Support XML Catalog Files in xsl:import in fn:transform()

[liamquin]

It turns out that the patch for issue #1405 hsa a rather obscure but very important problem - it doesn't work for xsl:import.

This turns out to be because Saxon (and possibly other XSLT implementations) only use the XML Catalog URI resolver for xsl:import when stylesheets are first compiled and then executed.

I hacked XsltTransform.java Transformer.transformer() to look like this,
Transformer tr = null;
TransformerFactory tf = TransformerFactory.newInstance();
if(cw != null) tf.setURIResolver(cw.getURIResolver());
Templates compiled = tf.newTemplates(ss);
return compiled.newTransformer();

This works wonderfully EXCEPT that there's no longer a transformer cache. To keep a cache, we'd either need to make sure the XSLT stylesheet was the same each time or to save just the result of TransformerFactory.newInstance(); is that worth it?

I didn't append a patch because of that question. It would also be possible to add a "basex:compile-stylesheet" boolean option to xslt:transform() to control the behaviour - being as BaseX is a database server i'm concerned about performance, but the performane characteristics will probably depend on the size and complexity of the stylesheet, and how much of it is used.

What do you think?

[gimsieke]

Thanks for investigating, Liam! Our usage scenarios often involve applying the same XSLT stylesheet to thousands of documents. Of course performance is always an issue. Therefore we’d like to see an option by which we can control the caching of compiled stylesheets that are also able to use the catalog for imports. Or rather, we actually don’t want to control the caching, we’d like to have it by default.

Let’s wait for Christian’s thoughts; I think he’s currently on a well-deserved vacation.

[ChristianGruen]

Thanks both for your feedback!

Based on Liam’s suggestions, I have committed a new version that:

• caches Templates instead of Transformer objects (I assume this is cleaner anyway);
• assigns the catalog URI resolver to both the Templates and the Transformer object (as it seems that each XSLT implementation seems to consider the resolver at a different stage).

Could you please try the new snapshot for me? I’d particularly be interested if we lose speed by caching the Templates (which are used in the next step to create Transformer instances).

[ChristianGruen]

Hi Liam, hi Gerrit, I hope the updated version matches your requirements. I’ll close this issue; feel free to reopen it.

[gimsieke]

Hi Christian, I was just about to test it.
I can confirm that it works, specifying for example
declare option db:catfile "path/relative/to/cwd/catalog.xml";.
Thank you both!

[ChristianGruen]

Great news, thanks Gerrit!

[liamquin]

On Mon, 2019-09-23 at 11:14 -0700, Gerrit Imsieke wrote: Hi Christian, I was just about to test it. I can confirm that it works, specifying for example `declare option db:catfile "path/relative/to/cwd/catalog.xml";`. Thank you both!

That's great! Thanks

-- Liam Quin, https://www.delightfulcomputing.com/ Available for XML/Document/Information Architecture/XSLT/ XSL/XQuery/Web/Text Processing/A11Y training, work & consulting. Barefoot Web-slave, antique illustrations: http://www.fromoldbooks.org