[cfe-commits] r65570 - in /cfe/trunk: include/clang/Analysis/ include/clang/Basic/ include/clang/Driver/ include/clang/Parse/ lib/Analysis/ lib/Basic/ lib/Driver/ lib/Parse/ lib/Rewrite/ lib/Sema/ test/Parser/ test/SemaTemplate/

[Chris Lattner]

On Feb 27, 2009, at 9:50 AM, Douglas Gregor wrote:

>>>
>>> +++ cfe/trunk/lib/Parse/ParseExpr.cpp Thu Feb 26 15:00:50 2009
>>> @@ -365,10 +365,19 @@
>>>
>>>   if (!LHS.isInvalid()) {
>>>     // Combine the LHS and RHS into the LHS (e.g. build AST).
>>> -      if (TernaryMiddle.isInvalid())
>>> +      if (TernaryMiddle.isInvalid()) {
>>> +        // If we're using '>>' as an operator within a template
>>> +        // argument list (in C++98), suggest the addition of
>>> +        // parentheses so that the code remains well-formed in C+ 
>>> +0x.
>>
>> It seems somewhat strange to me to handle this diag here instead of  
>> sema.  Are you sure it wouldn't be more elegant in sema?
>
> Yes, because it's a parsing issue, not a semantic issue. The '>>'  
> token is treated differently in a template argument list in C++98  
> vs. C++0x, and the parser encodes this behavior in  
> getBinOpPrecedence. However, we don't have the information to  
> diagnose this language inconsistency until we know that we've parsed  
> the '>>' as an operator and have both operands of the expression.

Ok,

Thanks Doug!

-Chris