[m-rev.] for review: disable the use of asm_fast grades on Windows

[Julien Fischer]

Hi Zoltan,

On Mon, 22 Jan 2024, Zoltan Somogyi wrote:

> On 2024-01-21 15:32 +11:00 AEDT, "Julien Fischer" <jfischer at opturion.com> wrote:
>> Compilation of the stage 3 library currently results in a segmentation
>> fault on Windows.
>
> We must have a setup (Mercury version, C compiler version etc) from the past
> that was known to bootstrap correctly. The last release should fit the bill,
> and more recent ROTDs may fit it as well. Whatever that last setup was,
> Mercury has obviously changed since then, but the C compiler being used
> may have changed as well. If you have a record of what C compiler version
> last compiled Mercury in a way that allowed a correct bootstrap test,
> you could try that same version with the current Mercury system,
> since that would tell us whether the problem is a change in Mercury
> or a change in the behavior of the C compiler. And in both cases,
> the problematic version change could in theory be isolated by binary
> search.

I will check, but I suspect this problem has been present since at least
the inception of the 64-bit Windows port (~2013) and that the various
issues with the bootcheck script on that platform I mentioned have
prevented us from detecting it.

For the record:

- I first noticed this last July (when the bootcheck script
   was fixed for Windows).

- The segmentation fault occurs regardless of the C compiler
   optimisation level.

- It occurs with at least GCC 11 throught to 13.
   (Reverting an older version of GCC on MSYS2 is non-trivial.)

- It occurs on both MinGW64 and Cygwin64.

- Both the reg.gc and none.gc do bootcheck successfull

Julien.