Using time on bash functions (not commands)

Question

How can one measure individual calls to bash functions from inside the bash file.

I have a program that I call using the command

eclipse -b col_solve.pl -e "myPred"

This call outputs some information, the last of which is SUCCESS or FAIL. I am writing a script that is called on a bunch of files in a directory, and for each of these files, outputs

• The name
• The status (SUCCESS or FAIL)
• and the (user) time it took to execute .

This is the code that I know works:

I use this to get the status (retrieving the last word in the output):

stat=
get_stat ( ){
    stat=$(awk '{print $NF}' <<< $1);
}

I use this to call the program :

run_eclipse_on ( ){
    get_stat "$(eclipse -b col_solve.pl -e "run(\"$1\")" )";
}

The problematic code is the following:

for i in `ls $1` ;  #where $1 is the directory containing the files
do
    tps=$(/usr/bin/time -f %U      \ #to get just the user time
         [run_eclipse_on $1/$i] ); # HERE it is! 
    echo $i::$stat::::$tps;  # gives, for ex: file_name::SUCCESS::::0.20
done

The culprit line is the one where the function is called. I tried surrounding it with `, {, [, $(, ' and ". Nothing worked...

Is it even possible...?

Answer 1

Use the time keyword instead of the external command. Using the keyword allows you to run time on any shell command, including function calls, not just on running a program. You can control the output format to some extent through the TIMEFORMAT variable.

TIMEFORMAT=%2U
time run_eclipse_on …
echo "$i::$stat"

The time output gets printed on its own line, though. Bash allows a trick: you can change TIMEFORMAT during the command, so you can stuff more things in there.

time { run_eclipse_on …; TIMEFORMAT="${i//%/%%}::${stat//%/%%}::%2U"; }

The output from time is printed to standard error. If you need it on standard output, just redirect with 2>&1. That will also redirect whatever the command printed on stderr, however. To preserve stderr, you can do some file descriptor shuffling.

{ time { {
      run_eclipse_on …;
      TIMEFORMAT=$stat::%2U;
    } 2>&3; } 2>&1; } 3>&2

Answer 2

It sounds like you want something like this:

#!/bin/bash

for f in "$1"/*; do
  time eclipse -b col_solve.pl -e "$f" | tail -n 1
done

Don't use functions in shell scripts unless you actually need them. The benefit of using shell scripting at all is that you can easily orchestrate other tools. Let the tools do the work, not the shell. Just use the shell to string together the other tools.

Further reading:

• In Bash, when to alias, when to script, and when to write a function?

Answer 3

While not the same as "user time", if elapsed time will suffice, another option is to save the start and end times and compute the time elapsed in your function call. The timer function listed here makes this easier.

Reproducing the function here for easier reference:

# Elapsed time.  Usage:
#
#   t=$(timer)
#   ... # do something
#   printf 'Elapsed time: %s\n' $(timer $t)
#      ===> Elapsed time: 0:01:12
#
#
#####################################################################
# If called with no arguments a new timer is returned.
# If called with arguments the first is used as a timer
# value and the elapsed time is returned in the form HH:MM:SS.
#
function timer()
{
    if [[ $# -eq 0 ]]; then
        echo $(date '+%s')
    else
        local  stime=$1
        etime=$(date '+%s')
        if [[ -z "$stime" ]]; then stime=$etime; fi
        dt=$((etime - stime))
        ds=$((dt % 60))
        dm=$(((dt / 60) % 60))
        dh=$((dt / 3600))
        printf '%d:%02d:%02d' $dh $dm $ds
    fi
}

In your case, you would include this function and then:

for i in *
do
    t=$(timer)
    run_eclipse_on $i
    elapsed=$(timer $t)
    echo $i::$stat::::$elapsed  # gives, for ex: file_name::SUCCESS::::0:00:03
done