commit 5a07adc5853c08fcc4a24c00f5e7d6c65fbe318d Author: Michael Raitza Date: Wed Dec 2 18:51:52 2020 +0100 Initial commit diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..aa1395f --- /dev/null +++ b/.gitignore @@ -0,0 +1,4 @@ +result +/.redo +**/.do_built* +**/*.did diff --git a/README.org b/README.org new file mode 100644 index 0000000..b6a40a0 --- /dev/null +++ b/README.org @@ -0,0 +1,18 @@ +#+TITLE: Nix flake +#+AUTHOR: spacefrogg +#+EMAIL: git@spacefrogg.net +#+options: toc:nil + +* Development + This comes with a self-contained redo toolchain to help in working with the + flake. It basically deploys any input flake into the =tmp= directory for easier + access. + + redo targets under =tmp/= come in three kinds: + - =clean= :: will remove all build products under =tmp/= + - == :: will make sure a store path to == exists and puts an + indirect root into =tmp/=. + - =.rw= :: will make sure a store path to == exists and will + copy its contents to =tmp/..rw= and will link to it from + =tmp/.rw=. (Rationale: On Linux, directories cannot denote reliable + targets due to their non-trivial relation to modification time.) diff --git a/do b/do new file mode 100755 index 0000000..909b5c6 --- /dev/null +++ b/do @@ -0,0 +1,446 @@ +#!/usr/bin/env sh +# +# A minimal alternative to djb redo that doesn't support incremental builds. +# For the full version, visit http://github.com/apenwarr/redo +# +# The author disclaims copyright to this source file and hereby places it in +# the public domain. (2010 12 14; updated 2019 02 24) +# +USAGE=" +usage: do [-d] [-x] [-v] [-c] + -d print extra debug messages (mostly about dependency checks) + -v run .do files with 'set -v' + -x run .do files with 'set -x' + -c clean up all old targets before starting + + Note: do is an implementation of redo that does *not* check dependencies. + It will never rebuild a target it has already built, unless you use -c. +" + +# CDPATH apparently causes unexpected 'cd' output on some platforms. +unset CDPATH + +# By default, no output coloring. +green="" +bold="" +plain="" + +if [ -n "$TERM" -a "$TERM" != "dumb" ] && tty <&2 >/dev/null 2>&1; then + green="$(printf '\033[32m')" + bold="$(printf '\033[1m')" + plain="$(printf '\033[m')" +fi + +# The 'seq' command is not available on all platforms. +_seq() { + local x=0 max="$1" + while [ "$x" -lt "$max" ]; do + x=$((x + 1)) + echo "$x" + done +} + +# Split $1 into a dir part ($_dirsplit_dir) and base filename ($_dirsplit_base) +_dirsplit() { + _dirsplit_base=${1##*/} + _dirsplit_dir=${1%$_dirsplit_base} +} + +# Like /usr/bin/dirname, but avoids a fork and uses _dirsplit semantics. +qdirname() ( + _dirsplit "$1" + dir=${_dirsplit_dir%/} + echo "${dir:-.}" +) + +_dirsplit "$0" +REDO=$(cd "$(env pwd -P)" && + cd "${_dirsplit_dir:-.}" && + echo "$PWD/$_dirsplit_base") +export REDO +_cmd=$_dirsplit_base + +DO_TOP= +if [ -z "$DO_BUILT" ]; then + export _do_opt_debug= + export _do_opt_exec= + export _do_opt_verbose= + export _do_opt_clean= +fi +while getopts 'dxvcj:h?' _opt; do + case $_opt in + d) _do_opt_debug=1 ;; + x) _do_opt_exec=x ;; + v) _do_opt_verbose=v ;; + c) _do_opt_clean=1 ;; + j) ;; # silently ignore, for compat with real redo + \?|h|*) printf "%s" "$USAGE" >&2 + exit 99 + ;; + esac +done +shift "$((OPTIND - 1))" +_debug() { + [ -z "$_do_opt_debug" ] || echo "$@" >&2 +} + +if [ -z "$DO_BUILT" -a "$_cmd" != "redo-whichdo" ]; then + DO_TOP=1 + if [ "$#" -eq 0 ] && [ "$_cmd" = "do" -o "$_cmd" = "redo" ]; then + set all # only toplevel redo has a default target + fi + export DO_STARTDIR="$(env pwd -P)" + # If starting /bin/pwd != $PWD, this will fix it. + # That can happen when $PWD contains symlinks that the shell is + # trying helpfully (but unsuccessfully) to hide from the user. + cd "$DO_STARTDIR" || exit 99 + export DO_BUILT="$PWD/.do_built" + if [ -z "$_do_opt_clean" -a -e "$DO_BUILT" ]; then + echo "do: Incremental mode. Use -c for clean rebuild." >&2 + fi + : >>"$DO_BUILT" + sort -u "$DO_BUILT" >"$DO_BUILT.new" + while read f; do + [ -n "$_do_opt_clean" ] && printf "%s\0%s.did\0" "$f" "$f" + printf "%s.did.tmp\0" "$f" + done <"$DO_BUILT.new" | + xargs -0 rm -f 2>/dev/null + mv "$DO_BUILT.new" "$DO_BUILT" + export DO_PATH="$DO_BUILT.dir" + export PATH="$DO_PATH:$PATH" + rm -rf "$DO_PATH" + mkdir "$DO_PATH" + for d in redo redo-ifchange redo-whichdo; do + ln -s "$REDO" "$DO_PATH/$d" + done + for d in redo-ifcreate redo-stamp redo-always redo-ood \ + redo-targets redo-sources; do + echo "#!/bin/sh" >"$DO_PATH/$d" + chmod a+rx "$DO_PATH/$d" + done +fi + + +# Chop the "file" part off a /path/to/file pathname. +# Note that if the filename already ends in a /, we just remove the slash. +_updir() +{ + local v="${1%/*}" + [ "$v" != "$1" ] && echo "$v" + # else "empty" which means we went past the root +} + + +# Returns true if $1 starts with $2. +_startswith() +{ + [ "${1#"$2"}" != "$1" ] +} + + +# Returns true if $1 ends with $2. +_endswith() +{ + [ "${1%"$2"}" != "$1" ] +} + + +# Prints $1 if it's absolute, or $2/$1 if $1 is not absolute. +_abspath() +{ + local here="$2" there="$1" + if _startswith "$1" "/"; then + echo "$1" + else + echo "$2/$1" + fi +} + + +# Prints $1 as a path relative to $PWD (not starting with /). +# If it already doesn't start with a /, doesn't change the string. +_relpath() +{ + local here="$2" there="$1" out= hadslash= + #echo "RP start '$there' hs='$hadslash'" >&2 + _startswith "$there" "/" || { echo "$there" && return; } + [ "$there" != "/" ] && _endswith "$there" "/" && hadslash=/ + here=${here%/}/ + while [ -n "$here" ]; do + #echo "RP out='$out' here='$here' there='$there'" >&2 + [ "${here%/}" = "${there%/}" ] && there= && break; + [ "${there#$here}" != "$there" ] && break + out=../$out + _dirsplit "${here%/}" + here=$_dirsplit_dir + done + there=${there#$here} + if [ -n "$there" ]; then + echo "$out${there%/}$hadslash" + else + echo "${out%/}$hadslash" + fi +} + + +# Prints a "normalized relative" path, with ".." resolved where possible. +# For example, a/b/../c will be reduced to just a/c. +_normpath() +( + local path="$1" relto="$2" out= isabs= + #echo "NP start '$path'" >&2 + if _startswith "$path" "/"; then + isabs=1 + else + path="${relto%/}/$path" + fi + set -f + IFS=/ + for d in ${path%/}; do + #echo "NP out='$out' d='$d'" >&2 + if [ "$d" = ".." ]; then + out=$(_updir "${out%/}")/ + else + out=$out$d/ + fi + done + #echo "NP out='$out' (done)" >&2 + out=${out%/} + if [ -n "$isabs" ]; then + echo "${out:-/}" + else + _relpath "${out:-/}" "$relto" + fi +) + + +# Prints a "real" path, with all symlinks resolved where possible. +_realpath() +{ + local path="$1" relto="$2" isabs= rest= + if _startswith "$path" "/"; then + isabs=1 + else + path="${relto%/}/$path" + fi + ( + for d in $(_seq 100); do + #echo "Trying: $PWD--$path" >&2 + if cd -P "$path" 2>/dev/null; then + # success + pwd=$(env pwd -P) + #echo " chdir ok: $pwd--$rest" >&2 + np=$(_normpath "${pwd%/}/$rest" "$relto") + if [ -n "$isabs" ]; then + echo "$np" + else + _relpath "$np" "$relto" + fi + break + fi + _dirsplit "${path%/}" + path=$_dirsplit_dir + rest="$_dirsplit_base/$rest" + done + ) +} + + +# List the possible names for default*.do files in dir $1 matching the target +# pattern in $2. We stop searching when we find the first one that exists. +_find_dofiles_pwd() +{ + local dodir="$1" dofile="$2" + _startswith "$dofile" "default." || dofile=${dofile#*.} + while :; do + dofile=default.${dofile#default.*.} + echo "$dodir$dofile" + [ -e "$dodir$dofile" ] && return 0 + [ "$dofile" = default.do ] && break + done + return 1 +} + + +# List the possible names for default*.do files in $PWD matching the target +# pattern in $1. We stop searching when we find the first name that works. +# If there are no matches in $PWD, we'll search in .., and so on, to the root. +_find_dofiles() +{ + local target="$1" dodir= dofile= newdir= + _debug "find_dofile: '$PWD' '$target'" + dofile="$target.do" + echo "$dofile" + [ -e "$dofile" ] && return 0 + + # Try default.*.do files, walking up the tree + _dirsplit "$dofile" + dodir=$_dirsplit_dir + dofile=$_dirsplit_base + [ -n "$dodir" ] && dodir=${dodir%/}/ + [ -e "$dodir$dofile" ] && return 0 + for i in $(_seq 100); do + [ -n "$dodir" ] && dodir=${dodir%/}/ + #echo "_find_dofiles: '$dodir' '$dofile'" >&2 + _find_dofiles_pwd "$dodir" "$dofile" && return 0 + newdir=$(_realpath "${dodir}.." "$PWD") + [ "$newdir" = "$dodir" ] && break + dodir=$newdir + done + return 1 +} + + +# Print the last .do file returned by _find_dofiles. +# If that file exists, returns 0, else 1. +_find_dofile() +{ + local files="$(_find_dofiles "$1")" + rv=$? + #echo "files='$files'" >&2 + [ "$rv" -ne 0 ] && return $rv + echo "$files" | { + while read -r linex; do line=$linex; done + printf "%s\n" "$line" + } +} + + +# Actually run the given $dofile with the arguments in $@. +# Note: you should always run this in a subshell. +_run_dofile() +{ + export DO_DEPTH="$DO_DEPTH " + export REDO_TARGET="$PWD/$target" + local line1 + set -e + read line1 <"$PWD/$dofile" || true + cmd=${line1#"#!/"} + if [ "$cmd" != "$line1" ]; then + set -$_do_opt_verbose$_do_opt_exec + exec /$cmd "$PWD/$dofile" "$@" + else + set -$_do_opt_verbose$_do_opt_exec + # If $dofile is empty, "." might not change $? at + # all, so we clear it first with ":". + :; . "$PWD/$dofile" + fi +} + + +# Find and run the right .do file, starting in dir $1, for target $2, +# providing a temporary output file as $3. Renames the temp file to $2 when +# done. +_do() +{ + local dir="$1" target="$1$2" tmp="$1$2.redo.tmp" tdir= + local dopath= dodir= dofile= ext= + if [ "$_cmd" = "redo" ] || + ( [ ! -e "$target" -o -d "$target" ] && + [ ! -e "$target.did" ] ); then + printf '%sdo %s%s%s%s\n' \ + "$green" "$DO_DEPTH" "$bold" "$target" "$plain" >&2 + dopath=$(_find_dofile "$target") + if [ ! -e "$dopath" ]; then + echo "do: $target: no .do file ($PWD)" >&2 + return 1 + fi + _dirsplit "$dopath" + dodir=$_dirsplit_dir dofile=$_dirsplit_base + if _startswith "$dofile" "default."; then + ext=${dofile#default} + ext=${ext%.do} + else + ext= + fi + target=$PWD/$target + tmp=$PWD/$tmp + cd "$dodir" || return 99 + target=$(_relpath "$target" "$PWD") || return 98 + tmp=$(_relpath "$tmp" "$PWD") || return 97 + base=${target%$ext} + tdir=$(qdirname "$target") + [ ! -e "$DO_BUILT" ] || [ ! -w "$tdir/." ] || + : >>"$target.did.tmp" + # $qtmp is a temporary file used to capture stdout. + # Since it might be accidentally deleted as a .do file + # does its work, we create it, then open two fds to it, + # then immediately delete the name. We use one fd to + # redirect to stdout, and the other to read from after, + # because there's no way to fseek(fd, 0) in sh. + qtmp=$DO_PATH/do.$$.tmp + ( + rm -f "$qtmp" + ( _run_dofile "$target" "$base" "$tmp" >&3 3>&- 4<&- ) + rv=$? + if [ $rv != 0 ]; then + printf "do: %s%s\n" "$DO_DEPTH" \ + "$target: got exit code $rv" >&2 + rm -f "$tmp.tmp" "$tmp.tmp2" "$target.did" + return $rv + fi + echo "$PWD/$target" >>"$DO_BUILT" + if [ ! -e "$tmp" ]; then + # if $3 wasn't created, copy from stdout file + cat <&4 >$tmp + # if that's zero length too, forget it + [ -s "$tmp" ] || rm -f "$tmp" + fi + ) 3>$qtmp 4<$qtmp # can't use "|| return" here... + # ...because "|| return" would mess up "set -e" inside the () + # on some shells. Running commands in "||" context, even + # deep inside, will stop "set -e" from functioning. + rv=$? + [ "$rv" = 0 ] || return "$rv" + mv "$tmp" "$target" 2>/dev/null + [ -e "$target.did.tmp" ] && + mv "$target.did.tmp" "$target.did" || + : >>"$target.did" + else + _debug "do $DO_DEPTH$target exists." >&2 + fi +} + + +# Implementation of the "redo" command. +_redo() +{ + local i startdir="$PWD" dir base + set +e + for i in "$@"; do + i=$(_abspath "$i" "$startdir") + ( + cd "$DO_STARTDIR" || return 99 + i=$(_realpath "$(_relpath "$i" "$PWD")" "$PWD") + _dirsplit "$i" + dir=$_dirsplit_dir base=$_dirsplit_base + _do "$dir" "$base" + ) + [ "$?" = 0 ] || return 1 + done +} + + +# Implementation of the "redo-whichdo" command. +_whichdo() +{ + _find_dofiles "$1" +} + + +case $_cmd in + do|redo|redo-ifchange) _redo "$@" ;; + redo-whichdo) _whichdo "$1" ;; + do.test) ;; + *) printf "do: '%s': unexpected redo command" "$_cmd" >&2; exit 99 ;; +esac +[ "$?" = 0 ] || exit 1 + +if [ -n "$DO_TOP" ]; then + if [ -n "$_do_opt_clean" ]; then + echo "do: Removing stamp files..." >&2 + [ ! -e "$DO_BUILT" ] || + while read f; do printf "%s.did\0" "$f"; done <"$DO_BUILT" | + xargs -0 rm -f 2>/dev/null + fi +fi diff --git a/tmp/.gitignore b/tmp/.gitignore new file mode 100644 index 0000000..beff5a9 --- /dev/null +++ b/tmp/.gitignore @@ -0,0 +1,3 @@ +* +!*.do +!.gitignore diff --git a/tmp/default.do b/tmp/default.do new file mode 100644 index 0000000..afb949a --- /dev/null +++ b/tmp/default.do @@ -0,0 +1,40 @@ +#!/bin/sh +# This do script creates an indirect root to an input flake + +exec >&2 + +redo-ifchange ../flake.nix ../flake.lock || true + +resolve_flake() { + STORE_PATH=$(nix flake info --json --inputs-from .. "$2" | jq -r .path) + + # Do this after 'info' to avoid unnecessary network activity. nix flake + # archive may try to download a flake although the store path is available. + if [ ! -d "$STORE_PATH" ]; then + nix flake archive --inputs-from .. "$2" + fi +} + +case $2 in + # Resolves cleanup targets by their softlinks + clean) + redo-always || true + find . -type l -print -exec realpath -P --relative-base="$PWD" {} \+ | + while read p; do + if [ "$p" = "${p#/}" ]; then + rm -rf -- "$p" + fi + done + ;; + *.rw) + resolve_flake "$@" + rm -rf -- ".$2" + cp --reflink=auto -r --no-preserve=mode "$STORE_PATH" ".$2" + ln -sf ".$2" "$3" + ;; + *) + resolve_flake "$@" + nix-store -r "$STORE_PATH" --indirect --add-root "$2" + [ "$2" = "$3" ] || mv "$2" "$3" + ;; +esac