Patchwatcher online

[James Hawkins]

On Tue, Aug 12, 2008 at 1:01 AM, Paul Vriens <paul.vriens.wine at gmail.com> wrote:
> Dan Kegel wrote:
>> On Mon, Aug 11, 2008 at 4:31 PM, Vijay Kiran Kamuju <infyquest at gmail.com> wrote:
>>> When running tests for the patch, I think we should just run the tests
>>> of the dlls that are affected direct;y or indirectly by that change.
>>> its running the tests for entire wine, which is very time consuming.
>>
>> True, but hey, it was easier to code.  And getting anything like this
>> working at all is pretty hard.  Figuring out which tests a give
>> patch affects is an extra challenge I'd rather not face just now.
>> Once it's up and working well we can refine it.
>
> I'd argue that testing just the affected dll is correct. What about things like
> patches to ntdll/kernel32/advapi32 (and the likes). They could influence far
> more tests then just the ones for it's own dll.
>

Which is an argument for why you should test the entire tree for each
patch.  If a patch to ntdll makes tests in another dll fail, the patch
should be rejected just as if the ntdll tests failed.  This is how
Julliard's work flow goes, so the test bot should do the same.

-- 
James Hawkins