Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84 SMI; site sun.uucp
Path: utzoo!watmath!clyde!burl!ulysses!allegra!mit-eddie!genrad!decvax!decwrl!sun!guy
From: guy@sun.uucp (Guy Harris)
Newsgroups: net.bugs.usg
Subject: "fgrep" is slow
Message-ID: <2734@sun.uucp>
Date: Fri, 30-Aug-85 20:17:37 EDT
Article-I.D.: sun.2734
Posted: Fri Aug 30 20:17:37 1985
Date-Received: Sun, 1-Sep-85 06:02:33 EDT
Distribution: net
Organization: Sun Microsystems, Inc.
Lines: 51

It's slow because it uses a subroutine to perform every(!) character
comparison.  The 4.2BSD one uses a macro; does V7's also use a macro?
Changing it to use a macro makes a major difference (like ~10 CPU seconds on
a 200,000 character file - that's ~40% of the CPU time on my Sun).

*** fgrep.c.BAK	Mon Aug 19 17:02:39 1985
--- fgrep.c	Fri Aug 30 16:04:56 1985
***************
*** 39,44
  extern	char *optarg;
  extern	int optind;
  
  main(argc, argv)
  char **argv;
  {

--- 39,48 -----
  extern	char *optarg;
  extern	int optind;
  
+ /* The following macro was inserted to allow for the "-i" option */
+ 
+ #define	same(a, b) ((a) == (b) || iflag && ((a) ^ (b)) == ' ' && letter(a) == letter(b))
+ 
  main(argc, argv)
  char **argv;
  {
***************
*** 356,368
  	}
  }
  
! /* The following functions were inserted to allow for the "-i" option */
! 
! same(a, b)
! register int a, b;
! {
! 	return (a == b || iflag && (a ^ b) == ' ' && letter(a) == letter(b));
! }
  
  letter(c)
  register int c;

--- 360,366 -----
  	}
  }
  
! /* The following function was inserted to allow for the "-i" option */
  
  letter(c)
  register int c;