Saxon

The remaining failures are the test cases on-empty-106X and on-empty-107X. There seems to be a problem here that is only indirectly related to tracing. When on-empty-106 is run without tracing, we get a warning:

Warning on line 6 column 43 of on-empty-106.xsl: SXWN9000 The result of the sequence constructor will never be empty, so xsl:on-empty instructions will never be evaluated, and xsl:on-non-empty instructions will always be evaluated

This means that the optimiser has basically made the decisions at compile time, which in turn means that the run-time logic is not being exercised by the test. The run-time logic looks incorrect, and the bug would be equally exposed by a test that put xsl:attribute inside a conditional rather than a trace instruction.

I have created tests on-empty-106e and on-empty-107e to demonstrate this: these are failing even without tracing enabled.

Fixed SignificantItemDetector so that calls on attribute() and namespace() invoke start() to indicate that non-empty content has been found.

With this change all tests in test set on-empty pass both with and without the -T option.