From 31a1d5cbf97bea81418d79c9831d48702682a499 Mon Sep 17 00:00:00 2001
From: CPunisher <1343316114@qq.com>
Date: Fri, 12 Sep 2025 11:11:08 +0800
Subject: [PATCH 1/5] wtf8

---
 crates/hstr/Cargo.toml                |    1 +
 crates/hstr/src/lib.rs                |    1 +
 crates/hstr/src/wtf8/mod.rs           | 1265 +++++++++++++++++++++++++
 crates/hstr/src/wtf8/not_quite_std.rs |  250 +++++
 4 files changed, 1517 insertions(+)
 create mode 100644 crates/hstr/src/wtf8/mod.rs
 create mode 100644 crates/hstr/src/wtf8/not_quite_std.rs

diff --git a/crates/hstr/Cargo.toml b/crates/hstr/Cargo.toml
index 1c9e615d348d..bb178d06a4e3 100644
--- a/crates/hstr/Cargo.toml
+++ b/crates/hstr/Cargo.toml
@@ -18,6 +18,7 @@ rkyv          = ["dep:rkyv"]
 serde         = ["dep:serde"]
 
 [dependencies]
+arrayvec              = { workspace = true }
 hashbrown             = { workspace = true }
 new_debug_unreachable = { workspace = true }
 once_cell             = { workspace = true }
diff --git a/crates/hstr/src/lib.rs b/crates/hstr/src/lib.rs
index 6f2c842bcf33..39e370db1348 100644
--- a/crates/hstr/src/lib.rs
+++ b/crates/hstr/src/lib.rs
@@ -22,6 +22,7 @@ mod global_store;
 mod tagged_value;
 #[cfg(test)]
 mod tests;
+pub mod wtf8;
 
 /// An immutable string which is cheap to clone, compare, hash, and has small
 /// size.
diff --git a/crates/hstr/src/wtf8/mod.rs b/crates/hstr/src/wtf8/mod.rs
new file mode 100644
index 000000000000..43374fdea1a8
--- /dev/null
+++ b/crates/hstr/src/wtf8/mod.rs
@@ -0,0 +1,1265 @@
+/*!
+
+Implementation of [the WTF-8 encoding](https://simonsapin.github.io/wtf-8/).
+
+This library uses Rust’s type system to maintain
+[well-formedness](https://simonsapin.github.io/wtf-8/#well-formed),
+like the `String` and `&str` types do for UTF-8.
+
+Since [WTF-8 must not be used
+for interchange](https://simonsapin.github.io/wtf-8/#intended-audience),
+this library deliberately does not provide access to the underlying bytes
+of WTF-8 strings,
+nor can it decode WTF-8 from arbitrary bytes.
+WTF-8 strings can be obtained from UTF-8, UTF-16, or code points.
+
+*/
+
+extern crate alloc;
+
+use alloc::{borrow::Cow, string::String, vec::Vec};
+use core::{
+    cmp::Ordering,
+    fmt, hash,
+    iter::{FromIterator, IntoIterator},
+    mem::transmute,
+    ops::Deref,
+    slice, str,
+};
+use std::iter::FusedIterator;
+
+use arrayvec::ArrayVec;
+
+mod not_quite_std;
+
+static UTF8_REPLACEMENT_CHARACTER: &[u8] = b"\xEF\xBF\xBD";
+
+/// A Unicode code point: from U+0000 to U+10FFFF.
+///
+/// Compare with the `char` type,
+/// which represents a Unicode scalar value:
+/// a code point that is not a surrogate (U+D800 to U+DFFF).
+#[derive(Eq, PartialEq, Ord, PartialOrd, Clone)]
+pub struct CodePoint {
+    value: u32,
+}
+
+impl Copy for CodePoint {}
+
+/// Format the code point as `U+` followed by four to six hexadecimal digits.
+/// Example: `U+1F4A9`
+impl fmt::Debug for CodePoint {
+    #[inline]
+    fn fmt(&self, formatter: &mut fmt::Formatter) -> Result<(), fmt::Error> {
+        write!(formatter, "U+{:04X}", self.value)
+    }
+}
+
+impl CodePoint {
+    /// Unsafely create a new `CodePoint` without checking the value.
+    ///
+    /// Only use when `value` is known to be less than or equal to 0x10FFFF.
+    #[inline]
+    #[allow(clippy::missing_safety_doc)]
+    pub unsafe fn from_u32_unchecked(value: u32) -> CodePoint {
+        CodePoint { value }
+    }
+
+    /// Create a new `CodePoint` if the value is a valid code point.
+    ///
+    /// Return `None` if `value` is above 0x10FFFF.
+    #[inline]
+    pub fn from_u32(value: u32) -> Option<CodePoint> {
+        match value {
+            0..=0x10ffff => Some(CodePoint { value }),
+            _ => None,
+        }
+    }
+
+    /// Create a new `CodePoint` from a `char`.
+    ///
+    /// Since all Unicode scalar values are code points, this always succeds.
+    #[inline]
+    pub fn from_char(value: char) -> CodePoint {
+        CodePoint {
+            value: value as u32,
+        }
+    }
+
+    /// Return the numeric value of the code point.
+    #[inline]
+    pub fn to_u32(&self) -> u32 {
+        self.value
+    }
+
+    /// Optionally return a Unicode scalar value for the code point.
+    ///
+    /// Return `None` if the code point is a surrogate (from U+D800 to U+DFFF).
+    #[inline]
+    pub fn to_char(&self) -> Option<char> {
+        match self.value {
+            0xd800..=0xdfff => None,
+            _ => Some(unsafe { char::from_u32_unchecked(self.value) }),
+        }
+    }
+
+    /// Return a Unicode scalar value for the code point.
+    ///
+    /// Return `'\u{FFFD}'` (the replacement character “�”)
+    /// if the code point is a surrogate (from U+D800 to U+DFFF).
+    #[inline]
+    pub fn to_char_lossy(&self) -> char {
+        self.to_char().unwrap_or('\u{FFFD}')
+    }
+}
+
+pub struct CharIter(ArrayVec<char, 12>);
+
+/// Ported from https://github.com/web-infra-dev/oxc/blob/99a4816ce7b6132b2667257984f9d92ae3768f03/crates/oxc_parser/src/lexer/mod.rs#L1349-L1374
+impl IntoIterator for CodePoint {
+    type IntoIter = CharIter;
+    type Item = char;
+
+    #[allow(unsafe_code)]
+    fn into_iter(self) -> Self::IntoIter {
+        //        // TODO: Check if this is correct
+        //        fn to_char(v: u8) -> char {
+        //            char::from_digit(v as _, 16).unwrap_or('0')
+        //        }
+
+        CharIter(match char::from_u32(self.value) {
+            Some(c) => {
+                let mut buf = ArrayVec::new();
+                // Safety: we can make sure that `buf` has enough capacity
+                unsafe {
+                    buf.push_unchecked(c);
+                }
+                buf
+            }
+            None => {
+                let mut buf = ArrayVec::new();
+
+                let high = self.value & 0xffff0000 >> 16;
+
+                let low = self.value & 0x0000ffff;
+
+                // The second code unit of a surrogate pair is always in the range from 0xDC00
+                // to 0xDFFF, and is called a low surrogate or a trail surrogate.
+                if !(0xdc00..=0xdfff).contains(&low) {
+                    // Safety: we can make sure that `buf` has enough capacity
+                    unsafe {
+                        buf.push_unchecked('\\');
+                        buf.push_unchecked('u');
+                        for c in format!("{high:x}").chars() {
+                            buf.push_unchecked(c);
+                        }
+                        buf.push_unchecked('\\');
+                        buf.push_unchecked('u');
+                        for c in format!("{low:x}").chars() {
+                            buf.push_unchecked(c);
+                        }
+                    }
+                } else {
+                    // `https://tc39.es/ecma262/#sec-utf16decodesurrogatepair`
+                    let astral_code_point = (high - 0xd800) * 0x400 + low - 0xdc00 + 0x10000;
+
+                    // Safety: we can make sure that `buf` has enough capacity
+                    unsafe {
+                        buf.push_unchecked('\\');
+                        buf.push_unchecked('u');
+                        for c in format!("{astral_code_point:x}").chars() {
+                            buf.push_unchecked(c);
+                        }
+                    }
+                }
+
+                buf
+            }
+        })
+    }
+}
+
+impl Iterator for CharIter {
+    type Item = char;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        if self.0.is_empty() {
+            None
+        } else {
+            Some(self.0.remove(0))
+        }
+    }
+}
+
+impl FusedIterator for CharIter {}
+
+/// An owned, growable string of well-formed WTF-8 data.
+///
+/// Similar to `String`, but can additionally contain surrogate code points
+/// if they’re not in a surrogate pair.
+#[derive(Eq, PartialEq, Ord, PartialOrd, Clone)]
+pub struct Wtf8Buf {
+    bytes: Vec<u8>,
+}
+
+impl Deref for Wtf8Buf {
+    type Target = Wtf8;
+
+    fn deref(&self) -> &Wtf8 {
+        unsafe { transmute(&*self.bytes) }
+    }
+}
+
+/// Format the string with double quotes,
+/// and surrogates as `\u` followed by four hexadecimal digits.
+/// Example: `"a\u{D800}"` for a string with code points [U+0061, U+D800]
+impl fmt::Debug for Wtf8Buf {
+    #[inline]
+    fn fmt(&self, formatter: &mut fmt::Formatter) -> Result<(), fmt::Error> {
+        Wtf8::fmt(self, formatter)
+    }
+}
+
+impl Wtf8Buf {
+    /// Create an new, empty WTF-8 string.
+    #[inline]
+    #[allow(clippy::new_without_default)]
+    pub fn new() -> Wtf8Buf {
+        Wtf8Buf { bytes: Vec::new() }
+    }
+
+    /// Create an new, empty WTF-8 string with pre-allocated capacity for `n`
+    /// bytes.
+    #[inline]
+    pub fn with_capacity(n: usize) -> Wtf8Buf {
+        Wtf8Buf {
+            bytes: Vec::with_capacity(n),
+        }
+    }
+
+    /// Create a WTF-8 string from an UTF-8 `String`.
+    ///
+    /// This takes ownership of the `String` and does not copy.
+    ///
+    /// Since WTF-8 is a superset of UTF-8, this always succeeds.
+    #[inline]
+    pub fn froming(string: String) -> Wtf8Buf {
+        Wtf8Buf {
+            bytes: string.into_bytes(),
+        }
+    }
+
+    /// Create a WTF-8 string from a potentially ill-formed UTF-16 slice of
+    /// 16-bit code units.
+    ///
+    /// This is lossless: calling `.to_ill_formed_utf16()` on the resulting
+    /// string will always return the original code units.
+    pub fn from_ill_formed_utf16(v: &[u16]) -> Wtf8Buf {
+        let mut string = Wtf8Buf::with_capacity(v.len());
+        for item in not_quite_std::decode_utf16(v.iter().cloned()) {
+            match item {
+                Ok(c) => string.push_char(c),
+                Err(s) => {
+                    // Surrogates are known to be in the code point range.
+                    let code_point = unsafe { CodePoint::from_u32_unchecked(s as u32) };
+                    // Skip the WTF-8 concatenation check,
+                    // surrogate pairs are already decoded by utf16_items
+                    not_quite_std::push_code_point(&mut string, code_point)
+                }
+            }
+        }
+        string
+    }
+
+    /// Reserves capacity for at least `additional` more bytes to be inserted
+    /// in the given `Wtf8Buf`.
+    /// The collection may reserve more space to avoid frequent reallocations.
+    ///
+    /// # Panics
+    ///
+    /// Panics if the new capacity overflows `usize`.
+    ///
+    /// # Example
+    ///
+    /// ```
+    /// let mut s = Wtf8Buf::new();
+    /// s.reserve(10);
+    /// assert!(s.capacity() >= 10);
+    /// ```
+    #[inline]
+    pub fn reserve(&mut self, additional: usize) {
+        self.bytes.reserve(additional)
+    }
+
+    /// Returns the number of bytes that this string buffer can hold without
+    /// reallocating.
+    ///
+    /// # Example
+    ///
+    /// ```
+    /// let s = Wtf8Buf::with_capacity(10);
+    /// assert!(s.capacity() >= 10);
+    /// ```
+    #[inline]
+    pub fn capacity(&self) -> usize {
+        self.bytes.capacity()
+    }
+
+    /// Append an UTF-8 slice at the end of the string.
+    #[inline]
+    pub fn push_str(&mut self, other: &str) {
+        self.bytes.extend_from_slice(other.as_bytes())
+    }
+
+    /// Append a WTF-8 slice at the end of the string.
+    ///
+    /// This replaces newly paired surrogates at the boundary
+    /// with a supplementary code point,
+    /// like concatenating ill-formed UTF-16 strings effectively would.
+    #[inline]
+    pub fn push_wtf8(&mut self, other: &Wtf8) {
+        match (
+            (*self).final_lead_surrogate(),
+            other.initial_trail_surrogate(),
+        ) {
+            // Replace newly paired surrogates by a supplementary code point.
+            (Some(lead), Some(trail)) => {
+                let len_without_lead_surrogate = self.len() - 3;
+                self.bytes.truncate(len_without_lead_surrogate);
+                let other_without_trail_surrogate = &other.bytes[3..];
+                // 4 bytes for the supplementary code point
+                self.bytes.reserve(4 + other_without_trail_surrogate.len());
+                self.push_char(decode_surrogate_pair(lead, trail));
+                self.bytes.extend_from_slice(other_without_trail_surrogate);
+            }
+            _ => self.bytes.extend_from_slice(&other.bytes),
+        }
+    }
+
+    /// Append a Unicode scalar value at the end of the string.
+    #[inline]
+    pub fn push_char(&mut self, c: char) {
+        not_quite_std::push_code_point(self, CodePoint::from_char(c))
+    }
+
+    /// Append a code point at the end of the string.
+    ///
+    /// This replaces newly paired surrogates at the boundary
+    /// with a supplementary code point,
+    /// like concatenating ill-formed UTF-16 strings effectively would.
+    #[inline]
+    pub fn push(&mut self, code_point: CodePoint) {
+        if let trail @ 0xdc00..=0xdfff = code_point.to_u32() {
+            if let Some(lead) = (*self).final_lead_surrogate() {
+                let len_without_lead_surrogate = self.len() - 3;
+                self.bytes.truncate(len_without_lead_surrogate);
+                self.push_char(decode_surrogate_pair(lead, trail as u16));
+                return;
+            }
+        }
+
+        // No newly paired surrogates at the boundary.
+        not_quite_std::push_code_point(self, code_point)
+    }
+
+    /// Shortens a string to the specified length.
+    ///
+    /// # Failure
+    ///
+    /// Fails if `new_len` > current length,
+    /// or if `new_len` is not a code point boundary.
+    #[inline]
+    pub fn truncate(&mut self, new_len: usize) {
+        assert!(not_quite_std::is_code_point_boundary(self, new_len));
+        self.bytes.truncate(new_len)
+    }
+
+    /// Consume the WTF-8 string and try to convert it to UTF-8.
+    ///
+    /// This does not copy the data.
+    ///
+    /// If the contents are not well-formed UTF-8
+    /// (that is, if the string contains surrogates),
+    /// the original WTF-8 string is returned instead.
+    pub fn into_string(self) -> Result<String, Wtf8Buf> {
+        match self.next_surrogate(0) {
+            None => Ok(unsafe { String::from_utf8_unchecked(self.bytes) }),
+            Some(_) => Err(self),
+        }
+    }
+
+    /// Consume the WTF-8 string and convert it lossily to UTF-8.
+    ///
+    /// This does not copy the data (but may overwrite parts of it in place).
+    ///
+    /// Surrogates are replaced with `"\u{FFFD}"` (the replacement character
+    /// “�”)
+    pub fn into_string_lossy(mut self) -> String {
+        let mut pos = 0;
+        loop {
+            match self.next_surrogate(pos) {
+                Some((surrogate_pos, _)) => {
+                    pos = surrogate_pos + 3;
+                    self.bytes[surrogate_pos..pos].copy_from_slice(UTF8_REPLACEMENT_CHARACTER);
+                }
+                None => return unsafe { String::from_utf8_unchecked(self.bytes) },
+            }
+        }
+    }
+}
+
+impl From<&str> for Wtf8Buf {
+    /// Create a WTF-8 string from an UTF-8 `&str` slice.
+    ///
+    /// This copies the content of the slice.
+    ///
+    /// Since WTF-8 is a superset of UTF-8, this always succeeds.
+    #[inline]
+    fn from(s: &str) -> Wtf8Buf {
+        Wtf8Buf {
+            bytes: s.as_bytes().to_vec(),
+        }
+    }
+}
+
+impl Wtf8Buf {
+    pub const fn as_bytes(&self) -> &[u8] {
+        self.bytes.as_slice()
+    }
+}
+
+/// Create a new WTF-8 string from an iterator of code points.
+///
+/// This replaces surrogate code point pairs with supplementary code points,
+/// like concatenating ill-formed UTF-16 strings effectively would.
+impl FromIterator<CodePoint> for Wtf8Buf {
+    fn from_iter<T: IntoIterator<Item = CodePoint>>(iterable: T) -> Wtf8Buf {
+        let mut string = Wtf8Buf::new();
+        string.extend(iterable);
+        string
+    }
+}
+
+/// Append code points from an iterator to the string.
+///
+/// This replaces surrogate code point pairs with supplementary code points,
+/// like concatenating ill-formed UTF-16 strings effectively would.
+impl Extend<CodePoint> for Wtf8Buf {
+    fn extend<T: IntoIterator<Item = CodePoint>>(&mut self, iterable: T) {
+        let iterator = iterable.into_iter();
+        let (low, _high) = iterator.size_hint();
+        // Lower bound of one byte per code point (ASCII only)
+        self.bytes.reserve(low);
+        for code_point in iterator {
+            self.push(code_point);
+        }
+    }
+}
+
+/// A borrowed slice of well-formed WTF-8 data.
+///
+/// Similar to `&str`, but can additionally contain surrogate code points
+/// if they’re not in a surrogate pair.
+pub struct Wtf8 {
+    bytes: [u8],
+}
+
+// FIXME: https://github.com/rust-lang/rust/issues/18805
+impl PartialEq for Wtf8 {
+    fn eq(&self, other: &Wtf8) -> bool {
+        self.bytes.eq(&other.bytes)
+    }
+}
+
+// FIXME: https://github.com/rust-lang/rust/issues/18805
+impl Eq for Wtf8 {}
+
+// FIXME: https://github.com/rust-lang/rust/issues/18738
+#[allow(clippy::non_canonical_partial_ord_impl)]
+impl PartialOrd for Wtf8 {
+    #[inline]
+    fn partial_cmp(&self, other: &Wtf8) -> Option<Ordering> {
+        self.bytes.partial_cmp(&other.bytes)
+    }
+
+    #[inline]
+    fn lt(&self, other: &Wtf8) -> bool {
+        self.bytes.lt(&other.bytes)
+    }
+
+    #[inline]
+    fn le(&self, other: &Wtf8) -> bool {
+        self.bytes.le(&other.bytes)
+    }
+
+    #[inline]
+    fn gt(&self, other: &Wtf8) -> bool {
+        self.bytes.gt(&other.bytes)
+    }
+
+    #[inline]
+    fn ge(&self, other: &Wtf8) -> bool {
+        self.bytes.ge(&other.bytes)
+    }
+}
+
+// FIXME: https://github.com/rust-lang/rust/issues/18738
+impl Ord for Wtf8 {
+    #[inline]
+    fn cmp(&self, other: &Wtf8) -> Ordering {
+        self.bytes.cmp(&other.bytes)
+    }
+}
+
+/// Format the slice with double quotes,
+/// and surrogates as `\u` followed by four hexadecimal digits.
+/// Example: `"a\u{D800}"` for a slice with code points [U+0061, U+D800]
+impl fmt::Debug for Wtf8 {
+    fn fmt(&self, formatter: &mut fmt::Formatter) -> Result<(), fmt::Error> {
+        formatter.write_str("\"")?;
+        let mut pos = 0;
+        loop {
+            match self.next_surrogate(pos) {
+                None => break,
+                Some((surrogate_pos, surrogate)) => {
+                    formatter.write_str(unsafe {
+                        str::from_utf8_unchecked(&self.bytes[pos..surrogate_pos])
+                    })?;
+                    write!(formatter, "\\u{{{surrogate:X}}}")?;
+                    pos = surrogate_pos + 3;
+                }
+            }
+        }
+        formatter.write_str(unsafe { str::from_utf8_unchecked(&self.bytes[pos..]) })?;
+        formatter.write_str("\"")
+    }
+}
+
+impl Wtf8 {
+    /// Create a WTF-8 slice from a UTF-8 `&str` slice.
+    ///
+    /// Since WTF-8 is a superset of UTF-8, this always succeeds.
+    #[inline]
+    pub fn from(value: &str) -> &Wtf8 {
+        unsafe { transmute(value.as_bytes()) }
+    }
+
+    /// Return the length, in WTF-8 bytes.
+    #[inline]
+    pub fn len(&self) -> usize {
+        self.bytes.len()
+    }
+
+    #[inline]
+    #[must_use]
+    pub fn is_empty(&self) -> bool {
+        self.len() == 0
+    }
+
+    /// Return a slice of the given string for the byte range [`begin`..`end`).
+    ///
+    /// # Failure
+    ///
+    /// Fails when `begin` and `end` do not point to code point boundaries,
+    /// or point beyond the end of the string.
+    #[inline]
+    pub fn slice(&self, begin: usize, end: usize) -> &Wtf8 {
+        // is_code_point_boundary checks that the index is in [0, .len()]
+        if begin <= end
+            && not_quite_std::is_code_point_boundary(self, begin)
+            && not_quite_std::is_code_point_boundary(self, end)
+        {
+            unsafe { not_quite_std::slice_unchecked(self, begin, end) }
+        } else {
+            not_quite_std::slice_error_fail(self, begin, end)
+        }
+    }
+
+    /// Return a slice of the given string from byte `begin` to its end.
+    ///
+    /// # Failure
+    ///
+    /// Fails when `begin` is not at a code point boundary,
+    /// or is beyond the end of the string.
+    #[inline]
+    pub fn slice_from(&self, begin: usize) -> &Wtf8 {
+        // is_code_point_boundary checks that the index is in [0, .len()]
+        if not_quite_std::is_code_point_boundary(self, begin) {
+            unsafe { not_quite_std::slice_unchecked(self, begin, self.len()) }
+        } else {
+            not_quite_std::slice_error_fail(self, begin, self.len())
+        }
+    }
+
+    /// Return a slice of the given string from its beginning to byte `end`.
+    ///
+    /// # Failure
+    ///
+    /// Fails when `end` is not at a code point boundary,
+    /// or is beyond the end of the string.
+    #[inline]
+    pub fn slice_to(&self, end: usize) -> &Wtf8 {
+        // is_code_point_boundary checks that the index is in [0, .len()]
+        if not_quite_std::is_code_point_boundary(self, end) {
+            unsafe { not_quite_std::slice_unchecked(self, 0, end) }
+        } else {
+            not_quite_std::slice_error_fail(self, 0, end)
+        }
+    }
+
+    /// Return the code point at `position` if it is in the ASCII range,
+    /// or `b'\xFF' otherwise.
+    ///
+    /// # Failure
+    ///
+    /// Fails if `position` is beyond the end of the string.
+    #[inline]
+    pub fn ascii_byte_at(&self, position: usize) -> u8 {
+        match self.bytes[position] {
+            ascii_byte @ 0x00..=0x7f => ascii_byte,
+            _ => 0xff,
+        }
+    }
+
+    /// Return an iterator for the string’s code points.
+    #[inline]
+    pub fn code_points(&self) -> Wtf8CodePoints {
+        Wtf8CodePoints {
+            bytes: self.bytes.iter(),
+        }
+    }
+
+    /// Try to convert the string to UTF-8 and return a `&str` slice.
+    ///
+    /// Return `None` if the string contains surrogates.
+    ///
+    /// This does not copy the data.
+    #[inline]
+    pub fn as_str(&self) -> Option<&str> {
+        // Well-formed WTF-8 is also well-formed UTF-8
+        // if and only if it contains no surrogate.
+        match self.next_surrogate(0) {
+            None => Some(unsafe { str::from_utf8_unchecked(&self.bytes) }),
+            Some(_) => None,
+        }
+    }
+
+    /// Lossily convert the string to UTF-8.
+    /// Return an UTF-8 `&str` slice if the contents are well-formed in UTF-8.
+    ///
+    /// Surrogates are replaced with `"\u{FFFD}"` (the replacement character
+    /// “�”).
+    ///
+    /// This only copies the data if necessary (if it contains any surrogate).
+    pub fn to_string_lossy(&self) -> Cow<str> {
+        let surrogate_pos = match self.next_surrogate(0) {
+            None => return Cow::Borrowed(unsafe { str::from_utf8_unchecked(&self.bytes) }),
+            Some((pos, _)) => pos,
+        };
+        let wtf8_bytes = &self.bytes;
+        let mut utf8_bytes = Vec::with_capacity(self.len());
+        utf8_bytes.extend_from_slice(&wtf8_bytes[..surrogate_pos]);
+        utf8_bytes.extend_from_slice(UTF8_REPLACEMENT_CHARACTER);
+        let mut pos = surrogate_pos + 3;
+        loop {
+            match self.next_surrogate(pos) {
+                Some((surrogate_pos, _)) => {
+                    utf8_bytes.extend_from_slice(&wtf8_bytes[pos..surrogate_pos]);
+                    utf8_bytes.extend_from_slice(UTF8_REPLACEMENT_CHARACTER);
+                    pos = surrogate_pos + 3;
+                }
+                None => {
+                    utf8_bytes.extend_from_slice(&wtf8_bytes[pos..]);
+                    return Cow::Owned(unsafe { String::from_utf8_unchecked(utf8_bytes) });
+                }
+            }
+        }
+    }
+
+    /// Convert the WTF-8 string to potentially ill-formed UTF-16
+    /// and return an iterator of 16-bit code units.
+    ///
+    /// This is lossless:
+    /// calling `Wtf8Buf::from_ill_formed_utf16` on the resulting code units
+    /// would always return the original WTF-8 string.
+    #[inline]
+    pub fn to_ill_formed_utf16(&self) -> IllFormedUtf16CodeUnits {
+        IllFormedUtf16CodeUnits {
+            code_points: self.code_points(),
+            extra: 0,
+        }
+    }
+
+    #[inline]
+    fn next_surrogate(&self, mut pos: usize) -> Option<(usize, u16)> {
+        let mut iter = self.bytes[pos..].iter();
+        loop {
+            let b = match iter.next() {
+                None => return None,
+                Some(&b) => b,
+            };
+            if b < 0x80 {
+                pos += 1;
+            } else if b < 0xe0 {
+                iter.next();
+                pos += 2;
+            } else if b == 0xed {
+                match (iter.next(), iter.next()) {
+                    (Some(&b2), Some(&b3)) if b2 >= 0xa0 => {
+                        return Some((pos, decode_surrogate(b2, b3)))
+                    }
+                    _ => pos += 3,
+                }
+            } else if b < 0xf0 {
+                iter.next();
+                iter.next();
+                pos += 3;
+            } else {
+                iter.next();
+                iter.next();
+                iter.next();
+                pos += 4;
+            }
+        }
+    }
+
+    #[inline]
+    fn final_lead_surrogate(&self) -> Option<u16> {
+        let len = self.len();
+        if len < 3 {
+            return None;
+        }
+        let seq = &self.bytes[len - 3..];
+        if seq[0] == 0xed && 0xa0 <= seq[1] && seq[1] <= 0xaf {
+            Some(decode_surrogate(seq[1], seq[2]))
+        } else {
+            None
+        }
+    }
+
+    #[inline]
+    fn initial_trail_surrogate(&self) -> Option<u16> {
+        let len = self.len();
+        if len < 3 {
+            return None;
+        }
+        let seq = &self.bytes[..3];
+        if seq[0] == 0xed && 0xb0 <= seq[1] && seq[1] <= 0xbf {
+            Some(decode_surrogate(seq[1], seq[2]))
+        } else {
+            None
+        }
+    }
+}
+
+#[inline]
+fn decode_surrogate(second_byte: u8, third_byte: u8) -> u16 {
+    // The first byte is assumed to be 0xED
+    0xd800 | (second_byte as u16 & 0x3f) << 6 | third_byte as u16 & 0x3f
+}
+
+#[inline]
+fn decode_surrogate_pair(lead: u16, trail: u16) -> char {
+    let code_point = 0x10000 + (((lead as u32 - 0xd800) << 10) | (trail as u32 - 0xdc00));
+    unsafe { char::from_u32_unchecked(code_point) }
+}
+
+/// Iterator for the code points of a WTF-8 string.
+///
+/// Created with the method `.code_points()`.
+#[derive(Clone)]
+pub struct Wtf8CodePoints<'a> {
+    bytes: slice::Iter<'a, u8>,
+}
+
+impl<'a> Iterator for Wtf8CodePoints<'a> {
+    type Item = CodePoint;
+
+    #[inline]
+    fn next(&mut self) -> Option<CodePoint> {
+        not_quite_std::next_code_point(&mut self.bytes)
+            .map(|value| unsafe { CodePoint::from_u32_unchecked(value) })
+    }
+
+    #[inline]
+    fn size_hint(&self) -> (usize, Option<usize>) {
+        let (len, _) = self.bytes.size_hint();
+        (len.saturating_add(3) / 4, Some(len))
+    }
+}
+
+#[derive(Clone)]
+pub struct IllFormedUtf16CodeUnits<'a> {
+    code_points: Wtf8CodePoints<'a>,
+    extra: u16,
+}
+
+impl<'a> Iterator for IllFormedUtf16CodeUnits<'a> {
+    type Item = u16;
+
+    #[inline]
+    fn next(&mut self) -> Option<u16> {
+        not_quite_std::next_utf16_code_unit(self)
+    }
+
+    #[inline]
+    fn size_hint(&self) -> (usize, Option<usize>) {
+        let (low, high) = self.code_points.size_hint();
+        // every code point gets either one u16 or two u16,
+        // so this iterator is between 1 or 2 times as
+        // long as the underlying iterator.
+        (low, high.and_then(|n| n.checked_mul(2)))
+    }
+}
+
+impl PartialEq<&'_ Wtf8> for Wtf8Buf {
+    fn eq(&self, other: &&Wtf8) -> bool {
+        **self == **other
+    }
+}
+
+impl PartialEq<Wtf8Buf> for &'_ Wtf8 {
+    fn eq(&self, other: &Wtf8Buf) -> bool {
+        **self == **other
+    }
+}
+
+impl hash::Hash for CodePoint {
+    #[inline]
+    fn hash<H: hash::Hasher>(&self, state: &mut H) {
+        self.value.hash(state)
+    }
+}
+
+impl hash::Hash for Wtf8Buf {
+    #[inline]
+    fn hash<H: hash::Hasher>(&self, state: &mut H) {
+        Wtf8::hash(self, state)
+    }
+}
+
+impl hash::Hash for Wtf8 {
+    #[inline]
+    fn hash<H: hash::Hasher>(&self, state: &mut H) {
+        state.write(&self.bytes);
+        0xfeu8.hash(state)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use alloc::{format, vec};
+    use core::mem::transmute;
+
+    use super::*;
+
+    #[test]
+    fn code_point_from_u32() {
+        assert!(CodePoint::from_u32(0).is_some());
+        assert!(CodePoint::from_u32(0xd800).is_some());
+        assert!(CodePoint::from_u32(0x10ffff).is_some());
+        assert!(CodePoint::from_u32(0x110000).is_none());
+    }
+
+    #[test]
+    fn code_point_to_u32() {
+        fn c(value: u32) -> CodePoint {
+            CodePoint::from_u32(value).unwrap()
+        }
+        assert_eq!(c(0).to_u32(), 0);
+        assert_eq!(c(0xd800).to_u32(), 0xd800);
+        assert_eq!(c(0x10ffff).to_u32(), 0x10ffff);
+    }
+
+    #[test]
+    fn code_point_from_char() {
+        assert_eq!(CodePoint::from_char('a').to_u32(), 0x61);
+        assert_eq!(CodePoint::from_char('💩').to_u32(), 0x1f4a9);
+    }
+
+    #[test]
+    fn code_point_to_string() {
+        assert_eq!(format!("{:?}", CodePoint::from_char('a')), "U+0061");
+        assert_eq!(format!("{:?}", CodePoint::from_char('💩')), "U+1F4A9");
+    }
+
+    #[test]
+    fn code_point_to_char() {
+        fn c(value: u32) -> CodePoint {
+            CodePoint::from_u32(value).unwrap()
+        }
+        assert_eq!(c(0x61).to_char(), Some('a'));
+        assert_eq!(c(0x1f4a9).to_char(), Some('💩'));
+        assert_eq!(c(0xd800).to_char(), None);
+    }
+
+    #[test]
+    fn code_point_to_char_lossy() {
+        fn c(value: u32) -> CodePoint {
+            CodePoint::from_u32(value).unwrap()
+        }
+        assert_eq!(c(0x61).to_char_lossy(), 'a');
+        assert_eq!(c(0x1f4a9).to_char_lossy(), '💩');
+        assert_eq!(c(0xd800).to_char_lossy(), '\u{FFFD}');
+    }
+
+    #[test]
+    fn wtf8buf_new() {
+        assert_eq!(Wtf8Buf::new().bytes, b"");
+    }
+
+    #[test]
+    fn wtf8buf_from() {
+        assert_eq!(Wtf8Buf::from("").bytes, b"");
+        assert_eq!(Wtf8Buf::from("aé 💩").bytes, b"a\xC3\xA9 \xF0\x9F\x92\xA9");
+    }
+
+    #[test]
+    fn wtf8buf_froming() {
+        assert_eq!(Wtf8Buf::froming(String::from("")).bytes, b"");
+        assert_eq!(
+            Wtf8Buf::froming(String::from("aé 💩")).bytes,
+            b"a\xC3\xA9 \xF0\x9F\x92\xA9"
+        );
+    }
+
+    #[test]
+    fn wtf8buf_from_ill_formed_utf16() {
+        assert_eq!(Wtf8Buf::from_ill_formed_utf16(&[]).bytes, b"");
+        assert_eq!(
+            Wtf8Buf::from_ill_formed_utf16(&[0x61, 0xe9, 0x20, 0xd83d, 0xd83d, 0xdca9]).bytes,
+            b"a\xC3\xA9 \xED\xA0\xBD\xF0\x9F\x92\xA9"
+        );
+    }
+
+    #[test]
+    fn wtf8buf_push_str() {
+        let mut string = Wtf8Buf::new();
+        assert_eq!(string.bytes, b"");
+        string.push_str("aé 💩");
+        assert_eq!(string.bytes, b"a\xC3\xA9 \xF0\x9F\x92\xA9");
+    }
+
+    #[test]
+    fn wtf8buf_push_char() {
+        let mut string = Wtf8Buf::from("aé ");
+        assert_eq!(string.bytes, b"a\xC3\xA9 ");
+        string.push_char('💩');
+        assert_eq!(string.bytes, b"a\xC3\xA9 \xF0\x9F\x92\xA9");
+    }
+
+    #[test]
+    fn wtf8buf_push() {
+        let mut string = Wtf8Buf::from("aé ");
+        assert_eq!(string.bytes, b"a\xC3\xA9 ");
+        string.push(CodePoint::from_char('💩'));
+        assert_eq!(string.bytes, b"a\xC3\xA9 \xF0\x9F\x92\xA9");
+
+        fn c(value: u32) -> CodePoint {
+            CodePoint::from_u32(value).unwrap()
+        }
+
+        let mut string = Wtf8Buf::new();
+        string.push(c(0xd83d)); // lead
+        string.push(c(0xdca9)); // trail
+        assert_eq!(string.bytes, b"\xF0\x9F\x92\xA9"); // Magic!
+
+        let mut string = Wtf8Buf::new();
+        string.push(c(0xd83d)); // lead
+        string.push(c(0x20)); // not surrogate
+        string.push(c(0xdca9)); // trail
+        assert_eq!(string.bytes, b"\xED\xA0\xBD \xED\xB2\xA9");
+
+        let mut string = Wtf8Buf::new();
+        string.push(c(0xd800)); // lead
+        string.push(c(0xdbff)); // lead
+        assert_eq!(string.bytes, b"\xED\xA0\x80\xED\xAF\xBF");
+
+        let mut string = Wtf8Buf::new();
+        string.push(c(0xd800)); // lead
+        string.push(c(0xe000)); // not surrogate
+        assert_eq!(string.bytes, b"\xED\xA0\x80\xEE\x80\x80");
+
+        let mut string = Wtf8Buf::new();
+        string.push(c(0xd7ff)); // not surrogate
+        string.push(c(0xdc00)); // trail
+        assert_eq!(string.bytes, b"\xED\x9F\xBF\xED\xB0\x80");
+
+        let mut string = Wtf8Buf::new();
+        string.push(c(0x61)); // not surrogate, < 3 bytes
+        string.push(c(0xdc00)); // trail
+        assert_eq!(string.bytes, b"\x61\xED\xB0\x80");
+
+        let mut string = Wtf8Buf::new();
+        string.push(c(0xdc00)); // trail
+        assert_eq!(string.bytes, b"\xED\xB0\x80");
+    }
+
+    #[test]
+    fn wtf8buf_push_wtf8() {
+        let mut string = Wtf8Buf::from("aé");
+        assert_eq!(string.bytes, b"a\xC3\xA9");
+        string.push_wtf8(Wtf8::from(" 💩"));
+        assert_eq!(string.bytes, b"a\xC3\xA9 \xF0\x9F\x92\xA9");
+
+        fn w(value: &[u8]) -> &Wtf8 {
+            unsafe { transmute(value) }
+        }
+
+        let mut string = Wtf8Buf::new();
+        string.push_wtf8(w(b"\xED\xA0\xBD")); // lead
+        string.push_wtf8(w(b"\xED\xB2\xA9")); // trail
+        assert_eq!(string.bytes, b"\xF0\x9F\x92\xA9"); // Magic!
+
+        let mut string = Wtf8Buf::new();
+        string.push_wtf8(w(b"\xED\xA0\xBD")); // lead
+        string.push_wtf8(w(b" ")); // not surrogate
+        string.push_wtf8(w(b"\xED\xB2\xA9")); // trail
+        assert_eq!(string.bytes, b"\xED\xA0\xBD \xED\xB2\xA9");
+
+        let mut string = Wtf8Buf::new();
+        string.push_wtf8(w(b"\xED\xA0\x80")); // lead
+        string.push_wtf8(w(b"\xED\xAF\xBF")); // lead
+        assert_eq!(string.bytes, b"\xED\xA0\x80\xED\xAF\xBF");
+
+        let mut string = Wtf8Buf::new();
+        string.push_wtf8(w(b"\xED\xA0\x80")); // lead
+        string.push_wtf8(w(b"\xEE\x80\x80")); // not surrogate
+        assert_eq!(string.bytes, b"\xED\xA0\x80\xEE\x80\x80");
+
+        let mut string = Wtf8Buf::new();
+        string.push_wtf8(w(b"\xED\x9F\xBF")); // not surrogate
+        string.push_wtf8(w(b"\xED\xB0\x80")); // trail
+        assert_eq!(string.bytes, b"\xED\x9F\xBF\xED\xB0\x80");
+
+        let mut string = Wtf8Buf::new();
+        string.push_wtf8(w(b"a")); // not surrogate, < 3 bytes
+        string.push_wtf8(w(b"\xED\xB0\x80")); // trail
+        assert_eq!(string.bytes, b"\x61\xED\xB0\x80");
+
+        let mut string = Wtf8Buf::new();
+        string.push_wtf8(w(b"\xED\xB0\x80")); // trail
+        assert_eq!(string.bytes, b"\xED\xB0\x80");
+    }
+
+    #[test]
+    fn wtf8buf_truncate() {
+        let mut string = Wtf8Buf::from("aé");
+        string.truncate(1);
+        assert_eq!(string.bytes, b"a");
+    }
+
+    #[test]
+    #[should_panic]
+    fn wtf8buf_truncate_fail_code_point_boundary() {
+        let mut string = Wtf8Buf::from("aé");
+        string.truncate(2);
+    }
+
+    #[test]
+    #[should_panic]
+    fn wtf8buf_truncate_fail_longer() {
+        let mut string = Wtf8Buf::from("aé");
+        string.truncate(4);
+    }
+
+    #[test]
+    fn wtf8buf_into_string() {
+        let mut string = Wtf8Buf::from("aé 💩");
+        assert_eq!(string.clone().into_string(), Ok(String::from("aé 💩")));
+        string.push(CodePoint::from_u32(0xd800).unwrap());
+        assert_eq!(string.clone().into_string(), Err(string));
+    }
+
+    #[test]
+    fn wtf8buf_into_string_lossy() {
+        let mut string = Wtf8Buf::from("aé 💩");
+        assert_eq!(string.clone().into_string_lossy(), String::from("aé 💩"));
+        string.push(CodePoint::from_u32(0xd800).unwrap());
+        assert_eq!(string.clone().into_string_lossy(), String::from("aé 💩�"));
+    }
+
+    #[test]
+    fn wtf8buf_from_iterator() {
+        fn f(values: &[u32]) -> Wtf8Buf {
+            values
+                .iter()
+                .map(|&c| CodePoint::from_u32(c).unwrap())
+                .collect::<Wtf8Buf>()
+        }
+        assert_eq!(
+            f(&[0x61, 0xe9, 0x20, 0x1f4a9]).bytes,
+            b"a\xC3\xA9 \xF0\x9F\x92\xA9"
+        );
+
+        assert_eq!(f(&[0xd83d, 0xdca9]).bytes, b"\xF0\x9F\x92\xA9"); // Magic!
+        assert_eq!(
+            f(&[0xd83d, 0x20, 0xdca9]).bytes,
+            b"\xED\xA0\xBD \xED\xB2\xA9"
+        );
+        assert_eq!(f(&[0xd800, 0xdbff]).bytes, b"\xED\xA0\x80\xED\xAF\xBF");
+        assert_eq!(f(&[0xd800, 0xe000]).bytes, b"\xED\xA0\x80\xEE\x80\x80");
+        assert_eq!(f(&[0xd7ff, 0xdc00]).bytes, b"\xED\x9F\xBF\xED\xB0\x80");
+        assert_eq!(f(&[0x61, 0xdc00]).bytes, b"\x61\xED\xB0\x80");
+        assert_eq!(f(&[0xdc00]).bytes, b"\xED\xB0\x80");
+    }
+
+    #[test]
+    fn wtf8buf_extend() {
+        fn e(initial: &[u32], extended: &[u32]) -> Wtf8Buf {
+            fn c(value: &u32) -> CodePoint {
+                CodePoint::from_u32(*value).unwrap()
+            }
+            let mut string = initial.iter().map(c).collect::<Wtf8Buf>();
+            string.extend(extended.iter().map(c));
+            string
+        }
+
+        assert_eq!(
+            e(&[0x61, 0xe9], &[0x20, 0x1f4a9]).bytes,
+            b"a\xC3\xA9 \xF0\x9F\x92\xA9"
+        );
+
+        assert_eq!(e(&[0xd83d], &[0xdca9]).bytes, b"\xF0\x9F\x92\xA9"); // Magic!
+        assert_eq!(
+            e(&[0xd83d, 0x20], &[0xdca9]).bytes,
+            b"\xED\xA0\xBD \xED\xB2\xA9"
+        );
+        assert_eq!(e(&[0xd800], &[0xdbff]).bytes, b"\xED\xA0\x80\xED\xAF\xBF");
+        assert_eq!(e(&[0xd800], &[0xe000]).bytes, b"\xED\xA0\x80\xEE\x80\x80");
+        assert_eq!(e(&[0xd7ff], &[0xdc00]).bytes, b"\xED\x9F\xBF\xED\xB0\x80");
+        assert_eq!(e(&[0x61], &[0xdc00]).bytes, b"\x61\xED\xB0\x80");
+        assert_eq!(e(&[], &[0xdc00]).bytes, b"\xED\xB0\x80");
+    }
+
+    #[test]
+    fn wtf8buf_debug() {
+        let mut string = Wtf8Buf::from("aé 💩");
+        string.push(CodePoint::from_u32(0xd800).unwrap());
+        assert_eq!(format!("{string:?}"), r#""aé 💩\u{D800}""#);
+    }
+
+    #[test]
+    fn wtf8buf_as_slice() {
+        assert_eq!(Wtf8Buf::from("aé"), Wtf8::from("aé"));
+    }
+
+    #[test]
+    fn wtf8_debug() {
+        let mut string = Wtf8Buf::from("aé 💩");
+        string.push(CodePoint::from_u32(0xd800).unwrap());
+        assert_eq!(format!("{:?}", &*string), r#""aé 💩\u{D800}""#);
+    }
+
+    #[test]
+    fn wtf8_from() {
+        assert_eq!(&Wtf8::from("").bytes, b"");
+        assert_eq!(&Wtf8::from("aé 💩").bytes, b"a\xC3\xA9 \xF0\x9F\x92\xA9");
+    }
+
+    #[test]
+    fn wtf8_len() {
+        assert_eq!(Wtf8::from("").len(), 0);
+        assert_eq!(Wtf8::from("aé 💩").len(), 8);
+    }
+
+    #[test]
+    fn wtf8_slice() {
+        assert_eq!(&Wtf8::from("aé 💩").slice(1, 4).bytes, b"\xC3\xA9 ");
+    }
+
+    #[test]
+    #[should_panic]
+    fn wtf8_slice_not_code_point_boundary() {
+        Wtf8::from("aé 💩").slice(2, 4);
+    }
+
+    #[test]
+    fn wtf8_slice_from() {
+        assert_eq!(
+            &Wtf8::from("aé 💩").slice_from(1).bytes,
+            b"\xC3\xA9 \xF0\x9F\x92\xA9"
+        );
+    }
+
+    #[test]
+    #[should_panic]
+    fn wtf8_slice_from_not_code_point_boundary() {
+        Wtf8::from("aé 💩").slice_from(2);
+    }
+
+    #[test]
+    fn wtf8_slice_to() {
+        assert_eq!(&Wtf8::from("aé 💩").slice_to(4).bytes, b"a\xC3\xA9 ");
+    }
+
+    #[test]
+    #[should_panic]
+    fn wtf8_slice_to_not_code_point_boundary() {
+        Wtf8::from("aé 💩").slice_from(5);
+    }
+
+    #[test]
+    fn wtf8_ascii_byte_at() {
+        let slice = Wtf8::from("aé 💩");
+        assert_eq!(slice.ascii_byte_at(0), b'a');
+        assert_eq!(slice.ascii_byte_at(1), b'\xFF');
+        assert_eq!(slice.ascii_byte_at(2), b'\xFF');
+        assert_eq!(slice.ascii_byte_at(3), b' ');
+        assert_eq!(slice.ascii_byte_at(4), b'\xFF');
+    }
+
+    #[test]
+    fn wtf8_code_points() {
+        fn c(value: u32) -> CodePoint {
+            CodePoint::from_u32(value).unwrap()
+        }
+        fn cp(string: &Wtf8Buf) -> Vec<Option<char>> {
+            string
+                .code_points()
+                .map(|c| c.to_char())
+                .collect::<Vec<_>>()
+        }
+        let mut string = Wtf8Buf::from("é ");
+        assert_eq!(cp(&string), vec![Some('é'), Some(' ')]);
+        string.push(c(0xd83d));
+        assert_eq!(cp(&string), vec![Some('é'), Some(' '), None]);
+        string.push(c(0xdca9));
+        assert_eq!(cp(&string), vec![Some('é'), Some(' '), Some('💩')]);
+    }
+
+    #[test]
+    fn wtf8_as_str() {
+        assert_eq!(Wtf8::from("").as_str(), Some(""));
+        assert_eq!(Wtf8::from("aé 💩").as_str(), Some("aé 💩"));
+        let mut string = Wtf8Buf::new();
+        string.push(CodePoint::from_u32(0xd800).unwrap());
+        assert_eq!(string.as_str(), None);
+    }
+
+    #[test]
+    fn wtf8_to_string_lossy() {
+        assert_eq!(Wtf8::from("").to_string_lossy(), Cow::Borrowed(""));
+        assert_eq!(
+            Wtf8::from("aé 💩").to_string_lossy(),
+            Cow::Borrowed("aé 💩")
+        );
+        let mut string = Wtf8Buf::from("aé 💩");
+        string.push(CodePoint::from_u32(0xd800).unwrap());
+        assert_eq!(string.to_string_lossy(), {
+            let o: Cow<str> = Cow::Owned(String::from("aé 💩�"));
+            o
+        });
+    }
+
+    #[test]
+    fn wtf8_to_ill_formed_utf16() {
+        let mut string = Wtf8Buf::from("aé ");
+        string.push(CodePoint::from_u32(0xd83d).unwrap());
+        string.push_char('💩');
+        assert_eq!(
+            string.to_ill_formed_utf16().collect::<Vec<_>>(),
+            vec![0x61, 0xe9, 0x20, 0xd83d, 0xd83d, 0xdca9]
+        );
+    }
+}
diff --git a/crates/hstr/src/wtf8/not_quite_std.rs b/crates/hstr/src/wtf8/not_quite_std.rs
new file mode 100644
index 000000000000..f7482be8d83e
--- /dev/null
+++ b/crates/hstr/src/wtf8/not_quite_std.rs
@@ -0,0 +1,250 @@
+//! The code in this module is copied from Rust standard library
+//! (the `std` crate and crates it is a facade for)
+//! at commit 16d80de231abb2b1756f3951ffd4776d681035eb,
+//! with the signature changed to use `Wtf8Buf`, `Wtf8`, and `CodePoint`
+//! instead of `String`, `&str`, and `char`.
+//!
+//! FIXME: if and when this is moved into the standard library,
+//! try to avoid the code duplication.
+//! Maybe by having private generic code that is monomorphized to UTF-8 and
+//! WTF-8?
+
+use core::{char, mem, slice};
+
+use super::{CodePoint, IllFormedUtf16CodeUnits, Wtf8, Wtf8Buf};
+
+// UTF-8 ranges and tags for encoding characters
+// Copied from 48d5fe9ec560b53b1f5069219b0d62015e1de5ba^:src/libcore/char.rs
+const TAG_CONT: u8 = 0b1000_0000;
+const TAG_TWO_B: u8 = 0b1100_0000;
+const TAG_THREE_B: u8 = 0b1110_0000;
+const TAG_FOUR_B: u8 = 0b1111_0000;
+const MAX_ONE_B: u32 = 0x80;
+const MAX_TWO_B: u32 = 0x800;
+const MAX_THREE_B: u32 = 0x10000;
+
+/// Copied from 48d5fe9ec560b53b1f5069219b0d62015e1de5ba^:src/libcore/char.rs
+#[inline]
+fn encode_utf8_raw(code: u32, dst: &mut [u8]) -> Option<usize> {
+    // Marked #[inline] to allow llvm optimizing it away
+    if code < MAX_ONE_B && !dst.is_empty() {
+        dst[0] = code as u8;
+        Some(1)
+    } else if code < MAX_TWO_B && dst.len() >= 2 {
+        dst[0] = (code >> 6 & 0x1f) as u8 | TAG_TWO_B;
+        dst[1] = (code & 0x3f) as u8 | TAG_CONT;
+        Some(2)
+    } else if code < MAX_THREE_B && dst.len() >= 3 {
+        dst[0] = (code >> 12 & 0x0f) as u8 | TAG_THREE_B;
+        dst[1] = (code >> 6 & 0x3f) as u8 | TAG_CONT;
+        dst[2] = (code & 0x3f) as u8 | TAG_CONT;
+        Some(3)
+    } else if dst.len() >= 4 {
+        dst[0] = (code >> 18 & 0x07) as u8 | TAG_FOUR_B;
+        dst[1] = (code >> 12 & 0x3f) as u8 | TAG_CONT;
+        dst[2] = (code >> 6 & 0x3f) as u8 | TAG_CONT;
+        dst[3] = (code & 0x3f) as u8 | TAG_CONT;
+        Some(4)
+    } else {
+        None
+    }
+}
+
+/// Copied from 48d5fe9ec560b53b1f5069219b0d62015e1de5ba^:src/libcore/char.rs
+#[inline]
+fn encode_utf16_raw(mut ch: u32, dst: &mut [u16]) -> Option<usize> {
+    // Marked #[inline] to allow llvm optimizing it away
+    if (ch & 0xffff) == ch && !dst.is_empty() {
+        // The BMP falls through (assuming non-surrogate, as it should)
+        dst[0] = ch as u16;
+        Some(1)
+    } else if dst.len() >= 2 {
+        // Supplementary planes break into surrogates.
+        ch -= 0x1_0000;
+        dst[0] = 0xd800 | ((ch >> 10) as u16);
+        dst[1] = 0xdc00 | ((ch as u16) & 0x3ff);
+        Some(2)
+    } else {
+        None
+    }
+}
+
+/// Copied from core::str::next_code_point
+#[inline]
+pub fn next_code_point(bytes: &mut slice::Iter<u8>) -> Option<u32> {
+    // Decode UTF-8
+    let x = match bytes.next() {
+        None => return None,
+        Some(&next_byte) if next_byte < 128 => return Some(next_byte as u32),
+        Some(&next_byte) => next_byte,
+    };
+
+    // Multibyte case follows
+    // Decode from a byte combination out of: [[[x y] z] w]
+    // NOTE: Performance is sensitive to the exact formulation here
+    let init = utf8_first_byte(x, 2);
+    let y = unwrap_or_0(bytes.next());
+    let mut ch = utf8_acc_cont_byte(init, y);
+    if x >= 0xe0 {
+        // [[x y z] w] case
+        // 5th bit in 0xE0 .. 0xEF is always clear, so `init` is still valid
+        let z = unwrap_or_0(bytes.next());
+        let y_z = utf8_acc_cont_byte((y & CONT_MASK) as u32, z);
+        ch = init << 12 | y_z;
+        if x >= 0xf0 {
+            // [x y z w] case
+            // use only the lower 3 bits of `init`
+            let w = unwrap_or_0(bytes.next());
+            ch = (init & 7) << 18 | utf8_acc_cont_byte(y_z, w);
+        }
+    }
+
+    Some(ch)
+}
+
+#[inline]
+fn utf8_first_byte(byte: u8, width: u32) -> u32 {
+    (byte & (0x7f >> width)) as u32
+}
+
+/// Return the value of `ch` updated with continuation byte `byte`.
+#[inline]
+fn utf8_acc_cont_byte(ch: u32, byte: u8) -> u32 {
+    (ch << 6) | (byte & CONT_MASK) as u32
+}
+
+#[inline]
+fn unwrap_or_0(opt: Option<&u8>) -> u8 {
+    match opt {
+        Some(&byte) => byte,
+        None => 0,
+    }
+}
+
+/// Mask of the value bits of a continuation byte
+const CONT_MASK: u8 = 0b0011_1111;
+
+/// Copied from String::push
+/// This does **not** include the WTF-8 concatenation check.
+#[inline]
+pub fn push_code_point(string: &mut Wtf8Buf, code_point: CodePoint) {
+    let cur_len = string.len();
+    // This may use up to 4 bytes.
+    string.reserve(4);
+
+    unsafe {
+        // Attempt to not use an intermediate buffer by just pushing bytes
+        // directly onto this string.
+        let slice = slice::from_raw_parts_mut(string.bytes.as_mut_ptr().add(cur_len), 4);
+        let used = encode_utf8_raw(code_point.to_u32(), slice).unwrap_or(0);
+        string.bytes.set_len(cur_len + used);
+    }
+}
+
+/// Copied from core::str::StrPrelude::is_char_boundary
+#[inline]
+pub fn is_code_point_boundary(slice: &Wtf8, index: usize) -> bool {
+    if index == slice.len() {
+        return true;
+    }
+    match slice.bytes.get(index) {
+        None => false,
+        Some(&b) => !(128u8..192u8).contains(&b),
+    }
+}
+
+/// Copied from core::str::raw::slice_unchecked
+#[inline]
+pub unsafe fn slice_unchecked(s: &Wtf8, begin: usize, end: usize) -> &Wtf8 {
+    mem::transmute(slice::from_raw_parts(
+        s.bytes.as_ptr().add(begin),
+        end - begin,
+    ))
+}
+
+/// Copied from core::str::raw::slice_error_fail
+#[inline(never)]
+pub fn slice_error_fail(s: &Wtf8, begin: usize, end: usize) -> ! {
+    assert!(begin <= end);
+    panic!("index {begin} and/or {end} in {s:?} do not lie on character boundary",);
+}
+
+/// Copied from core::str::Utf16CodeUnits::next
+pub fn next_utf16_code_unit(iter: &mut IllFormedUtf16CodeUnits) -> Option<u16> {
+    if iter.extra != 0 {
+        let tmp = iter.extra;
+        iter.extra = 0;
+        return Some(tmp);
+    }
+
+    let mut buf = [0u16; 2];
+    iter.code_points.next().map(|code_point| {
+        let n = encode_utf16_raw(code_point.to_u32(), &mut buf).unwrap_or(0);
+        if n == 2 {
+            iter.extra = buf[1];
+        }
+        buf[0]
+    })
+}
+
+/// Copied from src/librustc_unicode/char.rs
+pub struct DecodeUtf16<I>
+where
+    I: Iterator<Item = u16>,
+{
+    iter: I,
+    buf: Option<u16>,
+}
+
+/// Copied from src/librustc_unicode/char.rs
+#[inline]
+pub fn decode_utf16<I: IntoIterator<Item = u16>>(iterable: I) -> DecodeUtf16<I::IntoIter> {
+    DecodeUtf16 {
+        iter: iterable.into_iter(),
+        buf: None,
+    }
+}
+
+/// Copied from src/librustc_unicode/char.rs
+impl<I: Iterator<Item = u16>> Iterator for DecodeUtf16<I> {
+    type Item = Result<char, u16>;
+
+    fn next(&mut self) -> Option<Result<char, u16>> {
+        let u = match self.buf.take() {
+            Some(buf) => buf,
+            None => self.iter.next()?,
+        };
+
+        if !(0xd800..=0xdfff).contains(&u) {
+            // not a surrogate
+            Some(Ok(unsafe { char::from_u32_unchecked(u as u32) }))
+        } else if u >= 0xdc00 {
+            // a trailing surrogate
+            Some(Err(u))
+        } else {
+            let u2 = match self.iter.next() {
+                Some(u2) => u2,
+                // eof
+                None => return Some(Err(u)),
+            };
+            if !(0xdc00..=0xdfff).contains(&u2) {
+                // not a trailing surrogate so we're not a valid
+                // surrogate pair, so rewind to redecode u2 next time.
+                self.buf = Some(u2);
+                return Some(Err(u));
+            }
+
+            // all ok, so lets decode it.
+            let c = (((u - 0xd800) as u32) << 10 | (u2 - 0xdc00) as u32) + 0x1_0000;
+            Some(Ok(unsafe { char::from_u32_unchecked(c) }))
+        }
+    }
+
+    #[inline]
+    fn size_hint(&self) -> (usize, Option<usize>) {
+        let (low, high) = self.iter.size_hint();
+        // we could be entirely valid surrogates (2 elements per
+        // char), or entirely non-surrogates (1 element per char)
+        (low / 2, high)
+    }
+}

From ef7476791862370877a3754c20563acba0b17190 Mon Sep 17 00:00:00 2001
From: CPunisher <1343316114@qq.com>
Date: Fri, 12 Sep 2025 18:04:14 +0800
Subject: [PATCH 2/5] Store code point

---
 Cargo.lock                                    |   1 +
 crates/hstr/src/dynamic.rs                    |  31 ++-
 crates/hstr/src/global_store.rs               |   4 +-
 crates/hstr/src/wtf8/mod.rs                   |   5 -
 crates/swc_atoms/src/lib.rs                   |   8 +-
 .../swc_ecma_lexer/src/common/lexer/char.rs   | 103 +---------
 crates/swc_ecma_lexer/src/common/lexer/mod.rs | 191 +++++++-----------
 crates/swc_ecma_lexer/src/lexer/mod.rs        |   4 +-
 crates/swc_ecma_parser/src/lexer/mod.rs       |  20 +-
 crates/swc_ecma_parser/src/lexer/state.rs     |   6 +-
 10 files changed, 115 insertions(+), 258 deletions(-)

diff --git a/Cargo.lock b/Cargo.lock
index 320f44cc3a34..c00182f90ecd 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -2249,6 +2249,7 @@ checksum = "7f24254aa9a54b5c858eaee2f5bccdb46aaf0e486a595ed5fd8f86ba55232a70"
 name = "hstr"
 version = "2.0.1"
 dependencies = [
+ "arrayvec",
  "compact_str",
  "criterion",
  "hashbrown 0.14.5",
diff --git a/crates/hstr/src/dynamic.rs b/crates/hstr/src/dynamic.rs
index ad9244b19416..48a96f9da294 100644
--- a/crates/hstr/src/dynamic.rs
+++ b/crates/hstr/src/dynamic.rs
@@ -70,7 +70,12 @@ impl Default for AtomStore {
 impl AtomStore {
     #[inline(always)]
     pub fn atom<'a>(&mut self, text: impl Into<Cow<'a, str>>) -> Atom {
-        atom_in(self, &text.into())
+        self.atom_raw(text.into().as_bytes())
+    }
+
+    #[inline(always)]
+    pub fn atom_raw(&mut self, text: &[u8]) -> Atom {
+        atom_in(self, text)
     }
 
     fn gc(&mut self) {
@@ -94,7 +99,7 @@ pub fn global_atom_store_gc() {
     });
 }
 
-pub(crate) fn global_atom(text: &str) -> Atom {
+pub(crate) fn global_atom(text: &[u8]) -> Atom {
     GLOBAL_DATA.with(|global| {
         let mut store = global.borrow_mut();
 
@@ -104,7 +109,7 @@ pub(crate) fn global_atom(text: &str) -> Atom {
 
 /// This can create any kind of [Atom], although this lives in the `dynamic`
 /// module.
-fn atom_in<S>(storage: S, text: &str) -> Atom
+fn atom_in<S>(storage: S, text: &[u8]) -> Atom
 where
     S: Storage,
 {
@@ -115,7 +120,7 @@ where
         let tag = INLINE_TAG_INIT | ((len as u8) << LEN_OFFSET);
         let mut unsafe_data = TaggedValue::new_tag(tag);
         unsafe {
-            unsafe_data.data_mut()[..len].copy_from_slice(text.as_bytes());
+            unsafe_data.data_mut()[..len].copy_from_slice(text);
         }
         return Atom { unsafe_data };
     }
@@ -159,31 +164,25 @@ pub(crate) const fn inline_atom(text: &str) -> Option<Atom> {
 }
 
 trait Storage {
-    fn insert_entry(self, text: &str, hash: u64) -> Item;
+    fn insert_entry(self, text: &[u8], hash: u64) -> Item;
 }
 
 impl Storage for &'_ mut AtomStore {
-    fn insert_entry(self, text: &str, hash: u64) -> Item {
+    fn insert_entry(self, text: &[u8], hash: u64) -> Item {
         // If the text is too long, interning is not worth it.
         if text.len() > 512 {
-            return Item(ThinArc::from_header_and_slice(
-                Metadata { hash },
-                text.as_bytes(),
-            ));
+            return Item(ThinArc::from_header_and_slice(Metadata { hash }, text));
         }
 
         let (entry, _) = self
             .data
             .raw_entry_mut()
             .from_hash(hash, |key| {
-                key.header.header.hash == hash && key.slice.eq(text.as_bytes())
+                key.header.header.hash == hash && key.slice.eq(text)
             })
             .or_insert_with(move || {
                 (
-                    Item(ThinArc::from_header_and_slice(
-                        Metadata { hash },
-                        text.as_bytes(),
-                    )),
+                    Item(ThinArc::from_header_and_slice(Metadata { hash }, text)),
                     (),
                 )
             });
@@ -192,7 +191,7 @@ impl Storage for &'_ mut AtomStore {
 }
 
 #[inline(always)]
-fn calc_hash(text: &str) -> u64 {
+fn calc_hash(text: &[u8]) -> u64 {
     let mut hasher = FxHasher::default();
     text.hash(&mut hasher);
     hasher.finish()
diff --git a/crates/hstr/src/global_store.rs b/crates/hstr/src/global_store.rs
index 6136c747b6ee..c1fdee923cc0 100644
--- a/crates/hstr/src/global_store.rs
+++ b/crates/hstr/src/global_store.rs
@@ -6,7 +6,7 @@ macro_rules! direct_from_impl {
     ($T:ty) => {
         impl From<$T> for Atom {
             fn from(s: $T) -> Self {
-                global_atom(&s)
+                global_atom(s.as_bytes())
             }
         }
     };
@@ -18,6 +18,6 @@ direct_from_impl!(String);
 
 impl From<Box<str>> for crate::Atom {
     fn from(s: Box<str>) -> Self {
-        global_atom(&s)
+        global_atom(s.as_bytes())
     }
 }
diff --git a/crates/hstr/src/wtf8/mod.rs b/crates/hstr/src/wtf8/mod.rs
index 43374fdea1a8..57e2ed40eec7 100644
--- a/crates/hstr/src/wtf8/mod.rs
+++ b/crates/hstr/src/wtf8/mod.rs
@@ -122,11 +122,6 @@ impl IntoIterator for CodePoint {
 
     #[allow(unsafe_code)]
     fn into_iter(self) -> Self::IntoIter {
-        //        // TODO: Check if this is correct
-        //        fn to_char(v: u8) -> char {
-        //            char::from_digit(v as _, 16).unwrap_or('0')
-        //        }
-
         CharIter(match char::from_u32(self.value) {
             Some(c) => {
                 let mut buf = ArrayVec::new();
diff --git a/crates/swc_atoms/src/lib.rs b/crates/swc_atoms/src/lib.rs
index 07c584f10c8a..301c8cc9f7ef 100644
--- a/crates/swc_atoms/src/lib.rs
+++ b/crates/swc_atoms/src/lib.rs
@@ -18,6 +18,7 @@ use std::{
     rc::Rc,
 };
 
+pub use hstr::wtf8;
 use once_cell::sync::Lazy;
 use serde::Serializer;
 
@@ -264,11 +265,16 @@ impl AtomStoreCell {
     pub fn atom<'a>(&self, s: impl Into<Cow<'a, str>>) -> Atom {
         // evaluate the into before borrowing (see #8362)
         let s: Cow<'a, str> = s.into();
+        self.atom_raw(s.as_bytes())
+    }
+
+    #[inline]
+    pub fn atom_raw(&self, s: &[u8]) -> Atom {
         // SAFETY: We can skip the borrow check of RefCell because
         // this API enforces a safe contract. It is slightly faster
         // to use an UnsafeCell. Note the borrow here is short lived
         // only to this block.
-        unsafe { (*self.0.get()).atom(s) }
+        unsafe { Atom((*self.0.get()).0.atom_raw(s)) }
     }
 }
 
diff --git a/crates/swc_ecma_lexer/src/common/lexer/char.rs b/crates/swc_ecma_lexer/src/common/lexer/char.rs
index 53e433e4d7ee..c46c92452600 100644
--- a/crates/swc_ecma_lexer/src/common/lexer/char.rs
+++ b/crates/swc_ecma_lexer/src/common/lexer/char.rs
@@ -1,101 +1,4 @@
-use std::iter::FusedIterator;
-
-use arrayvec::ArrayVec;
-
-#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
-pub struct Char(u32);
-
-impl From<char> for Char {
-    fn from(c: char) -> Self {
-        Char(c as u32)
-    }
-}
-
-impl From<u32> for Char {
-    fn from(c: u32) -> Self {
-        Char(c)
-    }
-}
-
-pub struct CharIter(ArrayVec<char, 12>);
-
-/// Ported from https://github.com/web-infra-dev/oxc/blob/99a4816ce7b6132b2667257984f9d92ae3768f03/crates/oxc_parser/src/lexer/mod.rs#L1349-L1374
-impl IntoIterator for Char {
-    type IntoIter = CharIter;
-    type Item = char;
-
-    #[allow(unsafe_code)]
-    fn into_iter(self) -> Self::IntoIter {
-        //        // TODO: Check if this is correct
-        //        fn to_char(v: u8) -> char {
-        //            char::from_digit(v as _, 16).unwrap_or('0')
-        //        }
-
-        CharIter(match char::from_u32(self.0) {
-            Some(c) => {
-                let mut buf = ArrayVec::new();
-                // Safety: we can make sure that `buf` has enough capacity
-                unsafe {
-                    buf.push_unchecked(c);
-                }
-                buf
-            }
-            None => {
-                let mut buf = ArrayVec::new();
-
-                let high = self.0 & 0xffff0000 >> 16;
-
-                let low = self.0 & 0x0000ffff;
-
-                // The second code unit of a surrogate pair is always in the range from 0xDC00
-                // to 0xDFFF, and is called a low surrogate or a trail surrogate.
-                if !(0xdc00..=0xdfff).contains(&low) {
-                    // Safety: we can make sure that `buf` has enough capacity
-                    unsafe {
-                        buf.push_unchecked('\\');
-                        buf.push_unchecked('u');
-                        for c in format!("{high:x}").chars() {
-                            buf.push_unchecked(c);
-                        }
-                        buf.push_unchecked('\\');
-                        buf.push_unchecked('u');
-                        for c in format!("{low:x}").chars() {
-                            buf.push_unchecked(c);
-                        }
-                    }
-                } else {
-                    // `https://tc39.es/ecma262/#sec-utf16decodesurrogatepair`
-                    let astral_code_point = (high - 0xd800) * 0x400 + low - 0xdc00 + 0x10000;
-
-                    // Safety: we can make sure that `buf` has enough capacity
-                    unsafe {
-                        buf.push_unchecked('\\');
-                        buf.push_unchecked('u');
-                        for c in format!("{astral_code_point:x}").chars() {
-                            buf.push_unchecked(c);
-                        }
-                    }
-                }
-
-                buf
-            }
-        })
-    }
-}
-
-impl Iterator for CharIter {
-    type Item = char;
-
-    fn next(&mut self) -> Option<Self::Item> {
-        if self.0.is_empty() {
-            None
-        } else {
-            Some(self.0.remove(0))
-        }
-    }
-}
-
-impl FusedIterator for CharIter {}
+use swc_atoms::wtf8::CodePoint;
 
 /// Implemented for `char`.
 pub trait CharExt: Copy {
@@ -164,10 +67,10 @@ pub trait CharExt: Copy {
     }
 }
 
-impl CharExt for Char {
+impl CharExt for CodePoint {
     #[inline(always)]
     fn to_char(self) -> Option<char> {
-        char::from_u32(self.0)
+        CodePoint::to_char(&self)
     }
 }
 
diff --git a/crates/swc_ecma_lexer/src/common/lexer/mod.rs b/crates/swc_ecma_lexer/src/common/lexer/mod.rs
index 7be808407c43..d36ed93f5320 100644
--- a/crates/swc_ecma_lexer/src/common/lexer/mod.rs
+++ b/crates/swc_ecma_lexer/src/common/lexer/mod.rs
@@ -1,11 +1,14 @@
 use std::borrow::Cow;
 
-use char::{Char, CharExt};
+use char::CharExt;
 use either::Either::{self, Left, Right};
 use num_bigint::BigInt as BigIntValue;
 use smartstring::{LazyCompact, SmartString};
 use state::State;
-use swc_atoms::Atom;
+use swc_atoms::{
+    wtf8::{CodePoint, Wtf8Buf},
+    Atom,
+};
 use swc_common::{
     comments::{Comment, CommentKind},
     input::{Input, StringInput},
@@ -91,9 +94,13 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
     /// We know that the start and the end are valid
     unsafe fn input_slice(&mut self, start: BytePos, end: BytePos) -> &'a str;
     fn input_uncons_while(&mut self, f: impl FnMut(char) -> bool) -> &'a str;
-    fn atom<'b>(&self, s: impl Into<Cow<'b, str>>) -> swc_atoms::Atom;
+    fn atom_raw(&self, bytes: &[u8]) -> swc_atoms::Atom;
     fn push_error(&mut self, error: crate::error::Error);
 
+    fn atom<'b>(&self, s: impl Into<Cow<'b, str>>) -> swc_atoms::Atom {
+        self.atom_raw(s.into().as_bytes())
+    }
+
     #[inline(always)]
     #[allow(clippy::misnamed_getters)]
     fn had_line_break_before_last(&self) -> bool {
@@ -1119,10 +1126,9 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
         Ok(Self::Token::str(value, raw, self))
     }
 
-    fn read_unicode_escape(&mut self) -> LexResult<Vec<Char>> {
+    fn read_unicode_escape(&mut self) -> LexResult<CodePoint> {
         debug_assert_eq!(self.cur(), Some('u'));
 
-        let mut chars = Vec::with_capacity(4);
         let mut is_curly = false;
 
         self.bump(); // 'u'
@@ -1132,89 +1138,39 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
         }
 
         let state = self.input().cur_pos();
-        let c = match self.read_int_u32::<16>(if is_curly { 0 } else { 4 }) {
-            Ok(Some(val)) => {
-                if 0x0010_ffff >= val {
-                    char::from_u32(val)
-                } else {
-                    let start = self.cur_pos();
-
-                    self.error(
-                        start,
-                        SyntaxError::BadCharacterEscapeSequence {
-                            expected: if is_curly {
-                                "1-6 hex characters in the range 0 to 10FFFF."
-                            } else {
-                                "4 hex characters"
-                            },
-                        },
-                    )?
-                }
-            }
-            _ => {
-                let start = self.cur_pos();
-
-                self.error(
-                    start,
-                    SyntaxError::BadCharacterEscapeSequence {
-                        expected: if is_curly {
-                            "1-6 hex characters"
-                        } else {
-                            "4 hex characters"
-                        },
+        let Ok(Some(val)) = self.read_int_u32::<16>(if is_curly { 0 } else { 4 }) else {
+            let start = self.cur_pos();
+            self.error(
+                start,
+                SyntaxError::BadCharacterEscapeSequence {
+                    expected: if is_curly {
+                        "1-6 hex characters"
+                    } else {
+                        "4 hex characters"
                     },
-                )?
-            }
+                },
+            )?
         };
 
-        match c {
-            Some(c) => {
-                chars.push(c.into());
-            }
-            _ => {
-                unsafe {
-                    // Safety: state is valid position because we got it from cur_pos()
-                    self.input_mut().reset_to(state);
-                }
-
-                chars.push(Char::from('\\'));
-                chars.push(Char::from('u'));
-
-                if is_curly {
-                    chars.push(Char::from('{'));
-
-                    for _ in 0..6 {
-                        if let Some(c) = self.input().cur() {
-                            if c == '}' {
-                                break;
-                            }
-
-                            self.bump();
-
-                            chars.push(Char::from(c));
-                        } else {
-                            break;
-                        }
-                    }
-
-                    chars.push(Char::from('}'));
-                } else {
-                    for _ in 0..4 {
-                        if let Some(c) = self.input().cur() {
-                            self.bump();
-
-                            chars.push(Char::from(c));
-                        }
-                    }
-                }
-            }
-        }
+        let Some(code_point) = CodePoint::from_u32(val) else {
+            let start = self.cur_pos();
+            self.error(
+                start,
+                SyntaxError::BadCharacterEscapeSequence {
+                    expected: if is_curly {
+                        "1-6 hex characters in the range 0 to 10FFFF."
+                    } else {
+                        "4 hex characters"
+                    },
+                },
+            )?
+        };
 
         if is_curly && !self.eat(b'}') {
-            self.error(state, SyntaxError::InvalidUnicodeEscape)?
+            self.error(state, SyntaxError::InvalidUnicodeEscape)?;
         }
 
-        Ok(chars)
+        Ok(code_point)
     }
 
     #[cold]
@@ -1231,7 +1187,7 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
     fn read_tmpl_token(&mut self, start_of_tpl: BytePos) -> LexResult<Self::Token> {
         let start = self.cur_pos();
 
-        let mut cooked = Ok(String::new());
+        let mut cooked = Ok(Wtf8Buf::new());
         let mut cooked_slice_start = start;
         let raw_slice_start = start;
 
@@ -1288,7 +1244,7 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
                             Ok(self.atom(s))
                         } else {
                             consume_cooked!();
-                            cooked.map(|s| self.atom(s))
+                            cooked.map(|s| self.atom_raw(s.as_bytes()))
                         };
 
                         let end = self.input().cur_pos();
@@ -1313,7 +1269,7 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
                         Ok(self.atom(s))
                     } else {
                         consume_cooked!();
-                        cooked.map(|s| self.atom(s))
+                        cooked.map(|s| self.atom_raw(s.as_bytes()))
                     };
 
                     let end = self.input().cur_pos();
@@ -1333,7 +1289,7 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
                     }
 
                     if let Ok(ref mut cooked) = cooked {
-                        cooked.push('\n');
+                        cooked.push_char('\n');
                     }
                     cooked_slice_start = self.cur_pos();
                 }
@@ -1342,11 +1298,9 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
                     consume_cooked!();
 
                     match self.read_escaped_char(true) {
-                        Ok(Some(chars)) => {
+                        Ok(Some(code_point)) => {
                             if let Ok(ref mut cooked) = cooked {
-                                for c in chars {
-                                    cooked.extend(c);
-                                }
+                                cooked.push(code_point);
                             }
                         }
                         Ok(None) => {}
@@ -1365,7 +1319,7 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
     /// Read an escaped character for string literal.
     ///
     /// In template literal, we should preserve raw string.
-    fn read_escaped_char(&mut self, in_template: bool) -> LexResult<Option<Vec<Char>>> {
+    fn read_escaped_char(&mut self, in_template: bool) -> LexResult<Option<CodePoint>> {
         debug_assert_eq!(self.cur(), Some('\\'));
 
         let start = self.cur_pos();
@@ -1403,7 +1357,13 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
                 self.bump(); // 'x'
 
                 match self.read_int_u32::<16>(2)? {
-                    Some(val) => return Ok(Some(vec![Char::from(val)])),
+                    Some(val) => {
+                        return {
+                            // Safety: val is in 0x00..=0xFF
+                            let code_point = unsafe { CodePoint::from_u32_unchecked(val) };
+                            Ok(Some(code_point))
+                        };
+                    }
                     None => self.error(
                         start,
                         SyntaxError::BadCharacterEscapeSequence {
@@ -1427,7 +1387,7 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
                     match self.cur() {
                         Some(next) if next.is_digit(8) => c,
                         // \0 is not an octal literal nor decimal literal.
-                        _ => return Ok(Some(vec!['\u{0000}'.into()])),
+                        _ => return Ok(Some(CodePoint::from_char('\u{0000}'))),
                     }
                 } else {
                     c
@@ -1454,7 +1414,9 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
                                         .and_then(|value| value.checked_add(v as u8));
                                     match new_val {
                                         Some(val) => val,
-                                        None => return Ok(Some(vec![Char::from(value as char)])),
+                                        None => {
+                                            return Ok(Some(CodePoint::from_char(value as char)))
+                                        }
                                     }
                                 } else {
                                     value * 8 + v as u8
@@ -1462,7 +1424,7 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
 
                                 self.bump();
                             }
-                            _ => return Ok(Some(vec![Char::from(value as u32)])),
+                            _ => return Ok(Some(CodePoint::from_char(value as char))),
                         }
                     }};
                 }
@@ -1470,7 +1432,7 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
                 one!(false);
                 one!(true);
 
-                return Ok(Some(vec![Char::from(value as char)]));
+                return Ok(Some(CodePoint::from_char(value as char)));
             }
             _ => c,
         };
@@ -1480,7 +1442,7 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
             self.input_mut().bump();
         }
 
-        Ok(Some(vec![c.into()]))
+        Ok(Some(CodePoint::from_char(c)))
     }
 
     /// Expects current char to be '/'
@@ -1665,24 +1627,19 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
                         }
                     }
 
-                    let chars = self.read_unicode_escape()?;
-
-                    if let Some(c) = chars.first() {
-                        let valid = if first {
-                            c.is_ident_start()
-                        } else {
-                            c.is_ident_part()
-                        };
+                    let code_point = self.read_unicode_escape()?;
 
-                        if !valid {
-                            self.emit_error(start, SyntaxError::InvalidIdentChar);
-                        }
-                    }
+                    let valid = if first {
+                        code_point.is_ident_start()
+                    } else {
+                        code_point.is_ident_part()
+                    };
 
-                    for c in chars {
-                        buf.extend(c);
+                    if !valid {
+                        self.emit_error(start, SyntaxError::InvalidIdentChar);
                     }
 
+                    buf.extend(code_point);
                     slice_start = self.cur_pos();
                     continue;
                 }
@@ -1964,7 +1921,7 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
 
         let mut slice_start = self.input().cur_pos();
 
-        let mut buf: Option<String> = None;
+        let mut buf: Option<Wtf8Buf> = None;
 
         loop {
             let table = if quote == b'"' {
@@ -2005,7 +1962,7 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
                             self.input_slice(slice_start, value_end)
                         };
                         buf.push_str(s);
-                        self.atom(&*buf)
+                        self.atom_raw(buf.as_bytes())
                     } else {
                         let s = unsafe { self.input_slice(slice_start, value_end) };
                         self.atom(s)
@@ -2034,15 +1991,15 @@ pub trait Lexer<'a, TokenAndSpan>: Tokens<TokenAndSpan> + Sized {
                     };
 
                     if buf.is_none() {
-                        buf = Some(s.to_string());
+                        let mut new_buf = Wtf8Buf::with_capacity(s.len());
+                        new_buf.push_str(s);
+                        buf = Some(new_buf);
                     } else {
                         buf.as_mut().unwrap().push_str(s);
                     }
 
-                    if let Some(chars) = self.read_escaped_char(false)? {
-                        for c in chars {
-                            buf.as_mut().unwrap().extend(c);
-                        }
+                    if let Some(code_point) = self.read_escaped_char(false)? {
+                        buf.as_mut().unwrap().push(code_point);
                     }
 
                     slice_start = self.cur_pos();
diff --git a/crates/swc_ecma_lexer/src/lexer/mod.rs b/crates/swc_ecma_lexer/src/lexer/mod.rs
index 5833e87f7986..cbf06bdbc1fd 100644
--- a/crates/swc_ecma_lexer/src/lexer/mod.rs
+++ b/crates/swc_ecma_lexer/src/lexer/mod.rs
@@ -110,8 +110,8 @@ impl<'a> crate::common::lexer::Lexer<'a, TokenAndSpan> for Lexer<'a> {
     }
 
     #[inline(always)]
-    fn atom<'b>(&self, s: impl Into<std::borrow::Cow<'b, str>>) -> swc_atoms::Atom {
-        self.atoms.atom(s)
+    fn atom_raw(&self, s: &[u8]) -> swc_atoms::Atom {
+        self.atoms.atom_raw(s)
     }
 }
 
diff --git a/crates/swc_ecma_parser/src/lexer/mod.rs b/crates/swc_ecma_parser/src/lexer/mod.rs
index bb0cd16e1c55..391b6242910a 100644
--- a/crates/swc_ecma_parser/src/lexer/mod.rs
+++ b/crates/swc_ecma_parser/src/lexer/mod.rs
@@ -2,7 +2,7 @@
 
 use std::{char, iter::FusedIterator, rc::Rc};
 
-use swc_atoms::AtomStoreCell;
+use swc_atoms::{wtf8::Wtf8Buf, AtomStoreCell};
 use swc_common::{
     comments::Comments,
     input::{Input, StringInput},
@@ -113,8 +113,8 @@ impl<'a> swc_ecma_lexer::common::lexer::Lexer<'a, TokenAndSpan> for Lexer<'a> {
     }
 
     #[inline(always)]
-    fn atom<'b>(&self, s: impl Into<std::borrow::Cow<'b, str>>) -> swc_atoms::Atom {
-        self.atoms.atom(s)
+    fn atom_raw(&self, s: &[u8]) -> swc_atoms::Atom {
+        self.atoms.atom_raw(s)
     }
 }
 
@@ -332,7 +332,7 @@ impl Lexer<'_> {
         started_with_backtick: bool,
     ) -> LexResult<Token> {
         debug_assert!(self.cur() == Some(if started_with_backtick { '`' } else { '}' }));
-        let mut cooked = Ok(String::with_capacity(8));
+        let mut cooked = Ok(Wtf8Buf::with_capacity(8));
         self.bump(); // `}` or `\``
         let mut cooked_slice_start = self.cur_pos();
         let raw_slice_start = cooked_slice_start;
@@ -357,7 +357,7 @@ impl Lexer<'_> {
         while let Some(c) = self.cur() {
             if c == '`' {
                 consume_cooked!();
-                let cooked = cooked.map(|cooked| self.atoms.atom(cooked));
+                let cooked = cooked.map(|cooked| self.atoms.atom_raw(cooked.as_bytes()));
                 let raw = raw_atom(self);
                 self.bump();
                 return Ok(if started_with_backtick {
@@ -369,7 +369,7 @@ impl Lexer<'_> {
                 });
             } else if c == '$' && self.input.peek() == Some('{') {
                 consume_cooked!();
-                let cooked = cooked.map(|cooked| self.atoms.atom(cooked));
+                let cooked = cooked.map(|cooked| self.atoms.atom_raw(cooked.as_bytes()));
                 let raw = raw_atom(self);
                 self.input.bump_bytes(2);
                 return Ok(if started_with_backtick {
@@ -383,11 +383,9 @@ impl Lexer<'_> {
                 consume_cooked!();
 
                 match self.read_escaped_char(true) {
-                    Ok(Some(chars)) => {
+                    Ok(Some(code_point)) => {
                         if let Ok(ref mut cooked) = cooked {
-                            for c in chars {
-                                cooked.extend(c);
-                            }
+                            cooked.push(code_point);
                         }
                     }
                     Ok(None) => {}
@@ -416,7 +414,7 @@ impl Lexer<'_> {
                 self.bump();
 
                 if let Ok(ref mut cooked) = cooked {
-                    cooked.push(c);
+                    cooked.push_char(c);
                 }
                 cooked_slice_start = self.cur_pos();
             } else {
diff --git a/crates/swc_ecma_parser/src/lexer/state.rs b/crates/swc_ecma_parser/src/lexer/state.rs
index 18bf6880d597..d9da4d259475 100644
--- a/crates/swc_ecma_parser/src/lexer/state.rs
+++ b/crates/swc_ecma_parser/src/lexer/state.rs
@@ -553,13 +553,11 @@ impl Lexer<'_> {
                     continue;
                 }
                 self.bump(); // bump 'u'
-                let Ok(chars) = self.read_unicode_escape() else {
+                let Ok(code_point) = self.read_unicode_escape() else {
                     self.emit_error(self.cur_pos(), SyntaxError::InvalidUnicodeEscape);
                     break;
                 };
-                for c in chars {
-                    v.extend(c);
-                }
+                v.extend(code_point);
                 self.token_flags |= swc_ecma_lexer::lexer::TokenFlags::UNICODE;
             } else {
                 break;

From 158455edaa074f7d3d4445368667210a5ff0f463 Mon Sep 17 00:00:00 2001
From: CPunisher <1343316114@qq.com>
Date: Fri, 12 Sep 2025 18:21:07 +0800
Subject: [PATCH 3/5] as_wtf8_str

---
 crates/hstr/src/lib.rs      | 15 +++++++++++++++
 crates/hstr/src/wtf8/mod.rs |  5 +++++
 2 files changed, 20 insertions(+)

diff --git a/crates/hstr/src/lib.rs b/crates/hstr/src/lib.rs
index 39e370db1348..38a316c1dd6c 100644
--- a/crates/hstr/src/lib.rs
+++ b/crates/hstr/src/lib.rs
@@ -284,6 +284,21 @@ impl Atom {
             _ => unsafe { debug_unreachable!() },
         }
     }
+
+    fn as_wtf8_str(&self) -> &wtf8::Wtf8 {
+        match self.tag() {
+            DYNAMIC_TAG => unsafe {
+                let item = crate::dynamic::deref_from(self.unsafe_data);
+                wtf8::Wtf8::from_bytes(transmute::<&[u8], &'static [u8]>(&item.slice))
+            },
+            INLINE_TAG => {
+                let len = (self.unsafe_data.tag() & LEN_MASK) >> LEN_OFFSET;
+                let src = self.unsafe_data.data();
+                wtf8::Wtf8::from_bytes(&src[..(len as usize)])
+            }
+            _ => unsafe { debug_unreachable!() },
+        }
+    }
 }
 
 #[cfg(test)]
diff --git a/crates/hstr/src/wtf8/mod.rs b/crates/hstr/src/wtf8/mod.rs
index 57e2ed40eec7..5a4c73098b59 100644
--- a/crates/hstr/src/wtf8/mod.rs
+++ b/crates/hstr/src/wtf8/mod.rs
@@ -539,6 +539,11 @@ impl Wtf8 {
         unsafe { transmute(value.as_bytes()) }
     }
 
+    #[inline]
+    pub fn from_bytes(value: &[u8]) -> &Wtf8 {
+        unsafe { transmute(value) }
+    }
+
     /// Return the length, in WTF-8 bytes.
     #[inline]
     pub fn len(&self) -> usize {

From a6ce6f663fd68e16bac706012d055175166efec7 Mon Sep 17 00:00:00 2001
From: CPunisher <1343316114@qq.com>
Date: Mon, 15 Sep 2025 11:14:15 +0800
Subject: [PATCH 4/5] shear and doc test

---
 Cargo.lock                       | 1 -
 crates/hstr/src/wtf8/mod.rs      | 4 ++--
 crates/swc_ecma_lexer/Cargo.toml | 1 -
 3 files changed, 2 insertions(+), 4 deletions(-)

diff --git a/Cargo.lock b/Cargo.lock
index c00182f90ecd..8803bcb6b50d 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -5577,7 +5577,6 @@ dependencies = [
 name = "swc_ecma_lexer"
 version = "23.0.1"
 dependencies = [
- "arrayvec",
  "bitflags 2.6.0",
  "codspeed-criterion-compat",
  "either",
diff --git a/crates/hstr/src/wtf8/mod.rs b/crates/hstr/src/wtf8/mod.rs
index 5a4c73098b59..6a1dd537af2a 100644
--- a/crates/hstr/src/wtf8/mod.rs
+++ b/crates/hstr/src/wtf8/mod.rs
@@ -277,7 +277,7 @@ impl Wtf8Buf {
     /// # Example
     ///
     /// ```
-    /// let mut s = Wtf8Buf::new();
+    /// let mut s = hstr::wtf8::Wtf8Buf::new();
     /// s.reserve(10);
     /// assert!(s.capacity() >= 10);
     /// ```
@@ -292,7 +292,7 @@ impl Wtf8Buf {
     /// # Example
     ///
     /// ```
-    /// let s = Wtf8Buf::with_capacity(10);
+    /// let s = hstr::wtf8::Wtf8Buf::with_capacity(10);
     /// assert!(s.capacity() >= 10);
     /// ```
     #[inline]
diff --git a/crates/swc_ecma_lexer/Cargo.toml b/crates/swc_ecma_lexer/Cargo.toml
index b7628b4d7079..2fc4907991f9 100644
--- a/crates/swc_ecma_lexer/Cargo.toml
+++ b/crates/swc_ecma_lexer/Cargo.toml
@@ -25,7 +25,6 @@ typescript    = []
 verify        = ["swc_ecma_visit"]
 
 [dependencies]
-arrayvec    = { workspace = true }
 bitflags    = { workspace = true }
 either      = { workspace = true }
 num-bigint  = { workspace = true }

From acf8ffeb98ac62fe888fa19a381a7da43fb95c63 Mon Sep 17 00:00:00 2001
From: CPunisher <1343316114@qq.com>
Date: Mon, 15 Sep 2025 11:29:42 +0800
Subject: [PATCH 5/5] as_wtf8_str

---
 crates/hstr/src/lib.rs             | 30 +++++++++++++++---------------
 crates/swc_atoms/src/lib.rs        |  5 +++++
 crates/swc_ecma_codegen/src/lit.rs |  3 +--
 3 files changed, 21 insertions(+), 17 deletions(-)

diff --git a/crates/hstr/src/lib.rs b/crates/hstr/src/lib.rs
index 38a316c1dd6c..78fdbd23b703 100644
--- a/crates/hstr/src/lib.rs
+++ b/crates/hstr/src/lib.rs
@@ -250,6 +250,21 @@ impl Atom {
         }
         self.clone()
     }
+
+    pub fn as_wtf8_str(&self) -> &wtf8::Wtf8 {
+        match self.tag() {
+            DYNAMIC_TAG => unsafe {
+                let item = crate::dynamic::deref_from(self.unsafe_data);
+                wtf8::Wtf8::from_bytes(transmute::<&[u8], &'static [u8]>(&item.slice))
+            },
+            INLINE_TAG => {
+                let len = (self.unsafe_data.tag() & LEN_MASK) >> LEN_OFFSET;
+                let src = self.unsafe_data.data();
+                wtf8::Wtf8::from_bytes(&src[..(len as usize)])
+            }
+            _ => unsafe { debug_unreachable!() },
+        }
+    }
 }
 
 impl Atom {
@@ -284,21 +299,6 @@ impl Atom {
             _ => unsafe { debug_unreachable!() },
         }
     }
-
-    fn as_wtf8_str(&self) -> &wtf8::Wtf8 {
-        match self.tag() {
-            DYNAMIC_TAG => unsafe {
-                let item = crate::dynamic::deref_from(self.unsafe_data);
-                wtf8::Wtf8::from_bytes(transmute::<&[u8], &'static [u8]>(&item.slice))
-            },
-            INLINE_TAG => {
-                let len = (self.unsafe_data.tag() & LEN_MASK) >> LEN_OFFSET;
-                let src = self.unsafe_data.data();
-                wtf8::Wtf8::from_bytes(&src[..(len as usize)])
-            }
-            _ => unsafe { debug_unreachable!() },
-        }
-    }
 }
 
 #[cfg(test)]
diff --git a/crates/swc_atoms/src/lib.rs b/crates/swc_atoms/src/lib.rs
index 301c8cc9f7ef..f9114cc51640 100644
--- a/crates/swc_atoms/src/lib.rs
+++ b/crates/swc_atoms/src/lib.rs
@@ -76,6 +76,11 @@ impl Atom {
     pub fn as_str(&self) -> &str {
         &self.0
     }
+
+    #[inline]
+    pub fn as_wtf8_str(&self) -> &wtf8::Wtf8 {
+        self.0.as_wtf8_str()
+    }
 }
 
 impl Deref for Atom {
diff --git a/crates/swc_ecma_codegen/src/lit.rs b/crates/swc_ecma_codegen/src/lit.rs
index 1ebe6497faf6..15ab750d5f16 100644
--- a/crates/swc_ecma_codegen/src/lit.rs
+++ b/crates/swc_ecma_codegen/src/lit.rs
@@ -78,8 +78,7 @@ impl MacroNode for Str {
 
                 if es5_safe
                     && (!emitter.cfg.ascii_only || raw.is_ascii())
-                    && (!emitter.cfg.inline_script
-                        || !self.raw.as_ref().unwrap().contains("script"))
+                    && (!emitter.cfg.inline_script || !raw.contains("script"))
                 {
                     emitter.wr.write_str_lit(DUMMY_SP, raw)?;
                     return Ok(());