[m-dev.] rotd-2024-04-5 build failed / cannot compile lib in C# grade

[Julien Fischer]

On Sat, 6 Apr 2024, Zoltan Somogyi wrote:

> On 2024-04-06 16:01 +11:00 AEDT, "Julien Fischer" <jfischer at opturion.com> wrote:
>> The grade is asm_fast.gc.  We build the the source distribution in
>> hlc.gc.pregen, but above does not build the source distribution.
>> It builds and installs the compiler that is then used to build the
>> source distribution.
>
> OK. In that case, the problem is probably NOT in the MLDS code generator,
> but in the LLDS code generator, but for some reason, it miscompiles only
> code in the MLDS parts of the compiler.

That's one possibility; the other is that that specific version of GCC
is miscompiling some of the new switch code somewhere along the line.

>> Having just run through that process half a dozen times today, you
>> don't need to tell me ;-)  I'm going to attempt to reproduce the
>> problem on the 16-core server I gave you access to, and then investigate
>> further from there.
>
> Thanks.

Ok, I have reproduced the problem on the other server, which is
unsurprising since it's indentically configured other than have a more
civilized number of cores.

(FYI: rotd-2024-04-04 is installed in /usr/local/mercury/rotd-2024-04-04-gcc4).

>>> And what was the last time the process worked without an error?
>>
>> rotd-2024-04-04 (commit: 6ef9b3cf16209c829ef32f1a0a5bfe05113f07)
>
> That narrows down the scope of the problem. Since then I have commited
> only three minor diffs that shouldn't have caused an issue, but one clearly did.
> My guess would be 07db4ec97ee854a320a4369ea760320109453a68.
> I don't think you can reverse just that commit without first reversing the next two,
> but you can try backing out all of them, check that this works, and then reapply
> just 07db, and see whether that fails. Or just check out 07db and test it directly.

I will do so later this evening; I'll also try some different versions
of GCC in order to eliminate that as a factor.

Julien.