I am currently maintaining a large number of JS files and the dependency issue is growing over my head. Right now I have each function in a separate file and I manually maintain a database to work out the dependencies between functions.
This I would like to automate. For instance if I have the function f
Array.prototype.f = function() {};
which is referenced in another function g
MyObject.g = function() {
var a = new Array();
a.f();
};
I want to be able to detect that g is referencing f.
How do I go about this? Where do I start? Do I need to actually write a compiler or can I tweak Spidermonkey for instance? Did anyone else already do this?
Any pointers to get me started is very much appreciated
Thanks
Dok
Whilst you could theoretically write a static analysis tool that detected use of globals defined in other files, such as use of MyObject, you couldn't realistically track usage of prototype extension methods.
JavaScript is a dynamically-typed language so there's no practical way for any tool to know that a, if passed out of the g function, is an Array, and so if f() is called on it there's a dependency. It only gets determined what variables hold what types at run-time, so to find out you'd need an interpreter and you've made yourself a Turing-complete problem.
Not to mention the other dynamic aspects of JavaScript that completely defy static analysis, such as fetching properties by square bracket notation, the dreaded eval, or strings in timeouts or event handler attributes.
I think it's a bit of a non-starter really. You're probably better of tracking dependencies manually, but simplifying it by grouping related functions into modules which will be your basic unit of dependency tracking. OK, you'll pull in a few more functions that you technically need, but hopefully not too much.
It's also a good idea to namespace each module, so it's very clear where each call is going, making it easy to keep the dependencies in control manually (eg. by a // uses: ThisModule, ThatModule comment at the top).
Since extensions of the built-in prototypes are trickier to keep track of, keep them down to a bare minimum. Extending eg. Array to include the ECMAScript Fifth Edition methods (like indexOf) on browsers that don't already have them is a good thing to do as a basic fixup that all scripts will use. Adding completely new arbitrary functionality to existing prototypes is questionable.
Have you tried using a dependency manager like RequireJS or LabJS? I noticed no one's mentioned them in this thread.
From http://requirejs.org/docs/start.html:
Inside of main.js, you can use require() to load any other scripts you
need to run:
require(["helper/util"], function(util) {
//This function is called when scripts/helper/util.js is loaded.
//If util.js calls define(), then this function is not fired until
//util's dependencies have loaded, and the util argument will hold
//the module value for "helper/util".
});
You can nest those dependencies as well, so helper/util can require some other files within itself.
As #bobince already suggested, doing static analysis on a JavaScript program is a close to impossible problem to crack. Google Closure compiler does it to some extent but then it also relies on external help from JSDoc comments.
I had a similar problem of finding the order in which JS files should be concatenated in a previous project, and since there were loads of JS files, manually updating the inclusion order seemed too tedious. Instead, I stuck with certain conventions of what constitutes a dependency for my purposes, and based upon that and using simple regexp :) I was able to generated the correct inclusion order.
The solution used a topological sort algorithm to generate a dependency graph which then listed the files in the order in which they should be included to satisfy all dependencies. Since each file was basically a pseudo-class using MooTools syntax, there were only 3 ways dependencies could be created for my situation.
When a class Extended some other class.
When a class Implemented some other class.
When a class instantiated an object of some other class using the new keyword.
It was a simple, and definitely a broken solution for general purpose usage but it served me well. If you're interested in the solution, you can see the code here - it's in Ruby.
If your dependencies are more complex, then perhaps you could manually list the dependencies in each JS file itself using comments and some homegrown syntax such as:
// requires: Array
// requires: view/TabPanel
// requires: view/TabBar
Then read each JS file, parse out the requires comments, and construct a dependency graph which will give you the inclusion order you need.
It would be nice to have a tool that can automatically detect those dependencies for you and choose how they are loaded. The best solutions today are a bit cruder though. I created a dependency manager for my particular needs that I want to add to the list (Pyramid Dependency Manager). It has some key features which solve some unique use cases.
Handles other files (including inserting html for views...yes, you can separate your views during development)
Combines the files for you in javascript when you are ready for release (no need to install external tools)
Has a generic include for all html pages. You only have to update one file when a dependency gets added, removed, renamed, etc
Some sample code to show how it works during development.
File: dependencyLoader.js
//Set up file dependencies
Pyramid.newDependency({
name: 'standard',
files: [
'standardResources/jquery.1.6.1.min.js'
]
});
Pyramid.newDependency({
name:'lookAndFeel',
files: [
'styles.css',
'customStyles.css',
'applyStyles.js'
]
});
Pyramid.newDependency({
name:'main',
files: [
'createNamespace.js',
'views/buttonView.view', //contains just html code for a jquery.tmpl template
'models/person.js',
'init.js'
],
dependencies: ['standard','lookAndFeel']
});
Html Files
<head>
<script src="standardResources/pyramid-1.0.1.js"></script>
<script src="dependencyLoader.js"></script>
<script type="text/javascript">
Pyramid.load('main');
</script>
</head>
It does require you to maintain a single file to manage dependencies. I am thinking about creating a program that can automatically generate the loader file for you based on includes in the header but since it handles many different types of dependencies, maintaining them in one file might actually be better.
JSAnalyse uses static code analysis to detect dependencies between javascript files:
http://jsanalyse.codeplex.com/
It also allows you to define the allowed dependencies and to ensure it during the build, for instance. Of course, it cannot detect all dependencies because javascript is dynamic interpretet language which is not type-safe, like already mentioned. But it at least makes you aware of your javascript dependency graph and helps you to keep it under control.
I have written a tool to do something like this: http://github.com/damonsmith/js-class-loader
It's most useful if you have a java webapp and you structure your JS code in the java style. If you do that, it can detect all of your code dependencies and bundle them up, with support for both runtime and parse-time dependencies.
Related
I am building a Sails.js application using sails 1.2.3, node 10.15. I want to include a javascript module in my api/helpers/* directory, without sails automatically using it to try to create a helper. I.e. I have javascript objects that use helpers and are used in a helper, but are not helpers themselves; as in this image, where the module 'rules' is imported into the create-rule helper and the objects exported by this module are used within the helper.
By default, sails tries to load each file in the helpers/* directory as a helper, and throws if the underlying implementation does not match that of a valid helper:
ImplementationError: Failed to load helper `create-rule/rules/foo/index` into a Callable! Sorry, could not interpret "index" because its underlying implementation has a problem:
------------------------------------------------------
• Missing the `fn` property.
------------------------------------------------------
Hoping someone can help out! Let me know if more info is needed. Thanks in advance!
I don't quite understand what you are trying to do. In my humble opinion I would grab all object constructors and placed them as a single file in api/services. That will make it automatically available in all controllers. I would not allow my object's methods to use helpers by them selves (I even think you can't, at least easily). Then when you need a helper to use your object, just pass it as parameter. Anyway, again, in my humble opinion; you are structuring your code to fit all inside /helpers and that will make it extremely hard to develop. Let assume you manage to make it work all inside /helpers, only you without exception, will be able to understand what it does or how it works. Doesn't seem as a good idea.
Uncompressed jQuery file: http://code.jquery.com/jquery-2.0.3.js
jQuery Source code: https://github.com/jquery/jquery/blob/master/src/core.js
What are they doing to make it seem like the final output is not using Require.js under the hood? Require.js examples tells you to insert the entire library into your code to make it work standalone as a single file.
Almond.js, a smaller version of Require.js also tell you to insert itself into your code to have a standalone javascript file.
When minified, I don't care for extra bloat, it's only a few extra killobytes (for almond.js), but unminified is barely readable. I have to scroll all the way down, past almond.js code to see my application logic.
Question
How can I make my code to be similar to jQuery, in which the final output does not look like a Frankenweenie?
Short answer:
You have to create your own custom build procedure.
Long answer
jQuery's build procedure works only because jQuery defines its modules according to a pattern that allows a convert function to transform the source into a distributed file that does not use define. If anyone wants to replicate what jQuery does, there's no shortcut: 1) the modules have to be designed according to a pattern which will allow stripping out the define calls, and 2) you have to have a custom conversion function. That's what jQuery does. The entire logic that combines the jQuery modules into one file is in build/tasks/build.js.
This file defines a custom configuration that it passes to r.js. The important option are:
out which is set to "dist/jquery.js". This is the single
file produced by the optimization.
wrap.startFile which is set to "src/intro.js". This file
will be prepended to dist/jquery.js.
wrap.endFile which is set to "src/outro.js". This file will
be appended to dist/jquery.js.
onBuildWrite which is set to convert. This is a custom function.
The convert function is called every time r.js wants to output a module into the final output file. The output of that function is what r.js writes to the final file. It does the following:
If a module is from the var/ directory, the module will be
transformed as follows. Let's take the case of
src/var/toString.js:
define([
"./class2type"
], function( class2type ) {
return class2type.toString;
});
It will become:
var toString = class2type.toString;
Otherwise, the define(...) call is replace with the contents of the callback passed to define, the final return statement is stripped and any assignments to exports are stripped.
I've omitted details that do not specifically pertain to your question.
You can use a tool called AMDClean by gfranko https://www.npmjs.org/package/amdclean
It's much simpler than what jQuery is doing and you can set it up quickly.
All you need to do is to create a very abstract module (the one that you want to expose to global scope) and include all your sub modules in it.
Another alternative that I've recently been using is browserify. You can export/import your modules the NodeJS way and use them in any browser. You need to compile them before using it. It also has gulp and grunt plugins for setting up a workflow. For better explanations read the documentations on browserify.org.
Here's what I'm looking for:
I want to use the wonderful features of SIMPLE mode minification while disabling just one specific feature (disable local function inline).
UPDATE: The answer is NO, it's not possible given my setup. But for me there is a workaround given I am using Grails.
As #Chad has explained below, "This violates core assumptions of the compiler". See my UPDATE3 below for more info.
IN QUESTION FORM:
I'm using CompilationLevel.SIMPLE_OPTIMIZATIONS which does everything I want, except that it's inlining my local functions.
Is there any way around this? For example, is there a setting I can place in my JS files to tell Google Closure not to inline my local functions?
It would be cool to have some directives at the top of my javascript file such as:
// This is a JS comment...
// google.closure.compiler = [inlineLocalFunctions: false]
I'm developing a Grails app and using the Grails asset-pipeline plugin, which uses Google Closure Compiler (hereafter, Compiler). The plugin supports the different minification levels that Compiler supports via the Grails config grails.assets.minifyOptions. This allows for 'SIMPLE', 'ADVANCED', 'WHITESPACE_ONLY'.
AssetCompiler.groovy (asset-pipeline plugin) calls ClosureCompilerProcessor.process()
That eventually assigns SIMPLE_OPTIMIZATIONS on the CompilerOptions object. And by doing so, CompilerOptions.inlineLocalFunctions = true as a byproduct (this is hard coded behavior in Compiler). If I were to use WHITESPACE_ONLY the result would be inlineLocalFunctions=false.
So by using Asset Pipeline's 'SIMPLE' setting, local functions are being inlined and that is causing me trouble. Example: ExtJS ext-all-debug.js which uses lots of local functions.
SO post Is it possible to make Google Closure compiler *not* inline certain functions? provides some help. I can use its window['dontBlowMeAway'] = dontBlowMeAway trick to keep my functions from inlining. However I have LOTS of functions and I'm not about to manually do this for each one; nor would I want to write a script to do it for me. Creating a JS model and trying to identity local functions doesn't sound safe, fun nor fast.
The previous SO post directs the reader to https://developers.google.com/closure/compiler/docs/api-tutorial3#removal, where the window['bla'] trick is explained, and it works.
Wow thanks for reading this long.
Help? :-)
UPDATE1:
Okay. While spending all the effort in writing this question, I may have a trick that could work. Grails uses Groovy. Groovy makes method call interception easy using its MetaClass API.
I'm going to try intercepting the call to:
com.google.javascript.jscomp.Compiler.compile(
List<T1> externs, List<T2> inputs, CompilerOptions options)
My intercepting method will look like:
options.inlineLocalFunctions=false
// Then delegate call to the real compile() method
It's bed time so I'll have to try this later. Even so, it would be nice to solve this without a hack.
UPDATE2:
The response in a similar post (Is it possible to make Google Closure compiler *not* inline certain functions?) doesn't resolve my problem because of the large quantity of functions I need inlined. I've already explained this point.
Take the ExtJS file I cited above as an example of why the above similar SO post doesn't resolve my problem. Look at the raw code for ext-all-debug.js. Find the byAttribute() function. Then keep looking for the string "byAttribute" and you'll see that it is part of strings that are being defined. I am not familiar with this code, but I'm supposing that these string-based values of byAttribute are later being passed to JS's eval() function for execution. Compiler does not alter these values of byAttribute when it's part of a string. Once function byAttribute is inlined, attempts to call the function is no longer possible.
UPDATE3: I attempted two strategies to resolve this problem and both proved unsuccessful. However, I successfully implemented a workaround. My failed attempts:
Use Groovy method interception (Meta Object Protocol, aka MOP) to intercept com.google.javascript.jscomp.Compiler.compile().
Fork the closure-compiler.jar (make my own custom copy) and modify com.google.javascript.jscomp.applySafeCompilationOptions() by setting options.setInlineFunctions(Reach.NONE); instead of LOCAL.
Method interception doesn't work because Compiler.compile() is a Java class which is invoked by a Groovy class marked as #CompileStatic. That means Groovy's MOP is not used when process() calls Google's Compiler.compile(). Even ClosureCompilerProcessor.translateMinifyOptions() (Groovy code) can't be intercepted because the class is #CompileStatic. The only method that can be intercepted is ClosureCompilerProcessor.process().
Forking Google's closure-compiler.jar was my last ugly resort. But just like #Chad said below, simply inserting options.setInlineFunctions(Reach.NONE) in the right place didn't resurrect my inline JS functions names. I tried toggling other options such as setRemoveDeadCode=false to no avail. I realized what Chad said was right. I would end up flipping settings around and probably destroying how the minification works.
My solution: I pre-compressed ext-all-debug.js with UglifyJS and added them to my project. I could have named the files ext-all-debug.min.js to do it more cleanly but I didn't. Below are the settings I placed in my Grails Config.groovy:
grails.assets.minifyOptions = [
optimizationLevel: 'SIMPLE' // WHITESPACE_ONLY, SIMPLE or ADVANCED
]
grails.assets.minifyOptions.excludes = [
'**ext-all-debug.js',
'**ext-theme-neptune.js'
]
Done. Problem solved.
Keywords: minify, minification, uglify, UglifyJS, UglifyJS2
In this case, you would either need to make a custom build of the compiler or use the Java API.
However - disabling inlining is not enough to make this safe. Renaming and dead code elimination will also cause problems. This violates core assumptions of the compiler. This local function is ONLY referenced from within strings.
This code is only safe for the WHITESPACE_ONLY mode of the compiler.
Use the function constructor
var fnc = new Function("param1", "param2", "alert(param1+param2);");
Closure will leave the String literals alone.
See https://developer.mozilla.org/de/docs/Web/JavaScript/Reference/Global_Objects/Function
We get used to include separately some javascript files with using standard <script src="file.js"></script>, but... I really don't like such a way of including.
As for the developer with the C#/C++ background, many features for me are representing as a mess.
I really don't like the way including the js files, but I understand, that it's the most regular practice.
With such kind of thoughts, I suppose to try different possible ways.
As I thought, there are two possible ways, which are able to implement the including process:
1). Create dynamically tag in imaginary function - include(), which I can define in my prototype ( class imitation in js ).
But... It's nearly the same way as the standard including, but just with some dynamic logic stored in prototypes, which may dynamically load some needed modules of web application.
And that's why I dislike it, but it's an option and I must add this to the list.
2). Dynamic load of js file via AJAX/WebSockets/SPDY, that you have wanted or even hanlding the parts of the loaded files with the partion function execution (needed some logic or implementing some design patters) with the following execution via eval() function and preparing it in the new defined function include( args ); to make the js scenario looks like:
include( "models.js" );
include( "dbController.js" );
...
Yes... I know, that eval() is an evil in JavaScript world, but I don't know the name of the different function, which allows me the same functionallity.
Also... It may be too slow compared to original way of including and I repeat even vulnerable.
These are the two ways I can imagine, maybe you will suggest a better solution or persuade me from some bad thoughts.
Require.js is a JavaScript modulisation tool that does exactly what you want. You define modules in their own file like so:
define('moduleName', [ dependencies ], function () {
// module code
});
And load them asynchronously:
require('moduleName');
Or as dependencies to other modules:
define([ 'moduleName' ], function (myModule) {
// other module
});
It also has an optimisation tool which is used to compile your modules into one JavaScript file ready for production.
I was wondering if anyone had experience using require.js with the Adobe CQ5 platform. I'm writing a Chaplin.js(backbone-based) single page app that will be integrated into the rest of CQ5-based site we're working on. Chaplin requires the use of a module system like AMD/Common.js and I want to make sure my compiled javascript file will usable within CQ5's clientlibs. Is it as simple as adding require.js as a dependency in clientlibs prior to loading in my javascript modules? Someone's insight who has experience in doing this would be greatly appreciated.
I've implemented this as a solution of organize in a better way all the modules such as:
//public/js/modules/myModule.js
define('myModule',[ /* dependencies */] ,function( /* stuff here */ ))
and in the components such:
<% //components/component.jsp %>
<div>
<!-- stuff here -->
</div>
componentJS:
//components/component/clientlibs/js/component.js
require(['modules/myModule']);
and finally I've configured require (config.js) and I've stored the JSs modules in a new different design folder. Located the compiled JS after close </body> to guarantee the JS is always located at the bottom after the HTML.
<!-- page/body.jsp -->
...
<cq:includeClientLib js="specialclientlibs.footer"/>
</body>
solving with this the issue of have "ready" all the content before the JS is executed. I've had some problems to resolve with this async stuff managed for the clienlibs compilation tool, in production the problem was different, however, in development, the order in what CQ compiles the JS has produced me some lacks in terms of order of the JS. The problem really was a little bit more complex than the explanation because the number of JS was really big and the team too, but in simple terms it was the best way I've discovered so far..
The Idea
I think you can compile your Chaplin.js with one of the AMDShims to make it self contained, wraps every dependencies it needs inside a function scope, expose an entry point as global variable (which is a bad practise, but CQ do it all the time...) and then use the compiled.js inside a normal clientlib:
AMD Shims
https://github.com/jrburke/requirejs/wiki/AMD-API-Shims
Example
Here is an example of how we make the one of our libs self-contained:
https://github.com/normanzb/chuanr/blob/master/gruntfile.js
Basically, in source code level the lib "require"s the other modules just as usual. However after compiled, the generated chuanr.js file contains everything its needs, even wrapped a piece of lightweight AMD compatible implementation.
check out compiled file here: https://github.com/normanzb/chuanr/blob/master/Chuanr.js
and the source code: https://github.com/normanzb/chuanr/tree/master/src
Alternative
Alternatively rather than compile every lib you are using to be independent/self-contained, what we do in our project is simply use below amdshim instead of the real require.js. so on cq component level you can call into require() function as usual:
require(['foo/bar'], function(){});
The amd shim will not send the http request to the module immediately, instead it will wait until someone else loads the module.
and then at the bottom of the page, we collect all the dependencies, send the requirements to server side handler (scriptmananger) for dynamic merging (by internally calling into r.js):
var paths = [];
for (var path in amdShim.waiting){
paths.push(path);
}
document.write('/scriptmananger?' + paths.join(','));
The amdShim we are using: https://github.com/normanzb/amdshim/tree/exp/defer