SetUp functions for multiple test cases

[brooklineTom]

On Jan 20, 9:57 am, Roy Smith <r... at panix.com> wrote:
> In article
> <45b0bf56-673c-40cd-a27a-62f9943d9... at r41g2000prr.googlegroups.com>,
>  Georg Schmid <gspsch... at gmail.com> wrote:
>
> > I've just started working with unittests and already hit a snag. I
> > couldn't find out how to implement a setup function, that is executed
> > only _once_ before all of the tests. Specifically, I need this for
> > testing my database interface, and naturally I don't want to create a
> > new database in-memory and fill it with example data for every single
> > test case.
>
> Short answer -- there's no way to do it in the unittest framework.
>
> The True Unit Test Zealots will argue that all tests should be 100%
> independent of each other, which means there should be NO common state
> between test cases.  For that matter, they will also argue that unit tests
> should not interface with external resources like databases.  And they're
> right.
>
> On the other hand, the Real World Test Pragmatists will argue that this is
> just not practical in all cases.  Real programs have classes which interact
> with the outside world, and they need to get tested.  You could stub out
> the external resource, but that's a lot of work, and may introduce as many
> problems as it solves.  Sometimes, a big part of what you're testing is
> your understanding of the external world (i.e. "does this really work like
> it's documented?").
>
> Plus, some operations are just too expensive to do for every test case.  I
> don't know how long it takes to build your in-memory database.  If it takes
> one second, it probably makes sense to do it for every test case (unless
> you've got thousands of test cases).  If it takes 10 minutes, then it makes
> sense to do it once and deal with the fact that you're violating True Unit
> Test Dogma.
>
> Anyway, do what I do.  I run the tests with a:
>
> if __name__ == "__main__":
>    blah blah
>
> block at the bottom of the test file.  Just do your "do once" setup in that
> code block and store the result in a global.
>
> You might have your setUp() method re-assign the global to an instance
> variable and then your test cases can access it via self.whatever.  The
> reason for that is if at some point in the future you change your mind and
> decide to re-build the database in setUp() for each test, you just have to
> change setUp(), not touch the individual test cases.

I believe that the answer sought by the OP is the TestResource class
from the original Smalltalk unit test. Sadly, the various language-
specific clones -- including the Python unittest package -- followed
the example of the JUnit hack, which elided this (for a variety of
reasons, some good, some bad). There are two major obstacles in moving
SUnit from Smalltalk to Python: (1) SUnit takes full advantage of
class objects which still have only vestigial implementations in
Python, and (2) SUnit takes full advantage of Smalltalk's restartable
exception semantics.

I've implemented an alternative unit test package, but I requires the
rest of my framework (which addresses the above two restrictions) and
is therefore not yet ready for prime time.

The basic theme of TestResource is simple enough, once the rest of the
unit test framework is properly factored:
1. Tests always run within an instance of TestSuite.
2. The "run" method of test suite (with no arguments) collects and
creates whatever resources are needed by the tests within it.
3. The run method creates an instance of TestResult to collect the
results.
4. Each TestCase instance collects whatever (existing) resources it
needs.
5. When the "run" method (in step 2 above) finishes, all resources are
finalized.

Instances of TestResource have a "setUp" and "tearDown" method, just
like TestCase. These are called once, at steps 2 and 5 above.

I built my test frame from the original SUnit from KentBeck et al and
used an excellent reference Joseph Pelrine (http://www.metaprog.com/
ESUG/TestResources.pdf) to guide my work.