A truly modular frontend framework
To understand what I mean by truly modular just read the source
var document = require('global/document');
var hg = require('mercury');
var h = require('mercury').h;
function App() {
// define application state.
var state = hg.struct({
value: hg.value(0),
handles: hg.value(null)
});
// define handles that react to events.
state.handles.set(hg.handles({
clicks: incrementCounter
}, state));
return state;
}
// declare view rendering.
App.render = function render(state) {
return h('div.counter', [
'The state ', h('code', 'clickCount'),
' has value: ' + state.value + '.', h('input.button', {
type: 'button',
value: 'Click me!',
'ev-click': hg.event(state.handles.clicks)
})
]);
};
function incrementCounter(state) {
state.value.set(state.value() + 1);
}
hg.app(document.body, App(), App.render);
The following examples demonstrate how you can mix & match mercury with other frameworks. This is possible because mercury is fundamentally modular.
Disclaimer: The following are neither "good" nor "bad" ideas. Your milage may vary on using these ideas
mercury is similar to react, however it's larger in scope,
it is better compared against om or
quiescent
- mercury leverages
virtual-domwhich uses an immutable vdom structure - mercury comes with
observ-structwhich uses immutable data for your state atom - mercury is truly modular, you can trivially swap out subsets of it for other modules
- mercury source code itself is maintainable, the modules it uses are all small, well tested and well documented. you should not be afraid to use mercury in production as it's easy to maintain & fix.
- mercury encourages zero dom manipulation in your application code. As far as your application is concerned elements do not exist. This means you don't need to reference DOM elements when rendering or when handling events
- mercury is compact, it's 11kb min.gzip.js, that's smaller than backbone.
- mercury strongly encourages FRP techniques and discourages local mutable state.
- mercury is highly performant, it's faster then React / Om / ember+htmlbars in multiple benchmarks
TodoMVC benchmark
animation benchmark TodoMVC benchmark source - mercury comes with FP features like time-travel / easy undo out of the box.
- mercury is lean, it's an weekend's read at 2.5kloc. (virtual-dom is 1.1kloc, an evening's read.) compared to react which is almost 20kloc (a month's read)
mercury is a small glue layer that composes a set of modules
that solves a subset of the frontend problem.
If mercury is not ideal for your needs, you should check out
the individual modules and see if you can re-use something.
Alternatively if the default set of modules in mercury doesn't
work for you, you can just require other modules. It's possible
to for example, swap out vtree with
react or swap out observ-struct
with backbone
There are three pieces to mercury, Input (Controller), State (Model) and Render (View).
In a normal mercury app you define your top level Input which is a finite list of events.
You then define your top level state "atom". Generally you want a large fat state object for your entire application. We then wire all the events in Input up to some updating logic, i.e. every time an event occurs in Input we do some logic and then update the State.
Finally we define our Rendering logic as a single function that takes the entire state of our application and returns a virtual DOM representation of our UI. Every time the state changes we just call render and then update the DOM.
You may also need Output for your application, if we want to have some other side effect other then updating the UI, like sending a HTTP POST or writing to a websocket or persisting to indexedDB. Then we generally listen to changes in the state and have our side effect. Note that Render is just a specific subset of the Output of your application.
For the view layer mercury uses a set of modules that come together and make it easy to work with a Virtual DOM.
In mercury the view is just a function that takes your
application state and returns a virtual DOM representation.
This makes writing your view really easy, you just write it
top to bottom.
mercury then uses the following modules to make it really
performant to use the virtual DOM to update the real DOM.
vtree is the module that contains the data
structures for the virtual DOM. These are the primitive
objects and values that the rendering functions in a
mercury app will return.
vtree also contains the diffing algorithm used in
mercury. Mercury uses a diffing algorithm on a virtual DOM
to compute a minimal set of VPatch records that it can apply to the DOM.
vdom is the module that contains the create and
patch algorithm for turning the vtree data structures
into real DOM objects.
vdom/create-element is used to turn a
virtual DOM into a real DOM. this is used for the initial
rendering of your app.
vdom/patch is used to apply a series of VPatch
records to a real DOM element.
You can also use vdom and
min-document together on the server to
generate HTML strings. min-document is a
minimal fake DOM for use on the server, you can pass
vdom any document you want. In this case
min-document contains the logic to convert
its fake DOM into a HTML string.
virtual-hyperscript is a module that
makes it easier to create VTree nodes. It basically exports
a h() function that creates a DSL similar to jade
(just more brackets ;)).
virtual-hyperscript allows you to write
your views in an expressive manner.
vdom-thunk is a module that increases the
performance of building applications with a virtual DOM
system. One of the important parts of using a virtual DOM
and functional programming in general is to make extensive
use of caching.
You can use vdom-thunk to effectively memoize a
function that returns a virtual DOM node. This means if you
call it twice with the same arguments it will not re-evaluate
the function.
This basically means you only have to render that which has changed instead of rendering the entire virtual tree of your application.
It should be noted that vdom-thunk assumes
arguments are immutable and thus does an O(1) === check
to see whether the arguments has changed. This will only
work if your state is immutable. Thankfully,
observ-struct is immutable
main-loop is another optimization module for a
virtual DOM system. Normally you would re-create the virtual
tree every time your state changes. This is not optimum,
with main-loop you will only update your
virtual tree at most once per request animation frame.
main-loop basically gives you batching of your
virtual DOM changes, which means if you change your model
multiple times it will be rendered once asynchronously on
the next request animation frame.
In mercury we use immutable data structure primitives to
represent our model. Using immutable data structures allows
you to use the vdom-thunk optimization.
mercury uses an observable state representation so that you
can be notified of any changes.
Generally applications built with mercury will have a single top level state "atom". i.e. there is one large state object for your application and child components do not have local or hidden state. However we can directly embed the state of a child component in our top level state "atom" to achieve composition.
observ is the data structure used for observable
data. It allows you to create a value for which you can
listen for changes.
observ also comes with higher order functions
like observ/computed that can be used to
create new dependent observables. Generally these computed
observables cannot be directly mutated but instead change
when they data they rely on changes.
observ is basically an implementation of the
Signal type that is normally used in FRP.
observ-struct is an observable that contains an
object with a fixed number of keys. Generally the key-value
pairs in observ-struct are themselves
observables. You can change the value of any key in an
observ-struct and the top level object
will also change to be a new object with that key changed.
observ-struct uses shallow extension to ensure
that every time the struct changes you get a fresh immutable
object.
observ-array is an observable that contains
an array of observables. It's generally recommended that this a
heterogeneous array. You can change the value of any item in
the array and the top level array will also change to be a
new array.
observ-array uses shallow extension to ensure
that every time the array changes (an item changes or an
item is added or removed) you get a fresh immutable array.
observ-array has the benefit of being able
to add or remove items from the array, where as
observ-struct has a fixed number
of keys and you cannot add more keys to an
observ-struct
In mercury we model all the inputs to our application
explicitly. We define an input object that contains a bunch of
geval Event instances.
Somewhere else in our application we listen to the Input and run some logic and update our state when an event happens.
geval is our data structure for Events. it gives us
a way of listening to events and a way of publishing them.
Most of the time you will either create a computed Event that emits events based on some raw source, like winddow scroll events or a websocket. Or you can create a mutable Event which you pass to the UI renderer so it can emit events for dynamically created UI components.
geval is basically an implementation of the
Event type that is normally used in FRP.
dom-delegator is an event delegator that
allows you to seperate your event listeners from your
event emitters. It sets up global event listeners and
allow you to embed event handlers on your virtual DOM
elements without having to manage adding or removing
actual event listeners.
value-event allows you to create event
handlers that you can embed in a virtual DOM.
These event handlers work with both the native DOM and
dom-delegator.
value-event contains a set of higher order
functions that allows you to write to a value to a
geval Event when some user interaction occurs.
Using the higher order functions defined in
value-event (change, submit, etc. )
allows you to not have to write any DOM event handling
code in your application. value-event
takes care of all the reading from the DOM.
See the FAQ document
WIP. In lieu of documentation please see examples :(
npm install mercury
If you want to develop on mercury you can clone the code
git clone [email protected]:Raynos/mercury
cd mercury
npm install
npm testnpm testruns the testsnpm run jshintwill run jshint on the codenpm run discwill open discify (if globally installed)npm run buildwill build the html assets for gh-pagesnpm run exampleswill start a HTTP server that shows examplesnpm run distwill create a distributed version of mercury
A lot of the philosophy and design of mercury is inspired by
the following:
reactfor documenting and explaining the concept of a virtual DOM and its diffing algorithmomfor explaining the concept and benefits of immutable state and time travel.elmfor explaining the concept of FRP and having a reference implementation of FRP in JavaScript. I wrote a pre-cursor tomercurythat was literally a re-implementation ofelmin javascript (graphics)reflexfor demonstrating the techniques used to implement dynamic inputs.
- Raynos
- Matt-Esch
- neonstalwart
- parshap
- nrw