pico-client
v0.8.2
Published
pico client
Downloads
57
Readme
pico-client
pico app framework
Example
An obligatory Todo App to demotrate the framework. The example source code can be found here
Browser Compatibility
- Android Browser: compatible
- Chrome/Chromium: compatible
- Safari: compatible
- Safari ios: compatible
- Firefox: compatible
- IE 11: compatible
- IE 10 and below: not compatible
Features
Data Driven
A pico client is data driven, meaning that the application structure is not defined by script but by data. The data file is in json file format. The data file format is closely assemble to lisp syntax. for example
["name", "type", "content", "meta data"]
and the example of a complete project file
[
["component1", "view", [
["moduleA", "view", []],
["moduleB", "view", []]
]],
["component2", "view", []]
]
Lazy Load At Component Level
var
dep1 = require('vendorA/dep1'),
dep2 = require('vendorB/dep2')
this.load=function(){
dep1.doSeomthing()
dep2.doSomething()
}
Decentralized Configuration
pico archieved decentralized configuration by making spec files joinable during runtime.
A spec file with a entry point (a special js file) can be compiled to a bundle file, the build script generate single .js file based on the configuration file and the entry point
to add external spec to current spec, use dynamic require
load the external spec and spawn
to render it
require('/path/to/external/spec', (err, spec) => {
if (err) return console.error(err)
this.spawn(specMgr.create('id-here', 'view', [], spec))
})
Complete example at example/bundle
Create bundle
npx pclient-build
Manage environment config with dependency injection
move environment dependant config to a separate config file, inject the env config to main config in the main js file
var specMgr= require('p/specMgr')
var View= require('p/View')
var project = require('cfg/xin.json')
var env = require('cfg/dev.json')
var main
return function(){
specMgr.load(null, null, project, function(err, spec){
if (err) return console.error(err)
main = new View
main.spawnBySpec(spec, null, env)
})
}
Code Spliting
Code spliting is done at the project file level, each bundle can hae it own project file and project can be load lazily during runtime
Support circular dependencies
pico-client supported asynchronous module loading, therefore no circular dependencies issue
Syntax similar to commonjs and amd, easy to pickup
syntax of pico-client is heavely burrow from commonjs and amd to reduce learning curve
Support recursive view component
for recursive view component such tree view
// tree-node.js
return {
deps: {
'tree_node': 'view', // self referencing
},
create(deps, params){
this.spawn(deps.tree_node) // or, this.spawn(deps.tree_node, null, [deps.tree_node.splice(0, 3)]]
}
}
Caveat
sub-module readiness
pico modules are loaded in series orderly, what it means is module declared at top always load first and pico ensure it is loaded completely before loading the next module. One caveat is submodule may not ready during event call. for example project configuration as follow
[
["modA","view",[
["modA-subA","view",[]]
]],
["modB","view",[]]
]
if modB event call modA immediately in create function, modA-subA may not ready when modA receive the event. modA should listen to moduleAdded emitted by modA-subA before using any functionality from modA-subA
External editable static property
ModuleA
var obj={
a:1,
print:function(){
console.log(obj.a)
}
}
return obj
if ModuleA.print method was call in ModuleB
var modA=require('ModuleA')
modA.print() // 1
what if ModuleA.a was changed in ModuleB?
var modA=require('ModuleA')
modA.a='hello'
modA.print() // 1
print result is still 1, that's because 'hello' is set on modA placeholder, to make ModuleA.a an editable property, make this changes to ModuleA
var modA=require('ModuleA')
var obj={
a:1,
print:function(){
console.log(modA.a)
}
}
return obj