[bug] Nokogiri::XML::Document#css no longer matches namespaced XML attributes

[stanhu]

Please describe the bug

I'm not sure if this is a bug, but I noticed when trying to upgrade from nokogiri v1.16.8 to v1.17.2, that this behavior changed:

require 'nokogiri'

body = <<-XML
<OpenSearchDescription xmlns="http://a9.com/-/spec/opensearch/1.1/" xmlns:moz="http://www.mozilla.org/2006/browser/search/">
  <ShortName>Test</ShortName>
  <Description>Search Test</Description>
  <InputEncoding>UTF-8</InputEncoding>
  <Image width="16" height="16" type="image/x-icon">http://test.host/favicon.ico</Image>
  <Url type="text/html" method="get" template="http://test.host/search?search={searchTerms}"/>
  <moz:SearchForm>http://test.host/search</moz:SearchForm>
</OpenSearchDescription>
XML

doc = Nokogiri::XML.parse(body)

puts doc.css('OpenSearchDescription *').map(&:name)

In nokogiri v1.16.8, this returned:

ShortName
Description
InputEncoding
Image
Url
SearchForm

In nokogiri v1.17.2, SearchForm is excluded:

ShortName
Description
InputEncoding
Image
Url

I see three commits in lib/nokogiri/xml/searchable.rb that might be relevant:

• fb19c55
• a84ad7f
• 179d74d

Is this intentional? Do I need to change my selector to handle this?

[stanhu]

It seems I have two options. Option 1 is to drop namespaces:

doc.remove_namespaces!

Option 2 is to register the namespaces and update the query:

namespaces = {
  'default' => 'http://a9.com/-/spec/opensearch/1.1/',
  'moz' => 'http://www.mozilla.org/2006/browser/search/'
}

doc.xpath('//default:OpenSearchDescription/*', namespaces)

[flavorjones]

@stanhu Interesting! I'll take a look as soon as I can.

[flavorjones]

Behavior change was introduced in 179d74d:

commit 179d74d (HEAD)
Author: Mike Dalessio mike.dalessio@gmail.com
Date: 2024-06-11 12:03:12 -0400

refactor: move namespaces into the XPathVisitor

Previously this was passed into CSS::Parser.new and used during the
parsing phase. Now it's passed into the visitor and we do namespace
munging there, where it should be.

Note the namespaces has also been moved out of the top-level parser
cache key into XPathVisitor#config.

Finally, note that namespaces is nil by default (instead of being an
empty Hash) which should prevent some unnecessary object creation in
the most common use cases.

So this seems like it's a bug, I'll fix it.

[flavorjones]

Proposed fix at #3413

[flavorjones]

Backporting to v1.18.x for a bugfix release: #3414

[flavorjones]

Fixed in Release v1.18.2 / 2024-01-19 · sparklemotion/nokogiri