Skip to content

~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.

10.8 KB
Raw Blame
"""
Tests for missing coverage gaps in local_deep_research/config/llm_config.py.

Targets specific uncovered code paths:
- get_llm env var logic (LDR_USE_FALLBACK_LLM, LDR_TESTING_WITH_MOCKS,
  provider_has_config skipping, lmstudio/llamacpp local checks)
- get_llm ollama model-not-found, creation exception (now raises), enable_thinking true/false
- get_llm llamacpp no model path, invalid extension, directory with gguf suggestion
- _get_context_window_for_provider cloud unrestricted=False, local None window
- wrap_llm_without_think_tags context_limit injection, no-overwrite, model_name vs model
"""

import os
from unittest.mock import MagicMock, patch

import pytest

MODULE = "local_deep_research.config.llm_config"
# Context-window resolution delegates to the _helpers twin, which reads
# settings via thread_settings.get_setting_from_snapshot. Patch this module
# (not MODULE) for context-window reads; wrap_llm tests that also read
# rate_limiting settings must patch BOTH.
THREAD_SETTINGS = "local_deep_research.config.thread_settings"


def _settings_dict(overrides=None):
    """Base settings dict with sensible defaults for most tests."""
    base = {
        "search.tool": "searxng",
        "llm.model": "test-model",
        "llm.temperature": 0.5,
        "llm.provider": "ollama",
        "llm.local_context_window_size": 4096,
        "llm.context_window_unrestricted": True,
        "llm.supports_max_tokens": True,
        "llm.max_tokens": 4096,
        "rate_limiting.llm_enabled": False,
    }
    if overrides:
        base.update(overrides)
    return base


def _mock_get_setting(settings):
    """Return a side_effect callable that reads from the given dict."""
    return lambda key, default=None, **kw: settings.get(key, default)


class TestGetLlmFallbackEnvVar:
    """Tests for LDR_USE_FALLBACK_LLM / LDR_TESTING_WITH_MOCKS env var guard."""

    def test_fallback_env_var_skipped_when_testing_with_mocks(self):
        """When LDR_TESTING_WITH_MOCKS is set, fallback env var is ignored;
        now raises ValueError since fallback model was removed."""
        settings = _settings_dict({"llm.provider": "none"})
        with (
            patch.dict(
                os.environ,
                {"LDR_USE_FALLBACK_LLM": "1", "LDR_TESTING_WITH_MOCKS": "1"},
            ),
            patch(f"{MODULE}.is_llm_registered", return_value=False),
            patch(
                f"{MODULE}.get_setting_from_snapshot",
                side_effect=_mock_get_setting(settings),
            ),
        ):
            from local_deep_research.config.llm_config import get_llm

            with pytest.raises(ValueError):
                get_llm(provider="none", settings_snapshot=settings)

    def test_fallback_env_var_no_config_raises_error(self):
        """When provider has NO config, ValueError is raised (fallback model removed)."""
        settings = _settings_dict(
            {"llm.provider": "openai", "llm.openai.api_key": None}
        )
        with (
            patch.dict(os.environ, {"LDR_USE_FALLBACK_LLM": "1"}, clear=False),
            patch.dict(os.environ, {"LDR_TESTING_WITH_MOCKS": ""}, clear=False),
            patch(f"{MODULE}.is_llm_registered", return_value=False),
            patch(
                f"{MODULE}.get_setting_from_snapshot",
                side_effect=_mock_get_setting(settings),
            ),
        ):
            from local_deep_research.config.llm_config import get_llm

            with pytest.raises(ValueError):
                get_llm(provider="openai", settings_snapshot=settings)


class TestGetLlmOllamaEdgeCases:
    """Tests for ollama-specific edge cases in get_llm."""

    def _ollama_settings(self, overrides=None):
        s = _settings_dict(
            {
                "llm.provider": "ollama",
                "llm.ollama.url": "http://localhost:11434",
                "llm.ollama.enable_thinking": True,
            }
        )
        if overrides:
            s.update(overrides)
        return s


class TestGetContextWindowForProviderMissing:
    """Additional coverage for _get_context_window_for_provider edge cases."""

    def test_cloud_unrestricted_false_returns_configured_size(self):
        """Cloud provider with unrestricted=False returns configured size."""

        def fake_setting(key, default, settings_snapshot=None):
            if key == "llm.context_window_unrestricted":
                return False
            if key == "llm.context_window_size":
                return 64000
            return default

        with patch(
            f"{THREAD_SETTINGS}.get_setting_from_snapshot",
            side_effect=fake_setting,
        ):
            from local_deep_research.config.llm_config import (
                _get_context_window_for_provider,
            )

            result = _get_context_window_for_provider("anthropic")
            assert result == 64000
            assert isinstance(result, int)

    def test_local_provider_none_window_defaults_to_8192(self):
        """When local context window setting returns None, fallback to 8192."""
        with patch(
            f"{THREAD_SETTINGS}.get_setting_from_snapshot", return_value=None
        ):
            from local_deep_research.config.llm_config import (
                _get_context_window_for_provider,
            )

            result = _get_context_window_for_provider("ollama")
            assert result == 8192

    def test_cloud_restricted_none_window_defaults_to_128000(self):
        """When cloud restricted but context_window_size is None, fallback to 128000."""

        def fake_setting(key, default, settings_snapshot=None):
            if key == "llm.context_window_unrestricted":
                return False
            if key == "llm.context_window_size":
                return None
            return default

        with patch(
            f"{THREAD_SETTINGS}.get_setting_from_snapshot",
            side_effect=fake_setting,
        ):
            from local_deep_research.config.llm_config import (
                _get_context_window_for_provider,
            )

            result = _get_context_window_for_provider("openai")
            assert result == 128000


class TestWrapLlmContextLimitInjection:
    """Tests for context_limit injection in wrap_llm_without_think_tags."""

    def test_injects_context_limit_for_local_provider(self):
        """wrap_llm sets context_limit for local provider (llamacpp)."""
        llm = MagicMock()
        research_ctx = {}

        def fake_setting(key, default=None, settings_snapshot=None):
            if key == "rate_limiting.llm_enabled":
                return False
            if key == "llm.local_context_window_size":
                return 8192
            return default

        with (
            patch(
                f"{MODULE}.get_setting_from_snapshot", side_effect=fake_setting
            ),
            patch(
                f"{THREAD_SETTINGS}.get_setting_from_snapshot",
                side_effect=fake_setting,
            ),
        ):
            from local_deep_research.config.llm_config import (
                wrap_llm_without_think_tags,
            )

            wrap_llm_without_think_tags(
                llm, provider="llamacpp", research_context=research_ctx
            )
        assert research_ctx["context_limit"] == 8192

    def test_does_not_overwrite_existing_context_limit(self):
        """If research_context already has context_limit, it is NOT overwritten."""
        llm = MagicMock()
        research_ctx = {"context_limit": 32000}

        def fake_setting(key, default=None, settings_snapshot=None):
            if key == "rate_limiting.llm_enabled":
                return False
            if key == "llm.local_context_window_size":
                return 4096
            return default

        with patch(
            f"{MODULE}.get_setting_from_snapshot", side_effect=fake_setting
        ):
            from local_deep_research.config.llm_config import (
                wrap_llm_without_think_tags,
            )

            wrap_llm_without_think_tags(
                llm, provider="ollama", research_context=research_ctx
            )
        assert research_ctx["context_limit"] == 32000

    def test_model_name_attribute_used_for_token_callback(self):
        """When llm has model_name, it is used for preset_model on token callback."""
        llm = MagicMock()
        llm.model_name = "gpt-4o"
        llm.callbacks = None
        mock_counter = MagicMock()
        mock_callback = MagicMock()
        mock_counter.create_callback.return_value = mock_callback
        with (
            patch(f"{MODULE}.get_setting_from_snapshot", return_value=False),
            patch(
                "local_deep_research.metrics.TokenCounter",
                return_value=mock_counter,
            ),
        ):
            from local_deep_research.config.llm_config import (
                wrap_llm_without_think_tags,
            )

            wrap_llm_without_think_tags(llm, research_id=1, provider="openai")
        assert mock_callback.preset_model == "gpt-4o"

    def test_model_attribute_fallback_for_token_callback(self):
        """When llm has .model but NOT .model_name, .model is used."""
        llm = MagicMock(spec=["invoke", "callbacks", "model"])
        llm.model = "claude-3-haiku"
        llm.callbacks = None
        mock_counter = MagicMock()
        mock_callback = MagicMock()
        mock_counter.create_callback.return_value = mock_callback
        with (
            patch(f"{MODULE}.get_setting_from_snapshot", return_value=False),
            patch(
                "local_deep_research.metrics.TokenCounter",
                return_value=mock_counter,
            ),
        ):
            from local_deep_research.config.llm_config import (
                wrap_llm_without_think_tags,
            )

            wrap_llm_without_think_tags(
                llm, research_id=2, provider="anthropic"
            )
        assert mock_callback.preset_model == "claude-3-haiku"

    def test_cloud_unrestricted_does_not_inject_context_limit(self):
        """For cloud provider with unrestricted context, context_limit is not added."""
        llm = MagicMock()
        research_ctx = {}

        def fake_setting(key, default=None, settings_snapshot=None):
            if key == "rate_limiting.llm_enabled":
                return False
            if key == "llm.context_window_unrestricted":
                return True
            return default

        with (
            patch(
                f"{MODULE}.get_setting_from_snapshot", side_effect=fake_setting
            ),
            patch(
                f"{THREAD_SETTINGS}.get_setting_from_snapshot",
                side_effect=fake_setting,
            ),
        ):
            from local_deep_research.config.llm_config import (
                wrap_llm_without_think_tags,
            )

            wrap_llm_without_think_tags(
                llm, provider="openai", research_context=research_ctx
            )
        assert "context_limit" not in research_ctx