fix(emitter): keep an unset group_id unset when cloning

[LHMQ878]

Which issue(s) does this pull-request address?

Closes: #1581

Description

Emitter.clone() built the clone with the group id passed positionally through str():

cloned = Emitter(
    str(self._group_id),
    ...
)

group_id is str | None and defaults to None, so when it was unset the clone received the literal string "None". _create_event() stamps group_id=self._group_id onto every EventMeta, and child() does group_id=group_id or self._group_id, so the synthetic value propagated to every event emitted by the clone and by every emitter descended from it.

Measured on main (21284d7, Python 3.12.10):

expression	before	after
Emitter(namespace=["app"])._group_id	None	None
(await that.clone())._group_id	'None'	None
cl._group_id is None	False	True
bool(cl._group_id)	True	False
event.group_id from the clone	'None'	None
(await that.clone()).child(...)._group_id	'None'	None
clone with group_id="run-42"	'run-42'	'run-42'

Agents are the common case rather than an edge case. All nine clone() implementations do cloned.emitter = await self.emitter.clone() — the four adapter agents (a2a, acp, agentstack, watsonx_orchestrate), the four core agents (lite, react, requirement, tool_calling), and context.py:281 — and agents build their emitter via Emitter.root().child(namespace=["agent", "tool_calling"], creator=self, events=...) with no group id. Events off a cloned agent looked like this:

run.agent.tool_calling.start     group_id='None'
agent.tool_calling.done          group_id='None'

Anything grouping or filtering by group_id therefore saw a spurious "None" group for cloned agents: if meta.group_id: became true, meta.group_id is None became false, and a dict keyed by group id gained a "None" bucket. Forwarding meta.group_id to a tracing backend recorded the string as-is.

The fix passes the constructor arguments by keyword and drops the coercion, so the absent value stays absent. creator no longer needs the if self.creator else None guard — the attribute is already object | None, so the conditional was a no-op.

The TypeScript emitter does this.groupId = input?.groupId (typescript/src/emitter/emitter.ts:60) and preserves the absent value through child() (:80) and createSnapshot() (:254), so this was a Python-only divergence and no TypeScript change is needed.

Why the existing test missed it

test_clone constructs with group_id="test_group", and a real group id round-trips through str() unchanged, so the only broken path — the default None — was never exercised. Four tests are added next to it:

test	red before this change
test_clone_keeps_absent_group_id_absent	✅ AssertionError: assert 'None' is None
test_clone_emits_events_without_a_group_id	✅ assert ['None'] == [None]
test_child_of_clone_inherits_absent_group_id	✅ AssertionError: assert 'None' is None
test_clone_preserves_group_id	green both ways — regression guard for the real-id path

Checklist

General

• I have read the appropriate contributor guide: Python
  / [ ] TypeScript
• I have signed off on my commit: Python instructions / TypeScript instructions
• Commit messages and PR title follow conventional commits
• Appropriate label(s) added to PR: Python for Python changes, TypeScript for TypeScript changes — I cannot set labels on this repo; please add Python.

Code quality checks

• Code quality checks pass: ruff check and ruff format --check are clean on both changed files. mypy beeai_framework/emitter/emitter.py reports only four pre-existing findings unrelated to this diff (missing deprecated/cachetools stubs, utils/schema.py:67, backend/chat.py:489) — identical with the change reverted.

Testing

• Unit tests pass — python -m pytest -m unit, run on Windows/Python 3.12.10:

  	before	after
  passed	290	293
  failed	4	1
  skipped	4	4

  The 3 resolved failures are the new tests. The 1 remaining failure is the same in both runs and unrelated: tests/tools/filesystem/test_read_edit.py::test_tools_route_through_installed_backend fails with ToolInputValidationError: 'path' must be absolute, got: '/tmp/virtual.txt' — a POSIX path that is not absolute on Windows, so it is a platform artefact of my environment, not a regression.

  Two collection paths are excluded in my environment and in both runs equally: beeai_framework/adapters/transformers and tests/backend/providers/test_transformers.py, which fail to import with ModuleNotFoundError: No module named 'outlines_core.fsm' from the installed outlines package.

• E2E tests pass: mise test:e2e — not run; they require model-provider credentials I do not have. This change touches no provider code path.

• Tests are included (for bug fixes or new features)

Documentation

• Documentation is updated — no doc change needed; the fix restores the documented str | None contract rather than altering it.
• Embedme embeds code examples in docs — no examples touched.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.