken existing tests?
One solution could be to segregate those tests into a separate test suite
(available for all but primarily
for those working in the niche area that demands the special environment).
Moving this kind of tests would affect many people: they will see
separate suites,
try, ask questions...
If the test can be configured by a few people only who works on that
specific area and those people are aware of those tests why not just
print a log when the test is skipped?
Because the same set of people that will be bothered by separate suites
will have the same reaction to skipped tests.
This is why I advocate making a "separate tree" for the system tests -
make it clear that they are not the general unit tests...
It would not disturb most of the people because the test will pass in 'bad'
environment. But those, who know about these tests will sometimes grep
logs to validate configuration.
IMO, there's too much special information there, too much config. I'm a
simple person, and like things clean and simple. I don't like to mix
concerns when possible, and here's a place where it's definitely
possible to separate cleanly.
I don't see the downside.
geir
Thanks,
Mikhail
Alternatively, they could be
included as part of a general test suite but be purposely skipped over at
test execution time using a
test exclusion list understood by the test runner.
Best regards,
George
________________________________________
George C. Harley
Tim Ellison <[EMAIL PROTECTED]>
27/01/2006 08:53
Please respond to
[email protected]
To
[email protected]
cc
Subject
Re: [testing] code for exotic configurations
Anton Avtamonov wrote:
Note that I could create my own provider and test with it, but what I
would
really want is to test how my EncryptedPrivateKeyInfo works with
AlgorithmParameters from real provider as well as how my other classes
work
with real implementations of crypto Engines.
Thanks,
Mikhail.
Hi Mikhail,
There are 'system' and 'unit' tests. Traditionally, unit tests are of
developer-level. Each unit test is intended to test just a limited
piece of functionality separately from other sub-systems (test for one
fucntion, test for one class, etc). Such tests must create a desired
environment over the testing fucntionality and run the scenario in the
predefined conditions. Unit tests usually able to cover all scenarios
(execution paths) for the tested parts of fucntionality.
What are you talking about looks like 'system' testing. Such tests
usually run on the real environment and test the most often scenarious
(the reduntant set, all scenarios usually cannot be covered). Such
testing is not concentrated on the particular fucntionality, but
covers the work of the whole system.
A sample is: "run some demo application on some particular platform,
with some particular providers installed and perform some operations".
I think currently we should focus on 'unit' test approach since it is
more applicable during the development (so my advise is to revert your
tests to install 'test' providers with the desired behavior as George
proposed).
However we should think about 'system' scenarios which can be run on
the later stage and act as 'verification' of proper work of the entire
system.
I agree with all this. The unit tests are one style of test for
establishing the correctness of the code. As you point out the unit
tests typically require a well-defined environment in which to run, and
it becomes a judgment-call as to whether a particular test's
environmental requirements are 'reasonable' or not.
For example, you can reasonably expect all developers to have an
environment to run unit tests that has enough RAM and a writable disk
etc. such that if those things do not exist the tests will simply fail.
However, you may decide it is unreasonable to expect the environment to
include a populated LDAP server, or a carefully configured RMI server.
If you were to call that environment unreasonable then testing JNDI and
RMI would likely involve mock objects etc. to get good unit tests.
Of course, as you point out, once you are passing the unit tests you
also need the 'system' tests to ensure the code works in a real
environment. Usage scenarios based on the bigger system are good, as is
running the bigger system's test suite on our runtime.
Regards,
Tim
--
Anton Avtamonov,
Intel Middleware Products Division
--
Tim Ellison ([EMAIL PROTECTED])
IBM Java technology centre, UK.
