Cursors

Built upon transducers and Lenses, cursors are a great way to bridge value-oriented designs and object-oriented systems. It enables modularity by allowing one to write components that reference mutable and observable data abstracting away the shape of the data storage.

Types of cursors

Cursors are wrappers for models where you can get the current version of a model, watch it for changes, transform into other cursors and (for some types of cursors) replace the model with another version:

There are many types of cursors in Lager.

Based on their interfaces, one can divide cursors into read-only cursors, write-only cursors, and read-write cursors.

Based on their source of information, one can divide cursors into root cursors which directly holds the model, and derived cursors which does not directly hold the model.

Interfaces

  • lager::reader is the read-only cursor interface in lager. It provides get() and watch() functionalities:

    #include <lager/reader.hpp>
    
    // pseudo-code for the call signatures
    model_type lager::reader<model_type>::get();
    void lager::watch(lager::reader<model_type> cursor,
                      std::function<void(model_type)> callback);
    

    get() will return the value of the model in the cursor. watch() will make the cursor call callback every time the content in it has changed (checked using operator==()).

  • lager::writer is the write-only cursor interface in lager. It provides the set() and update() functionalities:

    #include <lager/writer.hpp>
    
    // pseudo-code for the call signatures
    void lager::writer<model_type>::set(model_type new_model);
    void lager::writer<model_type>::update(
        std::function<model_type(model_type)> callback);
    

    set() will replace the value of the model in the cursor. update() will call callback with the current value of the model in the cursor, and replace the value with what callback returns.

  • lager::cursor is the read-write cursor interface in lager. It inherits from lager::reader and lager::writer, and has the functionalities of both. The lager::cursor class is can be made available by:

    #include <lager/cursor.hpp>
    

Root cursors

There are four root cursors in lager. All these cursors can serve as the “single source of truth” for other cursors.

  • lager::state is a subclass of lager::cursor. One can create a lager::state by:

    #include <lager/state.hpp>
    
    auto state = lager::make_state(model_value);
    // or
    auto state = lager::make_state(model_value, tag_value);
    

    tag_value is a instance of one of lager::transactional_tag and lager::automatic_tag. If it is not provided, transactional_tag is used.

    The difference between these two are when the value in the state gets propagated (i.e. becomes accessible via get()). transactional_tag requires a lager::commit() call before the value gets propagated, while automatic_tag does not:

    using model = int;
    
    auto state = lager::make_state(model{});
    std::cout << state.get() << std::endl; // 0
    state.set(1);
    std::cout << state.get() << std::endl; // 0
    lager::commit(state);
    std::cout << state.get() << std::endl; // 1
    
    auto state2 = lager::make_state(model{}, lager::automatic_tag{});
    state2.set(2);
    std::cout << state2.get() << std::endl; // 2
    
  • lager::store is a subclass of lager::reader. It makes changes to models by dispatching Actions, instead of the set() function. One can create a lager::store by the following code. For more information, see store.

    #include <lager/store.hpp>
    
    auto store = lager::make_store<action>(
        model, event_loop, enhancers...);
    
  • lager::sensor is a subclass of lager::reader. It takes a function and use its result as the value of the underlying model.

    #include <lager/sensor.hpp>
    
    int foo = 5;
    auto func = [&] { return foo; };
    
    auto sensor = lager::make_sensor(func);
    

    One can make the sensor re-evaluate the function and update the value inside it. The re-evaluation only happens when lager::commit() is called on the sensor.

    #include <lager/commit.hpp>
    
    std::cout << sensor.get() << std::endl; // 5
    foo = 8;
    std::cout << sensor.get() << std::endl; // 5
    lager::commit(sensor);
    std::cout << sensor.get() << std::endl; // 8
    
  • lager::constant is a subclass of lager::reader. It takes a value and use it as the value of the underlying model. The value cannot be changed later.

    #include <lager/constant.hpp>
    
    int foo = 5;
    auto constant = lager::make_constant(5);
    
    // Always prints 5, as long as `constant` is not re-assigned
    std::cout << constant.get() << std::endl;
    

Derived cursors

Derived cursors are all cursors that are not root cursors. They are obtained by transforming other cursors using the methods described below.

Zooming with lenses

One can use zoom() method to zoom a cursor into another:

auto cursor_type<model_type>::zoom(lens<model_type, model_part>)
    -> maybe_other_cursor_type<model_part>;

For example:

#include <lager/state.hpp>

using map_t = immer::map<std::string, int>;
using arr_t = immer::array<std::string>;
struct whole {
    part p;
    map_t m;
    arr_t a;
};

lager::state<whole> state = lager::make_state(whole{});
lager::cursor<part> part_cursor =
    state.zoom(lager::lenses::attr(&whole::p));
lager::cursor<map_t> map_cursor =
    state.zoom(lager::lenses::attr(&whole::m));
lager::cursor<int> int_cursor =
    map_cursor.zoom(lager::lenses::at("foo"))
    .zoom(lager::lenses::or_default);
lager::cursor<std::string> str_cursor =
    state.zoom(lager::lenses::attr(&whole::a))
    .zoom(lager::lenses::at(0))
    .zoom(lager::lenses::value_or("no value"));

For convenience, one can also use the operator[], which takes a lens, key (index) or pointer to attribute. The latter two will be converted into a lens using lager::lenses::at and lager::lenses::attr automatically. The example above can also be written as:

lager::cursor<part> part_cursor =
    state[&whole::p];
lager::cursor<map_t> map_cursor =
    state[&whole::m];
lager::cursor<int> int_cursor =
    map_cursor["foo"][lager::lenses::or_default];
lager::cursor<std::string> str_cursor =
    state[&whole::a][0][lager::lenses::value_or("no value")];

Transformations

The xform() function is another way to transform the cursor. For read-only cursors (lager::reader), it takes one transducer (see zug for more information); for writable cursors (lager::writer and lager::cursor), it can take two to transform into another writable cursor, or take one to transform into a read-only cursor.

lager::reader<std::string> str = ...;
// One-way transformation for read-only cursors
lager::reader<int> str_length = str.xform(
    zug::map([](std::string x) { return x.size(); }));

lager::cursor<std::string> str = ...;

// Two-way transformation for writable cursors
lager::cursor<int> num = str.xform(
    zug::map([](std::string x) { return std::stoi(x); }),
    zug::map([](int x) { return std::to_string(x); })
);

// One-way transformation to make a read-only cursor
lager::reader<int> num2 = num.xform(
    zug::map([](int x) { return 2*x; }));

str.set("123");
// You need `lager::commit(state);`
// if you use transactional_tag
std::cout << num.get() << std::endl; // 123
num.set(42);
std::cout << str.get() << std::endl; // 42
std::cout << num2.get() << std::endl; // 84

Combinations

You can combine more than one cursors into one using with(). The resulted cursor will be of a std::tuple containing all the value types in the original cursors:

#include <lager/with.hpp>

lager::cursor<int> num = ...;
lager::cursor<std::string> str = ...;
lager::cursor<std::tuple<int, std::string>> dual =
    lager::with(num, str);

// If any of the cursors passed into with() are read-only,
// it will result in a read-only cursor.
lager::reader<std::string> str_ro = ...;
lager::reader<std::tuple<int, std::string>> dual_ro =
    lager::with(num, str_ro);

Using cursors

We will use a minimal example to show how cursors work. Suppose one wants to represent a house using the following models, actions and reducers:

#include <lager/util.hpp>

struct room
{
    bool light_on;
};

struct toggle_light_action {};
using room_action = std::variant<toggle_light_action>;

room update_room(room r, room_action a)
{
    return std::visit(lager::visitor{
        [=](toggle_light_action) {
            return room{ ! r.light_on };
        }
    }, a);
}

struct house
{
    immer::map<std::string, room> rooms;
};

struct change_room_action
{
    std::string id;
    room_action a;
};
struct add_room_action
{
    std::string id;
    room r;
};
using house_action = std::variant<change_room_action,
                                  add_room_action>;

house update(house h, house_action a)
{
    return std::visit(lager::visitor{
        [&](change_room_action a) {
            auto old_room = h.rooms[a.id];
            auto new_room = update_room(old_room, a.a);
            // For simplicity we do not add move semantics
            // here, but you should do in your own program
            h.rooms = h.rooms.set(a.id, new_room);
            return h;
        },
        [&](add_room_action a) {
            h.rooms = h.rooms.set(a.id, a.r);
            return h;
        }
    }, a);
}

Create the single source of truth

As discussed above, we have two choices for our single source of truth: lager::state or lager::store. If you are refactoring old code, it may be a good choice to use lager::state because it allows you to gradually lift the state up without rewriting everything at once. If you are developing new software, it may be worthy to to use lager::store to benefit from the use of actions.

Here, we will use lager::store as an example.

#include <lager/store.hpp>
#include <lager/event_loop/manual.hpp>

// Make an initial model
house initial_house;
initial_house.rooms = initial_house.rooms
    .set("kitchen", room{false});
initial_house.rooms = initial_house.rooms
    .set("bedroom", room{true});

auto store = lager::make_store<house_action>(
    initial_house,
    // Be sure to use a suitable event loop
    // that integrates into the rest of your program
    lager::with_manual_event_loop{});

Zooming the cursors

Suppose we want to access and watch the state of the kitchen. We can use the zoom() method to obtain a cursor just for that:

#include <lager/lenses/at.hpp>
#include <lager/lenses/attr.hpp>
#include <lager/lenses/optional.hpp>

lager::reader<room> kitchen_cursor = store
    .zoom(lager::lenses::attr(&house::rooms))
    .zoom(lager::lenses::at("kitchen"))
    // maybe you want to use some other
    // approach to deal with this std::optional
    .zoom(lager::lenses::or_default);

// You can now query for the state:
auto kitchen = kitchen_cursor.get();

auto kitchen_light_on = kitchen.light_on;

Using cursors in object-oriented views

Suppose we want to display our room in an object-oriented GUI library, we can make the widget receive a cursor to the room model and watch it for changes:

class room_component : public widget
{
    lager::reader<room> r;
    lager::reader<bool> light_on;
    label l;

    static std::string light_state(bool on) {
        return on ? "light is on" : "light is off";
    }
public:
    room_component(lager::reader<room> r, widget *parent = 0)
        : widget(parent)
        , r(r)
        , light_on(r[&room::light_on])
        , l(light_state(light_on.get()))
    {
        lager::watch(light_on, [&] (bool on) {
            l.set_text(light_state(on));
        });
    }
};

Note

A lager::watch(reader, ...) is bound to the reader object. It is simply an alias of reader.watch(...). This means that when the reader object goes out of scope, the watch is disposed. For example:

void setup_watch() {
    auto reader = my_store[&my_model::foo];

    lager::watch(reader, [] (auto value) {
        std::cout << value << std::endl;
    });
}

Because the reader is freed when setup_watch() returns, the watch is disposed and will never get called. Instead store the reader object in a class member or somewhere else where it lives for at least as long as the watch is necessary.

Dispatching actions

Of course, we do not want the GUI to only display the model. Instead, we would like to allow it make changes to our model. Here, since we are using lager::store as our single source of truth, we benefit from making changes through actions.

We dispatch actions through contexts. Here, lager::store is a context. We may directly dispatch actions via the store:

store.dispatch(change_room_action{"kitchen",
               toogle_light_action{}});

But for the room_component we have here, it may not be a great idea, because it breaks modularity. If we were to dispatch an action via store, the room component will need to know the room’s id. In other words, it has to know something about the house, rather than only know about the room itself. We would like to have a context that can dispatch a room_action, instead of a house_action:

ctx.dispatch(toogle_light_action{}); // what should ctx be?

Fortunately, lager provides a context conversion constructor that can be used here, and the only thing we would like to do is to provide a conversion function that converts a room_action into a house_action:

std::string room_id = "kitchen";
auto ctx = lager::context<room_action>(
    store,
    [=](room_action a) -> house_action {
        return change_room_action{ room_id, a };
    });

And now we can add a toggle button to our room component to control the light:

class room_component : public widget
{
    lager::reader<room> r;
    lager::reader<bool> light_on;
    lager::context<room_action> ctx;
    label l;
    button b;

    static std::string light_state(bool on) {
        return on ? "light is on" : "light is off";
    }
public:
    room_component(lager::reader<room> r,
                   lager::context<room_action> ctx,
                   widget *parent = 0)
        : widget(parent)
        , r(r)
        , light_on(r.zoom(lager::lenses::attr(&room::light_on)))
        , ctx(ctx)
        , l(light_state(light_on.get()))
        , b("Toogle light")
    {
        lager::watch(light_on, [&](bool on) {
            l.set_text(light_state(on));
        });

        b.clicked.connect([ctx=this->ctx]() {
            ctx.dispatch(toogle_light_action{});
        });
    }
};

Additional resources

To learn more about cursors, you can watch the C++ Russia 2019 Talk: Squaring the circle: value oriented design in an object oriented system (slides).