babel-plugin-transform-decorators-legacy

Babel Legacy Decorator plugin

This is a plugin for Babel 6 that is meant to replicate the old decorator behavior from Babel 5 in order to allow people to more easily transition to Babel 6 without needing to be blocked on updates to the decorator proposal or for Babel to re-implement it.

Babel >= 7.x

This plugin is specifically for Babel 6.x. If you're using Babel 7, this plugin is not for you. Babel 7's @babel/plugin-proposal-decorators officially supports the same logic that this plugin has, but integrates better with Babel 7's other plugins. You can enable this with

{
  "plugins": [
    ["@babel/plugin-proposal-decorators", { "legacy": true }],
  ]
}

in your Babel configuration. Note that legacy: true is specifically needed if you want to get the same behavior as transform-decorators-legacy because there are newer versions of the decorator specification coming out, and they do not behave the same way as this plugin does.

Installation & Usage

$ npm install --save-dev babel-plugin-transform-decorators-legacy

Add the following line to your .babelrc file:

{
    "plugins": ["transform-decorators-legacy"]
}

NOTE: Order of Plugins Matters!

If you are including your plugins manually and using transform-class-properties, make sure that transform-decorators-legacy comes before transform-class-properties.

/// WRONG
 
"plugins": [
  "transform-class-properties",
  "transform-decorators-legacy"
]
 
// RIGHT
 
"plugins": [
  "transform-decorators-legacy",
  "transform-class-properties"
]

Why "legacy"?

Decorators are still only a relatively new proposal, and they are (at least currently) still in flux. Many people have started to use them in their original form, where each decorator is essentially a function of the form

function(target, property, descriptor){}

This form is very likely to change moving forward, and Babel 6 did not wish to support the older form when it was known that it would change in the future. As such, I created this plugin to help people transition to Babel 6 without requiring them to drop their decorators or requiring them to wait for the new proposal update and then update all their code.

Best Effort

This plugin is a best effort to be compatible with Babel 5's transpiler output, but there are a few things that were difficult to reproduce, and a few things that were simply incorrect in Babel 5 with respect to the decorators proposal.

Two main things to mention as differences, though not things you are likely to encounter:

1. Decorators expressions are evaluated top to bottom, and executed bottom to top. e.g.

   function dec(id){
       console.log('evaluated', id);
       return (target, property, descriptor) => console.log('executed', id);
   }
   
   class Example {
       @dec(1)
       @dec(2)
       method(){}
   }
   

   In Babel 5, this would output:

   evaluated 2
   evaluated 1
   executed 2
   executed 1
   

   With this plugin, it will result in:

   evaluated 1
   evaluated 2
   executed 2
   executed 1
   

   which is what the spec dictates as the correct behavior and was incorrect in Babel 5.

2. Static class property initializers are evaluated once up front.

   If you decorate a static class property, you will get a descriptor with an initializer property. However whereas with Babel 5 this could be re-executed multiple times with potentially differing results, decorators-legacy will precompute the value and return an initializer that will return that value. e.g.

   function dec(target, prop, descriptor){
       let {initializer} = descriptor;
       delete descriptor.initializer;
       delete descriptor.writable;
   
       descriptor.get = function(){
           return initializer.call(this);
       };
   }
   
   var i = 0;
   
   class Example {
       @dec
       static prop = i++;
   }
   

   In Babel 5, every access to prop would increment i. In Babel 6, the very first value of i will be cached for future initializer calls.

   The spec is a little vague around how initializers work for repeat calls, and I'd consider calling an initializer multiple times to be a mistake in general, so hopefully this will not cause anyone trouble.

License

MIT (c) 2015