myfaces-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Werner Punz <werner.p...@gmail.com>
Subject proposed: myfaces jsf.js refactorings
Date Mon, 10 May 2010 08:09:50 GMT
Hello everyone, I have started somewhat of a cleanup regarding the 
jsf.js codebase which will be done in the following weeks.

First of all after fixing the bugs (which I will commit today), I came 
to the conclusion that our utils classes are somewhat in need of a fixup 
and generally some structures can be solved better.

What I want to do is following:

a) Split the utils classes into:
_Lang (former _LangUtils)
_Dom (former mostly utils)
_Communications (former AjaxUtils)

move some of the methods around and add a bunch of new ones regarding 
dom traversal (mostly iterators utilizing filter closures, which can be 
browser optimized, due to most browsers already have dom level 2 
filtering iterators in place).

Also we will add a new class _Core or _Runtime wich will initialize the 
basic runtime stuff, like namespacing, browser detection, configuration 
binding (which our system is more or less coupled with loosely)
  etc...

Note, most of this stuff already is in the codebase, but grown into the 
wrong place of the _LangUtils, or _Utils classes.

There also will be changes to the namespacing itself
currently we have the usualy if cascades to check for namespaces

I want to move that into something like _Core.reserveNamespace, so that 
the code will look like:

if(_Core.reserverNamespace("my.name.space")) {
	my.name.space = function() {};	
}

This is very close to what dojo.provide("my.name.space") does.


Also what I want to do to ease future development is to add prototype 
based inheritance which follows very closely the dojo pattern.

which means:

_Lang.extends("my.new.MyClass", parentClass, {
	/*note we have to rename the constructor name a little bit due 	to 
javascript limitations in prototype inheritance*/
	constructor_:function (arg1) {
		//we have to do it that way due to javascript limitations
		this._callSuper('constructor', arg1);
		....
	},
	myNewFunc: function(arg1, arg2) {
		//basically the same as this._callSuper("myNewFunc", arg2)
		this._inherited(arg2);
		....	
	}
});


Now for now I also would leave following option open if you want to do 
it in a more ide friendly way:
if(_Core.reserverNamespace("my.new.MyClass")) {
	my.new.MyClass = _Lang.extends(function (arg1) {
		this._callSuper('constructor', arg1);
	}, parent);
	my.new.MyClass.prototype.myNewFunc = function(arg1,arg2) {
		this._callSuper('myNewFunc', arg2);
		...	
	}
}

The second option is easier to grasp for IDEs, but the this._inherited() 
call will not be possible. But I want to support both options because 
the first option is much more dense and does not need a separate 
namespace checking, while the second one is more friendly to ides.
As for now inheritance is not used, we use delegation over loose 
coupling but for future extensions it would make sense to have it in.
Especially since a load of stuff from the transport layer can be reused.

Anyway:

I do not want to go for a full blown framework here, to keep the 
codebase as tight as possible. I just want to add
the needed stuff to make future extensions easier.

(Which will probably different transports like iframes, and websockets, 
queue control and timeout (which we already have in our codebase))

Problem is by using an existing js library (and that was the major 
reason why I did not do it firsthand) we will add around 20-100kbyte of 
extra code which is mostly pointless. Although I am a huge fan of dojo 
(less of jquery due to its structure), if we integrated dojo under our 
own namespace, we would need the core (50kbyte + xhr + probably 1-2 
other modules) which would add 50-70 kbyte in compressed form, 80% of 
stuff which we will never touch)

I however opted for a loose binding of the layers so that people can rip 
out our implementations of the layers and use whatever they like in a 
configuration manner. (it is possible to replace the entire xhr 
transport by your own implementation without overwriting the entire 
xhrCore classes for instance so that dojo or other transports 
theoretically could be plugged in)



Werner





Mime
View raw message