difmap: diff files under a Unix filter, like difmap -c 'jq .' f1 f2

difmap.sh

#!/bin/sh
set -eu

usage() {
    cat <<'EOF'
difmap: diff two files after mapping them through a shell filter

Examples:

    # How do the plaintexts of these encrypted files differ?
    difmap -c 'openssl enc -d -pbkdf2 -aes-256-ctr -pass env:KEY' a.enc b.enc

    # How has this minified JSON changed since my last commit?
    difmap -c 'jq -S .' -g @ path/to/file.json

    # How does this string change if I remove all the diacritics?
    printf 'r\xc3\xa9sum\xc3\xa9\n' | difmap -c cat -c 'iconv -t ascii//TRANSLIT' -

    # What string literals do these two binary files have in common?
    difmap -d 'comm -12' -c 'strings -n8 | sort -u' /usr/bin/vi /usr/bin/emacs

Usage:

    Diff two files:
        difmap <old-file> <new-file> [<flags>...]

    Diff a file between a Git tree(-ish) and the working tree:
        difmap -g <git-ref> <file> [<flags>...]

    Diff a file between two Git trees:
        difmap -g <old-git-ref> -g <new-git-ref> <file> [<flags>...]

    Diff two files at a Git tree:
        difmap -g <git-ref> <old-file> <new-file> [<flags>...]

    Diff two files at different Git trees:
        difmap -g <old-git-ref> <old-file> -g <new-git-ref> <new-file> [<flags>...]

    Diff a file against itself, under different filters:
        difmap <file> -c <cmd1> -c <cmd2> [<flags>...]

    Diff stdin against itself (read only once), under different filters:
        difmap -c <cmd1> -c <cmd2> - [<flags>...]

Flags:

    -c CMD  apply `sh -c "$CMD"` to each file before diffing
    -d CMD  use $CMD as diff program (e.g.: "nvim -d")
    --      further flags passed to diff program
EOF
}

die_usage() {
    printf >&2 'difmap: fatal: %s\n' "$@"
    printf '\n'
    usage >&2
    exit 1
}

parse_args() {
    diff=
    gitold=
    gitnew=
    cmd1=
    cmd2=
    fold=
    fnew=
    argn=$#
    while [ $# -gt 0 ]; do
        arg="$1"
        shift
        case "$arg" in
            -g)
                if [ $# = 0 ]; then die_usage '-g: missing argument'; fi
                ref="$(git rev-parse --short=12 --verify "$1")" && shift
                if [ -z "$ref" ]; then die_usage '-g: empty argument'; fi
                if [ -z "$gitold" ]; then gitold="$ref"
                elif [ -z "$gitnew" ]; then gitnew="$ref"
                else die_usage '-g: too many'; fi
                ;;
            -c)
                if [ $# = 0 ]; then die_usage '-c: missing argument'; fi
                cmd="$1" && shift
                if [ -z "$cmd" ]; then die_usage '-c: empty argument'; fi
                if [ -z "$cmd1" ]; then cmd1="$cmd"
                elif [ -z "$cmd2" ]; then cmd2="$cmd"
                else die_usage '-c: too many'; fi
                ;;
            -d)
                if [ $# = 0 ]; then die_usage '-d: missing argument'; fi
                if [ -z "$1" ]; then die_usage '-d: empty argument'; fi
                diff="$1" && shift
                ;;
            --)
                break
                ;;
            -?*)
                die_usage "${arg}: unknown flag"
                ;;
            *)
                if [ -z "$arg" ]; then die_usage '<file>: empty argument'; fi
                if [ -z "$fold" ]; then fold="$arg"
                elif [ -z "$fnew" ]; then fnew="$arg"
                else die_usage '<file>: too many'; fi
                ;;
        esac
    done
    if [ -z "$fold" ]; then die_usage 'too few sources'; fi
    if [ -z "$fnew" ] && [ -z "$gitold" ]; then fnew="$fold"; fi
    if [ -z "$cmd1" ]; then cmd1=cat; fi
    if [ -z "$cmd2" ]; then cmd2="$cmd1"; fi
    nshift=$(( argn - $# ))
}

main() {
    parse_args "$@"
    shift "$nshift"
    tmpdir=
    trap cleanup EXIT
    tmpdir="$(mktemp -d)"
    forcelabelold=
    forcelabelnew=
    if [ "$fold" = - ] || [ "$fnew" = - ]; then
        cat >"${tmpdir}/stdin"
        if [ "$fold" = - ]; then forcelabelold="$fold" && fold="${tmpdir}/stdin"; fi
        if [ "$fnew" = - ]; then forcelabelnew="$fnew" && fnew="${tmpdir}/stdin"; fi
    elif [ "$fold" = "$fnew" ] && [ -p "$fold" ]; then
        cat "$fold" >"${tmpdir}/pipe"
        forcelabelold="$fold"
        forcelabelnew="$fnew"
        fold="${tmpdir}/pipe"
        fnew="${tmpdir}/pipe"
    fi
    readold >"${tmpdir}/old"
    readnew >"${tmpdir}/new"
    if [ -n "$forcelabelold" ]; then labelold="$forcelabelold"; fi
    if [ -n "$forcelabelnew" ]; then labelnew="$forcelabelnew"; fi
    if [ "$cmd1" != cat ]; then labelold="$labelold | $cmd1"; fi
    if [ "$cmd2" != cat ]; then labelnew="$labelnew | $cmd2"; fi
    exec ${diff:-diff -u --color=auto --label "${labelold}" --label "${labelnew}"} \
        "${tmpdir}/old" "${tmpdir}/new" "$@"
}

filter() {
    sh -c "$1"
}

readold() {
    labelold=  # output variable
    if [ -n "$gitold" ]; then
        labelold="${gitold}:${fold}"
        git show "${gitold}:${fold}" | sh -c "$cmd1"
    else
        labelold="${fold}"
        cat -- "$fold" | sh -c "$cmd1"
    fi
}

readnew() {
    labelnew=  # output variable
    if [ -n "$gitnew" ]; then
        labelnew="${gitnew}:${fnew:-${fold}}"
        git show "${gitnew}:${fnew:-${fold}}" | sh -c "$cmd2"
    elif [ -n "$gitold" ]; then
        if [ -n "$fnew" ]; then
            labelnew="${gitold}:${fnew}"
            git show "${gitold}:${fnew}" | sh -c "$cmd2"
        else
            labelnew="${fold}"
            cat -- "$fold" | sh -c  # diff Git ref against working tree
        fi
    else
        labelnew="${fnew:-${fold}}"
        cat -- "${fnew:-${fold}}" | sh -c "$cmd2"
    fi
}

cleanup() {
    if [ -n "${tmpdir}" ]; then
        rm -f "${tmpdir}"/*
        rmdir "${tmpdir}"
    fi
}

main "$@"