There seem to be a number of issues with the recent GitHub repositories for METIS and ParMETIS published by KarypisLab. See the following and the issues referencing them:
In particular, some breaking changes have not been propagated correctly to downstream dependencies which leads to unanticipated behavior or breakage (MUMPS, SuperLU_DIST, and STRUMPACK all use METIS reordering). The consensus seems that package managers are sticking with the old v5.1.0 of METIS.
This PR reverts the METIS/ParMETIS builds to use the same versions used by PETSc:
These include a number of patches (https://github.com/KarypisLab/METIS/issues/6) and are widely tested to give Palace users more confidence. We also include a couple of minor patches on top of the PETSc forks to silence compiler warnings, including updating the minimum CMake version, fixing ambiguous indentation issues, resolving a bug where abs was called instead of fabs for a floating point argument, and fixing a potential buffer overrun in file IO.
There seem to be a number of issues with the recent GitHub repositories for METIS and ParMETIS published by KarypisLab. See the following and the issues referencing them:
In particular, some breaking changes have not been propagated correctly to downstream dependencies which leads to unanticipated behavior or breakage (MUMPS, SuperLU_DIST, and STRUMPACK all use METIS reordering). The consensus seems that package managers are sticking with the old v5.1.0 of METIS.
This PR reverts the METIS/ParMETIS builds to use the same versions used by PETSc:
These include a number of patches (https://github.com/KarypisLab/METIS/issues/6) and are widely tested to give Palace users more confidence. We also include a couple of minor patches on top of the PETSc forks to silence compiler warnings, including updating the minimum CMake version, fixing ambiguous indentation issues, resolving a bug where
abswas called instead offabsfor a floating point argument, and fixing a potential buffer overrun in file IO.