MAP_INPUT_ARRAY always fills an entry with "nan", even when no matches are found

When tupling with MATCH_INPUT_ARRAY, one obtains an array and a branch with the number of entries in the relation table for the given particle. However, when there is no relation found at all, this 'length' branch is still set to 1, and the arrays themselves are added with a "nan" entry. It is more intuitive to fill with an empty array, and set the length branch value to 0 - that already tells you there are no relations found, and avoids the extra use of "nan".

added Functors label

assigned to @amathad

Hi @ldufour , What you suggest would makes sense but would be in contradiction to what we have been saying so far i.e. "when there exists no relation an invalid_value of nan is tupled". However arrays are special and if you leave the array empty, as you suggest, we might end-up saving some small amount of space but this is very small amount.

Maybe ask for opinions: @erodrigu , @pkoppenb and @graven

The problem is not in disk space, but rather the meaning of the length of the array being the number of relations found. :-)

the

when there exists no relation an invalid_value of nan is tupled

makes sense for 'scalar' quantities where absence is not representable explicitly as such. But a (variable size) array is different, as it does have a natural 'state' which represents 'nothing here', namely zero size (*). So I completely agree with @ldufour on this one.

(*) one inefficient but correct way to implement std::optional<T> would be by using an std::vector<T> with the additional constraint that the number of entries is either zero or one...

Agree with it all. Have added an MR (Analysis!977 (merged)).

mentioned in merge request Analysis!977 (merged)

closed with merge request Analysis!977 (merged)

mentioned in commit Analysis@74696aad