[cfe-commits] [Review] First step towards PR13231 - vftable generation with -cxx-abi microsoft

[Timur Iskhodzhanov]

Thanks to John's example above I think I now understand what's going
on in the Itanium ABI and I think I know where to go now to fix the
incompatibility.

On Wed, Dec 5, 2012 at 6:58 PM, João Matos <ripzonetriton at gmail.com> wrote:
> I've researched MS ctor/dtor behaviour a bit, check PR 12784
> (http://llvm.org/bugs/show_bug.cgi?id=12784) for a simple file that shows
> the different types we will expect.
>
> On Wed, Dec 5, 2012 at 2:28 PM, Timur Iskhodzhanov <timurrrr at google.com>
> wrote:
>>
>> On Tue, Dec 4, 2012 at 8:09 PM, Timur Iskhodzhanov <timurrrr at google.com>
>> wrote:
>> >>> d) The complete/deleting dtor logic will change when we support MSVC
>> >>> dtor types, but not now
>> >>>    [shouldn't matter now that we still don't support virtual
>> >>> inheritance]
>> >>
>> >> You're thinking about complete vs. base destructors.  Complete vs.
>> >> deleting
>> >> destructors isn't about virtual inheritance;  it's about whether the
>> >> destructor
>> >> calls operator delete.
>> > Yeah, now I see it... [see below]
>> >
>> >> (This is important because the 'delete' operator uses
>> >> the operator delete from the most-derived class.)
>> >>
>> >> How does MSVC handle the difference between
>> >>   foo->~Foo(); // virtually dispatched but doesn't call operator delete
>> >> and
>> >>   delete foo; // virtually dispatched and calls operator delete
>> >> ?
>> > Clang with my patch behaves badly on such code.
>> > It seems to work fine at least on simple code where you don't use
>> > virtual destructors.
>> >
>> > Actually, the whole idea of base vs complete dtors doesn't apply to
>> > -cxx-abi microsoft.
>> > Anyways, had plans to work on ctor and dtor compatibility as a
>> > separate patch - is that OK?
>> > I've put a couple of more FIXME into the code for that. (in
>> > ::AddMethod and ::ComputeMethodVTableIndices).
>>
>> Attached is the patch which is similar to bug_13231_2 but puts the
>> complete dtor rather than deleting.
>> Both are basically broken anyways but at least you can choose what you
>> think fits better :)
>> Investigating compatible dtor behavior is definitely on my TODO-soon
>> list now [as a separate change if this works]
>>
>> _______________________________________________
>> cfe-commits mailing list
>> cfe-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>>
>
>
>
> --
> João Matos