Well, that's what it is doing.
Look, if the query doesn't request any columns from the base data sources - tables or other information models - then the calculation engine removes those columns (column pruning).
If no columns of a underlying data source are requested, there is no way that any records/tupels are being handed over to the next node.
The next node in this case is your projection where the computed column is added to every tupel that got fed into the projection node.
No columns from the base table selected means no tupels being fed into the projected which means: no computed columns to be added.
You could imagine this as an inner join between your base columns and the computed columns.
If no base columns are there to be joined/glued to, then you don't get any records back.
Hope this makes it a bit clearer.
- Lars