Re: [Classpathx-discuss] SAX conformance tests from Elliotte Rusty Harold

[David Brownell]

Maarten Coene wrote:
> Hi,
>
> Perhaps you already knew, but here is an interesting paper:
> http://www.idealliance.org/papers/dx_xmle04/papers/03-06-02/03-06-02.html
>
> it compares different SAX parsers, including the GNU JAXP parser.

Interesting.  Something broke badly in the GNU JAXP stuff; I
know the initial version didn't have those problems.  It passed
comfortably over 95% of the NIST tests (vs 46% as he reported),
with the main gaps bing the ones that cared about which Unicode
characters were allowed.  (Since the XML spec had an armload of
exceptions to its basic rules, which were changing along with
Java's own rules about Unicode classification.)

However I know the NIST folk did make some changes to how they
interpreted the XML spec, so maybe that caused some of those
problems too.  That work went so slow for a long time that I
thought it was dead.  (Much like the SAX conformance work for
Xerces ... it's good to know some of the failures have now
been fixed, but they're hardly new...)

I'm not quite sure what to make of some of his comments
though.  Among other things, the reason there's been not
a lot of need for SAX conformance tests is that when the
OUTPUT (!) test cases on the NIST suite are included,
there's really quite high coverage on basic behavior of
SAX parsers.  That small (~100+ testcases with jUnit)
test suite, misattributed (I wrote it!) covered most of
the other APIs.  It seems that he didn't consider those.

It's also LONG been the convention that RuntimeError is
never specified explicitly in calling conventions, so it
can always be thrown.  I'd agree that's the wrong thing
to do given malformed XML, so it's a bug; but his statement
was even stronger ("not legal to throw them, ever") and
called those conformance bugs.

- Dave