Re: [xml-dev] Concerned about the increasing reliance on XPath

[Stephen D Green <stephengreenubl@gmail.com>]

I tried to think of a way to write the XPath expression so
it gives the same result whatever the version of XPath
and it gets rather complex and obscure and not what I'd
be expecting a non-specialist developer to do or understand
well enough to be able to maintain it but it does evaluate
to true whatever the version and maybe (?) whatever the
namespace bindings

/*[local-name()='a']/@version-id = 0.4

What would I have to do to ensure it didn't get the same result
with another default namespace though? I guess I'd have to
make the expression even more complex:

/*[local-name()='a'][namespace-uri()='http://example.com']/@version-id = 0.4

I could hardly expect a nominally XML-savvy developer to
have to understand and maintian this. Then, if the expression
involved more than one element and mutliple namespaces...
----
Stephen D Green



On 9 May 2011 15:49, David Carlisle <davidc@nag.co.uk> wrote:
> On 09/05/2011 15:33, Stephen D Green wrote:
>>
>> An example of the reasons for my exacerbations with XPath
>> and my own questions about reliance on it
>>
>> I've been trying XPath expressions which would be tests I
>> could apply to some XML for validation during processing
>> (of a dataset in some ASP.NET, in this case, which seems
>> a sensible use case).
>>
>> Here is an example fragment
>>
>> <a xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"; version-id="0.4"
>>     xmlns="http://example.com";>
>> ...
>>
>> In XPath 1.0
>> /default:a/@version-id = 0.4 returns true
>> but  /a/@version-id = 0.4 returns false
>
> For the first case not to be an error you mush hav edefined the prefix
> "default" to be http://example.com
> The second case is as would be expected as there is no no-namespace element
> with local name a at the top level.
> /b/@version-id=0.4 would also be false.
>
>
>>
>> In XPath 2.0
>> /default:a/@version-id = 0.4 returns an error (prefix not bound)
>> and  /a/@version-id = 0.4 returns true
>
> You must have set the default xpath namepsace somewhere (and not defined the
> namespace binding for the prefix default:)
> so you only get those results depending on code you have not shown, and you
> would get the same results as xpath 1 by default.
>
>>
>> (I tried these out in Oxygen and I trust the results)
>>
>> In my own case I'd start off with XPath 1.0 as it is there out-of-the-box
>> in my particular toolbase but later I would imagine there might be a
>> reason to progress to XPath 2.0. It would look to me is if all the XPaths
>> would need review in that case and some awkward gotchas might occur
>> and developers might have to learn about lots of subtle and barely
>> explicable
>> nuances and differences between the versions, etc.
>
> the places where xpath2 gives different results are rather slight 9and even
> less if xpath2 is used in backward compatibility mode.
> The example you gave is not one of them.
> /a/@version-id = 0.4 returns false
> will be the xpath 2 behaviour unless you use the (xpath 2) feature of
> changing the default xpath namespace.
>
>
>>  I'd be reluctant to
>> encumber them with this by chosing this approach to validating the XML
>> beyond what I get with XSD but overall there doesn't seem to be an
>> easier/better alternative so I'd probably still go ahead.
>>
>> ----
>> Stephen D Green
>>
>
> The fact that changing a language from version 1 to version 2 might have
> some problems in edge cases is not really related to xpath though, you would
> get differences changing from xsd 1.0 to 1.1 or even just between xsd
> processors as they tend to implement darker corners of the xsd spec
> differently.
>
> David
>
> ________________________________________________________________________
> The Numerical Algorithms Group Ltd is a company registered in England
> and Wales with company number 1249803. The registered office is:
> Wilkinson House, Jordan Hill Road, Oxford OX2 8DR, United Kingdom.
>
> This e-mail has been scanned for all viruses by Star. The service is
> powered by MessageLabs.
> ________________________________________________________________________
>