Differences of JavaScript contexts
NOTE: some content in this wiki applies only to 0.12 and earlier versions. For official documentation on 0.13 and later, see http://docs.nwjs.io
Different windows of a node-webkit-based application have different JavaScript contexts, i.e. each window has its own global object and its own set of global constructors (such as Array
or Object
).
That's common practice among web browsers. It's a good thing because, for example:
-
when an object's prototype is replaced or augmented by some library (such as Prototype) or a simpler script, the analogous objects in other windows are unaffected nevertheless;
-
when a programmer makes a mistake (such as missing
new
before a poorly written constructor) and the bug affects (pollutes) the global scope, it still cannot affect larger areas (several windows); -
malicious applications cannot access confidential data structures in other windows.
Node modules in node-webkit run in their own shared Node context. (Shared by default; however, you may explicitly add 'new-instance': true
to the options of Window.open
if you need your new window to have a separate Node.js context.)
If the require()
method (of Node.js modules API) is used, then the required module runs in the Node's context. (When you call the require()
function or a function from some required module, the JS engine enters the Node's context and leaves it after the function returns.)
If HTML <script>
element (or jQuery's $.getScript()
, or any other similar method) is used in some window, then the script runs in the context of that window.
If the module is given as the value of the "node-main"
property of the application's manifest file, then the module runs in the Node's context but later has access to the window
object. (See the “node-main” article for details.)
Scripts running in the Node's context may use __dirname
variable to read the path of their file's directory.
The Node.js global
object is the global object in the Node's context. Any WebKit window's window
object is not the global object and even is not implicitly available in the Node's context (the special case of node-main is the only exception), i.e. you have to (explicitly) pass the window
object to your module's function if you need to access it.
That also means you cannot rely on alert()
(which is actually window.alert()
) for debugging. You may, however, use console.log()
; its output (and the output of other similar methods such as console.warn()
and console.error()
) is redirected to WebKit's console. You may see it in your “Developer Tools” window (on its “Console” tab).
You cannot use require('nw.gui')
(to access the node-webkit's GUI API) from the Node's context, because there's no GUI outside of a window.
Some other browser features (such as Worker
and WebSocket
interfaces) are also unavailable in the Node's context.
Relative paths in webkit context are resolved according to path of main HTML file (like all browsers do). Relative paths in node modules are resolved according to path of that module (like node.js always do). Just remember in which context you are.
For example if we have file /myApp/main.html
:
<html>
<head>
<!-- will be resolved according to this html file path -->
<script src="components/myComponent.js"></script>
</head>
<body>
<script>
// will be resolved according to this html file path
var hello = require('./libs/myLib');
// __dirname is not defined in webkit context, this is only node.js thing
console.log(__dirname); // undefined
</script>
</body>
</html>
In file /myApp/components/myComponent.js
we can do this:
// we are still in webkit context, so paths are still resolved according to main.html
var something = require('./util.js'); // will look for file /myApp/util.js NOT for /myApp/components/util.js
// __dirname still not defined
console.log(__dirname); // undefined
In file /myApp/libs/myLib.js
we can do this:
// here we are in node.js context, so paths are resolved according to this file
var something = require('./otherUtil.js'); // will look for file /myApp/libs/otherUtil.js
// __dirname is defined
console.log(__dirname); // '/myApp/libs'
While differences of contexts are generally beneficial, sometimes they may constitute a problem in your (or some other person's) code, and a need for a workaround arises.
The most common cause for such problems is the behaviour of the instanceof
operator in JavaScript. As you may see in MDN, the operation someValue instanceof someConstructor
tests whether an object has in its prototype chain the prototype
property of the given constructor. However, if someValue
is passed from a different JavaScript context, then it has its own line of ancestor objects, and the someValue instanceof someConstructor
check fails inevitably.
For example, a simple check someValue instanceof Array
cannot determine if a variable's value is an array's if it's passed from another context (see “Determining with absolute accuracy whether or not a JavaScript object is an array” for details).
The same problem arises when the .constructor
property is checked directly (for example, when someValue.constructor === Array
is used instead of someValue instanceof Array
).
The following constructors are children of the context-dependent global object, and thus their instances are affected:
-
Standard object types:
Array
,Boolean
,Date
,Function
,Number
,Object
,RegExp
,String
. -
Typed array types:
ArrayBuffer
,DataView
,Float32Array
,Float64Array
,Int16Array
,Int32Array
,Int8Array
,Uint16Array
,Uint32Array
,Uint8Array
. -
Error types:
Error
,EvalError
,RangeError
,ReferenceError
,SyntaxError
,TypeError
,URIError
.
There are several ways to work around this problem.
The easiest way to prevent context-related problems is to avoid using instanceof
when a value may come from another JavaScript context. For example, you may use Array.isArray
method to check whether a value is an array, and that method works reliably across contexts.
However, if such a convenient alternate method is not readily available, or when you face a problem in someone other's (not your own) code and patching that would need a hassle, then another workaround is necessary.
When you foresee passing a value to some other context, you may providently use a constructor from that context in order to construct your value. The value then would easily pass any instanceof
checks in that context.
For example, the well-known async
module used (in its code dated 2013-05-20; it was fixed later) numerous .constructor
checks (in lines 472, 505, 545, 675, 752) and thus failed whenever it encountered an array from another context. For example, if you run the following code from a node-webkit's window context,
require('async').waterfall([
function(callback){
console.log('1.');
callback(null, 'one', 'two');
},
function(arg1, arg2, callback){
console.log('2.');
callback(null, 'three');
},
function(arg1, callback){
console.log('3.');
callback(null, 'done');
}
], function (err, result) {
console.log('Fin.');
if( err ) throw err;
console.log(result);
});
it throws the Error: First argument to waterfall must be an array of functions
(erroneously thinking it's not an array).
Using the nwglobal
module, you may access the Node's context's Array
constructor and rewrite the above code:
require('async').waterfall( require('nwglobal').Array(
function(callback){
console.log('1.');
callback(null, 'one', 'two');
},
function(arg1, arg2, callback){
console.log('2.');
callback(null, 'three');
},
function(arg1, callback){
console.log('3.');
callback(null, 'done');
}
), function (err, result) {
console.log('Fin.');
if( err ) throw err;
console.log(result);
});
It makes the async
module happy.
However, in some cases you cannot (or won't) use the constructor directly to create your value. (For example, as you may see in MDN, using the Function
constructor is less efficient than declaring a function, and it also does not create a closure.) In such cases another workaround is necessary.
The non-standard (but widely implemented) __proto__
property of an object can be used (as you may see in MDN) to change the object's internal “Prototype” property (initially containing the prototype of its constructor).
When you foresee passing a value to some other context, you may providently replace the value's __proto__
property with a constructor from that context. The value then would easily pass any instanceof
checks in that context.
For example, the well-known async
module used (in its code dated 2013-05-20; it was fixed later) an instanceof Function
check (on line 428) and thus it failed whenever it encountered a function from another context. For example, if you run the following code from a node-webkit's window context,
var getData = function (callback) {
setTimeout(function(){
console.log('1.1: got data');
callback();
}, 300);
}
var makeFolder = function (callback) {
setTimeout(function(){
console.log('1.1: made folder');
callback();
}, 200);
}
var writeFile = function(callback) {
setTimeout(function(){
console.log('1.1: wrote file');
callback(null, 'myfile');
}, 300);
}
var emailFiles = function(callback, results) {
console.log('1.1: emailed file: '+results.writeFile);
callback(null, results.writeFile);
}
require('async').auto({
getData:getData ,
makeFolder:makeFolder,
writeFile: ['getData', 'makeFolder',writeFile],
emailFiles: ['writeFile',emailFiles]
}, function(err, results) {
console.log('1.1: err: '+ err);
console.log('1.1: results: '+ results);
});
it throws the has no method slice
error (erroneously thinking that the given value is not a function and thus it has to be an array, and then attempting to slice that “array”).
Using the nwglobal
module, you may access the Node's context's Function
constructor and rewrite the above code:
var getData = function (callback) {
setTimeout(function(){
console.log('1.1: got data');
callback();
}, 300);
}
getData.__proto__ = require('nwglobal').Function;
var makeFolder= function (callback) {
setTimeout(function(){
console.log('1.1: made folder');
callback();
}, 200);
}
makeFolder.__proto__ = require('nwglobal').Function;
var writeFile= function(callback) {
setTimeout(function(){
console.log('1.1: wrote file');
callback(null, 'myfile');
}, 300);
}
writeFile.__proto__ = require('nwglobal').Function;
var emailFiles= function(callback, results) {
console.log('1.1: emailed file: '+results.writeFile);
callback(null, results.writeFile);
}
emailFiles.__proto__ = require('nwglobal').Function;
require('async').auto({
getData:getData ,
makeFolder:makeFolder,
writeFile: ['getData', 'makeFolder',writeFile],
emailFiles: ['writeFile',emailFiles]
}, function(err, results) {
console.log('1.1: err: '+ err);
console.log('1.1: results: '+ results);
});
It makes the async
module happy.
Switching between WebKit's and Node's contexts takes some time.
In most cases this delay does not constitute a serious problem, but if you are in a WebKit's context and you are simply deferring some function's execution (in some hot spot where the same function is likely to be deferred hundreds of times; for example, when you draw some 500 objects giving the WebKit its chance to redraw the window after each object's appearance rather than forcing the user to stare on a blank screen mindlessly), then using Node's setImmediate
(exported from some Node.js module) can actually become less “immediate” than you would be happy to experience.
To work around this problem it's usually enough to define (in WebKit's context) and use David Baron's setZeroTimeout
function instead of Node's setImmediate
.
// Only add setZeroTimeout to the window object, and hide
// everything else in a closure.
(function() {
var timeouts = [];
var messageName = "zero-timeout-message";
// Like setTimeout, but only takes a function argument.
// There's no time argument (always 0) and no function's arguments
// (you have to use a closure if such arguments are necessary).
function setZeroTimeout(fn) {
timeouts.push(fn);
window.postMessage(messageName, "*");
}
function handleMessage(event) {
if (event.source == window && event.data == messageName) {
event.stopPropagation();
if (timeouts.length > 0) {
var fn = timeouts.shift();
fn();
}
}
}
window.addEventListener("message", handleMessage, true);
// Add the one thing we want added to the window object.
window.setZeroTimeout = setZeroTimeout;
})();
Using standard window.setTimeout(yourFunction, 0)
is less efficient because HTML5 standard defines its minimal timeout as 4 milliseconds even if 0
is given (i.e. even if you use setTimeout
only 250 times, you already get a whole second of extra delay in your application).