m

1 files changed, 280 insertions, 103 deletions

diff --git a/mkdeps.sh b/mkdeps.sh
index 9df3e5a..496c13e 100755
--- a/mkdeps.sh
+++ b/mkdeps.sh
@@ -4,163 +4,340 @@ 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() {
-	xargs realpath --relative-to=. | sort
+	## 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
 }
 
-pages() {
-	find src/pages/*/*.adoc -type f                          | normalize
+## 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'
+		page_translations
+	} | normalize | uniq
 }
 
-articles() {
-	find src/collections/*/*/*/ -type f -name '*.adoc'       | normalize
+## 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
 }
 
-slides() {
-	find src/slides/*/*/ -type f -name '*.adoc'              | normalize
+collections() {
+	grep -l '^:type: collection$' src/content/*/*/index.adoc |
+		normalize | only_masters | xargs -n1 dirname
 }
 
 indexes() {
-	find src/collections/*/*/index.adoc                      | normalize
+	all_collections | sed 's|$|/index.adoc|' | normalize
+}
+
+## Every collection there is, translations included: the rules
+## below emit a listing, a feed and a sortdata list for each, and
+## a translated collection needs all three as much as its master.
+all_collections() {
+	collections | sed 's|$|/index.adoc|' | with_translations |
+		sed 's|/index.adoc$||' | normalize
 }
 
 categories() {
-	grep -l '^:type: categories$' src/collections/*/*/*.adoc | normalize
+	grep -l '^:type: categories$' src/content/*/*/*.adoc |
+		normalize | only_masters | with_translations | normalize
+}
+
+articles() {
+	collections | xargs -I% 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="$(mktemp)"
+	{ homes; indexes; categories; articles; } | sort -u > "$claimed"
+	universe | grep -vxF -f "$claimed"
+	rm -f "$claimed"
 }
 
 media() {
-	find src/content/ \( -type f -and \! -type l \) -and \( \
+	# 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() {
 	media
 	tarballs
+	shortener
 }
 
-listings() {
-	indexes
-	categories
+## 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() {
+	if [ -z "${LANGS_CACHE:-}" ]; then
+		LANGS_CACHE="$(
+			find po -maxdepth 1 -name '*.po' |
+				sed 's|.*/||; s|\.po$||' | sort
+		)"
+		export LANGS_CACHE
+	fi
+	echo "$LANGS_CACHE"
 }
 
-files() {
-	pages
-	articles
-	listings
+## 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() {
+	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- "$(po4a_cache)" |
+		tr ' ' '\n' | grep . | normalize
+}
+
+## Just the pages among them.  The switcher offers a reader the
+## same page in another language; the data files under src/i18n
+## are not pages.
+page_translations() {
+	translations | grep '\.adoc$'
+}
+
+## Drops the translations from a list of paths, so a rule reads
+## the masters whether or not po4a has run.
+## po4a.cfg does not change while this runs, so parse it once:
+## every selector below asks for the same answer.
+po4a_cache() {
+	if [ -z "${PO4A_CACHE:-}" ]; then
+		PO4A_CACHE="$(mktemp)"
+		po4a_sets > "$PO4A_CACHE"
+		export PO4A_CACHE
+	fi
+	echo "$PO4A_CACHE"
+}
+
+translations_cache() {
+	if [ -z "${TRANSLATIONS_CACHE:-}" ]; then
+		TRANSLATIONS_CACHE="$(mktemp)"
+		translations > "$TRANSLATIONS_CACHE"
+		export TRANSLATIONS_CACHE
+	fi
+	echo "$TRANSLATIONS_CACHE"
+}
+
+only_masters() {
+	awk 'NR == FNR { t[$0]; next } !($0 in t)' \
+		"$(translations_cache)" -
+}
+
+## Reads masters on stdin, writes them and their translations.
+with_translations() {
+	awk -v sets="$(po4a_cache)" '
+		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[$0], g, " ")
+				for (i = 1; i <= n; i++) {
+					print g[i]
+				}
+			}
+		}
+	'
+}
+
+## Prime the caches here, in this shell.  Every pipeline stage
+## below runs in a subshell, so a cache filled inside one is gone
+## by the next: filling them once, up here, is what makes them
+## caches rather than seven identical computations.
+PO4A_CACHE="$(mktemp)"
+po4a_sets > "$PO4A_CACHE"
+TRANSLATIONS_CACHE="$(mktemp)"
+translations > "$TRANSLATIONS_CACHE"
+LANGS_CACHE="$(find po -maxdepth 1 -name '*.po' |
+	sed 's|.*/||; s|\.po$||' | sort)"
+export PO4A_CACHE TRANSLATIONS_CACHE LANGS_CACHE
+trap 'rm -f "$PO4A_CACHE" "$TRANSLATIONS_CACHE"' EXIT
 
 pages                                      | varlist      'pages.adoc'
+homes                                      | varlist      'homes.adoc'
 articles                                   | varlist   'articles.adoc'
-slides                                     | varlist     'slides.adoc'
+slides                                     | varlist 'slides.eslaides'
 categories                                 | varlist 'categories.adoc'
 indexes                                    | varlist    'indexes.adoc'
 indexes | sed 's|/index\.adoc$|/feed.xml|' | varlist 'feeds.xml'
-find src/content/img/ -name '*.svg'        | varlist 'images.svg'
+find src/content/img/ -name '*.svg' | sort | varlist 'images.svg'
 media                                      | varlist 'sources.media'
 tarballs                                   | varlist 'sources.tarballs'
 extras                                     | varlist 'sources.extras'
 find po/*.po po/*.pot                      | varlist 'sources.po'
+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.
+music | sed 's/^\(.*\)\.ly$/\1.pdf:\t\1.midi/'
 
-{
-	files   | sed 's/^\(.*\)\.adoc$/\1.html/'
-	files   | sed 's/^\(.*\)\.adoc$/\1.snippets/'
-	slides  | sed 's/^\(.*\)\.adoc$/\1.pdf/'
-	indexes | sed 's|^\(.*\)/index\.adoc$|\1/feed.xml|'
-	media   | sed 's/^\(.*\)$/\1.torrent/'
-} | sed 's/^\(.*\)$/\1.gz:\t\1/'
-printf '\n'
-
-files    | sed 's/^\(.*\)\.adoc$/\1.htmlbody\t\1.snippets\t\1.conf:\t\1.adoc/'
-files    | sed 's/^\(.*\)\.adoc$/\1.html:\t\1.conf\t\1.htmlbody/'
-printf '\n'
-
-files    | sed 's/^\(.*\)\.adoc$/\1.updatedat-check:\t\1.conf/'
-files    | sed 's/^\(.*\)\.adoc$/\1.links-internal-check:\t\1.links/'
-files    | sed 's/^\(.*\)\.adoc$/\1.caslinks:\t\1.links/'
-printf '\n'
-
-articles | sed 's/^\(.*\)\.adoc$/\1.feedentry:\t\1.conf\t\1.htmlbody/'
-articles | sed 's/^\(.*\)\.adoc$/\1.sortdata:\t\1.conf/'
-articles | sed 's/^\(.*\)\.adoc$/\1.categorydata:\t\1.conf/'
-printf '\n'
-
-listings | sed 's/^\(.*\)\.adoc$/\1.htmlheader\t\1.htmlfooter:\t\1.conf/'
-listings | sed 's/^\(.*\)\.adoc$/\1.htmllisting:\t\1.conf/'
-listings | sed 's/^\(.*\)\.adoc$/\1.html:\t\1.htmlheader\t\1.htmlfooter/'
-listings | sed 's/^\(.*\)\.adoc$/\1.html:\t\1.htmllisting\t\1.htmlbody/'
-printf '\n'
-
-media    | sed 's/^\(.*\)$/\1.torrent:\t\1/'
-printf '\n'
-
-slides   | sed 's/^\(.*\)\.adoc$/\1.ps:\t\1.adoc/'
-slides   | sed 's/^\(.*\)\.adoc$/\1.pdf:\t\1.ps/'
-printf '\n'
-
-
+## A short link points into the built tree, so on a fresh clone it
+## dangles, and make --- which stats through a symlink --- reports no
+## rule for a file that is right there.  Name the page each one points
+## at as what the link is made from.
+shortener | while read -r link; do
+	target="$(dirname "$link")/$(readlink "$link")"
+	printf '%s:\t%s\n' "$link" \
+		"$(printf '%s\n' "$target" | normalize)"
+done
 
-collectionentries() {
-	langlink="$1"
-	colllink="$2"
-	lang="$(basename "$langlink")"
-	c="$(printf '%s' "$colllink" | normalize)"
-	plural="$(cat src/names/categories/"$lang".txt)"
 
-	printf '\n\n'
-	name="$(basename "$c")"
-	art=articles."$lang"
-	find "$c"/*/ -type f -name '*.adoc' | varlist "$art.$name.adoc"
-	echo "$art.$name.sortdata     = \$($art.$name.adoc:.adoc=.sortdata)"
-	echo "$art.$name.indexentry   = \$($art.$name.adoc:.adoc=.indexentry)"
-	echo "$art.$name.feedentry    = \$($art.$name.adoc:.adoc=.feedentry)"
-	echo "$art.$name.categorydata = \$($art.$name.adoc:.adoc=.categorydata)"
+## 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.
+{
+	pages      | sed 's/^/page\t/'
+	homes      | sed 's/^/home\t/'
+	articles   | sed 's/^/article\t/'
+	indexes    | sed 's/^/index\t/'
+	categories | sed 's/^/category\t/'
+	slides     | sed 's/^/slide\t/'
+	media      | sed 's/^/media\t/'
+} | mkwb deps src/config.json
 
-	printf '%s/sortdata.txt:\tdeps.mk\n' "$c"
-	printf '\tprintf %s $(%s.%s.sortdata) > $@\n\n' "'%s\n'" "$art" "$name"
+## What po4a 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.
+translations | varlist 'translations'
 
-	listings="
-		feed.xml
-		index.htmllisting
-		$plural.htmllisting
-		$plural.txt
-	"
-	for lst in $listings; do
-		printf '%s/%s:\t%s/sortdata.txt\n'   "$c" "$lst" "$c"
-		printf '%s/%s:\t$(%s.%s.sortdata)\n' "$c" "$lst" "$art" "$name"
-	done
+## One run writes all of them, so they hang off a stamp rather
+## than each invoking po4a for itself.
+printf 'src/i18n.stamp: po/po4a.cfg %s\n' "$(find po -maxdepth 1 -name '*.po' | sort | tr '\n' ' ')"
+printf '\tpo4a po/po4a.cfg\n'
+printf '\ttouch $@\n\n'
+printf '$(translations): src/i18n.stamp\n\n'
 
-	printf      '%s/index.htmllisting\t'   "$c"
-	printf '%s/%s.htmllisting:\t'          "$c" "$plural"
-	printf '$(%s.%s.indexentry)\n'         "$art" "$name"
 
-	printf '%s/%s.txt:\t'                  "$c" "$plural"
-	printf '$(%s.%s.categorydata)\n'       "$art" "$name"
+## 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'"
+po4a_sets | grep '\.adoc' | sed "s/^/\t\t'/; s/\$/' \\\\/"
+printf '\t\t> $@\n\n'
 
-	printf '%s/%s.txt\t'                   "$c" "$plural"
-	printf '%s/feed.xml:\t'                "$c"
-	printf '$(%s.%s.feedentry)\n'          "$art" "$name"
 
-	printf '%s/%s.htmllisting\t'           "$c" "$plural"
-	printf '%s/%s.xml:\t'                  "$c" "$plural"
-	printf '%s/%s.txt\n'                   "$c" "$plural"
+## The canonical rules, inlined so no include is needed
+## at make time.
+echo
+mkwb rules
 
-	printf '%s/%s.xml.gz:\t%s/%s.xml\n' "$c" "$plural" "$c" "$plural"
-}
 
-
-for langlink in src/collections/*; do
-	for colllink in "$langlink"/*; do
-		collectionentries "$langlink" "$colllink"
-	done
-done
+## Everything mkwb renders reads the site's wording table, which
+## po4a writes.  The rules above define the variables, so this
+## comes after them.  It has to name what renders rather than
+## what a page is made of: an article never writes a .htmlheader
+## of its own, so naming that missed every one of them.
+printf '\n$(sources.html) $(sources.json): src/i18n.stamp\n'
+printf '$(listings.htmllisting) $(categories.xml): src/i18n.stamp\n'
+printf '$(articles.feedentry): src/i18n.stamp\n'