Merge pull request #41 from tamnd/feat/zim-search-index

Store each page under its real title in a packed ZIM
2026-06-17 18:23:21 +07:00
parent 8bb1e344d2 6655a434b9
commit 3117c4c55c
4 changed files with 124 additions and 6 deletions
@@ -6,6 +6,17 @@ All notable changes to kage are recorded here. The format follows

 ## [Unreleased]

+### Changed
+
+- Each saved page is now stored in a packed ZIM under its own `<title>` instead of its URL path, so a ZIM reader's search box suggests pages by their readable title.
+  Typing into Kiwix's search now offers "Five Founders" or "Female Founders" rather than a filename, because the title pointer list a reader's suggestion search walks carries the real page titles.
+  A page with no `<title>` still falls back to its path, and the per-page title also flows into the `title` column of `kage parquet export`.
+
+### Added
+
+- The packing guide now documents how search works on a kage archive: title suggestions in any ZIM reader, and full-text search of page bodies through `kage parquet export` and DuckDB.
+  A Xapian full-text index is deliberately not written, since Xapian is GPL and kage is MIT.
+
 ## [0.3.4] - 2026-06-17

 ### Fixed
@@ -41,7 +41,18 @@ kage open paulgraham.com.zim            # read it back with kage
 kiwix-serve paulgraham.com.zim          # or serve it with Kiwix at http://localhost
 ```

-You can also double-click the file in the [Kiwix desktop app](https://kiwix.org/en/applications/), or load it on Kiwix for Android or iOS to read your mirror on your phone. kage writes the metadata the format and `zimcheck` treat as mandatory, including the title, description, and the favicon Kiwix shows as the book icon in its library, so the archive shows up properly rather than as an untitled, iconless entry. One caveat: kage does not write the full-text search index that Kiwix's own packs ship with, so browsing works everywhere while in-reader search is limited.
+You can also double-click the file in the [Kiwix desktop app](https://kiwix.org/en/applications/), or load it on Kiwix for Android or iOS to read your mirror on your phone. kage writes the metadata the format and `zimcheck` treat as mandatory, including the title, description, and the favicon Kiwix shows as the book icon in its library, so the archive shows up properly rather than as an untitled, iconless entry.
+
+### Searching a packed mirror
+
+Each page is stored under its own real `<title>`, so the search box in Kiwix (and any other ZIM reader) suggests pages by their readable title: type `five` into a paulgraham.com archive and it offers "Five Founders" and "Female Founders", not a filename. This title search works in every reader with no extra index.
+
+What kage does not write is a Xapian full-text index, the separate search database that lets Kiwix match words inside a page's body. Xapian is GPL and kage is MIT, so linking it in would change kage's license; rather than do that, kage leaves full-text search to its own columnar export. `kage parquet export mirror.zim` writes one row per page with a `text` column, which [DuckDB](https://duckdb.org) searches with its `fts` extension or a plain `ILIKE`, fully offline and ranked:
+
+```bash
+kage parquet export paulgraham.com.zim -o paulgraham.parquet
+duckdb -c "SELECT url FROM 'paulgraham.parquet' WHERE text ILIKE '%schlep blindness%'"
+```

 ## A self-contained binary

@@ -354,3 +354,80 @@ func TestHandler(t *testing.T) {
 		t.Errorf("GET missing = %d, want 404", code)
 	}
 }
+
+// titleOf returns the stored entry title for a content url, scanning entries in
+// url order since the reader exposes titles only through EntryAt.
+func titleOf(t *testing.T, r *zim.Reader, url string) string {
+	t.Helper()
+	for i := uint32(0); i < r.Count(); i++ {
+		e, err := r.EntryAt(i)
+		if err != nil {
+			continue
+		}
+		if e.Namespace == zim.NamespaceContent && e.URL == url {
+			return e.Title
+		}
+	}
+	t.Fatalf("no content entry for %q", url)
+	return ""
+}
+
+// TestBuildZIMPageTitles checks that each HTML page entry carries its own
+// <title> (collapsed to one line), that a page with no title falls back to its
+// url, and that a non-HTML asset keeps the url as its title.
+func TestBuildZIMPageTitles(t *testing.T) {
+	root := t.TempDir()
+	host := filepath.Join(root, "example.com")
+	files := map[string]string{
+		"index.html":       "<!doctype html><title>Home</title><h1>Hi</h1>",
+		"essay/index.html": "<!doctype html><title>\n  A Long\n  Title  </title><p>x</p>",
+		"bare/index.html":  "<!doctype html><h1>No title here</h1>",
+		"logo.png":         "\x89PNGfake",
+	}
+	for rel, body := range files {
+		p := filepath.Join(host, filepath.FromSlash(rel))
+		if err := os.MkdirAll(filepath.Dir(p), 0o755); err != nil {
+			t.Fatal(err)
+		}
+		if err := os.WriteFile(p, []byte(body), 0o644); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	out := filepath.Join(t.TempDir(), "titles.zim")
+	if _, _, err := BuildZIM(host, ZIMOptions{Out: out, Date: "2026-06-14"}); err != nil {
+		t.Fatalf("BuildZIM: %v", err)
+	}
+	r, err := zim.Open(out)
+	if err != nil {
+		t.Fatalf("Open: %v", err)
+	}
+	defer func() { _ = r.Close() }()
+
+	cases := map[string]string{
+		"index.html":       "Home",
+		"essay/index.html": "A Long Title", // newlines and runs collapsed to single spaces
+		"bare/index.html":  "bare/index.html",
+		"logo.png":         "logo.png",
+	}
+	for url, want := range cases {
+		if got := titleOf(t, r, url); got != want {
+			t.Errorf("title of %q = %q, want %q", url, got, want)
+		}
+	}
+}
+
+func TestCollapseSpaces(t *testing.T) {
+	cases := map[string]string{
+		"  hello  world ":      "hello world",
+		"line\n\tone\r\n  two": "line one two",
+		"":                     "",
+		"   ":                  "",
+		"single":               "single",
+	}
+	for in, want := range cases {
+		if got := collapseSpaces(in); got != want {
+			t.Errorf("collapseSpaces(%q) = %q, want %q", in, got, want)
+		}
+	}
+}
@@ -170,11 +170,16 @@ func buildWriter(mirrorDir string, opts ZIMOptions) (*zim.Writer, *clusterCache,
 			return err
 		}
 		mime := MimeForExt(rel)
+		title := ""
 		if mime == "text/html" {
 			htmlPages = append(htmlPages, rel)
+			// Store the page's real <title> on its entry, not the URL path, so a
+			// ZIM reader's search box and suggestions match the readable title.
+			// An empty result leaves the writer to fall back to the url.
+			title = htmlTitleOfBytes(data)
 		}
 		counts[mime]++
-		w.AddContent(zim.NamespaceContent, rel, "", mime, data)
+		w.AddContent(zim.NamespaceContent, rel, title, mime, data)
 		return nil
 	})
 	if walkErr != nil {
@@ -240,16 +245,30 @@ func htmlTitleOf(mirrorDir, mainURL string) string {
 	if mainURL == "" {
 		return ""
 	}
-	f, err := os.Open(filepath.Join(mirrorDir, filepath.FromSlash(mainURL)))
+	data, err := os.ReadFile(filepath.Join(mirrorDir, filepath.FromSlash(mainURL)))
 	if err != nil {
 		return ""
 	}
-	defer func() { _ = f.Close() }()
-	doc, err := html.Parse(f)
+	return htmlTitleOfBytes(data)
+}
+
+// htmlTitleOfBytes returns the first <title> in the given HTML with surrounding
+// and internal whitespace collapsed, or "" if the bytes do not parse or carry
+// no title. Page entries use it so a ZIM reader shows the page's real title
+// instead of its URL path.
+func htmlTitleOfBytes(data []byte) string {
+	doc, err := html.Parse(bytes.NewReader(data))
 	if err != nil {
 		return ""
 	}
-	return strings.TrimSpace(findTitle(doc))
+	return collapseSpaces(findTitle(doc))
+}
+
+// collapseSpaces trims s and replaces every run of whitespace (including the
+// newlines a wrapped <title> can carry) with a single space, so a title reads
+// as one clean line in a reader's library and search results.
+func collapseSpaces(s string) string {
+	return strings.Join(strings.Fields(s), " ")
 }

 // findTitle returns the text of the first <title> element in depth-first order.