Testing is done. I've found the bug and a workaround.
The XPath expression gets compiled in
JAXPXPathExpression.afterPropertiesSet() as long as
JAXPXPathExpression.xPathExpression is NULL. Since
JAXPXPathExpression.afterPropertiesSet() gets called from the constructor
JAXPXPathExpression(String xpath) any further calls to
JAXPXPathExpression.set... don't change the compiled XPath expression. This
is the case, if you call JAXPXPathExpression.setNamespaceContext() and Xalan
won't know about the namespaces, when it evaluates the XPath expression.
The workaround is: Don't use the constructor JAXPXPathExpression(String
xpath). Use the default constructor and call the setter methods. Afterwards
make sure, that JAXPXPathExpression.afterPropertiesSet() gets called (by
Spring or by calling it directly). If you want to change any attribute, that
is used in the XPath expression evaluation, of an instance of
JAXPXPathExpression after construction, set the attribute
JAXPXPathExpression.xPathExpression to NULL, set other attributes and
finally call JAXPXPathExpression.afterPropertiesSet().
With these changes the ContentBasedRouter works correctly, as long as you
are not using non US-ASCII characters like 'ü' or 'ä'. That's the second
problem and I'm afraid this is a major problem when you're using ServiceMix
on non-english systems.
The problems comes from the class
org.apache.servicemix.jbi.jaxp.StringSource and its method
StringSource.getInputStream() that's used very often in ServiceMix (e.g.
JAXPXPathExpression):
public InputStream getInputStream() {
return new ByteArrayInputStream(text.getBytes());
}
According to the JDK documentation, the method String.getBytes() uses the
default character set of the system to transform/encode the characters of a
string to a byte array. On my system the default character set is "cp1252".
Every caller of StringSource.getInputStream(), that has to decode the bytes
from the input stream back to a string, has to use the default character
set. I'm quite sure, that there is a lot of code in ServiceMix (see
SourceTransformer), that doesn't set the character encoding. Unfortunately
I've no workaround for this problem. Avoid to use
StringSource.getInputStream(). Use JDK's reader implementations.
Kind regards
Jürgen
--
View this message in context:
http://www.nabble.com/NamespaceContext-in-XPath-based-content-router-t1512305.html#a4119455
Sent from the ServiceMix - User forum at Nabble.com.
