#!/bin/sh
set -eu

export LANG=POSIX.UTF-8


## -m so a path need not exist: po4a's outputs are named here
## before po4a has written them --- naming them is what the rules
## being generated are for --- and realpath without it fails on
## the first one, taking the whole script with it under set -e.
## For a path that does exist the answer is the same either way.
normalize() {
	## realpath(1) did this, but it is a page of string work and
	## this runs it dozens of times: doing it here saves a
	## process and most of the time.
	awk '
		{
			n = split($0, c, "/")
			top = 0
			for (i = 1; i <= n; i++) {
				if (c[i] == "" || c[i] == ".") {
					continue
				}
				if (c[i] == ".." && top > 0 &&
				    out[top] != "..") {
					top--
					continue
				}
				out[++top] = c[i]
			}
			p = ""
			for (i = 1; i <= top; i++) {
				p = p (i > 1 ? "/" : "") out[i]
			}
			if (p != "") {
				print p
			}
		}
	' | sort
}

## Every selector below is asked for more than once --- "pages"
## alone is built on four of the others, and the tagged list at
## the end asks for all seven again --- over a tree of a couple of
## thousand files.  So each is computed once, by the loop at the
## end of this section, into a file named after it; everything
## else reads it back with memo().  The answer has to live in a
## file rather than a variable: every stage of a pipeline runs in
## a subshell, and what one sets there is gone by the next.
CACHEDIR="$(mktemp -d)"
trap 'rm -rf "$CACHEDIR"' EXIT

memo() {
	cat "$CACHEDIR/$1"
}

## Every page there is: what the repository carries, plus what
## po4a writes from it.  A translation need not exist yet -- on a
## fresh clone none of them do -- so it is named, not found.
universe() {
	{
		find src/content -type f -name '*.adoc'
		memo page_translations
	} | normalize | uniq
}

## The classifying rules read the masters, which the repository
## always carries; a translation is whatever its master was.
homes() {
	find src/content/*/index.adoc -type f | normalize |
		only_masters | with_translations | normalize
}

collections() {
	## dirname(1) did the last step, one process per collection
	## and once for every time this was asked; every path here has
	## a trailing component to drop, so sed says it in one.
	grep -l '^:type: collection$' src/content/*/*/index.adoc |
		normalize | only_masters | sed 's|/[^/]*$||'
}

indexes() {
	memo all_collections | sed 's|$|/index.adoc|' | normalize
}

## Every collection there is, translations included: the rules
## below emit a listing, a feed and a sortdata list for each, and
## a translated collection needs all three as much as its master.
all_collections() {
	memo collections | sed 's|$|/index.adoc|' | with_translations |
		sed 's|/index.adoc$||' | normalize
}

categories() {
	grep -l '^:type: categories$' src/content/*/*/*.adoc |
		normalize | only_masters | with_translations | normalize
}

articles() {
	## One find over every collection at once: with "xargs -I%"
	## it was one find per collection, for the same answer.
	# shellcheck disable=SC2046 # a collection path holds no blanks
	set -- $(memo collections)
	[ $# -gt 0 ] || return 0
	find "$@" -mindepth 3 -name '*.adoc' |
		normalize | only_masters | with_translations | normalize
}

slides() {
	find src/content -type f -name '*.eslaides'              | normalize
}

## Whatever no other rule claimed, at any depth.
pages() {
	claimed="$CACHEDIR/claimed"
	sort -u "$CACHEDIR/homes" "$CACHEDIR/indexes" \
		"$CACHEDIR/categories" "$CACHEDIR/articles" > "$claimed"
	memo universe | grep -vxF -f "$claimed"
}

media() {
	# src/content/music is excluded: those .ogg files are score
	# renderings, not published media, and the old site never
	# offered them over BitTorrent.
	find src/content/ -path 'src/content/music' -prune -or \
		\( \( -type f -and \! -type l \) -and \( \
			-name '*.flac' -or \
			-name '*.ogg'  -or \
			-name '*.webm'     \
		\) -print \) | normalize
}

## A glob that matches nothing makes find exit non-zero, which
## under set -e would end the script.  There is no music yet.
music() {
	find src/content/music/*.ly 2>/dev/null | normalize || true
}

tarballs() {
	find src/content/ \( -type f -and \! -type l \) -and \( \
			-name '*.tar.gz'   \
		\) | normalize
}

## Short paths that stand in for a page's own, kept as symlinks
## beside it.  Guarded like music() above: a site need not have any,
## and find exits non-zero on a directory that is not there.
shortener() {
	find src/content/l -type l 2>/dev/null | normalize || true
}

extras() {
	memo media
	memo tarballs
	memo shortener
}

## Which pages are one another's translations.  po4a already says
## so, set by set, so read it there rather than keep a second list
## that can quietly disagree with the first.
## The languages po4a translates into: whichever .po files exist.
langs() {
	find po -maxdepth 1 -name '*.po' |
		sed 's|.*/||; s|\.po$||' | sort
}

## One stanza per line, master first, then the files po4a writes
## from it.  "$lang:" stands for every language, and "add_$lang:"
## names an addendum rather than an output.
po4a_sets() {
	memo langs | tr '\n' ' ' | {
		## read hits EOF on the last field and reports it, which
		## under set -e would end the block before awk ran.
		read -r all || true
		awk -v langs="$all" '
			/^\[type:/ { set = "" }
			/^\[type:/ || cont {
				line = $0
				sub(/^\[type:[^]]*\][ \t]*/, "", line)
				cont = (line ~ /\\$/)
				sub(/[ \t]*\\$/, "", line)
				n = split(line, f, /[ \t]+/)
				nl = split(langs, L, " ")
				for (i = 1; i <= n; i++) {
					if (f[i] == "" || f[i] ~ /^add_/) continue
					if (f[i] ~ /^\$lang:/) {
						sub(/^\$lang:/, "", f[i])
						for (j = 1; j <= nl; j++) {
							t = f[i]
							gsub(/\$lang/, L[j], t)
							set = set " " t
						}
						continue
					}
					sub(/^[a-z]+:/, "", f[i])
					set = set (set == "" ? "" : " ") f[i]
				}
				if (!cont && set != "") { print set; set = "" }
			}
		' po/po4a.cfg
	}
}

## Everything po4a writes: every field of a stanza but the first.
translations() {
	cut -d' ' -f2- "$CACHEDIR/po4a_sets" |
		tr ' ' '\n' | grep . | normalize
}

## Just the pages among them.  The switcher offers a reader the
## same page in another language; the data files under src/i18n
## are not pages.
page_translations() {
	memo translations | grep '\.adoc$'
}

## Drops the translations from a list of paths, so a rule reads
## the masters whether or not po4a has run.
only_masters() {
	awk 'NR == FNR { t[$0]; next } !($0 in t)' \
		"$CACHEDIR/translations" -
}

## Reads masters on stdin, writes them and their translations.
with_translations() {
	awk -v sets="$CACHEDIR/po4a_sets" '
		BEGIN {
			while ((getline line < sets) > 0) {
				n = split(line, f, /[ \t]+/)
				for (i = 2; i <= n; i++) {
					t[f[1]] = t[f[1]] " " f[i]
				}
			}
		}
		{
			print
			if ($0 in t) {
				n = split(t[$0], g, " ")
				for (i = 1; i <= n; i++) {
					print g[i]
				}
			}
		}
	'
}

## Computed here, in this shell, one selector after another.
## Every stage of a pipeline runs at the same time as the stages
## beside it, and two stages of one selector can ask for the same
## answer --- only_masters and with_translations both read what
## po4a writes --- so an answer computed where it is first needed
## could be read while it was still half-written.  The order is
## the order they are built on one another; put one too early and
## it reads a file that is not there, and says so.
for sel in langs po4a_sets translations page_translations universe \
	collections all_collections indexes homes categories articles \
	slides media tarballs shortener music pages; do
	"$sel" > "$CACHEDIR/$sel"
done

memo pages                                      | varlist      'pages.adoc'
memo homes                                 | varlist      'homes.adoc'
memo articles                              | varlist   'articles.adoc'
memo slides                                | varlist 'slides.eslaides'
memo categories                            | varlist 'categories.adoc'
memo indexes                               | varlist    'indexes.adoc'
memo indexes | sed 's|/index\.adoc$|/feed.xml|' | varlist 'feeds.xml'
find src/content/img/ -name '*.svg' | sort | varlist 'images.svg'
memo media                                 | varlist 'sources.media'
memo tarballs                              | varlist 'sources.tarballs'
extras                                     | varlist 'sources.extras'
find po/*.po po/*.pot                      | varlist 'sources.po'
memo music                                 | varlist 'music.ly'
find src/content/music/*.ogg 2>/dev/null   | varlist 'music.ogg'
find src/content/music/*.ly.include 2>/dev/null | \
	varlist 'music.include'

# lilypond emits the score alongside the MIDI, from one run.
memo music | sed 's/^\(.*\)\.ly$/\1.pdf:\t\1.midi/'

## A short link points into the built tree, so on a fresh clone it
## dangles, and make --- which stats through a symlink --- reports no
## rule for a file that is right there.  Name the page each one points
## at as what the link is made from.
memo shortener | while read -r link; do
	target="$(dirname "$link")/$(readlink "$link")"
	printf '%s:\t%s\n' "$link" \
		"$(printf '%s\n' "$target" | normalize)"
done


## What this site has, for mkwb to say what each artifact is
## built from.  Finding them is the site's half: only it knows
## where its pages live, or that it keeps decks and no music.
{
	memo pages | sed 's/^/page\t/'
	memo homes | sed 's/^/home\t/'
	memo articles | sed 's/^/article\t/'
	memo indexes | sed 's/^/index\t/'
	memo categories | sed 's/^/category\t/'
	memo slides | sed 's/^/slide\t/'
	memo media | sed 's/^/media\t/'
} | mkwbg deps src/config.json

## What the translator writes.  Naming them here is what lets them
## be derived: deps.mk is written before make runs, so a translation
## that is not named cannot be built, and would have to be
## committed instead.
memo translations | varlist 'translations'

## One run writes all of them, so they hang off a stamp rather
## than each invoking the translator for itself.  --no-update
## because a build writes translations and never catalogues; "make
## i18n" is what brings those up to the masters.
printf 'src/i18n.sentinel: po/po4a.cfg %s\n' "$(find po -maxdepth 1 -name '*.po' | sort | tr '\n' ' ')"
printf '\tmaritacag --no-update po/po4a.cfg\n'
printf '\ttouch $@\n\n'
printf '$(translations): src/i18n.sentinel\n\n'


## The map mkwb conf reads to link a page to its translations,
## written out here so it cannot drift from po4a.cfg.
printf '\ntranslations.txt = src/translations.txt\n\n'
printf 'src/translations.txt: deps.mk\n'
printf '\tprintf %s \\\n' "'%s\\n'"
memo po4a_sets | grep '\.adoc' | sed "s/^/\t\t'/; s/\$/' \\\\/"
printf '\t\t> $@\n\n'


## The canonical rules, inlined so no include is needed
## at make time.
echo
mkwbg rules


## Everything mkwb renders reads the site's wording table, which
## po4a writes.  The rules above define the variables, so this
## comes after them.  It has to name what renders rather than
## what a page is made of: an article never writes a .htmlheader
## of its own, so naming that missed every one of them.
printf '\n$(sources.html) $(sources.json): src/i18n.sentinel\n'
printf '$(listings.htmllisting) $(categories.xml): src/i18n.sentinel\n'
printf '$(articles.feedentry): src/i18n.sentinel\n'