* Build-Depends: Drop versioned constraint on fftw-dev and libgsl-dev.

Changes-By: deb-scrub-obsolete

This commit updates NEWS, ChangeLog and configure
related files for version 0.3.19.

This commit deletes the original version of 'src/transform.f90',
and updates 'src/Makefile.am' to copy either
'src/transform_single.f90' or 'src/transform_double.f90' into
place as a temporary copy of 'src/transform.f90', just before
compiling, depending on whether '--enable-single' is selected or
not, respectively, at the './configure' step. This should
fix the compile bug #957565 reported in Debian [1], according
to which compilation is successful with gcc-9, but fails with
gcc-10, which is fussier about old-style single-vs-double
precision hacks.

[1] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=957565

Since this is presently a side branch, it is safe, during
testing, to replace the original `regression-test-0.3.7.9.sh` by
the new `regression-test-0.3.7.9_haveopenmpi.sh`, and to remove
the latter since it is no longer needed (and available in the git
history).

Since mpgrafic uses an MPI library, any regression test has to
check the MPI environment. The existing regression test has
mainly only been tested on Debian machines.

If the environment sets the environment variable HAVE_OPENMPI to
the one-character string `1`, then this experimental new version
of the test, `regression-test-0.3.7.9_haveopenmpi.sh`, should
run the test assuming that openmpi is available. Any other cases
should run as before. If this script appears to work, then it
will replace the old one.

modified:   debian/changelog
modified:   debian/compat
modified:   debian/control

See ChangeLog, NEWS for updates; TODO includes a comment on LFS
risks in builds on i386.

Allow optionally using an input file named on the command line to
replace reading from standard input. This brings mpgrafic more into
line with what users might expect, and bypasses the openmpi/hurd bug,
in which openmpi cannot handle reading from standard input:
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=919335 .  The
regression test has been shifted to using the filename as a file
instead of as standard input, so that automated testing will not block
the installation of mpgrafic on Debian GNU/Hurd.

Some of the fortran source files need to wait until others are
compiled before they themselves can be compiled, so this info is now
added to src/Makefile.am, so that `make -j N` (to allow N>1 processors)
should work.

The word 'mesh' in context of cosmological structure formation simulations
risks confusion - particle and mesh simulations are fundamentally different
techniques, even though they're often hybridised together, so 'mesh' is
removed in this commit (only present in one obsolete comment and one
non-obsolete comment). A bit more explanation of what "grid" means has been
added.

debian/rules: rm `--with autotools-dev` since
dh_update_autotools_config is now run by default by `dh`;
override_dh_auto_test made conditional, since `nocheck` in
DEB_BUILD_OPTIONS has to allow builds without running tests;

debian/patches/series: add 02-force-regression-test-fail.patch to the
patch series in commented form; this patch is present for convenience
for maintainers/developers who wish to neatly force the regression
test to fail (this is explained in the description of the patch);

debian/control: rm redundant line 'Testsuite: autopkgtest' because
debian/tests/control exists; standards-version bumped up to 4.3.0;
vcs-* s/alioth/salsa/g

The top Makefile.am needs a line with ACLOCAL_AMFLAGS = -I m4,
and the 2.0.2-2 version of openmpi doesn't seem to have
the symlink /usr/bin/mpicc to point to mpicc.openmpi . So a safe
hack is to add mpicc.openmpi as an alternative binary, which is
done in this commit.

As explained in guides such as
http://freesoftwaremagazine.com/articles/brief_introduction_to_gnu_autotools/ ,
the preferred method of including local additions of m4 scripts is
to put them in a subdirectory called m4/ . That's done in this commit.

This reverts commit 891b94aa,
because it inserted a bug instead of fixing one.

since size_t can be 32-bit even on 64-bit machines;
HAVE_USE_MPI_F08 preprocessor macro test added for "use mpi_f08" in .f90 files,
with fallback to "include 'mpif.h'"