ForPath: For-Comprehensions with Effect Paths

"Though this be madness, yet there is method in't."

-- William Shakespeare, Hamlet

And so it is with for-comprehensions: what appears to be arcane syntax hides a deeply methodical approach to composing sequential operations.

The Problem: Bridging Two Worlds

The standard For class provides powerful for-comprehension syntax, but it operates on raw Kind<M, A> values and requires explicit Monad instances. When working with the Effect Path API, this creates friction:

// Using standard For with Path types requires extraction and rewrapping
Kind<MaybeKind.Witness, Integer> kindResult = For.from(maybeMonad, path1.run().kind())
    .from(a -> path2.run().kind())
    .yield((a, b) -> a + b);

MaybePath<Integer> result = Path.maybe(MAYBE.narrow(kindResult));

The intent is clear, but the ceremony obscures it. ForPath eliminates this friction:

// ForPath works directly with Path types
MaybePath<Integer> result = ForPath.from(path1)
    .from(a -> path2)
    .yield((a, b) -> a + b);

The comprehension accepts Path types and returns Path types. No manual extraction, no rewrapping, no boilerplate.

Entry Points

ForPath provides entry points for each supported Path type:

All nine path types support up to 12 chained bindings. The when() guard operation is only available for Path types backed by MonadZero, which can represent emptiness or failure.

Core Operations

Generators: .from()

The .from() operation extracts a value from the current step and chains to a new Path-producing computation. This is the monadic bind (flatMap) in disguise.

MaybePath<String> result = ForPath.from(Path.just("Alice"))
    .from(name -> Path.just(name.length()))         // a = "Alice", b = 5
    .from(t -> Path.just(t._1() + ":" + t._2()))    // t is Tuple2<String, Integer>
    .yield((name, len, combined) -> combined);      // "Alice:5"

Each .from() adds a new value to the accumulating tuple, making all previous values available to subsequent steps.

Value Bindings: .let()

The .let() operation computes a pure value from accumulated results without introducing a new effect. It's equivalent to map that carries the value forward.

MaybePath<String> result = ForPath.from(Path.just(10))
    .let(a -> a * 2)                    // b = 20 (pure calculation)
    .let(t -> t._1() + t._2())          // c = 30 (can access tuple)
    .yield((a, b, c) -> "Sum: " + c);   // "Sum: 30"

Guards: .when()

For Path types with MonadZero (MaybePath, OptionalPath, NonDetPath), the .when() operation filters results. When the predicate returns false, the computation short-circuits to the monad's zero value (Nothing, empty, etc.).

MaybePath<Integer> evenOnly = ForPath.from(Path.just(4))
    .when(n -> n % 2 == 0)              // passes: 4 is even
    .yield(n -> n * 10);                // Just(40)

MaybePath<Integer> filtered = ForPath.from(Path.just(3))
    .when(n -> n % 2 == 0)              // fails: 3 is odd
    .yield(n -> n * 10);                // Nothing

Projection: .yield()

Every comprehension ends with .yield(), which maps the accumulated values to a final result. You can access values individually or as a tuple:

// Individual parameters
.yield((a, b, c) -> a + b + c)

// Or as a tuple for many values
.yield(t -> t._1() + t._2() + t._3())

Optics Integration

ForPath integrates with the Focus DSL for structural navigation within comprehensions.

Extracting with .focus()

The .focus() operation uses a FocusPath to extract a nested value:

record User(String name, Address address) {}
record Address(String city, String postcode) {}

// Create lenses for each field
Lens<User, Address> addressLens = Lens.of(
    User::address, (u, a) -> new User(u.name(), a));
Lens<Address, String> cityLens = Lens.of(
    Address::city, (a, c) -> new Address(c, a.postcode()));

// Compose paths with via() for nested access
FocusPath<User, Address> addressPath = FocusPath.of(addressLens);
FocusPath<User, String> userCityPath = addressPath.via(FocusPath.of(cityLens));

MaybePath<String> result = ForPath.from(Path.just(user))
    .focus(userCityPath)                        // extract city directly
    .yield((user, city) -> city.toUpperCase());

Alternatively, chain focus operations where the second takes a function:

FocusPath<User, Address> addressPath = FocusPath.of(addressLens);

MaybePath<String> result = ForPath.from(Path.just(user))
    .focus(addressPath)                         // extract address -> Steps2
    .focus(t -> t._2().city())                  // extract city from tuple
    .yield((user, address, city) -> city.toUpperCase());

Pattern Matching with .match()

The .match() operation uses an AffinePath for optional extraction. When the focus is absent, the comprehension short-circuits for MonadZero types:

sealed interface Result permits Success, Failure {}
record Success(String value) implements Result {}
record Failure(String error) implements Result {}

AffinePath<Result, Success> successPath = AffinePath.of(
    Affine.of(
        r -> r instanceof Success s ? Optional.of(s) : Optional.empty(),
        (r, s) -> s
    )
);

MaybePath<String> result = ForPath.from(Path.just((Result) new Success("data")))
    .match(successPath)                         // extract Success
    .yield((r, success) -> success.value().toUpperCase());
// Just("DATA")

MaybePath<String> empty = ForPath.from(Path.just((Result) new Failure("error")))
    .match(successPath)                         // fails to match
    .yield((r, success) -> success.value());
// Nothing

When to Use ForPath vs For

For monad transformer stacks, the standard For class remains the appropriate choice as it works directly with the transformer's Kind representation.

./gradlew :hkj-benchmarks:jmh --includes=".*ForPathVTaskBenchmark.*"

Previous: Composition Patterns Next: ForPath Examples