Feature: drill-through

[rpbouman]

This is a proposal to have a feature that would let you see the actual records from the source dataset that contribute to a particular selection from the result.

[rpbouman]

This feature could be implemented by having a few distinct features that would be useful in their own right:

• "keep" feature on the axes: User selects a few tuples, and chooses "keep" (probably from the context menu). This would result in a filter being added that selects only the tuples that were selected
• "relational row id" or "source row id" derived attribute. This would be a special virtual attribute that assigns a unique key (preferably, rownumber() or sequential int) to each row in the source data set. This item would normally not be part of any analytical query. But once the user adds this item to an existing analytical query, it would force grouping on the lowest level, thus "de-aggregating" the result. If we'd use rownumber() to implement it, we have to be careful to assign that before applying any conditions. Basically, selecting this item while also filtering implies generating an extra "stage" (cte) in our query logic, where we first select the raw data along with the rownumber(), and only then apply the filter.