That makes me know that, we we we we're doin'
We had the right idea in the beginning
And and we just need to maintain our focus, and elevate
We what we do we update our formulas
We have certain formulas but we update em (oh right)
with the times, and everything y'know
And and so, ya know
The rhyme style is elevated
The style of beats is elevated
but it's still Guru and Premier
And it's always a message involved
Gang Starr - You Know My Steez
Keechma/next is the second iteration of the Keechma framework. In its scope, it's similar to Integrant - a data driven, state management framework for single page apps.
Keechma/next is built on top of experience of building ~50 (non trivial) apps in the span of 5 years, in agency setting. You could say that its features were forged in infernal flames of agency work. In these 5 years, we've built client facing web apps, B2C mobile apps, back offices and animation heavy landing pages. This allowed us to define features we need to build apps in a sustainable and agile fashion.
Original Keechma framework was based on Reagent, and it was relying on Reagent's reactive atom to keep UI synchronized with the application state. Keechma/next is self contained framework, and has no dependency on any UI frameworks. It can be easily integrated directly with React - through the hooks system. It features automatic state -> UI synchronization in a system inspired by the reactive atom.
Keechma/next's allows you to build apps that are explicit, deterministic and predictable. This is achieved with the architecture that is somewhat different from the other ClojureScript frameworks:
- No shared app-state atom
- Integration of state reads and state writes
- Automatic state synchronization that can be effectful
In Keechma/next your main building block is the controller. Controllers are composed in the app state graph, and they can depend on each other. UI can subscribe to the controller state. Each controller has it's own state, with read and write access, and each controller has read access to it's parent controllers'. Controllers can react to events. When the controller's state changes all descendant controllers are automatically synchronized. Each controller controls a key in the app-state map.
Controllers are implemented with multi methods which are dispatched on the controller keyword. Controller keywords must be derived from the :keechma/controller
keyword.
(derive :counter :keechma/controller)
(ns example.controller
(:require [keechma.next.core :as core]
[keechma.next.controller :as ctrl]))
(derive :counter :keechma/controller)
(defmethod ctrl/start :counter [_ _ _ _]
0)
(defmethod ctrl/handle :counter [{:keys [state*]} event _]
(case event
:inc (swap! state* inc)
nil))
(def app
{:keechma/controllers {:counter {:keechma.controller/params true}}})
(def app-instance (core/start! app))
(core/get-derived-state app-instance) ;; {:counter 0}
(core/dispatch app-instance :counter :inc)
(core/get-derived-state app-instance) ;; {:counter 1}
In this example, we've created a very simple counter controller:
- It's state is initialized with value
0
- Whenever the event
:inc
is dispatched to the:counter
controller it's state value is incremented
Keechma/next app is defined as a map, where we glue together all the controllers in the app. Reconciliation of the state is strict and synchronous.
Let's expand the example. In this case we'll add the :counter-2
controller whose value will be :counter
controller's value multiplied by 2.
(ns example.controller
(:require [keechma.next.core :as core]
[keechma.next.controller :as ctrl]))
(derive :counter :keechma/controller)
(defmethod ctrl/start :counter [_ _ _ _]
0)
(defmethod ctrl/handle :counter [{:keys [state*]} event _]
(case event
:inc (swap! state* inc)
nil))
(derive :counter-2 :keechma/controller)
(defmethod ctrl/derive-state :counter-2 [_ _ {:keys [counter]}]
(* 2 counter))
(def app
{:keechma/controllers {:counter {:keechma.controller/params true}
:counter-2 {:keechma.controller/params true
:keechma.controller/deps [:counter]}}})
(def app-instance (core/start! app))
(core/get-derived-state app-instance) ;; {:counter 0 :counter-2 0}
(core/dispatch app-instance :counter :inc)
(core/get-derived-state app-instance) ;; {:counter 1 :counter-2 2}
In this example, :counter-2
controller has no internal state. Instead it's state is derived from it's parents' state. ctrl/derive-state
method allows you to derive the public state based on the controller's internal state and the state of its parents.
Let's add another controller to the mix:
(ns example.controller
(:require [keechma.next.core :as core]
[keechma.next.controller :as ctrl]))
(derive :counter :keechma/controller)
(defmethod ctrl/start :counter [_ _ _ _]
0)
(defmethod ctrl/handle :counter [{:keys [state*]} event _]
(case event
:inc (swap! state* inc)
nil))
(derive :counter-2 :keechma/controller)
(defmethod ctrl/derive-state :counter-2 [_ _ {:keys [counter]}]
(* 2 counter))
(derive :counter-3 :keechma/controller)
(defmethod ctrl/start :counter-3 [_ _ _ _]
0)
(defmethod ctrl/handle :counter-3 [{:keys [state*]} event _]
(case event
:inc (swap! state* inc)
nil))
(defmethod ctrl/derive-state :counter-3 [_ state {:keys [counter counter-2]}]
(+ state counter counter-2))
(def app
{:keechma/controllers {:counter {:keechma.controller/params true}
:counter-2 {:keechma.controller/params true
:keechma.controller/deps [:counter]}
:counter-3 {:keechma.controller/params true
:keechma.controller/deps [:counter :counter-2]}}})
(def app-instance (core/start! app))
(core/get-derived-state app-instance) ;; {:counter 0 :counter-2 0 :counter-3 0}
(core/dispatch app-instance :counter :inc)
(core/get-derived-state app-instance) ;; {:counter 1 :counter-2 2 :counter-3 3}
(core/dispatch app-instance :counter-3 :inc)
(core/get-derived-state app-instance) ;; {:counter 1 :counter-2 2 :counter-3 4}
In this case :counter-3
has it's own internal state, and it's using it in combination with its parents' states to derive its public state.
If you look at the app map in the last two examples, you'll notice that controllers define its dependencies. This allows Keechma/next to know which controllers need to be informed of the state change. Although you're not able to access whole app-state atom, you can rely on this mechanism to ensure that your controllers have the correct data pushed to them when they need it.
There are no limitations on what can be stored in the controller state. In these examples, we had simple values stored, but you can store whatever you want - it's a normal Clojure atom. In fact, you'll probably have a few controllers that manage bigger chunks of state, and a bunch of controllers that will only manage smaller pieces.
There are no limitations to effects either. Event handlers can mutate state, perform AJAX requests or do whatever else is required to implement your app.
In all these examples, controllers' params were set to true
. Controller's params determine current state of the controller. It can be in :stopped
or :running
state (don't confuse it with controller's state atom). When params are set to a truthy value, or to a function that returns a truthy value, controllers will be started. Let's make another small example:
(ns example.controller
(:require [keechma.next.core :as core]
[keechma.next.controller :as ctrl]))
(derive :counter :keechma/controller)
(defmethod ctrl/start :counter [_ _ _ _]
0)
(defmethod ctrl/handle :counter [{:keys [state*]} event _]
(case event
:inc (swap! state* inc)
nil))
(derive :counter-2 :keechma/controller)
(defmethod ctrl/derive-state :counter-2 [_ _ {:keys [counter]}]
(* 2 counter))
(def app
{:keechma/controllers {:counter {:keechma.controller/params true}
:counter-2 {:keechma.controller/params (fn [deps] (even? (:counter deps)))
:keechma.controller/deps [:counter]}}})
(def app-instance (core/start! app))
(core/get-derived-state app-instance) ;; {:counter 0 :counter-2 0}
(core/dispatch app-instance :counter :inc)
(core/get-derived-state app-instance) ;; {:counter 1}
(core/dispatch app-instance :counter :inc)
(core/get-derived-state app-instance) ;; {:counter 2 :counter-2 4}
If you compare the :counter-2
controller implementation, it is identical to the previous examples. But, in this case we've passed a function as the controller's params. This function will be called whenever any parent state is changed. If that function returns true, controller will be started, otherwise stopped. You can notice that after second :inc
dispatch, :counter-2
key is not present in the derived state map. This is because the controller is not running at that moment (:counter
value is 1
, so params function returned false). If the params function returned truthy value on previous call, and returns a truthy value again, these values will be compared. If they are different controller will be restarted (stopped and started again). This behavior drastically simplifes your code, because you don't have to handle cases like these in your own code. Params function determines the life time of a controller.
Prev Params | Current Params | Prev == Current | Actions |
---|---|---|---|
falsy | falsy | - | Do nothing (controller is in :stopped state) |
truthy | falsy | - | Stop the current controller instance |
falsy | truthy | - | Start a new controller instance |
truthy | truthy | false | Stop the current controller instance and start a new one |
truthy | truthy | true | Dispatch :keechma.on/deps-change event to the running controller instance |
There also exists the ability to change a controller type between reconcilliation cycles which can also affect their lifecycle, for more information refer to this expanded table.
In examples so far, we've used only one controller variant - singleton. Singleton controllers are great when you have only one instance of controller running in the system. For instance, you could have a :router
controller, :jwt
or :current-user
controller. Another variant is identity controllers. Identity controllers use a composite key [:user :some-identity]
. They are useful when you want to mount some controller multiple times in the same app. For all other purposes they are identical to singleton controllers. Last variant is factory controller. Factory controllers are not mounted directly, instead they can dynamically produce configuration for controllers that should be mounted. This is useful when you have to dynamically mount controllers based on some data that is not available in development time. For instance you might have a Kanban board, and you want to mount a controller for each column. All these controllers will be of same type, but each will be its own instance. Factory controllers' configuration looks different from the singleton and identity controllers:
{:keechma/controllers
{:counter-1 {:keechma.controller/params true}
[:counter-2] {:keechma.controller.factory/produce
(fn [{:keys [counter-1]}]
(->> (range counter-1 (+ 2 counter-1))
(map (fn [i] [(inc i) {:keechma.controller/params 1}]))
(into {})))
:keechma.controller/deps [:counter-1]}}}
- Factory controllers' key is a single element vector. This is controller type.
- Instead of
:keechma.controller/params
attribute, they have:keechma.controller.factory/produce
attribute which should be a function.
Produce function should return a map where keys are ids and values are controller config maps. These will be merged with the parent map to produce final configuration. For instance if the produce function returned
{1 {:keechma.controller/params 1} 2 {:keechma.controller/params 1}}
Keechma/next would mount two controllers: [:counter-2 1]
and [:counter-2 2]
. Produce function is called at the same time when the params function would be called (when parents' state changes), and Keechma/next will make a diff between the current and the previous value, and determine which controllers have to be started, stopped and removed.
Controller variant is determined based on the configuration (controller key and configuration map), and you can have the same controller mounted as a singleton, identity or factory controller without any changes to the controller implementation.
Controllers have a defined lifecycle which is guaranteed by Keechma/next. Lifecycles reduce need for a lot of boilerplate code that deals with resource initialization and teardown. Each controller is implemented with a small number of multimethods (consult the documentation in controller.cljs file). Controllers should be "thin", they should only act as a glue code between the UI, various APIs and domain code. Controllers are unopinionated on purpose, you get full access to the state atom, so you can implement features in any way you need. Keechma/next will call appropriate methods when controller's state changes - you can use these methods to implement your logic.
ctrl/prep
method is called for each controller. This method should return a modified controller config map. This map will be used to initialize all instances of the controller (controllers are not initialized at this point yet)
ctrl/init
method is called. This method should return a modified controller config map which will be used for that controller instance. Use this method to initialize any resources (for instance start a go-loop or connect to a websocket)ctrl/start
method is called. This method should return initial controller state:keechma.on/start
event is dispatched to the controller. This will result with the call to thectrl/handle
methodctrl/derive-state
method is called to compute the derived state
This happens if the params function returned same truthy value like on previous invocation.
:keechma.on/deps-change
event is dispatched to the controller. This will result with the call to thectrl/handle
method. Payload is a map that contains only the keys that changed from the previous invocation.ctrl/derive-state
method is called to compute new derived state. If this state is different from the current value (cached in the app-state) all descendant controllers are reconciled
This happens if the controller was running and params function returned different value than on the previous invocation
:keechma.on/stop
event is dispatched to the controller. This will result with the call to thectrl/handle
methodctrl/stop
method is called. This method should return final state. This state will be passed to thectrl/start
function when the next instance is startedctrl/terminate
method is called. Use this method to teardown any resources initialized in thectrl/init
method
ctrl/handle
method is called with the event and payloadctrl/derive-state
method is called to compute new derived state. If this state is different from the current value (cached in the app-state) all descendant controllers are reconciled
* This is not 100% correct because there are some optimizations around when the ctrl/derive-state
method is called, but this method will be called if the controller's state changes
Lifecycle functions allow you to focus on your domain code and drastically reduce the number of events that are not domain related.
As your app grows, there will be features that are local. For instance, you might have a number of controllers that are related to the user profile app area. Keechma/next allows you to group these controllers into a subapp. Subapps have their own equivalent of params
- :keechma.app/should-run?
function. When this function returns a truthy value, Keechma/next will reconcile its child controllers. These controllers still have their own :keechma.controller/params
and :keechma.controller/deps
that determine when a controller should run. Use :keechma.app/should-run?
function to extract the shared params logic. Subapps can be nested so you can create app hierarchies that make sense for your app.
{:keechma/controllers {:user-role {:keechma.controller/params true }}
:keechma/apps
{:public {:keechma.app/should-run? (fn [{:keys [user-role]}] (= :guest user-role))
:keechma.app/deps [:user-role]
:keechma/controllers {:posts {:keechma.controller/params true
:keechma.controller/type :public-posts}}}
:user {:keechma.app/should-run? (fn [{:keys [user-role]}] (= :user user-role))
:keechma.app/deps [:user-role]
:keechma/controllers {:posts {:keechma.controller/params true
:keechma.controller/type :user-posts}}}}}
In this example we have three apps defined. The main app - with the :user-role
controller, the :public
app with the :posts
controller and the :user
app with the :posts
controller. You've probably noticed that both the :public
and the :user
apps have the same - :posts
controller defined. Also, each of these controllers explicitly sets the :keechma.controller/type
attribute - controllers are registered on the same key, but have a different type. This is possible because these apps are not able to run at the same time (their :keechma.app/should-run?
function prevents this). Keechma/next's reconciliation is synchronous so these apps will be started / stopped in the same cycle - UI will only notice the data change. This enables simpler controllers, where they can only implement feature set that makes sense for them. UI can stay dumb, and you don't have to have any conditional logic in the UI as long as the controllers react to same events. For instance, upvote event could cause the :users
:posts
controller to send the upvote request to the server, while :public
:users
controller could redirect the user to the registration page.
In future, we'll add support for the :keechma.app/load
function which will allow you to return app config map from a function with the support for Promises. This will enable code splitting for subapps.
Keechma/next's reconciliation is transacted. This means that any controllers that are dirtied as a result of some state change will be reconciled before next round of reconciliation starts. If any events are dispatched during the reconciliation, and these events cause a state change, controllers dirtied by this state change will be reconciled in the next round (run to completion semantics).
* This is not 100% correct because of some exceptions caused by the optimizations in the reconciliation code
State changes must be wrapped inside a transaction. Synchronous calls to the ctrl/handle
method are automatically wrapped inside a transaction, but there are cases where you might have an async call that will mutate the state as a result. In that case, wrap this callback in the ctrl/transact
manually.
(defmethod ctrl/handle :login [ctrl cmd payload]
(log-cmd! ctrl cmd payload)
(case cmd
:do-login (js/setTimeout #(ctrl/transact ctrl
(fn []
(ctrl/dispatch ctrl :token :update-token "TOKEN")
(ctrl/dispatch ctrl :current-user :update-user {:id 1 :username "retro"}))))
nil))
Keechma/next has no integration with the UI layer in the core library. But, Keechma/next has support for subscriptions which can be used to integrate your UI library of choice (these are 100-200 lines of code and can be generalized). Keechma/next provides core/subscribe
function that can be used to subscribe to the controller's state:
(ns example.controller
(:require [keechma.next.core :as core]
[keechma.next.controller :as ctrl])
(derive :counter :keechma/controller)
(defmethod ctrl/start :counter [_ _ _ _]
0)
(defmethod ctrl/handle :counter [{:keys [state*]} event _]
(case event
:inc (swap! state* inc)
nil))
(def app
{:keechma/controllers {:counter {:keechma.controller/params true}}})
(defn subscription-handler [state] (println state))
(def app-instance (core/start! app))
(def unsubscribe (core/subscribe app-instance :counter subscription-handler))
(core/dispatch app :counter :inc) ;; Subscription function will be called here synchronously
(unsubscribe) ;; Removes the subscription
Subscriptions are synchronously called after the reconciliation is done.
Why prevent access to the whole app-state? Why integrate state reads and writes?
In our experience as apps grow, implicit dependencies make them more and more rigid. You start by managing a piece of data in one place, but as the features are added, more and more places read and write in the same location. You can approach this by having more events flowing in the system, so each handler manages only it's own piece of state, but that doesn't solve the problem it just moves it into a different place. Instead of data dependencies, you now have implicit dependencies between events.
We wanted a way to always manage one piece of data in one place. In Keechma/next controllers manage data under a key in the map. This data is owned by that controller. When we've concluded that data ownership is a good idea, we needed a way to depend on other pieces of data. This is why the controllers are organized in a dependency graph. When controller's state changes all descendant controllers will be reconciled, and their derived state recomputed.
Controllers owning their data allows us to do a lot of repetitive, boilerplate work for you. For instance, if you have some async action that is mutating the state in its callback, and the controller is stopped during the async action, Keechma/next will prevent accidental state change. Each controller instance gets it's own atom, which is detached from the app when the controller is stopped.
Another benefit is reduction of events flowing in the system. If you have only events as your building block, it's easy to create events that result in more events that result in more events. This kind of implementation can become hard to manage (and to hold in your head while you're developing or debugging). Keechma/next ensures that controllers are started at the right time with right dependencies. You can focus on your domain logic instead on complex event chains. In our opinion, events should be used to communicate domain actions, and their number should be small.
Keechma/next removes the need for manual synchronization which can happen when you have separate state atoms. In our opinion this is best of both worlds - data ownership and automatic synchronization.
Keechma/next removes dependency on Reagent and React. It's a standalone library that can be integrated with any UI framework that might surface in the future. Reagent is a great library, and we've succesfuly pushed many applications into production with it, but React is becoming better, and a lot of features that were exclusive to Reagent are now possible with raw React. Keechma/next implements subset of reactive features that we need. We believe that this architecture can enable simple, understandable apps that have a very rich feature set. Integrant was a great inspiration, and in our experience with it, it became obvious how powerful it is to be able to have an overview of the app in one place. As mentioned previously, Keechma is developed in an agency context, and our constraints may not be 100% shared with you, but it enabled us to iterate quickly and to learn from our mistakes. Keechma/next is the result of thousands of hours we've spent working on single page and mobile apps in a very dynamic environment.
Depending on the size of your app(s), and the rhythm of your development, you might not see immediate need for the architecture provided by Keechma/next. But, Keechma/next shouldn't be judged by the code you write, instead think about all the code you won't have to write. Predictable and deterministic apps are not trivial to implement, and in our opinion Keechma/next provides everything you need to pull it of. Keechma/next is a contained library, without any global state leaking, so you can try it out without major commitment. If you do try it, feel free to reach out to us on #keechma channel on the Clojurians slack.
Very Big Things sponsors Keechma development, and is offering design and development services. If you ever need help with your app, feel free to reach out to us, and work with the Keechma core team on your next app.
Keechma/next is inspired by many libraries in Clojure ecosystem (and beyond). Here's a list of some of the codebases we've researched and studied during the development:
- Fork this repo
- run
clojure -A:deps:test
from terminal - Fix something and make sure to visit http://localhost:8021/ see tests are green.
- Push your changes and do a PR.
Copyright © 2020 Mihael Konjevic, Tibor Kranjcec
Distributed under the MIT License.