feat(web): lazy per-script font loading for international CAD text

The web build has no system-font access, so CAD text in non-Latin scripts
(Cyrillic, Greek, CJK, …) rendered as nothing — build_fallback was a hard
None on wasm (#141). Instead, ship per-script Noto subsets under web/fonts
(one alphabet per file, SIL OFL 1.1) and fetch them lazily over HTTP the
first time a drawing uses that script, then outline glyphs with ttf-parser.

- web_font: Script enum + script_of (char → script font), a per-script
  store with lazy fetch, and a pending queue drained by the app loop.
- CJK is split by language (chinese/japanese/korean). Han ideographs share
  code points but differ by language, so the shared block is routed by the
  document's $DWGCODEPAGE (932→JP, 949→KR, GB/936→CN); kana is always
  Japanese, Hangul always Korean. Re-tessellates when the language changes.
- build_fallback (wasm) outlines from the fetched subset; clear_fallback_cache
  lets glyphs that missed while a font was in flight reappear on load.
- PollWebFonts subscription (web only) + WebFontLoaded message drive fetches.
- Trunk copy-dir serves web/fonts; nothing is bundled into the native binary
  (desktop keeps using system fonts).

Co-Authored-By: Claude Opus 4.8 <[email protected]>

src/scene/text/mod.rs

@@ -2,4 +2,5 @@ pub mod lff;
 pub mod font_face;
 pub mod sysfont;
 pub mod ttf_glyph;
+pub mod web_font;
 pub mod complex_lt;

src/scene/text/ttf_glyph.rs

@@ -263,11 +263,35 @@ pub fn fallback_glyph(ch: char) -> Option<Arc<Glyph>> {
     built
 }
 
-/// Web: no system fonts → no cosmic-text fallback (LFF-missing glyphs simply
-/// do not render).
+/// Drop the cached fallback glyphs so the next lookup re-resolves. The web
+/// build calls this when a per-script font finishes loading: glyphs that
+/// resolved to `None` while the font was still in flight then get a real
+/// outline. (#141)
+pub fn clear_fallback_cache() {
+    fallback_cache().lock().unwrap().clear();
+}
+
+/// Web: outline the glyph from the lazily-fetched per-script Noto subset that
+/// covers it. Returns `None` while that font is still loading (the char renders
+/// once it arrives and the fallback cache is cleared). (#141)
 #[cfg(target_arch = "wasm32")]
-fn build_fallback(_ch: char) -> Option<Arc<Glyph>> {
-    None
+fn build_fallback(ch: char) -> Option<Arc<Glyph>> {
+    let script = crate::scene::text::web_font::script_of(ch)?;
+    let bytes = crate::scene::text::web_font::request(script)?;
+    let face = ttf_parser::Face::parse(&bytes, 0).ok()?;
+    let gid = face.glyph_index(ch)?;
+    let k = cap_scale(&face);
+    let advance = face.glyph_hor_advance(gid).unwrap_or(0) as f32 * k;
+    let mut fl = OutlineFlattener::new(k);
+    face.outline_glyph(gid, &mut fl);
+    fl.flush();
+    if fl.contours.is_empty() {
+        return None;
+    }
+    Some(Arc::new(Glyph {
+        strokes: fl.contours,
+        advance,
+    }))
 }
 
 #[cfg(not(target_arch = "wasm32"))]

src/scene/text/web_font.rs

@@ -0,0 +1,199 @@
+//! Per-script web font store + lazy fetch (#141).
+//!
+//! The desktop build outlines glyphs for non-LFF scripts from the user's
+//! installed system fonts (see `ttf_glyph::build_fallback`). The web build has
+//! no system-font access, so it instead lazily fetches a per-script Noto subset
+//! — served from `fonts/<script>.ttf`, one alphabet per file — the first time a
+//! drawing uses that script, then outlines glyphs from it. Splitting per script
+//! keeps each download small (Latin/Cyrillic/Greek ~50–100 KB; CJK loads only
+//! when a CJK drawing is opened).
+//!
+//! The store and fetch are web-only; the desktop side keeps no-op stubs so the
+//! shared call sites (`ttf_glyph`, the app message loop) compile unchanged.
+
+use std::sync::atomic::{AtomicU8, Ordering};
+
+/// A script we ship a Noto subset for. [`script_of`] maps a char to one.
+///
+/// CJK is split by language — Chinese, Japanese and Korean each get their own
+/// file. Their ideographs (Han, U+4E00–9FFF) share the same code points but
+/// differ in glyph shape, so the shared block is routed by the document's
+/// language (see [`set_cjk_lang_from_codepage`]); kana is always Japanese and
+/// Hangul always Korean.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
+pub enum Script {
+    Latin,
+    Cyrillic,
+    Greek,
+    Arabic,
+    Hebrew,
+    Thai,
+    Devanagari,
+    Chinese,
+    Japanese,
+    Korean,
+}
+
+impl Script {
+    /// Same-origin asset path the web build fetches this script's font from.
+    /// Must match the files produced by `web/fonts/generate.sh`.
+    pub fn asset(self) -> &'static str {
+        match self {
+            Script::Latin => "fonts/latin.ttf",
+            Script::Cyrillic => "fonts/cyrillic.ttf",
+            Script::Greek => "fonts/greek.ttf",
+            Script::Arabic => "fonts/arabic.ttf",
+            Script::Hebrew => "fonts/hebrew.ttf",
+            Script::Thai => "fonts/thai.ttf",
+            Script::Devanagari => "fonts/devanagari.ttf",
+            Script::Chinese => "fonts/chinese.ttf",
+            Script::Japanese => "fonts/japanese.ttf",
+            Script::Korean => "fonts/korean.ttf",
+        }
+    }
+}
+
+/// Language used to render shared Han ideographs: 0 = Chinese, 1 = Japanese,
+/// 2 = Korean. Set from the active document's code page.
+static CJK_LANG: AtomicU8 = AtomicU8::new(0);
+
+fn cjk_lang() -> Script {
+    match CJK_LANG.load(Ordering::Relaxed) {
+        1 => Script::Japanese,
+        2 => Script::Korean,
+        _ => Script::Chinese,
+    }
+}
+
+/// Point the shared-Han routing at a language based on a DWG/DXF code page
+/// (`$DWGCODEPAGE`), e.g. `ANSI_932` → Japanese, `ANSI_949` → Korean, GB/936 or
+/// anything else → Chinese. Returns `true` if the language changed (the caller
+/// then clears the glyph cache and re-tessellates).
+pub fn set_cjk_lang_from_codepage(code_page: &str) -> bool {
+    let c = code_page.to_ascii_uppercase();
+    let lang = if c.contains("932") || c.contains("SJIS") || c.contains("SHIFT") {
+        1 // Japanese (Shift-JIS)
+    } else if c.contains("949") || c.contains("KOR") || c.contains("UHC") {
+        2 // Korean
+    } else {
+        0 // Chinese (936 / GB / 950 / Big5) or non-CJK default
+    };
+    CJK_LANG.swap(lang, Ordering::Relaxed) != lang
+}
+
+/// The script font that covers `ch`, or `None` for control / uncovered code
+/// points. Ranges mirror the subset unicode ranges in `web/fonts/generate.sh`.
+pub fn script_of(ch: char) -> Option<Script> {
+    Some(match ch as u32 {
+        0x0000..=0x024F | 0x1E00..=0x1EFF | 0x2000..=0x206F => Script::Latin,
+        0x0370..=0x03FF | 0x1F00..=0x1FFF => Script::Greek,
+        0x0400..=0x052F | 0x2DE0..=0x2DFF | 0xA640..=0xA69F => Script::Cyrillic,
+        0x0590..=0x05FF | 0xFB1D..=0xFB4F => Script::Hebrew,
+        0x0600..=0x06FF | 0x0750..=0x077F | 0x08A0..=0x08FF | 0xFB50..=0xFDFF | 0xFE70..=0xFEFF => {
+            Script::Arabic
+        }
+        0x0900..=0x097F => Script::Devanagari,
+        // Hangul → always Korean; kana → always Japanese.
+        0x1100..=0x11FF | 0x3130..=0x318F | 0xAC00..=0xD7A3 => Script::Korean,
+        0x3040..=0x30FF | 0x31F0..=0x31FF => Script::Japanese,
+        // Shared Han + CJK symbols + fullwidth → routed by the document language.
+        0x3000..=0x303F | 0x3400..=0x4DBF | 0x4E00..=0x9FFF | 0xF900..=0xFAFF | 0xFF00..=0xFFEF => {
+            cjk_lang()
+        }
+        _ => return None,
+    })
+}
+
+// ── Web store ───────────────────────────────────────────────────────────────
+
+#[cfg(target_arch = "wasm32")]
+mod imp {
+    use super::Script;
+    use std::collections::HashMap;
+    use std::sync::{Arc, Mutex, OnceLock};
+
+    enum State {
+        Loading,
+        Loaded(Arc<Vec<u8>>),
+        Failed,
+    }
+
+    #[derive(Default)]
+    struct Store {
+        states: HashMap<Script, State>,
+        /// Scripts requested but not yet fetched; the app drains this and kicks
+        /// off the fetch tasks.
+        pending: Vec<Script>,
+    }
+
+    fn store() -> &'static Mutex<Store> {
+        static S: OnceLock<Mutex<Store>> = OnceLock::new();
+        S.get_or_init(|| Mutex::new(Store::default()))
+    }
+
+    /// Loaded font bytes for `script`, or `None` — queueing a fetch the first
+    /// time a script is missed so the app loop can load it.
+    pub fn request(script: Script) -> Option<Arc<Vec<u8>>> {
+        let mut s = store().lock().unwrap();
+        match s.states.get(&script) {
+            Some(State::Loaded(b)) => Some(b.clone()),
+            Some(_) => None, // Loading or Failed — don't re-queue.
+            None => {
+                s.states.insert(script, State::Loading);
+                s.pending.push(script);
+                None
+            }
+        }
+    }
+
+    /// Drain the scripts awaiting a fetch.
+    pub fn take_pending() -> Vec<Script> {
+        std::mem::take(&mut store().lock().unwrap().pending)
+    }
+
+    /// Record a fetch result: `Some(bytes)` on success, `None` on failure.
+    pub fn insert(script: Script, bytes: Option<Vec<u8>>) {
+        let st = match bytes {
+            Some(b) => State::Loaded(Arc::new(b)),
+            None => State::Failed,
+        };
+        store().lock().unwrap().states.insert(script, st);
+    }
+
+    /// Fetch a script font over HTTP from the same origin.
+    pub async fn fetch(script: Script) -> Result<Vec<u8>, String> {
+        use wasm_bindgen::JsCast;
+        use wasm_bindgen_futures::JsFuture;
+        let win = web_sys::window().ok_or("no window")?;
+        let resp_val = JsFuture::from(win.fetch_with_str(script.asset()))
+            .await
+            .map_err(|e| format!("{e:?}"))?;
+        let resp: web_sys::Response = resp_val.dyn_into().map_err(|_| "bad response".to_string())?;
+        if !resp.ok() {
+            return Err(format!("HTTP {}", resp.status()));
+        }
+        let ab = JsFuture::from(resp.array_buffer().map_err(|e| format!("{e:?}"))?)
+            .await
+            .map_err(|e| format!("{e:?}"))?;
+        Ok(js_sys::Uint8Array::new(&ab).to_vec())
+    }
+}
+
+#[cfg(not(target_arch = "wasm32"))]
+mod imp {
+    use super::Script;
+    use std::sync::Arc;
+
+    pub fn request(_script: Script) -> Option<Arc<Vec<u8>>> {
+        None
+    }
+    pub fn take_pending() -> Vec<Script> {
+        Vec::new()
+    }
+    pub fn insert(_script: Script, _bytes: Option<Vec<u8>>) {}
+    pub async fn fetch(_script: Script) -> Result<Vec<u8>, String> {
+        Err("web only".into())
+    }
+}
+
+pub use imp::{fetch, insert, request, take_pending};