configure: switch hashlib usage from md5 to sha256

and avoid "usedforsecurity = False" as its incompatible with Python <3.9.

Replace -with-mpi-f90module-visibility with -with-mpi-f90module=mpi_f08

Add PETSC_INT_KIND and PETSC_MPIINT_KIND

Reported-by: M.Weiland@epcc.ed.ac.uk

Use the same getAPI.py and generatefortranbindings.py scripts for both PETSc and SLEPc

Includes some cleanup of the scripts

checkbadSource: check style of config/*

configure: add --with-devicelanguage=(C|C++)

Fix configure test for `cxxabi.h`

The `cxxabi.h` header is C++, so use the C++ compiler to check for it.

Merge remote-tracking branch 'origin/release'

configure: test if fortran pointers can be initialized in the declaration

ensure that Fortran Bool is interoperable

The kind C_BOOL for logical from ISO_C_binding only defines the size of
the logical as 1 Byte. It does not define which values (bit patterns) are
used to en

ensure that Fortran Bool is interoperable

The kind C_BOOL for logical from ISO_C_binding only defines the size of
the logical as 1 Byte. It does not define which values (bit patterns) are
used to encode true and false. In stdbool.h from C, true is defined as 1
and false as 0, but these values are not guaranteed by the Fortran standard.

There are (at least) two contemporary Fortran compilers that use a
difference convention: Intel Fortran (ifort and ifx) and NVIDIA Fortran
(nvfortran, former PGI). Both represent true as all bits set, i.e. -1 for
a signed integer or 255 for an unsigned integer.

Intel Fortran can be made interoperable with the "-standard-semantics" or
"-fpscomp logicals". The latter is now set by default to ensure
interoperability while avoiding effects such as modified name mangling
rules and performance degradations due to the former.

NVIDIA Fortran can be made interoperable with the "-Munixlogical" flag.
This is set manually in the "linux-cuda-single-cxx" pipeline. This is
now set by default.

show more ...