Secondary output

The term secondary output refers to values produced using the xsl:result-document instruction. As with the principal output, applications can control both the destination of secondary output, and its format.

When running from the command line, secondary outputs are always serialized, using the serialization parameters supplied on the xsl:result-document instruction itself, or in the xsl:output declaration referenced in its format attribute. Serialization parameters supplied on the command line (for example !indent=yes) only take effect if the xsl:result-document chooses to use the default output format.

By default, secondary outputs are written to a file whose name is established by resolving the relative URI given in the @href attribute of xsl:result-document against the base output URI of the transformation. The base output URI defaults to the destination of the principal output.

The destination and format of secondary outputs can be influenced using the Java or C# API:

• The JAXP interface was designed for XSLT 1.0 and it has never been updated to handle secondary results. Applications that produce multiple results should therefore use s9api.

• Various Java interfaces, such as the s9api Xslt30Transformer, have a method setResultDocumentHandler() which accepts a Java Function<URI, Destination>, that is, a function that accepts an absolute URI and returns a Destination for the output. For example, you might write:

  transformer.setResultDocumentHandler(uri -> processor.newSerializer(....));

  When writing to the secondary result document is complete, Saxon will call the method Destination.close(), whose effect depends on the kind of destination. For example, in the case of a Serializer, the close() method will close any input stream that Saxon created, but not one that the user created, unless the method Serializer.setCloseOnCompletion() is called to request this.

  You can also call the method Destination.onClose() to supply an application-specific action to be performed when the destination is closed. For example, if the URI is a mailto: URI, you might want to attach the serialized result to an email message and send it to a nominated recipient.

  The older OutputURIResolver mechanism is retained for backwards compatibility. This is limited by the fact that it returns a JAXP Result object, which doesn't allow, for example, for returning a "raw" result in which the XSLT results are not wrapped in a document node.

• Various C# interfaces, such as the Saxon.Api Xslt30Transformer, have a property ResultDocumentHandler whose value is a delegate that accepts a relative URI and base URI, and returns an IDestination for the output. For example, you might write:

  transformer.ResultDocumentHandler = (href, base) => processor.NewSerializer(....);

  You can set the property Destination.OnClose to supply an application-specific action to be performed when the destination is closed. For example, if the URI is a mailto: URI, you might want to attach the serialized result to an email message and send it to a nominated recipient.

In the case of an xsl:result-document instruction with no href attribute, the output is sent to the principal output destination, which must not then be used for anything else. The main reason for this is to allow serialization options to be set dynamically.

Because xsl:result-document is an instruction that by its nature has side-effects, which runs against the grain of XSLT's pure functional language philosophy, the XSLT specification defines a number of rules constraining what the instruction can do. Notably (a) you can't write more than one output to the same URI, and (b) you can't write to a URI that the transformation also reads from. Using a ResultDocumentHandler, it's very easy to circumvent these rules (for example by using URIs containing an arbitrary query part, and having the result handler drop the query part), but the consequences might be unpredictable behaviour if instructions in the stylesheet are executed in an unexpected order.

Saxon-EE evaluates the xsl:result-document instruction asynchronously, unless multi-threading is explicitly disabled. This means that the transformation might be building several output files in parallel. This will cause problems if the stylesheet uses extension functions that have side-effects.