DocumentFragment#element_children doesn't return children (MRI)

[kylev]

It appears to usually return zero even when elements are present. Failing test case on MRI (1.9.3-p484), passes on JRuby (1.7.11).

def test_element_children
  doc = Nokogiri::HTML::DocumentFragment.parse("   <div>  </div>\n   ")
  assert doc.element_children.count == 1
end

Given the recent change to whitespace handling in e593e96, element_children has become more important to handle only element Nodes.

[kylev]

Prehaps even more clear, one would expect these to by symmetrical:

def test_element_children
  doc = Nokogiri::HTML::DocumentFragment.parse("   <div>  </div>\n   ")
  assert doc.element_children.count == doc.children.select(:element?).count
end

[flavorjones]

Thanks for reporting this, and for supplying a test case. I'll take a look!

[kylev]

Note, it seems to affect Nokogiri::XML::DocumentFragment as well.

[kylev]

It looks like a libxml2 bug or problem. The implementation for the xmlFirstElementChild doesn't list XML_DOCUMENT_FRAG_NODE among its types that it will look for element nodes within.

I think this is an oversite, so I may work up a fix for them, and maybe we can backport.

Grr.

[kylev]

Patch proposed. We'll see how responsive they are. It's possible it isn't supported for some good reason that isn't apparent to me.

[kylev]

I have a branch with a proposed patch and the above test, but I don't want to turn it into a pull request until I've heard from the libxml2 people. I don't know if there is some purposeful reason they left this functionality out. I'd hate to diverge Nokogiri's libxml2 from upstream.

[kylev]

Upstream libxml2 has replied to the bug by applying the patch to head. #1144 should be safe to apply to Nokogiri now as it won't diverge from libxml2.

[knu]

Thanks for the investigation and the feedback to upstream.
Can this be closed now?

[kylev]

Yup. All done. I anticipate the 1.6.4 release!