On 01.09.2005 06:04:45 Manuel Mall wrote:
> We now have over 200 layout engine test cases in the repository which is 
> great.

Wow! I just hope nobody holds a grudge against me for introducing that
facility and pushing people to use it. ;-)

> However, with this ever growing number I wonder if we should put 
> some more structure to it. There is a real chance that we get more and 
> more duplication just because people wouldn't know which tests are 
> doing what so one starts writing new ones which may already be covered.

I agree.

> I don't want to suggest some complex system with the associated 
> management,  setup and on-going compliance overhead. But what about 
> some simple naming system along the following lines: 
> 
> Most current tests (not all) cover a particular feature and can be 
> described by the fo they target, the property they exercise and the 
> particular aspect of that combination they test. Therefore giving each 
> test file a name constructed like <fo name>[-<property name>]?
> [-<feature>]?[<serial number>]?.xml., e.g.
> 
> table-padding-relative.xml will test relative padding values on a 
> fo:table element.
> 
> Yes, this will give us some longer names but will make looking for a 
> particular test much easier as simple directory search/sort/filter 
> operations will do. It will also reduce the number of tests which are 
> identified just by a different non descript number, that is things like 
> padding1.xml, padding2.xml will be replaced by something more 
> meaningful. And yes, it will not cover every case especially once we 
> get into tests which deal with the interaction of multiple fos and 
> properties.

Sounds good. This was bound to produce problems when it reached a
certain size. One additional suggestion, though:

It would be good to separate feature tests from regression tests. The
latter could, for example, contain the Bugzilla number if a Bugzilla
issue is associated with it. I've thought about this myself a number of
times. I wonder if we should also separate the tests into multiple
directories.

> If agreement is found on this I am not sure what the best way to 
> actually do it with svn is. One way would be for someone (probably 
> me :-)) to rename all the files and for a committer to simple delete 
> everything in that directory in svn and submit all the renamed files as 
> new. That would loose some history but I don't think its a big deal for 
> these testcases.

Certainly not the best of approaches doing that via patches. I'd like to
keep the history, so this means someone with commit access will have to
do it. I'm sure we'll find a solution to that. It just takes a little
more time.


Jeremias Maerki

Reply via email to