From 3b843708b097d6c4ff96d374900282ddf5142f0d Mon Sep 17 00:00:00 2001 From: shivam Date: Sat, 18 Jul 2026 20:17:57 +0000 Subject: [PATCH] fix(bedrock): degrade gracefully on malformed tool-call arguments split_concatenated_json_objects re-raised JSONDecodeError on genuinely malformed (non-concatenated) tool-call arguments, which propagated out of _convert_to_bedrock_tool_call_invoke and turned every replayed Bedrock conversation into a 500. Catch the decode error, keep whatever complete objects parsed, log a warning, and let the caller fall back to input={} so the conversation continues. Fixes #18667 Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- .../prompt_templates/common_utils.py | 27 +++++++++--- ...ore_utils_prompt_templates_common_utils.py | 30 +++++++++++-- ...llm_core_utils_prompt_templates_factory.py | 43 +++++++++++++++++++ 3 files changed, 89 insertions(+), 11 deletions(-) diff --git a/litellm/litellm_core_utils/prompt_templates/common_utils.py b/litellm/litellm_core_utils/prompt_templates/common_utils.py index 538d5f650ef..db3856ce86b 100644 --- a/litellm/litellm_core_utils/prompt_templates/common_utils.py +++ b/litellm/litellm_core_utils/prompt_templates/common_utils.py @@ -1679,16 +1679,19 @@ def split_concatenated_json_objects(raw: str) -> List[Dict[str, Any]]: This helper uses ``json.JSONDecoder.raw_decode()`` to walk the string and extract each JSON object individually. + The walk degrades gracefully: if the string is malformed or truncated + (e.g. a stream that ended mid-tool-call), whatever complete objects were + parsed before the bad tail are returned and the remainder is discarded + with a warning, rather than raising. The sole caller + (``_convert_to_bedrock_tool_call_invoke``) treats an empty result as + ``input={}`` so the conversation can continue instead of hard-failing. + Returns ------- list[dict] A list of parsed dicts – one per JSON object found. If *raw* is - empty or whitespace-only, an empty list is returned. - - Raises - ------ - json.JSONDecodeError - If the string contains text that cannot be parsed as JSON at all. + empty, whitespace-only, or wholly unparseable, an empty list is + returned. """ import json @@ -1708,7 +1711,17 @@ def split_concatenated_json_objects(raw: str) -> List[Dict[str, Any]]: if idx >= length: break - obj, end_idx = decoder.raw_decode(raw, idx) + try: + obj, end_idx = decoder.raw_decode(raw, idx) + except json.JSONDecodeError as e: + verbose_logger.warning( + "split_concatenated_json_objects: discarding unparseable tool-call " + "arguments tail after %d complete object(s); error=%s at char %d", + len(results), + e, + idx, + ) + break if isinstance(obj, dict): results.append(obj) else: diff --git a/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_common_utils.py b/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_common_utils.py index 1b1db634ed2..6d14d283b84 100644 --- a/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_common_utils.py +++ b/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_common_utils.py @@ -251,10 +251,32 @@ def test_split_concatenated_json_non_dict_value(): assert result == [{}] -def test_split_concatenated_json_invalid_raises(): - """Completely invalid JSON raises JSONDecodeError.""" - with pytest.raises(json.JSONDecodeError): - split_concatenated_json_objects("not json at all") +def test_split_concatenated_json_wholly_invalid_returns_empty(): + """ + Wholly unparseable JSON degrades to an empty list instead of raising. + + Regression for https://github.com/BerriAI/litellm/issues/18667: a raise + here propagated out of `_convert_to_bedrock_tool_call_invoke` and turned + every replayed conversation into a 500. + """ + assert split_concatenated_json_objects("not json at all") == [] + + +def test_split_concatenated_json_malformed_object_returns_empty(): + """ + A single malformed object (missing comma between keys) degrades to an + empty list rather than raising `Expecting ',' delimiter`. + """ + assert split_concatenated_json_objects('{"location": "Boston" "unit": "celsius"}') == [] + + +def test_split_concatenated_json_salvages_prefix_before_truncated_tail(): + """ + Complete objects parsed before an unparseable/truncated tail are kept; + only the bad tail is discarded. + """ + result = split_concatenated_json_objects('{"a": 1}{"b": 2}{"c":') + assert result == [{"a": 1}, {"b": 2}] # --------------------------------------------------------------------------- diff --git a/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py b/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py index bcda88ea609..17f680df4d9 100644 --- a/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py +++ b/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py @@ -2286,6 +2286,49 @@ def test_bedrock_tool_call_invoke_non_dict_arguments(): assert result[0]["toolUse"]["input"] == {} +def test_bedrock_tool_call_invoke_malformed_json_does_not_raise(): + """ + Regression for https://github.com/BerriAI/litellm/issues/18667. + + When the model emits malformed JSON in tool-call arguments (here a + missing comma between keys), replaying that history must NOT raise + `Unable to convert openai tool calls ... Expecting ',' delimiter`. + It degrades to an empty-object input so the conversation can continue. + """ + tool_calls = [ + { + "id": "toolu_abc123", + "type": "function", + "function": { + "name": "get_weather", + "arguments": '{"location": "Boston" "unit": "celsius"}', + }, + } + ] + result = _convert_to_bedrock_tool_call_invoke(tool_calls) + assert len(result) == 1 + assert result[0]["toolUse"]["toolUseId"] == "toolu_abc123" + assert result[0]["toolUse"]["name"] == "get_weather" + assert result[0]["toolUse"]["input"] == {} + + +def test_bedrock_tool_call_invoke_salvages_valid_prefix_before_truncated_tail(): + """ + A valid leading object followed by a truncated tail keeps the valid + object rather than dropping everything or raising. + """ + tool_calls = [ + { + "id": "call_partial", + "type": "function", + "function": {"name": "shell", "arguments": '{"cmd": "ls"}{"cmd":'}, + } + ] + result = _convert_to_bedrock_tool_call_invoke(tool_calls) + assert len(result) == 1 + assert result[0]["toolUse"]["input"] == {"cmd": "ls"} + + def test_make_valid_bedrock_tool_name_preserves_hyphens(): assert make_valid_bedrock_tool_name("my-tool") == "my-tool" assert (