[cfe-commits] r101953 - /cfe/trunk/lib/AST/Decl.cpp

[Fariborz Jahanian]

On Apr 20, 2010, at 11:27 PM, Douglas Gregor wrote:

>
> On Apr 20, 2010, at 11:13 PM, Chris Lattner wrote:
>
>>>>> Can you explain a bit about why this makes you uncomfortable?  
>>>>> From my perspective, all of the Objective-C declarations have  
>>>>> external linkage since they are visible from multiple  
>>>>> translation units.
>>>>
>>>> Except for meta-data for the class (and its metaclass), othere  
>>>> meta-data have non-external linkage.
>>>> Content of all non-external metadata, however,  can be inspected  
>>>> with various APIs. In that respect they are 'external'.
>>>> But not wrt linkage.
>>>
>>> I'm looking at linkage more from the language perspective. For  
>>> example, say we have:
>>>
>>> @interface A
>>> @end
>>>
>>> A *an_A;
>>>
>>> Is this code well-formed? In C++, it actually depends on the  
>>> linkage of A: if A has no linkage, then the program is ill-formed  
>>> because an_A does have linkage. Also, one could imagine  
>>> instantiating a template over an Objective-C class (or pointer to  
>>> one), e.g.,
>>>
>>> std::vector<A *> a_objects;
>>>
>>> which also requires A to have linkage.
>>
>> How about "if 'a' has a defined/stable mangling, then it has  
>> linkage".  Is this close enough to be a useful guideline?
>
>
> It's close, although it doesn't capture entities declared in  
> anonymous namespaces, which have external linkage (from the language  
> perspective) but no stable mangling.

One exception may be the protocols as they do not participate in  
mangling and they cannot be directly accessed, etc.

- Fariborz

>
>
> 	- Doug
>
>