[cfe-commits] RFC: Half floating point support

[Eli Friedman]

On Wed, Oct 5, 2011 at 12:13 AM, Anton Korobeynikov
<anton at korobeynikov.info> wrote:
> Hello Eli,
>
>> This looks suspicious: specifically, it destroys the invariant in C++
>> that preinc/dec return an lvalue.  Also, it doesn't appear to handle
>> ++x and x+=1 equivalently.
> How it can be modelled then? Here we have unary expression applied
> directly to decl reference.

The cast does need to get there somehow in C mode (in C++, we should
get an appropriate AST without that change).  However you do it, it
does need to cover the x+=1 case, though.

If you follow John's advice and do the promotion in
UsualUnaryConversions, it should pick up this case without extra code.

-Eli