[Bug]:Reconstruction error in `test_embedding_function_config_roundtrip`

[sarahec]

What happened?

While building for nixpkgs, TestEmbeddingFunctionSchemas.test_embedding_function_config_roundtrip[simple_ef] fails with:

E       AssertionError: Configs don't match after recreation for simple_ef
E       assert <MagicMock na...727821456000'> == {'dim': <Magi...33900358720'>}
E
E         Full diff:
E         + <MagicMock name='mock.get_config()' id='140727821456000'>
E         - {
E         -     'dim': <MagicMock name='mock.get_config().__getitem__()' id='140733900358720'>,
E         - }

Versions

Chromadb 1.3.6, Python 3.13.9, on x86_64-Linux and aarch64-Darwin

Relevant log output

_ TestEmbeddingFunctionSchemas.test_embedding_function_config_roundtrip[simple_ef] _
[gw2] linux -- Python 3.12.12 /nix/store/fdibxyh7xcmqrc172y78awzhxs292gq1-python3-3.12.12/bin/python3.12

self = <test_embedding_function_schemas.TestEmbeddingFunctionSchemas object at 0x7fff29356360>
ef_name = 'simple_ef'
mock_embeddings = <function mock_embeddings.<locals>._mock_embeddings at 0x7ffdbfcc62a0>
mock_common_deps = <_pytest.monkeypatch.MonkeyPatch object at 0x7ffdbfced670>

    @pytest.mark.parametrize("ef_name", get_embedding_function_names())
    def test_embedding_function_config_roundtrip(
        self,
        ef_name: str,
        mock_embeddings: Callable[[Documents], Embeddings],
        mock_common_deps: MonkeyPatch,
    ) -> None:
        """Test embedding function configuration roundtrip"""
        ef_class = known_embedding_functions[ef_name]

        # Create an autospec of the embedding function class
        mock_ef = create_autospec(ef_class, instance=True)

        # Mock the __call__ method
        mock_call = MagicMock(return_value=mock_embeddings(["test"]))
        mock_ef.__call__ = mock_call

        # For chroma-cloud-qwen, mock get_config to return valid data
        if ef_name == "chroma-cloud-qwen":
            from chromadb.utils.embedding_functions.chroma_cloud_qwen_embedding_function import (
                ChromaCloudQwenEmbeddingModel,
                CHROMA_CLOUD_QWEN_DEFAULT_INSTRUCTIONS,
            )

            mock_ef.get_config.return_value = {
                "api_key_env_var": "CHROMA_API_KEY",
                "model": ChromaCloudQwenEmbeddingModel.QWEN3_EMBEDDING_0p6B.value,
                "task": "nl_to_code",
                "instructions": CHROMA_CLOUD_QWEN_DEFAULT_INSTRUCTIONS,
            }

        # Mock the class constructor to return our mock instance
        mock_common_deps.setattr(
            ef_class, "__new__", lambda cls, *args, **kwargs: mock_ef
        )

        # Create instance with minimal args (constructor will be mocked)
        ef_instance = ef_class()

        # Get the config (this will use the real method)
        config = ef_instance.get_config()

        # Test recreation from config
        new_instance = ef_class.build_from_config(config)
        new_config = new_instance.get_config()

        # Configs should match
>       assert (
            config == new_config
        ), f"Configs don't match after recreation for {ef_name}"
E       AssertionError: Configs don't match after recreation for simple_ef
E       assert <MagicMock na...727821456000'> == {'dim': <Magi...33900358720'>}
E
E         Full diff:
E         + <MagicMock name='mock.get_config()' id='140727821456000'>
E         - {
E         -     'dim': <MagicMock name='mock.get_config().__getitem__()' id='140733900358720'>,
E         - }

chromadb/test/utils/test_embedding_function_schemas.py:82: AssertionError

[ramfrancis0x1]

Can u either (a) set an explicit dict return for simple_ef too, or (b) stop mocking new and instead instantiate the real EF (mock only external deps)?

Then send the output. Thanks

[sarahec]

Can u either (a) set an explicit dict return for simple_ef too, or (b) stop mocking new and instead instantiate the real EF (mock only external deps)?

Then send the output. Thanks

Hi Ram,

The code is all chromadb's, not mine. It looks like toString() is pretty-printing the mock (with newlines and indentation) and this doesn't match the all-inline comparison value.

What would happen if you tested specific fields in the mock instead?