Github user paul-rogers commented on the issue:
https://github.com/apache/drill/pull/1138
As @arina-ielchiieva points out, this change backs out plan-time knowledge
of schema. This may not affect run-time accuracy. However, it does mean that
queries can be planned, based on not knowing types, that fail at runtime when
types are learned. This seems more like a bug that a feature. In general, we
should use all information available. It is not helpful to ignore information
if doing so results in poorer user experience.
---
