fix(closure): patchOnProperty with exact eventNames as possible

[JiaLiPassion]

related to #722.

We should try to patch onProperties with exact eventNames as possible.

in this PR, I separate the eventNames into several categories.

1. globalEventHandlerEventNames, eventNames in GlobalEventHandlers
2. documentEventNames, eventNames in Document
3. window, eventNames in Window
4. element, eventNames in Element
5. webgl, eventNames in WebGlContextEvent
6. form, eventNames in FormElement
7. details, eventNames in DetailsElement
8. XMLHttpRequest, eventNames in XMLHttpRequest
9. IDBIndex, eventNames in IDBIndex
10. WebSocket, eventNames in WebSocket

So we can patch each target with exact eventNames.

For example, we patch HTMLElement with
globalEventHandlers + element + webgl + form + details,
and we patch Window with
globalEventHandlers + window.

[JiaLiPassion]

Need to double check the event names and the target.

[JiaLiPassion]

@mhevery , just as #722, we may need to patch exact event names as possible instead of use patchOnProperties(XXX.Prototype, null), please review whether the idea is OK or not, I will continue to check the event names.

[mhevery]

Could we have a fixed set rather than performing this complex operation on each startup?

[JiaLiPassion]

@mhevery, yes, in my first commit, my goal is to make a fixed set, but it seems sometimes patch non exist property may cause errors, I will do more research and describe it more clear here.

I have modified the code to patch fixed set of event names.
And there is a potential risk (the risk also exists in the previous code), in some version of Browser, some property doesn't exists. for example,

in IE9, onprogress don't exist in XMLHttpRequest's prototype, so the code will
return false in following code without zone.js patch.

'onprogress' in (new XMLHttpRequest());

after zone.js patch in this PR, the code above will return true.

In previous code, we patch XMLHttpRequest with patchOnProperties(XMLHttpRequest.prototype, null), so it will only patch the event names which exists in XMLHttpRequest.prototype, that's why I added the getEventNames logic to make sure we only patch the properties exist in target.

[mhevery]

I don't think so.

The way I imagine this to work is that we have a list of all properties, and than we only patch those properties, but only if the property already exist. But we don't look for extra ones.

[JiaLiPassion]

@mhevery , got it, thanks! I also think we should do like that, I will modify it and do some tests.

I have modified the code, patchProperty will only patch the property which exist in obj, please review.

[JiaLiPassion]

should patch window object in IE correctly.
This has been fixed with passing Object.getPrototypeOf(window) to patchProperty method.

[mhevery]

The fact that we are getting all of the properties of the window, does not seem right. Why can't we have a fixed list here as well?

[JiaLiPassion]

@mhevery , the reason is in IE/Edge, the onProp don't exist in window but in WindowPrototype.

The code below describe the issue.

const desc = Object.getOwnPropertyDescriptor(window, 'onload');
// in IE
console.log(desc); // will be undefined.
// in chrome/firefox
console.log(desc); // will be {configurable:true,get/set....}

So in new patchProperty method, we only patch the prop which exist in the target object, so I pass the prototype of window for the existency check.
https://github.com/JiaLiPassion/zone.js/blob/a787bf8ca6922465a6456c4182d7c61f46fb0f4f/lib/common/utils.ts#L73

export function patchProperty(obj: any, prop: string, prototype?: any) {
  let desc = Object.getOwnPropertyDescriptor(obj, prop);
  if (!desc && prototype) {
    // when patch window object, use prototype to check prop exist or not
    const prototypeDesc = Object.getOwnPropertyDescriptor(prototype, prop);
    if (prototypeDesc) {
      desc = {enumerable: true, configurable: true};
    }
  }
  // if the descriptor not exists or is not configurable
  // just return
  if (!desc || !desc.configurable) {
    return;
  }
....