Re: [xml] Problem with data in interleave in RELAX NG validation

[Nikolai Weibull <now disu se>]

Hi!

OK, I managed to decode it somewhat.  The issue seems to be that 
we build groups of what can be matched by the interleave, but that 
these groups don’t include data, list, and value elements, only 
element and text elements.  This patch extends 
xmlRelaxNGGetElements so that it can return these elements for us 
in xmlRelaxNGComputeInterleaves.  Then we make sure to 
updatexmlRelaxNGNodeMatchesList as well so that it accepts the 
correct types.

The testsuite passes and my test below does as well.

I’m a bit surprised that interleaves simply wouldn’t allow for 
data, list, and value elements previously, so I’m wondering if 
there was a reason for the code to be the way it was and that the 
fix should be placed somewhere else or if it was simply an 
oversight.  Either way, this does seem to be the correct solution. 
If someone could confirm that this solution is what we’re looking 
for, I’ll add some proper test cases and apply another merge 
request on git.gnome.org.

Best regards,
 Nikolai

---
relaxng.c | 16 ++++++++++++----
1 file changed, 12 insertions(+), 4 deletions(-)

diff --git a/relaxng.c b/relaxng.c
index 8306e546..3ed03ff4 100644
--- a/relaxng.c
+++ b/relaxng.c
@@ -3993,7 +3993,7 @@ 
xmlRelaxNGGenerateAttributes(xmlRelaxNGParserCtxtPtr ctxt,
 * xmlRelaxNGGetElements:
 * @ctxt:  a Relax-NG parser context
 * @def:  the definition definition
- * @eora:  gather elements (0) or attributes (1)
+ * @eora:  gather elements (0), attributes (1) or elements and 
text (2)
 *
 * Compute the list of top elements a definition can generate
 *
@@ -4019,7 +4019,12 @@ 
xmlRelaxNGGetElements(xmlRelaxNGParserCtxtPtr ctxt,
    while (cur != NULL) {
        if (((eora == 0) && ((cur->type == XML_RELAXNG_ELEMENT) 
        ||
                             (cur->type == XML_RELAXNG_TEXT))) ||
-            ((eora == 1) && (cur->type == 
            XML_RELAXNG_ATTRIBUTE))) {
+            ((eora == 1) && (cur->type == XML_RELAXNG_ATTRIBUTE)) 
||
+            ((eora == 2) && ((cur->type == XML_RELAXNG_DATATYPE) 
||
+                             (cur->type == XML_RELAXNG_ELEMENT) 
||
+                             (cur->type == XML_RELAXNG_LIST) ||
+                             (cur->type == XML_RELAXNG_TEXT) ||
+                             (cur->type == XML_RELAXNG_VALUE)))) 
{
            if (ret == NULL) {
                max = 10;
                ret = (xmlRelaxNGDefinePtr *)
@@ -4374,7 +4379,7 @@ xmlRelaxNGComputeInterleaves(void *payload, 
void *data,
        if (cur->type == XML_RELAXNG_TEXT)
            is_mixed++;
        groups[nbgroups]->rule = cur;
-        groups[nbgroups]->defs = xmlRelaxNGGetElements(ctxt, cur, 
        0);
+        groups[nbgroups]->defs = xmlRelaxNGGetElements(ctxt, cur, 
2);
        groups[nbgroups]->attrs = xmlRelaxNGGetElements(ctxt, 
        cur, 1);
        nbgroups++;
        cur = cur->next;
@@ -9280,7 +9285,10 @@ xmlRelaxNGNodeMatchesList(xmlNodePtr node, 
xmlRelaxNGDefinePtr * list)
                return (1);
        } else if (((node->type == XML_TEXT_NODE) ||
                    (node->type == XML_CDATA_SECTION_NODE)) &&
-                   (cur->type == XML_RELAXNG_TEXT)) {
+                   ((cur->type == XML_RELAXNG_DATATYPE) ||
+                    (cur->type == XML_RELAXNG_LIST) ||
+                    (cur->type == XML_RELAXNG_TEXT) ||
+                    (cur->type == XML_RELAXNG_VALUE))) {
            return (1);
        }
        cur = list[i++];
--
2.19.1


Nikolai Weibull, 2018-10-13 00:23:

> Hi!
>
> This remains unfixed.  I have absolutely no idea what’s going on 
> in
> the interleave validation code.  Daniel, could you please put 
> together
> some minor documentation on how the interleave validation code 
> works?
> It’s very complicated.
>
> Thank you,
>
>  Nikolai
>
> Nikolai Weibull, 2018-09-09 21:26:
>
> > Hi!
> >
> > Given the following input RELAX NG grammar:
> >
> > <grammar xmlns="http://relaxng.org/ns/structure/1.0";
> >         datatypeLibrary="http://www.w3.org/2001/XMLSchema-datatypes";>
> >  <start>
> >    <element name="a">
> >      <interleave>
> >        <attribute name="b"/>
> >        <data type="string"/>
> >      </interleave>
> >    </element>
> >  </start>
> > </grammar>
> >
> > and the following input document a.xml:
> >
> > <a b="1">c</a>
> >
> > xmllint reports:
> >
> > a.xml:1: element a: Relax-NG validity error : Element a has 
> > extra
> > content: text
> > a.xml fails to validate
> >
> > Changing the interleave to a group solves the issue, so the 
> > problem
> > is
> > with how interleaves are validated.
> >
> > I looked at xmlRelaxNGValidateInterleave() and I sadly have no 
> > idea
> > what’s going on.  Please point me in the right direction and 
> > I’ll
> > gladly write a patch.
> >
> >  Nikolai