DTD.validate(xmlroot) has no reason in .error_log why validation fails

[jenisys]

DTD.validate() returns false, but shows no reason in the DTD.error_log attribute.
This case occurs for example when an XML ID attribute value is used twice in an XML document.
"xmllint --valid ..." (that is based on libxml2, also used by lxml) shows these kind of errors.

NOTE:
Eclipse XML Editor also shows these kind of problems as error markers.

VERSION-INFO:
  Python : (2, 6, 6, 'final', 0)
  lxml.etree : (2, 3, 2, 0)
  libxml used : (2, 7, 3)
  libxml compiled : (2, 7, 3)
  libxslt used : (1, 1, 24)
  libxslt compiled : (1, 1, 24)

[jenisys (jenisys)]

Add example.

[jenisys (jenisys)]

XMLLINT OUTPUT:
xmllint --valid persons.with_duplicate_ID.xml
persons.with_duplicate_ID.xml:7: element person: validity error : ID pid.1 already defined
    <person id="pid.1" description="OOPS, same ID as Alice">
                                                           ^

[scoder (scoder)]

Looks like a bug in libxml2 to me. It has this code in valid.c:

   2636 if ((ctxt != NULL) && (ctxt->error != NULL)) {
   2637 xmlErrValidNode(ctxt, attr->parent, XML_DTD_ID_REDEFINED,
   2638 "ID %s already defined\n",
   2639 value, NULL, NULL);
   2640 }

lxml does not currently set ctxt->error, because all errors are collected globally. The check above prevents libxml2 from reporting the error at all in that case, although it would work if the check wasn't there. Meaning, it's really just exactly this error message that does not go through. Thanks for stumbling over it. :)

[scoder (scoder)]

Work-around implemented here:

https://github.com/lxml/lxml/commit/bcd02706b18c20f90ed23fce10089c3b3f7b0b83

[scoder (scoder)]

Fixed in lxml 3.3.2.