Conference turris::digital_unix

9865.0. "Awk and Memory fault" by NNTPD::"[email protected]" (Adrian Morrisson) Mon May 19 1997 03:43

Hi
	I have a customer who has written an awk script. This script runs ok under
gawk and aix awk but under nawk/awk it drops a huge core file and gives a 
1104 Memory fault message at the end. The problem occures under 3.2x and 4.0x
of Digital Unix. 
Checking the core file shows nawk bombing at regcomp.


The awk script is below but the data files are at
http://www.stl.dec.com/~adrian/Q52975

Can anyone see an obvious problem with this script before I impt this?

Thanks kindly


Adrian Morrisson

[ -s list ] || { echo "No list file ?!!?" ; exit 2 ; }
[ -s ./db.excluded ] || { echo "No db.excluded file ?!!?" ; exit 2 ; }

nawk 'BEGIN { states = "seek"
             lg = "pickup.log"
             mf = "db.new_misc"
             xf = "db.remain"
             objcnt = 0 ; cnta = 0
             i = 0 ; ac = 0 ; j = 0
             list = ARGV[1] ; delete ARGV[1]
             while ( getline < list > 0 )  lst[$1] = ++i
             print "Loaded " i " relevant table names !\n" > lg
             close(list)
           }

     $0 ~ /^[   ]*--/  { next }
     NF == 0           { next }
     $0 ~ /^\{ TABLE / { next }

     states == "found" { print $0 > mf
                         if ( $0 ~ /;/ ) states = "seek"
                         next    }
     states == "doco" { if ( $0 ~ /;/ ) states = "seek"
                        next    }
     { for ( i in lst ) {
         match ($0, i)
         if ( RLENGTH > 0 ) {
           if ( substr($0, ((RSTART-1)), 1)       ~ /[_0-9a-zA-Z]/  ||
                substr($0, ((RSTART+RLENGTH)), 1) ~ /[_0-9a-zA-Z]/ ) {
             continue
           }
           states = "found"
           objcnt++
           print "Discovered _____  " i > lg
           print "Processing _____  \n" $0
           break
         }
       }  # for...
       if ( states == "found" )  {
         for ( j=1; j<=ac; j++ ) {
           print arr[j] > mf
           delete arr[j]  }
         ac = 0
         print $0 > mf
         if ( $0 ~ /;/ )  states = "seek"
       }
       else  {
         if ( $0 ~ /;/ )  {
           for ( j=1; j<=ac; j++ ) {
             print arr[j] > xf
             delete arr[j]  }
           ac = 0
           print $0 > xf
         }
         else  arr[++ac] = $0
       }
     }  # section...
     ac  > 100 { print "(" NR "): statement line counter over 100 !" >
"xx.out" }
     { next }
     $0 ~ /^[ ]*grant / { n = split ( $3, aa, "." )
                         tbl = aa[n]
                         if ( tbl in lst )  {
                           print $2 "(" tbl ")"
                           print "Processed _____  " tbl > lg
                           viewcnt++
                           print $0 > cf
                           print "drop " $2 "  " $3 " ;" > df
                           states = "view"
                         }
                         else { print "Skipped >>>>>>>" $3 > lg
                                print $0 > xf
                                states = "skip"    }
                         if ( $0 ~ /;/ )  states = "seek"
                         next
                       }
     END { print "\nCounted " objcnt " discoveries of listed names." > lg
           print   "GRANT   SQL redone =  " cnta  > lg
         }' \
  list ./db.excluded

Stack Trace~

~_regcomp_C(0x12000ae70, 0x140045a40, 0x140045a80, 0x14000a738, 0x14)
[0x3ff800ee79c]
  __regcomp(0x140045a80, 0x14000a738, 0x14, 0x14001db70, 0x12000ae8c)
[0x3ff800e426c]
(dbx) 

[Posted by WWW Notes gateway]

Your customer's awk program creates an array called "lst" from records in
a file called "list". I have noticed using Digital UNIX awk/nawk, that
these array variables will not always be in the same sequence as in the
file from which they came. If maintaining the sequence is important, then
you must use gawk. This problem is mentioned in Dale Dougherty's book "sed
& awk". He describes it as a known problem in awk, and one that is
"implementation dependent" in nawk.

Also, the customer uses the "lst" array and the match function in the same
code block. In the Dale Dougherty book, it seems odd that the two match function
programming examples in the chapter on nawk are actually presented using gawk.