Removing all spaces, tabs, newlines, etc from a variable?

Question

This is the error I am getting and it's failing because of a variable whose value is supposed to be 2 (I am getting this using a select * from tabel). I am getting spaces in that variable.

+ 0 != 
         2
./setjobs[19]: 0:  not found.

How do I remove all those spaces or a newline from that variable? Can tr, sed, or anything help?

This what I am doing:

set_jobs_count=$(echo  "set heading off;
      select count(*) from oppar_db
      where ( oppar_db_job_name, oppar_db_job_rec ) in ($var) ;" | \
  sqlplus -s ${OP_ORA_USER}/${OP_ORA_PASS}@$OPERATIONAL_DB_NAME)

---

This works as suggested:

| sed 's/[[:space:]]//g'

But I still obtain a value like :

  set_jobs_count=
  2

Solution 1

The reason sed 's/[[:space:]]//g' leaves a newline in the output is because the data is presented to sed a line at a time. The substitution can therefore not replace newlines in the data (they are simply not part of the data that sed sees).

Instead, you may use tr

tr -d '[:space:]'

which will remove space characters, form feeds, new-lines, carriage returns, horizontal tabs, and vertical tabs.

Solution 2

In ksh, bash or zsh:

set_jobs_count=…
set_jobs_count=${set_jobs_count//[[:space:]]/}

In any Bourne-like shell, you can remove leading and trailing whitespace and normalize all intermediate whitespace to a single space like this:

set -f
set -- $set_jobs_count
set_jobs_count="$*"
set +f

set -f turns off globbing; if you know that the data contains none of the characters \[?*, you can omit it.

That assumes $IFS contains its default value of space, tab, newline. Other whitespace characters (CR, VT, FF...) are left alone.

Solution 3

As already suggested, use "tr -d '[:space:]'", BUT if that doesn't work, possibly on an older Unix system, one needs to cover all these characters:

\011 HT    '\t' (horizontal tab)
\012 LF    '\n' (new line)
\013 VT    '\v' (vertical tab)
\014 FF    '\f' (form feed)
\015 CR    '\r' (carriage ret)
\040 SPACE ' '  (space)

with this:

tr -d '\011\012\013\014\015\040'

However, perhaps there are other non-visible characters garbling the output, The result from 'select count(*) ...' should be only an integer, so to leave only digits:

tr -dc '0-9'

However, that won't work if there are other characters after the first digits, and then more digits. So, to only return the first digit sequence found:

perl -ane '/\d+/ && print($&) && exit'

## example: will return '11', but raw output looks like '22 33'
(echo -e '\a11\b\b22 33') | perl -ane '/\d+/ && print($&) && exit'

Comments

[bsd]

You can cast a string to an int in the select statement. How that is done depends on database, Sybase, Oracle, MySQL, etc.

[Ezequiel]

how do i do that, i have oracle 9i

[bsd]

using sed it's | sed 's/[[:space:]]//g' to collapse whitespace

[Ezequiel]

thanks works up to some extent but still the values of variable comes like set_jobs_count= 2

[Edd Steel]

Depending on the shell, you may be able to do it without any external tools. See stackoverflow.com/a/3352015/587717

[Terry Wang]

In my case tr -d '[:blank:]' does a better job (than [:space:]) because it take case of tab characters as well.

[Kusalananda]

@TerryWang [:space:] matches tabs too.

[Gilles 'SO- stop being evil']

@BinaryZebra The globbing happens at set -- $set_jobs_count. set_jobs_count=$* is equivalent to set_jobs_count="$@" since $* and $@ are only equivalent when unquoted and the right-hand side of an assignment is parsed the same way as a double-quoted string.

[Kusalananda]

Do you have any idea of when the standard [:space:] character class first came into use, given that you refer to unnamed systems that may not have it? I'm assuming it must be after the introduction of Perl, as you suggest using it on these systems. I've found no reference about this and the POSIX standard does not contain historical annotations.

[Admin]

Note this approach is POSIX compliant, meaning it's very portable across shells pubs.opengroup.org/onlinepubs/009696899/utilities/tr.html