Andrey Borisko suggested to use the disk cache to retrieve the resources. My solution is not that efficient, but you don't need to decompress text files.
I use XMLHttpRequest to retrieve all resources after I registered them with the resource.received
event handler. I then filter the resources into images, css and fonts. The current limitation is that remote resource paths that contain something like ../
or ./
are not handled correctly.
I retrieve the current page content with getHTML
and iterate over all captured resources to replace the path used in the markup, that is identified by a portion of the complete resource URL, with a randomly generated file name. The file extension is created from the content type of the resource. It is converted using mimeType from this gist.
Since CSS files may contain background images or fonts, they have to be processed before saving to disk. The provided loadResource
function loads the resource, but does not save it.
Since XMLHttpRequest to download the resources the script has to be invoked with the --web-security=false
flag:
casperjs script.js --web-security=false
script.js
var casper = require("casper").create();
var utils = require('utils');
var fs = require('fs');
var mimetype = require('./mimetype'); // URL provided below
var cssResources = [];
var imgResources = [];
var fontResources = [];
var resourceDirectory = "resources";
var debug = false;
fs.removeTree(resourceDirectory);
casper.on("remote.message", function(msg){
this.echo("remote.msg: " + msg);
});
casper.on("resource.error", function(resourceError){
this.echo("res.err: " + JSON.stringify(resourceError));
});
casper.on("page.error", function(pageError){
this.echo("page.err: " + JSON.stringify(pageError));
});
casper.on("downloaded.file", function(targetPath){
if (debug) this.echo("dl.file: " + targetPath);
});
casper.on("resource.received", function(resource){
// don't try to download data:* URI and only use stage == "end"
if (resource.url.indexOf("data:") != 0 && resource.stage == "end") {
if (resource.contentType == "text/css") {
cssResources.push({obj: resource, file: false});
}
if (resource.contentType.indexOf("image/") == 0) {
imgResources.push({obj: resource, file: false});
}
if (resource.contentType.indexOf("application/x-font-") == 0) {
fontResources.push({obj: resource, file: false});
}
}
});
// based on http://docs.casperjs.org/en/latest/modules/casper.html#download
casper.loadResource = function loadResource(url, method, data) {
"use strict";
this.checkStarted();
var cu = require('clientutils').create(utils.mergeObjects({}, this.options));
return cu.decode(this.base64encode(url, method, data));
};
function escapeRegExp(string) {
// from https://stackoverflow.com/a/1144788/1816580
return string.replace(/([.*+?^=!:${}()|[]/\])/g, "\$1");
}
function replaceAll(find, replace, str) {
// from https://stackoverflow.com/a/1144788/1816580
return str.replace(find, replace);
}
var wrapFunctions = [
function wrapQuot1(s){
return '"' + s + '"';
},
function wrapQuot2(s){
return "'" + s + "'";
},
function csswrap(s){
return '(' + s + ')';
}
];
function findAndReplace(doc, resources, resourcesReplacer) {
// change page on the fly
resources.forEach(function(resource){
var url = resource.obj.url;
// don't download again
if (!resource.file) {
// set random filename and download it **or** call further processing which in turn will load ans write to disk
resource.file = resourceDirectory+"/"+Math.random().toString(36).slice(2)+"."+mimetype.ext[resource.obj.contentType];
if (typeof resourcesReplacer != "function") {
if (debug) casper.echo("download resource (" + resource.obj.contentType + "): " + url + " to " + resource.file);
casper.download(url, resource.file, "GET");
} else {
resourcesReplacer(resource);
}
}
wrapFunctions.forEach(function(wrap){
// test the resource url (growing from the back) with a string in the document
var lastURL;
var lastRegExp;
var subURL;
// min length is 4 characters
for(var i = 0; i < url.length-5; i++) {
subURL = url.substring(i);
lastRegExp = new RegExp(escapeRegExp(wrap(subURL)), "g");
if (doc.match(lastRegExp)) {
lastURL = subURL;
break;
}
}
if (lastURL) {
if (debug) casper.echo("replace " + lastURL + " with " + resource.file);
doc = replaceAll(lastRegExp, wrap(resource.file), doc);
}
});
});
return doc;
}
function capturePage(){
// remove all <script> and <base> tags
this.evaluate(function(){
Array.prototype.forEach.call(document.querySelectorAll("script"), function(scr){
scr.parentNode.removeChild(scr);
});
Array.prototype.forEach.call(document.querySelectorAll("base"), function(scr){
scr.parentNode.removeChild(scr);
});
});
// TODO: remove all event handlers in html
var page = this.getHTML();
page = findAndReplace(page, imgResources);
page = findAndReplace(page, cssResources, function(cssResource){
var css = casper.loadResource(cssResource.obj.url, "GET");
css = findAndReplace(css, imgResources);
css = findAndReplace(css, fontResources);
fs.write(cssResource.file, css, "wb");
});
fs.write("page.html", page, "wb");
}
casper.start("http://www.themarysue.com/").wait(3000).then(capturePage).run(function(){
this.echo("DONE");
this.exit();
});
The magic happens in findAndReplace
. capturePage
is completely synchronous so it can be dropped anywhere without much head ache.
URL for mimetype.js