Schema-Aware XSLT from the Command Line

To run a schema-aware transformation from the command line, use the com.saxonica.Transform command instead of the usual net.sf.saxon.Transform. This has an additional option -val:strict to request strict validation of the source document, or -val:lax for lax validation. This applies not only to the principal source document loaded from the command line, but to all documents loaded via the doc() and document() functions.

The schemas to be used to validate these source documents can be specified either by using the xsl:import-schema declaration in the stylesheet, or using xsi:schemaLocation (or xsi:noNamespaceSchemaLocation) attributes within the source documents themselves, or by using the -xsd option on the command line.

Validating the source document has several effects. Most obviously, it will cause the transformation to fail if the document is invalid. It will also cause default values for attributes and elements to be expanded, so they will appear to the stylesheet as if they were present on the source document. In addition, element and attribute nodes that have been validated will be annotated with a type. This enables operations to be performed in a type-safe way. This may cause error messages, for example if you try to use an xs:decimal value as an argument to a function that expects a string. It may also cause some operations to produce different results: for example when using elements or attributes that have been given a list type in the schema, the typed value of the node will appear in the stylesheet as a sequence rather than as a single string value.

Saxon-EE also allows you to validate result documents (both final result documents and temporary trees), using the validation and type attributes. For details of these, refer to the XSLT 2.0 specification. Validation of result documents is done on-the-fly, so if the stylesheet attempts to produce invalid output, you will usually get an error message that identifies the offending instruction in the stylesheet. Type annotations on final result documents are lost if you send the output to a standard JAXP Result object (whether it's a StreamResult, SAXResult, or DOMResult, but they remain available if you capture the output in a Saxon Receiver or in a DOMResult that encapsulates a Saxon NodeInfo. For details of the way in which type annotations are represented in the Saxon implementation of the data model, see the JavaDoc documentation. The getSchemaType() method on a NodeInfo object returns a SchemaType object representing the type.

The -outval:recover option on the command line causes validation errors encountered in processing a final result tree to be treated as warnings, allowing processing to continue. This allows more than one error to be reported in a single run. The result document is serialized as if validation were successful, but with XML comments inserted to show where the validation errors were found. This option does not necessarily recover from all validation errors, for example at present it does not recover from errors in uniqueness or referential constraints. It applies only to result trees validated using the validation attribute of xsl:result-document.

With the schema-aware version of Saxon, type declarations (the as attribute on elements such as xsl:function, xsl:variable, and xsl:param) can refer to schema-defined types, for example you can write <xsl:variable name="a" as="schema-element(ipo:invoice)"/>. You can also use the element() and attribute() tests to select nodes by their schema type in path expressions and match patterns.

Saxon does a certain amount of static analysis of the XSLT and XPath code based on schema information. For example, if a template rule is defined with a match pattern such as match="schema-element(invoice)", then it will check any path expressions used in the template rule to ensure that they are valid against the schema when starting from invoice as the context node. Similarly, if the result type of a template rule or function is declared using an as attribute, then Saxon will check any literal result elements in the body of the template or function to ensure that they are consistent with this declared type. This analysis can reveal many simple user errors at compile time that would otherwise result in run-time errors or simply in incorrect output. But this is only possible if the source code explicitly declares the types of parameters, template and function results, and match patterns.