[bug#40691] [PATCH v3 1/1] gnu: Add gromacs.

[Ludovic Courtès]

Hi Vincent,

Vincent Legoll <address@hidden> skribis:

>> It may be reasonable to just skip it.
>
> Hope you don't mind trading a test suite failure for a (harmless)
> test suite build warning...
>
> How's the crude disabling in the attached patch v3 look to you ?
>
> And does it fix the failure ?
> This test is not failing here (kvm on ryzen).

Oh that’s guix-daemon inside QEMU/KVM?  Then what can happen, unlike on
real hardware, is that hwloc doesn’t discrepancies between the actual
number of cores and the info it gathers from /sys or whatever.

> From 660371d5a3e7c7c763fc123357ddd6c31f730553 Mon Sep 17 00:00:00 2001
> From: Vincent Legoll <address@hidden>
> Date: Sat, 18 Apr 2020 00:14:56 +0200
> Subject: [PATCH] gnu: Add gromacs.
>
> * gnu/packages/bioinformatics.scm (gromacs): New variable.

[...]

> +         (add-after 'unpack 'disable-hwloc-test
> +          (lambda _
> +             ;; This test warns about the build host hardware
> +             (substitute* "src/gromacs/hardware/tests/hardwaretopology.cpp"
> +               (("TEST\\(HardwareTopologyTest, HwlocExecute\\)")
> +                "void __guix_disabled()"))

I was thinking it’d be nice to use whatever mechanism gtest has to mark
a test as skipped but I didn’t find anything, so it’s probably OK like
this.

However, there’s another issue I hadn’t noticed (apologies!):
src/external bundles a few things like googletest, tinyxml2, zlib
(indirectly), etc.

Could you check if we can use our own dependencies instead of the
bundled ones?

For gtest it’s good if we can unpack the source of our ‘googletest’
package in place of the bundled copy, but it’s optional (I think there
are cases where we found this wasn’t possible for some reason).

Thank you!

Ludo’.