[ecodiv]

PR fixes #8062

t.rast.mapcalc's documented semantic-label band references (inputs=strds.S2_8,strds.S2_4) regressed on main and fail with:

ERROR: Space time dataset <strds.S2_8> of type <raster> not found

The issue was with:

_parse_id() (added in #7670) splits name.semantic_label only when stds_type == "strds". But the call path is
t.rast.mapcalc → dataset_mapcalculator(..., "raster", ...) → open_old_stds(..., "raster") → _parse_id(stds_type="raster"), so the .label is never split off. In 8.4 the split was unconditional, so it worked.

This can hopefully be fixed with:

Accept the raster map element types ("raster"/"rast") alongside "strds", consistent with try_get_stds(), which already maps them to a raster STRDS. 3D-raster and vector STDS stay excluded (no semantic labels).

Added doctests for _parse_id and registered open_stds in the temporal doctest suite. Minimal reproducer (fails before, passes after):

r.mapcalc expression="nir = 100"; r.support map=nir semantic_label=S2_8
r.mapcalc expression="red = 50";  r.support map=red semantic_label=S2_4
t.create type=strds temporaltype=absolute output=test title=t description=t
# register nir,red at 2020-01-01, then:
t.rast.mapcalc inputs=test.S2_8,test.S2_4 output=ndvi basename=ndvi \
  expression="float(test.S2_8 - test.S2_4) / (test.S2_8 + test.S2_4)"

Disclaimer: created with help of Claud
But also tested it manually on my own dataset. With this fix, it works again.

[petrasovaa]

Caught by AI review - since parse_id is also used in check_new_stds, one more suggestion to restore 8.4 behavior when user outputs a dataset name with a dot:

In check_new_stds, treat a non-None semantic label as an error:

name, mapset, semantic_label = _parse_id(name, stds_type=stds_type)
if semantic_label:
    msgr.fatal(
        _("Illegal dataset name <{}>. Character '.' not allowed.").format(name)
    )

The message should quote the original full name, so capture it before the parse.

[ninsbl]

Sorry for breaking it and thanks for fixing.