Re: Running Mauve tests with JUnit

[David Gilbert <david.gilbert@object-refinery.com>]

Hi Dalibor,

I had meant to keep this on the Mauve lists, but I'll reply to the 
Classpath list also...

Dalibor Topic wrote:

>On Wed, Mar 08, 2006 at 10:40:55AM +0000, David Gilbert wrote:
>  
>
>>I didn't get any feedback about this...anyone think it is a good/bad idea?
>>    
>>
>excellent idea in my opinion. Have you looked at graydon's junit mauve
>bridge at
>http://sources.redhat.com/ml/mauve-discuss/2003-q4/msg00003.html ?
>
>  
>
I hadn't seen Graydon's bridge class, thanks for the link (and I should 
do more research next time).  Looking over it, it has the advantage that 
it doesn't require any existing Mauve testlets to be modified (and we 
have a lot of testlets), but the disadvantage that it doesn't buy you 
much in terms of integration with IDEs (you still have to generate the 
test list ['classes'] file, for instance, which is the major stumbling 
block that people seem to have when trying to run Mauve). 

By modifying the Mauve testlets in the way that I proposed, you can (for 
example) run a single test in Eclipse just by selecting the source file 
and clicking 'Run as --> JUnit test'.  I figured that was the sort of 
thing people were expecting.

JUnit does seem to me to be less flexible in terms of selecting subsets 
of tests, and it's approach of reporting a pass/fail for each test 
method (only) makes it, in my opinion, less suitable for the type of 
testing we are doing on GNU Classpath.  But I was careful in the 
"conversion" to retain the Mauve testlets so that we can continue 
running the tests in the traditional (Mauve) way.

>If we do something like that. I'd like to see the junit code from
>freenet merged in, to keep it simple to run mauve without external
>dependencies.
>  
>
Agreed.  I didn't have much trouble getting the tests to compile against 
the freenet code (a  basic GPLed implementation of the JUnit API for 
those that don't know what it is) but didn't get any meaningful output 
from running the tests against it yet.  I don't think that will be too 
hard to resolve.

Regards,

Dave