On Sun, Jan 03, 2016 at 10:24:47AM +0100, Georg Baum wrote: > Scott Kostyshak wrote: > > > On Fri, Jan 01, 2016 at 01:22:53PM +0100, Kornel Benko wrote: > >> > >> What we have now is _one_ step in this direction. > > > > Dedicated tests requires understanding of the underlying mechanism. I do > > not know much at all about lyx2lyx (I'm not sure Kornel does either?). > > I agree that the unspecific tests have been useful, but nevertheless I would > not spend further time on expanding them.
I see. > _Any_ non-trivial test requires understanding of the underlying > mechanism (and the unspecific tests are far from trivial already), as > this thread shows: Without understanding it looked like a bug, but > with understanding it turned out to be a false alarm. Yes, but I fully expect there to be false alarms. It is just a question of whether the signal to noise ratio is acceptable. I am not happy that I wasted your time with this false alarm, but I still think that overall it has been worth it. > > But despite this lack of knowledge we were able to discover bugs through > > the use of the less-than-ideal tests. I would love to have dedicated > > tests but for them I think we need the people that implement features or > > fix bugs to make the tests. > > True. Therefore it is IMHO better to work on the infrastructure for such > dedicated tests (adding the machinery to test individual lyx2lyx steps for > example) instead of doing more work on existing files in the hope that this > might discover more bugs. I would have added a dedicated test for the 500- > >501 change for example if this would have been as simple as adding a file > and adding it to a list. Yes, this is a good point. I agree. Hopefully we can make progress on unit tests in the 2.3 cycle, for both lyx2lyx and our C++ code and anything else. I would be happy to eventually learn the testing framework and try to contribute. > PS: Happy new year to all! Happy new year! Scott
signature.asc
Description: PGP signature
