XML Schema 1.1 assert constraints diverge when referenced elements are absent

Answer

The divergence is caused by two factors: the XPath 2.0 evaluation context is defined as the element or attribute the xs:assert decorates, and the specification does not normatively prescribe how a dynamic XPath error (e.g., trying to access a missing node) should affect validation.

Confirmed facts

• The XSD 1.1 Part 1 specification states that the context item for an assertion is the element or attribute to which the assertion is attached, not the document root.
• The spec defines an assertion failure only when the expression evaluates to false or an empty sequence; it is silent on errors raised during evaluation of the XPath expression itself.
• Implementations such as Xerces and Saxon may treat any XPath runtime error as a fatal validation error, while others may suppress the error and continue, leading to inconsistent results.

Likely explanation

Because the standard leaves error handling unspecified, processors are free to map a dynamic XPath error to a validation failure, to ignore it, or to raise an internal exception. This freedom explains why some validators report a failure when an optional element is absent, whereas others silently accept the schema.

Steps to verify behavior

1. Create a minimal XSD that contains an xs:assert referencing an optional child element or attribute.
2. Prepare two XML instances: one where the optional element is present and one where it is absent.
3. Validate each instance with Xerces, Saxon, and any other XSD 1.1 processor you intend to support, recording whether the processor reports a validation error, treats the assert as failed, or continues without error.
4. Repeat the test with an assert that uses a function known to raise a dynamic error on an empty sequence (e.g., count(//b)) and with a guarded version that checks exists() first.
5. Compare the outcomes to see how each processor handles the error and whether the evaluation context is the decorated element.

Recommendation for schema authors

To obtain consistent results, avoid XPath expressions that can raise dynamic errors on missing nodes. Use explicit existence checks such as not(exists(b)) or test="not(exists(b)) or b/@x = 'y'" instead of bare references. This defensive style works across processors because empty sequences are treated as false rather than as errors.