Github user orhankislal commented on the issue:
https://github.com/apache/madlib/pull/325
Another alternative is to integrate the madlib deployment into the IC/DC.
What I mean is similar to how PostGIS runs its unit tests. IC/DC creates a
temporary database/schema, deploys the MADlib over there, runs the tests as
usual and then removes the temporary database/schema. This will inevitably
increase the running time of IC/DC but I believe it will be more stable. Since
we assume that the user might be using the madlib deployment schema, it is also
possible that they drop and/or recreate UDTs, UDFs and UDAs. Our IC/DC does not
account for a case like that and will probably fail.
It will also mean that a user can run IC/DC before they deploy it to their
target database. I would assume most users are already using a similar workflow
(temp database -> deploy MADlib -> run IC -> deploy MADlib on actual target)
---
