perf(dictgen): Remove ascii/unicode branch from trie

crates/dictgen/src/gen.rs

@@ -40,7 +40,11 @@ impl<'g> DictGen<'g> {
     }
 
     pub fn ordered_map(self) -> crate::OrderedMapGen<'g> {
-        crate::OrderedMapGen { gen: self }
+        crate::OrderedMapGen {
+            gen: self,
+            unicode: true,
+            unicase: true,
+        }
     }
 
     pub fn trie(self) -> crate::TrieGen<'g> {

crates/dictgen/src/insensitive.rs

@@ -113,3 +113,97 @@ impl<'b, 'a: 'b> phf_shared::PhfBorrow<InsensitiveStr<'b>> for InsensitiveStr<'a
         self
     }
 }
+
+/// `UniCase` look-alike that avoids const-fn so large tables don't OOM
+#[derive(Copy, Clone)]
+pub struct InsensitiveAscii<'s>(pub &'s str);
+
+impl<'s> InsensitiveAscii<'s> {
+    pub fn convert(self) -> unicase::Ascii<&'s str> {
+        unicase::Ascii::new(self.0)
+    }
+
+    pub fn into_inner(self) -> &'s str {
+        self.0
+    }
+
+    pub fn is_empty(self) -> bool {
+        self.0.is_empty()
+    }
+
+    pub fn len(self) -> usize {
+        self.0.len()
+    }
+}
+
+impl<'s> From<unicase::Ascii<&'s str>> for InsensitiveAscii<'s> {
+    fn from(other: unicase::Ascii<&'s str>) -> Self {
+        Self(other.into_inner())
+    }
+}
+
+impl<'s2> PartialEq<InsensitiveAscii<'s2>> for InsensitiveAscii<'_> {
+    #[inline]
+    fn eq(&self, other: &InsensitiveAscii<'s2>) -> bool {
+        self.convert() == other.convert()
+    }
+}
+
+impl Eq for InsensitiveAscii<'_> {}
+
+impl PartialOrd for InsensitiveAscii<'_> {
+    fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
+        Some(self.cmp(other))
+    }
+}
+
+impl Ord for InsensitiveAscii<'_> {
+    fn cmp(&self, other: &Self) -> std::cmp::Ordering {
+        self.convert().cmp(&other.convert())
+    }
+}
+
+impl core::hash::Hash for InsensitiveAscii<'_> {
+    #[inline]
+    fn hash<H: core::hash::Hasher>(&self, hasher: &mut H) {
+        self.convert().hash(hasher);
+    }
+}
+
+impl core::fmt::Debug for InsensitiveAscii<'_> {
+    #[inline]
+    fn fmt(&self, fmt: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        core::fmt::Debug::fmt(self.into_inner(), fmt)
+    }
+}
+
+impl core::fmt::Display for InsensitiveAscii<'_> {
+    #[inline]
+    fn fmt(&self, fmt: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        core::fmt::Display::fmt(self.into_inner(), fmt)
+    }
+}
+
+#[cfg(feature = "map")]
+impl phf_shared::PhfHash for InsensitiveAscii<'_> {
+    #[inline]
+    fn phf_hash<H: core::hash::Hasher>(&self, state: &mut H) {
+        core::hash::Hash::hash(self, state);
+    }
+}
+
+#[cfg(feature = "map")]
+impl phf_shared::FmtConst for InsensitiveAscii<'_> {
+    fn fmt_const(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        f.write_str("dictgen::InsensitiveAscii(")?;
+        self.into_inner().fmt_const(f)?;
+        f.write_str(")")
+    }
+}
+
+#[cfg(feature = "map")]
+impl<'b, 'a: 'b> phf_shared::PhfBorrow<InsensitiveAscii<'b>> for InsensitiveAscii<'a> {
+    fn borrow(&self) -> &InsensitiveAscii<'b> {
+        self
+    }
+}

crates/dictgen/src/ordered_map.rs

@@ -1,10 +1,22 @@
 #[cfg(feature = "codegen")]
 pub struct OrderedMapGen<'g> {
     pub(crate) gen: crate::DictGen<'g>,
+    pub(crate) unicase: bool,
+    pub(crate) unicode: bool,
 }
 
 #[cfg(feature = "codegen")]
 impl OrderedMapGen<'_> {
+    pub fn unicase(mut self, yes: bool) -> Self {
+        self.unicase = yes;
+        self
+    }
+
+    pub fn unicode(mut self, yes: bool) -> Self {
+        self.unicode = yes;
+        self
+    }
+
     pub fn write<'d, W: std::io::Write, V: std::fmt::Display>(
         &self,
         file: &mut W,
@@ -14,7 +26,7 @@ impl OrderedMapGen<'_> {
         data.sort_unstable_by_key(|v| unicase::UniCase::new(v.0));
 
         let name = self.gen.name;
-        let key_type = "dictgen::InsensitiveStr<'static>";
+        let key_type = self.key_type();
         let value_type = self.gen.value_type;
 
         let mut smallest = usize::MAX;
@@ -29,11 +41,7 @@ impl OrderedMapGen<'_> {
             smallest = std::cmp::min(smallest, key.len());
             largest = std::cmp::max(largest, key.len());
 
-            let key = if key.is_ascii() {
-                format!("dictgen::InsensitiveStr::Ascii({key:?})")
-            } else {
-                format!("dictgen::InsensitiveStr::Unicode({key:?})")
-            };
+            let key = self.key_new(key);
 
             writeln!(file, "      {key},")?;
         }
@@ -51,6 +59,28 @@ impl OrderedMapGen<'_> {
 
         Ok(())
     }
+
+    fn key_type(&self) -> &'static str {
+        match (self.unicase, self.unicode) {
+            (true, true) => "dictgen::InsensitiveStr<'static>",
+            (true, false) => "dictgen::InsensitiveAscii<'static>",
+            (false, _) => "&'static str",
+        }
+    }
+
+    fn key_new(&self, key: &str) -> String {
+        match (self.unicase, self.unicode) {
+            (true, true) => {
+                if key.is_ascii() {
+                    format!("dictgen::InsensitiveStr::Ascii({key:?})")
+                } else {
+                    format!("dictgen::InsensitiveStr::Unicode({key:?})")
+                }
+            }
+            (true, false) => format!("dictgen::InsensitiveAscii({key:?})"),
+            (false, _) => format!("{key:?}"),
+        }
+    }
 }
 
 pub struct OrderedMap<K: 'static, V: 'static> {
@@ -73,6 +103,20 @@ impl<V> OrderedMap<crate::InsensitiveStr<'_>, V> {
     }
 }
 
+impl<V> OrderedMap<crate::InsensitiveAscii<'_>, V> {
+    #[inline]
+    pub fn find(&self, word: &'_ unicase::Ascii<&str>) -> Option<&'static V> {
+        if self.range.contains(&word.len()) {
+            self.keys
+                .binary_search_by_key(word, |key| key.convert())
+                .map(|i| &self.values[i])
+                .ok()
+        } else {
+            None
+        }
+    }
+}
+
 impl<V> OrderedMap<&str, V> {
     #[inline]
     pub fn find(&self, word: &'_ &str) -> Option<&'static V> {

crates/dictgen/src/trie.rs

@@ -75,7 +75,7 @@ impl<V> Trie<V> {
                     // Unsafe: Everything before has been proven to be ASCII, so this should be
                     // safe.
                     let remaining = unsafe { core::str::from_utf8_unchecked(remaining) };
-                    let remaining = unicase::UniCase::ascii(remaining);
+                    let remaining = unicase::Ascii::new(remaining);
                     return t.find(&remaining);
                 }
             }
@@ -91,7 +91,7 @@ pub struct TrieNode<V: 'static> {
 
 pub enum TrieChild<V: 'static> {
     Nested(&'static [Option<&'static TrieNode<V>>; 26]),
-    Flat(&'static crate::OrderedMap<crate::InsensitiveStr<'static>, V>),
+    Flat(&'static crate::OrderedMap<crate::InsensitiveAscii<'static>, V>),
 }
 
 #[cfg(feature = "codegen")]
@@ -179,6 +179,7 @@ mod codegen {
                         .name(&children_name)
                         .value_type(value_type)
                         .ordered_map()
+                        .unicode(false)
                         .write(file, table_input)?;
                 }
             }