Performance asymmetry with overlays

Performance asymmetry with overlays

[Satyaki Das]

[-- Attachment #1: Type: text/plain, Size: 1126 bytes --]


There seems to be an asymmetry in performance when overlays are
used depending on initial cursor position in a buffer. Here is a
test that demonstrates it:

0. Save the attached file as overlay-test.el

1. Start emacs with:
      emacs -q --no-site-file -l overlay-test.el

2. Open a large file. I tried this with the NEWS file by doing:
      C-h n

3. Go to the end of the buffer (with M->) and do:
      M-x overlay-test RET
   It runs very quickly (on my 800MHz Pentium 3 it takes 0.22 sec)

4. Remove the overlays with: M-x overlay-test-reset.

5. Go to the beginning of the buffer and run overlay-test again.
   This time it takes about 71.5 sec!

[I got the times by using elp]

I am running CVS Emacs from yesterday on a Redhat GNU/Linux 9.0
on a Pentium 3. I get similar results with Emacs-21.3.1 built on
the same machine. I also tried on a different Solaris machine
with Emacs-21.2 and got similar results.

I did profile overlay-test. It seems that when cursor is at top of
the buffer initially, make-overlay and forward-line both take much
longer than in the other case.

Can this be improved?

Thanks,
Satyaki


[-- Attachment #2: Test for Overlays --]
[-- Type: text/plain, Size: 484 bytes --]

(defvar overlay-test-string "%")

(defun overlay-test ()
  (interactive)
  (let ((start (point-min))
        (end (point-max))
        ov)
    (goto-char start)
    (beginning-of-line)
    (while (and (not (eobp)) (< (point) end))
      (setq ov (make-overlay (point) (1+ (point))))
      (overlay-put ov 'display overlay-test-string)
      (forward-line))))

(defun overlay-test-reset ()
  (interactive)
  (dolist (ov (overlays-in (point-min) (point-max)))
    (delete-overlay ov)))

[-- Attachment #3: Type: text/plain, Size: 142 bytes --]

_______________________________________________
Emacs-devel mailing list
Emacs-devel@gnu.org
http://mail.gnu.org/mailman/listinfo/emacs-devel

Re: Performance asymmetry with overlays

[Luc Teirlinck]

Apparently your particular test runs a lot faster with overlays
centered at point-max.  I do not know whether that is really a general
rule, however.  Somebody who knows more about the internal
implementation of overlays than I do should comment on that.

The folllowing variation on your test contains only one single added
line:

(overlay-recenter (point-max)) 

With this change, execution speed seems to be essentially independent
of the starting position.

(defun overlay-test ()
  (interactive)
  (let ((start (point-min))
        (end (point-max))
        ov)
    (goto-char start)
    (overlay-recenter (point-max))
    (beginning-of-line)
    (while (and (not (eobp)) (< (point) end))
      (setq ov (make-overlay (point) (1+ (point))))
      (overlay-put ov 'display overlay-test-string)
      (forward-line))))

Re: Performance asymmetry with overlays

[Satyaki Das]

Luc Teirlinck <teirllm@dms.auburn.edu> writes:

> Apparently your particular test runs a lot faster with overlays
> centered at point-max.  I do not know whether that is really a general
> rule, however.  Somebody who knows more about the internal
> implementation of overlays than I do should comment on that.
> 
> The folllowing variation on your test contains only one single added
> line:
> 
> (overlay-recenter (point-max)) 
> 
> With this change, execution speed seems to be essentially independent
> of the starting position.

Thanks, that performs much better! I needed an overlay-recenter in
the other function too. Otherwise that becomes slow.

Satyaki