chore: merge round 5 search fixes

website/astro.config.mjs

@@ -108,26 +108,37 @@ export default defineConfig({
 			// ten times its text, so "dolphin" finds Dolphin, and the file names of
 			// a system page (its `files` field, written by sitestarlight.py) count
 			// half, so "scph5501" finds the Sony page without a file name winning
-			// every other query. Two fields name the page's subject in other words
-			// and weigh half a title: `aliases`, the ids setups file a system under
-			// (psx, dreamcast, gba), and `devices`, the handhelds and boxes a
-			// setup's own site names (R36S, Miyoo Mini). `consoles`, the consoles a
-			// manufacturer page holds (Neo Geo, WonderSwan Color), weighs as much
-			// as a title. `terms`, the words of a task or a message on screen (bios
-			// not found, which mame version, check my dump), weighs three times
-			// that: a field match scores below a title match, and the cards of a
-			// part page keep their own weight, so below 30 the MAME parts, whose
-			// card titles name MAME files, came before the version table for
-			// "which mame version". All of them are
-			// attributes on an empty element, so no text of theirs reaches an
-			// excerpt.
+			// every other query. `devices`, the handhelds and boxes a setup's own
+			// site names (R36S, Miyoo Mini), weighs half a title. `aliases`, the ids
+			// setups file a system under (psx, dreamcast, gba), weighs a title: with
+			// file names written as their parts, half a title put GBA.emu and two
+			// more emulators above the Nintendo page for "gba bios". `terms`, the
+			// words of a task or a message on screen (bios not found, which mame
+			// version, check my dump), weighs three titles: a field match scores
+			// below a title match, and the cards of a part page keep their own
+			// weight, so below 30 the MAME parts, whose card titles name MAME files,
+			// came before the version table for "which mame version".
+			//
+			// A field adds the same score to every page whose field holds the
+			// word, whatever else the field holds, and a query word also matches
+			// the start of a longer one. `consoles`, the consoles a system page
+			// answers for, weighs ten titles: below that an emulator named after
+			// its console (Saturn.emu, BK, PIECE) came before the console. The
+			// PlayStation 2 page holds every word of "playstation bios", so
+			// `exact` repeats, on its own page, a console whose every word another
+			// page's consoles hold (PlayStation, Nintendo DS, Nintendo Super
+			// System). It weighs two titles: enough over the other page, and less
+			// than the "3" of "playstation 3 bios" adds to the Sony page. Measured
+			// on the "<console> bios" query of every system page and on
+			// tools/search_gate.py. All of them are attributes on an empty
+			// element, so no text of theirs reaches an excerpt.
 			pagefind: {
 				ranking: {
 					pageLength: 0.1,
 					termFrequency: 0.5,
 					termSaturation: 1.4,
 					termSimilarity: 9,
-					metaWeights: { title: 10, files: 0.5, aliases: 5, devices: 5, consoles: 10, terms: 30 },
+					metaWeights: { title: 10, files: 0.5, aliases: 10, devices: 5, consoles: 100, exact: 20, terms: 30 },
 				},
 			},
 			plugins: [

website/src/routeData.ts

@@ -1,7 +1,8 @@
 // What each page knows of its place in the site, computed once per page
 // from what scripts/sitestarlight.py writes beside the sidebar:
 // - breadcrumbs: the tab and the groups above the page, each linked to the
-//   page or the section that introduces it, never to a sibling page;
+//   page or the section that introduces it, never to a sibling page. A
+//   group of names by their first letter (A, B) is no crumb;
 // - previous and next: the reading order of `reading`. Start here runs on
 //   into Your setup, every other tab is read on its own, and a system,
 //   emulator or platform page has none;
@@ -19,7 +20,7 @@ import sidebar from './generated/sidebar.json' with { type: 'json' };
 
 type Entry = StarlightRouteData['sidebar'][number];
 type Link = Extract<Entry, { type: 'link' }>;
-type Group = { href: string | null; prune: boolean };
+type Group = { href: string | null; prune: boolean; crumb?: boolean };
 export type Crumb = { label: string; href?: string };
 
 const BASE = import.meta.env.BASE_URL.replace(/\/?$/, '/');
@@ -55,7 +56,11 @@ const trail = (entries: Entry[], labels: string[]): Crumb[] | undefined => {
 		const here = [...labels, entry.label];
 		const below = trail(entry.entries, here);
 		if (below) {
-			const href = groupOf(here)?.href;
+			const group = groupOf(here);
+			// A group of names by their first letter is a way through the
+			// list, not a place: the crumb stops at the list.
+			if (group?.crumb === false) return below;
+			const href = group?.href;
 			return [{ label: entry.label, href: href ? withBase(href) : undefined }, ...below];
 		}
 	}

website/src/scripts/site.ts

@@ -784,11 +784,11 @@ const hashLookupRow = (drawer: HTMLElement, value: string) => {
 
 // Pagefind lists a page's title twice: once for the page and once for the
 // anchor of its h1 (#_top). The second title is hidden, its excerpt kept.
-// The `files`, `aliases`, `devices`, `consoles` and `terms` fields weigh in
+// The `files`, `aliases`, `devices`, `consoles`, `exact` and `terms` fields weigh in
 // the ranking and are never shown: a result is not a list of 120 names. Pagefind reuses
 // its nodes from one query to the next, so both are worked out again on
 // every change.
-const UNSHOWN_FIELDS = new Set(['files', 'aliases', 'devices', 'consoles', 'terms']);
+const UNSHOWN_FIELDS = new Set(['files', 'aliases', 'devices', 'consoles', 'exact', 'terms']);
 
 const tidyResults = (root: Element) => {
 	for (const title of root.querySelectorAll<HTMLElement>('.pagefind-ui__result-nested > .pagefind-ui__result-title')) {

website/tools/search_gate.py

@@ -63,6 +63,26 @@ QUERIES: dict[str, tuple[tuple[str, ...], int]] = {
     "psx bios": (("systems/sony",), 3),
     "dc bios missing": (("systems/sega", "emulators/flycast/"), 3),
     "gba bios": (("systems/nintendo",), 3),
+    # A console whose page answers "No file needed" says BIOS only where the
+    # search does not read: its `terms` field carries "<name> BIOS".
+    "snes bios": (("systems/nintendo-snes/",), 1),
+    "nes bios": (("systems/nintendo-nes/",), 1),
+    "gamecube bios": (("systems/nintendo-gamecube/",), 1),
+    "game boy bios": (("systems/nintendo-game-boy/",), 1),
+    "3ds bios": (("systems/nintendo-3ds/",), 1),
+    # A console another one's name begins with, both ways: the longer name's
+    # page holds every word of the shorter query. GBA, GBC and the
+    # PlayStation 3 are sections of their maker's page; Saturn.emu is an
+    # emulator named after its console.
+    "playstation 2 bios": (("systems/sony-playstation-2/",), 1),
+    "playstation 3 bios": (("systems/sony/",), 1),
+    "game boy advance bios": (("systems/nintendo/",), 1),
+    "game boy color bios": (("systems/nintendo/",), 1),
+    "nintendo ds bios": (("systems/nintendo-nintendo-ds/",), 1),
+    "nintendo dsi bios": (("systems/nintendo-nintendo-dsi/",), 1),
+    "wii bios": (("systems/nintendo/",), 1),
+    "wii u bios": (("systems/nintendo-wii-u/",), 1),
+    "saturn bios": (("systems/sega-saturn/",), 1),
     # The words an emulator prints, and the page that answers them.
     "bios not found": (("wiki/troubleshooting/",), 3),
     "please insert playstation cd-rom": (("wiki/troubleshooting/",), 3),
@@ -74,12 +94,20 @@ QUERIES: dict[str, tuple[tuple[str, ...], int]] = {
     # Checking a dump is what the Find a file page does.
     "check my dump": (("files/",), 1),
     "md5 check": (("files/",), 1),
+    "is my bios correct": (("files/",), 3),
+    # A folder is checked by the setup's command, which troubleshooting runs.
+    "check my bios folder": (("wiki/troubleshooting/",), 3),
     # A tool author's words: the data reference or the installer's.
     "manifest": (("data/", "wiki/installer/"), 3),
     "events": (("data/", "wiki/installer/"), 3),
     "exit code": (("data/", "wiki/installer/"), 3),
     "rate limit": (("data/", "wiki/installer/"), 3),
-    "ios": (("wiki/faq/",), 3),
+    "schema": (("data/",), 3),
+    "json schema": (("data/",), 3),
+    "license": (("data/", "wiki/legal/"), 3),
+    "check bios folder script": (("data/",), 3),
+    # The iPhone and iPad answer: copy the files to the app, no terminal.
+    "ios": (("which-pack/", "wiki/faq/"), 3),
     # A device name: the guide of the handhelds that run it.
     "r36s": (("platforms/", "which-pack/"), 3),
     "steam deck": (("platforms/emudeck/", "platforms/retrodeck/", "which-pack/"), 1),