Merge pull request #301 from Kijewski/pr-digits

Speed-up HTML escaping a bit

rinja/src/html.rs

@@ -41,13 +41,10 @@ pub(crate) fn write_escaped_char(mut dest: impl fmt::Write, c: char) -> fmt::Res
 }
 
 /// Returns the decimal representation of the codepoint if the character needs HTML escaping.
-#[inline(always)]
+#[inline]
 fn get_escaped(byte: u8) -> Option<[u8; 2]> {
     match byte {
-        MIN_CHAR..=MAX_CHAR => match TABLE.lookup[(byte - MIN_CHAR) as usize] {
-            0 => None,
-            escaped => Some(escaped.to_ne_bytes()),
-        },
+        MIN_CHAR..=MAX_CHAR => Some(TABLE.0[(byte - MIN_CHAR) as usize]?.to_bytes()),
         _ => None,
     }
 }
@@ -93,24 +90,17 @@ const MAX_CHAR: u8 = {
 /// Number of codepoints between the lowest and highest character that needs escaping, incl.
 const CHAR_RANGE: usize = (MAX_CHAR - MIN_CHAR + 1) as usize;
 
-struct Table {
-    _align: [usize; 0],
-    lookup: [u16; CHAR_RANGE],
-}
+#[repr(align(64))]
+struct Table([Option<Digits>; CHAR_RANGE]);
 
 /// For characters that need HTML escaping, the codepoint is formatted as decimal digits,
 /// otherwise `b"\0\0"`. Starting at [`MIN_CHAR`].
-const TABLE: Table = {
-    let mut table = Table {
-        _align: [],
-        lookup: [0; CHAR_RANGE],
-    };
+const TABLE: &Table = &{
+    let mut table = Table([None; CHAR_RANGE]);
     let mut i = 0;
     while i < CHARS.len() {
         let c = CHARS[i];
-        let h = c / 10 + b'0';
-        let l = c % 10 + b'0';
-        table.lookup[(c - MIN_CHAR) as usize] = u16::from_ne_bytes([h, l]);
+        table.0[(c - MIN_CHAR) as usize] = Some(Digits::new(c));
         i += 1;
     }
     table
@@ -120,6 +110,45 @@ const TABLE: Table = {
 const ESCAPED_BUF_INIT: [u8; 8] = *b"&#__;\0\0\0";
 const ESCAPED_BUF_LEN: usize = b"&#__;".len();
 
+/// All possible decimal representations of codepoints that need escaping in HTML / XML.
+///
+/// Using this type instead of e.g. `Option<NonZeroU16>` allows rustc to select any and all of the
+/// 65,531 unused representations of this type as niche, which can can help speeding up the
+/// generated byte code. If not all needed representations were present, then the `TABLE` definition
+/// would fail to compile.
+#[derive(Debug, Clone, Copy)]
+#[repr(u16)]
+enum Digits {
+    /// `'"'`
+    V34 = u16::from_ne_bytes(*b"34"),
+    /// `'&'`
+    V38 = u16::from_ne_bytes(*b"38"),
+    /// `'\''`
+    V39 = u16::from_ne_bytes(*b"39"),
+    /// `'<'`
+    V60 = u16::from_ne_bytes(*b"60"),
+    /// `'>'`
+    V62 = u16::from_ne_bytes(*b"62"),
+}
+
+impl Digits {
+    #[inline]
+    const fn to_bytes(self) -> [u8; 2] {
+        (self as u16).to_ne_bytes()
+    }
+
+    const fn new(v: u8) -> Self {
+        match v {
+            34 => Self::V34,
+            38 => Self::V38,
+            39 => Self::V39,
+            60 => Self::V60,
+            62 => Self::V62,
+            _ => panic!(),
+        }
+    }
+}
+
 #[test]
 #[cfg(feature = "alloc")]
 fn test_simple_html_string_escaping() {