Hi,
my last response to the bug was meant for a different bug, sorry for
that.
Am 04.11.2025 10:13 schrieb Graham Inggs:
On Tue, 4 Nov 2025 at 08:32, <[email protected]> wrote:
with the attached patches Jochen and I could successfully separated
the
Kokkos (v4.5.1) package from the Trilinos (v16.1.0) package.
What problem are you trying to solve?
The main idea of the provided diffs is to resolve this bug so that the
Kokkos package
can transition to testing.
Why not use kokkos that is already packaged in trilinos?
The motivation for separating the Kokkos package from the Trilinos
package would
be that Kokkos is developed independently of Trilinos. That would be
more obvious
if a libkokkos-* package is available. That could also be achieved with
a virtual
package or using e.g. 'Provides: libkokkos' in the
'libtrilinos-kokkos-package'.
CMake projects which rely on 'find_package(Kokkos)' also works with
'libtrilinos-kokkos-dev'.
I have just tested that.
In favor of a separate Kokkos package would be that Kokkos is released
more frequently
than Trilinos. Making the latest version of Kokkos available in unstable
and testing is easier
if Kokkos is available as a separate package. Trilinos could be build
using
'Tpetra_IGNORE_KOKKOS_COMPATIBILITY=ON'.
If the CI/buildd does not fail the Kokkos package could be accepted.
Am 4.11.2025 06:56 schrieb Christian Alexander Glusa
On Tue, 04 Nov 2025 07:32:26 +0000, Unnamed wrote:
Should the same approach be taken for Kokkos Kernels?
To keep treat Kokkos and Kokkos Kernels consistently that seems
reasonable to me.
Probably one of the maintainers can say more about this.
Best
Nils
