quickjs-oxide-regexp 0.4.0

//! Checksum-pinned Unicode 17 case conversion used by String intrinsics.
//!
//! This is a direct Rust port of QuickJS's `lre_case_conv`, `lre_is_cased`,
//! `lre_is_case_ignorable`, UTF-16 traversal, and Final_Sigma context check.
//! It intentionally does not use Rust `char` casing, whose Unicode version is
//! coupled to the toolchain and whose scalar-only API cannot preserve lone
//! UTF-16 surrogates.

use super::{CodeUnitBuf, CodeUnits, MAX_STRING_LEN, StringError, is_in_table};
#[cfg(any(test, feature = "test-hooks"))]
use std::cell::Cell;

mod tables {
    include!("generated/unicode/unicode_case_tables.rs");
}

const RUN_TYPE_U: u32 = 0;
const RUN_TYPE_L: u32 = 1;
const RUN_TYPE_UF: u32 = 2;
const RUN_TYPE_LF: u32 = 3;
const RUN_TYPE_UL: u32 = 4;
const RUN_TYPE_LSU: u32 = 5;
const RUN_TYPE_U2L_399_EXT2: u32 = 6;
const RUN_TYPE_UF_D20: u32 = 7;
const RUN_TYPE_UF_D1_EXT: u32 = 8;
const RUN_TYPE_U_EXT: u32 = 9;
const RUN_TYPE_LF_EXT: u32 = 10;
const RUN_TYPE_UF_EXT2: u32 = 11;
const RUN_TYPE_LF_EXT2: u32 = 12;
const RUN_TYPE_UF_EXT3: u32 = 13;

#[cfg(any(test, feature = "test-hooks"))]
thread_local! {
    static FAIL_CASE_RESERVATION_AFTER: Cell<Option<usize>> = const { Cell::new(None) };
}

/// Force the next QuickJS-equivalent case buffer allocation to fail. The hook
/// is thread-local so parallel test binaries cannot consume one another's
/// injected failure.
#[cfg(any(test, feature = "test-hooks"))]
pub fn fail_next_case_reservation_for_test() {
    fail_case_reservation_after_for_test(0);
}

#[cfg(any(test, feature = "test-hooks"))]
fn fail_case_reservation_after_for_test(successful_reservations: usize) {
    FAIL_CASE_RESERVATION_AFTER.with(|armed| {
        assert!(
            armed.replace(Some(successful_reservations)).is_none(),
            "case reservation failure was already armed"
        );
    });
}

fn try_reserve_exact<T>(values: &mut Vec<T>, additional: usize) -> Result<(), StringError> {
    #[cfg(any(test, feature = "test-hooks"))]
    if FAIL_CASE_RESERVATION_AFTER.with(|armed| match armed.get() {
        None => false,
        Some(0) => {
            armed.set(None);
            true
        }
        Some(remaining) => {
            armed.set(Some(remaining - 1));
            false
        }
    }) {
        return Err(StringError::OutOfMemory);
    }
    values
        .try_reserve_exact(additional)
        .map_err(|_| StringError::OutOfMemory)
}

enum CaseStorage {
    Latin1(Vec<u8>),
    Utf16(Vec<u16>),
}

/// Narrow-first, fallible equivalent of QuickJS's `StringBuffer`. `size` is
/// kept separately from the host Vec capacity because QuickJS grows its
/// logical buffer by 3/2 and widens the complete logical allocation.
struct CaseStringBuffer {
    storage: CaseStorage,
    size: usize,
    limit: usize,
}

impl CaseStringBuffer {
    fn new(initial_size: usize, limit: usize) -> Result<Self, StringError> {
        let limit = limit.min(MAX_STRING_LEN);
        let size = initial_size.min(limit);
        let mut values = Vec::new();
        try_reserve_exact(&mut values, size)?;
        Ok(Self {
            storage: CaseStorage::Latin1(values),
            size,
            limit,
        })
    }

    fn len(&self) -> usize {
        match &self.storage {
            CaseStorage::Latin1(values) => values.len(),
            CaseStorage::Utf16(values) => values.len(),
        }
    }

    fn grow_size(&self, new_len: usize) -> Result<usize, StringError> {
        if new_len > self.limit {
            return Err(StringError::TooLong);
        }
        Ok(new_len.max(self.size.saturating_mul(3) / 2).min(self.limit))
    }

    fn reserve_to<T>(values: &mut Vec<T>, size: usize) -> Result<(), StringError> {
        // Enter the reservation path even if the host allocator previously
        // returned spare capacity: QuickJS tracks its own logical `size`.
        try_reserve_exact(values, size.saturating_sub(values.len()))
    }

    fn widen(&mut self, size: usize) -> Result<(), StringError> {
        let CaseStorage::Latin1(narrow) =
            std::mem::replace(&mut self.storage, CaseStorage::Latin1(Vec::new()))
        else {
            unreachable!("case buffer was widened twice")
        };
        let mut wide = Vec::new();
        if let Err(error) = try_reserve_exact(&mut wide, size) {
            self.storage = CaseStorage::Latin1(narrow);
            return Err(error);
        }
        wide.extend(narrow.iter().copied().map(u16::from));
        self.storage = CaseStorage::Utf16(wide);
        self.size = size;
        Ok(())
    }

    fn push_code_unit(&mut self, unit: u16) -> Result<(), StringError> {
        let len = self.len();
        let new_len = len.checked_add(1).ok_or(StringError::TooLong)?;
        if new_len > self.limit {
            return Err(StringError::TooLong);
        }

        if len >= self.size {
            let new_size = self.grow_size(new_len)?;
            match &mut self.storage {
                CaseStorage::Latin1(_) if unit > u16::from(u8::MAX) => {
                    self.widen(new_size)?;
                }
                CaseStorage::Latin1(values) => {
                    Self::reserve_to(values, new_size)?;
                    self.size = new_size;
                }
                CaseStorage::Utf16(values) => {
                    Self::reserve_to(values, new_size)?;
                    self.size = new_size;
                }
            }
        } else if matches!(self.storage, CaseStorage::Latin1(_)) && unit > u16::from(u8::MAX) {
            self.widen(self.size)?;
        }

        match &mut self.storage {
            CaseStorage::Latin1(values) => values.push(unit as u8),
            CaseStorage::Utf16(values) => values.push(unit),
        }
        Ok(())
    }

    fn push_code_point(&mut self, code_point: u32) -> Result<(), StringError> {
        debug_assert!(code_point <= 0x10_ffff);
        if code_point < 0x1_0000 {
            self.push_code_unit(code_point as u16)
        } else {
            let adjusted = code_point - 0x1_0000;
            self.push_code_unit(0xd800 | (adjusted >> 10) as u16)?;
            self.push_code_unit(0xdc00 | (adjusted & 0x3ff) as u16)
        }
    }

    fn finish(self) -> CodeUnitBuf {
        match self.storage {
            CaseStorage::Latin1(values) => CodeUnitBuf::Latin1(values),
            CaseStorage::Utf16(values) => CodeUnitBuf::Utf16(values),
        }
    }
}

#[derive(Clone, Copy)]
struct CaseMapping {
    code_points: [u32; 3],
    len: usize,
}

impl CaseMapping {
    const fn one(code_point: u32) -> Self {
        Self {
            code_points: [code_point, 0, 0],
            len: 1,
        }
    }

    const fn two(first: u32, second: u32) -> Self {
        Self {
            code_points: [first, second, 0],
            len: 2,
        }
    }

    const fn three(first: u32, second: u32, third: u32) -> Self {
        Self {
            code_points: [first, second, third],
            len: 3,
        }
    }
}

fn case_conversion_entry(
    code_point: u32,
    conversion: u32,
    index: usize,
    entry: u32,
) -> CaseMapping {
    let run_type = (entry >> 4) & 0xf;
    let data = ((entry & 0xf) << 8) | u32::from(tables::CASE_CONV_TABLE2[index]);
    let run_start = entry >> 15;
    let is_lower = u32::from(conversion != 0);
    let mut converted = code_point;

    match run_type {
        RUN_TYPE_U | RUN_TYPE_L | RUN_TYPE_UF | RUN_TYPE_LF => {
            if conversion == (run_type & 1) || run_type >= RUN_TYPE_UF && conversion == 2 {
                let mapped_start = tables::CASE_CONV_TABLE1[data as usize] >> 15;
                converted = code_point - run_start + mapped_start;
            }
        }
        RUN_TYPE_UL => {
            let offset = code_point - run_start;
            if (offset & 1) == 1 - is_lower {
                converted = (offset ^ 1) + run_start;
            }
        }
        RUN_TYPE_LSU => {
            let offset = code_point - run_start;
            if offset == 1 {
                converted = (i64::from(code_point) + (2 * i64::from(is_lower) - 1)) as u32;
            } else if offset == (1 - is_lower) * 2 {
                converted = (i64::from(code_point) + (2 * i64::from(is_lower) - 1) * 2) as u32;
            }
        }
        RUN_TYPE_U2L_399_EXT2 => {
            if conversion != 0 {
                converted = code_point - run_start
                    + u32::from(tables::CASE_CONV_EXT[(data & 0x3f) as usize]);
            } else {
                return CaseMapping::two(
                    code_point - run_start + u32::from(tables::CASE_CONV_EXT[(data >> 6) as usize]),
                    0x399,
                );
            }
        }
        RUN_TYPE_UF_D20 => {
            if conversion != 1 {
                converted = data + u32::from(conversion == 2) * 0x20;
            }
        }
        RUN_TYPE_UF_D1_EXT => {
            if conversion != 1 {
                converted =
                    u32::from(tables::CASE_CONV_EXT[data as usize]) + u32::from(conversion == 2);
            }
        }
        RUN_TYPE_U_EXT | RUN_TYPE_LF_EXT => {
            if is_lower == run_type - RUN_TYPE_U_EXT {
                converted = u32::from(tables::CASE_CONV_EXT[data as usize]);
            }
        }
        RUN_TYPE_UF_EXT2 => {
            if conversion != 1 {
                return CaseMapping::two(
                    code_point - run_start + u32::from(tables::CASE_CONV_EXT[(data >> 6) as usize]),
                    u32::from(tables::CASE_CONV_EXT[(data & 0x3f) as usize]),
                );
            }
        }
        RUN_TYPE_LF_EXT2 => {
            if conversion != 0 {
                return CaseMapping::two(
                    code_point - run_start + u32::from(tables::CASE_CONV_EXT[(data >> 6) as usize]),
                    u32::from(tables::CASE_CONV_EXT[(data & 0x3f) as usize]),
                );
            }
        }
        RUN_TYPE_UF_EXT3 => {
            if conversion != 1 {
                return CaseMapping::three(
                    u32::from(tables::CASE_CONV_EXT[(data >> 8) as usize]),
                    u32::from(tables::CASE_CONV_EXT[((data >> 4) & 0xf) as usize]),
                    u32::from(tables::CASE_CONV_EXT[(data & 0xf) as usize]),
                );
            }
        }
        _ => unreachable!("invalid QuickJS case-conversion run type"),
    }
    CaseMapping::one(converted)
}

fn case_conversion(code_point: u32, to_lower: bool) -> CaseMapping {
    if code_point < 128 {
        let converted = if to_lower && (u32::from(b'A')..=u32::from(b'Z')).contains(&code_point) {
            code_point + u32::from(b'a' - b'A')
        } else if !to_lower && (u32::from(b'a')..=u32::from(b'z')).contains(&code_point) {
            code_point - u32::from(b'a' - b'A')
        } else {
            code_point
        };
        return CaseMapping::one(converted);
    }

    let mut lower = 0;
    let mut upper = tables::CASE_CONV_TABLE1.len();
    while lower < upper {
        let middle = (lower + upper) / 2;
        let entry = tables::CASE_CONV_TABLE1[middle];
        let run_start = entry >> 15;
        let run_len = (entry >> 8) & 0x7f;
        if code_point < run_start {
            upper = middle;
        } else if code_point >= run_start + run_len {
            lower = middle + 1;
        } else {
            return case_conversion_entry(code_point, u32::from(to_lower), middle, entry);
        }
    }
    CaseMapping::one(code_point)
}

/// Apply QuickJS's RegExp-specific canonicalization using the checksum-pinned
/// Unicode 17 tables. Legacy expressions use the single-code-point uppercase
/// rule; `u`/`v` expressions use Unicode simple case folding.
pub fn regexp_canonicalize(code_point: u32, unicode: bool) -> u32 {
    if code_point < 128 {
        if unicode && (u32::from(b'A')..=u32::from(b'Z')).contains(&code_point) {
            return code_point + u32::from(b'a' - b'A');
        }
        if !unicode && (u32::from(b'a')..=u32::from(b'z')).contains(&code_point) {
            return code_point - u32::from(b'a' - b'A');
        }
        return code_point;
    }

    let mut lower = 0;
    let mut upper = tables::CASE_CONV_TABLE1.len();
    while lower < upper {
        let middle = (lower + upper) / 2;
        let entry = tables::CASE_CONV_TABLE1[middle];
        let run_start = entry >> 15;
        let run_len = (entry >> 8) & 0x7f;
        if code_point < run_start {
            upper = middle;
        } else if code_point >= run_start + run_len {
            lower = middle + 1;
        } else {
            let mapping =
                case_conversion_entry(code_point, if unicode { 2 } else { 0 }, middle, entry);
            if mapping.len == 1 {
                let mapped = mapping.code_points[0];
                return if unicode || mapped >= 128 {
                    mapped
                } else {
                    code_point
                };
            }
            if unicode {
                return match code_point {
                    0xfb06 => 0xfb05,
                    0x1fd3 => 0x0390,
                    0x1fe3 => 0x03b0,
                    _ => code_point,
                };
            }
            return code_point;
        }
    }
    code_point
}

const CASE_U_RUN_MASK: u16 = (1_u16 << RUN_TYPE_U)
    | (1_u16 << RUN_TYPE_UF)
    | (1_u16 << RUN_TYPE_UL)
    | (1_u16 << RUN_TYPE_LSU)
    | (1_u16 << RUN_TYPE_U2L_399_EXT2)
    | (1_u16 << RUN_TYPE_UF_D20)
    | (1_u16 << RUN_TYPE_UF_D1_EXT)
    | (1_u16 << RUN_TYPE_U_EXT)
    | (1_u16 << RUN_TYPE_UF_EXT2)
    | (1_u16 << RUN_TYPE_UF_EXT3);

const CASE_F_RUN_MASK: u16 = (1_u16 << RUN_TYPE_UF)
    | (1_u16 << RUN_TYPE_LF)
    | (1_u16 << RUN_TYPE_UL)
    | (1_u16 << RUN_TYPE_LSU)
    | (1_u16 << RUN_TYPE_U2L_399_EXT2)
    | (1_u16 << RUN_TYPE_LF_EXT)
    | (1_u16 << RUN_TYPE_LF_EXT2)
    | (1_u16 << RUN_TYPE_UF_D20)
    | (1_u16 << RUN_TYPE_UF_D1_EXT)
    | (1_u16 << RUN_TYPE_UF_EXT2)
    | (1_u16 << RUN_TYPE_UF_EXT3);

/// Canonicalize inclusive ranges with QuickJS's RegExp case-folding
/// algorithm. Unlike a full-domain expansion, this visits only code points in
/// the case-conversion mask (1,585 points for Unicode 17).
pub fn regexp_canonicalize_range_pairs(ranges: &[(u32, u32)], unicode: bool) -> Vec<(u32, u32)> {
    let ranges = normalize_half_open_ranges(
        ranges
            .iter()
            .filter(|(start, end)| start <= end)
            .map(|(start, end)| (*start, end.saturating_add(1)))
            .collect(),
    );
    let mask = regexp_case_change_mask(unicode);
    let affected = intersect_half_open_ranges(&ranges, &mask);
    let unaffected = subtract_half_open_ranges(&ranges, &mask);

    let mut canonicalized = Vec::new();
    canonicalized.reserve(affected.iter().fold(0_usize, |count, (start, end)| {
        count.saturating_add((*end - *start) as usize)
    }));
    for (start, end) in affected {
        canonicalized.extend((start..end).map(|code_point| {
            let mapped = regexp_canonicalize(code_point, unicode);
            (mapped, mapped + 1)
        }));
    }
    canonicalized.extend(unaffected);
    normalize_half_open_ranges(canonicalized)
        .into_iter()
        .map(|(start, end)| (start, end - 1))
        .collect()
}

fn regexp_case_change_mask(unicode: bool) -> Vec<(u32, u32)> {
    let run_mask = if unicode {
        CASE_F_RUN_MASK
    } else {
        CASE_U_RUN_MASK
    };
    let mut ranges = Vec::new();
    for entry in tables::CASE_CONV_TABLE1 {
        let run_type = (entry >> 4) & 0xf;
        if run_mask & (1_u16 << run_type) == 0 {
            continue;
        }
        let mut start = entry >> 15;
        let len = (entry >> 8) & 0x7f;
        match run_type {
            RUN_TYPE_UL => {
                if !unicode {
                    start += 1;
                }
                for offset in (0..len).step_by(2) {
                    ranges.push((start + offset, start + offset + 1));
                }
            }
            RUN_TYPE_LSU => {
                if unicode {
                    ranges.push((start, start + 1));
                }
                ranges.push((start + 1, start + 2));
                if !unicode {
                    ranges.push((start + 2, start + 3));
                }
            }
            _ => ranges.push((start, start + len)),
        }
    }
    normalize_half_open_ranges(ranges)
}

fn intersect_half_open_ranges(left: &[(u32, u32)], right: &[(u32, u32)]) -> Vec<(u32, u32)> {
    let mut intersection = Vec::new();
    let mut left_index = 0;
    let mut right_index = 0;
    while left_index < left.len() && right_index < right.len() {
        let (left_start, left_end) = left[left_index];
        let (right_start, right_end) = right[right_index];
        let start = left_start.max(right_start);
        let end = left_end.min(right_end);
        if start < end {
            intersection.push((start, end));
        }
        if left_end <= right_end {
            left_index += 1;
        } else {
            right_index += 1;
        }
    }
    intersection
}

fn subtract_half_open_ranges(ranges: &[(u32, u32)], excluded: &[(u32, u32)]) -> Vec<(u32, u32)> {
    let mut result = Vec::new();
    let mut excluded_index = 0;
    for &(start, end) in ranges {
        while excluded_index < excluded.len() && excluded[excluded_index].1 <= start {
            excluded_index += 1;
        }
        let mut cursor = start;
        let mut index = excluded_index;
        while index < excluded.len() && excluded[index].0 < end {
            let (excluded_start, excluded_end) = excluded[index];
            if cursor < excluded_start {
                result.push((cursor, excluded_start.min(end)));
            }
            cursor = cursor.max(excluded_end);
            if cursor >= end {
                break;
            }
            index += 1;
        }
        if cursor < end {
            result.push((cursor, end));
        }
    }
    result
}

fn normalize_half_open_ranges(mut ranges: Vec<(u32, u32)>) -> Vec<(u32, u32)> {
    ranges.retain(|(start, end)| start < end);
    ranges.sort_unstable();
    let mut normalized: Vec<(u32, u32)> = Vec::with_capacity(ranges.len());
    for (start, end) in ranges {
        if let Some((_, previous_end)) = normalized.last_mut()
            && start <= *previous_end
        {
            *previous_end = (*previous_end).max(end);
        } else {
            normalized.push((start, end));
        }
    }
    normalized
}

fn is_cased(code_point: u32) -> bool {
    let mut lower = 0;
    let mut upper = tables::CASE_CONV_TABLE1.len();
    while lower < upper {
        let middle = (lower + upper) / 2;
        let entry = tables::CASE_CONV_TABLE1[middle];
        let run_start = entry >> 15;
        let run_len = (entry >> 8) & 0x7f;
        if code_point < run_start {
            upper = middle;
        } else if code_point >= run_start + run_len {
            lower = middle + 1;
        } else {
            return true;
        }
    }
    is_in_table(code_point, &tables::CASED_TABLE, &tables::CASED_INDEX)
}

fn is_case_ignorable(code_point: u32) -> bool {
    is_in_table(
        code_point,
        &tables::CASE_IGNORABLE_TABLE,
        &tables::CASE_IGNORABLE_INDEX,
    )
}

fn next_code_point<U: CodeUnits + ?Sized>(input: &U, index: &mut usize) -> u32 {
    let first = input
        .unit_at(*index)
        .expect("case conversion advanced past the input");
    *index += 1;
    if (0xd800..=0xdbff).contains(&first)
        && input
            .unit_at(*index)
            .is_some_and(|second| (0xdc00..=0xdfff).contains(&second))
    {
        let second = input
            .unit_at(*index)
            .expect("checked trail surrogate disappeared");
        *index += 1;
        0x1_0000 + ((u32::from(first) - 0xd800) << 10) + (u32::from(second) - 0xdc00)
    } else {
        u32::from(first)
    }
}

/// Return zero before the first character, matching QuickJS `string_prevc`.
fn previous_code_point<U: CodeUnits + ?Sized>(input: &U, index: &mut usize) -> u32 {
    if *index == 0 {
        return 0;
    }
    *index -= 1;
    let second = input
        .unit_at(*index)
        .expect("case conversion retreated past the input");
    if (0xdc00..=0xdfff).contains(&second) && *index > 0 {
        let first = input
            .unit_at(*index - 1)
            .expect("checked lead surrogate disappeared");
        if (0xd800..=0xdbff).contains(&first) {
            *index -= 1;
            return 0x1_0000 + ((u32::from(first) - 0xd800) << 10) + (u32::from(second) - 0xdc00);
        }
    }
    u32::from(second)
}

fn is_final_sigma<U: CodeUnits + ?Sized>(input: &U, sigma_position: usize) -> bool {
    let mut index = sigma_position;
    let previous = loop {
        let code_point = previous_code_point(input, &mut index);
        if !is_case_ignorable(code_point) {
            break code_point;
        }
    };
    if !is_cased(previous) {
        return false;
    }

    let mut index = sigma_position + 1;
    loop {
        if index >= input.unit_len() {
            return true;
        }
        let code_point = next_code_point(input, &mut index);
        if !is_case_ignorable(code_point) {
            return !is_cased(code_point);
        }
    }
}

/// Convert one raw UTF-16 string using the exact Unicode tables and contextual
/// lowercase rule bundled by QuickJS 2026-06-04. `upper = true` selects upper
/// case. Locale arguments are intentionally outside this kernel because
/// QuickJS's four String methods ignore them completely. `max_len` is
/// clamped to [`MAX_STRING_LEN`]. An empty input returns empty Latin-1
/// storage without reserving (callers that keep string identity return the
/// input itself instead).
pub fn convert_case_with_limit<U: CodeUnits + ?Sized>(
    input: &U,
    upper: bool,
    max_len: usize,
) -> Result<CodeUnitBuf, StringError> {
    if input.unit_len() == 0 {
        return Ok(CodeUnitBuf::Latin1(Vec::new()));
    }

    let mut output = CaseStringBuffer::new(input.unit_len(), max_len)?;
    let mut index = 0;
    while index < input.unit_len() {
        let position = index;
        let code_point = next_code_point(input, &mut index);
        let mapping = if !upper && code_point == 0x3a3 && is_final_sigma(input, position) {
            CaseMapping::one(0x3c2)
        } else {
            case_conversion(code_point, !upper)
        };
        for mapped in &mapping.code_points[..mapping.len] {
            output.push_code_point(*mapped)?;
        }
    }
    Ok(output.finish())
}

#[cfg(test)]
mod tests {
    use super::{
        case_conversion, convert_case_with_limit, fail_case_reservation_after_for_test,
        fail_next_case_reservation_for_test, is_case_ignorable, is_cased,
        normalize_half_open_ranges, regexp_canonicalize, regexp_canonicalize_range_pairs,
        regexp_case_change_mask, tables,
    };
    use crate::unicode::{CodeUnitBuf, MAX_STRING_LEN, StringError};

    fn js(text: &str) -> Vec<u16> {
        text.encode_utf16().collect()
    }

    fn convert(input: &[u16], upper: bool) -> CodeUnitBuf {
        convert_case_with_limit(input, upper, MAX_STRING_LEN).unwrap()
    }

    fn text(output: &CodeUnitBuf) -> String {
        String::from_utf16_lossy(&output.to_utf16())
    }

    #[test]
    fn generated_tables_are_pinned_to_quickjs_unicode_17() {
        assert_eq!(
            tables::SOURCE_SHA256,
            "cf782bc7a07549e976f606bd3cb8555858482b279574554dcb8d46412986006c"
        );
        assert_eq!(tables::CASE_CONV_TABLE1.len(), 378);
        assert_eq!(tables::CASE_CONV_TABLE2.len(), 378);
        assert_eq!(tables::CASE_CONV_EXT.len(), 58);
        assert_eq!(tables::CASED_TABLE.len(), 190);
        assert_eq!(tables::CASED_INDEX.len(), 18);
        assert_eq!(tables::CASE_IGNORABLE_TABLE.len(), 785);
        assert_eq!(tables::CASE_IGNORABLE_INDEX.len(), 75);
    }

    #[test]
    fn regexp_canonicalization_matches_legacy_and_unicode_rules() {
        assert_eq!(regexp_canonicalize(u32::from(b'a'), false), u32::from(b'A'));
        assert_eq!(regexp_canonicalize(u32::from(b'A'), false), u32::from(b'A'));
        assert_eq!(regexp_canonicalize(u32::from(b'A'), true), u32::from(b'a'));
        assert_eq!(regexp_canonicalize(u32::from(b'a'), true), u32::from(b'a'));

        // Unicode simple folding admits these two non-ASCII word characters;
        // legacy canonicalization must not map either one down into ASCII.
        assert_eq!(regexp_canonicalize(0x017f, false), 0x017f);
        assert_eq!(regexp_canonicalize(0x212a, false), 0x212a);
        assert_eq!(regexp_canonicalize(0x017f, true), u32::from(b's'));
        assert_eq!(regexp_canonicalize(0x212a, true), u32::from(b'k'));

        assert_eq!(regexp_canonicalize(0xfb06, true), 0xfb05);
        assert_eq!(regexp_canonicalize(0x1fd3, true), 0x0390);
        assert_eq!(regexp_canonicalize(0x1fe3, true), 0x03b0);
    }

    #[test]
    fn regexp_case_change_masks_match_quickjs_unicode_17_shape() {
        for (unicode, expected_ranges, expected_points) in
            [(false, 634, 1_580_u32), (true, 637, 1_585_u32)]
        {
            let mask = regexp_case_change_mask(unicode);
            assert_eq!(mask.len(), expected_ranges);
            assert_eq!(
                mask.iter().map(|(start, end)| end - start).sum::<u32>(),
                expected_points,
            );
        }
    }

    #[test]
    fn range_canonicalization_matches_pointwise_quickjs_mapping() {
        fn naive(ranges: &[(u32, u32)], unicode: bool) -> Vec<(u32, u32)> {
            let mapped = ranges
                .iter()
                .flat_map(|(start, end)| *start..=*end)
                .map(|code_point| {
                    let mapped = regexp_canonicalize(code_point, unicode);
                    (mapped, mapped + 1)
                })
                .collect();
            normalize_half_open_ranges(mapped)
                .into_iter()
                .map(|(start, end)| (start, end - 1))
                .collect()
        }

        let probes = [
            (0x0041, 0x007a),
            (0x017e, 0x0181),
            (0x0390, 0x03b0),
            (0x1fd3, 0x1fe3),
            (0xfb05, 0xfb06),
            (0x10400, 0x1044f),
        ];
        for unicode in [false, true] {
            assert_eq!(
                regexp_canonicalize_range_pairs(&probes, unicode),
                naive(&probes, unicode),
            );
            let complete_mask = regexp_case_change_mask(unicode)
                .into_iter()
                .map(|(start, end)| (start, end - 1))
                .collect::<Vec<_>>();
            assert_eq!(
                regexp_canonicalize_range_pairs(&complete_mask, unicode),
                naive(&complete_mask, unicode),
            );
        }

        assert_eq!(
            regexp_canonicalize_range_pairs(&[(0xfb06, 0xfb06)], true),
            [(0xfb05, 0xfb05)],
        );
        assert_eq!(
            regexp_canonicalize_range_pairs(&[(0x1fd3, 0x1fd3)], true),
            [(0x0390, 0x0390)],
        );
        assert_eq!(
            regexp_canonicalize_range_pairs(&[(0x1fe3, 0x1fe3)], true),
            [(0x03b0, 0x03b0)],
        );
    }

    #[test]
    fn full_unicode_range_canonicalization_stays_compact() {
        let ranges = regexp_canonicalize_range_pairs(&[(0, 0x10_ffff)], true);
        assert!(ranges.len() <= 700);
        for code_point in [
            0,
            u32::from(b'A'),
            u32::from(b'a'),
            0x017f,
            0x212a,
            0x1fd3,
            0xfb06,
            0x10_ffff,
        ] {
            let mapped = regexp_canonicalize(code_point, true);
            assert!(
                ranges
                    .iter()
                    .any(|(start, end)| mapped >= *start && mapped <= *end),
                "missing canonical representative {mapped:#x}",
            );
        }
    }

    #[test]
    fn conversion_ports_ascii_simple_extended_and_multicodepoint_runs() {
        let input = js("aZ ß ffi ʼn İ 𐐀𐐨");
        assert_eq!(text(&convert(&input, true)), "AZ SS FFI ʼN İ 𐐀𐐀");
        assert_eq!(text(&convert(&input, false)), "az ß ffi ʼn i\u{307} 𐐨𐐨");

        let greek = js("ΐ");
        assert_eq!(text(&convert(&greek, true)), "Ι\u{308}\u{301}");
    }

    #[test]
    fn lowercase_uses_quickjs_final_sigma_context() {
        for (source, expected) in [
            ("ΟΣ", "ος"),
            ("ΟΣΑ", "οσα"),
            ("Σ", "σ"),
            ("A\u{301}Σ", "a\u{301}ς"),
            ("AΣ\u{301}", "aς\u{301}"),
            ("AΣ\u{301}B", "aσ\u{301}b"),
            ("AΣ'", "aς'"),
            ("AΣ'B", "aσ'b"),
        ] {
            let source = js(source);
            assert_eq!(text(&convert(&source, false)), expected);
        }
    }

    #[test]
    fn raw_utf16_surrogates_are_preserved_while_pairs_are_decoded() {
        let input = [0xd800, 0x61, 0xdc00, 0xd801, 0xdc00, 0x5a, 0xdfff];
        assert_eq!(
            convert(&input, true).to_utf16(),
            [0xd800, 0x41, 0xdc00, 0xd801, 0xdc00, 0x5a, 0xdfff]
        );
        assert_eq!(
            convert(&input, false).to_utf16(),
            [0xd800, 0x61, 0xdc00, 0xd801, 0xdc28, 0x7a, 0xdfff]
        );
    }

    #[test]
    fn buffer_starts_narrow_then_widens_or_stays_narrow_from_content() {
        let narrows = js("Ÿ");
        let narrowed = convert(&narrows, false);
        assert_eq!(text(&narrowed), "ÿ");
        assert!(matches!(narrowed, CodeUnitBuf::Latin1(_)));

        let widens = [0xff_u8];
        let widened = convert_case_with_limit(&widens[..], true, MAX_STRING_LEN).unwrap();
        assert_eq!(text(&widened), "Ÿ");
        assert!(matches!(widened, CodeUnitBuf::Utf16(_)));
    }

    #[test]
    fn empty_identity_length_limit_and_reservation_failure_match_string_buffer() {
        let empty: [u16; 0] = [];
        fail_next_case_reservation_for_test();
        let converted_empty = convert_case_with_limit(&empty[..], true, 0).unwrap();
        assert_eq!(converted_empty, CodeUnitBuf::Latin1(Vec::new()));

        // The empty fast path did not consume the injected allocation failure.
        let ascii = js("a");
        assert_eq!(
            convert_case_with_limit(&ascii[..], true, 1),
            Err(StringError::OutOfMemory)
        );

        let sharp_s = js("ß");
        assert_eq!(
            convert_case_with_limit(&sharp_s[..], true, 1),
            Err(StringError::TooLong)
        );
        assert_eq!(
            text(&convert_case_with_limit(&sharp_s[..], false, 1).unwrap()),
            "ß"
        );
    }

    #[test]
    fn widening_and_growth_reservation_failures_are_recoverable() {
        let widens = js("ÿ");
        fail_case_reservation_after_for_test(1);
        assert_eq!(
            convert_case_with_limit(&widens[..], true, MAX_STRING_LEN),
            Err(StringError::OutOfMemory)
        );

        let expands = js("ß");
        fail_case_reservation_after_for_test(1);
        assert_eq!(
            convert_case_with_limit(&expands[..], true, MAX_STRING_LEN),
            Err(StringError::OutOfMemory)
        );
        assert_eq!(
            text(&convert_case_with_limit(&expands[..], true, MAX_STRING_LEN).unwrap()),
            "SS"
        );
    }

    #[test]
    fn cased_and_case_ignorable_property_boundaries_are_stable() {
        for code_point in [0x41, 0x2b0, 0x345, 0x1f88, 0x10400, 0x1d7c4] {
            assert!(is_cased(code_point), "U+{code_point:04X}");
        }
        for code_point in [0, 0x20, 0x300, 0x1f600, 0x10ffff] {
            assert!(!is_cased(code_point), "U+{code_point:04X}");
        }
        for code_point in [0x27, 0x2e, 0x300, 0x5bd, 0x200d, 0xe0100] {
            assert!(is_case_ignorable(code_point), "U+{code_point:04X}");
        }
        for code_point in [0, 0x41, 0x5be, 0x1f600, 0x10ffff] {
            assert!(!is_case_ignorable(code_point), "U+{code_point:04X}");
        }
    }

    #[test]
    fn table_mapping_has_expected_unicode_17_samples() {
        let lower = case_conversion(0x0130, true);
        assert_eq!(&lower.code_points[..lower.len], [0x69, 0x307]);
        let upper = case_conversion(0xfb03, false);
        assert_eq!(&upper.code_points[..upper.len], [0x46, 0x46, 0x49]);
        assert_eq!(case_conversion(0x10d70, false).code_points[0], 0x10d50);
        assert_eq!(case_conversion(0x10d50, true).code_points[0], 0x10d70);
    }

    #[test]
    fn exhaustive_kernel_fingerprint_matches_pinned_quickjs_c() {
        use crate::unicode::tests::hash_u32;

        let mut case_hash = 14_695_981_039_346_656_037_u64;
        let mut property_hash = 14_695_981_039_346_656_037_u64;
        let mut cased_count = 0_u32;
        let mut ignorable_count = 0_u32;
        for code_point in 0..=0x10_ffff {
            let cased = is_cased(code_point);
            let ignorable = is_case_ignorable(code_point);
            cased_count += u32::from(cased);
            ignorable_count += u32::from(ignorable);
            property_hash = hash_u32(property_hash, code_point);
            property_hash = hash_u32(property_hash, u32::from(cased));
            property_hash = hash_u32(property_hash, u32::from(ignorable));

            for (conversion, to_lower) in [(0, false), (1, true)] {
                let mapping = case_conversion(code_point, to_lower);
                case_hash = hash_u32(case_hash, code_point);
                case_hash = hash_u32(case_hash, conversion);
                case_hash = hash_u32(case_hash, mapping.len as u32);
                for mapped in &mapping.code_points[..mapping.len] {
                    case_hash = hash_u32(case_hash, *mapped);
                }
            }
        }

        assert_eq!(cased_count, 4_632);
        assert_eq!(ignorable_count, 2_794);
        assert_eq!(property_hash, 0x0f20_329d_dcab_a159);
        assert_eq!(case_hash, 0x24ab_cb80_d55c_587e);
    }
}