How can I reload a python class if file is changed on disk?

Question

To support runtime changes of parameters stored in source class file and used as an object with fields, how can I check if the source file for the object was modified since runtime started or since the last time it was reloaded, and reload the class and make a new instance of the object?

Answer 1

This method seems to work:

def reload_class_if_modified(obj:object, every:int=1)->object:
    """
    Reloads an object if the source file was modified since runtime started or since last reloaded

    :param obj: the original object
    :param every: only check every this many times we are invoked

    :returns: the original object if classpath file has not been modified 
              since startup or last reload time, otherwise the reloaded object
    """
    reload_class_if_modified.counter+=1
    if reload_class_if_modified.counter>1 and reload_class_if_modified.counter%every!=0:
        return obj
    try:
        module=inspect.getmodule(obj)
        cp=Path(module.__file__)
        mtime=cp.stat().st_mtime
        classname=type(obj).__name__

        if (mtime>reload_class_if_modified.start_time and (not (classname in reload_class_if_modified.dict))) \
                or ((classname in reload_class_if_modified.dict) and mtime>reload_class_if_modified.dict[classname]):
            importlib.reload(module)
            class_ =getattr(module,classname)
            o=class_()
            reload_class_if_modified.dict[classname]=mtime
            return o
        else:
            return obj
    except Exception as e:
        logger.error(f'could not reload {obj}: got exception {e}')
        return obj

reload_class_if_modified.dict=dict()
reload_class_if_modified.start_time=time()
reload_class_if_modified.counter=0

Use it like this:

import neural_mpc_settings
from time import sleep as sleep
g=neural_mpc_settings()
while True:
    g=reload_class_if_modified(g, every=10)
    print(g.MIN_SPEED_MPS, end='\r')
    sleep(.1)

where neural_mpc_settings is

class neural_mpc_settings():
    MIN_SPEED_MPS = 5.0

When I change neural_mpc_settings.py on disk, the class is reloaded and the new object returned reflects the new class fields.

Answer 2

You might want to consider using a library like watchdog, which would let you trigger a handler whenever the file is changed. Instead of collocating your parameters with the code, you could stored them in a data file, with a data loader method that was called on startup and whenever the underlying data file was changed.