From owner-freebsd-bugs  Fri Apr  4 02:09:32 1997
Return-Path: <owner-bugs>
Received: (from root@localhost)
          by freefall.freebsd.org (8.8.5/8.8.5) id CAA00298
          for bugs-outgoing; Fri, 4 Apr 1997 02:09:32 -0800 (PST)
Received: from spinner.DIALix.COM (root@spinner.dialix.com [192.203.228.67])
          by freefall.freebsd.org (8.8.5/8.8.5) with ESMTP id CAA00261;
          Fri, 4 Apr 1997 02:08:14 -0800 (PST)
Received: from spinner.DIALix.COM (peter@localhost.DIALix.oz.au [127.0.0.1])
          by spinner.DIALix.COM (8.8.5/8.8.5) with ESMTP id RAA01247;
          Fri, 4 Apr 1997 17:52:16 +0800 (WST)
Message-Id: <199704040952.RAA01247@spinner.DIALix.COM>
X-Mailer: exmh version 2.0gamma 1/27/96
To: Eugeny Kuzakov <kev@l321.omsk.net.ru>
cc: "Alexander V. Lukyanov" <lav@video.yars.free.net>, bugs@freebsd.org,
        ache@freebsd.org
Subject: Re: sscanf is slow on 2.2.1 
In-reply-to: Your message of "Fri, 04 Apr 1997 16:16:11 +0700."
             <Pine.BSD.3.95.970404161401.26348B-100000@l321.omsk.net.ru> 
Date: Fri, 04 Apr 1997 17:52:15 +0800
From: Peter Wemm <peter@spinner.DIALix.COM>
Sender: owner-bugs@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

Eugeny Kuzakov wrote:
> On Thu, 3 Apr 1997, Alexander V. Lukyanov wrote:
> 
> > The program below works awfully slowly on FreeBSD 2.2.1, and quite normally
> > on Solaris 2.5, Linux, Digital Unix etc.
> > 
> > Linux on i586 100Mhz:
> > real    0m26.020s
> > user    0m24.240s
> > sys     0m0.400s
> > 
> > FreeBSD 2.1.5 on i586 66Mhz:
> >       104.06 real        51.97 user         0.00 sys
> My machine:
> 	P100/32MB/FreeBSD 2.1/Proxy SQUID/router to ISP (33600Kb)/30
> logined users:
> 
>        30.65 real        29.33 user         0.02 sys
> 8-()
> 
> > 
> > FreeBSD 2.2.1 on i586 133Mhz:
> >       737.01 real       588.48 user         0.43 sys
> > 
> > More than 10 times slower than 2.1.5 on a weaker system.
> > Obviously, that time is spent inside libc.


Hmm, most likely it's in the locale stuff.. :-(  Here's a suspicious set 
of diffs from -current realtive to 2.1:

@@ -712,15 +729,19 @@
                         * we just stored in the table (c).
                         */
                        n = *fmt;
-                       if (n == ']' || n < c) {
+                       if (n == ']' || __collate_range_cmp (n, c) < 0) {
                                c = '-';
                                break;  /* resume the for(;;) */
                        }
                        fmt++;
-                       do {            /* fill in the range */
-                               tab[++c] = v;
-                       } while (c < n);
+                       /* fill in the range */
+                       for (i = 0; i < 256; i ++)
+                               if (   __collate_range_cmp (c, i) < 0
+                                   && __collate_range_cmp (i, n) <= 0
+                                  )
+                               tab[i] = v;
 #if 1  /* XXX another disgusting compatibility hack */
+                       c = n;
                        /*
                         * Alas, the V7 Unix scanf also treats formats

The commit log message was:
----------------------------
revision 1.5
date: 1997/01/16 07:36:14;  author: ache;  state: Exp;  lines: +13 -8
Use collate for national [a-z]-like ranges

Should go in 2.2
----------------------------

Ouch..

-Peter

> > ---
> > #include <stdio.h>
> > int main()
> > {
> > int i;
> > char str[256];
> > for(i=0; i<1000000; i++)
> > 	sscanf("abcde","%255[a-zA-Z]",str);
> > return 0;
> > }
> > ---
> >    Alexander.
> > 
> 
> 	Best wishes, Eugeny Kuzakov
> 		Laboratory 321 ( Omsk, Russia )
> 		kev@l321.omsk.net.ru

Cheers,
-Peter