hack/tools/scripts/llm-key-hunter/results/_cache/content/chayapats_jarn.json

{"repo": "chayapats/jarn", "files": {"tests/test_providers_extra.py": {"1ad4cbfb15914a5505cc5e0ebc8d7d27d983f377": {"t": 1786068440.871415, "n": 18709, "h": "68dc9cc0a48c1755", "c": "\"\"\"Extended provider coverage (groq/deepseek/xai openai-compatible; google/mistral).\"\"\"\n\nfrom __future__ import annotations\n\nfrom unittest.mock import patch\n\nimport pytest\n\nfrom jarn.config.schema import Config, ProviderConfig, ProviderType, RoutingConfig\nfrom jarn.providers import ModelFactory\n\n\ndef _factory(ptype, **prov):\n    cfg = Config(\n        default_profile=\"p\",\n        providers={\"p\": ProviderConfig(type=ptype, **prov)},\n        routing=RoutingConfig(main=\"p/model\"),\n    )\n    return ModelFactory(cfg)\n\n\n@pytest.mark.parametrize(\"ptype,base\", [\n    (ProviderType.GROQ, \"https://api.groq.com/openai/v1\"),\n    (ProviderType.DEEPSEEK, \"https://api.deepseek.com\"),\n    (ProviderType.TOGETHER, \"https://api.together.xyz/v1\"),\n    (ProviderType.FIREWORKS, \"https://api.fireworks.ai/inference/v1\"),\n    (ProviderType.XAI, \"https://api.x.ai/v1\"),\n    (ProviderType.OPENAI_COMPATIBLE, \"http://localhost:8000/v1\"),\n])\ndef test_openai_compatible_providers(ptype, base):\n    factory = _factory(ptype, api_key=\"k\", base_url=base)\n    with patch(\"langchain.chat_models.init_chat_model\") as m:\n        m.return_value = object()\n        factory.build(\"p/model\")\n    assert m.call_args.kwargs[\"model_provider\"] == \"openai\"\n    assert m.call_args.kwargs[\"base_url\"] == base\n    assert m.call_args.kwargs[\"api_key\"] == \"k\"\n    # Ask for streamed usage so cost/token tracking isn't blind on servers (LM\n    # Studio, vLLM, …) that omit usage unless stream_options.include_usage is set.\n    assert m.call_args.kwargs[\"stream_usage\"] is True\n\n\n@pytest.mark.parametrize(\"ptype,provider_str\", [\n    (ProviderType.GOOGLE, \"google_genai\"),\n    (ProviderType.MISTRAL, \"mistralai\"),\n])\ndef test_dedicated_providers(ptype, provider_str):\n    factory = _factory(ptype, api_key=\"k\")\n    with patch(\"langchain.chat_models.init_chat_model\") as m:\n        m.return_value = object()\n        factory.build(\"p/model\")\n    assert m.call_args.kwargs[\"model_provider\"] == provider_str\n    # stream_usage is an OpenAI-only kwarg — don't pass it to other model classes.\n    assert \"stream_usage\" not in m.call_args.kwargs\n\n\ndef test_all_provider_types_have_defaults():\n    from jarn.config.defaults import ALL_PROVIDERS, DEFAULT_MODELS, PROVIDER_ENV_VARS\n\n    for p in ALL_PROVIDERS:\n        assert p in DEFAULT_MODELS, f\"{p} missing default models\"\n    # Every cloud provider should suggest an env var.\n    from jarn.config.defaults import CLOUD_PROVIDERS\n    for p in CLOUD_PROVIDERS:\n        assert p in PROVIDER_ENV_VARS\n\n\n# -- P2.B: suggest_slug did-you-mean ----------------------------------------\n\ndef test_suggest_slug_dot_to_dash_anthropic():\n    \"\"\"Dot-form slug on Anthropic provider yields a dash-form suggestion.\"\"\"\n    from jarn.config.schema import ProviderType\n    from jarn.providers.models import suggest_slug\n\n    result = suggest_slug(ProviderType.ANTHROPIC, \"claude-opus-4.8\")\n    assert result is not None\n    assert \"claude-opus-4-8\" in result\n    assert \"dashes\" in result or \"Anthropic\" in result\n\n\ndef test_suggest_slug_dash_to_dot_openrouter():\n    \"\"\"Dash-form slug on OpenRouter provider yields a dot-form suggestion.\"\"\"\n    from jarn.config.schema import ProviderType\n    from jarn.providers.models import suggest_slug\n\n    result = suggest_slug(ProviderType.OPENROUTER, \"anthropic/claude-opus-4-8\")\n    # The candidate is anthropic/claude-opus-4.8 which matches OR default slugs\n    assert result is not None\n    assert \"4.8\" in result or \"claude-opus-4.8\" in result\n\n\ndef test_suggest_slug_no_match_returns_none():\n    \"\"\"Completely unknown slug returns None (no false positive).\"\"\"\n    from jarn.config.schema import ProviderType\n    from jarn.providers.models import suggest_slug\n\n    result = suggest_slug(ProviderType.ANTHROPIC, \"gpt-5-ultra-xyz-fake\")\n    assert result is None\n\n\ndef test_suggest_slug_already_correct_returns_none():\n    \"\"\"Correct dash-form slug on Anthropic returns None (no spurious suggestion).\"\"\"\n    from jarn.config.schema import ProviderType\n    from jarn.providers.models import suggest_slug\n\n    # \"claude-opus-4-8\" is correct for Anthropic; swapping to \"claude-opus-4.8\"\n    # would be the OR form — but the function should still return something or None\n    # without crashing. We verify it does not return the same slug as suggestion.\n    result = suggest_slug(ProviderType.ANTHROPIC, \"claude-opus-4-8\")\n    # It may or may not suggest the OR form; it must not crash and must not\n    # suggest the same slug unchanged.\n    if result is not None:\n        assert \"claude-opus-4-8\" not in result or \"claude-opus-4.8\" in result\n\n\ndef test_model_resolution_error_includes_suggestion():\n    \"\"\"ModelResolutionError message includes did-you-mean when slug is swappable.\"\"\"\n    from unittest.mock import patch\n\n    from jarn.config.schema import Config, ProviderConfig, ProviderType, RoutingConfig\n    from jarn.providers.models import ModelFactory, ModelResolutionError\n\n    # Set up an Anthropic provider with a dot-form slug (wrong for Anthropic).\n    cfg = Config(\n        default_profile=\"anthropic\",\n        providers={\"anthropic\": ProviderConfig(type=ProviderType.ANTHROPIC, api_key=\"fake-key\")},\n        routing=RoutingConfig(main=\"anthropic/claude-opus-4.8\"),\n    )\n    factory = ModelFactory(cfg)\n\n    def _raise(*a, **kw):\n        raise ValueError(\"model not found\")\n\n    with patch(\"langchain.chat_models.init_chat_model\", side_effect=_raise), pytest.raises(ModelResolutionError) as exc_info:\n        factory.build(\"anthropic/claude-opus-4.8\")\n\n    msg = str(exc_info.value)\n    assert \"did you mean\" in msg\n    assert \"claude-opus-4-8\" in msg\n\n\n# -- list_remote_models (local endpoint discovery; HTTP fully mocked) --------\n\n\nclass _FakeResp:\n    def __init__(self, payload, *, status_ok=True):\n        self._payload = payload\n        self._ok = status_ok\n\n    def raise_for_status(self):\n        if not self._ok:\n            import httpx\n\n            raise httpx.HTTPStatusError(\"boom\", request=None, response=None)\n\n    def json(self):\n        return self._payload\n\n\ndef _capture_get(payload, *, status_ok=True):\n    \"\"\"Return (patched_get, urls) — records every URL the discovery probe hits.\"\"\"\n    urls: list[str] = []\n\n    def _get(url, *args, **kwargs):\n        urls.append(url)\n        return _FakeResp(payload, status_ok=status_ok)\n\n    return _get, urls\n\n\ndef test_list_remote_models_ollama_parses_tags():\n    from jarn.providers import list_remote_models\n\n    payload = {\"models\": [{\"name\": \"qwen3-coder:30b\"}, {\"name\": \"llama3:8b\"}]}\n    get, urls = _capture_get(payload)\n    prov = ProviderConfig(type=ProviderType.OLLAMA, base_url=\"http://localhost:11434\")\n    with patch(\"httpx.get\", get):\n        out = list_remote_models(prov)\n    assert out == [\"qwen3-coder:30b\", \"llama3:8b\"]\n    assert urls == [\"http://localhost:11434/api/tags\"]\n\n\ndef test_list_remote_models_lmstudio_parses_v1_models():\n    from jarn.providers import list_remote_models\n\n    payload = {\"data\": [{\"id\": \"qwen2.5-coder\"}, {\"id\": \"phi-4\"}]}\n    get, urls = _capture_get(payload)\n    # base_url already carries /v1 (normalize_base_url convention).\n    prov = ProviderConfig(type=ProviderType.LMSTUDIO, base_url=\"http://localhost:1234/v1\")\n    with patch(\"httpx.get\", get):\n        out = list_remote_models(prov)\n    assert out == [\"qwen2.5-coder\", \"phi-4\"]\n    assert urls == [\"http://localhost:1234/v1/models\"]\n\n\ndef test_list_remote_models_openai_compatible_appends_v1_when_bare():\n    from jarn.providers import list_remote_models\n\n    payload = {\"data\": [{\"id\": \"local-model\"}]}\n    get, urls = _capture_get(payload)\n    prov = ProviderConfig(type=ProviderType.OPENAI_COMPATIBLE, base_url=\"http://localhost:8000\")\n    with patch(\"httpx.get\", get):\n        out = list_remote_models(prov)\n    assert out == [\"local-model\"]\n    assert urls == [\"http://localhost:8000/v1/models\"]\n\n\ndef test_list_remote_models_unreachable_returns_empty():\n    \"\"\"A connection error must degrade to [] (manual entry), never raise.\"\"\"\n    import httpx\n\n    from jarn.providers import list_remote_models\n\n    def _boom(*a, **k):\n        raise httpx.ConnectError(\"no endpoint\")\n\n    prov = ProviderConfig(type=ProviderType.OLLAMA, base_url=\"http://localhost:11434\")\n    with patch(\"httpx.get\", _boom):\n        assert list_remote_models(prov) == []\n\n\ndef test_slug_hint_is_provider_aware():\n    \"\"\"The did-you-mean hint names the right convention per provider and is not\n    misleadingly anthropic/openrouter-specific for other providers.\"\"\"\n    from jarn.providers.models import _slug_hint\n\n    assert \"dashes\" in _slug_hint(ProviderType.ANTHROPIC)\n    assert \"dots\" in _slug_hint(ProviderType.OPENROUTER)\n    generic = _slug_hint(ProviderType.GOOGLE)\n    assert \"OpenRouter uses dots\" not in generic\n    assert \"dot-vs-dash\" in generic\n\n\ndef test_list_remote_models_http_error_returns_empty():\n    from jarn.providers import list_remote_models\n\n    get, _ = _capture_get({\"models\": []}, status_ok=False)\n    prov = ProviderConfig(type=ProviderType.OLLAMA, base_url=\"http://localhost:11434\")\n    with patch(\"httpx.get\", get):\n        assert list_remote_models(prov) == []\n\n\ndef test_list_remote_models_no_base_url_returns_empty():\n    from jarn.providers import list_remote_models\n\n    # No network call should be made when there's no endpoint to hit.\n    def _fail(*a, **k):\n        raise AssertionError(\"must not hit the network without a base_url\")\n\n    prov = ProviderConfig(type=ProviderType.OLLAMA, base_url=None)\n    with patch(\"httpx.get\", _fail):\n        assert list_remote_models(prov) == []\n\n\ndef test_list_remote_models_non_local_provider_returns_empty():\n    from jarn.providers import list_remote_models\n\n    def _fail(*a, **k):\n        raise AssertionError(\"must not probe cloud providers\")\n\n    prov = ProviderConfig(type=ProviderType.ANTHROPIC, base_url=\"https://api.anthropic.com\")\n    with patch(\"httpx.get\", _fail):\n        assert list_remote_models(prov) == []\n\n\ndef test_disallowed_provider_extra_rejected_at_load(tmp_path):\n    import yaml\n\n    from jarn.config import ConfigError\n    from jarn.config.loader import load_config\n\n    gp = tmp_path / \"g.yaml\"\n    gp.write_text(\n        yaml.safe_dump(\n            {\n                \"providers\": {\n                    \"openai\": {\n                        \"type\": \"openai\",\n                        \"api_key\": \"sk-test\",\n                        \"default_headers\": {\"X-Evil\": \"yes\"},\n                    }\n                }\n            }\n        ),\n        encoding=\"utf-8\",\n    )\n    with pytest.raises(ConfigError, match=\"unknown extra\"):\n        load_config(global_path=gp, project_path=None)\n\n\ndef test_allowed_provider_extra_reaches_constructor():\n    factory = _factory(ProviderType.OPENAI, api_key=\"k\", extra={\"timeout\": 42})\n    with patch(\"langchain.chat_models.init_chat_model\") as m:\n        m.return_value = object()\n        factory.build(\"p/model\")\n    assert m.call_args.kwargs[\"timeout\"] == 42\n\n\ndef test_provider_headers_reach_constructor():\n    factory = _factory(\n        ProviderType.OPENAI,\n        api_key=\"k\",\n        headers={\"Authorization\": \"Bearer secret\"},\n    )\n    with patch(\"langchain.chat_models.init_chat_model\") as m:\n        m.return_value = object()\n        factory.build(\"p/model\")\n    assert m.call_args.kwargs[\"default_headers\"] == {\"Authorization\": \"Bearer secret\"}\n\n\ndef test_list_remote_models_malformed_payload_returns_empty():\n    from jarn.providers import list_remote_models\n\n    # Missing keys / wrong shapes must not raise.\n    get, _ = _capture_get({\"unexpected\": \"shape\"})\n    prov = ProviderConfig(type=ProviderType.LMSTUDIO, base_url=\"http://localhost:1234/v1\")\n    with patch(\"httpx.get\", get):\n        assert list_remote_models(prov) == []\n\n\ndef test_list_remote_models_passes_auth_headers():\n    from jarn.providers import list_remote_models\n\n    payload = {\"data\": [{\"id\": \"local-model\"}]}\n    captured: list[dict] = []\n\n    def _get(url, *args, **kwargs):\n        captured.append(kwargs.get(\"headers\") or {})\n        return _FakeResp(payload)\n\n    prov = ProviderConfig(\n        type=ProviderType.OPENAI_COMPATIBLE,\n        base_url=\"http://localhost:8000/v1\",\n        api_key=\"secret-key\",\n        headers={\"X-Custom\": \"yes\"},\n    )\n    with patch(\"httpx.get\", _get):\n        assert list_remote_models(prov) == [\"local-model\"]\n    assert captured[0][\"Authorization\"] == \"Bearer secret-key\"\n    assert captured[0][\"X-Custom\"] == \"yes\"\n\n\ndef test_cache_invalidation():\n    \"\"\"After invalidate_cache, build() re-resolves the provider with fresh config.\"\"\"\n    from unittest.mock import MagicMock\n\n    from jarn.config.schema import RoutingConfig\n    from jarn.providers.models import ModelFactory\n\n    cfg = Config(\n        default_profile=\"p\",\n        providers={\"p\": ProviderConfig(type=ProviderType.OPENAI, api_key=\"old-key\")},\n        routing=RoutingConfig(main=\"p/model\"),\n    )\n    factory = ModelFactory(cfg)\n    first = MagicMock(name=\"first\")\n    second = MagicMock(name=\"second\")\n\n    with patch(\"langchain.chat_models.init_chat_model\", side_effect=[first, second]) as m:\n        assert factory.build(\"p/model\") is first\n        cfg.providers[\"p\"].api_key = \"new-key\"\n        factory.invalidate_cache()\n        assert factory.build(\"p/model\") is second\n\n    assert m.call_count == 2\n    assert m.call_args_list[0].kwargs[\"api_key\"] == \"old-key\"\n    assert m.call_args_list[1].kwargs[\"api_key\"] == \"new-key\"\n\n\n# -- remote_context_window (local-model context size for the gauge) ---------\n\n\ndef test_remote_context_window_lmstudio_prefers_loaded():\n    \"\"\"LM Studio: query the native /api/v0 API and prefer loaded over max.\"\"\"\n    from jarn.providers import remote_context_window\n\n    prov = ProviderConfig(type=ProviderType.LMSTUDIO, base_url=\"http://localhost:1234/v1\")\n    payload = {\"data\": [\n        {\"id\": \"other\", \"loaded_context_length\": 999},\n        {\"id\": \"qwen\", \"loaded_context_length\": 8192, \"max_context_length\": 32768},\n    ]}\n    get, urls = _capture_get(payload)\n    with patch(\"httpx.get\", get):\n        assert remote_context_window(prov, \"qwen\") == 8192\n    assert urls == [\"http://localhost:1234/api/v0/models\"]  # native API, not /v1\n\n\ndef test_remote_context_window_lmstudio_falls_back_to_max():\n    from jarn.providers import remote_context_window\n\n    prov = ProviderConfig(type=ProviderType.LMSTUDIO, base_url=\"http://localhost:1234/v1\")\n    get, _ = _capture_get({\"data\": [{\"id\": \"qwen\", \"max_context_length\": 16384}]})\n    with patch(\"httpx.get\", get):\n        assert remote_context_window(prov, \"qwen\") == 16384\n\n\ndef test_remote_context_window_ollama_reads_model_info():\n    from jarn.providers import remote_context_window\n\n    prov = ProviderConfig(type=ProviderType.OLLAMA, base_url=\"http://localhost:11434\")\n    payload = {\"model_info\": {\"general.architecture\": \"qwen2\", \"qwen2.context_length\": 4096}}\n    posts: list[str] = []\n\n    def _post(url, *a, **k):\n        posts.append(url)\n        return _FakeResp(payload)\n\n    with patch(\"httpx.post\", _post):\n        assert remote_context_window(prov, \"qwen2.5\") == 4096\n    assert posts == [\"http://localhost:11434/api/show\"]\n\n\ndef test_remote_context_window_model_not_found_returns_none():\n    from jarn.providers import remote_context_window\n\n    prov = ProviderConfig(type=ProviderType.LMSTUDIO, base_url=\"http://localhost:1234/v1\")\n    get, _ = _capture_get({\"data\": [{\"id\": \"other\", \"loaded_context_length\": 8192}]})\n    with patch(\"httpx.get\", get):\n        assert remote_context_window(prov, \"qwen\") is None\n\n\ndef test_remote_context_window_unreachable_returns_none():\n    \"\"\"A connection error must degrade to None (hide the gauge), never raise.\"\"\"\n    import httpx\n\n    from jarn.providers import remote_context_window\n\n    prov = ProviderConfig(type=ProviderType.LMSTUDIO, base_url=\"http://localhost:1234/v1\")\n\n    def _boom(*a, **k):\n        raise httpx.ConnectError(\"no endpoint\")\n\n    with patch(\"httpx.get\", _boom):\n        assert remote_context_window(prov, \"qwen\") is None\n\n\n# ---------------------------------------------------------------------------\n# T-4-8 (fix round 1) — the JARN_DEMO canned model is WIRED into the build path\n# ---------------------------------------------------------------------------\n\n\ndef test_demo_mode_wires_canned_model_into_build(monkeypatch):\n    \"\"\"JARN_DEMO=1 makes ``build_main()`` return the canned demo model with NO\n    real provider config and NO API key, and it yields a canned response with no\n    network.  Env unset: ``build_main()`` does NOT return the fake — it raises\n    (no model configured), proving the fake is never silently substituted.\n\n    This is the wiring proof: the reviewer found the demo functions had zero\n    callers in the build path (dead code).  This test consumes the wiring.\n    \"\"\"\n    from langchain_core.messages import HumanMessage\n\n    from jarn.providers.models import ModelFactory, ModelResolutionError\n\n    # A config with NO providers and NO configured model: a real build must fail.\n    cfg = Config(providers={}, routing=RoutingConfig())\n\n    # --- env unset: no fake substitution; build_main raises (nothing configured) ---\n    monkeypatch.delenv(\"JARN_DEMO\", raising=False)\n    with pytest.raises(ModelResolutionError):\n        ModelFactory(cfg).build_main()\n\n    # --- JARN_DEMO=1: canned model returned despite empty config / no API key ---\n    monkeypatch.setenv(\"JARN_DEMO\", \"1\")\n    model = ModelFactory(cfg).build_main()\n    assert model._llm_type == \"jarn-demo-canned\", (\n        \"build_main() must return the canned demo model when JARN_DEMO=1\"\n    )\n    # It yields a canned response with NO network and NO API key.\n    reply = model.invoke([HumanMessage(content=\"add input validation to server.py\")])\n    assert reply.content, \"demo model must return non-empty canned content\"\n\n\ndef test_demo_model_bind_tools_is_noop_and_survives(monkeypatch):\n    \"\"\"The demo model must tolerate ``bind_tools`` (deepagents/langgraph calls it)\n    — a raw GenericFakeChatModel raises NotImplementedError there.\"\"\"\n    monkeypatch.setenv(\"JARN_DEMO\", \"1\")\n    cfg = Config(providers={}, routing=RoutingConfig())\n    model = ModelFactory(cfg).build_main()\n    bound = model.bind_tools([])  # must not raise\n    assert bound is not None\n\n\ndef test_demo_money_shot_has_edit_tool_call():\n    \"\"\"The scripted demo must drive the money-shot DIFF via a write_file/edit_file\n    tool call (not just prose), so the recorded GIF shows a real diff.\"\"\"\n    from jarn.providers.models import _demo_messages\n\n    msgs = _demo_messages()\n    tool_calls = [\n        tc for m in msgs for tc in (getattr(m, \"tool_calls\", None) or [])\n    ]\n    assert any(\n        tc[\"name\"] in (\"write_file\", \"edit_file\") for tc in tool_calls\n    ), \"the demo script must include a write_file/edit_file tool call for the diff step\"\n"}}}}