Review and resolve minor TODO items

* Workaround for lmstudio.js schema export bug is no longer needed
* Tool call parameters are validated, update comments accordingly
* Test suite covers the essential handle APIs in various test cases
* Adding signal subscription support is speculative, not definite

Author: ncoghlan

src/lmstudio/json_api.py

@@ -95,9 +95,7 @@
     LlmTool,
     LlmToolUseSettingToolArray,
     ModelInfo,
-    # TODO: Import this once the definition is fixed in lmstudio.js
-    #       https://github.com/lmstudio-ai/lmstudio.js/pull/222
-    # ModelInstanceInfo,
+    ModelInstanceInfo,
     ModelSearchOptsDict,
     ModelSpecifier,
     ModelSpecifierDict,
@@ -197,10 +195,6 @@
     EmbeddingRpcUnloadModelParameter | LlmRpcUnloadModelParameter
 )
 
-# TODO: Drop this workaround once the lmstudio.js schema export is fixed:
-#       https://github.com/lmstudio-ai/lmstudio.js/pull/222
-ModelInstanceInfo: TypeAlias = EmbeddingModelInstanceInfo | LlmInstanceInfo
-
 
 class ModelSessionTypes(Generic[TLoadConfig]):
     """Helper class to group related types for code sharing across model namespaces."""
@@ -516,8 +510,8 @@ def __init__(self, logger: StructuredLogger) -> None:
         self._last_channel_id = 0
         self._pending_calls: dict[int, TQueue] = {}
         self._last_call_id = 0
-        # TODO: add `_last_subscriber_id` when adding signal support
-        # TODO: add `_active_subscriptions` when adding signal support
+        # `_active_subscriptions` (if we add signal support)
+        # `_last_subscriber_id` (if we add signal support)
         self._logger = logger
 
     def all_queues(self) -> Iterator[TQueue]:
@@ -1288,6 +1282,7 @@ def request_tool_call(
             )
             result = _ToolCallResultData(content=err_msg, tool_call_id=tool_call_id)
             return lambda: result
+        # Validate parameters against their specification
         params_struct, implementation = client_tool
         raw_kwds = request.arguments
         try:
@@ -1300,7 +1295,7 @@ def request_tool_call(
             return lambda: result
         kwds = to_builtins(parsed_kwds)
 
-        # TODO: Validate parameters against their specification
+        # Allow caller to schedule the tool call request for background execution
         def _call_requested_tool() -> _ToolCallResultData:
             call_result = implementation(**kwds)
             return _ToolCallResultData(

tests/async/test_model_handles_async.py

@@ -45,10 +45,6 @@ async def test_completion_llm_handle_async(model_id: str, caplog: LogCap) -> Non
     assert response.content
 
 
-# TODO: also test `complete_stream`, `respond`, and `respond_stream`,
-#       as all the wrappers are implemented independently of each other
-
-
 @pytest.mark.asyncio
 @pytest.mark.lmstudio
 @pytest.mark.parametrize("model_id", (EXPECTED_EMBEDDING, EXPECTED_EMBEDDING_ID))

tests/sync/test_model_handles_sync.py

@@ -51,10 +51,6 @@ def test_completion_llm_handle_sync(model_id: str, caplog: LogCap) -> None:
     assert response.content
 
 
-# TODO: also test `complete_stream`, `respond`, and `respond_stream`,
-#       as all the wrappers are implemented independently of each other
-
-
 @pytest.mark.lmstudio
 @pytest.mark.parametrize("model_id", (EXPECTED_EMBEDDING, EXPECTED_EMBEDDING_ID))
 def test_embedding_handle_sync(model_id: str, caplog: LogCap) -> None: