Re: make-hash-table :size

[Oliver Scholz]

Pascal Bourguignon <spam@mouse-potato.com> writes:

> Kai Grossjohann <kai@emptydomain.de> writes:
>
>> Joost Kremers <joostkremers@yahoo.com> writes:
>> 
>> > i'm writing a program in emacs that makes rather extensive use of small
>> > hash tables, about a dozen elemens or so. i'm going over my code again, to
>> > see if i can improve things here and there, and i suddenly realise that
>> > MAKE-HASH-TABLE takes a :size argument.
>> 
>> I wonder if hash tables are really faster than alists in this case.
>> Have you tried both?
>
> Well, it seems that yes, non-empty emacs hash tables are faster than
> alists...
[...]

For about a dozen elements? I doubt that. I would assume that the
small overhead in hash key access is likely to make alists more
efficient for such few elements. And indeed, profiling supports that
assumption:


(defconst test-alist
  (let ((lst (make-list 14 '(lirum . larum))))
    (append lst '((alpha . beta)))))

(defconst test-hash
  (let ((hash (make-hash-table :size 15 :test 'eq)))
    (puthash 'alpha 'beta hash)
    hash))

(defun test-hash-access ()
  (gethash 'alpha test-hash))

(defun test-alist-access ()
  (cdr (assq 'alpha test-alist)))


(defun test-run ()
  (interactive)
  (dolist (func '(test-hash-access test-alist-access))
    (unless (byte-code-function-p
             (symbol-function func))
      (byte-compile func)))
  (elp-instrument-function 'test-hash-access)
  (elp-instrument-function 'test-alist-access)
  (dotimes (ignore-me 10000)
    (test-hash-access)
    (test-alist-access))
  (elp-results))


On my Emacs 21.3 on a 1.5 GHtz Athlon M-x test-run gives:

Function Name      Call Count  Elapsed Time  Average Time
=================  ==========  ============  ============
test-hash-access   10000       0.09          9e-006
test-alist-access  10000       0.04          4e-006



    Oliver
-- 
13 Fructidor an 212 de la Révolution
Liberté, Egalité, Fraternité!