Re: Gregor's EXC-C14N interop samples (WAS:RE: Moving exc-c14n forward: your response is needed!)

On Thursday 18 April 2002 07:03, Gregor Karlinger wrote:
> > 3. I've tried Gregor's examples and, with one exception
> > (example 4, regular c14n, GrandChild, xml:fool should not be
> > there?), I think everything works.

Yesterday, I was able to confirm all of my examples with Gregor's except 
for normal c14n in example 1.

> I guess that you mean example 1 since this is the only one
> bearing an xml:fool attribute.

Yes, this is the problem I had, and I think Gregor is (unfortunately 
<grin/>) right:

http://mail.python.org/pipermail/xml-sig/2002-April/007646.html

However, I also discovered a bug from Gregor's example 1 [a]. Presently, 
c14n.py does the following:
1. takes any xml_attrs (e.g., xml:lang) from previous recursions not 
rendered and adds them to local xml_attrs.
2. if the element we are processing is in the subset, it renders the 
accumulated xml_attrs as part of that element and and resets xml_attrs=[] 
3. it recurses.

This has the effect of rendering all XML attributes that are "in affect" 
since the last ancestor was rendered. Sounds sensible, but it's not correct 
according to the the spec [b]. To fix this, I think we'd have to test 
whether the immediate parent was in the nodeset and keep a dictionary 
(instead of a list) of *all* ancestors' xml:foo attributes (not just since 
the nearest rendered one), with the *nearest* value. (Moving from a list to 
a dictionary is akin to what I did for the namespace bug.) This bug exists 
on in c14n, not exc-c14n.