[SoC] Project proposal: Improving the testing workflow

[Damien Cassou]

2007/3/20, Andrew P. Black <black at cs.pdx.edu>:
> On 20 Mar 2007, at 3:46, Damien Cassou wrote:
>> - Linkage: it should be possible to know what tests must be run when a
>> method is modified. This should not be annoying for the developer (no
>> message box for example). The tests must NOT be run automatically.

> Why must the test NOT be run automatically?

While working on Dakar Testing, I interviewed Smalltalk engineers from
Georg Heeg eK. They told me it was far from a good idea to run tests
automatically. Here are the reasons I remember:

- you may be in the middle of a refactoring and you know your tests
will fail. No need to slow down things to get a known result
- to prevent from slowing down the developer, a solution to running
tests automatically would be to use an other process. This should be
avoided as much as possible. Processes are often sources of difficult
to find and correct bugs.

>Erik Ernst has shown dramatic
> benefits form running the test continuously and automatically with novice
> programmers.

Are you referring to JaTack?

>   Certainly, it should be possible to turn automatic testing
> (on and) off.  But once you have the dependencies figures out, why not
> provide the option of running all the tests that might have changed?


I want to provide that option but currently I plan to let the user
launch the changed method manually (a click on a "run all pending
tests" button).

-- 
Damien Cassou