Query in v2

In v1, I modeled my query API for ichiban/prolog after database/sql. As I’m working on v2, I’m experimenting with an improved query API with newer Go features.

With recent Go features, including generic methods introduced in the newest Go 1.27, the query API can be as simple as a for range loop over Query().

for r, err := range p.Query[struct {
	Who prolog.Atom
}](ctx, `mortal(Who).`) {
	if err != nil {
		panic(err)
	}
	
	fmt.Printf("Who = %s\n", r.Who) // ==> Who = socrates
}

Compared with v1, which requires Query(), Next(), Scan(), Err(), and Close(), this is considerably simpler.

sols, err := p.Query(`mortal(Who).`)
if err != nil {
	panic(err)
}
defer sols.Close()

for sols.Next() {
	var s struct {
		Who string
	}
	if err := sols.Scan(&s); err != nil {
		panic(err)
	}

	fmt.Printf("Who = %s\n", s.Who) // ==> Who = socrates
}
if err := sols.Err(); err != nil {
	panic(err)
}

The new standard iterator abstraction iter.Seq2[K, V], introduced in Go 1.23, does the heavy lifting. It allows Query() to hide the explicit calls to Next(), Err(), and Close() from the caller. Together with range-over-function support, it lets callers iterate over results using a single for range loop.

Generics, introduced in Go 1.18 and extended to support generic methods in 1.27, allow Query() to return results in a specified Go type, eliminating the need for an explicit Scan().

The goal hasn’t changed since v1: to make Prolog feel natural to Go developers. Go itself has simply given me better tools to achieve it.