dirstat

#!/bin/bash

# take output of either "tar -tvf" or "find ... -ls" via stdin, summarise.

# requires GNU datamash

# stdin: as described above, or if not a pipe, will run `find -ls`
[[ -t 0 ]] && { find -ls | $0 "$@"; exit $?; }

usage() {
    cat <<'EOF'
Usage:
    dirstat -d N            # depth; default 2
    dirstat -s s|f|d        # sort by size, file count, or directory count; default "s"
EOF
    exit 1
}

[[ $1 = -h ]] && usage

# output: the output is lines like this
#   <size> <#files> <#dirs> path
# where "path" is truncated to the depth requested, and the numbers are
# summations for path and its descendents.

# Basically, think of ncdu, but with entries flattened out to a given depth
# (default 2), and show file and directory counts along with size.

# For more info, see https://github.com/sitaramc/notes/blob/master/dirstat.mkd

DEPTH=2
SORT=s
while getopts 's:d:' c
do
    case $c in
        d) DEPTH=$OPTARG ;;
        s) SORT=$OPTARG ;;
    esac
done

case $SORT in
    # convert to the field number in the eventual output
    s ) SORT=1 ;;
    f ) SORT=2 ;;
    d ) SORT=3 ;;
esac

_cut() {
    # takes the output of "find -ls", which looks like this:

    #   5636574      4 drwx------   6  sitaram  sitaram      4096 Sep 17 06:41 .
    #   5636154      4 drwx------   9  sitaram  sitaram      4096 Sep 17 06:41 ./contrib
    #   5636155      4 drwx------   2  sitaram  sitaram      4096 Sep 17 06:41 ./contrib/commands
    #   5636156      8 -rwx------   1  sitaram  sitaram      4587 Sep 17 06:41 ./contrib/commands/compile-1
    #   5636157     32 -rwx------   1  sitaram  sitaram     29575 Sep 17 06:41 ./contrib/commands/ukm

    # and converts it into (note the hard tabs)

    #   0	0	1	.
    #   0	0	1	contrib
    #   0	0	1	contrib/commands
    #   4587	1	0	contrib/commands/compile-1
    #   29575	1	0	contrib/commands/ukm

    # where column 1 is the size, column 2 is "1" for files, "0" otherwise,
    # and column 3 is the opposite ("1" for directories, "0" otherwise).

    # As a bonus, it can automagically deal with the output of "tar -tvf" too!

    perl -ne '
        chomp;
        unless ($nf) {
            # this will execute only on the first input line
            ($tf,$sf,$nf) = (2,6,10);                           # find -ls output fields
            ($tf,$sf,$nf) = (0,2,5) if /^[-d][rwxXst-]{9}/;     # tar -tvf output fields
        }
        next unless s/([-d])[rwxXst-]{9}/$1/;
        @f = split " ", $_, $nf+1;
        ($t, $s, $n) = @f[$tf, $sf, $nf];
        $n =~ s(^\./)(); $n =~ s(/$)(); $n ||= ".";
        if ($t eq "-") {
            print "$s\t1\t0\t$n\n"
        } else {
            print "0\t0\t1\t$n\n"
        }
    ' | cut -f1-$1 -d/
    # and at the end it also chops off trailing path components beyond the
    # requested depth
}

MB() {
    # the first field in the line is total bytes; convert to MB
    perl -lpe 's(^ *(\d+))($1 ? sprintf "%8.3f", $1/1_000_000 : "    -   ")e'
}

_head() {
    printf "%04s %23s %15s %7s\n" "MB" "NFiles" "NDirs" "Path"
    [ -t 1 ] && head -20
    [ -t 1 ] || cat
}

_cut $DEPTH |
    datamash --format=%12.0f -s -g 4 sum 3 sum 2 sum 1 | datamash reverse |
    sort -k$SORT,${SORT}nr |
    MB |
    _head