Current state

tgis.init() resolves the session with gs.gisenv() and no env, then caches it in module-level globals in core.py (current_mapset, current_location, current_gisdbase, tgis_backend, tgis_database*, message_interface, c_library_interface, ...). None of the 64 grass.script calls in python/grass/temporal/ (50 run_command, 7 parse_command, 5 gisenv, 2 read_command) pass env=.

Elsewhere in the package the environment can be explicit: gs.run_command(env=...), Tools(session=session), gs.setup.init(..., env=...).

Describe the solution you'd like

tgis.init(env=session.env)
dbif = tgis.SQLDatabaseInterfaceConnection(env=session.env)

with env threaded through to the grass.script calls underneath. env=None would keep reading os.environ as it does now, so existing code and the global session stay valid.

It would also make it possible to use more than one session in a process, which the rest of the API allows.

Scope

This is not a small change. At minimum it needs:

• init() to accept env and not rely on gs.gisenv() against the global environment,
• the session carried on the objects which currently rely on the globals (SQLDatabaseInterfaceConnection, the AbstractDataset hierarchy, ...) rather than module-level state in core.py,
• env passed at the 64 grass.script call sites,
• the RPC-backed interfaces (CLibrariesInterface, the messenger), which spawn subprocesses that need the environment too.

So it probably wants to be staged.

Describe alternatives you've considered

A test which needs a session outside the global environment can mirror it into os.environ with monkeypatch, which is what #7732 does. It works, but every caller has to repeat it.

Additional context

This came up while debugging r.coin on the OSGeo4W CI (#7732).