[SOLVED] Getting test function to work with regex in bash

Well, after having been stuck with this a couple hours now I decided to ask some help.

I'm currently reading The Linux Command Line and when I tried to do practice (chapter 28) which the book recommended I got stuck with series of errors resulting from bad syntax I can't seem to get rid of.

I have a set of small functions/programs which I'm trying to just rewrite with (worse, judging by few comments I found from web) different syntax.

Book has this:

 if [[ "$int" =~ ^-?[0-9]+$ ]];

which I'm trying to write with older version, test-function.

Book gives a hint to use grep as evaluator, but since I can't get this evaluation to work without it I haven't went that far yet.

If I want to rewrite it with test-function, do I have to use test with expr or is it just something along the lines of "test $foo = regex_i_didnt_get_right"?

Also.. Am I being stuck with trivialities here or is this relevant in the grand scheme of things (learning Linux).

The legacy test command, or its alias "[", simply does not do regex. I have no way to see the book, but what I presume the book is suggesting is to use either expr or grep instead of the test command and just let the if construct use that command's return code.

if grep -q '^-?[0-9]+$' <<<$int

if expr "$int" : '-?[0-9]+'

Note the subtle difference in the regex in the expr command. That command always treats the regex as "anchored", so the "^" and "$" anchors should not be included.

Have $int in quotes, to prevent unwanted expansions.
The ? and + repeaters are defined in ERE only, that is egrep or grep -E

if egrep -q '^-?[0-9]+$' <<< "$int"

If your shell does not have the <<< operator, the following is equivalent

if echo "$int" | egrep -q '^-?[0-9]+$'

expr does BRE like grep. A fix is to emulate the ? and + with \{m,n\}.
Further, expr is vulnerable e.g. if $int contains an expr operator like ":".
A fix is to prepend a simple character, here X, on both sides of the : operator.

if expr X"$int" : X'-\{0,1\}[0-9]\{1,\}'

The BRE after the : operator works from the beginning so the ^ can be omitted.

Thank you both for help/info. Especially that echo piping to grep looks like it could come in handy sometimes. Had a good nights sleep and was able to figure it out and write them in a way that didn't result in errors.

If you had to omit compounds when using conditionals which you think would be better to use, expr or grep? I find expr a bit easier to read in code, so I'd go for that if Id have to, but newer syntax seems way more convenient and easier to read.

Book was right, though... This was a good exercise. It made me read a whole lot more than what was written in the book itself.

I would disagree about the shell not handling regex as I have used it extensively.
I have found that if you need to do as complicated regex it is best to place it in a variable.

regex='^-?[0-9]+$'

if [[ "$int" =~ $regex ]]

You will not that the "regex" parameter is unquoted, this is important

Originally Posted by grail I would disagree about the shell not handling regex as I have used it extensively. I have found that if you need to do as complicated regex it is best to place it in a variable. regex='^-?[0-9]+$' if [[ "$int" =~ $regex ]] if test '(egrep '^-?[0-9]+$' <<<"$int")'; then You will not that the "regex" parameter is unquoted, this is important

But this is compound representation, I was asking about older version of the command which apparently doesn't do it. Would probably have to dig deeper into google search to find older documentation for that, but I'll trust information given by rcnichols that it's not supported.

Code I used to rewrite first example from the book started like this:

if test '(egrep '^-?[0-9]+$' <<<"$int")'; then

Example start was piece I mentioned at the start of thread. Suggestion to put regexes into variables is a good one though, should be common sense but I still have sometimes a habit of not using enough variables in my code. Would probably need some bigger project to really teach me the hard way...

Often a case-esac is an alternative.
But the ERE must be converted to a shell glob. After all the | is an OR, giving it a bit more capability.

case $int in
(?*[!0-9]*|[!-0-9]*|"")
  echo false
;;
(*)
  echo true
esac