Saxon XSLT and XQuery Processor

I think it depends on what components you have in your pipeline. There's a
component somewhere that will manufacture a prefix for the attribute, but
it's probably possible to assemble a pipeline in which that component is
absent. My usual rule of thumb is that sorting out namespace constistency is
the task of the NamespaceReducer component.

(I'm making guesses here about the nature of the application as being some
kind of pipeline assembled from Saxon components.)

Regards,

Michael Kay
http://www.saxonica.com/
http://twitter.com/michaelhkay  

> -----Original Message-----
> From: Norman Walsh [mailto:ndw@...] 
> Sent: 14 February 2010 21:05
> To: saxon-help@...
> Subject: [saxon] Attribute in an explicit namespace
> 
> Hi,
> 
> If I construct a QName w/o a prefix:
> 
>   QName aname = new QName("http://example.com/", "name")
> 
> and use that as the name of an attribute, when Saxon 
> serializes the result, it appears not to manufacture a prefix 
> for the attribute.
> 
> Is that what you'd expect, or is it likely that I'm doing 
> something wrong in either the creation of the attribute or in 
> the call to the serializer?
> 
> (If the answer isn't obvious, I'll construct a proper test 
> case and send it in, just let me know.)
> 
>                                         Be seeing you,
>                                           norm
> 
> --
> Norman Walsh <ndw@...> | Correct me if I'm wrong, but 
> hasn't the
> http://nwalsh.com/            | fine line between sanity and madness
>                               | gotten finer?--George Price
>

Norman Walsh wrote:

  Hi Norm,

> If I construct a QName w/o a prefix:

>   QName aname = new QName("http://example.com/", "name")

> and use that as the name of an attribute

  Of course I let Mike give the details about Saxon, but that's
an interesting one.  You are actually asking for an attribute in
a specific namepace but with a QName without prefix.  What is, if
I am right, not possible in XML.

  Shouldn't you instead create a prefix in the calling application:

    new QName("http://example.com/", "pr:name")

  Regards,

-- 
Florent Georges
http://www.fgeorges.org/

"Michael Kay" <mike@...> writes:
> I think it depends on what components you have in your pipeline. There's a
> component somewhere that will manufacture a prefix for the attribute, but
> it's probably possible to assemble a pipeline in which that component is
> absent. My usual rule of thumb is that sorting out namespace constistency is
> the task of the NamespaceReducer component.

Ok. I don't think I've worked very hard to setup an explicit pipeline, but
perhaps my memory is hazy. I'll construct a small test case to see if I
can work it out.

                                        Be seeing you,
                                          norm

-- 
Norman Walsh <ndw@...> | Traditions are group efforts to keep
http://nwalsh.com/            | the unexpected from happening.--Barbara
                              | Tober

Florent Georges <lists@...> writes:
>> If I construct a QName w/o a prefix:
>
>>   QName aname = new QName("http://example.com/", "name")
>
>> and use that as the name of an attribute
>
>   Of course I let Mike give the details about Saxon, but that's
> an interesting one.  You are actually asking for an attribute in
> a specific namepace but with a QName without prefix.  What is, if
> I am right, not possible in XML.

True, but for better or worse, you can do this in XProc:

  <p:add-attribute match="//div" attribute-value="bar"
                   attribute-name="foo"
                   attribute-namespace="http://example.com/"/>

My thinking is that if you do, the processor should pick some unused
prefix.

I can probably work out the unused prefix at the XProc step level, but
if I can get Saxon to do the work for me when it serializes, so much
the better. :-)

                                        Be seeing you,
                                          norm

-- 
Norman Walsh <ndw@...> | Ahhh. A man with a sharp wit. Someone
http://nwalsh.com/            | ought to take it away from him before
                              | he cuts himself.--Peter da Silva

> I can probably work out the unused prefix at the XProc step 
> level, but if I can get Saxon to do the work for me when it 
> serializes, so much the better. :-)

Saxon won't do namespace fixup as it serializes - the serializer assumes
that's its input satisfies all constraints. The namespace fixup has to be
done earlier.

Looking at the code, I think Saxon allocates prefixes for attribute nodes
right in the logic for xsl:attribute instructions or XQuery computed
attribute constructors, so I think you'll have to replicate that logic
yourself. Putting a NamespaceReducer into the pipeline in front of the
serializer will ensure that the namespace gets declared. There's a helper
method in NamePool, suggestPrefixForUri() which does what it says, but this
doesn't check that the prefix isn't a duplicate; Saxon does that check in
the ComplexContentOutputter, which you probably don't want to incorporate in
your pipeline either. Logically it should probably be done in the
NamespaceReducer, but changing the code in this area will always introduce a
bug or two, so I prefer to leave it alone.

Regards,

Michael Kay
http://www.saxonica.com/
http://twitter.com/michaelhkay

Norman Walsh wrote:

> True, but for better or worse, you can do this in XProc:

>   <p:add-attribute match="//div" attribute-value="bar"
>                    attribute-name="foo"
>                    attribute-namespace="http://example.com/"/>

  Isn't it required to use attribute-prefix here?  BTW that's why
I proposed to use a literal QName as the value of *-name :-) (and
in this case I think it would even make sense to allow NCName
only when *-namespace is not set...)

  Regards,

-- 
Florent Georges
http://www.fgeorges.org/

On 15 February 2010 15:18, Florent Georges <lists@...> wrote:
> Norman Walsh wrote:
>
>> True, but for better or worse, you can do this in XProc:
>
>>   <p:add-attribute match="//div" attribute-value="bar"
>>                    attribute-name="foo"
>>                    attribute-namespace="http://example.com/"/>
>
>  Isn't it required to use attribute-prefix here?  BTW that's why
> I proposed to use a literal QName as the value of *-name :-) (and
> in this case I think it would even make sense to allow NCName
> only when *-namespace is not set...)

If you don't mind me wading into this too without following the whole
thread, feel free to ignore it :) it's interesting that you don't use
the same way XSLT works, for example:

<p:add-attribute match="//div" attribute-value="bar"
   attribute-name="f:foo" xmlns:f="foo_ns" xmlns="bar_ns"/>

....having the namespace defined in the XML that is the config file.




-- 
Andrew Welch
http://andrewjwelch.com
Kernow: http://kernowforsaxon.sf.net/

Andrew Welch wrote:


> it's interesting that you don't use the same way XSLT works


  That's the last email I post in this thread, as we are becoming really OT, but just to mention the following thread on this same subject:


    http://markmail.org/thread/tx3myfodnckdox7w

  Regards,

-- 
Florent Georges
http://www.fgeorges.org/

"Michael Kay" <mike@...> writes:
> Looking at the code, I think Saxon allocates prefixes for attribute nodes
> right in the logic for xsl:attribute instructions or XQuery computed
> attribute constructors, so I think you'll have to replicate that logic
> yourself.

Ok. Fair 'nough. Thanks!

                                        Be seeing you,
                                          norm

-- 
Norman Walsh <ndw@...> | Ahhh. A man with a sharp wit. Someone
http://nwalsh.com/            | ought to take it away from him before
                              | he cuts himself.--Peter da Silva