Created attachment 292613 [details]
reproducer

When the XML Reader API is used with XML_PARSE_NOCDATA and XML Schema validation, then LibXML2 appears to drop CDATA sections. The text it returns is as if the CDATA sections were empty.

The attached program works on this XML code:

"<test>one <![CDATA[two]]> three</test>"

It parses this code three times with different combinations of validation and NOCDATA, and lists the nodes it encounters. It should find the same content whether validation is enabled or not. This is the actual output:

validating, not decoding CDATA:
  start tag
  text: "one "
  CDATA: "two"
  text: " three"
  end tag

not validating, decoding CDATA:
  start tag
  text: "one two three"
  end tag

validating, decoding CDATA:
  start tag
  text: "one  three"
  end tag

1 errors detected