fix: log non-OpenAI format= assumption once at init (#1502)

mellea/backends/openai.py

@@ -195,7 +195,9 @@ def __init__(
 
         # Use provided parameters or fall back to environment variables
         self._api_key = api_key
-        self._base_url = base_url
+        # Resolve env here (not only in the SDK) so _server_type / init logging
+        # see the same host the client will actually call.
+        self._base_url = base_url or os.getenv("OPENAI_BASE_URL")
 
         # Validate that we have the required configuration
         if self._api_key is None and os.getenv("OPENAI_API_KEY") is None:
@@ -205,7 +207,7 @@ def __init__(
                 "  2. Pass it as a parameter: OpenAIBackend(api_key='your-key-here')"
             )
 
-        if self._base_url is None and os.getenv("OPENAI_BASE_URL") is None:
+        if self._base_url is None:
             MelleaLogger.get_logger().warning(
                 "OPENAI_BASE_URL or base_url is not set.\n"
                 "The openai SDK is going to assume that the base_url is `https://api.openai.com/v1`"
@@ -216,6 +218,16 @@ def __init__(
             if self._base_url is not None
             else _ServerType.OPENAI
         )  # type: ignore
+        if self._server_type != _ServerType.OPENAI:
+            MelleaLogger.get_logger().info(
+                "Mellea assumes you are NOT using the OpenAI platform, and that "
+                "other model providers have less strict requirements on supporting "
+                "JSON schemas passed into `format=`. If you encounter a server-side "
+                "error when using format=, then you found an exception to this "
+                "assumption. Please open an issue at "
+                "github.com/generative-computing/mellea with the stack trace and "
+                "your inference engine / model provider."
+            )
 
         self._openai_client_kwargs = self.filter_openai_client_kwargs(**kwargs)
 
@@ -968,9 +980,6 @@ async def _generate_from_chat_context_standard(
                     },
                 }
             else:
-                MelleaLogger.get_logger().info(
-                    "Mellea assumes you are NOT using the OpenAI platform, and that other model providers have less strict requirements on supporting JSON schemas passed into `format=`. If you encounter a server-side error following this message, then you found an exception to this assumption. Please open an issue at github.com/generative_computing/mellea with this stack trace and your inference engine / model provider."
-                )
                 extra_params["response_format"] = {
                     "type": "json_schema",
                     "json_schema": {

test/backends/test_openai_unit.py

@@ -7,6 +7,7 @@
 _simplify_and_merge, and _make_backend_specific_and_remove.
 """
 
+import os
 from unittest.mock import AsyncMock, MagicMock, PropertyMock, patch
 
 import pytest
@@ -433,6 +434,123 @@ class Answer(pydantic.BaseModel):
     assert "guided_json" in extra_body or "structured_outputs" in extra_body
 
 
+# --- #1502: non-OpenAI format= warning only at init ---
+
+_FORMAT_ASSUMPTION = "NOT using the OpenAI platform"
+
+
+def _info_msgs(mock_logger) -> list[str]:
+    return [str(c.args[0]) for c in mock_logger.info.call_args_list if c.args]
+
+
+def test_non_openai_format_assumption_logged_once_at_init():
+    mock_logger = MagicMock()
+    with (
+        patch(
+            "mellea.backends.openai.MelleaLogger.get_logger", return_value=mock_logger
+        ),
+        patch(
+            "mellea.backends.openai.is_vllm_server_with_structured_output",
+            return_value=False,
+        ),
+    ):
+        OpenAIBackend(
+            model_id="gpt-4o", api_key="fake-key", base_url="http://localhost:9999/v1"
+        )
+        OpenAIBackend(
+            model_id="gpt-4o", api_key="fake-key", base_url="http://localhost:9999/v1"
+        )
+
+    matches = [m for m in _info_msgs(mock_logger) if _FORMAT_ASSUMPTION in m]
+    assert len(matches) == 2  # once per backend instance
+
+
+def test_openai_platform_skips_format_assumption_log():
+    mock_logger = MagicMock()
+    # Unset OPENAI_BASE_URL for this test: after resolving env into _base_url,
+    # a leftover non-OpenAI env would make the no-base_url construction log.
+    # These backends point at api.openai.com, so mock the vLLM version probe:
+    # __init__ calls is_vllm_server_with_structured_output unconditionally,
+    # which would otherwise make a real GET to api.openai.com/version.
+    with (
+        patch(
+            "mellea.backends.openai.MelleaLogger.get_logger", return_value=mock_logger
+        ),
+        patch(
+            "mellea.backends.openai.is_vllm_server_with_structured_output",
+            return_value=False,
+        ),
+        patch.dict(os.environ),
+    ):
+        os.environ.pop("OPENAI_BASE_URL", None)
+        OpenAIBackend(
+            model_id="gpt-4o", api_key="fake-key", base_url="https://api.openai.com/v1"
+        )
+        OpenAIBackend(model_id="gpt-4o", api_key="fake-key")
+
+    matches = [m for m in _info_msgs(mock_logger) if _FORMAT_ASSUMPTION in m]
+    assert matches == []
+
+
+def test_format_assumption_log_honors_openai_base_url_env():
+    """Env-only non-OpenAI base_url must still classify as non-OpenAI at init."""
+    mock_logger = MagicMock()
+    with (
+        patch(
+            "mellea.backends.openai.MelleaLogger.get_logger", return_value=mock_logger
+        ),
+        patch(
+            "mellea.backends.openai.is_vllm_server_with_structured_output",
+            return_value=False,
+        ),
+        patch.dict(
+            os.environ, {"OPENAI_BASE_URL": "http://localhost:9999/v1"}, clear=False
+        ),
+    ):
+        OpenAIBackend(model_id="gpt-4o", api_key="fake-key")
+
+    matches = [m for m in _info_msgs(mock_logger) if _FORMAT_ASSUMPTION in m]
+    assert len(matches) == 1
+
+
+async def test_format_assumption_not_relogged_per_generation():
+    """#1502: the notice must not repeat on every format= generation."""
+    import pydantic
+
+    from mellea.core.base import CBlock
+    from mellea.stdlib.context import ChatContext
+
+    class Answer(pydantic.BaseModel):
+        value: int
+
+    backend = OpenAIBackend(
+        model_id="gpt-4o", api_key="fake-key", base_url="http://localhost:9999/v1"
+    )
+    ctx = ChatContext().add(CBlock(value="q"))
+    resp = MagicMock()
+    resp.choices = [MagicMock()]
+    resp.choices[0].message.content = "{}"
+    resp.choices[0].message.role = "assistant"
+
+    mock_logger = MagicMock()
+    with (
+        patch(
+            "mellea.backends.openai.MelleaLogger.get_logger", return_value=mock_logger
+        ),
+        patch.object(
+            backend._async_client.chat.completions, "create", new_callable=AsyncMock
+        ) as create,
+    ):
+        create.return_value = resp
+        for _ in range(3):
+            await backend.generate_from_chat_context(
+                CBlock(value="q"), ctx, _format=Answer, model_options={}
+            )
+
+    msgs = [str(c.args[0]) for c in mock_logger.info.call_args_list if c.args]
+    assert [m for m in msgs if _FORMAT_ASSUMPTION in m] == []
+
+
 def test_default_extra_body_applied_when_no_per_call_override():
     """Construction-time default_extra_body is present in every merged result."""
     backend = OpenAIBackend(