perf: Loop over the bash variable directly instead of starting subprocesses.

[JeffFaer]

Today, the loop is printing the completions one value per line and then reading one line to a variable. We can just skip the whole IO rigamarole by looping over the completions directly.

I had added this optimization to #2126 but it doesn't look like it got picked up by #1743.

[marckhouzam]

Will this cause problems if some completions have a space in them?

[JeffFaer]

It shouldn't. The quotes around the array variable should prevent word splitting like that.

$ arr=( "one" "two" "three four" "five")
$ for n in "${arr[@]}"; do
  echo "$n"
done
$ for n in ${arr[@]}; do
  echo "$n"
done

[marckhouzam]

Ok. Give me a bit of time to revive the completion tests that I haven't looked at in a while and run them.
Ref: https://github.com/marckhouzam/cobra-completion-testing

[Luap99]

I did a quick spot check with podman and it seems to work correctly even if the values have spaces in them.

[JeffFaer]

Anecdotally, this appears to shave off something like 10ms:

#!/usr/bin/env bash

arr=()
for ((i=0; i<1000; i++)); do
    arr+=($i)
done

trials=1

loop_arr_1() {
    sum=0
    for ((j=0; j<trials; j++)); do
        for i in "${arr[@]}"; do
            ((sum+=i)) || true
        done
    done
}

loop_arr_2() {
    sum=0
    for ((j=0; j<trials; j++)); do
        while IFS='' read -r i; do
            ((sum+=i)) || true
        done < <(printf '%s\n' "${arr[@]}")
    done
}

time loop_arr_1
time loop_arr_2

real	0m0.003s
user	0m0.003s
sys	0m0.000s

real	0m0.016s
user	0m0.013s
sys	0m0.004s

[marckhouzam]

The regressions tests pass.
LGTM