In the past weeks I've had a significant number of failing Travis CI
builds, the majority of which was failing in something that looked like
a docker build.
Is it correct and intended that Travis builds, or at least updates, a
docker image for every test? Maybe this step could be split out of the
standard tests and moved to a dedicated task? This may have the down-
side that tests would use whatever status the latest official docker
image has, which may or may not be up-to-date with upstream ressoures.
But it brings a significant speedup, much better error resilience and
last, not least, a more reproducible test experience.
Can anyone comment about the current design?
All the best,
Mario
