Scalar Subqueries

If q is a query with value type T, and if:

T is a single-column type (this is checked at run time) and
q produces exactly one row (this is checked at run time),

then scalar(q) builds and returns an exprn_mapper<T>, which executes by executing q and taking q's output as its value.

In this context, q is known as a scalar subquery.

Scalar subqueries (like the subqueries that we pass to exists()) are allowed to use mappers that belong to an outer context.

We saw a typical use of scalar() in the introduction:

exprn_mapper<int64_t>
score(const exprn_mapper<serial> &movie_id) {
    return scalar(
        critics
        .where(critics->favorite_movie == movie_id)
        .select(count_all)
    );
}

It is typical in these ways:

The scalar subquery's output comes from an aggregate function. Here it is the function constant count_all, but any aggregate function is normal. This ensures that the scalar subquery will produce exactly one row.
The scalar subquery uses mappers that belong to an outer context. This is not required, but it's something that useful scalar subqueries tend to do.

The example also shows a typical feature of my coding style: When I'm building a query that depends on an outer context to provide a mapper, I write a C++ function for the purpose, with an argument to advertise the dependency.