saxon performance

[Vilo Repan]

I use saxon in cooperation with xsom parser. Yesterday I was profiling my
application and I found some performance issues in saxon. In method
net.sf.saxon.dom.DOMObjectModel.getDocumentBuilder(Result) code always create
DocumentBuilderFactory and DocumentBuilder object and then docBuilder =
dfactory.newDocumentBuilder();

It take 40-50% of time to parse xsd schema I tried to move it to a static
block, and create final class member DocumentBuilder like this:

        private static final DocumentBuilder docBuilder;

    static {

        try {

            DocumentBuilderFactory dfactory = DocumentBuilderFactory.newInstance();

            docBuilder = dfactory.newDocumentBuilder();

                } catch (ParserConfigurationException e) {

                        throw new RuntimeException("Couldn't create document builder, reason: " + e.getMessage(), e);

                }

    }

then in getDocumentBuilder(Result) I only call

Document out = docBuilder.newDocument();

with that change I made about 40-50% faster xsd schema parsing. I use saxon as
maven dependency, now as private version. It would be great if this fix,
properly tested could be in next scheduled release...

Can somebody have a look at this?

Thanks.

[Michael Kay]

I don't know the internals of DSOM, and I'm not sure how much control you have
over the way Saxon is used. Saxon with DOM is not a good combination; there
are many inefficiencies. It's much better (by a factor of 10) to use Saxon's
internal tree model. If you do want Saxon's output as a DOMResult, it's much
better to initialise the DOMResult with a Document node and allow Saxon to
build the tree underneath this DOMResult; if you don't do this, Saxon has to
find an implementation of DOM, and it uses the JAXP DocumentBuilderFactory
mechanism to do this (it doesn't include its own DOM implementation).

Although this path is always going to be very inefficient, you are right to
point out that if people do multiple transformations, we should only need to
do the search for a DocumentBuilderFactory once, and I will make this change.

[Vilo Repan]

AFAIK I don't have any control over xsom or saxon, but I'll check it anyway.
Thanks for that change.
When can I expect that change will be in maven repository?

Thanks.

[Michael Kay]

Saxon is not distributed via Maven. We produce periodic maintenance releases
published here on SourceForge. I'm not planning to put this change into a
maintenance release - the general policy is to treat performance improvements
as enhancements rather than bug fixes and only issue them in the next major
release (i.e. 9.4) - which will be ready when it's ready.

[Vilo Repan]

Thanks for quick responses.