element name case changes when added to document with different encoding

[dcorrigan]

Hi,

It looks like element names are serialized in a case-insensitive way if the element is added as a string to a document that declares a different encoding:

# encoding: utf-8
f = <<-EOT
<?xml version="1.0" encoding="us-ascii"?>
<root>
  <p>foo</p>
</root>
EOT
r = Nokogiri.XML(f)
p = r.at_css('p').replace('<yYy>©bar</yYy>')
puts r.to_s

This produces:

<?xml version="1.0" encoding="us-ascii"?>
<root>
  <yyy>&#169;bar</yyy>
</root>

I tested this on nokogiri 1.6.3.1, libxml2 2.8.0. Is this a libxml2 problem, maybe?

[knu]

Seems replace(Nokogiri::XML::DocumentFragment.parse('<yYy>©bar</yYy>')) works around the problem. node.replace(string) internally calls node.fragment(string), which is where to start investigation.

[dcorrigan]

Thanks, @knu. I'll use the work-around for now.

[flavorjones]

Apologies for not replying for such an embarrassingly long time.

Fragment parsing is a bit complicated because libxml2 doesn't recover from errors while parsing a fragment in the context of a specific node (e.g., node.parse(...) as opposed to Nokogiri::XML::DocumentFragment.parse(...)).

When an error (like the encoding error here, or any other syntax error) is encountered, Nokogiri's fallback behavior is to parse the fragment outside the context of the node. Unfortunately, though, it was using a hardcoded class to do this: Nokogiri::HTML4::DocumentFragment. This uses the HTML parser and causes all elements to be downcased.

See #2388 for fix.