Saxon

Confirmed that with 9.9, the expected name="" attribute is present on line 400 of the output, and with 10.2 it is absent; and that there are other similar differences in the output files. Which suggests, prima facie, that one of the releases is wrong (unless of course it turns out that the code is dependent on something that's not well-defined, like order of attributes).

This is a complex stylesheet and working out what it does, and where the difference arises, is a pretty large undertaking. Is there any chance of trying to reduce it to a simpler repro? Without having any idea what the data means or how it is being processed, this is quite hard to do. As I mention, there are legitimate reasons why two different processors can give different output.

The stylesheet is using the xsl:where-populated instruction, and the implementation of this instruction changed in 10.0 so it has both pull-mode and push-mode implementations (in 9.9 it was always pull mode). Since the semantics of xsl:where-populated examine whether an attribute value is zero-length, this certainly seems like an area to investigate.

Confirmed that in 10.2 the WherePopulatedOutputter is suppressing the copying of an empty name attribute. Now we need to work out whether it is correct to do so...

No, the logic of WherePopulatedOutputter.attribute() is wrong. It should only suppress an empty attribute when level==0, that is, when an instruction in the sequence constructor contained within xsl:where-populated directly delivers a parentless attribute node. It should not suppress empty attributes deeper in the tree.

I was no longer online yesterday, but apparently - despite the somewhat complex xslt - you were able to find the cause. Thanks for the quick result!

Bug fix applied in the Saxon 10.3 maintenance release