Multi-document mode produces events only for first document

[display-none]

When trying to parse a stream of documents, woodstox produces events for first document and after an end-document event goes into end-of-input state and reports that it has no more events. Seems like it's not the intended behaviour.

For example the following code will print "Document start" only once.

import com.ctc.wstx.api.WstxInputProperties;
import com.google.common.collect.Lists;
import org.codehaus.stax2.XMLEventReader2;
import org.codehaus.stax2.XMLInputFactory2;

import javax.xml.stream.XMLStreamException;
import javax.xml.stream.events.XMLEvent;
import java.io.StringReader;
import java.util.List;

import static org.apache.commons.lang3.StringUtils.*;

public class Test {

    private static final String MULTIDOC =
            "<?xml version='1.0'?><root>text</root><!--comment-->\n"
            +"<?xml version='1.0'?><root>text</root><?proc instr>\n"
            +"<?xml version='1.0'?><root>text</root><!--comment-->"
            +"<?xml version='1.0' encoding='UTF-8'?><root>text</root><!--comment-->"
            +"<?xml version='1.0' standalone='yes'?><root>text</root><!--comment-->"
            +"<?xml version='1.0'?><root>text</root><!--comment-->";


    public static void main(String[] args) throws XMLStreamException {
        new Test().woodstox();
    }

    public void woodstox() throws XMLStreamException {
        XMLInputFactory2 factory = (XMLInputFactory2) XMLInputFactory2.newInstance();
        factory.setProperty(WstxInputProperties.P_INPUT_PARSING_MODE, WstxInputProperties.PARSING_MODE_DOCUMENTS);

        XMLEventReader2 xmlEventReader = (XMLEventReader2) factory.createXMLEventReader(new StringReader(MULTIDOC));
        while (xmlEventReader.hasNextEvent()) {
            XMLEvent xmlEvent = xmlEventReader.nextEvent();
            if (xmlEvent.isStartDocument()) {
                System.out.println("Document start");
            }
        }
    }

}

woodstox 5.0.3, stax2 4.0.0

[cowtowncoder]

That does seem unexpected. I think most testing has been done with token streams; I don't use event streams myself, ever (there is not much benefit to using, just overhead), so it is possible that something with interaction there might be presenting problems. Perhaps event construction logic does not take multi-doc mode into account, for example.

On plus side this might something relatively easy to fix.

[cowtowncoder]

I can reproduce this, added failing test.

[cowtowncoder]

I can see how this occurs, but not sure how easy it would be to resolve. Part of the problem is with use of Stax2 reference implementation, which can't rely on mode (since it does not know of it).
But I think I can look into this more after releasing intermediate 5.1.0 version.