parser: Only enable SAX2 if there are SAX2 element handlers

This reverts part of commit 235b15a5 for backward compatibility and
adds some comments trying to clarify the whole mess.

Fixes #623.

include/libxml/parser.h

@@ -745,6 +745,19 @@ struct _xmlSAXHandler {
     setDocumentLocatorSAXFunc setDocumentLocator;
     startDocumentSAXFunc startDocument;
     endDocumentSAXFunc endDocument;
+    /*
+     * `startElement` and `endElement` are only used by the legacy SAX1
+     * interface and should not be used in new software. If you really
+     * have to enable SAX1, the preferred way is set the `initialized`
+     * member to 1 instead of XML_SAX2_MAGIC.
+     *
+     * For backward compatibility, it's also possible to set the
+     * `startElementNs` and `endElementNs` handlers to NULL.
+     *
+     * You can also set the XML_PARSE_SAX1 parser option, but versions
+     * older than 2.12.0 will probably crash if this option is provided
+     * together with custom SAX callbacks.
+     */
     startElementSAXFunc startElement;
     endElementSAXFunc endElement;
     referenceSAXFunc reference;
@@ -758,8 +771,14 @@ struct _xmlSAXHandler {
     getParameterEntitySAXFunc getParameterEntity;
     cdataBlockSAXFunc cdataBlock;
     externalSubsetSAXFunc externalSubset;
+    /*
+     * `initialized` should always be set to XML_SAX2_MAGIC to enable the
+     * modern SAX2 interface.
+     */
     unsigned int initialized;
-    /* The following fields are extensions available only on version 2 */
+    /*
+     * The following members are only used by the SAX2 interface.
+     */
     void *_private;
     startElementNsSAX2Func startElementNs;
     endElementNsSAX2Func endElementNs;

parser.c

@@ -863,7 +863,14 @@ xmlDetectSAX2(xmlParserCtxtPtr ctxt) {
     if (ctxt == NULL) return;
     sax = ctxt->sax;
 #ifdef LIBXML_SAX1_ENABLED
-    if ((sax) && (sax->initialized == XML_SAX2_MAGIC))
+    /*
+     * Only enable SAX2 if there SAX2 element handlers, except when there
+     * are no element handlers at all.
+     */
+    if ((sax) && (sax->initialized == XML_SAX2_MAGIC) &&
+        ((sax->startElementNs != NULL) ||
+         (sax->endElementNs != NULL) ||
+         ((sax->startElement == NULL) && (sax->endElement == NULL))))
         ctxt->sax2 = 1;
 #else
     ctxt->sax2 = 1;

testrecurse.c

@@ -646,6 +646,8 @@ static void
 initSAX(xmlParserCtxtPtr ctxt) {
     ctxt->sax->startElementNs = NULL;
     ctxt->sax->endElementNs = NULL;
+    ctxt->sax->startElement = NULL;
+    ctxt->sax->endElement = NULL;
     ctxt->sax->characters = NULL;
     ctxt->sax->cdataBlock = NULL;
     ctxt->sax->ignorableWhitespace = NULL;