From 6af917e869409c76be36bbb9b2b8e9bea1bfeaa1 Mon Sep 17 00:00:00 2001
From: "Dylan M. Taylor" <dylan@dylanmtaylor.com>
Date: Tue, 3 Nov 2026 20:27:51 +0000
Subject: [PATCH] Spelling: stop an affix header's count at the end of the
 file, share aliased dictionaries

---
 crates/engine/src/spell.rs | 42 +++++++++++++++++++++++++++++++++-----
 1 file changed, 37 insertions(+), 5 deletions(-)

diff --git a/crates/engine/src/spell.rs b/crates/engine/src/spell.rs
index 2dda5c6f4..a04710eb4 100644
--- a/crates/engine/src/spell.rs
+++ b/crates/engine/src/spell.rs
@@ -171,7 +171,9 @@ impl Dictionary {
                 ["COMPOUNDWORDMAX", n, ..] => d.compound_max = n.parse().unwrap_or(0),
                 ["REP", n] => {
                     for _ in 0..n.parse::<usize>().unwrap_or(0) {
-                        let r: Vec<&str> = lines.next().unwrap_or("").split_whitespace().collect();
+                        // The count is the file's claim: stop when the records run out.
+                        let Some(line) = lines.next() else { break };
+                        let r: Vec<&str> = line.split_whitespace().collect();
                         if let [_, a, b, ..] = r.as_slice() {
                             d.rep.push((a.replace('_', " "), b.replace('_', " ")));
                         }
@@ -182,7 +184,8 @@ impl Dictionary {
                     let Some(flag) = one_flag(Some(flag), mode) else { continue };
                     let cross = cross.eq_ignore_ascii_case("y");
                     for _ in 0..n.parse::<usize>().unwrap_or(0) {
-                        let r: Vec<&str> = lines.next().unwrap_or("").split_whitespace().collect();
+                        let Some(line) = lines.next() else { break };
+                        let r: Vec<&str> = line.split_whitespace().collect();
                         let [_, _, strip, add, rest @ ..] = r.as_slice() else { continue };
                         let (add, cont) = match add.split_once('/') {
                             Some((a, c)) => (a, parse_flags(c, mode)),
@@ -576,6 +579,15 @@ fn cache() -> &'static Cache {
     CACHE.get_or_init(Default::default)
 }
 
+/// Language strings remembered by [`dictionary`]; a string past it is resolved again on each call.
+const MAX_CACHED_LANGS: usize = 256;
+
+/// Parsed dictionaries by their affix file, so aliases of one language share it.
+fn by_files() -> &'static Mutex<HashMap<PathBuf, Arc<Dictionary>>> {
+    static M: OnceLock<Mutex<HashMap<PathBuf, Arc<Dictionary>>>> = OnceLock::new();
+    M.get_or_init(Default::default)
+}
+
 /// The dictionary for `lang` (a BCP-47 tag such as `en-US`), loaded once per process.
 pub fn dictionary(lang: &str) -> Option<Arc<Dictionary>> {
     let cache = cache();
@@ -586,10 +598,22 @@ pub fn dictionary(lang: &str) -> Option<Arc<Dictionary>> {
         return d.clone();
     }
     let loaded = find_in(&search_dirs(), lang).and_then(|(a, d)| {
-        let (a, d) = (std::fs::read(a).ok()?, std::fs::read(d).ok()?);
-        Some(Arc::new(Dictionary::parse(&a, &d)))
+        // Tags that resolve to the same files share one parsed dictionary.
+        if let Ok(m) = by_files().lock()
+            && let Some(dict) = m.get(&a)
+        {
+            return Some(dict.clone());
+        }
+        let (bytes_a, bytes_d) = (std::fs::read(&a).ok()?, std::fs::read(d).ok()?);
+        let dict = Arc::new(Dictionary::parse(&bytes_a, &bytes_d));
+        if let Ok(mut m) = by_files().lock() {
+            m.insert(a, dict.clone());
+        }
+        Some(dict)
     });
-    if let Ok(mut c) = cache.lock() {
+    if let Ok(mut c) = cache.lock()
+        && c.len() < MAX_CACHED_LANGS
+    {
         c.insert(key, loaded.clone());
     }
     loaded
@@ -880,6 +904,14 @@ SFX D Y 1\nSFX D 0 ed/S .\n";
         assert!(d.suggest("workingslides", 5).contains(&"working slides".to_string()));
     }
 
+    /// An affix header can claim billions of records; parsing stops where the file does.
+    #[test]
+    fn a_header_count_past_the_records_ends_with_them() {
+        let aff = b"SFX S Y 4000000000\nSFX S 0 s .\nREP 4000000000\nREP a b\n";
+        let d = Dictionary::parse(aff, b"1\ncat/S\n".as_slice());
+        assert!(d.check("cats"));
+    }
+
     /// #1175: compound splitting is quadratic in the word's length, so a word longer than
     /// `MAX_CHECK_CHARS` is not checked; shorter ones still are.
     #[test]