chatterer 0.1.12__py3-none-any.whl → 0.1.14__py3-none-any.whl

chatterer/utils/code_agent.py

@@ -7,13 +7,44 @@ from typing import (
     NamedTuple,
     Optional,
     Self,
+    Sequence,
 )
 
 from langchain_core.runnables.config import RunnableConfig
 
+from ..messages import LanguageModelInput, SystemMessage
+
 if TYPE_CHECKING:
     from langchain_experimental.tools import PythonAstREPLTool
 
+# --- Constants ---
+DEFAULT_CODE_GENERATION_PROMPT = (
+    "You are equipped with a Python code execution tool.\n"
+    "Your primary goal is to generate Python code that effectively solves the *specific, immediate sub-task* required to progress towards the overall user request. The generated code and its resulting output will be automatically added to our conversation history.\n"
+    "\n"
+    "Guidelines for Optimal Tool Use:\n"
+    "- Conciseness and Efficiency: Write code that directly addresses the current need. Avoid unnecessary complexity, computations, or data loading. Tool execution has resource limits.\n"
+    "- Targeted Action: Focus only on the code required for the *next logical step*. Do not attempt to solve the entire problem in one code block if it involves multiple steps.\n"
+    "- Error Handling: Implement basic error handling (e.g., `try-except`) for operations that might fail (like file access or network requests, if applicable).\n"
+    "- Context Awareness: Assume the code runs in a stateful environment where variables and imports might persist from previous executions (unless explicitly cleared).\n"
+    "- Self-Contained Execution: Ensure the code block is runnable as provided. Define necessary variables within the block if they aren't guaranteed to exist from prior context.\n"
+    "\n"
+    "Output Format:\n"
+    "Return *only* a JSON object containing the Python code:\n"
+    '{\n  "code": "<your_python_code_here>"\n}\n\n'
+)
+
+DEFAULT_FUNCTION_REFERENCE_PREFIX_PROMPT = (
+    "The following Python functions are available in the global scope for you to use directly in your code.\n"
+    "You do not need to define these functions; simply call them as needed.\n"
+    "Use these functions only when they directly help in solving the current task. You are not obligated to use them.\n"
+)
+
+DEFAULT_FUNCTION_REFERENCE_SEPARATOR = "\n---\n"  # Separator to distinguish different function references
+
+
+# --- Helper Classes and Functions ---
+
 
 class FunctionSignature(NamedTuple):
     name: str
@@ -24,57 +55,63 @@ class FunctionSignature(NamedTuple):
     def from_callable(cls, callables: Optional[Callable[..., object] | Iterable[Callable[..., object]]]) -> list[Self]:
         if callables is None:
             return []
-        if callable(callables):
+        # Correctly handle single callable case
+        if isinstance(callables, Callable) and not isinstance(callables, type):  # Exclude classes if not intended
             return [cls._from_callable(callables)]
-        return [cls._from_callable(callable) for callable in callables]
+        # Handle iterables
+        if isinstance(callables, Iterable):
+            return [cls._from_callable(c) for c in callables]
+        # If it's neither a callable nor an iterable of callables, return empty
+        return []
 
     @classmethod
-    def _from_callable(cls, callable: Callable[..., object]) -> Self:
+    def _from_callable(cls, callable_obj: Callable[..., object]) -> Self:
         """
         Get the name and signature of a function as a string.
         """
-        # Determine if the function is async
-        is_async_func = inspect.iscoroutinefunction(callable)
+        is_async_func = inspect.iscoroutinefunction(callable_obj)
         function_def = "async def" if is_async_func else "def"
 
-        # Determine the function name based on the type of callable
-        if inspect.isfunction(callable):
-            # For regular Python functions, use __code__.co_name
-            function_name = callable.__code__.co_name
-        elif hasattr(callable, "name"):
-            # For StructuredTool or similar objects with a 'name' attribute
-            function_name = callable.name  # type: ignore
-        elif hasattr(callable, "__name__"):
-            # For other callables with a __name__ attribute
-            function_name = callable.__name__
+        if inspect.isfunction(callable_obj):
+            function_name = callable_obj.__code__.co_name
+        elif hasattr(callable_obj, "name") and isinstance(getattr(callable_obj, "name"), str):
+            function_name = getattr(callable_obj, "name")
+        elif hasattr(callable_obj, "__name__"):
+            function_name = callable_obj.__name__
         else:
-            # Fallback to the class name if no name is found
-            function_name = type(callable).__name__
+            function_name = type(callable_obj).__name__
+
+        try:
+            signature_str = str(inspect.signature(callable_obj))
+        except ValueError:  # Handles built-ins or others without inspectable signatures
+            signature_str = "(...)"  # Placeholder signature
+
+        signature = f"{function_def} {function_name}{signature_str}:"
+        docstring = inspect.getdoc(callable_obj)
 
-        # Build the signature string
-        signature = f"{function_def} {function_name}{inspect.signature(callable)}:"
-        docstring = inspect.getdoc(callable)
         if docstring:
             docstring = f'"""{docstring.strip()}"""'
-            return cls(
-                name=function_name, callable=callable, signature=f"{signature}\n{textwrap.indent(docstring, '    ')}"
-            )
+            full_signature = f"{signature}\n{textwrap.indent(docstring, '    ')}"
         else:
-            return cls(name=function_name, callable=callable, signature=signature)
+            full_signature = signature
+
+        return cls(name=function_name, callable=callable_obj, signature=full_signature)
 
     @classmethod
     def as_prompt(
         cls,
         function_signatures: Iterable[Self],
-        prefix: Optional[str] = "You can use the pre-made functions below without defining them:\n",
-        sep: str = "\n---\n",
+        prefix: Optional[str] = DEFAULT_FUNCTION_REFERENCE_PREFIX_PROMPT,  # Use constant
+        sep: str = DEFAULT_FUNCTION_REFERENCE_SEPARATOR,  # Use constant
     ) -> str:
         """
-        Generate a prompt string from a list of callables.
+        Generate a prompt string from a list of function signatures.
         """
+        if not function_signatures:
+            return ""
         body: str = sep.join(fsig.signature for fsig in function_signatures)
         if prefix:
-            return f"{prefix}{body}"
+            return f"{prefix}\n{body}"  # Add newline for clarity
         return body
 
 
@@ -92,13 +129,14 @@ class CodeExecutionResult(NamedTuple):
         **kwargs: object,
     ) -> Self:
         """
-        Execute code using the Python Code Execution Language Model.
+        Execute code using the Python REPL tool.
         """
         if repl_tool is None:
             repl_tool = get_default_repl_tool()
         if function_signatures:
             insert_callables_into_global(function_signatures=function_signatures, repl_tool=repl_tool)
-        output = str(repl_tool.invoke(code, config=config, **kwargs))  # pyright: ignore[reportUnknownMemberType]
+        # Ensure kwargs are passed correctly if needed by invoke
+        output = str(repl_tool.invoke(code, config=config))  # pyright: ignore[reportUnknownMemberType]
         return cls(code=code, output=output)
 
     @classmethod
@@ -111,28 +149,89 @@ class CodeExecutionResult(NamedTuple):
         **kwargs: object,
     ) -> Self:
         """
-        Execute code using the Python Code Execution Language Model asynchronously.
+        Execute code using the Python REPL tool asynchronously.
         """
         if repl_tool is None:
             repl_tool = get_default_repl_tool()
         if function_signatures:
             insert_callables_into_global(function_signatures=function_signatures, repl_tool=repl_tool)
-        output = str(await repl_tool.ainvoke(code, config=config, **kwargs))  # pyright: ignore[reportUnknownMemberType]
+        # Ensure kwargs are passed correctly if needed by ainvoke
+        output = str(await repl_tool.ainvoke(code, config=config))  # pyright: ignore[reportUnknownMemberType]if not needed by base ainvoke
         return cls(code=code, output=output)
 
 
 def get_default_repl_tool() -> "PythonAstREPLTool":
-    from langchain_experimental.tools import PythonAstREPLTool
+    """Initializes and returns a default PythonAstREPLTool instance."""
+    try:
+        from langchain_experimental.tools import PythonAstREPLTool
 
-    return PythonAstREPLTool()
+        # You might want to configure specific globals/locals here if needed
+        return PythonAstREPLTool()
+    except ImportError:
+        raise ImportError(
+            "PythonAstREPLTool requires langchain_experimental. Install with: pip install langchain-experimental"
+        )
 
 
 def insert_callables_into_global(
     function_signatures: Iterable[FunctionSignature], repl_tool: "PythonAstREPLTool"
 ) -> None:
     """Insert callables into the REPL tool's globals."""
-    repl_globals: Optional[dict[str, object]] = repl_tool.globals  # pyright: ignore[reportUnknownMemberType]
-    if repl_globals is None:
-        repl_tool.globals = {fsig.name: fsig.callable for fsig in function_signatures}
+    # Accessing globals might depend on the specific REPL tool implementation.
+    # This assumes a .globals attribute exists and is a dict.
+    if not hasattr(repl_tool, "globals") or not isinstance(repl_tool.globals, dict):  # pyright: ignore[reportUnknownMemberType]
+        # Handle cases where .globals is not available or not a dict
+        # Maybe initialize it or log a warning/error
+        repl_tool.globals = {}  # Or handle appropriately
+
+    # Safely update globals
+    current_globals: dict[object, object] = repl_tool.globals  # pyright: ignore[reportUnknownMemberType]
+    for fsig in function_signatures:
+        current_globals[fsig.name] = fsig.callable
+    # No need to reassign if globals is mutable (dict)
+    # repl_tool.globals = current_globals
+
+
+def _add_message_first(messages: LanguageModelInput, prompt_to_add: str) -> LanguageModelInput:
+    """Prepends a SystemMessage to the beginning of the message list/string."""
+    if not prompt_to_add:  # Don't add empty prompts
+        return messages
+
+    if isinstance(messages, str):
+        # Prepend with a newline for separation
+        return f"{prompt_to_add}\n\n{messages}"
+    elif isinstance(messages, Sequence):
+        # Create a mutable copy if it's a tuple
+        msg_list = list(messages)
+        msg_list.insert(0, SystemMessage(content=prompt_to_add))
+        return msg_list
+    # Handle LangChain Core BaseMessagePromptTemplate or similar if needed
+    # elif hasattr(messages, 'to_messages'):
+    #    msg_list = messages.to_messages()
+    #    msg_list.insert(0, SystemMessage(content=prompt_to_add))
+    #    return msg_list # Or return a new prompt template if required
     else:
-        repl_globals.update({fsig.name: fsig.callable for fsig in function_signatures})
+        # Fallback or raise error for unsupported types
+        raise TypeError(f"Unsupported message input type: {type(messages)}")
+
+
+def augment_prompt_for_toolcall(
+    function_signatures: Iterable[FunctionSignature],
+    messages: LanguageModelInput,
+    prompt_for_code_invoke: Optional[str] = DEFAULT_CODE_GENERATION_PROMPT,
+    function_reference_prefix: Optional[str] = DEFAULT_FUNCTION_REFERENCE_PREFIX_PROMPT,
+    function_reference_seperator: str = DEFAULT_FUNCTION_REFERENCE_SEPARATOR,
+) -> LanguageModelInput:
+    """Adds function references and code invocation prompts to the messages."""
+    # Add function references first (if any)
+    func_prompt = FunctionSignature.as_prompt(
+        function_signatures, function_reference_prefix, function_reference_seperator
+    )
+    if func_prompt:
+        messages = _add_message_first(messages=messages, prompt_to_add=func_prompt)
+
+    # Then add the main code invocation prompt (if provided)
+    if prompt_for_code_invoke:
+        messages = _add_message_first(messages=messages, prompt_to_add=prompt_for_code_invoke)
+
+    return messages

chatterer/utils/imghdr.py

@@ -0,0 +1,148 @@
+"""
+Recognize image file formats based on their first few bytes (base64-encoded).
+Originally derived from Python's imghdr, modified for base64 inputs.
+"""
+
+import base64
+import math
+from typing import Callable, List, Literal, Optional
+
+ImageType = Literal["jpeg", "png", "gif", "tiff", "rgb", "pbm", "pgm", "ppm", "rast", "xbm", "bmp", "webp", "exr"]
+
+tests: List[Callable[[bytes], Optional[ImageType]]] = []
+
+
+def register_test(func: Callable[[bytes], Optional[ImageType]]) -> Callable[[bytes], Optional[ImageType]]:
+    tests.append(func)
+    return func
+
+
+def decode_prefix(b64_data: str, prefix_bytes: int = 32) -> bytes:
+    needed_chars = math.ceil(prefix_bytes * 4 / 3)
+    truncated_data = b64_data[:needed_chars]
+
+    try:
+        return base64.b64decode(truncated_data)
+    except Exception:
+        return base64.b64decode(b64_data)
+
+
+def what(b64_data: str) -> Optional[ImageType]:
+    """
+    base64 인코딩된 문자열에 포함된 이미지의 타입을 반환한다.
+
+    :param b64_data: 이미지 데이터를 담은 base64 문자열.
+    :return: 이미지 포맷 문자열 (예: "jpeg", "png", "gif", 등) 또는 인식되지 않으면 None.
+    """
+    h: bytes = decode_prefix(b64_data, prefix_bytes=32)
+
+    for tf in tests:
+        res = tf(h)
+        if res:
+            return res
+    return None
+
+
+# --- 테스트 함수들 --- #
+
+
+@register_test
+def test_jpeg(h: bytes) -> Optional[ImageType]:
+    if len(h) >= 10 and h[6:10] in (b"JFIF", b"Exif"):
+        return "jpeg"
+    elif h.startswith(b"\xff\xd8\xff\xdb"):
+        return "jpeg"
+    return None
+
+
+@register_test
+def test_png(h: bytes) -> Optional[ImageType]:
+    if h.startswith(b"\x89PNG\r\n\x1a\n"):
+        return "png"
+    return None
+
+
+@register_test
+def test_gif(h: bytes) -> Optional[ImageType]:
+    if h.startswith(b"GIF87a") or h.startswith(b"GIF89a"):
+        return "gif"
+    return None
+
+
+@register_test
+def test_tiff(h: bytes) -> Optional[ImageType]:
+    if h[:2] in (b"MM", b"II"):
+        return "tiff"
+    return None
+
+
+@register_test
+def test_rgb(h: bytes) -> Optional[ImageType]:
+    if h.startswith(b"\x01\xda"):
+        return "rgb"
+    return None
+
+
+@register_test
+def test_pbm(h: bytes) -> Optional[ImageType]:
+    if len(h) >= 3 and h[0] == ord(b"P") and h[1] in b"14" and h[2] in b" \t\n\r":
+        return "pbm"
+    return None
+
+
+@register_test
+def test_pgm(h: bytes) -> Optional[ImageType]:
+    if len(h) >= 3 and h[0] == ord(b"P") and h[1] in b"25" and h[2] in b" \t\n\r":
+        return "pgm"
+    return None
+
+
+@register_test
+def test_ppm(h: bytes) -> Optional[ImageType]:
+    if len(h) >= 3 and h[0] == ord(b"P") and h[1] in b"36" and h[2] in b" \t\n\r":
+        return "ppm"
+    return None
+
+
+@register_test
+def test_rast(h: bytes) -> Optional[ImageType]:
+    if h.startswith(b"\x59\xa6\x6a\x95"):
+        return "rast"
+    return None
+
+
+@register_test
+def test_xbm(h: bytes) -> Optional[ImageType]:
+    if h.startswith(b"#define "):
+        return "xbm"
+    return None
+
+
+@register_test
+def test_bmp(h: bytes) -> Optional[ImageType]:
+    if h.startswith(b"BM"):
+        return "bmp"
+    return None
+
+
+@register_test
+def test_webp(h: bytes) -> Optional[ImageType]:
+    if len(h) >= 12 and h.startswith(b"RIFF") and h[8:12] == b"WEBP":
+        return "webp"
+    return None
+
+
+@register_test
+def test_exr(h: bytes) -> Optional[ImageType]:
+    if h.startswith(b"\x76\x2f\x31\x01"):
+        return "exr"
+    return None
+
+
+if __name__ == "__main__":
+    example_png_base64 = (
+        "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mP8/5+BAQAE/wH+U6az4wAAAABJRU5ErkJggg=="
+    )
+
+    fmt = what(example_png_base64)
+    print(f"Detected format: {fmt}")  # Expected: png