Re: Testable assertion tagging for W3C specifications

Alex Rousskov writes:
>From technical point of view, "pointing to normative sentence in a
>Recommendation" does not imply a need for structure or W3C
>documentation standard. An external document can point to normative
>sentences using a variety of already available techniques, which will
>depend on the format of the Recommendation and on the test tool
>preferences.

>In the extreme case, one can use byte offsets and a simple rendering
>engine to highlight relevant citations. In many cases, approaches
>like XPath/XPointer could be used....

We've already tried this with the first wave of XSLT/XPath tests
being released through OASIS. See
http://www.oasis-open.org/committees/xslt/index.shtml
for the overview. In my opinion, and I'm the person who has done
more work on these XPointer-style references than anyone, it is a
lot harder than you suggest. You want to be able to query the test
suite in ways like:
Find all the xsl:number tests where level="any" and count is
defaulted.
This query can only work on the XPointers if everyone who encodes
their test cases uses *exactly the same* pointer to mean "count is
defaulted" and so forth. The pointers look like
id(number)/ulist[1]/item[2]/p[1]/text()[1]
but it isn't always obvious that text()[1] is the one to use, since
further strings up to, say, text()[7] may refer to the same aspect
of the count attribute. Of course, other sentences in other p and
ulist elements may also appear to be the ones that best express
that aspect of count. I could rave on, but you get the idea.

Scott is suggesting that we experiment with XMLSpec or generally
come from that direction. As you can see above, we have already
experimented with retro-fitting pointers onto a document (XSLT 1.0
Rec) that was not written to isolate testable sentences, and it
was basically too much work for too little benefit.
.................David Marston

Received on Monday, 27 May 2002 22:15:55 UTC