[m-rev.] for review: fix mantis bug 514

[Julien Fischer]

Hi Zoltan,

On Fri, 24 Jul 2020, Zoltan Somogyi wrote:

> 2020-07-24 15:52 GMT+10:00 "Julien Fischer" <jfischer at opturion.com>:
>> I will try it out with GCC 10 over the weekend.  (I don't recall if I
>> have a VM lying about that has GCC 10; if not I will create one.)
>
> Thanks.
>
>> We should probably look at defaulting to -fno-common for GCC < 10 (and
>> clang too if required).  The GCC manual says there may be a performance
>> impact with -fcommon.
>
> Yes, we should, but we need the Microsoft C compiler to have something
> equivalent as well.

I can't see anything in its documentation, but my guess is that it may
not support C style tentative definitions anyway as its really a C++
compiler.  Regardless, that shouldn't stop us enabling it on platforms
where we know it is supported.

>> Add a comment at the head of runtime/mercury_builtin_types.h pointing out that
>> all these places need to be updated if the arity of a builtin type changes.
>
> There is already a very similar comment just above the block of definitions :-(

Hmmm, the amount of time I needed to find that comment suggests its not
in a prominent enough place.  I think we need something at the top of
the mercury_builtin_types.h saying "When you add a new builtin type you *must*
update X, Y, Z".

Julien.