Re: [xml] Schema validity failure for valid document

[Kasimier Buchcik <kbuchcik 4commerce de>]

Hi (back from vacations),


Daniel Veillard wrote:
> On Mon, Jan 03, 2005 at 05:46:38PM +0100, cazic gmx net wrote:
[...]
> > >  well, at the time of the error, all libxml2 had left from the Schemas
> > > content model is a compiled automata. Sometimes that automata can be
> > > reserialized as a string making sense to a human, we don't have an API for
> >
> > This would be great!
> >
> >
> > > this (yet that could be added). It's also sometime hard to find out what
>
>
>   Reserializing the content model to a string (as DTD errors do now) is
> not generally possible for all regexp and schemas content model. Instead
> I propose an API to extract the last error value, and the list of potential
> accepted values.
>   Not yet added to the header but I commited to xmlregexp.c the following
> routine:
> /**
>  * xmlRegExecErrInfo:
>  * @exec: a regexp execution context generating an error
>  * @string: return value for the error string
>  * @nbval: pointer to the number of accepted values IN/OUT
>  * @values: pointer to the array of acceptable values
>  *
>  * Extract error informations from the regexp execution, the parameter
>  * @string will be updated with the value pushed and not accepted,
>  * the parameter @values must point to an array of @nbval string pointers
>  * on return nbval will contain the number of possible strings in that
>  * state and the @values array will be updated with them. The string values
>  * returned will be freed with the @exec context and don't need to be
>  * deallocated.
>  *
>  * Returns: 0 in case of success or -1 in case of error.
>  */
> int
> xmlRegExecErrInfo(xmlRegExecCtxtPtr exec, const xmlChar **string,
>                   int *nbval, xmlChar **values)
>
>   For example suppose you validate 
>
>     <doc><b/><c/><d/></doc>
>
> with test/schemas/all_1.xsd (where doc content model is defined as an
> <all> with <a><b><c>, then when calling xmlRegExecErrInfo on the 
> associated regexp after the error is noticed, then you will get
>    string = "b"
>    nbval = 1
>    values[0] = "a"
> on return.
> This should be the foundation for:
>    - better error reporting
>    - schemas driven editing capabilities
>    - better DTD driven editing capabilities
> for the 2 last examples rather than on-error report, one would 
> probably need potential next values from a current state of the
> xmlRegExecCtxtPtr, this is actually trivial to add based on the same
> kind of code, something like
>
> int xmlRegExecNextValues(xmlRegExecCtxtPtr exec, int *nbval, xmlChar **values)
>
> would be the associated API.
>
>   Thoughts ? I will probably export xmlRegExecErrInfo and xmlRegExecNextValues
> from the xmlregexp.h header soon. But special code to improve the Schemas
> error reports would still be needed, Kasimier does this fit your need ?

Cool!
A namespace aware version would be needed as well.
Proposal:
  xmlRegExecErrInfo2(xmlRegExecCtxtPtr exec,
    const xmlChar **string,
    const xmlChar **string2, <-- the namespace name
    int *nbval,
    xmlChar **values,
    xmlChar **values2) <-- the array of namespace names

In the case of the XML Schema engine, @string and @string2 would be
already known.

Occurence:

Content model: (a, b*, c)
input: <a/><d/>
would xmlRegExecErrInfo return "b" for values[0], or
rather "c", since the automaton should have passed the
b* ?

This leads to the question if/how to report occurence information;
at least if the element is mandatory or not.

Content model: (a, b+, c)
input: <a/><b/><d/>
At the point of the error prone input of "d" it seems cruicial to
report that the allowed input at this point can be
an optional "b" or a "c". Don't know if this info can be provided by the
regex engine.

Any ideas?

Greetings,

Kasimier