Hi Hans, On Sun, Feb 4, 2018 at 10:30 PM, Hans Verkuil <hverkuil@xxxxxxxxx> wrote: > Hi Gustavo, Alexandre, > > As you may have seen I have been extending the v4l2-compliance utility with tests > for v4l-subdevX and mediaX devices. In the process of doing that I promptly > found a bunch of bugs. Mostly little things that are easy to fix, but much > harder to find without proper tests. > > You are both working on new API additions and I want to give a heads-up that > I want to have patches for v4l2-compliance that test the new additions to a > reasonable extent. > > I say 'reasonable' because I suspect that e.g. in-fence support might be hard > to test. But out-fences can definitely be tested and for in-fences you can > at minimum test what happens when you give it an invalid file descriptor. > > For the request API is it obviously too early to start writing tests, but > as the API crystallizes you may consider starting to work on it. > > I've decided to be more strict about this based on my experiences. I'm more > than happy to assist and give advice, especially since this is a bit of a late > notice, particularly for Gustavo. > > But every time we skip this step it bites us in the ass later on. > > It is also highly recommended to add fence/request support to the vivid and > vim2m drivers. It makes it much easier to find regressions in the API due to > future changes since you don't need 'real' hardware for these drivers. > > Again my apologies for the late notice, but it is better to make these tests > now while you are working on the new feature, rather than doing it much later > and finding all sorts of gremlins in the API. I completely agree with your reasoning and will consider updating v4l2-compliance as soon as we have something stable on the user-facing side (which hopefully should be close by now). Alex.