Extend scan syntax to support project and filter

[julianhyde]

We propose to extend the syntax of a scan (from expression or join) in a query so that you can easily project a subset of fields and/or apply equality filters. After this issue is fixed,

from e as {deptno = 10, empno, job, ...} from scott.emps;

will be equivalent to

from e in (
  from e in scott.emps
    where e.deptno = 10
    yield {e.empno, e.job});
> val it =
>   [{empno=7782,job="MANAGER"},{empno=7839,job="PRESIDENT"},
>    {empno=7934,job="CLERK"}] : {empno:int, job:string} bag

Note that the output contains fields empno and job because empno is shorthand for empno = empno, and therefore a new field empno is bound (and similarly job). deptno = 10 causes the input to be filtered, but deptno is not returned. sal is not mentioned and therefore it is not returned.

The pattern-matching syntax (field = value to filter, ... for fields we don't want) already works but the records have been unpacked. There is no e for the employee record. With the as syntax, e as {empno, job, ...}, the query step will have one named field e, which is more convenient.

Consider Standard ML:

val i as {j = 1, k = k, ...} = {j = 1, k = "a", l = true};
> val i = {j=1,k="a",l=true} : {j:int, k:string, l:bool}
> val k = "a" : string

This expression binds both i and k, and does not remove the fields j and l from the record. Our proposal is somewhat at odds with this.

There seems to be a bug relating to field ordering. For example, in

from e as {job = "CLERK", deptno, job, ...} in scott.emps;
> java.lang.ClassCastException: class java.lang.Float cannot be cast to
> class java.lang.Integer (java.lang.Float and java.lang.Integer are in
> module java.base of loader 'bootstrap')

I suspect that values from the sal field are being incorrectly assigned to the deptno field.