Home / ubuntu / Questions / 29412
Accepted
moata_u
Asked: 2011-03-08 08:15:17 +0800 CST

How to get nicer error-messages in this bash-script?

  • 772

I'm trying to catch any error when run a command in order to write a log-file / report

I've tried this code:

    function valid (){

    if [ $? -eq 0 ]; then
     echo "$var1" ": status : OK" 
     else echo "$var1" ": status : ERROR"   
    fi
   }

function save(){

 sed -i "/:@/c connection.url=jdbc:oracle:thin:@$ip:1521:$dataBase" $search
 var1="adding database ip"
 valid $var1

 sed -i "/connection.username/c connection.username=$name" #$search
 var1="addning database SID"
 valid $var1
}
save

The output looks like this:

adding database ip : status : OK
sed: no input file

But I want it to look like this: 

adding database ip : status : OK
sed: no input file : status : ERROR"

or this:

adding database ip : status : OK
addning database SID : status : ERROR"

I've been trying, but it's not working with me. :(

command-line

1 Answers

  1. Best Answer

    For the former:

    # run sed, capture any output
var1="$(sed -i "/connection.username/c connection.username=$name" $search)"

    For the latter:

    # run sed, discard all output but keep exit status
sed -i "/connection.username/c connection.username=$name" $search >/dev/null 2>&1

    That said, valid() is... strange, to say the least. I would write it as

    # valid label [status]
# check status ($? if not provided) and log success/failure
function valid {
  if [[ ${2-$?} == 0 ]]; then
    echo "$1 : status : OK"
  else
    echo "$1 : status : ERROR"
  fi
}

    Actually, I would do it a bit differently from the start:

    # log label command [arguments...]
# run command, output success/failure to stderr.  If label is empty, use the
# command name:  'log "" echo hi' uses 'echo' as the label.
# log entries look like
#    label1 : status : OK
#    label2 : status : ERROR
#    Error output from foo:
#      Baz is broken; fix and try again.
log() {
  # save off label
  local label="${1:-$2}"
  shift # this removes $1 and shifts $2... to $1... so "$@" works later
  # run command, capture output
  # $(command) runs 'command' and substitutes its output
  # "$@" preserves quoting; $* would turn argument "foo bar" into two
  # arguments foo bar
  err="$("$@")"
  if [[ $? == 0 ]]; then
    status=OK
  else
    status=ERROR
  fi
  echo "$label : status : $status" >&2 # standard error
  if [[ $status == ERROR ]]; then
    # log output from command, indented for clarity
    echo "Error output from $2:"
    echo "$err" | sed 's/^/  /' >&2
  fi
}

save() {
  # this sed command is pedantically correct; if GNU sed lets you
  # do single-line 'c' commands as in the original, then go for it.
  # The backslash-return after the label argument is just for clarity;
  # 'log "foo" bar' works just as well.
  log "adding database ip" \
    sed -i "/:@/c\\
connection.url=jdbc:oracle:thin:@$ip:1521:$dataBase\\
." "$search"
  # using the original (GNU?) sed syntax
  log "adding database SID" \
    sed -i "/connection.username/c connection.username=$name" "$search"
}

save

    I would also include a timestamp and program ID, etc. in a real program.

    You should probably explore the Advanced Bash Scripting Guide to learn more about writing shell scripts. The UNIX Programming Environment's shell programming chapters don't cover bash extensions to the original Bourne shell, but are still useful for learning the "zen" of shell scripting.

    • 4