开发者

How can I implement an expiring LRU cache in elisp?

开发者 https://www.devze.com 2023-03-19 05:02 出处:网络
I have data that consists of the following three components: a_path a_key a_value =f(a_path, a_key) a_value is expensive to calculate, so I want to calcula开发者_StackOverflowte it infrequently. I

I have data that consists of the following three components:

  • a_path
  • a_key
  • a_value =f(a_path, a_key)

a_value is expensive to calculate, so I want to calcula开发者_StackOverflowte it infrequently. In an ideal world, that would be only when it's going to change. So, my requirements for this cache are as follows:

  • LRU cache with configurable maximum size
  • Keyed on (a_path, a_key)
  • Ability to expire an entry based on age (recalculate every hour or so, for example)
  • Ability to expire an entry based on expiry_func(a_path, a_key)

My googling has failed me here; I find a lot of Java sites even when searching for "elisp LRU cache".


Here's most of what you want: a fixed-size least-recently-used cache with O(1) lookup, O(1) insertion, and O(1) deletion.

It's slightly tricky to get all of these operations to be O(1), hence this slightly elaborate implementation. I combine a hash-table (for fast lookup) with a doubly-linked list of items (for fast removal, re-ordering, and finding the oldest element).

(require 'cl)

(defstruct lru-cache max-size size newest oldest table)

(defstruct lru-item key value next prev)

(defun lru-remove-item (item lru)
  (let ((next (lru-item-next item))
        (prev (lru-item-prev item)))
    (if next (setf (lru-item-prev next) prev) 
      (setf (lru-cache-newest lru) prev))
    (if prev (setf (lru-item-next prev) next)
      (setf (lru-cache-oldest lru) next))))

(defun lru-insert-item (item lru)
  (let ((newest (lru-cache-newest lru)))
    (setf (lru-item-next item) nil (lru-item-prev item) newest)
    (if newest (setf (lru-item-next newest) item)
      (setf (lru-cache-oldest lru) item))
    (setf (lru-cache-newest lru) item)))

;;; Public interface starts here.

(defun* lru-create (&key (size 65) (test 'eql))
  "Create a new least-recently-used cache and return it.
Takes keyword arguments
:SIZE the maximum number of entries (default: 65).
:TEST a hash table test (default 'EQL)."
  (make-lru-cache 
   :max-size size
   :size 0
   :newest nil
   :oldest nil
   :table (make-hash-table :size size :test test)))

(defun lru-get (key lru &optional default)
  "Look up KEY in least-recently-used cache LRU and return
its associated value.
If KEY is not found, return DEFAULT which defaults to nil."
  (let ((item (gethash key (lru-cache-table lru))))
    (if item
        (progn
          (lru-remove-item item lru)
          (lru-insert-item item lru)
          (lru-item-value item))
      default)))

(defun lru-rem (key lru)
  "Remove KEY from least-recently-used cache LRU."
  (let ((item (gethash key (lru-cache-table lru))))
    (when item
      (remhash (lru-item-key item) (lru-cache-table lru))
      (lru-remove-item item lru)
      (decf (lru-cache-size lru)))))

(defun lru-put (key value lru)
  "Associate KEY with VALUE in least-recently-used cache LRU.
If KEY is already present in LRU, replace its current value with VALUE."
  (let ((item (gethash key (lru-cache-table lru))))
    (if item
        (setf (lru-item-value item) value)
      (when (eql (lru-cache-size lru) (lru-cache-max-size lru))
        (lru-rem (lru-item-key (lru-cache-oldest lru)) lru))
      (let ((newitem (make-lru-item :key key :value value)))
        (lru-insert-item newitem lru)
        (puthash key newitem (lru-cache-table lru))
        (incf (lru-cache-size lru))))))

 ;;; Exercise for the reader: implement lru-clr and lru-map to complete the
 ;;; analogy with hash tables.

For your application that's keyed on pairs, you probably want to supply :test 'equal to lru-create. Or see Defining Hash Comparisons if you need something special.

I'll let you figure out how to do the time-based expiry; it should be straightforward from here.

(If anyone knows a simpler way to implement this while keeping the operations running in constant time, I'd be very interested to see it.)

0

精彩评论

暂无评论...
验证码 换一张
取 消