[cfe-commits] [patch] Fixed for function specifiers

[Douglas Gregor]

On May 14, 2009, at 1:48 PM, Chris Lattner wrote:

>
> On May 14, 2009, at 10:41 AM, Douglas Gregor wrote:
>
>>
>> 	inline int f(int), g;
>>
>> I suggest that we introduce fix-it hints that remove the erroneous  
>> specifier, e.g., for inline on non-functions, use:
>>
>> 	Diag(D.getIdentifierLoc(), diag::err_inline_non_function)
>> 		<< CodeModificationHint::CreateRemoval(SourceRange(D.getDeclSpec 
>> ().getInlineSpecLoc()))
>> 		<< SourceRange(D.getDeclSpec().getInlineSpecLoc());
>>
>
> Why?  I think fixit hints should only be used when we really know  
> what the user meant.  I don't think there is any obvious  
> interpretation that the user "accidentally typed inline on a  
> variable".  If we saw this in practice, we should reject it, but I  
> don't think there is a clear indicator of what the user *meant*


Yeah, I jumped the gun here. The SourceRange(D.getDeclSpec 
().getInlineSpecLoc()) should remain, so that we highlight the  
problematic specifier, but Chris is right: there isn't enough context  
here to be sure that eliminating the specifier is the right thing to do.

	- Doug