Why does this xpath fail using lxml in python?

问题:

Here is an example web page I am trying to get data from. http://www.makospearguns.com/product-p/mcffgb.htm

The xpath was taken from chrome development tools, and firepath in firefox is also able to find it, but using lxml it just returns an empty list for 'text'.

from lxml import html
import requests

site_url = 'http://www.makospearguns.com/product-p/mcffgb.htm'
xpath = '//*[@id="v65-product-parent"]/tbody/tr[2]/td[2]/table[1]/tbody/tr/td/table/tbody/tr[2]/td[2]/table/tbody/tr[1]/td[1]/div/table/tbody/tr/td/font/div/b/span/text()'

page = requests.get(site_url)
tree = html.fromstring(page.text) 
text = tree.xpath(xpath)

Printing out the tree text with

print(tree.text_content().encode('utf-8'))

shows that the data is there, but it seems the xpath isn't working to find it. Is there something I am missing? Most other sites I have tried work fine using lxml and the xpath taken from chrome dev tools, but a few I have found give empty lists.

回答1:

1. Browsers frequently change the HTML

Browsers quite frequently change the HTML served to it to make it "valid". For example, if you serve a browser this invalid HTML:

<table>
  <p>bad paragraph</p>
  <tr><td>Note that cells and rows can be unclosed (and valid) in HTML
</table>

To render it, the browser is helpful and tries to make it valid HTML and may convert this to:

<p>bad paragraph</p>
<table>
  <tbody>
    <tr>
      <td>Note that cells and rows can be unclosed (and valid) in HTML</td>
    </tr>
  </tbody>
</table>

The above is changed because <p>aragraphs cannot be inside <table>s and <tbody>s are recommended. What changes are applied to the source can vary wildly by browser. Some will put invalid elements before tables, some after, some inside cells, etc...

2. Xpaths aren't fixed, they are flexible in pointing to elements.

Using this 'fixed' HTML:

<p>bad paragraph</p>
<table>
  <tbody>
    <tr>
      <td>Note that cells and rows can be unclosed (and valid) in HTML</td>
    </tr>
  </tbody>
</table>

If we try to target the text of <td> cell, all of the following will give you approximately the right information:

//td
//tr/td
//tbody/tr/td
/table/tbody/tr/td
/table//*/text()

And the list goes on...

however, in general browser will give you the most precise (and least flexible) XPath that lists every element from the DOM. In this case:

/table[0]/tbody[0]/tr[0]/td[0]/text()

3. Conclusion: Browser given Xpaths are usually unhelpful

This is why the XPaths produced by developer tools will frequently give you the wrong Xpath when trying to use the raw HTML.

The solution, always refer to the raw HTML and use a flexible, but precise XPath.

Examine the actual HTML that holds the price:

<table border="0" cellspacing="0" cellpadding="0">
    <tr>
        <td>
            <font class="pricecolor colors_productprice">
                <div class="product_productprice">
                    <b>
                        <font class="text colors_text">Price:</font>
                        <span itemprop="price">$149.95</span>
                    </b>
                </div>
            </font>
            <br/>
            <input type="image" src="/v/vspfiles/templates/MAKO/images/buttons/btn_updateprice.gif" name="btnupdateprice" alt="Update Price" border="0"/>
        </td>
    </tr>
</table>

If you want the price, there is actually only one place to look!

//span[@itemprop="price"]/text()

And this will return:

$149.95

回答2:

The xpath is simply wrong

Here is snippet from the page:

<form id="vCSS_mainform" method="post" name="MainForm" action="/ProductDetails.asp?ProductCode=MCFFGB" onsubmit="javascript:return QtyEnabledAddToCart_SuppressFormIE();">
      <img src="/v/vspfiles/templates/MAKO/images/clear1x1.gif" width="5" height="5" alt="" /><br />
      <table width="100%" cellpadding="0" cellspacing="0" border="0" id="v65-product-parent">
        <tr>
          <td colspan="2" class="vCSS_breadcrumb_td"><b>
&nbsp; 
<a href="http://www.makospearguns.com/">Home</a> > 

You can see, that element with id being "v65-product-parent" is of typetableand has subelementtr`.

There can be only one element with such id (otherwise it would be broken xml).

The xpath is expecting tbody as child of given element (table) and there is none in whole page.

This can be tested by

>>> "tbody" in page.text
False

How Chrome came to that XPath?

If you simply download this page by

$ wget http://www.makospearguns.com/product-p/mcffgb.htm

and review content of it, it does not contain a single element named tbody

But if you use Chrome Developer Tools, you find some.

How it comes here?

This often happens, if JavaScript comes into play and generates some page content when in the browser. But as LegoStormtroopr noted, this is not our case and this time it is the browser, which modifies document to make it correct.

How to get content of page dynamically modified within browser?

You have to give some sort of browser a chance. E.g. if you use selenium, you would get it.

byselenium.py

from selenium import webdriver
from lxml import html

url = "http://www.makospearguns.com/product-p/mcffgb.htm"
xpath = '//*[@id="v65-product-parent"]/tbody/tr[2]/td[2]/table[1]/tbody/tr/td/table/tbody/tr[2]/td[2]/table/tbody/tr[1]/td[1]/div/table/tbody/tr/td/font/div/b/span/text()'

browser = webdriver.Firefox()
browser.get(url)
html_source = browser.page_source
print "test tbody", "tbody" in html_source

tree = html.fromstring(html_source) 
text = tree.xpath(xpath)
print text

what prints

$ python byselenimum.py 
test tbody True
['$149.95']

Conclusions

Selenium is great when it comes to changes within browser. However it is a bit heavy tool and if you can do it simpler way, do it that way. Lego Stormrtoopr have proposed such a simpler solution working on simply fetched web page.

回答3:

I had a similar issue (Chrome inserting tbody elements when you do Copy as XPath). As others answered, you have to look at the actual page source, though the browser-given XPath is a good place to start. I've found that often, removing tbody tags fixes it, and to test this I wrote a small Python utility script to test XPaths:

#!/usr/bin/env python
import sys, requests
from lxml import html
if (len(sys.argv) < 3):
     print 'Usage: ' + sys.argv[0] + ' url xpath'
     sys.exit(1)
else:
    url = sys.argv[1]
    xp = sys.argv[2]

page = requests.get(url)
tree = html.fromstring(page.text)
nodes = tree.xpath(xp)

if (len(nodes) == 0):
     print 'XPath did not match any nodes'
else:
     # tree.xpath(xp) produces a list, so always just take first item
     print (nodes[0]).text_content().encode('ascii', 'ignore')

(that's Python 2.7, in case the non-function "print" didn't give it away)