[vtk-developers] Interim policy for image baseline changes in master while topics are under review

[Marcus D. Hanwell]

Hi,

We had a discussion about possible solutions to the baseline images
changing in master, whilst I think we all agree that external data
using the process developed for ITK is the ideal solution we do not
currently have the time/resources to accomplish this. It also looks
like there was widespread rejection of using submodules as an interim
solution, and rebasing/merging master in was suggested.

After looking into that I don't think it is a viable solution, or it
would require more work than I first realized to make it work. Perhaps
Brad may know of a clever solution I didn't see, but making ctest
merge/rebase and correctly report merge conflicts does not appear to
be easy - I even dug into the C++ code in CMake looking for a better
solution. As regression images for existing tests don't happen that
often I would like to propose a simpler, although perhaps less
satisfying solution...

If you submit a topic to Gerrit for review and it has apparently
unrelated test failures and you/the reviewers are worried about this
then rebase and push again. I would like to strongly discourage
abandoning a topic and creating a new one for the rebase, while
rebasing makes it difficult to see what really changed disconnecting
reviews across multiple topics is far worse in my opinion.

I would rather not spend too much time coming up with a clever
solution to creating temporary merged heads with possible
conflicts/reporting these to the dashboard when we already have a good
technical solution to this problem - content based hashing using
external data. Feel free to swoop in with something cleverer and
easier that I missed though ;-)

Marcus