Better method to get a side-value from awk

Question

I have a script whose main purpose is to gather some information and output it in a table. The primary part is an awk script:

awk '
    { 
      # do some stuff, including calculating dwt
      printf(format, a, b, c, d)
    }
    END {
      # pass on dwt
    }
' inputfile

The main purpose of the awk is to create and show the table. But there is a side value dwt that it also calculates that I need elsewhere in the main script, and I am trying to figure out the best way to pass it out without disrupting the table output.

There are two ways I know I can do this:

1. Save the value to a temporary file: END { print dwt > "tempfile" } then read it outside read dwt <tempfile; rm -f tempfile. But even with more care taken to avoid clobbering existing files than shown here, I prefer to avoid this - if nothing else, I'd rather not leave temp files lying around just because a job got interrupted at the wrong time.
2. Send the value to stdout as well, but flagged. Pipe stdout into a following routine that catches and directs the flagged output appropriately, but sends the rest on:

    awk '
       ...
        END { 
           print "dwt:" dwt 
        }
     ' inputfile | while read line; do
        if [[ $line = dwt:* ]]; then
           dwt="${line#dwt:}"
        else
           echo "$line"
        fi
     done

But that seems contrived and inelegant.

I am wondering if anyone knows a better method. I've experimented with using a different file descriptor, but so far have not managed to get that to work. I have not figured out how to get the information out of the file descriptor and into the dwt environment variable without disrupting stdout as well.

Answer 1

Here's one technique:

• print dwt on stdout in the END
• capture the awk output into an array
• extract the last element of the array into a variable in the shell process
• print the rest of the array

$ seq 5 > inputfile
$ readarray -t output < <(
    awk '
        { print "table", $0; dwt += $1 }
        END {print dwt}
    ' inputfile
)
$ dwt=${output[-1]}
$ echo "dwt = $dwt"
dwt = 15
$ unset output[-1]
$ printf "%s\n" "${output[@]}"
table 1
table 2
table 3
table 4
table 5

---

OK, ksh without readarray: your shell script can look like:

awk '
    { 
      # do some stuff, including calculating dwt
      printf(format, a, b, c, d)
    }
    END {
      # pass on dwt
      print dwt
    }
' inputfile  |&
# ...........^^

typeset -a output
while IFS= read -r -p line; do output+=( "$line" ); done
# .................^^

dwt=${output[-1]}
unset output[-1]
printf "%s\n" "${output[@]}"

# do stuff with $dwt ...

From my ksh93 man page:

The symbol |& causes asynchronous execution of the preceding pipeline with a two-way pipe established to the parent shell; the standard input and output of the spawned pipeline can be written to and read from by the parent shell [...] by using -p option of the built-in commands read and print described later.

Answer 2

A bit clumsy but you could print shell commands in awk an evaluate them in in ksh:

eval $( echo -e 1\\nA\\n2\\n3\\n4 |
  awk  'NR == 1 {printf"%s%s%s;","export var1=",$0, "\n"}
        NR > 2 {printf"%s%s%s;","echo -e '", $0 , "'\n" }'
)

Output:

2
3
4

$ echo $var1
1

Although a line-by-line shell evaluation might not be the simplest and in case of some strings, the echo might give funny results. Be aware of this danger!, Hard quotes should avoid this, though:

eval $( echo -e 1\\nA\\n2\\n3\\n4 |
  awk  'NR == 1 {printf"%s%s%s;","export var1=",$0, "\n"}
        NR > 2 {printf"%s%s%s;","echo -e \x27", $0 , "\x27\n" }'
)