Re: BUG #8469: Xpath behaviour unintuitive / arguably wrong

On 2013-10-05 Broce Monjian wrote:
> On Fri, Oct 4, 2013 at 10:20:46PM +0200, Stefan Kaltenbrunner wrote:
> > > > http://www.postgresql.org/message-id/201106291934.23089.rsmogura@softperience.eu
> > >
> > > There are two other similar bug reports on this from February and March
> > > of this year:
> > >
> > > http://www.postgresql.org/message-id/E1U1FKL-0002rD-RO@wrihigleys.postgresql.org
> > I think that should be:
> > http://www.postgresql.org/message-id/E1U1FKL-0002rD-RO@wrigleys.postgresql.org

> > > http://www.postgresql.org/message-id/E1UHyUw-0001oj-HE@wrigleys.postgresql.org

> > > Someone who knows XML needs to take leadership on this and propose a
> > > patch.

> Added to TODO.

I'm the one who submitted the patch which changed XPATH's behaviour. While I agree
that some might find it counter-intuitive for an XPATH expression like //node/text()
to return quoted output, there's really no way around that given the way the postgres
type system works.

A general XPATH function needs to return XML[], not TEXT[], since an XPATH can not
only select text and attribute nodes, but whole XML subtrees. But if it returns XML,
it NEEDS to return something that's actually valid content for a value of type XML,
not plain test masquerading as XML. Otherwise, basic guarantees are violated, like
for example that xmlin(xmlout(value)) yields value (and not an error).

So unfortunate as it may be, I really see no way around the current behaviour. We
could (and should) however

* Provide an way to actually unescape text contained in XML fields. I planned to
post a followup patch doing that when I submitted the patch for XPATH, but then
stuff came up and… oh well… :-(

* Add a big, fat warning to the docs about this.

best regards,
Florian Pflug

PS: I'm not subscribed to pgsql-bugs, so please keep me on the CC list