feat(string): implement RopeString with thin vtable and lazy flattening

cli/src/debug/string.rs

@@ -1,6 +1,6 @@
 use boa_engine::{
     Context, JsNativeError, JsObject, JsResult, JsValue, NativeFunction, js_string,
-    object::ObjectInitializer, property::Attribute, string::JsStrVariant,
+    object::ObjectInitializer, property::Attribute, string::JsStr,
 };
 
 fn storage(_: &JsValue, args: &[JsValue], _: &mut Context) -> JsResult<JsValue> {
@@ -34,9 +34,10 @@ fn encoding(_: &JsValue, args: &[JsValue], _: &mut Context) -> JsResult<JsValue>
     };
 
     let str = string.as_str();
-    let encoding = match str.variant() {
-        JsStrVariant::Latin1(_) => "latin1",
-        JsStrVariant::Utf16(_) => "utf16",
+    let encoding = match str {
+        JsStr::Latin1(_) => "latin1",
+        JsStr::Utf16(_) => "utf16",
+        JsStr::Rope(_) => "rope",
     };
     Ok(js_string!(encoding).into())
 }
@@ -55,9 +56,10 @@ fn summary(_: &JsValue, args: &[JsValue], context: &mut Context) -> JsResult<JsV
     };
 
     let storage = if string.is_static() { "static" } else { "heap" };
-    let encoding = match string.as_str().variant() {
-        JsStrVariant::Latin1(_) => "latin1",
-        JsStrVariant::Utf16(_) => "utf16",
+    let encoding = match string.as_str() {
+        JsStr::Latin1(_) => "latin1",
+        JsStr::Utf16(_) => "utf16",
+        JsStr::Rope(_) => "rope",
     };
 
     let summary = ObjectInitializer::new(context)

core/engine/src/builtins/date/utils.rs

@@ -1,6 +1,5 @@
 use crate::{JsStr, JsString, context::HostHooks, js_string, value::IntegerOrInfinity};
 use boa_macros::js_str;
-use boa_string::JsStrVariant;
 use std::slice::Iter;
 use std::str;
 use std::{borrow::Cow, iter::Peekable};
@@ -753,22 +752,22 @@ pub(super) fn pad_six(t: u32, output: &mut [u8; 6]) -> JsStr<'_> {
 /// [spec-format]: https://tc39.es/ecma262/#sec-date-time-string-format
 pub(super) fn parse_date(date: &JsString, hooks: &dyn HostHooks) -> Option<i64> {
     // All characters must be ASCII so we can return early if we find a non-ASCII character.
-    let owned_js_str = date.as_str();
-    let date = match owned_js_str.variant() {
-        JsStrVariant::Latin1(s) => {
+    let date = match date.as_flat_str() {
+        JsStr::Latin1(s) => {
             if !s.is_ascii() {
                 return None;
             }
             // SAFETY: Since all characters are ASCII we can safely convert this into str.
             Cow::Borrowed(unsafe { str::from_utf8_unchecked(s) })
         }
-        JsStrVariant::Utf16(s) => {
+        JsStr::Utf16(s) => {
             let date = String::from_utf16(s).ok()?;
             if !date.is_ascii() {
                 return None;
             }
             Cow::Owned(date)
         }
+        JsStr::Rope(_) => unreachable!("rope should be flattened by as_flat_str"),
     };
 
     // Date Time String Format: 'YYYY-MM-DDTHH:mm:ss.sssZ'

core/engine/src/builtins/intl/segmenter/iterator.rs

@@ -128,7 +128,7 @@ impl SegmentIterator {
                 .downcast_ref::<Segmenter>()
                 .js_expect("segment iterator object should contain a segmenter")
                 .ok()?;
-            let mut segments = segmenter.native.segment(string.variant());
+            let mut segments = segmenter.native.segment(string.as_str());
             // the first elem is always 0.
             segments.next();
             segments

core/engine/src/builtins/intl/segmenter/mod.rs

@@ -15,7 +15,7 @@ use crate::{
     string::StaticJsStrings,
 };
 use boa_gc::{Finalize, Trace};
-use boa_string::JsStrVariant;
+use boa_string::JsStr;
 use icu_collator::provider::CollationDiacriticsV1;
 use icu_locale::Locale;
 use icu_segmenter::{
@@ -63,12 +63,9 @@ impl NativeSegmenter {
 
     /// Segment the passed string, returning an iterator with the index boundaries
     /// of the segments.
-    pub(crate) fn segment<'l, 's>(
-        &'l self,
-        input: JsStrVariant<'s>,
-    ) -> NativeSegmentIterator<'l, 's> {
+    pub(crate) fn segment<'l, 's>(&'l self, input: JsStr<'s>) -> NativeSegmentIterator<'l, 's> {
         match input {
-            JsStrVariant::Latin1(input) => match self {
+            JsStr::Latin1(input) => match self {
                 Self::Grapheme(g) => {
                     NativeSegmentIterator::GraphemeLatin1(g.as_borrowed().segment_latin1(input))
                 }
@@ -79,7 +76,7 @@ impl NativeSegmenter {
                     NativeSegmentIterator::SentenceLatin1(s.as_borrowed().segment_latin1(input))
                 }
             },
-            JsStrVariant::Utf16(input) => match self {
+            JsStr::Utf16(input) => match self {
                 Self::Grapheme(g) => {
                     NativeSegmentIterator::GraphemeUtf16(g.as_borrowed().segment_utf16(input))
                 }
@@ -90,6 +87,10 @@ impl NativeSegmenter {
                     NativeSegmentIterator::SentenceUtf16(s.as_borrowed().segment_utf16(input))
                 }
             },
+            JsStr::Rope(rope) => {
+                let flat = boa_string::vtable::rope::flatten_rope(rope.header);
+                self.segment(flat.as_str())
+            }
         }
     }
 }

core/engine/src/builtins/intl/segmenter/segments.rs

@@ -89,7 +89,7 @@ impl Segments {
         // 8. Let startIndex be ! FindBoundary(segmenter, string, n, before).
         // 9. Let endIndex be ! FindBoundary(segmenter, string, n, after).
         let (range, is_word_like) = {
-            let mut segments = segmenter.native.segment(segments.string.variant());
+            let mut segments = segmenter.native.segment(segments.string.as_str());
             std::iter::from_fn(|| segments.next().map(|i| (i, segments.is_word_like())))
                 .tuple_windows()
                 .find(|((i, _), (j, _))| (*i..*j).contains(&n))

core/engine/src/builtins/number/globals.rs

@@ -8,7 +8,6 @@ use crate::{
 };
 
 use boa_macros::js_str;
-use boa_string::JsStrVariant;
 
 /// Builtin javascript 'isFinite(number)' function.
 ///
@@ -359,13 +358,14 @@ pub(crate) fn parse_float(
         return Ok(JsValue::nan());
     }
 
-    let value = match trimmed_string.variant() {
-        JsStrVariant::Latin1(s) => fast_float2::parse_partial::<f64, _>(s),
-        JsStrVariant::Utf16(s) => {
+    let value = match input_string.as_flat_str() {
+        JsStr::Latin1(s) => fast_float2::parse_partial::<f64, _>(s),
+        JsStr::Utf16(s) => {
             // TODO: Explore adding direct UTF-16 parsing support to fast_float2.
             let s = String::from_utf16_lossy(s);
             fast_float2::parse_partial::<f64, _>(s.as_bytes())
         }
+        JsStr::Rope(_) => unreachable!("rope should be flattened by as_flat_str"),
     };
 
     Ok(value.map_or_else(

core/engine/src/builtins/regexp/mod.rs

@@ -18,7 +18,7 @@ use crate::{
     object::{CONSTRUCTOR, JsObject, internal_methods::get_prototype_from_constructor},
     property::Attribute,
     realm::Realm,
-    string::{CodePoint, CommonJsStringBuilder, JsStrVariant, StaticJsStrings},
+    string::{CodePoint, CommonJsStringBuilder, JsStr, StaticJsStrings},
     symbol::JsSymbol,
     value::JsValue,
 };
@@ -1157,20 +1157,20 @@ impl RegExp {
 
         // 13.b. Let inputIndex be the index into input of the character that was obtained from element lastIndex of S.
         // 13.c. Let r be matcher(input, inputIndex).
-        let r: Option<regress::Match> = match (full_unicode, input.as_str().variant()) {
-            (true | false, JsStrVariant::Latin1(_)) => {
+        let r: Option<regress::Match> = match (full_unicode, input.as_flat_str()) {
+            (true | false, JsStr::Latin1(input)) => {
                 // TODO: Currently regress does not support latin1 encoding.
-                let input = input.to_vec();
-
+                let input = input.iter().map(|&b| u16::from(b)).collect::<Vec<u16>>();
                 // NOTE: We can use the faster ucs2 variant since there will never be two byte unicode.
                 matcher.find_from_ucs2(&input, last_index as usize).next()
             }
-            (true, JsStrVariant::Utf16(input)) => {
+            (true, JsStr::Utf16(input)) => {
                 matcher.find_from_utf16(input, last_index as usize).next()
             }
-            (false, JsStrVariant::Utf16(input)) => {
+            (false, JsStr::Utf16(input)) => {
                 matcher.find_from_ucs2(input, last_index as usize).next()
             }
+            (_, JsStr::Rope(_)) => unreachable!("rope should be flattened by as_flat_str"),
         };
 
         let Some(match_value) = r else {

core/engine/src/builtins/string/mod.rs

@@ -653,18 +653,19 @@ impl String {
         let this = this.require_object_coercible()?;
 
         // 2. Let S be ? ToString(O).
-        let mut string = this.to_string(context)?;
+        let mut strings = Vec::with_capacity(args.len() + 1);
+        strings.push(this.to_string(context)?);
 
         // 3. Let R be S.
         // 4. For each element next of args, do
         for arg in args {
             // a. Let nextString be ? ToString(next).
             // b. Set R to the string-concatenation of R and nextString.
-            string = js_string!(&string, &arg.to_string(context)?);
+            strings.push(arg.to_string(context)?);
         }
 
         // 5. Return R.
-        Ok(JsValue::new(string))
+        Ok(JsValue::new(JsString::concat_array_strings(&strings)))
     }
 
     /// `String.prototype.repeat( count )`

core/engine/src/module/loader/mod.rs

@@ -3,6 +3,8 @@ use std::cell::RefCell;
 use std::path::{Component, Path, PathBuf};
 use std::rc::Rc;
 
+#[cfg(target_family = "windows")]
+use cow_utils::CowUtils;
 use dynify::dynify;
 use rustc_hash::FxHashMap;
 
@@ -62,7 +64,7 @@ pub fn resolve_module_specifier(
 
     // On Windows, also replace `/` with `\`. JavaScript imports use `/` as path separator.
     #[cfg(target_family = "windows")]
-    let specifier = cow_utils::CowUtils::cow_replace(&*specifier, '/', "\\");
+    let specifier = specifier.cow_replace('/', "\\");
 
     let short_path = Path::new(&*specifier);
 

core/engine/src/string.rs

@@ -62,10 +62,10 @@ macro_rules! js_string {
         $crate::string::JsString::from($s)
     };
     ( $x:expr, $y:expr ) => {
-        $crate::string::JsString::concat($crate::string::JsStr::from($x), $crate::string::JsStr::from($y))
+        $crate::string::JsString::concat(&$crate::string::JsString::from($x), &$crate::string::JsString::from($y))
     };
     ( $( $s:expr ),+ ) => {
-        $crate::string::JsString::concat_array(&[ $( $crate::string::JsStr::from($s) ),+ ])
+        $crate::string::JsString::concat_array_strings(&[ $( $crate::string::JsString::from($s) ),+ ])
     };
 }
 
@@ -166,15 +166,15 @@ mod tests {
         let x = js_string!("hello");
         let z = js_string!("world");
 
-        let xy = js_string!(&x, &JsString::from(Y));
+        let xy = js_string!(x.clone(), JsString::from(Y));
         assert_eq!(&xy, utf16!("hello, "));
         assert_eq!(xy.refcount(), Some(1));
 
-        let xyz = js_string!(&xy, &z);
+        let xyz = js_string!(xy.clone(), z.clone());
         assert_eq!(&xyz, utf16!("hello, world"));
         assert_eq!(xyz.refcount(), Some(1));
 
-        let xyzw = js_string!(&xyz, &JsString::from(W));
+        let xyzw = js_string!(xyz.clone(), JsString::from(W));
         assert_eq!(&xyzw, utf16!("hello, world!"));
         assert_eq!(xyzw.refcount(), Some(1));
     }

core/engine/src/vm/opcode/concat/mod.rs

@@ -21,7 +21,7 @@ impl ConcatToString {
             let val = context.vm.get_register(value.into()).clone();
             strings.push(val.to_string(context)?);
         }
-        let s = JsString::concat_array(&strings.iter().map(JsString::as_str).collect::<Vec<_>>());
+        let s = JsString::concat_array_strings(&strings);
         context.vm.set_register(string.into(), s.into());
         Ok(())
     }

core/runtime/src/text/encodings.rs

@@ -22,8 +22,17 @@ pub(crate) mod utf8 {
 }
 
 pub(crate) mod utf16le {
+    use boa_engine::string::JsStr;
     use boa_engine::{JsString, js_string};
 
+    #[allow(dead_code)]
+    pub(crate) fn encode(input: &JsString) -> Vec<u8> {
+        match input.as_str() {
+            JsStr::Latin1(l) => l.iter().flat_map(|c| [*c, 0]).collect(),
+            JsStr::Utf16(s) => bytemuck::cast_slice(s).to_vec(),
+            JsStr::Rope(_) => input.iter().flat_map(u16::to_le_bytes).collect(),
+        }
+    }
     pub(crate) fn decode(mut input: &[u8], strip_bom: bool) -> JsString {
         if strip_bom {
             input = input.strip_prefix(&[0xFF, 0xFE]).unwrap_or(input);
@@ -48,8 +57,17 @@ pub(crate) mod utf16le {
 }
 
 pub(crate) mod utf16be {
+    use boa_engine::string::JsStr;
     use boa_engine::{JsString, js_string};
 
+    #[allow(dead_code)]
+    pub(crate) fn encode(input: &JsString) -> Vec<u8> {
+        match input.as_str() {
+            JsStr::Latin1(l) => l.iter().flat_map(|c| [0, *c]).collect(),
+            JsStr::Utf16(s) => s.iter().flat_map(|b| b.to_be_bytes()).collect::<Vec<_>>(),
+            JsStr::Rope(_) => input.iter().flat_map(u16::to_be_bytes).collect(),
+        }
+    }
     pub(crate) fn decode(mut input: Vec<u8>, strip_bom: bool) -> JsString {
         if strip_bom && input.starts_with(&[0xFE, 0xFF]) {
             input.drain(..2);