On 14/08/2012 22:46, Kin-man Chung wrote:
> On 08/14/12 14:18, Mark Thomas wrote:
>> On 10/08/2012 23:13, Kin-man Chung wrote:
>>
>>> There are some feedbacks from the public review. One of them is a
>>> request for adding namespaces to idendifiers.
>>>
>>> I know CDI has been using a.b.c.d as namespaced identifiers, but that
>>> creates ambiguities in both the syntax and semantics.
>>>
>>> Since we already have namespace for functions, its only logical to use
>>> the same syntax and notation for identifiers. For example,
>>>
>>> ns:obj.foo denotes the property foo of an object with a
>>> namespace ns.
>>>
>>> Obviously, namespaces are only allowed for top level objections.
>>>
>>> Note that there is an ambiguity with the ?: operator.
>>>
>>> What does a? b: d : e mean, a? (b:c): e or a? b: (c: d)?
>>>
>>> I think we already have similar problem with namespaced function in :?
>>> operators, so maybe we can reosslve this similarily. Let think about
>>> this more.
>>>
>> Could we just define a precedence order and require use of () to
>> change it?
>>
> I was thinking about this also. It probably makes sense to have
> namespace higher precedence than :? operator.
I was leaning the other way but having thought about it some more, I
think that is the better approach. That said, I reserve the right to
change my mind if writing the parser is a lot easier the other way ;)
>>> How can we fit this into our ELResolver? When an
>>> ELResolver.getValue(Object base, Object property) is invoked, property
>>> is the identifier, as before, but instead of a null base, we can
>>> encapsulate the namespace in "base". So the code will be like
>>>
>>> Object getValue(ELContext context, Object base, Object property) {
>>> String namespace = null;
>>> if (base != null&& base instance ELNameSpace)
>>> namespace = ((ELNameSpace)base).getNameSpace();
>>> }
>>> ...
>>>
>>> What do you think?
>>>
>> Seems like a bit of a hack but I don't have any better ideas right now.
>>
> The other option is to leave base null, and to have property a String of
> the form "ns:foo". I think this is more nature.
Agreed.
Mark