In some circumstances, it's useful to be able in inspect the parameters with which an instance of functools.lru_cache was instantiated. It's currently possible to recover the cache's maxsize via the .cache_info() method, but there's no way to recover the value passed for typed, which controls whether the lru_cache's cache is partitioned by input type.

This came up in the context of cloudpickle, a library that tries to extend pickle to support more types (in particular, interactively-defined functions and classes) for use-cases like cluster computing.

It's currently not possible to pickle an lru-cache decorated function that's defined in main (which includes, e.g. a Jupyter Notebook). We can almost fix this with a pure library solution (see cloudpipe/cloudpickle#309), but we're currently blocked by the fact that there's no way to recover the value that was passed for typed. Exposing a .typed attribute on the extension type for lru_cached functions fixes this

For more discussion, see the above linked PR, along with cloudpipe/cloudpickle#178.

I think it's a good idea to expose the full arguments that people use in lru_cache()

I have already make a PR for this issue

but here's a problem. add a new field to cache_info maybe cause some special problem for the third-party libs what're dependent the cache_info

I'm fine with adding a new function:

   >>> f.cache_parameters()
   {'maxsize': 200, 'typed'=False}

I think add new function would be a better way

I will make a new PR

It is easy:

diff --git a/Lib/functools.py b/Lib/functools.py
index 3192bd02d9..52c07db749 100644
--- a/Lib/functools.py
+++ b/Lib/functools.py
@@ -499,7 +499,9 @@ def lru_cache(maxsize=128, typed=False):
         # The user_function was passed in directly via the maxsize argument
         user_function, maxsize = maxsize, 128
         wrapper = _lru_cache_wrapper(user_function, maxsize, typed, _CacheInfo)
-        return update_wrapper(wrapper, user_function)
+        func = update_wrapper(wrapper, user_function)
+        func.cache_parameters = lambda: {'maxsize': maxsize, 'typed': typed}
+        return func
     elif maxsize is not None:
         raise TypeError(
             'Expected first argument to be an integer, a callable, or None')

But there are many design questions. Why method instead of just attribute?

        func.cache_parameters = {'maxsize': maxsize, 'typed': typed}

Or maybe just add the "typed" attribute?

        func.typed = typed

Also I consider adding the more general "make_key" parameter to lru_cache(). The "typed" parameter would just specify the default value for "make_key".

I think to modify in lru_cache should be good in normal circumstance

But here's maybe some people modify the wrapped object that underlying in lru_cache

So I prefer to add a new function in _functools.c?

I have already make a new function like this

static PyObject *
lru_cache_cache_parameters(lru_cache_object *self)
{
    PyObject *cache_parameters =  PyDict_New();
    PyDict_SetItemString(cache_parameters, "maxsize", PyLong_FromSsize_t(self->maxsize));
    PyDict_SetItemString(cache_parameters, "typed", self->typed == 0 ? Py_False : Py_True);
    return cache_parameters;
}

The rule used in the lru_cache implementation is: do not write in C that can be written in Python.

Yes, you are right, we shouldn't consider about the unstandard using way. I will update my PR

BTW, what're means for the "make_key" parameter?

New changeset 051ff52 by Raymond Hettinger (Manjusaka) in branch 'master':
bpo-38565: add new cache_parameters method for lru_cache (GH-16916)
051ff52

Scott, thank you for the suggestion.
Serhiy, thank for pointing to the simplest implementation.
Zheaoli, thanks for the patch.

Raymond, thanks for fixing many errors for my patch!