On Oct 14, 2010, at 5:57 PM, Dan Gohman wrote:
>
> On Oct 14, 2010, at 5:05 PM, Douglas Gregor wrote:
>
>>
>> On Oct 14, 2010, at 5:01 PM, Dan Gohman wrote:
>>
>>> }
>>>
>>> + // For now, treat all pointers as equivalent to each other.
>>> + if (Ty->isPointerType())
>>> + return MetadataCache[Ty] = getTBAAInfoForNamedType("TBAA.pointer",
>>> Char);
>>> +
>>
>> Basing TBAA info on strings seems... brittle. Perhaps you want to create a
>> TBAA metedata builder of some sort, which has a simple C++ interface hiding
>> the various strings behind it?
>
>
> The strings have no purpose other than uniquifying the MDNodes; the
> backend never looks at them. Front-ends can pick whatever strings
> make sense for themselves. We could introduce a C++ interface hiding
> the strings within clang itself, but CodeGenTBAA.cpp is only 90 lines
> long right now, so it doesn't seem urgent.
Ah, okay. I thought the names were part of the interface with the back end.
- Doug
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
