[cfe-commits] [PATCH] differentiate between non-POD and incomplete with regard to va_arg

[Eli Friedman]

On Mon, Jun 13, 2011 at 10:08 PM, David Majnemer
<david.majnemer at gmail.com> wrote:
> On Mon, Jun 13, 2011 at 5:04 PM, David Majnemer
> <david.majnemer at gmail.com> wrote:
>> On Mon, Jun 13, 2011 at 4:16 AM, David Majnemer
>> <david.majnemer at gmail.com> wrote:
>>> Hi,
>>>
>>> Right now, the code base is correct in saying that many types, like
>>> void, are non-POD when passed in to va_arg.
>>>
>>> While it is true that "struct X" is not a POD type, we do not say why
>>> it is not a POD type.
>>> I think that it would be more clear to say that the type is
>>> incomplete, this had a side effect of making a bit more sense in C99
>>> mode where non-POD is not really part of the lingo.
>>>
>>> The attached patch should make it a little more clear. It also moves a
>>> preexisting test to a more appropriate location.
>>>
>>> --
>>> David Majnemer
>>>
>>
>> After talking some folks in #llvm and thinking about it a bit, I have
>> decided to implement the following:
>>
>> The type must be complete per RequireCompleteType, otherwise issue an
>> error diagnostic
>> The type must be non-abstract per RequireNonAbstractType, otherwise
>> issue an error diagnostic
>> The type must be a POD per isPODType, otherwise issue a warning
>> diagnostic (-Wnon-pod-varargs)
>>
>> I have attached the new patch
>>
>> --
>> David Majnemer
>>
>
> A small update to the patch: don't return ExprError when the Diag is
> just a warning and keep the formatting within 80 columns.
>
> I had someone look it over in #llvm, awaiting the OK over here.

Looks fine.

-Eli