[cfe-commits] r144159 - in /cfe/trunk: lib/Sema/TreeTransform.h test/SemaCXX/pseudo-destructors.cpp

[Douglas Gregor]

On Nov 8, 2011, at 6:26 PM, Eli Friedman wrote:

> On Tue, Nov 8, 2011 at 6:19 PM, Douglas Gregor <dgregor at apple.com> wrote:
>> Author: dgregor
>> Date: Tue Nov  8 20:19:47 2011
>> New Revision: 144159
>> 
>> URL: http://llvm.org/viewvc/llvm-project?rev=144159&view=rev
>> Log:
>> Don't crash when transforming an ill-formed pseudo-destructor
>> expression. Fixes PR11339.
>> 
>> Modified:
>>    cfe/trunk/lib/Sema/TreeTransform.h
>>    cfe/trunk/test/SemaCXX/pseudo-destructors.cpp
>> 
>> Modified: cfe/trunk/lib/Sema/TreeTransform.h
>> URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Sema/TreeTransform.h?rev=144159&r1=144158&r2=144159&view=diff
>> ==============================================================================
>> --- cfe/trunk/lib/Sema/TreeTransform.h (original)
>> +++ cfe/trunk/lib/Sema/TreeTransform.h Tue Nov  8 20:19:47 2011
>> @@ -7218,7 +7218,7 @@
>>     if (!DestroyedTypeInfo)
>>       return ExprError();
>>     Destroyed = DestroyedTypeInfo;
>> -  } else if (ObjectType->isDependentType()) {
>> +  } else if (!ObjectType.isNull() && ObjectType->isDependentType()) {
>>     // We aren't likely to be able to resolve the identifier down to a type
>>     // now anyway, so just retain the identifier.
>>     Destroyed = PseudoDestructorTypeStorage(E->getDestroyedTypeIdentifier(),
>> 
>> Modified: cfe/trunk/test/SemaCXX/pseudo-destructors.cpp
>> URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/SemaCXX/pseudo-destructors.cpp?rev=144159&r1=144158&r2=144159&view=diff
>> ==============================================================================
>> --- cfe/trunk/test/SemaCXX/pseudo-destructors.cpp (original)
>> +++ cfe/trunk/test/SemaCXX/pseudo-destructors.cpp Tue Nov  8 20:19:47 2011
>> @@ -68,3 +68,12 @@
>>   x0.~X0();
>>  }
>> 
>> +namespace PR11339 {
>> +  template<class T>
>> +  void destroy(T* p) {
>> +    p->~T(); // ok
>> +    p->~oops(); // expected-error{{expected the class name after '~' to name a destructor}}
> 
> This isn't the same error we get for the non-dependent case:
> 
> <stdin>:1:22: error: 'oops' does not refer to a type name in
> pseudo-destructor expression; expected the name of type 'int'
> void f(int* p) { p->~oops(); }
>                     ^
> 
> Not sure how much that matters, though.


Yeah, we're missing some resolution of the identifier at template instantiation time. I don't think it has any impact on semantics (just diagnostics), so it's fairly low priority.

	- Doug