[cfe-dev] deque<Incomplete Type> support

[Peter Collingbourne]

On Tue, Nov 12, 2013 at 09:16:23PM -0500, Zhihao Yuan wrote:
> On Tue, Nov 12, 2013 at 6:26 PM, Dimitry Andric <dimitry at andric.com> wrote:
> > Now there might
> > be a problem when somebody else expects something different. :-)
> 
> So we only fix the cases without this possibility :)
> 
> >>  http://llvm.org/bugs/show_bug.cgi?id=9351
> >
> > Well, if this can be fixed in a similar way, and Howard thinks it is
> > worthwhile, I don't see why not.  But personally I would still refrain
> > from relying on it...
> 
> If more people rely on it, it's gonna be a good reason to
> standardize it :)
> 
> Here is my proposed change: remove the _BlockSize template
> parameter from __deque_iterator.  This does not decrease the
> identify of __deque_iterator since it's templated on _ValueType
> which covers the_BlockSize difference already.
> 
> Patch is attached.  Tests (+ a new one) passed.

Discussion of a similar patch, which was rejected due to ABI concerns:

http://www.mail-archive.com/cfe-commits@cs.uiuc.edu/msg80909.html
http://lists.cs.uiuc.edu/pipermail/cfe-dev/2013-October/032316.html

If you would like to make this change I would suggest doing it via an ABI
migration.

Thanks,
-- 
Peter