2

I am having difficulty understanding how exceptions are handled when code is fetched dynamically via AJAX and executed via eval. With clientside javascript, it is rather simple, if I have a piece of code such as this

var j = 'some string';
j.propA.x++;

this will raise an exception because propA, which is of type undefined does not have an x. Furthermore, the exception raised is very easy to understand.

Now lets put the above code in a text file, lets call it test.js, and store it on the server. Now lets load it dynamically with Ajax. I am using the following code to load it dynamically

dojo.xhrGet({
  url: 'load.php',
  handleAs: "javascript",
  content : {
    fileName : 'test.js'
  },
  load: function(returnValue) {
    /*Do Something*/
  },
  error: function(errorMessage) {
    /*Report Error*/
  }
});

Here is a very basic php script for loading the file and returning it as javascript code

<?php
  $fileName = $_GET['fileName'];
  $handle = fopen($fileName , 'r');
  $script = fread($handle, filesize($fileName));
  fclose($handle);
  echo $script;
?>

In the above dojo.xhrGet call, the error property can be set to a function to display the error message, here is an example of some of the many ways this can be done.

error: function(errorMessage) {
  console.error(errorMessage);
  console.error(errorMessage.arguments);
  console.error(errorMessage.message);
  console.error(errorMessage.stack);
  console.error(errorMessage.type);
}

Below is an example of the output. Although this output is for a different problem, it highlights how incomprehensible it is:

Cannot read property 'x' of undefined
TypeError: Cannot read property 'x' of undefined
    at eval at <anonymous> (http://o.aolcdn.com/dojo/1.6/dojo/dojo.xd.js:14:3088)
    at Object.load (http://192.168.1.8/easel.js:166:6)
    at http://o.aolcdn.com/dojo/1.6/dojo/dojo.xd.js:14:89998
    at _144 (http://o.aolcdn.com/dojo/1.6/dojo/dojo.xd.js:14:36518)
    at _142 (http://o.aolcdn.com/dojo/1.6/dojo/dojo.xd.js:14:36328)
    at [object Object].<anonymous> (http://o.aolcdn.com/dojo/1.6/dojo/dojo.xd.js:14:36994)
    at _144 (http://o.aolcdn.com/dojo/1.6/dojo/dojo.xd.js:14:36780)
    at _142 (http://o.aolcdn.com/dojo/1.6/dojo/dojo.xd.js:14:36328)
    at [object Object].<anonymous> (http://o.aolcdn.com/dojo/1.6/dojo/dojo.xd.js:14:36994)
    at Object.resHandle (http://o.aolcdn.com/dojo/1.6/dojo/dojo.xd.js:14:92730)
non_object_property_load

I am assuming dojo.xd.js:14 is the line where the eval statement is.

If one knows what they are looking for, the above might suffice. However, is there an easier, or at least a more productive way to deal with exceptions arising in eval?


Here is a somewhat similar question.


Phikin provided a good solution to this problem below so I gave him the bounty. Using his solution, I got an output which looked something like this (I cut it down a bit)

    ReferenceError in JS Code detected: (url: module.require.php?module=MainMenu.Bg_S)
easel.js:211Error Message: ReferenceError: apple is not defined
easel.js:213(function(){
return function(args){
dojo.require("Shape");

Module.assert('MainMenu_V');

/**
* The rectangular background of the Main View
* @property MainMenuBg_S
* @type Shape
**/
new Shape({
    /**
    * Unique descriptive name used when later accessing this shape via '$$()'
    * @param name
    * @type String
    **/
    name : 'MainMenu.Bg_S' , 
    /**
    * Left side of this rectangle
    * @param x
    * @type Number
    **/             
    x : $$('MainMenu_V').x , 

    /**
    * Top of this rectangle
    * @param y
    * @type Number
    **/ 
    y : $$('MainMenu_V').y , 

    /**
    * Width of this rectangle
    * @param w
    * @type Number
    **/ 
    w : $$('MainMenu_V').w , 

    /**
    * Height of this rectangle
    * @param h
    * @type Number
    **/ 
    h : $$('MainMenu_V').h , 

    /**
    * Type of this Shape
    * @param h
    * @type Number
    **/ 
    type : shapeType.RECTANGLE , 

    /**
    * Generate function which contains all the graphics instructions, as well as the contexts
    * to preload and initialize. This is currently under development. Backgrounds should NEVER
    * have mouse events associated with them as a redraw of a background implies a redraw of 
    * every single displayObject infront of the background.
    * @param generate
    * @type method
    **/   
    generate : function (){
        var x = this.x << 0 , y = this.y << 0 , h = this.h << 0 , w = this.w << 0 , a = this.a;

        this.graphics(contextID.LEAVE).lf([hsl(180,100,60,0.9),hsl(180,100,20,0.75)],[0,1],0,h/2,w,h/2).dr(x,y,w,h).ef();
        this.graphics(contextID.ENTER).lf([hsl(135,100,40,0.9),hsl(135,100,20,0.75)],[0,1],0,h/2,w,h/2).dr(x,y,w,h).ef();
        this.graphics(contextID.CLICK).lf([hsl(90,100,40,0.9),hsl(90,50,20,0.75)],[0,1],0,h/2,w,h/2).dr(x,y,w,h).ef();
        this.graphics(contextID.RCLICK).lf([hsl(90,110,40,0.9),hsl(80,60,20,0.45)],[0,1],0,h/2,w,h/2).dr(x,y,w,h).ef();
        this.graphics(contextID.DBLCLICK).lf([hsl(45,100,40,0.9),hsl(45,100,20,0.75)],[0,1],0,h/2,w,h/2).dr(x,y,w,h).ef();
        this.graphics(contextID.DBLRCLICK).lf([hsl(10,100,40,0.9),hsl(10,100,20,0.75)],[0,1],0,h/2,w,h/2).dr(x,y,w,h).ef();
        this.graphics(contextID.LPRESS).lf([hsl(110,25,40,0.9),hsl(110,25,20,0.75)],[0,1],0,h/2,w,h/2).dr(x,y,w,h).ef();
        this.graphics(contextID.RPRESS).lf([hsl(110,50,40,0.9),hsl(110,50,20,0.75)],[0,1],0,h/2,w,h/2).dr(x,y,w,h).ef();
        this.graphics(contextID.SCROLL).lf([hsl(110,50,40,0.9),hsl(110,50,20,0.75)],[0,1],0,h/2,w,h/2).dr(x,y,w,h).ef();

        if (debugFlags.BOUNDINGBOX()){
            this.graphics(contextID.ENTER).ss(2).s(rgba(0,255,0,a)).dr(this.boundingBox.softBounds.L +4<<0, this.boundingBox.softBounds.T +4<<0, this.boundingBox.softBounds.w-8<<0 , this.boundingBox.softBounds.h-8<<0).es();
            this.graphics(contextID.ENTER).ss(2).s(rgba(255,0,0,a)).dr(this.boundingBox.bounds.L +4<<0, this.boundingBox.bounds.T +4<<0, this.boundingBox.bounds.w-8<<0 , this.boundingBox.bounds.h-8<<0).es();
            this.graphics(contextID.ENTER).f(rgba(0,0,255,a)).dc(this.boundingBox.points[0].x+4 , this.boundingBox.points[0].y+4 , 4).ef();
            this.graphics(contextID.ENTER).f(rgba(0,0,255,a)).dc(this.boundingBox.points[1].x-8 , this.boundingBox.points[1].y+4 , 4).ef();
            this.graphics(contextID.ENTER).f(rgba(0,0,255,a)).dc(this.boundingBox.points[2].x-8 , this.boundingBox.points[2].y-8 , 4).ef();
            this.graphics(contextID.ENTER).f(rgba(0,0,255,a)).dc(this.boundingBox.points[3].x+4 , this.boundingBox.points[3].y-8 , 4).ef();
        }
    },

    /**
    * Arguments to pass to the mouse initialization function. These will get mixed in (via
    * dojo.mixin) to the mouse object. To increase performance, the signalOrderIn has been set to
    * NOHIT. This will limit the number of redraws (remember background redraws are extremely
    * expensive as they require redrawing everything in the container). The signalOrderOut is 
    * then set to BLOCK to prvent anything behind the background from receiving mouse signals
    * (this is actually unecessary as the only think behind the background is, and always should
    * be, the container, which itself has signalOrderIn and signalOrderOut set to NOHIT and BLOCK
    * respectively).
    * @param mouse
    * @type Object
    **/ 
    mouse : {
        _signalOrderIN : signalFlags.NOHIT ,
        _signalOrderOUT : signalFlags.BLOCK
    } ,

    /** 
    * All views are initially loaded via Ajax. Generally, views do not have any preconditions, beyond
    * that the stage be present. They can, however, and generally do, have modules they require. These
    * are called after this view has been created and loaded (load() function call). They are called
    * in the order of the sub arrays. In the example below:
    * [[A , B , C , D , E , F , G]]
    * The 7 modules are requested in that order, but, due to Ajax, they can be loaded in any order. 
    * In the below example, on the other hand:
    * [[A] , [B , C , D , E , F , G]]
    * Modules B-G depend on module A, therefore, module A is ordered to be loaded first. 
    * @property providedModules
    * @type Array[Array[String]]
    * @protected
    **/
    providedModules : [[]] ,

    /** 
    * Carries out all the initializations when loading the module
    * @method load
    * @protected
    **/
    load : function (){
                0/apple;
        $$('MainMenu_V').addChild(this);
    } ,

    /** 
    * Carries out all memory deallocation when leaving the module (generally only necessary if modules
    * were loaded but not added to stage as in the case with cached bitmaps)
    * @method leave
    * @protected
    **/
    leave : function (){
    }
});
$$('MainMenu.Bg_S')._code="dojo.require(\"Shape\");...";
};
}());

easel.js:217Error triggered by: function (_2bd){return err.call(args,_2bd,_2b7);}
easel.js:220XHR Object:
easel.js:221
Object
args: Object
handleAs: "javascript"
query: null
url: "module.require.php?module=MainMenu.Bg_S"
xhr: XMLHttpRequest
__proto__: Object
easel.js:222Error Object:
easel.js:223
ReferenceError
arguments: Array[1]
message: "—"
stack: "—"
type: "not_defined"
__proto__: Error
dojo.xd.js:14
ReferenceError
arguments: Array[1]
message: "—"
stack: "—"
type: "not_defined"
__proto__: Error
dojo.xd.js:14
ReferenceError
arguments: Array[1]
message: "—"
stack: "—"
type: "not_defined"
__proto__: Error

The only thing it's missing, that I need, is the ability to indicate what line the problem occurred.

Community
  • 1
  • 1
puk
  • 16,318
  • 29
  • 119
  • 199

2 Answers2

2

Here is a snipped that detectes non-network related errors from an xhr-get request and outputs some information about it in the console.

There is an extra isEvalError() function that goes through all eval-error types... which I am not really proud of. A nicer way could be by getting the parent object of the errorMessage sub-classes. I think you can ditch isEvalError() generally, because there shouldn´t be any other error possible in this block.

function isEvalError(errorMessage){
    return errorMessage.name == "RangeError" ||
    errorMessage.name == "ReferenceError" ||
    errorMessage.name == "SyntaxError" ||
    errorMessage.name == "URIError" ||
    errorMessage.name == "TypeError";
}

var foo = dojo.xhrGet({
    url: 'stacko.js',
    handleAs: "javascript",
    load: function(returnValue) {
        console.log("load: "+returnValue);
    },
    error: function(errorMessage,ioargs) {

        //request worked fine, this must be a non-network related error
        if(ioargs.xhr.readyState == 4 && ioargs.xhr.status == 200) {


            if(isEvalError(errorMessage)){
                //show eval-error, url request & the JS code that causes the exception
                //eval-error types: RangeError,ReferenceError,SyntaxError, URIError, TypeError
                console.error(errorMessage.name+" in JS Code detected: (url: "+ioargs.url+")")
                console.error("Error Message: "+ errorMessage);

                console.error(ioargs.xhr.responseText);
            }

            //a little reflection - if u want to know who triggered this error 
            //(although in this case the output is not very helpful )
            console.error("Error triggered by: "+arguments.callee.caller.toString());

            //last but not least log the error & the xhr-request object for more information
            console.error("XHR Object:");
            console.error(ioargs);
            console.error("Error Object:");
            console.error(errorMessage);


        }


    }
});
  • OK So I coded it and lets see how it goes...Surprisingly `0/0;` does not generate an exception. `0/apple;` does however. And the output is surprisingly informative! Thanks. I'll give you the bounty. But there is still one thing missing. It doesn't list the number where the error occurs. I will elaborate in the original question – puk Oct 26 '11 at 22:53
  • Thank you! Perhaps this helps to get the linenumber: http://stackoverflow.com/questions/3488994/javascript-eval-exception-line-number I will try it this weekend – Philipp Kinschel Oct 27 '11 at 09:02
1

It really depends on what you mean "productive way to deal with exceptions". If all you need to do is view the exception contents, a simple

console.log(errorMessage);

will allow you to effortlessly inspect the error object in decent browser like Chrome or Firefox (with Firebug). (Instead of forcing you to do a ton of console.log statements)

An annoying thing about Dojo exceptions inside asynchronous code is that they are always caught and handled so most browser debuggers ignore them. A notable exception to this rule is Chrome, where you can tell the debugger to pause on all exceptions.


BTW: I don't see how Javascript exceptions and Dojo have anything to do with PHP in this case, since they occur on the client side and there is nothing the server can do about them. Also, what the hell are you doing by sending Javascript code in the AJAX? Most of the time a client does a request it will be for data, in plain text, JSON or XML.

hugomg
  • 68,213
  • 24
  • 160
  • 246
  • I don't know the OP is saying that JS is being returned by the AJAX request, just that the handler will call an `error:` handler. My own thought is to use `trigger_error()` in PHP to shape and send a JSON `error_object` back to the caller and first check (in the client-side JS) if the JSON is this kind of response. – Jared Farrish Oct 23 '11 at 21:49
  • @Jared: I still don't see where PHP comes into this. The error callback in dojo.xhrGet will be called if any (javascript) exception is thrown in the process of acquiring the request, either from executing the code from a handleAs:javascript case or from an exception explicitely thrown by Dojo due to a timeout or a bad HTTP response code. At no point after the exception is thrown does to client stop to tell that to the server (unless you explicitely do so yourself) – hugomg Oct 23 '11 at 21:59
  • If the request is looking for a resource that is served by PHP, and the result that Javascript is accessing is PHP, wouldn't the way that the server (using PHP) responded be meaningful? (Although, error should "trigger" off the http response sent back by the server, which could be manipulated.) – Jared Farrish Oct 23 '11 at 22:02
  • The error callback will be called depending on the return code: http://stackoverflow.com/questions/5721949/what-is-considered-and-ajax-request-error-in-dojo – hugomg Oct 23 '11 at 22:05
  • Right, which can be [controlled by PHP](http://stackoverflow.com/questions/4797274/how-to-send-a-status-code-in-php-without-maintaining-an-array-of-status-names). – Jared Farrish Oct 23 '11 at 22:07
  • @missingno 'Also, what the hell are you doing by sending Javascript code in the AJAX?' I am not sending, but receiving, and I am remotely fetching code. Imagine there are 1,000,000,000 scripts. Instead of loading ALL of them, the user selects one, and it is fetched dynamically. Am I wrong to load this as `handleAs: "javascript"` – puk Oct 23 '11 at 22:45
  • Normally one would use the Dojo module system (dojo.provide, dojo.require, etc...) to do script loading and cacheing. Anyway, it is hard to reach a solid conclusion in this case since there are so many different reasons to load scripts at runtime and each one of them is better served by a different solution. – hugomg Oct 23 '11 at 22:49
  • Yes this question is poorly worded, I posted it a long time ago, I have since come to realize that this question has more to do with eval. I will edit the question somewhat. – puk Oct 23 '11 at 22:49
  • @missingno `productive way to deal with exceptions` would in the above example point out that the problem `Cannot read property 'x' of undefined` but ALSO point out it has to do with `j.propA.x++;` so I don't have to go and change every line that has `x` in it to figure out where the problem is. – puk Oct 23 '11 at 23:00