Re: FAIL: tests/math.test on Debian armhf

[Vagrant Cascadian]

On 2022-05-08, Jan Nieuwenhuizen wrote:
> Vagrant Cascadian writes:
>> On 2022-05-06, Vagrant Cascadian wrote:
>>> The host machine was running a 64-bit capable kernel if that matters for
>>> some reason, but it was done in a 32-bit userland chroot (Debian
>>> armhf). This also only appeared to fail when using "./configure
>>> --with-bootstrap", the gcc-only build of mes passed the tests in the
>>> same environment (at least once ... unless it was a fluke).
>>
>> I tried with a 32-bit kernel and 32-bit armhf userland with the same
>> results; math.test fails when built with "./configure --with-bootstrap"
>> and succeeds with the gcc-only build without --with-bootstrap passed.
>
> Thanks for your bug report.
>
> It took me some time, but yes, I'm able to reproduce it.  The result
> seems to be non-deterministic for mes-mescc.

Well, at least it is not just me!


>> I presume this is a difference in mes-mescc vs. mes-gcc ...
>
> Mes-0.24 can also be compiled with M2-Planet.  Both mes-gcc and mes-m2
> are OK, but mes-mescc's answer seems to vary between mes-mescc runs
> (note: not per call!).  See below.
>
> It looks like a pretty interesting bug.  Danny, would you like to have a
> look?

Oh fun!

I might end up uploading mes to Debian with only mes-gcc on armhf for
now, as mes-mescc seems to work fine for i386/x86 last I tried...


live well,
  vagrant

signature.asc
Description: PGP signature