langflow-ai / langflow

⛓️ Langflow is a visual framework for building multi-agent and RAG applications. It's open-source, Python-powered, fully customizable, model and vector store agnostic.

Home Page:http://www.langflow.org

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

[BUG] Unknown bug when using dynamic fields

yamonkjd opened this issue · comments

I am implementing a design that allows users to easily create tools using the Langchain StructuresTool. However, the first time I wrote the code as a custom component, the LangflowUI crashed. Upon restarting, the constructor works normally.

Here’s a description of the class:

  1. Receives user code input.

  2. In StructedTool, selects a class for the tool parameters to be provided as a schema via a dropdown.

  3. Writes the tool code.

image

This is Toolcode

class CalculatorInput(BaseModel):
    a: int = Field(description="first number")
    b: int = Field(description="second number")


def multiply(a: int, b: int) -> int:
    """Multiply two numbers."""
    return a * b


calculator = StructuredTool.from_function(
    func=multiply,
    name="Calculator",
    description="multiply numbers",
    args_schema=CalculatorInput,
    return_direct=True,
    # coroutine= ... <- you can specify an async method if desired as well
)

This is my customcomponent

from typing import Any, Dict, List, Callable
import ast
from langchain.agents import Tool
from langchain.tools import StructuredTool
from langflow.interface.custom.custom_component import CustomComponent
from langflow.schema.dotdict import dotdict
from langchain.pydantic_v1 import BaseModel, Field
 
class CodeStructuredTool(CustomComponent):
    display_name = "Code To Tool"
    description = "structuredtool dataclass code to tool"
    documentation = "https://python.langchain.com/docs/modules/tools/custom_tools/#structuredtool-dataclass"
    icon = "🐍"
    field_order = ["tool_code", "name", "description", "return_direct", "tool_function", "tool_class"]
 
    def build_config(self) -> Dict[str, Any]:
        return {
            "tool_code": {
                "display_name": "Tool Code",
                "info": "Enter the dataclass code.",
                "placeholder": "def my_function(args):\n    pass",
                "multiline": True,
                "refresh_button": True,
            },
            "name": {
                "display_name": "Tool Name",
                "info": "Enter the name of the tool.",
            },
            "description": {
                "display_name": "Description",
                "info": "Provide a brief description of what the tool does.",
            },
            "return_direct": {
                "display_name": "Return Directly",
                "info": "Should the tool return the function output directly?",
            },
            "tool_function": {
                "display_name": "Tool Function",
                "info": "Select the function for additional expressions.",
                "options": [],
                "refresh_button": True,
            },
            "tool_class": {
                "display_name": "Tool Class",
                "info": "Select the class for additional expressions.",
                "options": [],
                "refresh_button": True,
                "required": False,
            },
        }
        
    def parse_source_name(self, code:str)-> Dict:
        parsed_code = ast.parse(code)
        class_names = [node.name for node in parsed_code.body if isinstance(node, ast.ClassDef)]
        function_names = [node.name for node in parsed_code.body if isinstance(node, ast.FunctionDef)]
        return {"class": class_names, "function": function_names}

 
    def update_build_config(self, build_config: dotdict, field_value: Any, field_name: str | None = None) -> dotdict:
        if field_name == "tool_code" or field_name == "tool_function" or field_name == "tool_class":
            try:
                names = self.parse_source_name(build_config.tool_code.value)
                build_config.tool_class.options = names["class"]
                build_config.tool_function.options = names["function"]
            except Exception as e:
                self.status = f"Failed to extract class names: {str(e)}"
                build_config.tool_class.options = ["파싱실패", str(e)]
                build_config.tool_function.options = []
        return build_config
        
    async def build(self, tool_code: Code, name: str, description: str, tool_function: List[str], return_direct: bool, tool_class: List[str]=None) -> Tool:
        local_namespace = {}
        exec(tool_code, globals(), local_namespace)
        
        func = local_namespace[tool_function]
        _class = None
        
        if tool_class:
            _class = local_namespace[tool_class]
        
        tool = StructuredTool.from_function(
            func=func,
            args_schema=_class,
            name=name,
            description=description,
            return_direct=return_direct
        )
        return tool

Additionally, when using dynamic fields and setting realtime_refresh to True, it seems there may be a conflict with the Langflow UI history back, causing the code, text, or certain entered values to revert to their previous state (especially when selecting specific values from a dropdown menu as a List type).

While manually pressing the refresh button works fine, we hope it would function without having to press the button.