[xml] Re: [xmlsec] exc c14n w/o comments bug

[Aleksey Sanin]

DISCLAIMER: the LibXML2 behavior described bellow is how I remember it.
Unfortunately I don't have time right now to check it and make sure
that this is exactly how it works. I'll try to debug the problem tonight
and then I should have more details.


>>Help is always welcome :) I suspect that the problem is caused by
>>the fact that LibXML2 attribute namespace has a different memory
>>representation from the namespace declaration. In this particular
>>case the problem is that attribute namespace *is not* returned in
>>the list of nodes selected by XPath expression.
> 
> 
>   Hum ... are you suggesting it's a bug ? The XPath data model clearly
> differentiates attributes from namespace.

I am talking about attribute namespace and it is different from "namespace
declaration". For example, if I remember everything correctly then you 
parse the following element

     <test:something xmlns:test="http://www.example.org/test"
                     test:data="aaa"/>

with LibXML2 then the xmlNodePtr for "test:data" attribute will have a
pointer to namespace xmlNodePtr and it will be *different* from the
xmlNodePtr created for "xmlns:test" namespace declaration.

Now if we make the example slightly more complex:

    <test1:something xmlns:test1="http://www.example.org/test1
                     xmlns:test2="http://www.example.org/test2">
	<test2:somethingelse test1:data="aaa">
	</test2:somethingelse>
    </test1:something>

and suppose that we have XPath expression that selects all attributes
in the test2:something element

	//test1:something//test2:somethingelse@*

In this case, the xmlNodePtr that represents test1 namespace in the
test1:data attribute would not be included in the resulting XPath
nodeset. I am not sure if it is a problem with LibMXL2 XPath engine
or not.

> 
>>The change is more or less trivial but it requires *very* carefull
>>testing. My estimates is that change itself would take about an hour
>>and testing would take another 3-4 hours. And it's more than I have
>>these days :(
> 
> 
>   How is the weird serialization generated ? Is the redundant
> xmlns information pushed as an attribute instead of a namespace ?
> Maybe there is some fixup needed at the libxml2 tree API too.

The problem is that during exc-c14n I have to check if particular
namespace is visible or not. This works just great for element
namespaces but for attributes it fails because of the reasons described
above.

Aleksey