Re: Reproducible Research Hackathon: Friday, July 3rd

[Ludovic Courtès]

Hi Konrad,

Konrad Hinsen <konrad.hinsen@fastmail.net> skribis:

> Indeed. The details are here:
>
>   https://gitlab.inria.fr/guix-hpc/guix-past/-/issues/1

Oooh, thank you!  It looks like an “interesting” bug, one of those that
can help make the case for precise software environment control.  :-)

bonfacemunyoki@gmail.com (Bonface M. K.) skribis:

> That's strange. To get the right results, you'd have to do a `2L ** 64`.
> When I tried `2 ** 63` I got `-9223372036854775808`. There's also an
> overflow error. Here's a snippet of what fails from
> Python-2.4.6/Lib/test:
>
> ```
>     # If this fails, probably using a strict IEEE-754 conforming libm, and x
>     # is +Inf afterwards.  But Python wants overflows detected by default.
>     try:
>         x = math.exp(1000000000)
>     except OverflowError:
>         pass
>     else:
>         raise TestFailed("overflowing exp() didn't trigger OverflowError")
> ```
>
> Maybe there's an overflow somewhere and we'd have to tweak libm? I'm
> speculating though. I'd have to investigate this later.

Uh, weird!  We could check whether building Python with ‘-fwrapv’ helps.
See also <https://lwn.net/Articles/511259/>.

Ludo’.