Validate embedding_types elements before preserving them

The embedding_types merge only checked list-ness and non-emptiness, so a
caller extra containing non-string elements (e.g. [{"x": 1}]) was
preserved and would be sent as an invalid Cohere wire payload,
contradicting the "malformed falls back to [float]" comment. Require a
non-empty list of non-empty strings before preserving it; anything else
falls back to ["float"].

Author: chris-colinsky

src/openarmature/retrieval/providers/cohere.py

@@ -700,8 +700,12 @@ def _build_request_body(
         # only when a caller's embedding_types omits it (§8.4). A malformed /
         # empty embedding_types extra falls back to ["float"].
         caller_types = request_extras.get("embedding_types")
-        if isinstance(caller_types, list) and caller_types:
-            embedding_types = list(cast("list[Any]", caller_types))
+        if (
+            isinstance(caller_types, list)
+            and caller_types
+            and all(isinstance(t, str) and t for t in cast("list[object]", caller_types))
+        ):
+            embedding_types = list(cast("list[str]", caller_types))
             if "float" not in embedding_types:
                 embedding_types.append("float")
         else:

tests/unit/test_retrieval_provider.py

@@ -1184,18 +1184,22 @@ def handler(req: httpx.Request) -> httpx.Response:
 
     cfg_both = EmbeddingRuntimeConfig.model_validate({"embedding_types": ["float", "int8"]})
     cfg_int8 = EmbeddingRuntimeConfig.model_validate({"embedding_types": ["int8"]})
+    cfg_bad = EmbeddingRuntimeConfig.model_validate({"embedding_types": [{"x": 1}]})
     provider = _cohere_embed_provider(handler)
     # Caller adds int8 alongside float -> both reach the wire (float NOT clobbered).
     await provider.embed(["x"], config=cfg_both)
     # Caller omits float -> "float" is appended (the mapping must read embeddings.float).
     await provider.embed(["x"], config=cfg_int8)
     # No embedding_types extra -> the ["float"] default.
     await provider.embed(["x"])
+    # Malformed (non-string elements) -> falls back to ["float"], not sent verbatim.
+    await provider.embed(["x"], config=cfg_bad)
     await provider.aclose()
 
     assert captured[0]["embedding_types"] == ["float", "int8"]
     assert captured[1]["embedding_types"] == ["int8", "float"]
     assert captured[2]["embedding_types"] == ["float"]
+    assert captured[3]["embedding_types"] == ["float"]
 
 
 async def test_cohere_embed_per_chunk_count_mismatch_raises_despite_matching_total() -> None: