Re: ksh's select, almost, in portable shell

[Peter Seebach]

In message <address@hidden>, Ralf Wildenhues writes:
>> I enclose func_select below.  Weaknesses:
>> 1.  Newline after prompt.  (Solving this portably seemed like too much work.)

>Grab the settings of ECHO_C, ECHO_N, ECHO_T from Autoconf, and use them.

ECHO_T?  New on me.

I could probably do this, but wanted to keep this as small and self-contained
as I could.

>> func_select () {
>>   func_select_args=0
>>   if expr "$1" : "^[_a-zA-Z][_a-zA-Z0-9]*$" > /dev/null; then

>FWIW, expr expressions are always anchored, so there is no need for more
>anchoring (another instance below).  But can't you avoid a fork here,
>like
>  case $1 in
>    [!_a-zA-Z]* | *[!_a-zA-Z0-9]*) usage "$1";;
>  esac

Oooh, clever.

>'shift N' is not so portable.

I have been unable to find a living shell without it.

>Just drop the 1.

But yeah.

>>   for func_select_arg
>>   do
>>     func_select_args=`expr $func_select_args + 1`
>>     eval func_select_a_$func_select_args=\$func_select_arg
>>   done

>This again forks a lot.  I'd use a shell function to factor between
>$(()) and expr, so that only shells without the former use the latter
>and fork (Autotest and Libtool do this).  Likewise below.

That's a thought.

>>         if expr "$REPLY" : '^[1-9][0-9]*$' > /dev/null; then

>case $REPLY in
>  0* | *[!0-9]* ) bad ... ;;
>esac

Very clever.

I had not thought of the use of [!foo] as a replacement for a regex [foo]*.

... And I also didn't realize (though I should have) that spaces were allowed
in case patterns.  I think I was thrown off by

case a in
x y) ;;
esac

But of course, spaces around | are probably sane.  Neat!

-s