
net.sf.saxon.functions.DocumentFn Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of Saxon-HE Show documentation
Show all versions of Saxon-HE Show documentation
The XSLT and XQuery Processor
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Copyright (c) 2015 Saxonica Limited.
// This Source Code Form is subject to the terms of the Mozilla Public License, v. 2.0.
// If a copy of the MPL was not distributed with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
// This Source Code Form is "Incompatible With Secondary Licenses", as defined by the Mozilla Public License, v. 2.0.
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
package net.sf.saxon.functions;
import net.sf.saxon.Configuration;
import net.sf.saxon.Controller;
import net.sf.saxon.event.Builder;
import net.sf.saxon.event.PipelineConfiguration;
import net.sf.saxon.event.Receiver;
import net.sf.saxon.event.Sender;
import net.sf.saxon.expr.*;
import net.sf.saxon.expr.parser.Location;
import net.sf.saxon.expr.parser.PathMap;
import net.sf.saxon.expr.sort.DocumentOrderIterator;
import net.sf.saxon.expr.sort.GlobalOrderComparer;
import net.sf.saxon.lib.FeatureKeys;
import net.sf.saxon.lib.ParseOptions;
import net.sf.saxon.lib.RelativeURIResolver;
import net.sf.saxon.lib.StandardErrorHandler;
import net.sf.saxon.om.*;
import net.sf.saxon.trans.Err;
import net.sf.saxon.trans.NonDelegatingURIResolver;
import net.sf.saxon.trans.XPathException;
import net.sf.saxon.tree.tiny.Statistics;
import net.sf.saxon.tree.tiny.TinyBuilder;
import net.sf.saxon.value.Cardinality;
import net.sf.saxon.value.Whitespace;
import javax.xml.transform.Source;
import javax.xml.transform.SourceLocator;
import javax.xml.transform.TransformerException;
import javax.xml.transform.URIResolver;
import javax.xml.transform.dom.DOMSource;
import javax.xml.transform.stream.StreamSource;
import java.net.URI;
import java.net.URISyntaxException;
/**
* Implements the XSLT document() function
*/
public class DocumentFn extends SystemFunction implements Callable {
/**
* Determine the static cardinality
*/
public int getCardinality(Expression[] arguments) {
Expression expression = arguments[0];
if (Cardinality.allowsMany(expression.getCardinality())) {
return StaticProperty.ALLOWS_ZERO_OR_MORE;
} else {
return StaticProperty.ALLOWS_ZERO_OR_ONE;
}
// may have to revise this if the argument can be a list-valued element or attribute
}
/**
* Get the static properties of this expression (other than its type). The result is
* bit-signficant. These properties are used for optimizations. In general, if
* property bit is set, it is true, but if it is unset, the value is unknown.
* @param arguments
*/
public int getSpecialProperties(Expression[] arguments) {
return StaticProperty.ORDERED_NODESET |
StaticProperty.PEER_NODESET |
StaticProperty.NON_CREATIVE;
// Declaring it as a peer node-set expression avoids sorting of expressions such as
// document(XXX)/a/b/c
// The document() function might appear to be creative: but it isn't, because multiple calls
// with the same arguments will produce identical results.
}
/**
* Determine whether two expressions are equivalent
*/
// @Override
// public boolean equals(Object o) {
// return (o instanceof DocumentFn) && super.equals(o)
// && ExpressionTool.equalOrNull(getStaticBaseURIString(), ((DocumentFn) o).getStaticBaseURIString());
// }
// /**
// * iterate() handles evaluation of the function:
// * it returns a sequence of Document nodes
// */
//
// /*@NotNull*/
// public SequenceIterator iterate(XPathContext context) throws XPathException {
// int numArgs = getArity();
//
// SequenceIterator hrefSequence = getArg(0).iterate(context);
// String baseURI = null;
// if (numArgs == 2) {
// // we can trust the type checking: it must be a node
// NodeInfo base = (NodeInfo) getArg(1).evaluateItem(context);
// baseURI = base.getBaseURI();
// if (baseURI == null) {
// throw new XPathException("The second argument to document() is a node with no base URI", "XTDE1162");
// }
// }
//
// DocumentMappingFunction map = new DocumentMappingFunction(context);
// map.baseURI = baseURI;
// map.stylesheetURI = getStaticBaseURIString();
// map.locator = this;
//
// ItemMappingIterator iter = new ItemMappingIterator(hrefSequence, map);
//
// Expression expression = getArg(0);
// if (Cardinality.allowsMany(expression.getCardinality())) {
// return new DocumentOrderIterator(iter, GlobalOrderComparer.getInstance());
// // this is to make sure we eliminate duplicates: two href's might be the same
// } else {
// return iter;
// }
// }
/**
* Evaluate the expression
*
* @param context the dynamic evaluation context
* @param arguments the values of the arguments, supplied as SequenceIterators
* @return the result of the evaluation, in the form of a SequenceIterator
* @throws net.sf.saxon.trans.XPathException
* if a dynamic error occurs during the evaluation of the expression
*/
public Sequence call(XPathContext context, Sequence[] arguments) throws XPathException {
int numArgs = getArity();
SequenceIterator hrefSequence = arguments[0].iterate();
String baseURI = null;
if (numArgs == 2) {
// we can trust the type checking: it must be a node
NodeInfo base = (NodeInfo) arguments[1].head();
baseURI = base.getBaseURI();
if (baseURI == null) {
throw new XPathException("The second argument to document() is a node with no base URI", "XTDE1162");
}
}
DocumentMappingFunction map = new DocumentMappingFunction(context);
map.baseURI = baseURI;
map.stylesheetURI = getStaticBaseUriString();
//map.locator = this;
ItemMappingIterator iter = new ItemMappingIterator(hrefSequence, map);
//if (Cardinality.allowsMany(expression.getCardinality())) {
return SequenceTool.toLazySequence(new DocumentOrderIterator(iter, GlobalOrderComparer.getInstance()));
// this is to make sure we eliminate duplicates: two href's might be the same
// } else {
// return SequenceTool.toLazySequence(iter);
// }
}
private static class DocumentMappingFunction implements ItemMappingFunction {
public String baseURI;
public String stylesheetURI;
public Location locator;
public XPathContext context;
public DocumentMappingFunction(XPathContext context) {
this.context = context;
}
public Item mapItem(Item item) throws XPathException {
String b = baseURI;
if (b == null) {
if (item instanceof NodeInfo) {
b = ((NodeInfo) item).getBaseURI();
} else {
b = stylesheetURI;
}
}
try {
return makeDoc(item.getStringValue(), b, null, context, locator, false);
} catch (XPathException xerr) {
if (xerr.getErrorCodeLocalPart().equals("XTRE1160")) {
// Invalid fragment identifier: error code changes in XSLT 3.0
if (context.getController().getExecutable().isAllowXPath30()) {
xerr.setErrorCode("XTDE1160");
}
// Report a "recoverable error"
try {
context.getController().recoverableError(xerr);
} catch (XPathException err2) {
throw xerr;
}
// If recovering, the recovery action is to ignore the fragment identifier
String href = item.getStringValue();
int hash = href.indexOf('#');
href = href.substring(0, hash);
return makeDoc(href, b, null, context, locator, false);
} else if (xerr.getErrorCodeLocalPart().equals("XTDE1162")) {
// non-recoverable error
throw xerr;
} else {
// Other errors
if (!xerr.hasBeenReported()) {
xerr.maybeSetLocation(locator);
String code = (xerr.getException() instanceof URISyntaxException) ? "FODC0005" : "FODC0002";
xerr.maybeSetErrorCode(code);
try {
context.getController().recoverableError(xerr);
} catch (XPathException err2) {
throw xerr;
}
} else {
throw xerr;
}
}
return null;
}
}
}
/**
* Supporting routine to load one external document given a URI (href) and a baseURI. This is used
* in the normal case when a document is loaded at run-time (that is, when a Controller is available)
*
*
* @param href the relative URI
* @param baseURI the base URI
* @param options parse options to be used. May be null.
* @param c the dynamic XPath context
* @param locator used to identify the location of the instruction in event of error
* @param silent if true, errors should not be notified to the ErrorListener @return the root of the constructed document, or the selected element within the document
* if a fragment identifier was supplied
* @throws XPathException if reading or parsing the document fails
*/
public static NodeInfo makeDoc(String href, String baseURI, ParseOptions options, XPathContext c, Location locator, boolean silent)
throws XPathException {
Configuration config = c.getConfiguration();
// If the href contains a fragment identifier, strip it out now
//System.err.println("Entering makeDoc " + href);
int hash = href.indexOf('#');
String fragmentId = null;
if (hash >= 0) {
if (hash == href.length() - 1) {
// # sign at end - just ignore it
href = href.substring(0, hash);
} else {
fragmentId = href.substring(hash + 1);
href = href.substring(0, hash);
if (!NameChecker.isValidNCName(fragmentId)) {
XPathException de = new XPathException("The fragment identifier " + Err.wrap(fragmentId) + " is not a valid NCName");
de.setErrorCode("XTRE1160");
de.setXPathContext(c);
de.setLocator(locator);
throw de;
}
}
}
Controller controller = c.getController();
if (controller == null) {
throw new XPathException("doc() function is not available in this environment");
}
// Resolve relative URI
DocumentURI documentKey = computeDocumentKey(href, baseURI, c);
// see if the document is already loaded
TreeInfo doc = config.getGlobalDocumentPool().find(documentKey);
if (doc != null) {
return doc.getRootNode();
}
DocumentPool pool = controller.getDocumentPool();
//noinspection SynchronizationOnLocalVariableOrMethodParameter
synchronized (controller) {
doc = pool.find(documentKey);
if (doc != null) {
return getFragment(doc, fragmentId, c, locator);
}
// check that the document was not written by this transformation
if (!controller.checkUniqueOutputDestination(documentKey)) {
pool.markUnavailable(documentKey);
XPathException err = new XPathException(
"Cannot read a document that was written during the same transformation: " + documentKey);
err.setXPathContext(c);
err.setErrorCode("XTRE1500");
err.setLocator(locator);
throw err;
}
if (pool.isMarkedUnavailable(documentKey)) {
XPathException err = new XPathException(
"Document has been marked not available: " + documentKey);
err.setXPathContext(c);
err.setErrorCode("FODC0002");
err.setLocator(locator);
throw err;
}
}
try {
// Get a Source from the URIResolver
Source source = resolveURI(href, baseURI, documentKey.toString(), c);
//System.err.println("URI resolver returned " + source.getClass() + " " + source.getSystemId());
source = config.getSourceResolver().resolveSource(source, config);
//System.err.println("Resolved source " + source.getClass() + " " + source.getSystemId());
TreeInfo newdoc;
if (source instanceof NodeInfo || source instanceof DOMSource) {
NodeInfo startNode = controller.prepareInputTree(source);
newdoc = startNode.getTreeInfo();
} else {
Builder b = controller.makeBuilder();
if (b instanceof TinyBuilder) {
((TinyBuilder) b).setStatistics(Statistics.SOURCE_DOCUMENT_STATISTICS);
}
Receiver s = b;
if (options == null) {
options = new ParseOptions(b.getPipelineConfiguration().getParseOptions());
options.setStripSpace(Whitespace.XSLT);
options.setSchemaValidationMode(controller.getSchemaValidationMode());
}
if (silent) {
StandardErrorHandler eh = new StandardErrorHandler(controller.getErrorListener());
eh.setSilent(true);
options.setErrorHandler(eh);
}
if (controller.getExecutable().stripsInputTypeAnnotations()) {
s = controller.getConfiguration().getAnnotationStripper(s);
}
PathMap map = controller.getPathMapForDocumentProjection();
if (map != null) {
PathMap.PathMapRoot pathRoot = map.getRootForDocument(documentKey.toString());
if (pathRoot != null && !pathRoot.isReturnable() && !pathRoot.hasUnknownDependencies()) {
options.addFilter(config.makeDocumentProjector(pathRoot));
}
}
s.setPipelineConfiguration(b.getPipelineConfiguration());
try {
Sender.send(source, s, options);
newdoc = b.getCurrentRoot().getTreeInfo();
b.reset();
} finally {
if (options.isPleaseCloseAfterUse()) {
ParseOptions.close(source);
}
}
}
// At this point, we have built the document. But it's possible that another thread
// has built the same document and put it in the document pool. So we do another
// check on the document pool, and if this has happened, we discard the document
// we have just built and use the one from the pool instead.
//noinspection SynchronizationOnLocalVariableOrMethodParameter
synchronized (controller) {
doc = pool.find(documentKey);
if (doc != null) {
return getFragment(doc, fragmentId, c, locator);
}
controller.registerDocument(newdoc, documentKey);
controller.addUnavailableOutputDestination(documentKey);
}
return getFragment(newdoc, fragmentId, c, locator);
} catch (TransformerException err) {
pool.markUnavailable(documentKey);
XPathException xerr = XPathException.makeXPathException(err);
xerr.maybeSetLocation(locator);
String code = (err.getException() instanceof URISyntaxException) ? "FODC0005" : "FODC0002";
xerr.maybeSetErrorCode(code);
throw xerr;
}
}
/**
* Call the URIResolver to resolve a URI
*
* @param href the supplied relative URI, stripped of any fragment identifier
* @param baseURI the base URI
* @param documentKey the absolute URI if already available, or null otherwise
* @param context the dynamic context
* @return a Source representing the document to be read
* @throws XPathException
*/
public static Source resolveURI(String href, String baseURI, String documentKey, XPathContext context)
throws XPathException {
URIResolver resolver = context.getURIResolver();
Source source;
if (baseURI == null) {
try {
URI uri = new URI(href);
if (!uri.isAbsolute()) {
throw new XPathException("Relative URI passed to document() function (" + href +
"); but no base URI is available", "XTDE1162");
}
} catch (URISyntaxException e) {
throw new XPathException("Invalid URI passed to document() function: " + href, "FODC0005");
}
}
try {
if (resolver instanceof RelativeURIResolver && documentKey != null) {
source = ((RelativeURIResolver) resolver).dereference(documentKey);
} else {
source = resolver.resolve(href, baseURI);
}
} catch (Exception ex) {
//ex.printStackTrace();
XPathException de = new XPathException("Exception thrown by URIResolver", ex);
if (context.getConfiguration().getBooleanProperty(FeatureKeys.TRACE_EXTERNAL_FUNCTIONS)) {
ex.printStackTrace();
}
throw de;
}
if(source instanceof StreamSource && ((StreamSource)source).getInputStream() == null && ((StreamSource)source).getReader() == null) {
String uri = source.getSystemId();
resolver = context.getController().getStandardURIResolver();
try {
source = resolver.resolve(uri, "");
} catch (TransformerException ex) {
throw XPathException.makeXPathException(ex);
}
}
// if a user URI resolver returns null, try the standard one
// (Note, the standard URI resolver never returns null)
if (source == null && !(resolver instanceof NonDelegatingURIResolver) ) {
resolver = context.getController().getStandardURIResolver();
try {
if (resolver instanceof RelativeURIResolver && documentKey != null) {
source = ((RelativeURIResolver) resolver).dereference(documentKey);
} else {
source = resolver.resolve(href, baseURI);
}
} catch (TransformerException ex) {
throw XPathException.makeXPathException(ex);
}
}
return source;
}
/**
* Compute a document key
*/
protected static DocumentURI computeDocumentKey(String href, String baseURI, XPathContext c) throws XPathException {
// Resolve relative URI
Controller controller = c.getController();
URIResolver resolver = controller.getURIResolver();
if (resolver == null) {
resolver = controller.getStandardURIResolver();
}
return computeDocumentKey(href, baseURI, resolver);
}
/**
* Compute a document key (an absolute URI that can be used to see if a document is already loaded)
*
* @param href the relative URI
* @param baseURI the base URI
* @param resolver the URIResolver
*/
public static DocumentURI computeDocumentKey(String href, String baseURI, URIResolver resolver) throws XPathException {
String documentKey;
if (resolver instanceof RelativeURIResolver) {
// If this is the case, the URIResolver is responsible for absolutization as well as dereferencing
try {
documentKey = ((RelativeURIResolver) resolver).makeAbsolute(href, baseURI);
} catch (TransformerException e) {
documentKey = '/' + href;
}
} else {
// Saxon takes charge of absolutization, leaving the user URIResolver to handle dereferencing only
href = ResolveURI.escapeSpaces(href);
if (baseURI == null) { // no base URI available
try {
// the href might be an absolute URL
documentKey = (new URI(href)).toString();
} catch (URISyntaxException err) {
// it isn't; but the URI resolver might know how to cope
documentKey = '/' + href;
}
} else if (href.isEmpty()) {
// common case in XSLT, which java.net.URI#resolve() does not handle correctly
documentKey = baseURI;
} else {
try {
URI uri = new URI(baseURI).resolve(href);
documentKey = uri.toString();
} catch (URISyntaxException err) {
documentKey = baseURI + "/../" + href;
} catch (IllegalArgumentException err) {
documentKey = baseURI + "/../" + href;
}
}
}
return new DocumentURI(documentKey);
}
/**
* Supporting routine to load one external document given a URI (href) and a baseURI. This is used
* when the document is pre-loaded at compile time.
*
* @param href the relative URI. This must not contain a fragment identifier
* @param baseURI the base URI
* @param config the Saxon configuration
* @param locator used to identify the location of the instruction in event of error. May be null.
* @return the root of the constructed document, or the selected element within the document
* if a fragment identifier was supplied
*/
public static NodeInfo preLoadDoc(String href, String baseURI, Configuration config, SourceLocator locator)
throws XPathException {
int hash = href.indexOf('#');
if (hash >= 0) {
throw new XPathException("Fragment identifier not supported for preloaded documents");
}
// Resolve relative URI
String documentKey;
URIResolver resolver = config.getURIResolver();
if (resolver instanceof RelativeURIResolver) {
try {
documentKey = ((RelativeURIResolver) resolver).makeAbsolute(href, baseURI);
} catch (TransformerException e) {
documentKey = '/' + href;
baseURI = "";
}
} else {
if (baseURI == null) { // no base URI available
try {
// the href might be an absolute URL
documentKey = new URI(href).toString();
} catch (URISyntaxException err) {
// it isn't; but the URI resolver might know how to cope
documentKey = '/' + href;
baseURI = "";
}
} else if (href.isEmpty()) {
// common case in XSLT, which java.net.URI#resolve() does not handle correctly
documentKey = baseURI;
} else {
try {
URI uri = new URI(baseURI).resolve(href);
documentKey = uri.toString();
} catch (URISyntaxException err) {
documentKey = baseURI + "/../" + href;
} catch (IllegalArgumentException err) {
documentKey = baseURI + "/../" + href;
}
}
}
// see if the document is already loaded
TreeInfo doc = config.getGlobalDocumentPool().find(documentKey);
if (doc != null) {
return doc.getRootNode();
}
try {
// Get a Source from the URIResolver
URIResolver r = resolver;
Source source = null;
if (r != null) {
try {
source = r.resolve(href, baseURI);
} catch (Exception ex) {
XPathException de = new XPathException("Exception thrown by URIResolver", ex);
if (config.getBooleanProperty(FeatureKeys.TRACE_EXTERNAL_FUNCTIONS)) {
ex.printStackTrace();
}
de.setLocator(locator);
throw de;
}
}
// if a user URI resolver returns null, try the standard one
// (Note, the standard URI resolver never returns null)
if (source == null && !(r instanceof NonDelegatingURIResolver)) {
r = config.getSystemURIResolver();
source = r.resolve(href, baseURI);
}
//System.err.println("URI resolver returned " + source.getClass() + " " + source.getSystemId());
source = config.getSourceResolver().resolveSource(source, config);
//System.err.println("Resolved source " + source.getClass() + " " + source.getSystemId());
TreeInfo newdoc = config.buildDocumentTree(source);
config.getGlobalDocumentPool().add(newdoc, documentKey);
return newdoc.getRootNode();
} catch (TransformerException err) {
XPathException xerr = XPathException.makeXPathException(err);
xerr.setLocator(locator);
xerr.setErrorCode("FODC0002");
throw new XPathException(err);
}
}
/**
* Copy the documents identified by this expression to a given Receiver. This method is used only when it is
* known that the documents are being copied, because there is then no problem about node identity.
*
* @param context the XPath dynamic context
* @param args the arguments to the document() function
* @param out the destination to which the documents will be sent
*/
public void sendDocuments(XPathContext context, Expression[] args, Receiver out) throws XPathException {
SequenceIterator hrefSequence = args[0].iterate(context);
String explicitBaseURI = null;
if (getArity() == 2) {
// we can trust the type checking: it must be a node
NodeInfo base = (NodeInfo) args[1].evaluateItem(context);
explicitBaseURI = base.getBaseURI();
}
ParseOptions options = new ParseOptions();
while (true) {
Item href = hrefSequence.next();
if (href == null) {
break;
}
String base;
if (explicitBaseURI == null) {
if (href instanceof NodeInfo) {
base = ((NodeInfo) href).getBaseURI();
} else {
base = getStaticBaseUriString();
}
} else {
base = explicitBaseURI;
}
sendDoc(href.getStringValue(), base, context, null, out, options);
}
}
/**
* Supporting routine to push one external document given a URI (href) and a baseURI to a given Receiver.
* This method cannot handle fragment identifiers
*
* @param href the relative URI
* @param baseURL the base URI
* @param c the XPath dynamic context
* @param locator used to identify the lcoation of the instruction in case of error
* @param out the destination where the document is to be sent
*/
public static void sendDoc(String href, String baseURL, XPathContext c,
Location locator, Receiver out,
ParseOptions parseOptions) throws XPathException {
PipelineConfiguration pipe = out.getPipelineConfiguration();
if (pipe == null) {
pipe = c.getController().makePipelineConfiguration();
out.setPipelineConfiguration(pipe);
}
// Resolve relative URI
String documentKey;
if (baseURL == null) { // no base URI available
try {
// the href might be an absolute URL
documentKey = new URI(href).toString();
} catch (URISyntaxException err) {
// it isn't; but the URI resolver might know how to cope
documentKey = '/' + href;
baseURL = "";
}
} else if (href.isEmpty()) {
// common case in XSLT, which java.net.URI#resolve() does not handle correctly
documentKey = baseURL;
} else {
try {
URI url = new URI(baseURL).resolve(href);
documentKey = url.toString();
} catch (URISyntaxException err) {
documentKey = baseURL + "/../" + href;
} catch (IllegalArgumentException err) {
documentKey = baseURL + "/../" + href;
}
}
Controller controller = c.getController();
// see if the document is already loaded
TreeInfo doc = controller.getDocumentPool().find(documentKey);
Source source = null;
if (doc != null) {
source = doc.getRootNode();
} else {
try {
// Get a Source from the URIResolver
URIResolver r = controller.getURIResolver();
if (r != null) {
source = r.resolve(href, baseURL);
}
// if a user URI resolver returns null, try the standard one
// (Note, the standard URI resolver never returns null)
if (source == null) {
r = controller.getStandardURIResolver();
source = r.resolve(href, baseURL);
}
if (source instanceof NodeInfo || source instanceof DOMSource) {
NodeInfo startNode = controller.prepareInputTree(source);
source = startNode.getRoot();
}
} catch (TransformerException err) {
XPathException xerr = XPathException.makeXPathException(err);
xerr.setLocator(locator);
xerr.maybeSetErrorCode("FODC0005");
throw xerr;
}
}
ParseOptions options = new ParseOptions(parseOptions);
options.setStripSpace(Whitespace.XSLT);
if (controller.getExecutable().stripsInputTypeAnnotations()) {
out = controller.getConfiguration().getAnnotationStripper(out);
}
if (controller.getConfiguration().isTiming()) {
controller.getConfiguration().getLogger().info("Streaming input document " + source.getSystemId());
}
out.setPipelineConfiguration(pipe);
try {
Sender.send(source, out, options);
} catch (XPathException e) {
e.maybeSetLocation(locator);
e.maybeSetErrorCode("FODC0002");
throw e;
}
}
/**
* Resolve the fragment identifier within a URI Reference.
* Only "bare names" XPointers are recognized, that is, a fragment identifier
* that matches an ID attribute value within the target document.
*
* @param doc the document node
* @param fragmentId the fragment identifier (an ID value within the document)
* @param context the XPath dynamic context
* @return the element within the supplied document that matches the
* given id value; or null if no such element is found.
*/
private static NodeInfo getFragment(TreeInfo doc, String fragmentId, XPathContext context, SourceLocator locator)
throws XPathException {
// TODO: we only support one kind of fragment identifier. The rules say
// that the interpretation of the fragment identifier depends on media type,
// but we aren't getting the media type from the URIResolver.
if (fragmentId == null) {
return doc.getRootNode();
}
if (!NameChecker.isValidNCName(fragmentId)) {
XPathException err = new XPathException("Invalid fragment identifier in URI");
err.setXPathContext(context);
err.setErrorCode("XTRE1160");
err.setLocator(locator);
try {
context.getController().recoverableError(err);
} catch (XPathException dynamicError) {
throw err;
}
return doc.getRootNode();
}
return doc.selectID(fragmentId, false);
}
}
© 2015 - 2025 Weber Informatics LLC | Privacy Policy