Hi,
thanks a lot for being so diligent.
On 12/16/25 11:47, Adrian Bunk wrote:
I am closing this as it prevents migration and is actually resolved.
The bug was caused by incompatibilities in scotch 7.0 in earlier packages.
These are resolved now and the packages
builds with the current version 7.0.7-1.
Everything builds now, but the following looks like a regression to me:
https://buildd.debian.org/status/fetch.php?pkg=opm-simulators&arch=amd64&ver=2025.10%2Bds-4&stamp=1765648983&raw=0
...
-- Found PTScotch: /usr/include/scotch
-- Found Scotch: /usr/include/scotch
-- Found METIS: /usr/lib/x86_64-linux-gnu/libmetis.so
-- Looking for parmetis.h
-- Looking for parmetis.h - found
-- Found ParMETIS: /usr/include/parmetis
...
CMake Warning at
/usr/lib/x86_64-linux-gnu/cmake/opm-grid/opm-grid-config.cmake:365 (message):
Incompatible value "1" of variable "IS_SCOTCH_METIS_HEADER"
Call Stack (most recent call first):
/usr/share/opm/cmake/Modules/OpmFind.cmake:147 (find_package)
/usr/share/opm/cmake/Modules/OpmPackage.cmake:195
(find_and_append_package_to)
opm-simulators-prereqs.cmake:69 (find_package_deps)
CMakeLists.txt:167 (include)
...
I think this is due to different Debian package versions of scotch being
used for opm-grid (7.0.10-2) and opm-simulators (7.0.10-7).
The former did not install metis.h in /usr/include/metis while the
latter does.
While I don't think that there is an big problem with this, we should
probably still recompile opm-grid to have both modules use
the same defines.
The define is uses in a header and source file, but the API should be
unchanged. It will skip some better setting even if scotch is found.
Can someone trigger a binary rebuild of opm-grid or do we need an
unchanged source upload with a version bump?
Best,
Markus
