Re: [HACKERS] proposal - Default namespaces for XPath expressions(PostgreSQL 11) - Mailing list pgsql-hackers

Pavel Stehule <pavel.stehule@gmail.com> writes:
> [ xml-xpath-default-ns-7.patch ]

At Andrew's prompting, I took a look over this patch.  I don't know much
of anything about XML, so I have no idea as to standards compliance here,
but I do have some comments:

* I'm fairly uncomfortable with the idea that we're going to maintain
our own XPath parser.  That seems like a recipe for lots of future
work ... and the code is far too underdocumented for anyone to actually
maintain it.  (Case in point: _transformXPath's arguments are not
documented at all, and in fact there's hardly a word about what it's
even supposed to *do*.)

* I think the business with "pgdefnamespace.pgsqlxml.internal" is just
plain awful.  It's a wart, and I don't think it's even saving you any
code once you account for all the places where you have to throw errors
for somebody trying to use that as a regular name.  This should be done
with out-of-band signaling if possible.  The existing convention above
this code is that a NULL pointer means a default namespace; can't that
be continued throughout?  If that's not practical, can you pick a string
that simply can't syntactically be a namespace name?  (In the SQL world,
for example, an empty string is an illegal identifier so that that could
be used for the purpose.  But I don't know if that applies to XML.)
Or perhaps you can build a random name that is chosen just to make it
different from any of the listed namespaces?  If none of those work,
I think passing around an explicit "isdefault" boolean alongside the name
would be better than having a reserved word.

The string used like default namespace name should be valid XML namespace name, because it is injected to XPath expression and it is passed to libxml2 as one namespace name. libxml2 requires not null valid value.