Nightly test runs from 2008-09-24_01:15:01

[Andreas Pakulat]

On 24.09.08 16:01:21, David Nolden wrote:
> Just a general concern: While I like all the testing very much to find
> regressions, we still have to practice a tradeoff between creating
> tests and keeping stuff hackable. KDevelop is still in a very early
> phase, and a lot is going to change, so imo it doesn't make sense to
> create tests for _everything_, especially not for those parts that
> will anyway be turned upside down. It seems that this applies to large
> parts of the user interface. Imo tests should only be created for
> stuff that we consider _ready_, else the tests will just increase the
> resistance to finish these parts.

Actually tests are the _only_ thing needed when you want to start
turning a component upside-down. Else you never know what stuff you
break with it - unless you do extensive amount of manual testing. I'm
not saying you should have a 100% coverage (no matter which measurement
you use for that) for a component, but you should have at least the
common cases covered.

I'm not saying we should put all development on hold right now and do
only writing of tests. But doing a major rework of shell, interfaces or
outputview is insane without having a couple of tests in place before
that.

If writing a test for something is too hard such that you consider doing
something else instead of writing the test and then adjusting whatever
code you wanted to adjust, then something is wrong with our
testframework. And we should improve that part. I'm actually planning to
do some work in that direction for shell/ today and at the weekend.

Andreas
 
-- 
It may or may not be worthwhile, but it still has to be done.