For 5.0 - add in cachedproperty?

[whardier]

I might be able to help contribute here. Didn't know if this was already planned: Similar: https://github.com/python/cpython/blob/3.10/Lib/functools.py#L934

[tkem]

AFAIK, @cached_property is just a method executed once and its result then "cached" indefinitely. I don't see how a cachetools implementation could improve on that?

[whardier]

The decorator for a property is different.

Take for instance:

import time

from cachetools import TTLCache, cached, cachedmethod


cache = TTLCache(maxsize=3, ttl=5)


class Upstream(dict):
    #@cache(cache=cache) # Nope
    #@cachedmethod(cache=cache) # Nope
    @property
    #@cachedmethod(cache=cache) # Nope
    #@cache(cache=cache) # Nope
    def frequently_accessed_thing(self) -> int:
        return time.time_ns()


upstream = Upstream()

print(upstream.frequently_accessed_thing)
print(upstream.frequently_accessed_thing)

The decorators don't properly account for what's needed to properly represent a property as this does:

import time
from functools import cached_property


class Upstream(dict):
    @cached_property
    def frequently_accessed_thing(self) -> int:
        return time.time_ns()


upstream = Upstream()

print(upstream.frequently_accessed_thing)
print(upstream.frequently_accessed_thing)

The only way to properly integrate an LRU or TTL cache via a property is to use an inner function (or get really explicit)

import time
from cachetools import cached, TTLCache

cache = TTLCache(maxsize=100, ttl=60)


class Upstream(dict):
    @property
    def frequently_accessed_thing(self) -> int:
        @cached(cache=cache)
        def _frequently_accessed_thing():
            return time.time_ns()

        return _frequently_accessed_thing()


upstream = Upstream()

print(upstream.frequently_accessed_thing)
print(upstream.frequently_accessed_thing)

So cachetools could provide a cachedproperty decorator akin to functools.cached_property in order to help this work.

[tkem]

So what would be the actual benefit of this, compared too the standard library @cached_property decorator?
The only difference I can see is when using a TTLCache, you could specify a timeout after which the cached property value would be updated. However, that could also (and maybe easier) be achieved by adding a ttl option to the standard library @cached_property decorator.

[whardier]

Being able to work directly with an externally managed cache store is the goal. We are using cachetools to help deal with things that aren't provided by the standard lib.

[whardier]

OK. I read the docs and I'm sorry for spinning this up. Thanks @tkem - I have a decorator in place wrapping cached(...) in that is using a hashed key attribute for the function name.. works like a champ and easy to implement without patching this lib.

[tkem]

@whardier: Great to hear you got this solved! Would you mind posting your solution here (or providing a link if it's available elsewhere) for future reference?