test: add unit tests for stdlib/requirements (#814)

[planetf1]

Misc PR

Type of PR

• Bug Fix
• New Feature
• Documentation
• Other

Description

• Link to Issue: Fixes test: unit tests for stdlib/requirements (tool_reqs, requirement, md) #814

36 unit tests for stdlib/requirements/ — pure validation logic, no model or backend needed. Runs in ~6s.

Why these tests exist

mellea/stdlib/requirements/ was at 42% line coverage from the full HPC test suite (all backends, 1034 tests). However much of that came from e2e tests gated behind ollama markers that never run in CI. The unit-only coverage for the three target files was far lower.

These are all deterministic functions — parsing, validation, type coercion — where unit tests provide high regression value. As with the granite formatter tests (#818), these won't pick up new issues necessarily, but will help prevent us from unintentionally breaking the code.

Coverage change (unit/integration tests only)

File	Stmts	Before	After
tool_reqs.py	44	18%	99%
requirement.py	45	~50%	100%
md.py	45	~30%	96%

Excluding guardian.py (deprecated, 0%, out of scope), the active requirements module is now at 96% from unit tests alone. For reference the full HPC run (including e2e) had the whole module at 42%.

What's covered

File	Tests	Covers
test_reqlib_tools.py	14	_name2str edge case, uses_tool (present/absent/no calls/callable/check_only), tool_arg_validator (valid/failed/missing tool/missing arg/no calls/all-tools mode)
test_requirement.py	17	requirement_check_to_bool (threshold logic, missing key, invalid JSON), reqify/req/check shorthands, simple_validate with None output, LLMaJRequirement, ALoraRequirement init
test_reqlib_markdown.py	15	as_markdown_list edge cases (paragraph, mixed content, empty, single item), _md_list/_md_table validation wrappers, table edge cases

Marker fix

test_requirement.py had a module-level pytestmark = [pytest.mark.ollama, pytest.mark.e2e] that incorrectly gated the existing simple_validate unit tests behind ollama. Moved to per-function decorators on the two async tests that actually need them.

Design decisions

• Real ChatContext + ModelOutputThunk objects with canned data — no mocking of the context pipeline
• ModelToolCall.func uses unittest.mock.Mock (only the func field, not the data structures)
• ALoraRequirement tests patch Intrinsic.__init__ to avoid hitting the adapter catalogue

Testing

• Tests added to the respective file if code was changed
• New code has 100% coverage if code as added
• Ensure existing tests and github automation passes (a maintainer will kick off the github automation when the rest of the PR is populated)

[github-actions]

The PR description has been updated. Please fill out the template for your PR to be reviewed.

[psschwei]

Solid test PR — the marker fix alone is worth merging, and the 36 new unit tests are well-structured with good edge case coverage. Two minor suggestions below, neither blocking.

[psschwei]

Suggestion (non-blocking): There's a gap in tool_arg_validator coverage when tool_name=None and none of the tool calls contain the target arg_name. The production code silently returns True in that case (the for loop finishes without failing), which is arguably a latent bug. A test documenting this would be useful:

def test_tool_arg_validator_no_tool_name_arg_missing_everywhere():
    ctx = _ctx_with_tool_calls({"tool_a": _make_tool_call("tool_a", {"y": 5})})
    req = tool_arg_validator("x must be positive", tool_name=None, arg_name="x", validation_fn=lambda v: v > 0)
    result = req.validation_fn(ctx)
    assert result.as_bool() is True  # documents current (possibly surprising) behavior

Could be a follow-up issue too if you'd rather not expand scope here.

---

While you're here: there's a typo "Valiudation" in tool_reqs.py (appears twice in error reason strings around lines 109 and 119). Not introduced by this PR, but easy to fix if you're touching this area.