Natural Preference for Minification


1.23.0-esprima3 • Public • Published


NPM version Build Status Coverage Status Dependency Status

AssetGraph is an extensible, node.js-based framework for manipulating and optimizing web pages and web applications. It's the core of the third generation of the production builder tool we are using at for some of our web apps.

If you just want to get started with the basics, read Peter Müller - Getting started with Assetgraph.

If you are looking for a prepackaged build system take a look at Assetgraph-builder.

Check out the slides from a presentation of AssetGraph held at the Öresund JavaScript Meetup on June 16th, 2011.

The complete AssetGraph-based build system mentioned in the slides can be found here.

Assets and relations

All web build tools, even those that target very specific problems, have to get a bunch of boring stuff right just to get started, such as loading files from disc, parsing and serializing them, charsets, inlining, finding references to other files, resolution of and updating urls, etc.

The observation that inspired the project is that most of these tasks can be viewed as graph problems, where the nodes are the assets (HTML, CSS, images, JavaScript...) and the edges are the relations between them, e.g. anchor tags, image tags, favorite icons, css background-image properties and so on.

An AssetGraph object is a collection of assets (nodes) and the relations (edges) between them. It's a basic data model that allows you to populate, query, and manipulate the graph at a high level of abstraction. For instance, if you change the url of an asset, all relations pointing at it are automatically updated.

Additionally, each individual asset can be inspected and massaged using a relevant API: DOM for HTML (using jsdom, CSSOM for CSS (using NV's CSSOM module, and an abstract syntax tree for JavaScript (powered by UglifyJS' parser).

AssetGraph represents inline assets the same way as non-inline ones, so eg. inline scripts, stylesheets, and images specified as data: urls are also first-class nodes in the graph. This means that you don't need to dig into the HTML of the containing asset to manipulate them. An extreme example would be an Html asset with a conditional comment with an inline stylesheet with an inline image, which are modelled as 4 separate assets:

<!DOCTYPE html>
  <!--[if !IE]> -->
    <style type='text/css'>
      body {
        background-image: url(data:image/gif;base64,R0lGODlhAQABAID/AMDAwAAAACH5BAEAAAAALAAAAAABAAEAAAICRAEAOw==);
  <!-- <![endif]-->

These are some of the supported assets and associated relation types:


<a>, <link rel="stylesheet|shortcut icon|alternate">, <script>, <style>, <html manifest="..."> <img>, <video>, <audio>, <applet>, <embed>, <esi:include>, <iframe>


background-image: url(...), @import url(...), behavior: url(...), filter: AlphaImageLoader(src='...')


AMD/RequireJS require and define, CommonJS require(...), homegrown INCLUDE syntax for specifying requirements, and homegrown GETSTATICURL(...) and GETTEXT(...) syntax for referencing external files


(same as for HTML)

Cache manifest

Entries in the CACHE, NETWORK and FALLBACK sections




  • Build an AssetGraph programmatically or load it from disk or a remote server via http.
  • Find explicit dependencies between JavaScript and CSS and roll them out as <script> and <link rel='stylesheet'> tags in your HTML. For now require.js/AMD, the ExtJS 4 syntax and a homegrown INCLUDE syntax are supported, but the parsing phase can be adapted to almost any syntax. Support for more script loaders will be added on demand.
  • Bundle and inline CSS and JavaScript.
  • Create a cache manifest with references to all the assets your web app needs to be usable offline.
  • Move all CSS, JavaScript, image assets etc. to a static dir and rename them to md5.extension so the web server can be configured to set a far-future Cache-Control.
  • Help getting your static assets on a CDN by allowing you to easily rewrite all references to them.
  • Use Graphviz to visualize your dependencies at any step.
  • Using the separate assetgraph-sprite transform: Optimize CSS background images by creating sprite images. The spriting is guided by a set of custom CSS properties with a -ag-sprite prefix.


Make sure you have node.js and npm installed, then run:

$ npm install assetgraph

API documentation

A work in progress. Look here.

Querying the graph

AssetGraph supports a flexible syntax for finding assets and relations in a populated graph using the findAssets and findRelations methods. Both methods take a query object as the first argument. Below are some basic examples.

Get an array containing all assets in the graph:

var allAssets = assetGraph.findAssets();

Find assets by type:

var htmlAssets = assetGraph.findAssets({type: 'Html'});

Find assets by matching a regular expression against the url:

var localImageAssets = assetGraph.findAssets({url: /^file:.*\.(?:png|gif|jpg)$/});

Find assets by predicate function:

var orphanedJavaScriptAssets = assetGraph.findAssets(function (asset) {
    return asset.type === 'JavaScript' && assetGraph.findRelations({to: asset}).length === 0;

Find all HtmlScript (<script src=...> and inline <script>) relations:

var allHtmlScriptRelations = assetGraph.findRelations({type: 'HtmlScript'});

Query objects have "and" semantics, so all conditions must be met for a multi-criteria query to match:

var textBasedAssetsOnGoogleCom = assetGraph.findAssets({
    isText: true,
    url: /^https?:\/\/(?:www\.)google\.com\//

Find assets by existence of incoming relations (experimental feature):

var importedCssAssets = assetGraph.findAssets({type: 'Css', incoming: {type: 'CssImport'}})

Relation queries can contain nested asset queries when querying the to and from properties.

Find all HtmlAnchor (<a href=...>) relations pointing at local images:

    type: 'HtmlAnchor',
    to: {isImage: true, url: /^file:/}

Transforms and workflows

AssetGraph comes with a collection of premade "transforms" that you can use as high level building blocks when putting together your build procedure. Most transforms work on a set of assets or relations and usually accept a query object so they can be scoped to work on only a specific subset of the graph.

Usually you'll start by loading some initial assets from disc or via http using the loadAssets transform, then get the related assets added using the populate transform, then do the actual processing. Eventually you'll probably write the resulting assets back to disc.

Thus the skeleton looks something like this:

var AssetGraph = require('assetgraph');
new AssetGraph({root: '/the/root/directory/'})
    .loadAssets('*.html') // Load all Html assets in the root dir
    .populate({followRelations: {type: 'HtmlAnchor'}}) // Follow <a href=...>
    // More work...
    .writeAssetsToDisc({type: 'Html'}) // Overwrite existing files
    .run(function (err, assetGraph) {
        // Done!

In the following sections the built-in transforms are documented individually:


Add a CacheManifest asset to each Html asset in the graph (or to all Html assets matched by queryObj if provided). The cache manifests will contain relations to all assets reachable by traversing the graph through relations other than HtmlAnchor.

assetGraph.bundleRelations(queryObj[, strategyName])

Bundle the Css and JavaScript assets pointed to by the relations matched by queryObj.

The strategyName (string) parameter can be either:

oneBundlePerIncludingAsset (the default)

Each unique asset pointing to one or more of the assets being bundled will get its own bundle. This can lead to duplication if eg. several Html assets point to the same sets of assets, but guarantees that the number of http requests is kept low.


Create as many bundles as needed, optimizing for combined byte size of the bundles rather than http requests. Warning: Not as well tested as oneBundlePerIncludingAsset.

Note that a conditional comment within an Html asset conveniently counts as a separate including asset, so in the below example ie.css and all.css won't be bundled together:

<![iIE]><link rel='stylesheet' href='ie.css'><![endif]-->
<link rel='stylesheet' href='all.css'>

The created bundles will be placed at the root of the asset graph with names derived from their unique id (for example file://root/of/graph/124.css) and will replace the original assets.


Finds all CoffeeScript assets in the graph (or those specified by queryObj), compiles them to JavaScript assets and replaces the originals.


Finds all Less assets in the graph (or those specified by queryObj), compiles them to Css assets and replaces the originals.


Finds all Scss assets in the graph (or those specified by queryObj), compiles them to Css assets and replaces the originals.

assetGraph.compressJavaScript([queryObj[, compressorName[, compressorOptions]]])

Compresses all JavaScript assets in the graph (or those specified by queryObj).

The compressorName (string) parameter can be either:

uglifyJs (the default and the fastest)

The excellent UglifyJS compressor. If provided, the compressorOptions object will be passed to UglifyJS' ast_squeeze command.


Yahoo's YUICompressor though Tim-Smart's node-yuicompressor module. If provided, the compressorOptions object will be passed as the second argument to require('yui-compressor').compile.


Google's Closure Compiler through Tim-Smart's node-closure module. If provided, the compressorOptions object will be passed as the second argument to require('closure-compiler').compile.


Finds all Html assets in the graph (or those specified by queryObj), finds all CssImport relations (@import url(...)) in inline and external CSS and converts them to HtmlStyle relations directly from the Html document.

Effectively the inverse of assetGraph.convertHtmlStylesToInlineCssImports.


<style type='text/css'>
    @import url(print.css) print;
    @import url(foo.css);
    body {color: red;}

is turned into:

<link rel='stylesheet' href='print.css' media='print'>
<link rel='stylesheet' href='foo.css'>
<style type='text/css'>
    body {color: red;}


Finds all Html assets in the graph (or those specified by queryObj), finds all outgoing, non-inline HtmlStyle relations (<link rel='stylesheet' href='...'>) and turns them into groups of CssImport relations (@import url(...)) in inline stylesheets. A maximum of 31 CssImports will be created per inline stylesheet.


<link rel='stylesheet' href='foo.css'>
<link rel='stylesheet' href='bar.css'>

is turned into:

<style type='text/css'>
    @import url(foo.css);
    @import url(bar.css);

This is a workaround for the limit of 31 stylesheets in Internet Explorer <= 8. This transform allows you to have up to 31*31 stylesheets in the development version of your HTML and still have it work in older Internet Explorer versions.


Uses the Graphviz dot command to render the current contents of the graph and writes the result to fileName. The image format is automatically derived from the extension and can be any of these. Using .svg is recommended.

Requires Graphviz to be installed, sudo apt-get install graphviz on Debian/Ubuntu.

assetGraph.executeJavaScriptInOrder(queryObj[, context])

Experimental: For each JavaScript asset in the graph (or those matched by queryObj), find all reachable JavaScript assets and execute them in order.

If the context parameter is specified, it will be used as the execution context. Otherwise a new context will be created using vm.createContext.


Finds all inline relations in the graph (or those matched by queryObj) and makes them external. The file names will be derived from the unique ids of the assets.

For example:

<script>foo = 'bar';</script> 
<style type='text/css'>body {color: maroon;}</style> 

could be turned into:

<script src='4.js'></script>
<link rel='stylesheet' href='5.css'>


Finds all Html assets in the graph (or those matched by queryObj), finds all JavaScript and Css assets reachable through HtmlScript, HtmlStyle, JavaScriptOneInclude, and JavaScriptExtJsRequire relations and rolls them out as plain HtmlScript (<script src='...'>) and HtmlStyle (<link rel='stylesheet' href='...'>) relations.

If your project uses deeply nested INCLUDE statements, this transform allows you to create a "development version" that works in a browser. Refer to the buildDevelopment script from AssetGraph-builder.

For example:


where foo.js contains:

var blah = 'baz';

is turned into:

    <link rel='stylesheet' href='quux.css'>
<script src='bar.js'></script>
<script src='foo.js'></script>

assetGraph.inlineCssImagesWithLegacyFallback([queryObj[, sizeThreshold]])

Finds all Html assets in the graph (or those matched by queryObj), finds all directly reachable Css assets, and converts the outgoing CssImage relations (background-image etc.) to data: urls, subject to these criteria:

  1. If sizeThreshold is specified, images with a greater byte size won't be inlined.

  2. To avoid duplication, images referenced by more than one CssImage relation won't be inlined.

  3. A CssImage relation pointing at an image with an inline GET parameter will always be inlined (eg. background-image: url(foo.png?inline);). This takes precedence over the first two criteria.

If any image is inlined an Internet Explorer-only version of the stylesheet will be created and referenced from the Html asset in a conditional comment.

For example:

    .run(funtion (err, assetGraph) {...});

where assetGraph contains an Html asset with this fragment:

<link rel='stylesheet' href='foo.css'>

and foo.css contains:

body {background-image: url(small.png);}

will be turned into:

<!--[if IE]><link rel="stylesheet" href="foo.css"><![endif]-->
<!--[if !IE]>--><link rel="stylesheet" href="1234.css"><!--<![endif]-->

where 1234.css is a copy of the original foo.css with the images inlined as data: urls:

body {background-image: url(data;image/png;base64,iVBORw0KGgoAAAANSUhE...)}

The file name 1234.css is just an example. The actual asset file name will be derived from the unique id of the copy and be placed at the root of the assetgraph.


Inlines all relations in the graph (or those matched by queryObj). Only works on relation types that support inlining, for example HtmlScript, HtmlStyle, and CssImage.


assetGraph.inlineRelations({type: ['HtmlStyle', 'CssImage']});

where assetGraph contains an Html asset with this fragment:

<link rel='stylesheet' href='foo.css'>

and foo.css contains:

body {background-image: url(small.png);}

will be turned into:

<style type='text/css'>body {background-image: url(data;image/png;base64,iVBORw0KGgoAAAANSUhE...)}</style> 

Note that foo.css and the CssImage will still be modelled as separate assets after being inlined, so they can be manipulated the same way as when they were external.

assetGraph.loadAssets(fileName|wildcard|url|Asset[, ...])

Add new assets to the graph and make sure they are loaded. Several syntaxes are supported, for example:

assetGraph.loadAssets('a.html', 'b.css'); // Relative to assetGraph.root
assetGraph.loadAssets(new AssetGraph.JavaScript({
    url: "",
    text: "var foo = bar;" // The source is specified, won't be loaded

file:// urls support wildcard expansion:

assetGraph.loadAssets('file:///foo/bar/*.html'); // Wildcard expansion
assetGraph.loadAssets('*.html'); // assetGraph.root must be file://...


Compute the MD5 sum of every asset in the graph (or those specified by queryObj) and remove duplicates. The relations pointing at the removed assets are updated to point at the copy that is kept.

For example:


where assetGraph contains an Html asset with this fragment:

    <style type='text/css'>body {background-image: url(foo.png);}</style> 
    <img src='bar.png'>

will be turned into the following if foo.png and bar.png are identical:

    <style type='text/css'>body {background-image: url(foo.png);}</style> 
    <img src='foo.png'>

and the bar.png asset will be removed from the graph.


Minify all assets in the graph, or those specified by queryObj. Only has an effect for asset types that support minification, and what actually happens also varies:

Html and Xml

Pure-whitespace text nodes are removed immediately.

Json, JavaScript, and Css

The asset gets marked as minified (isPretty is set to false), which doesn't affect the in-memory representation (asset.parseTree), but is honored when the asset is serialized. For JavaScript this only governs the amount of whitespace (UglifyJS' beautify parameter); for how to apply variable renaming and other compression techniques see assetGraph.compressJavaScript.

Compare to assetGraph.prettyPrintAssets.

assetGraph.moveAssets(queryObj, newUrlFunctionOrString)

Change the url of all assets matching queryObj. If the second argument is a function, it will be called with each asset as the first argument and the assetGraph instance as the second and the url of the asset will be changed according to the return value:

  • If a falsy value is returned, nothing happens; the asset keeps its current url.
  • If a non-absolute url is returned, it is resolved from assetGraph.root.
  • If the url ends in a slash, the file name part of the old url is appended.

Move all Css and Png assets to a root-relative url:

assetGraph.moveAssets({type: 'Css'}, '/images/');

If the graph contains and assetGraph.root is file:///my/local/dir/, the resulting url will be file:///my/local/dir/images/bar.css.

Move all non-inline JavaScript and Css assets to either or, preserving the current file name part of their url:

assetGraph.moveAssets({type: ['JavaScript', 'Css'], isInline: false}, function (asset, assetGraph) {
    return "" + asset.type.toLowerCase() + "/" + asset.fileName;

The assets are moved in no particular order. Compare with assetGraph.moveAssetsInOrder.

assetGraph.moveAssetsInOrder(queryObj, newUrlFunctionOrString)

Does the same as assetGraph.moveAssets, but makes sure that the "leaf assets" are moved before the assets that have outgoing relations to them.

The typical use case for this is when you want to rename assets to <hashOfContents>.<extension> while making sure that the hashes of the assets that have already been moved don't change as a result of updating the urls of the related assets after the fact.

Here's a simplified example taken from buildProduction in AssetGraph-builder.

assetGraph.moveAssetsInOrder({type: ['JavaScript', 'Css', 'Jpeg', 'Gif', 'Png']}, function (asset) {
    return '/static/' + asset.md5Hex.substr(0, 10) + asset.extension;

If a graph contains an Html asset with a relation to a Css asset that again has a relation to a Png asset, the above snippet will always move the Png asset before the Css asset, thus making it safe to compute the md5 of the respective assets when the function is invoked.

Obviously this only works for graphs (or subsets of graphs) that don't contain cycles, and if that's not the case, an error will be thrown.


Add assets to the graph by recursively following "dangling relations". This is the preferred way to load a complete web site or web application into an AssetGraph instance after using assetGraph.loadAssets to add one or more assets to serve as the starting point for the population. The loading of the assets happens in parallel.

The options object can contain these properties:

from: queryObj

Specifies the set assets of assets to start populating from (defaults to all assets in the graph).

followRelations: queryObj

Limits the set of relations that are followed. The default is to follow all relations.

onError: function (err, assetGraph, asset)

If there's an error loading an asset and an onError function is specified, it will be called, and the population will continue. If not specified, the population will stop and pass on the error to its callback. (This is poorly thought out and should be removed or redesigned).

concurrency: Number

The maximum number of assets that can be loading at once (defaults to 100).


new AssetGraph()
        followRelations: {type: 'HtmlAnchor', to: {url: /\/[bc]\.html$/}}
    .run(function (err, assetGraph) {
        // Done!

If a.html links to b.html, and b.html links to c.html (using <a href="...">), all three assets will be in the graph after assetGraph.populate is done. If c.html happens to link to d.html, d.html won't be added.


Pretty-print all assets in the graph, or those specified by queryObj. Only has an effect for asset types that support pretty printing (JavaScript, Css, Html, Xml, and Json).

The asset gets marked as pretty printed (isPretty is set to true), which doesn't affect the in-memory representation (asset.parseTree), but is honored when the asset is serialized. For Xml, and Html, however, the existing whitespace-only text nodes in the document are removed immediately.

Compare to assetGraph.minifyAssets.


// Pretty-print all Html and Css assets:
assetGraph.prettyPrintAssets({type: ['Html', 'Css']});

assetGraph.removeAssets([queryObj[, detachIncomingRelations]])

Remove all assets in the graph, or those specified by queryObj, along with their incoming relations. If detachIncomingRelations is set to true, the incoming relations will also be detached (removed from the parse tree of the source asset). This is not supported by all relation types.


var AssetGraph = require('assetgraph');
new AssetGraph()
    // Add a Html asset with an inline Css asset:
    .loadAssets(new AssetGraph.Html({
        text: '<html><head><style type="text/css">body {color: red;}</style></head></html>'
    // Remove the inline Css asset and detach the incoming HtmlStyle relation:
    .removeAssets({type: 'Css'}, true)
    // Now the graph only contains the Html asset (without the <style> element):
    .writeAssetsToStdout({type: 'Html'})
    // '<html><head></head></html>'
    .run(function (err, assetGraph) {
        // Done!

assetGraph.removeRelations([queryObj, [options]])

Remove all relations in the graph, or those specified by queryObj.

The options object can contain these properties:

detach: Boolean

Whether to also detach the relations (remove their nodes from the parse tree of the source asset). Only supported for some relation types. Defaults to false.

unresolved: Boolean

Whether to remove unresolved relations too ("dangling" ones whose target assets aren't in the graph). Defaults to false.

removeOrphan: Boolean

Whether to also remove assets that become "orphans" as a result of removing their last incoming relation.

assetGraph.setAssetContentType(queryObj, contentType)

Updates the contentType property of all assets matching queryObj. After an asset is loaded, the contentType property is only kept around as a handy piece of metadata, so updating it has no side effects. It's mostly useful if want to upload a "snapshot" of an AssetGraph to a WebDAV server or similar.

assetGraph.setAssetEncoding(queryObj, newEncoding)

Changes the encoding (charset) of the assets matched by queryObj to encoding (utf-8, windows-1252, TIS-620, etc.). Only works for text-based assets. Affects the rawSrc property of the asset, the decoded text property remains unchanged.

Uses node-iconv to do the actual text conversion, so make sure the charset is supported.

As a convenient side effect, Html assets with a <head> element will get a <meta http-equiv="Content-Type" content="..."> appended specifying the new encoding. If such a <meta> already exists, it will be updated.


var AssetGraph = require('assetgraph');
new AssetGraph()
    // Add a Html asset with an inline Css asset:
    .loadAssets(new AssetGraph.Html({
        text: '<html><head></head>æ</html>'
    .setAssetEncoding({type: 'Html'}, 'iso-8859-1')
    .writeAssetsToStdout({type: 'Html'})
    // <html><head></head><meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1"></head>�</html>
    .run(function (err, assetGraph) {
        // Done!

assetGraph.setAssetExtension(queryObj, extension)

Changes the extension part of the urls of all non-inline assets matching queryObj to extension. The extension should include the leading dot like the require('path').extname() function.


var AssetGraph = require('assetgraph');
new AssetGraph()
    .setAssetExtension({type: 'Html'}, '.bar')
    .run(function (err, assetGraph) {
        if (err) throw err;
        console.log(assetGraph.findAssets({type: 'Html'})[0].url); // ''
        // Done!


Sets the width and height attributes of the img elements underlying all HtmlImage relations, or those matching queryObj. Only works when the image pointed to by the relation is in the graph.


var AssetGraph = require('assetgraph');
new AssetGraph()
    // assetGraph.findAssets({type: 'Html'})[0].text === '<body><img src="foo.png"></body>'
    // assetGraph.findAssets({type: 'Html'})[0].text === '<body><img src="foo.png" width="29" height="32"></body>'
    .run(function (err, assetGraph) {
        // Done!


Starts watching all non-inline file:// assets (or those matching queryObj) as they're added to the graph, and reruns all the following transformations when a source file is changed on disc.

Used to power buildDevelopment --watch in AssetGraph-builder. Should be considered experimental.


Dumps an ASCII table with some basic stats about all the assets in the graph (or those matching queryObj) in their current state.


       Ico   1   1.1 KB
       Png  28 196.8 KB
       Gif 145 129.4 KB
      Json   2  60.1 KB
       Css   2 412.6 KB
JavaScript  34   1.5 MB
      Html   1   1.3 KB
    Total: 213   2.2 MB

assetGraph.writeAssetsToDisc(queryObj, outRoot[, root])

Writes the assets matching queryObj to disc. The outRoot parameter must be a file:// url specifying the directory where the files should be output. The optional root parameter specifies the url that you want to correspond to the outRoot directory (defaults to the root property of the AssetGraph instance).

Directories will be created as needed.


var AssetGraph = require('assetgraph');
new AssetGraph({root: ''})
    // Will write the two assets to /my/output/dir/quux/foo.html and /my/output/dir/baz.html:
    .writeAssetsToDisc({type: 'Html'} 'file:///my/output/dir/', '')
    .run(function (err, assetGraph) {
        // Done!


Writes all assets in the graph (or those specified by queryObj) to stdout. Mostly useful for piping out a single asset.


AssetGraph is licensed under a standard 3-clause BSD license -- see the LICENSE-file for details.




npm i assetgraph-esprima

Downloadsweekly downloads









last publish


  • avatar
Report a vulnerability