Re: annotation-xml and annotation encoding

Le 17 déc. 07 à 09:23, Henri Sivonen a écrit :
> As far as I can tell, MathML 2.0 doesn't define a mechanism that's  
> allow implementations to implementations to use interoperable  
> values for the encoding attribute on <annotation> and <annotation- 
> xml>. The spec gives four tokens leaving their meaning implicit:  
> MathML-Presentation, MathML-Content, TeX and OpenMath.

In the examnples, right ? There is, indeed, no central table of  
suggested encoding values for well-known data-types.

> In the MathML 3.0 draft, the encoding attribute on <annotation>  
> seems to take a MIME type, such as text/latex or text/maple, or a  
> product name token like Maple, Mathematica or TeX.

I believe that the order should be:
- try to use a value that's documented the spec
- if there's none such use a mime-type

> In the MathML 3.0 draft the encoding attribute on <annotation-xml>  
> is said to take a namespace URI but examples use tokens such as  
> OpenMath.

I don't remember seeing this... In the spec to come out soon, this  
has gone away, I think. All of this has been revised quite much,  
especially under the light of clipboard operations. So I would wait  
for the draft that is coming soon where, at least:
- there are recommended-implications for clipboard-flavors
- annotations can be taken in a more general sense than the semantic  
alternative using annotation's key that is a symbol (e.g has-type, ...).

Also, the table of recommended names for presentation, content, and  
generic flavors will be included later in a unified manner since it  
used to be somewhat inconsistent (should be  MathML-Content, MathML- 
Presentation, MathML).

> Using a namespace URI as the encoding attribute value seems  
> redundant and unnecessary. Why wouldn't the consuming application  
> inspect the namespace of the child element?

Sure.

> <annotation> and <annotation-xml> appear to be so vaguely defined  
> that I have to doubt their interoperable implementability. Have  
> they been implemented in applications that consumes MathML? If they  
> have been implemented, have they been implemented interoperably? If  
> they are now interoperably implemented, it would be good for the  
> spec to define how to consume them in the way that is interoperable.

We really need to share more about defining the interoperability.
The new draft takes a stab at it with clipboard operations and  
"detached annotations" and I hope it'll bring more interop but  
interoperability can only be achieved if it suits all exchange  
purposes all concerned parties have in mind.

> Le 17 déc. 07 à 11:23, Max Berger a écrit :
> OpenOffice uses:
> <math:annotation math:encoding="StarMath 5.0">...</math:annotation> to
> describe the "source" for the MathML encoded in its ODF (OpenDocument
> Formula) files. This works very well within Openoffice and related
> products, which use this information when re-reading files.

Is is not an error to use math:encoding attribute name ?
Except for xml:xx or xlink:xx, which are imported attributes from  
other specs, no attribute in the MathML-spec use namespaced  
attributes, do I mistake ?
I see indeed that the draft of April at:
   http://www.w3.org/TR/2007/WD-MathML3-20070427/chapter5.html#id.5.4.2
does contain such an example. This is gone in the next draft.


paul