Saxon

Revisiting this, debugging the unit test on SaxonJ 10.6.

First observation is that the whole validation pipeline is duplicated - we are actually doing the validation twice.

The first DocumentValidator is created by SchemaValidatorImpl#692; the second by Sender.send() at line 121.

I have fixed this somewhat pragmatically by registering the DocumentValidator as a component on the PipelineConfiguration, and changing Sender.send() so it doesn't create a new DocumentValidator if there is already one on the pipeline.

Next observation is that the (changing) systemIDs on element nodes are actually being passed through to the target document correctly; the test is failing because the systemId on the document node is wrong.

Having some trouble with this.

If we set a base URI (or system ID) on the builder used to construct the result tree (validated tree), then it is used for all nodes in that tree. If we don't, then element nodes are getting the correct base URI set, but the document node isn't -- because in the Receiver interface, the startDocument event doesn't have a Location argument.

The problem with the setting useEventLocation=true is that the startDocument event has no event location.

I have finally found a solution to this as follows.

(a) When creating a LocationCopier, supply a system ID.

(b) In TinyBuilder.open(), if there is a LocationCopier on the PipelineConfiguration, use its systemId as the systemId of the TinyBuilder

(c) When creating a LocationCopier in Sender.sendDocumentInfo(), set the systemId on the LocationCopier.

This basically compensates for the fact that the Receiver.startDocument() call does not have a Location argument, and therefore has no direct way of transmitting the base URI of the document node; instead we pass it indirectly via the LocationCopier.

The solution doesn't work "as is" on the 11 branch, because the CopyInformee on the PipelineConfiguration is now a simple lambda expression.

Leaving bug as resolved until fix applied to the Saxon 10 maintenance release.