Re: PostgreSQL vs SQL/XML Standards

[Markus Winand]

> On 2018-11-6, at 15:23 , Pavel Stehule <pavel.steh...@gmail.com> wrote:
> 
> 
> 
> po 29. 10. 2018 v 11:45 odesílatel Pavel Stehule <pavel.steh...@gmail.com> 
> napsal:
> 
> 
> po 29. 10. 2018 v 10:11 odesílatel Pavel Stehule <pavel.steh...@gmail.com> 
> napsal:
> Hi
> 
> čt 25. 10. 2018 v 21:47 odesílatel Alvaro Herrera <alvhe...@2ndquadrant.com> 
> napsal:
> On 2018-Oct-25, Pavel Stehule wrote:
> 
> > I am thinking so I can fix some issues related to XMLTABLE. Please, send me
> > more examples and test cases.
> 
> Please see Markus Winand's patch that I referenced upthread.
> 
> here is a fix of some XMLTABLE mentioned issues.
> 
> this update allows cast boolean to numeric types from XPath expressions
> 
> Attached patch solves some cast issues mentioned by Chap. It solves issue 
> reported by Markus. I didn't use Markus's code, but it was inspiration for 
> me. I found native solution from libxml2.
> 
> Regards
> 
> Pavel

Better than my patch.

But I think the chunk in xml_xmlnodetoxmltype of my patch is still needed — in 
one way or the other (see below).

# select * from xmltable('*' PASSING '<e>pre<!--c1--><?pi 
arg?><![CDATA[&ent1]]><n2>&amp;deep</n2>post</e>' COLUMNS x XML PATH 'node()');
                    x
-----------------------------------------
 prec1arg&amp;ent1<n2>&amp;deep</n2>post
(1 row)

Output is not the original XML.

I dug a little further and found another case that doesn’t looks right even 
with my change to xml_xmlnodetoxmltype applied:

# select * from xmltable('*' PASSING '<e>pre<!--c1--><?pi 
arg?><![CDATA[&ent1]]><n2>&amp;deep</n2>post</e>' COLUMNS x XML PATH '/');
             x
---------------------------
 pre&amp;ent1&amp;deeppost
(1 row)

Oracle gives in both cases XML.

To fix that I included XML_DOCUMENT_NODE in the list of nodes that use 
xmlNodeDump. Now I wonder if that logic should be reversed to use the 
xmlXPathCastNodeToString branch in a few selected cases but default to the 
branch xmlNodeDump for all other cases?

I guess those few cases might be XML_ATTRIBUTE_NODE and XML_TEXT_NODE. 
Regression tests are happy with that approach but I don’t think that proves a 
lot.

-markus

diff --git a/src/backend/utils/adt/xml.c b/src/backend/utils/adt/xml.c
index 37d85f7..7c1f884 100644
--- a/src/backend/utils/adt/xml.c
+++ b/src/backend/utils/adt/xml.c
@@ -3682,7 +3682,7 @@ xml_xmlnodetoxmltype(xmlNodePtr cur, PgXmlErrorContext 
*xmlerrcxt)
 {
        xmltype    *result;

-       if (cur->type == XML_ELEMENT_NODE)
+       if (cur->type != XML_ATTRIBUTE_NODE && cur->type != XML_TEXT_NODE)
        {
                xmlBufferPtr buf;
                xmlNodePtr      cur_copy;