[Gzz] Re: Tests (was: Re: [Gzz-commits] gzz/gzz/view LinebrokenCellContentView.java)

[Benja Fallenstein]

Tuomas Lukka wrote:
> On Fri, Jan 10, 2003 at 09:01:34PM +0100, Benja Fallenstein wrote:
> > Well, fact: Nobody has taken the time to design the view framework so 
> > that it can be tested sanely. If you're so serious about this, someone 
> > should delete the whole gzz.view hierarchy right now and we should 
> > re-write it, test first. (It is allowed to delete anything that won't 
> > break tests, by your reasoning, so you wouldn't even have to write a PEG.)
>
>
> In the extreme case, yes. Of course, deletion of a class for no reason
> wouldn't be looked at nicely; there needs to be *some* reason for the action.
[...]
> What I want to make sure is that people don't start being afraid 
> of changing things: did you read mudyc's change to gfx/demo/gldemo.py?
> That is a symptom of something bad starting to happen when people
> are afraid of changing things: everything starts getting more
> complicated because no-one dares to change or delete anything
> and just adds another variation.

Ok, I agree with this put-into-perspective statement :)

> Why is the view hierarchy not testable in its current form? 

Hm, I think I can explain easier by giving some examples of what we need:

- Mock fonts (could be a mock TextStyle class maybe) for which we know
  the return values in advance, so that we don't depend on the fonts
  and particularities of one specific computer/GraphicsAPI.
- A way to test vob hierarchies easily. It should be possible to give
  the expected vob hierarchy declaratively.
- A way to test, 'which vob will be shown at window coordinates (x,y)?'
  This should allow testing of the *results* of vob hierarchies
  without actually looking into the hierarchies themselves.

That's what I can think of at the moment, but there may be more... 
anyways, with tools like these, testing should become substantially easier.

Fortunately, we do already have a structure that doesn't paint directly 
on the screen, so if the vobs work (needs to be separately tested), we 
only need to test that the right vob structure is created (making our 
task much easier).

> How should it be rewritten?

I don't know. Thinking more about it, additional functionality like the 
above may even suffice...

> > It may be more sensible to actually make a plan how we can get the 
> > currently untested parts of the code in tested form. 
>
> That would be really good. Should we just take a list of classes and assign
> them to people in TODO?

I guess that could work... hmmm... I think I'd prefer a "plan gzz.view 
testing" item after the ht'03 deadline :-)

- Benja