|
| 1 | +from __future__ import annotations |
| 2 | + |
| 3 | +import warnings |
| 4 | + |
| 5 | +import pytest |
| 6 | + |
| 7 | +from instructor.mode import Mode |
| 8 | +from instructor.utils.providers import Provider |
| 9 | + |
| 10 | +try: |
| 11 | + from instructor.v2.core import ( |
| 12 | + DEPRECATED_MODE_MAPPING, |
| 13 | + mode_registry, |
| 14 | + normalize_mode, |
| 15 | + reset_deprecation_warnings, |
| 16 | + ) |
| 17 | +except ModuleNotFoundError: |
| 18 | + pytest.skip("v2 module not available", allow_module_level=True) |
| 19 | + |
| 20 | + |
| 21 | +@pytest.fixture(autouse=True) |
| 22 | +def reset_warnings(): |
| 23 | + """Reset deprecation warnings before each test.""" |
| 24 | + reset_deprecation_warnings() |
| 25 | + yield |
| 26 | + reset_deprecation_warnings() |
| 27 | + |
| 28 | + |
| 29 | +@pytest.mark.parametrize( |
| 30 | + "provider,mode,expected", |
| 31 | + [ |
| 32 | + # GenAI legacy modes |
| 33 | + (Provider.GENAI, Mode.GENAI_TOOLS, Mode.TOOLS), |
| 34 | + (Provider.GENAI, Mode.GENAI_JSON, Mode.JSON), |
| 35 | + (Provider.GENAI, Mode.GENAI_STRUCTURED_OUTPUTS, Mode.JSON), |
| 36 | + # Anthropic legacy modes |
| 37 | + (Provider.ANTHROPIC, Mode.ANTHROPIC_TOOLS, Mode.TOOLS), |
| 38 | + (Provider.ANTHROPIC, Mode.ANTHROPIC_JSON, Mode.MD_JSON), |
| 39 | + (Provider.ANTHROPIC, Mode.ANTHROPIC_PARALLEL_TOOLS, Mode.PARALLEL_TOOLS), |
| 40 | + # OpenAI legacy modes |
| 41 | + (Provider.OPENAI, Mode.FUNCTIONS, Mode.TOOLS), |
| 42 | + (Provider.OPENAI, Mode.TOOLS_STRICT, Mode.TOOLS), |
| 43 | + (Provider.OPENAI, Mode.JSON_O1, Mode.JSON_SCHEMA), |
| 44 | + # Note: Mode.JSON is NOT deprecated - it's used by GenAI as a valid mode |
| 45 | + ( |
| 46 | + Provider.OPENAI, |
| 47 | + Mode.RESPONSES_TOOLS_WITH_INBUILT_TOOLS, |
| 48 | + Mode.RESPONSES_TOOLS, |
| 49 | + ), |
| 50 | + # Mistral legacy modes |
| 51 | + (Provider.MISTRAL, Mode.MISTRAL_TOOLS, Mode.TOOLS), |
| 52 | + (Provider.MISTRAL, Mode.MISTRAL_STRUCTURED_OUTPUTS, Mode.JSON_SCHEMA), |
| 53 | + # Cohere legacy modes |
| 54 | + (Provider.COHERE, Mode.COHERE_TOOLS, Mode.TOOLS), |
| 55 | + (Provider.COHERE, Mode.COHERE_JSON_SCHEMA, Mode.JSON_SCHEMA), |
| 56 | + # xAI legacy modes |
| 57 | + (Provider.XAI, Mode.XAI_TOOLS, Mode.TOOLS), |
| 58 | + (Provider.XAI, Mode.XAI_JSON, Mode.MD_JSON), |
| 59 | + # Generic modes should pass through unchanged |
| 60 | + (Provider.GENAI, Mode.TOOLS, Mode.TOOLS), |
| 61 | + (Provider.GENAI, Mode.JSON, Mode.JSON), |
| 62 | + (Provider.ANTHROPIC, Mode.TOOLS, Mode.TOOLS), |
| 63 | + (Provider.ANTHROPIC, Mode.JSON, Mode.JSON), |
| 64 | + (Provider.OPENAI, Mode.TOOLS, Mode.TOOLS), |
| 65 | + (Provider.OPENAI, Mode.JSON_SCHEMA, Mode.JSON_SCHEMA), |
| 66 | + (Provider.OPENAI, Mode.MD_JSON, Mode.MD_JSON), |
| 67 | + (Provider.OPENAI, Mode.PARALLEL_TOOLS, Mode.PARALLEL_TOOLS), |
| 68 | + (Provider.OPENAI, Mode.RESPONSES_TOOLS, Mode.RESPONSES_TOOLS), |
| 69 | + ], |
| 70 | +) |
| 71 | +def test_normalize_mode(provider: Provider, mode: Mode, expected: Mode): |
| 72 | + """Test that mode normalization works correctly.""" |
| 73 | + result = normalize_mode(provider, mode) |
| 74 | + assert result == expected |
| 75 | + |
| 76 | + |
| 77 | +@pytest.mark.parametrize( |
| 78 | + "provider,expected_modes", |
| 79 | + [ |
| 80 | + (Provider.GENAI, [Mode.TOOLS, Mode.JSON]), |
| 81 | + ( |
| 82 | + Provider.OPENAI, |
| 83 | + [ |
| 84 | + Mode.TOOLS, |
| 85 | + Mode.JSON_SCHEMA, |
| 86 | + Mode.MD_JSON, |
| 87 | + Mode.PARALLEL_TOOLS, |
| 88 | + Mode.RESPONSES_TOOLS, |
| 89 | + ], |
| 90 | + ), |
| 91 | + ], |
| 92 | +) |
| 93 | +def test_handlers_registered_with_generic_modes( |
| 94 | + provider: Provider, expected_modes: list[Mode] |
| 95 | +): |
| 96 | + """Test that handlers are registered with generic modes. |
| 97 | +
|
| 98 | + Note: This test requires provider handlers to be registered. |
| 99 | + It will be fully enabled in PR 3+ when providers are added. |
| 100 | + """ |
| 101 | + # Skip if no modes are registered for this provider yet |
| 102 | + registered_modes = mode_registry.get_modes_for_provider(provider) |
| 103 | + if not registered_modes: |
| 104 | + pytest.skip(f"No handlers registered for {provider.value} yet") |
| 105 | + |
| 106 | + for mode in expected_modes: |
| 107 | + if mode in registered_modes: |
| 108 | + assert mode_registry.is_registered(provider, mode) |
| 109 | + |
| 110 | + |
| 111 | +@pytest.mark.parametrize( |
| 112 | + "provider,legacy_mode,expected_mode", |
| 113 | + [ |
| 114 | + (Provider.GENAI, Mode.GENAI_TOOLS, Mode.TOOLS), |
| 115 | + (Provider.GENAI, Mode.GENAI_JSON, Mode.JSON), |
| 116 | + (Provider.GENAI, Mode.GENAI_STRUCTURED_OUTPUTS, Mode.JSON), |
| 117 | + (Provider.OPENAI, Mode.FUNCTIONS, Mode.TOOLS), |
| 118 | + (Provider.OPENAI, Mode.TOOLS_STRICT, Mode.TOOLS), |
| 119 | + (Provider.OPENAI, Mode.JSON_O1, Mode.JSON_SCHEMA), |
| 120 | + ( |
| 121 | + Provider.OPENAI, |
| 122 | + Mode.RESPONSES_TOOLS_WITH_INBUILT_TOOLS, |
| 123 | + Mode.RESPONSES_TOOLS, |
| 124 | + ), |
| 125 | + ], |
| 126 | +) |
| 127 | +def test_backwards_compatibility( |
| 128 | + provider: Provider, legacy_mode: Mode, expected_mode: Mode |
| 129 | +): |
| 130 | + """Test that old provider-specific modes still work. |
| 131 | +
|
| 132 | + Note: This test requires provider handlers to be registered. |
| 133 | + It will be fully enabled in PR 3+ when providers are added. |
| 134 | + """ |
| 135 | + # Skip if provider doesn't have any handlers registered yet |
| 136 | + registered_modes = mode_registry.get_modes_for_provider(provider) |
| 137 | + if not registered_modes: |
| 138 | + pytest.skip(f"No handlers registered for {provider.value} yet") |
| 139 | + |
| 140 | + # The normalized mode should be registered |
| 141 | + assert mode_registry.is_registered(provider, expected_mode) |
| 142 | + |
| 143 | + # Legacy mode should normalize and find the same handler |
| 144 | + legacy_handler = mode_registry.get_handler_class(provider, legacy_mode) |
| 145 | + expected_handler = mode_registry.get_handler_class(provider, expected_mode) |
| 146 | + |
| 147 | + assert legacy_handler is not None |
| 148 | + assert expected_handler is not None |
| 149 | + assert legacy_handler == expected_handler |
| 150 | + |
| 151 | + |
| 152 | +@pytest.mark.parametrize( |
| 153 | + "provider,mode,expected_replacement", |
| 154 | + [ |
| 155 | + (Provider.OPENAI, Mode.FUNCTIONS, Mode.TOOLS), |
| 156 | + (Provider.OPENAI, Mode.TOOLS_STRICT, Mode.TOOLS), |
| 157 | + (Provider.OPENAI, Mode.JSON_O1, Mode.JSON_SCHEMA), |
| 158 | + # Note: Mode.JSON is NOT deprecated - it's used by GenAI as a valid mode |
| 159 | + (Provider.ANTHROPIC, Mode.ANTHROPIC_TOOLS, Mode.TOOLS), |
| 160 | + (Provider.GENAI, Mode.GENAI_TOOLS, Mode.TOOLS), |
| 161 | + (Provider.MISTRAL, Mode.MISTRAL_TOOLS, Mode.TOOLS), |
| 162 | + (Provider.COHERE, Mode.COHERE_TOOLS, Mode.TOOLS), |
| 163 | + (Provider.XAI, Mode.XAI_TOOLS, Mode.TOOLS), |
| 164 | + ], |
| 165 | +) |
| 166 | +def test_deprecated_mode_emits_warning( |
| 167 | + provider: Provider, mode: Mode, expected_replacement: Mode |
| 168 | +): |
| 169 | + """Test that using deprecated modes emits a deprecation warning.""" |
| 170 | + with warnings.catch_warnings(record=True) as w: |
| 171 | + warnings.simplefilter("always") |
| 172 | + result = normalize_mode(provider, mode) |
| 173 | + |
| 174 | + # Should have emitted a warning |
| 175 | + assert len(w) == 1 |
| 176 | + assert issubclass(w[0].category, DeprecationWarning) |
| 177 | + assert mode.name in str(w[0].message) |
| 178 | + assert expected_replacement.name in str(w[0].message) |
| 179 | + assert "v3.0" in str(w[0].message) |
| 180 | + |
| 181 | + # Should still return the correct normalized mode |
| 182 | + assert result == expected_replacement |
| 183 | + |
| 184 | + |
| 185 | +@pytest.mark.parametrize("provider", [Provider.OPENAI, Provider.ANTHROPIC]) |
| 186 | +def test_deprecated_mode_warning_only_once(provider: Provider): |
| 187 | + """Test that deprecation warning is only shown once per mode.""" |
| 188 | + with warnings.catch_warnings(record=True) as w: |
| 189 | + warnings.simplefilter("always") |
| 190 | + |
| 191 | + # First call should warn |
| 192 | + normalize_mode(provider, Mode.FUNCTIONS) |
| 193 | + assert len(w) == 1 |
| 194 | + |
| 195 | + # Second call should not warn again |
| 196 | + normalize_mode(provider, Mode.FUNCTIONS) |
| 197 | + assert len(w) == 1 # Still only 1 warning |
| 198 | + |
| 199 | + # Different mode should warn |
| 200 | + normalize_mode(provider, Mode.TOOLS_STRICT) |
| 201 | + assert len(w) == 2 |
| 202 | + |
| 203 | + |
| 204 | +@pytest.mark.parametrize( |
| 205 | + "provider,modes", |
| 206 | + [ |
| 207 | + (Provider.OPENAI, [Mode.TOOLS, Mode.JSON_SCHEMA, Mode.MD_JSON]), |
| 208 | + (Provider.GENAI, [Mode.TOOLS, Mode.JSON]), |
| 209 | + (Provider.ANTHROPIC, [Mode.TOOLS, Mode.JSON]), |
| 210 | + ], |
| 211 | +) |
| 212 | +def test_generic_mode_no_warning(provider: Provider, modes: list[Mode]): |
| 213 | + """Test that using generic modes does not emit warnings.""" |
| 214 | + with warnings.catch_warnings(record=True) as w: |
| 215 | + warnings.simplefilter("always") |
| 216 | + |
| 217 | + for mode in modes: |
| 218 | + normalize_mode(provider, mode) |
| 219 | + |
| 220 | + # No warnings should be emitted for generic modes |
| 221 | + assert len(w) == 0 |
| 222 | + |
| 223 | + |
| 224 | +def test_deprecated_mode_mapping_is_complete(): |
| 225 | + """Test that all provider-specific modes are in the deprecated mapping.""" |
| 226 | + # These are the provider-specific modes that should be deprecated |
| 227 | + # Note: Mode.JSON is NOT deprecated - it's used by GenAI as a valid mode |
| 228 | + expected_deprecated = { |
| 229 | + # OpenAI legacy |
| 230 | + Mode.FUNCTIONS, |
| 231 | + Mode.TOOLS_STRICT, |
| 232 | + # Mode.JSON is NOT deprecated - it's used by GenAI |
| 233 | + Mode.JSON_O1, |
| 234 | + Mode.RESPONSES_TOOLS_WITH_INBUILT_TOOLS, |
| 235 | + # Anthropic |
| 236 | + Mode.ANTHROPIC_TOOLS, |
| 237 | + Mode.ANTHROPIC_JSON, |
| 238 | + Mode.ANTHROPIC_PARALLEL_TOOLS, |
| 239 | + # GenAI |
| 240 | + Mode.GENAI_TOOLS, |
| 241 | + Mode.GENAI_JSON, |
| 242 | + Mode.GENAI_STRUCTURED_OUTPUTS, |
| 243 | + # Mistral |
| 244 | + Mode.MISTRAL_TOOLS, |
| 245 | + Mode.MISTRAL_STRUCTURED_OUTPUTS, |
| 246 | + # Cohere |
| 247 | + Mode.COHERE_TOOLS, |
| 248 | + Mode.COHERE_JSON_SCHEMA, |
| 249 | + # xAI |
| 250 | + Mode.XAI_TOOLS, |
| 251 | + Mode.XAI_JSON, |
| 252 | + # Fireworks |
| 253 | + Mode.FIREWORKS_TOOLS, |
| 254 | + Mode.FIREWORKS_JSON, |
| 255 | + # Cerebras |
| 256 | + Mode.CEREBRAS_TOOLS, |
| 257 | + Mode.CEREBRAS_JSON, |
| 258 | + # Writer |
| 259 | + Mode.WRITER_TOOLS, |
| 260 | + Mode.WRITER_JSON, |
| 261 | + # Bedrock |
| 262 | + Mode.BEDROCK_TOOLS, |
| 263 | + Mode.BEDROCK_JSON, |
| 264 | + # Perplexity |
| 265 | + Mode.PERPLEXITY_JSON, |
| 266 | + # VertexAI |
| 267 | + Mode.VERTEXAI_TOOLS, |
| 268 | + Mode.VERTEXAI_JSON, |
| 269 | + Mode.VERTEXAI_PARALLEL_TOOLS, |
| 270 | + # Gemini |
| 271 | + Mode.GEMINI_TOOLS, |
| 272 | + Mode.GEMINI_JSON, |
| 273 | + # OpenRouter |
| 274 | + Mode.OPENROUTER_STRUCTURED_OUTPUTS, |
| 275 | + } |
| 276 | + |
| 277 | + # Check that all expected deprecated modes are in the mapping |
| 278 | + for mode in expected_deprecated: |
| 279 | + assert mode in DEPRECATED_MODE_MAPPING, ( |
| 280 | + f"Mode.{mode.name} should be in DEPRECATED_MODE_MAPPING" |
| 281 | + ) |
0 commit comments