That's interesting, I didn't know. However, I think my original point still stands. If you're using PIs to make DC annotations, how would you "annotate the annotations" so to speak if nested PIs are not permitted?

Eric van der Vlist suggested:

But wouldn't that completely defeat the original point of the PIs, which was that they're "outside" the scope of validation? So what happens if I now inherit an unmodifiable schema that specifies what PIs are allowed where -- how do I add the custom new PIs necessary for my application? I think we've come full circle. :-)

My suggestion, though it's probably not practical: Specify any processing instructions in your own XML vocabulary in a separate file, linking it to the original through appropriate XPath expressions. When processing, you need to specify the original file and any additional processing instruction file(s). A smart parser could probably integrate the instructions into the primary XML stream on the fly, if desired.

Advantages: You can validate both the original document and the PI document, each to their own schema. Your PIs can take advantage of all XML features. A single document can have many sets of PIs for different purposes, even from the same namespace/vocabulary.

Disadvantages: Multiple files. Hand-editing gets ugly/impossible. Needs customized parsers to make it transparent.

-- P.