[PATCH] Fix an assertion failure trying to emit a trivial destructor in ObjC++

[Ben Langmuir]

> On Sep 16, 2014, at 5:40 PM, Ben Langmuir <blangmuir at apple.com> wrote:
> 
> 
>> On Sep 16, 2014, at 5:08 PM, John McCall <rjmccall at apple.com <mailto:rjmccall at apple.com>> wrote:
>> 
>> On Sep 16, 2014, at 5:06 PM, Ben Langmuir <blangmuir at apple.com <mailto:blangmuir at apple.com>> wrote:
>>>> On Sep 16, 2014, at 1:25 PM, John McCall <rjmccall at apple.com <mailto:rjmccall at apple.com>> wrote:
>>>> 
>>>> On Sep 16, 2014, at 12:54 PM, Richard Smith <richard at metafoo.co.uk <mailto:richard at metafoo.co.uk>> wrote:
>>>>> On Tue, Sep 16, 2014 at 12:47 PM, John McCall <rjmccall at apple.com <mailto:rjmccall at apple.com>> wrote:
>>>>> On Sep 16, 2014, at 12:11 PM, Richard Smith <richard at metafoo.co.uk <mailto:richard at metafoo.co.uk>> wrote:
>>>>>> On Tue, Sep 16, 2014 at 11:45 AM, John McCall <rjmccall at apple.com <mailto:rjmccall at apple.com>> wrote:
>>>>>> On Sep 15, 2014, at 9:47 AM, Ben Langmuir <blangmuir at apple.com <mailto:blangmuir at apple.com>> wrote:
>>>>>> > Hi John,
>>>>>> >
>>>>>> > This patch fixes the assertion failure I talked to you about in Objective C++ codegen.  It turned out to have nothing to do with templates.
>>>>>> >
>>>>>> >    Fix an assertion failure trying to emit a trivial destructor in ObjC++
>>>>>> >
>>>>>> >    If a base class declares a destructor, we will add the implicit
>>>>>> >    destructor for the subclass in
>>>>>> >    ActOnFields -> AddImplicitlyDeclaredMembersToClass
>>>>>> >
>>>>>> >    But in Objective C++, we did not compute whether we have a trivial
>>>>>> >    destructor until after that in
>>>>>> >    CXXRecordDecl::completeDefinition()
>>>>>> >
>>>>>> >    This was leading to a mismatch between the class, which thought it had
>>>>>> >    no trivial destructor, and the CXXDestructorDecl, which considered
>>>>>> >    itself trivial.
>>>>>> 
>>>>>> I feel like hasTrivialDestructor should return the right value here.  I understand (and am saddened by) the hack about not setting PlainOldData until completeDefinition, but maybe we can set/clear the rest of the bits eagerly?
>>>>>> 
>>>>>> Why do we have to delay setting the PlainOldData flag?
>>>>> 
>>>>> There is a diagnostic which wants to warn about structs that are only POD in non-ARC modes.
>>>>> 
>>>>> Thanks, I suspected something along those lines. Perhaps we could track both properties and still perform the calculation eagerly:
>>>>> 
>>>>> -  bool isPOD() const { return data().PlainOldData; }
>>>>> +  bool isPOD() const { return data().PlainOldData && !data().HasARCObjectMember; }
>>>>> +  bool wouldHaveBeenPODIfItWerentForYouMeddlingKids() const { return data().PlainOldData; }
>>>> 
>>>> That works for me, or we could even give it its own bit in the definition data; it’s not like we aren’t tracking a number of other things there for similar purposes.
>>>> 
>>>> John.
>>> 
>>> John and I took a look and it turns out we killed the warning in question as part of removing -Warc-abi.  I’ve attached an updated patch that just eagerly sets these bits in addedMember so we will get the correct value inside AddImplicitlyDeclaredMembersToClass.
>> 
>> Looks great to me; thanks, Ben.
>> 
>> John.
> 
> Thanks John.
> 
> Richard, did you have any other feedback, or shall I commit?

ping

> 
> Ben

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20140919/4553702c/attachment.html>