feat(cli): expose missing EmissionsTracker options in 'codecarbon monitor' (#1273) - #1298

Fixes #1273

Problem

codecarbon monitor --help only exposed 7 options (measure-power-secs, api-call-interval, api, offline, country-iso-code, region, log-level), but EmissionsTracker.__init__ accepts 28+ parameters. Users running from the CLI had no way to set commonly needed options like pue, wue, gpu_ids, or force_cpu_power without dropping down to the Python API or a config file.

Solution

Added 8 new flags to the monitor() command in codecarbon/cli/main.py:

Design decisions

• Unset options are omitted, not passed as None. New options default to None at the CLI level, but are filtered out of tracker_args before reaching EmissionsTracker. This preserves the existing config-file → environment-variable → hardcoded-default precedence chain. A user who sets pue=1.4 in ~/.codecarbon.config but doesn't pass --pue on the CLI will still get 1.4, not None.
• gpu-ids is accepted as a comma-separated string ("0,1") and parsed into a list (["0", "1"]) before forwarding, matching what EmissionsTracker expects.

Tests added

3 new tests in tests/cli/test_cli_main.py:

1. test_monitor_help_lists_new_tracker_options — asserts all 8 flags appear in --help output (uses COLUMNS=200 env to prevent typer from truncating long flag names).
2. test_monitor_offline_forwards_new_options_to_tracker — invokes the full CLI with all new flags set, verifies each value reaches OfflineEmissionsTracker.__init__ with the correct type and value.
3. test_monitor_offline_omits_unset_optional_tracker_args — invokes with only --offline --country-iso-code FRA, verifies none of the 8 new keys appear in tracker_args.

All existing tests continue to pass (27/27 + 3 new = 30 total, the pre-existing test_monitor_run_and_monitor failure is a known Windows-only issue where echo is a shell built-in, not a standalone executable — passes on Linux CI).

Thank you for tackling #1273, and apologies that we ended up with two PRs on the same issue — that is on us for not triaging it sooner.

We are moving forward with #1275, which was opened first and covers the full set of CLI-representable tracker options (28 against the 8 here), plus the interaction between --experiment-id and the stored config.

Two things in your PR are better than what #1275 has, and we would like to carry them over with credit to you: the comma-separated --gpu-ids parsing into a list, and the monitor --help test that asserts the flags actually appear. Are you happy for us to bring those across?

Please do keep contributing — happy to point you at an unclaimed issue if you would like one.