| 2005-04-03 Juergen Donnerstag | |
| Once we introduced wicket namespace to handle wicket tags and attributes | |
| like <wicket:link ..> and <a wicket:id="myLabel" ..> wicket-example tests | |
| failed. Tracing the error down through jWebUnit and HttpUnit I found that | |
| HttpUnit calls ParseHTML.getDOM() for each assertXXX() like test. getDOM() | |
| calls Node.cloneNode( /* deep */ true) to make a copy of the DOM tree | |
| (obviously to protected the DOM against accidental changes). As Xerces-2.6.2 | |
| is the underlying XML implementation, the Xerces2 implementation of cloneNode() | |
| gets called. And cloneNode() is causing all the problems. From my point the | |
| cloneNode() (resp. importNode) implementation has several bugs with respect to | |
| namespaces (though I have not studied the xml spec in detail). | |
| a) cloneNode() first creates an empty Document but does not copy the | |
| strictErrorHandling() attribute from the original document. As default | |
| for this attribute is true, strict error handling is always enable | |
| during clone node and there is no way to disable it. | |
| Because html with namespace is not valid xml and since we do not have a | |
| xhtml + wicket DTD, wicket's output with wicket namespace is non-valid xml. | |
| (of course you can strip wicket tags and attrs from output, but for | |
| debugging purposes you'll still need it). Thus Xerces throws a | |
| NAMESPACE_ERR exception. In order to circumvent that problem I copied | |
| the HtmlDocumentImpl.java and changed cloneNode() to always disable | |
| error handling while cloning. | |
| b) Unfortunately cloneNode() still didn't work. Tags like | |
| <span wicket:id="myLabel" id="xxxx"> failed. The clone always looked | |
| like <span wicket:id="myLabel"> and id="xxxx" was swallowed. The reason, | |
| again, was not 100% compliant XML. The wicket attribute did not have a namespace | |
| URI (though I specific <html xmlns:wicket="http://...") and thus the | |
| namespace was simply neglected. Because it was neglected Xerces now | |
| thinks that wicket:id == id and simply replaces the first occurence | |
| with the second. The outcome was that always on attribute was swallowed. | |
| Because 2.6.2 is the newest version but has been available for more than | |
| a year, I checked the current CVS head as well. The problems however | |
| still seem to be there. | |
| As I didn't find a better solution I eventually made a "local" copy of | |
| HttpUnit:ParsedHTML.java and modified getDOM() to not make copy but to | |
| use the original DOM tree. | |
| I'm not even sure a XHTML + Wicket DTD will solve our problem completely. | |
| The reason being that ALL wicket markup must than be XHTML compliant. | |
| As plain (old) HTML however should still be supported ... | |
| In case anyone reads this and nows a better solution, please let me know. |