[saket0187]

Description

This PR fixes a duplicated command in the history of t.rast.neighbors
output and migrates its shell test & gunittest to pytest.

AI summary

Fix

Running with --overwrite but without -e on an existing output STRDS
recorded the command twice in the dataset history (two identical entries with
the same timestamp):

t.rast.neighbors input=A output=B basename=b
t.rast.neighbors input=A output=B basename=b --overwrite
t.info -h input=B

The history update ran whenever the output existed before the run
(if output_exists:). Without -e, --overwrite deletes and recreates the
dataset, and creating it already records the command, so it was appended a
second time. The condition now matches the branch that reopens the existing
dataset (output_exists and flags["e"]).

Only overwriting without -e changes. I ran every combination of new and
existing output, with and without -e and --overwrite, before and after the
change: exit codes, registered maps, and all other histories are identical.

Shell test migration

test.t.rast.neighbors.sh ran the tool with several methods, sizes, nprocs
values, and with and without -n, but did not check any results. It is
replaced by pytest tests in temporal/t.rast.neighbors/tests/ which use small
deterministic data:

• test_method: average, maximum, and minimum give the expected value for
  each map, with maps processed in parallel (nprocs=2)
• test_null_maps: a result with only null cells is registered with -n,
  otherwise it is not registered and its raster is removed
• test_overwrite_records_command_once: regression test for the fix
  (fails without it: assert 2 == 1)
• test_extend_appends_command: extending keeps the history and adds the new
  command

Implemented with the assistance of Claude.
@ninsbl

[ninsbl]

Good to see you are active also after GSoC @saket0187
I like that you replace the shell based smoke-tests with proper result tests. I do not think it is necessary to test all options of r.neighbors also in t.rast.neighbors, but it would be good if you could test the t.rast.neighbors specific parts (like setting semantic labels, e-flag or suffix or the like)...

[petrasovaa]

Can you please check whether this could be faster? Module-scoped fixture might be better. Also there is already gunittest, so I wonder if there is any unnecessery overlap.

[saket0187]

Can you please check whether this could be faster? Module-scoped fixture might be better. Also there is already gunittest, so I wonder if there is any unnecessery overlap.

With these changes, the tests run about twice as fast. I also removed the whole gunittest and moved its remaining checks to pytest, so everything it tested is now covered in one way or another.

[petrasovaa]

Thanks!