TypeTuple ABI mangling

[Iain Buclaw via Digitalmars-d]

On 27 July 2014 08:14, Iain Buclaw <ibuclaw at gdcproject.org> wrote:
> On 27 July 2014 00:20, Iain Buclaw <ibuclaw at gdcproject.org> wrote:
>> On 26 July 2014 21:06, H. S. Teoh via Digitalmars-d
>> <digitalmars-d at puremagic.com> wrote:
>>> On Sat, Jul 26, 2014 at 07:44:47PM +0000, Iain Buclaw via Digitalmars-d wrote:
>>>> Hi,
>>>>
>>>> TypeTuples have a special mangling convention in the compiler.
>>>>
>>>> As per the docs (ref: http://dlang.org/abi.html)
>>>>
>>>> TypeTuple:
>>>>     B Number Arguments
>>>>
>>>>
>>>> However after some cursory tests, it seems that the most obvious examples
>>>> of, passing a tuple via parameters, returning a tuple, declaring a tuple
>>>> variable, etc... all get lowered before mangling.
>>>>
>>>> So at first glance, this mangling convention is redundant.
>>>>
>>>> Unless of course someone can post an example which triggers mangling
>>>> of TypeTuples...
>>> [...]
>>>
>>> Huh, interesting. I tried various wrapping/unwrapping tricks and passing
>>> things via alias parameters, but none of them triggered TypeTuple
>>> mangling. Either I missed something obvious, or perhaps this mangling is
>>> never actually emitted? Not sure.
>>>
>>
>> I've tried something similar using D1 too, and get the same results.
>>
>> I'd be willing to do an experiment and remove TypeTuple mangling from
>> the language by force and see what ICE's.
>
> I changed Ttuple = 'B' to Ttuple = '@' and waited for assembly errors to happen.
>
> Finally got them when compiling std.outbuffer unittests.
>
> Reducing....


Ah ha!  D-style native varargs use them!

void printf(...) { }
void main()
{
    printf;    // printf(_arguments_typeinfo);
}

Here, the _arguments_typeinfo is a TypeInfo_Tuple.

Regards
Iain.