In the vectorized execution engine we use the CASE operator to support several expressions. We do so during the vectorized planning, but it seems more appropriate to do the transformation during the normalization stage of the optimization.

I believe that the execution in the row-by-row engine will not suffer performance-wise, so we'll be able to remove the implementations of Eval methods for the corresponding expressions.

In particular, the following constructs are implemented in the vectorized planning:

• WHERE a OR b
  • into WHERE CASE WHEN a THEN true WHEN b THEN true ELSE false END
• COALESCE(a, b, ...)
  • into CASE WHEN a IS DISTINCT FROM NULL THEN a WHEN b IS DISTINCT FROM NULL THEN b ... END
    - IF(a, b, c)
    - into CASE WHEN a THEN b ELSE c END
    - NULLIF(a, b)
    - into CASE WHEN a = b THEN NULL ELSE a END.

Jira issue: CRDB-13769