ARCH: this panics when a given column is not in the schema, but since we already validated this schema during the ReturnType call, we should be fine.

OPTIM: we're doing a lot of type shenanigans at runtime - when we evaluate a function on each stripe, we do the same tree of operations - this applies not just here, but in projections.go as well - e.g.

NewFunction is one of the very few constructors as we have to do some fiddling here.

TODO(quoting): rules are quite non-transparent - unify and document somehow.

+.

TODO(next): is it really that AND and OR have the same precedence?.

myFunction(X).

==, !=.

>, <, <=, >=.

foo.bar.

-X or NOT X.

*.

Query describes what we want to retrieve from a given dataset There are basically four places you need to edit (and test!) in order to extend this: 1) The engine itself needs to support this functionality (usually a method on Dataset or column.Chunk) 2) The query method has to be able to translate query parameters to the engine 3) The query endpoint handler needs to be able to process the incoming body to the Query struct (the Unmarshaler should mostly take care of this) 4) The HTML/JS frontend needs to incorporate this in some way.