I think that what James Hawkins had meant something like this:
Apply patch one, run the tests. If patch one fixed any of them, remove the todo_wine's for those tests as a part of patch one. Otherwise, leave them todo_wine.
Apply patch two, run the tests. If patch two fixed any of them, remove the todo_wine's for those tests as a part of patch two. Etc.
This way, if someone builds wine at any step of applying the five patches then the todo_wine's will reflect the current status of if the functions work or not. I think this is important for regression testing.
Of course, this will cause troubles if the patches are applied out-of-order or the later patches are applied without the first ones - but I think that's acceptable if you note that they should be committed in order.
--Matt Finicum
On 12/15/06, Clinton Stimpson cjstimpson@utwire.net wrote:
I can change the tests a bit, and change the currently empty functions to return E_NOTIMPL instead of S_OK. Then I can do it piecemeal. Is that how y'all want it?
Clint
Clinton Stimpson wrote:
Ok. There are 4 functions that have to be implemented at the same time in order to not break any tests, because of how the tests were written. A few days ago, I sent a single patch that implemented those 4 functions, including an update of the tests. It wasn't accepted, and it was suggested to break the patch up. But, I can't break it up without breaking the tests. I do have more patches coming after this batch, and those will be smaller and atomic.
So I guess I'm back to asking why my original patch wasn't accepted. ?? Should I resend it?
Thanks, Clinton
James Hawkins wrote:
On 12/14/06, Clinton Stimpson cjstimpson@utwire.net wrote:
Hi,
Part 5 of 5. Remove many todo_wine's from the tests, now that the functions are implemented.
You have to remove the todo_wine's in the same patch that fixes the tests, or the tests will fail for at least one commit. Patches have to be atomic and error free, and a failing test is an error.