aw.py_gen

Function makeing and calling

aw.py_gen.extract_function_from_code(code_str: str) → callable

Safely extract a function from a code string.

Parameters:

code_str – String containing a Python function definition

Returns:

The function object

Raises:

ValueError – If no function is found or multiple functions exist

Example

>>> code = 'def add(a, b):\n    return a + b'
>>> func = extract_function_from_code(code)
>>> func(2, 3)
5

aw.py_gen.make_code_generator(template: str = '\nYou are a Python code generator. Your task is to create a complete, executable Python function.\n\n## TASK DESCRIPTION\n{task}\n\n## OUTPUT SCHEMA\nThe function should return a value that conforms to this JSON schema:\n```json\n{output_schema}\n```\n\nIf output_schema is null or empty, the function can return any appropriate value.\n\n## REQUIREMENTS\n\n### Function Signature\n- Function name: `{name}`\n- Parameters: Extract from the task description (indicated by {{braces}})\n- All parameters should have type hints where possible\n- Use clear, descriptive parameter names\n\n### Code Quality\n- Write ONLY the function definition (no imports in function body, no examples, no explanations)\n- Include a minimal docstring (one line describing what it does)\n- Use type hints for parameters and return value\n- Handle edge cases appropriately (e.g., empty inputs, None values)\n- Use built-in functions and standard library where possible\n- Keep the code simple and readable\n\n### Output Format\n- If returning a dict matching a schema, ensure all required fields are present\n- If the schema specifies types, ensure they match (e.g., "number" → int or float)\n- Return values directly without unnecessary wrapping\n\n### Code Style\n- Follow PEP 8 conventions\n- Use meaningful variable names\n- Prefer comprehensions over loops where readable\n- No print statements or side effects unless the task requires them\n\n## CRITICAL INSTRUCTIONS\n- Respond with ONLY the function definition\n- Do NOT include: imports, examples, explanations, markdown formatting, or test code\n- Do NOT wrap the code in markdown code blocks\n- The response must be valid Python that can be directly executed with `exec()`\n\n## EXAMPLE FORMAT (for reference only, do not include this in your response)\n```python\ndef example_function(param1: int, param2: str) -> dict:\n    """Brief description of what this does."""\n    # function body\n    return {{"result": param1}}\n```\n\nNOW GENERATE THE FUNCTION.\n', code_schema: dict = {'name': 'generated_python_function', 'schema': {'properties': {'code': {'description': 'Complete Python function definition as a string', 'type': 'string'}}, 'required': ['code'], 'type': 'object'}}, prompt_json_function_maker=None)

Create a code generation function from a template and schema.

Parameters:

• template – The prompt template for code generation

• code_schema – JSON schema defining the output format

• prompt_json_function_maker – Factory function (defaults to oa.prompt_json_function)

Returns:

A function that generates code from task descriptions

Example

>>> from oa import prompt_json_function
>>> write_code = make_code_generator(
...     prompt_json_function_maker=prompt_json_function
... )
>>> result = write_code(
...     task='Add {a} and {b}',
...     output_schema='{"type": "object", "properties": {"sum": {"type": "number"}}}',
...     name='add_numbers'
... )
>>> func = extract_function_from_code(result['code'])
>>> func(2, 3)
{'sum': 5}

aw.py_gen.task_to_function(task: str, output_schema: dict | str | None = None, name: str = 'generated_function', code_generator=None, **generator_kwargs) → callable

End-to-end: Convert a task description to an executable function.

Parameters:

• task – Natural language description of the function’s purpose

• output_schema – JSON schema for the return value (optional)

• name – Name for the generated function

• code_generator – Code generation function (creates one if None)

• **generator_kwargs – Additional arguments for code generator

Returns:

Executable Python function

Example

>>> func = task_to_function(
...     task='Multiply {x} and {y}',
...     output_schema='{"type": "object", "properties": {"product": {"type": "number"}}}',
...     name='multiply'
... )
>>> func(3, 4)
{'product': 12}