Re: Numerically sorted arguments (in shell)

On 19/06/2024 at 14:11, Janis Papanagnou wrote:

On 19.06.2024 14:40, Chris Elvidge wrote:

On 18/06/2024 at 18:04, Janis Papanagnou wrote:

I've just tried a Unix tools based solution (with sed, sort, cut).
[...]
[...], and for more than two numeric
fields it can be modified to dynamically construct the sed pattern,
the sort option list, and the cut parameter, once at the beginning;
that way we could have a tool for arbitrary amounts of numeric keys in
the file name.
>
Note: this program doesn't handle pathological filenames (newlines).
>

>
If you're happy not handling pathological filenames:

 Well, typically I can indeed ignore them. But it's better of course
to avoid situations where processing is compromised by such names.
 

>
for (( i=1; i<=50; i++ )); do for (( j=2; j<=120; j+=3 )); do touch
"a${i}b${j}c.txt"; done; done
to create the files.
>
exnums() { j="$(sed 's/[^[:digit:]]\+/ /g' <<<"$@")"; printf '%s%s\n'
"$j" "$@"; }
function replaces all non-digit sequences with a space, prints digit
sequence(s) and original input.
>
for i in *; do exnums "$i"; done | sort -k1n -k2n -k3n -k4n | awk
'{print $NF}'
sort doesn't seem to care how many -k you use, fields separated with space.
awk prints the last field of the input.
>
This "seems" to work with all manner of filenames from PNN.htm (as your
original sequence) to p323dc45g12.htm, p324dc45g12.htm, p333dc45g12.htm
Seems to work in ksh, too.

 I tried the approach I outlined above... (here just echo'ing the
created parts)...
  N=${1:-1}
sed_a="[^0-9]*\([0-9]\+\)[^0-9]*"
sed_r="\1\t"
sort_a="-k1n"
for (( n=2; n<=N; n++ ))
do
     sed_a+="\([0-9]\+\)[^0-9]*"
     sed_r+="\${n}\t"
     sort_a+=" -k${n}n"
done
cut_a="-f$((N+1))-"
 echo "# The following commands would be connected by pipes:"
echo "sed 's/${sed_a}/${sed_r}&/'"
echo "sort -t$'\t' ${sort_a}"
echo "cut ${cut_a}"
  Janis
 

Your way is still restricted to filenames with a known number of sets of digits, though (AFAICS). I.e. you pass N rather than finding it.
But it takes a long time to do it my way, a call to sed for each filename, so I tried to cut down the time taken to do this and came up with:
bash: exnums() { shopt -s extglob; j="${@//+([^[:digit:]])/ }"; printf '%s%s\n' "$j" "$@"; }
ksh: exnums() { j="${@//+([^[:digit:]])/ }"; printf '%s%s\n' "$j" "$@"; }
ksh seems to do the extglob needed for bash natively.
removing the sed calls from exnum changes the time taken from 37 secs to under 1 sec with 2000+ files
ksh is faster than bash, ksh 50% of the bash time taken.
Substituting a tab for the replacement space in j= and -t$'\t' in sort would seem to allow spaces in filenames, too, as you originally had it.
--
Chris Elvidge, England
SUBSTITUTE TEACHERS ARE NOT SCABS