Proposal: allow member access operator to accept a Struct?, Object?, or Pair? on the LHS

[peterhuene]

Currently the member access operator accepts Struct, Object, calls, and, as mentioned elsewhere in the spec, Pair.

However, it does not accept a Struct?, Object?, or Pair? and a type coercion from T? -> T rightfully does not exist.

As a result of this limitation, the spec and ecosystem tooling suggest calling the select_first function passing an array literal containing only the optional value e.g. select_first([x]). If x is None, then a runtime error is generated by the engine as defined in the spec for the select_first function.

To me, the select_first workaround seems overly verbose and incurs the overhead of a naive array value allocation and function call when x is not None just to access its members.

While I'm relatively new to WDL and I lack the historical context for the design of the member access operator, it seems like it could explicitly accept T? (where T is a type that is otherwise accepted for the member access operator) on the left hand side of the expression and raise a runtime error if the value is None. This would lead to better readability of WDL source too: x.y instead of select_first([x]).y, which would effectively have the same semantics.

[patmagee]

related to #551