Hi, all. I’m building an analytics system. In this system, tables are created by the user-provided metadata on the fly, so there’s no way to know the column names in compile time. Even though, I still want the composability of Ecto query. The problem is, I can’t find a proper way to work with absolutely dynamic column names.
For example, when I want to add a where clause to some query, I can’t write
where(query, [q], field(q, ^column_name) > 100)
because the column_name is a string, not an atom.
But it should not be an atom since it was generated by user-specified metadata and thus has infinitely many possibilities. If I use String.to_atom/1 to convert the strings to atoms, the atom table can blow up over time.
By the way, fragment(^column_name) and fragment("#{column_name}") do not work either because fragment does not allow the first argument to be pinned or be a string with interpolation.
Is there a way to do such a job in Ecto, or should I use another library to achieve my goal?
Thanks.
