Performant parsing of pages with Node.js and XPath

问题:

I'm into some web scraping with Node.js. I'd like to use XPath as I can generate it semi-automatically with several sorts of GUI. The problem is that I cannot find a way to do this effectively.

1. jsdom is extremely slow. It's parsing 500KiB file in a minute or so with full CPU load and a heavy memory footprint.
2. Popular libraries for HTML parsing (e.g. cheerio) neither support XPath, nor expose W3C-compliant DOM.
3. Effective HTML parsing is, obviously, implemented in WebKit, so using phantom or casper would be an option, but those require to be running in a special way, not just node <script>. I cannot rely on the risk implied by this change. For example, it's much more difficult to find how to run node-inspector with phantom.
4. Spooky is an option, but it's buggy enough, so that it didn't run at all on my machine.

What's the right way to parse an HTML page with XPath then?

回答1:

You can do so in several steps.

1. Parse HTML with parse5. The bad part is that the result is not DOM. Though it's fast enough and W3C-compiant.
2. Serialize it to XHTML with xmlserializer that accepts DOM-like structures of parse5 as input.
3. Parse that XHTML again with xmldom. Now you finally have that DOM.
4. The xpath library builds upon xmldom, allowing you to run XPath queries. Be aware that XHTML has its own namespace, and queries like //a won't work.

Finally you get something like this.

const fs = require('mz/fs');
const xpath = require('xpath');
const parse5 = require('parse5');
const xmlser = require('xmlserializer');
const dom = require('xmldom').DOMParser;

(async () => {
    const html = await fs.readFile('./test.htm');
    const document = parse5.parse(html.toString());
    const xhtml = xmlser.serializeToString(document);
    const doc = new dom().parseFromString(xhtml);
    const select = xpath.useNamespaces({"x": "http://www.w3.org/1999/xhtml"});
    const nodes = select("//x:a/@href", doc);
    console.log(nodes);
})();

回答2:

Libxmljs is currently the fastest implementation (something like a benchmark) since it's only bindings to the LibXML C-library which supports XPath 1.0 queries:

var libxmljs = require("libxmljs");
var xmlDoc = libxmljs.parseXml(xml);
// xpath queries
var gchild = xmlDoc.get('//grandchild');

However, you need to sanitize your HTML first and convert it to proper XML. For that you could either use the HTMLTidy command line utility (tidy -q -asxml input.html), or if you want it to keep node-only, something like xmlserializer should do the trick.

回答3:

I have just started using npm install htmlstrip-native which uses a native implementation to parse and extract the relevant html parts. It is claiming to be 50 times faster than the pure js implementation (I have not verified that claim).

Depending on your needs you can use html-strip directly, or lift the code and bindings to make you own C++ used internally in htmlstrip-native

If you want to use xpath, then use the wrapper already avaialble here; https://www.npmjs.org/package/xpath

回答4:

I think Osmosis is what you're looking for.

• Uses native libxml C bindings
• Supports CSS 3.0 and XPath 1.0 selector hybrids
• Sizzle selectors, Slick selectors, and more
• No large dependencies like jQuery, cheerio, or jsdom

• HTML parser features

  • Fast parsing
  • Very fast searching
  • Small memory footprint

• HTML DOM features

  • Load and search ajax content
  • DOM interaction and events
  • Execute embedded and remote scripts
  • Execute code in the DOM

Here's an example:

osmosis.get(url)
    .find('//div[@class]/ul[2]/li')
    .then(function () {
        count++;
    })
    .done(function () {
        assert.ok(count == 2);
        assert.done();
    });

回答5:

There might be never a right way to parse HTML pages. A very first review on web scraping and crawling shows me that Scrapy can be a good candidate for your need. It accepts both CSS and XPath selectors. In the realm of Node.js, we have a pretty new module node-osmosis. This module is built upon libxmljs so that it is supposed to support both CSS and XPath although I did not find any example using XPath.