Use the components for field identifiers of a variable of type
`shp_triplet` instead of storing these field identifiers in variables
of modules `init_shapefiles_m` and `read_field_indices_m`. The dummy
argument hshp of procedure `read_field_indices` becomes
`intent(inout)`.

Create derived type `shp_triplet`. Motivation: we need to manipulate
two triplets of shapefiles in program `eddy_graph`. Use everywhere a
variable of type `shp_triplet` instead of the three variables
`hshp_extremum`, `hshp_outermost`, `hshp_max_speed` when all three of
them appear. Rename character variable `shp_triplet` to
`shp_triplet_dir`.

In procedure `dispatch_snapshot`, rename dummy argument m to rank,
clearer.

Rename program `test_get_snapshot` to `test_get_dispatch_snap` and
make it call `dispatch_snapshot` instead of `send_snapshot`. The only
difference in result is that the program now writes files
`isolated_nodes.txt` and `number_eddies.csv`.

Check in program `test_overlap` that it is run on a single MPI process.

Change order of dummy arguments of procedure `get_snapshot` to make it
more convenient to call with keywords. Rename dummy argument m to
rank, clearer.

Bug fix in `short_tests_.json`: run `test_read_snapshot` and
`test_overlap` with `mpiexec -n 1`, else the programs may not be able
to read standard input.

Add scripts to compare with Matlab output and script to concatenate shapefiles.

On our way to a generalization of script `filter.py`: use a function
to define the filter.

In script `plot_snapshot.py`, accept a triplet of shapefiles without
`ishape_last.txt`.

Add a test for program `test_get_snapshot` where one of the processes
does a reception in procedure `get_snapshot`.

Bug fix in main unit `test_get_snapshot`: we send and receive date k.

Add script for concatenation of triplets of shapefiles.

Generalize script `plot_snapshot.py` to read a single date in a
triplet of shapefiles containing several dates.

Bug fix in `Tests/CMakeLists.txt`: missing source file.

Include four dates in `Tests/Input/Region_1/SHP_triplet_noise`, in
preparation for a test with `test_get_snapshot`. Main unit
`test_nearby_extr` must then read the number of shapes for the first
date in `ishape_last.txt`.

Prepare for a run of `test_get_snapshot` which receives a
snapshot. Generalize main unit `test_get_snapshot`: it is now
parallelized; it reads a date index from stdin; it sends a snapshot
from process 1 to process 0 if it is run with 2 processes.

The logic of the overlap program will be that all processes access the
triplet of input shapefiles. So modify subroutine
`read_field_indices`: process 0 reads field indices and broadcasts
them. This implies that programs `test_overlap`, `test_read_snapshot`
and `test_read_eddy` use MPI.

Generalize `algorithm.py`.

In program `test_send_recv`, read input SHP triplet from command line
instead of requiring it to be present in the current directory with
the name `SHP_triplet_old`. Generalize `test_send_recv`: read
`ishape_last` from file. In `test_send_recv`, all processes must now
access the triplet of shapefiles since process 0 reads field
indices. For the sake of uniformity, read and broadcast k1 from
process 0.

First compilation and test of procedure `get_snapshot`. Change order
of dummy arguments of procedure `get_snapshot`: make it more
convenient for call with keywords. Replace dummy arguments periodic
and `dist_lim` by dummy argument copy. Add dummy arguments
`ishape_last` and k1. Promote `hshp_extremum`, `hshp_outermost`,
`hshp_max_speed` from local variables to dummy arguments.

In program `test_read_snapshot`, read input SHP triplet from command
line instead of requiring it to be present in the current directory
with the name `SHP_triplet_old`.

Remove directory `Tests/Input/Region_1/SHP_triplet`, which was not used.

Change of idea: make date index an intent in argument of procedure
`read_snapshot`. Replace dummy arguments `ishape_first` and scalar
`ishape_last` by k1 and array `ishape_last`. It is now the task of
`read_snapshot` to find the ishapes to read. This seems clearer.

Change of idea: instead of storing `date_index` and `ishape_last` in a
file `contents.csv`, store only `ishape_last` in a file
`ishape_last.txt`. The `date_index` column was useless and it
complicated reading the file. The first value of `date_index` can be
read from one of the three DBF files. Also, to roll with the flow, let
the lower bound of `ishape_last` in main unit `test_overlap` be 1
instead of k1.

Change order of actual arguments of `dbf_read_attribute_03` following
revision of `Shapelib_03`.

Bug fix in procedure `dispatch_snapshot`: the rank of MPI processes is 0-based.

Bug fix: remove `extr_map_outermost.nc`, which was not used since
commit ad420d25.

Organize better the directory `Input`: create sub-directories for regions.

Big change: use SHP triplets that can contain several snapshots. So
program `test_overlap` reads a single SHP triplet. The dates that
`test_overlap` should read are now specified in the input
namelist. SHP triplet directory must now contain a file named
`contents.csv`. `test_overlap` reads `contents.csv` into variables k1
and `ishape_last`. So there is no longer any need to read the number
of entities in one of the shapefiles.

For each region, concatenate SHP triplets for dates 2006/01/01 and
2006/01/02.

Read namelist from process 0 in program `test_send_recv`: it is
simpler since it is the default setting of mpiexec and I do not see
any drawback.

In programs `test_read_snapshot` and `test_send_recv`, just hard code
for now the ishapes we want to read in the SHP triplet.

Generalize `read_snapshot`. Add dummy argments `ishape_first` and
`ishape_last`. This allows getting a snaphot from a triplet of
shapefiles containing several snapshots. Consequence: move the call to
`shp_get_info_03` from `read_snapshot` to `test_overlap`,
`test_read_snapshot` and `test_send_recv`.

Remove opening and closing of shapefiles from procedure
`get_snapshot`. Since a triplet of shapefiles can contain several
snapshots, we do not want to open and close the same shapefiles
several times from the same MPI process.

Pass argument copy to `read_snapshot` rather than arguments periodic
and `dist_lim`. Motivation: do not recompute copy at each call of
`read_snapshot`. `read_snapshot` is going to be called many times in
overlap program, with always the same value of copy. And we do not
need the values of periodic and `dist_lim` separately in
`read_snapshot`.

Complete `send_snapshot` and `recv_snapshot` by sending and receiving
an entire snapshot. So we need nlon, nlat and copy in `recv_snapshot`
to allocate `extr_map`. It is better to pass arguments nlon, nlat and
copy than sending and receiving the shape of `extr_map` with MPI, of
course, since the shape is the same for all snapshots. In program
`test_send_recv`, write the entire received snaphot.

Bug fix in `Tests/CMakeLists.txt`: several targets missed
`${netcdff_INCLUDE_DIR}`.

Create procedure `write_snapshot` from code in main program
`test_read_snapshot`, because we need to do this also in program
`test_send_recv`. Note the dissymmetry with `read_snapshot`, which
does not include the opening and closing of the shapefiles. That is
because we anticipate calling `read_snapshot` several times with the
same shapefiles in program overlap.

Change order of dummy arguments of procedure `write_extr_map`. More
convenient for calling it with keywords from `write_snapshot`.

Simplify cmake files, taking advantage of improved packaging of
libraries. This also much simplifies the configuration process: the
user typically has only two directories to specify: `CMAKE_PREFIX_PATH` and
`fortrangis_INCLUDE_DIR`. Also `Shapelib_03` no longer depends on EZMPI,
since revision 129a0cc.

Add test for `test_send_recv`.

We need the dummy version of EZMPI for sequential programs linking
with `Shapelib_03`.

Replace `include_directories` by `target_include_directories` in
`Tests/CMakeLists.txt`. Motivation: some targets do no use ezmpi. This
implies adding directories to `target_include_directories` for
`extraction_eddies`.

In `send_snapshot` and `recv_snapshot`, send and receive a whole
eddy. Read and write a snapshot in `test_send_recv`.

In procedures `recv_snapshot` and `send_snapshot`, use library EZMPI
instead of MPI directly.

Rename `receive_snapshot` to `recv_snapshot` for consistency with MPI
calls. In procedure `send_snapshot`, invert order of dummy arguments
tag and dest for consistency with MPI calls.

Fill in procedures `receive_snapshot` and `send_snapshot`. This is
just a beginning with part of an eddy. Add test program.

hsp arguments in procedure `read_snapshot` can be declared with
`intent(in)` since they are `intent(in)` in `read_eddy` since commit d815bf76.

Bug fixes in procedure `get_snapshot`: remove duplicated declaration
of nlon, nlat; add argument `dist_lim`, which must be passed to `read_snapshot`.

In procedure `get_snapshot`, read the triplet of shapefiles from a
directory named with the date index (for the now moment, we will see
later for something more efficient). Check that the date index in the
shapefiles is the date index required.

In procedure `get_var`, take advantage of new procedure `nf95_get_missing`.

Update input files for program `test_overlap` because the program now
reads `delta_in` and `delta_out`.

Add a test for the logic of `delta_in` and `delta_out` in subroutine
overlap. Add corresponding input snapshots.

Add possiblity of specifying `delta_in` and `delta_out` in program
`test_overlap`.

Add possibility to renumber eddies in script `filter.py`.

Read `SHP_triplet` directories as arguments in script `filter.py`.

Move directory `Tests/Input/Extraction_eddies_region_5` to
`Tests/Input/Region_5_2006_01_01`.

Bug fix in procedure overlap: `delta_out` and `delta_in` should not increase.

In script `eddy_dump.py`, do not assume that there is a single date
index. So find ishape by iterating through values of date index and
eddy index.

Add a test of overlap in `short_tests.json`.

In procedure `write_eddy`, no need of special treatment for
interpolated eddy: the outermost contours and max-speed contours are
defined as null.

Read name of input file on the command line in script `read_overlap.py`.

Add tests with program `test_overlap` in `short_tests.json`.

Bug fix in main unit `test_overlap`: we have to call
`init_interpolated_eddy`.  In main unit `test_overlap`, change the
initial value of `flow(2:max_delta)%number_eddies` for more clarity in
the results. Bug fix in main unit `test_overlap`: valid isolated
eddies of snasphot 2 are in `flow(max_delta + 1)`, not `flow(2)`.

Rename `test_successive_overlap` to `test_overlap`: we will also use
it for non-successive overlap. Accordingly, add an input variable
`max_delta` and change the size of flow to `max_delta` + 1. We read
from snapshots `flow(1)` and `flow(max_delta + 1)`. For
`flow(2:max_delta)`, we only need to define the component
`number_eddies`.

In tests, remove use of `$test_old_dir`, use data from
`$src_dir/Tests/Input` instead. Motivation: run tests for the first
time more easily.

Change the order of dummy arguments of `write_overlap`.

Add tests for program `test_write_overlap`. In program
`test_write_overlap`, make the eddy identifiers distinct.

Create procedure `init_interpolated_eddy` so that initialization is
not repeated needlessly at each call of `write_overlap`. (Problem was
mentioned in message of commit 10f339d2.)

Use `SHP_triplet` instead of Snapshot for the name of the directory
containing the three shapefiles extremum, `outermost_contour` and
`max_speed_contour`. This is meant as a general name, analogous to a
new file type. Generally, the shapefiles may contain eddies at
different dates, not just a snapshot.

Add options for the NAG compiler.

Replace `target_include_directories` by `include_directories` in
`Tests/CMakeLists.txt`. The file was becoming too long.

Remove unused items in use statements.

Bug fix in `Tests/Input/successive_overlap_region_5_nml.txt`: remove a comma.

Bug fix in procedure overlap: procedure `interpolate_eddy` does not
exist any longer.

In tests, remove use of `$test_old_dir`, use data from
`$src_dir/Tests/Input` instead. Motivation: run tests for the first
time more easily.

Change order of dummy arguments of procedure
`write_overlap`. Motivation: call with keywords in
`test_write_overlap`.

Add main program `test_write_overlap` and corresponding tests in
`short_tests.json`.

New script `triplet_dump.sh`.

Add `.cmake` files. The idea is to have a build system independent of
the environment. This allows us to have a customized version of
`ConfigureCompilerFlags.cmake` without the `-ffpe-trap=invalid`
option.

Move the inner part of procedure overlap to a separate procedure
`write_overlap`. overlap was too complex with 6 nested constructs. The
downside is that the initialization of `e` is repeated for every
non-successive overlap. If necessary, we can later add an
initialization on first entry (but an initialization at declaration is
not possible). `write_overlap` has some consistency: all the arguments
are `intent(in)`, scalars or small vectors of intrinsic type, and
`write_overlap` only writes to files. All modifications to flow are
done outside of `write_overlap`.

Remove procedure `interpolate_eddy`. I realized that the longitude of
one of the overlapping eddies must be shifted. It seemed awkward to
pass variable m from overlap as an argument to
`interpolate_eddy`. Passing separately the useful components of
overlapping eddies made the procedure bulky, not worth creating for
three lines of computation.

As a consequence, inline the interpolation of eddies in procedure
overlap. An advantage is that part of the processing that was in
`interpolate_eddy` can migrate out of the interpolation loop.

As another consequence, remove program `test_interpolate_eddy` and
correponding test.

Upgrade procedure `candidate_overlap` to deal with non-successive
overlap. Add dummy argument delta.