ferroni 1.5.0

// api.rs - Idiomatic Rust API for Ferroni.
//
// Wraps the C-ported internals (onig_new, onig_search, etc.) with
// Rust-native types: Regex, RegexBuilder, Match, Captures, FindIter.

use std::cell::RefCell;
use std::ops::Range;
use std::time::Duration;

use crate::encodings::utf8::ONIG_ENCODING_UTF8;
use crate::error::RegexError;
use crate::oniguruma::*;
use crate::regcomp::onig_new;
use crate::regexec::{
    OnigMatchParam, onig_name_to_backref_number, onig_new_match_param, onig_search,
    onig_search_with_param,
};
use crate::regint::RegexType;
use crate::regsyntax::OnigSyntaxOniguruma;

thread_local! {
    /// One reusable capture region per thread. Taking the value out keeps
    /// nested and re-entrant searches independent; returning the larger region
    /// retains the most useful allocation when result values overlap.
    static CACHED_REGION: RefCell<Option<OnigRegion>> = const { RefCell::new(None) };
}

fn take_cached_region() -> OnigRegion {
    CACHED_REGION
        .try_with(|cached| cached.borrow_mut().take())
        .ok()
        .flatten()
        .unwrap_or_default()
}

fn cache_region(region: OnigRegion) {
    // Drop can run after this thread-local's destructor during thread
    // teardown. In that case, simply let the uncached region be freed.
    let _ = CACHED_REGION.try_with(|cached| {
        let mut cached = cached.borrow_mut();
        let region_capacity = region.beg.capacity() + region.end.capacity();
        let should_replace = match cached.as_ref() {
            Some(current) => region_capacity > current.beg.capacity() + current.end.capacity(),
            None => true,
        };
        if should_replace {
            *cached = Some(region);
        }
    });
}

fn timeout_to_millis(timeout: Duration) -> u64 {
    if timeout.is_zero() {
        return 0;
    }

    let millis = timeout.as_millis();
    let millis = millis + u128::from(!timeout.subsec_nanos().is_multiple_of(1_000_000));
    millis.clamp(1, u64::MAX as u128) as u64
}

/// Limits for a single search.
///
/// Every limit left unset keeps its process-wide setting (for example from
/// [`onig_set_time_limit`](crate::regexec::onig_set_time_limit)). A limit set
/// to zero turns that limit off for this search, as in Oniguruma.
///
/// Pass the options to the `*_with` search methods of [`Regex`], such as
/// [`Regex::find_with`]. Unlike the plain methods, they report a search that
/// stopped at a limit as an error instead of as "no match".
///
/// # Examples
///
/// ```
/// use std::time::Duration;
/// use ferroni::prelude::*;
///
/// let re = Regex::new(r"(a+)+b").unwrap();
/// let options = SearchOptions::new().timeout(Duration::from_millis(50));
///
/// assert_eq!(re.find_with("aab", options).unwrap().unwrap().as_str(), "aab");
///
/// let hostile = "a".repeat(40);
/// assert!(re.find_with(&hostile, options).is_err());
/// ```
#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)]
pub struct SearchOptions {
    timeout: Option<Duration>,
    retry_limit_in_match: Option<u64>,
    retry_limit_in_search: Option<u64>,
    match_stack_limit: Option<u32>,
}

impl SearchOptions {
    /// Options that keep every process-wide limit.
    pub const fn new() -> Self {
        SearchOptions {
            timeout: None,
            retry_limit_in_match: None,
            retry_limit_in_search: None,
            match_stack_limit: None,
        }
    }

    /// Stop the search with [`RegexError::TimeLimitOver`] once `timeout` has
    /// passed. `Duration::ZERO` removes the time limit for this search.
    ///
    /// The engine reads the clock every 512 backtracks, so the search can run
    /// slightly past the deadline. Durations are rounded up to whole
    /// milliseconds.
    pub const fn timeout(mut self, timeout: Duration) -> Self {
        self.timeout = Some(timeout);
        self
    }

    /// Stop the search with [`RegexError::RetryLimitInMatchOver`] after `limit`
    /// backtracks at one start position. Zero removes the limit.
    pub const fn retry_limit_in_match(mut self, limit: u64) -> Self {
        self.retry_limit_in_match = Some(limit);
        self
    }

    /// Stop the search with [`RegexError::RetryLimitInSearchOver`] after `limit`
    /// backtracks in total. Zero removes the limit.
    pub const fn retry_limit_in_search(mut self, limit: u64) -> Self {
        self.retry_limit_in_search = Some(limit);
        self
    }

    /// Stop the search with [`RegexError::MatchStackLimitOver`] once the
    /// backtracking stack holds `limit` entries. Zero removes the limit.
    pub const fn match_stack_limit(mut self, limit: u32) -> Self {
        self.match_stack_limit = Some(limit);
        self
    }

    fn match_param(self) -> OnigMatchParam {
        let mut match_param = onig_new_match_param();
        if let Some(timeout) = self.timeout {
            match_param.time_limit = timeout_to_millis(timeout);
        }
        if let Some(limit) = self.retry_limit_in_match {
            match_param.retry_limit_in_match = limit;
        }
        if let Some(limit) = self.retry_limit_in_search {
            match_param.retry_limit_in_search = limit;
        }
        if let Some(limit) = self.match_stack_limit {
            match_param.match_stack_limit = limit;
        }
        match_param
    }
}

/// A compiled regular expression.
///
/// # Examples
///
/// ```
/// use ferroni::api::Regex;
///
/// let re = Regex::new(r"\d+").unwrap();
/// assert!(re.is_match("hello 42"));
///
/// let m = re.find("hello 42").unwrap();
/// assert_eq!(m.as_str(), "42");
/// assert_eq!(m.start(), 6);
/// assert_eq!(m.end(), 8);
/// ```
pub struct Regex {
    inner: RegexType,
}

impl Regex {
    /// Compile a pattern using default options (Oniguruma syntax, UTF-8, no flags).
    pub fn new(pattern: &str) -> Result<Regex, RegexError> {
        Self::new_bytes(pattern.as_bytes())
    }

    /// Compile a pattern from raw bytes using default options.
    pub fn new_bytes(pattern: &[u8]) -> Result<Regex, RegexError> {
        let inner = onig_new(
            pattern,
            ONIG_OPTION_NONE,
            &ONIG_ENCODING_UTF8,
            &OnigSyntaxOniguruma,
        )?;
        Ok(Regex { inner })
    }

    /// Create a [`RegexBuilder`] for fine-grained control over compilation.
    pub fn builder(pattern: &str) -> RegexBuilder {
        RegexBuilder::new(pattern)
    }

    /// Return the first match in `text`, or `None` if no match.
    ///
    /// A search that stops at a process-wide limit (time, retry or stack) is
    /// reported as no match. Use [`Regex::find_with`] with [`SearchOptions`] to tell
    /// the two apart.
    pub fn find<'t>(&self, text: &'t str) -> Option<Match<'t>> {
        self.find_bytes(text.as_bytes())
    }

    /// Return the first match in `text` (as bytes), or `None` if no match.
    pub fn find_bytes<'t>(&self, text: &'t [u8]) -> Option<Match<'t>> {
        let (result, region) = onig_search(
            &self.inner,
            text,
            text.len(),
            0,
            text.len(),
            Some(take_cached_region()),
            ONIG_OPTION_NONE,
        );
        let region = region?;
        if result < 0 {
            cache_region(region);
            return None;
        }
        if region.num_regs < 1 {
            cache_region(region);
            return None;
        }
        let m = Match::from_region(text, region.beg[0], region.end[0]);
        cache_region(region);
        m
    }

    /// Return the first match in `text` under per-search `options`.
    ///
    /// # Errors
    ///
    /// Returns the [`RegexError`] of a limit the search reached, such as
    /// [`RegexError::TimeLimitOver`], instead of treating it as "no match".
    pub fn find_with<'t>(
        &self,
        text: &'t str,
        options: SearchOptions,
    ) -> Result<Option<Match<'t>>, RegexError> {
        self.find_bytes_with(text.as_bytes(), options)
    }

    /// Return the first match in `text` (as bytes) under per-search `options`.
    ///
    /// # Errors
    ///
    /// Returns the [`RegexError`] of a limit the search reached.
    pub fn find_bytes_with<'t>(
        &self,
        text: &'t [u8],
        options: SearchOptions,
    ) -> Result<Option<Match<'t>>, RegexError> {
        let (result, region) =
            self.search_with(text, 0, text.len(), Some(take_cached_region()), options)?;
        let Some(region) = region else {
            return Ok(None);
        };
        if result < 0 || region.num_regs < 1 {
            cache_region(region);
            return Ok(None);
        }
        let m = Match::from_region(text, region.beg[0], region.end[0]);
        cache_region(region);
        Ok(m)
    }

    /// Check whether `text` matches the pattern anywhere.
    ///
    /// A search that stops at a process-wide limit (time, retry or stack) is
    /// reported as no match. Use [`Regex::is_match_with`] with [`SearchOptions`] to tell
    /// the two apart.
    pub fn is_match(&self, text: &str) -> bool {
        self.is_match_bytes(text.as_bytes())
    }

    /// Check whether `text` (as bytes) matches the pattern anywhere.
    pub fn is_match_bytes(&self, text: &[u8]) -> bool {
        let (result, _) = onig_search(
            &self.inner,
            text,
            text.len(),
            0,
            text.len(),
            None,
            ONIG_OPTION_NONE,
        );
        result >= 0
    }

    /// Check whether `text` matches under per-search `options`.
    ///
    /// # Errors
    ///
    /// Returns the [`RegexError`] of a limit the search reached.
    pub fn is_match_with(&self, text: &str, options: SearchOptions) -> Result<bool, RegexError> {
        self.is_match_bytes_with(text.as_bytes(), options)
    }

    /// Check whether `text` (as bytes) matches under per-search `options`.
    ///
    /// # Errors
    ///
    /// Returns the [`RegexError`] of a limit the search reached.
    pub fn is_match_bytes_with(
        &self,
        text: &[u8],
        options: SearchOptions,
    ) -> Result<bool, RegexError> {
        let (result, _) = self.search_with(text, 0, text.len(), None, options)?;
        Ok(result >= 0)
    }

    /// Return the first match with all capture groups, or `None`.
    ///
    /// A search that stops at a process-wide limit (time, retry or stack) is
    /// reported as no match. Use [`Regex::captures_with`] with [`SearchOptions`] to tell
    /// the two apart.
    pub fn captures<'t>(&'t self, text: &'t str) -> Option<Captures<'t>> {
        self.captures_bytes(text.as_bytes())
    }

    /// Return the first match with all capture groups (bytes), or `None`.
    pub fn captures_bytes<'t>(&'t self, text: &'t [u8]) -> Option<Captures<'t>> {
        let (result, region) = onig_search(
            &self.inner,
            text,
            text.len(),
            0,
            text.len(),
            Some(take_cached_region()),
            ONIG_OPTION_NONE,
        );
        let region = region?;
        if result < 0 {
            cache_region(region);
            return None;
        }
        Some(Captures {
            text,
            region,
            regex: self,
        })
    }

    /// Return the first match with all capture groups under per-search
    /// `options`.
    ///
    /// # Errors
    ///
    /// Returns the [`RegexError`] of a limit the search reached.
    pub fn captures_with<'t>(
        &'t self,
        text: &'t str,
        options: SearchOptions,
    ) -> Result<Option<Captures<'t>>, RegexError> {
        self.captures_bytes_with(text.as_bytes(), options)
    }

    /// Return the first match with all capture groups (bytes) under
    /// per-search `options`.
    ///
    /// # Errors
    ///
    /// Returns the [`RegexError`] of a limit the search reached.
    pub fn captures_bytes_with<'t>(
        &'t self,
        text: &'t [u8],
        options: SearchOptions,
    ) -> Result<Option<Captures<'t>>, RegexError> {
        let (result, region) =
            self.search_with(text, 0, text.len(), Some(take_cached_region()), options)?;
        let Some(region) = region else {
            return Ok(None);
        };
        if result < 0 {
            cache_region(region);
            return Ok(None);
        }
        Ok(Some(Captures {
            text,
            region,
            regex: self,
        }))
    }

    /// Iterate over all non-overlapping matches in `text`.
    ///
    /// Iteration ends early when a search stops at a process-wide limit. Use
    /// [`Regex::find_iter_with`] with [`SearchOptions`] to see that error.
    pub fn find_iter<'r, 't>(&'r self, text: &'t str) -> FindIter<'r, 't> {
        FindIter {
            regex: self,
            text: text.as_bytes(),
            last_end: 0,
            last_was_empty: false,
            region: take_cached_region(),
        }
    }

    /// Iterate over all non-overlapping matches in `text` (as bytes).
    pub fn find_iter_bytes<'r, 't>(&'r self, text: &'t [u8]) -> FindIter<'r, 't> {
        FindIter {
            regex: self,
            text,
            last_end: 0,
            last_was_empty: false,
            region: take_cached_region(),
        }
    }

    /// Iterate over all non-overlapping matches in `text` under per-search
    /// `options`.
    ///
    /// The options apply to each search the iterator runs. A search that
    /// reaches a limit yields its error once and ends the iteration.
    pub fn find_iter_with<'r, 't>(
        &'r self,
        text: &'t str,
        options: SearchOptions,
    ) -> TryFindIter<'r, 't> {
        self.find_iter_bytes_with(text.as_bytes(), options)
    }

    /// Iterate over all non-overlapping matches in `text` (as bytes) under
    /// per-search `options`.
    pub fn find_iter_bytes_with<'r, 't>(
        &'r self,
        text: &'t [u8],
        options: SearchOptions,
    ) -> TryFindIter<'r, 't> {
        TryFindIter {
            regex: self,
            text,
            last_end: 0,
            last_was_empty: false,
            region: take_cached_region(),
            options,
            finished: false,
        }
    }

    fn search_with(
        &self,
        text: &[u8],
        start: usize,
        range: usize,
        region: Option<OnigRegion>,
        options: SearchOptions,
    ) -> Result<(i32, Option<OnigRegion>), RegexError> {
        let (result, region) = onig_search_with_param(
            &self.inner,
            text,
            text.len(),
            start,
            range,
            region,
            ONIG_OPTION_NONE,
            &options.match_param(),
        );
        if result < 0 && result != ONIG_MISMATCH {
            if let Some(region) = region {
                cache_region(region);
            }
            return Err(RegexError::from(result));
        }
        Ok((result, region))
    }

    /// Return the number of capture groups in the pattern (excluding group 0).
    pub fn captures_len(&self) -> usize {
        self.inner.num_mem as usize
    }

    /// Access the underlying `RegexType` for advanced / C-style usage.
    pub fn as_raw(&self) -> &RegexType {
        &self.inner
    }
}

impl std::fmt::Debug for Regex {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("Regex").finish_non_exhaustive()
    }
}

// === RegexBuilder ===

/// Builder for compiling a [`Regex`] with custom options.
///
/// # Examples
///
/// ```
/// use ferroni::api::Regex;
///
/// let re = Regex::builder(r"hello world")
///     .case_insensitive(true)
///     .build()
///     .unwrap();
/// assert!(re.is_match("Hello World"));
/// ```
pub struct RegexBuilder {
    pattern: Vec<u8>,
    options: OnigOptionType,
    syntax: &'static OnigSyntaxType,
}

impl RegexBuilder {
    /// Create a new builder for the given pattern.
    pub fn new(pattern: &str) -> Self {
        RegexBuilder {
            pattern: pattern.as_bytes().to_vec(),
            options: ONIG_OPTION_NONE,
            syntax: &OnigSyntaxOniguruma,
        }
    }

    /// Enable or disable case-insensitive matching.
    pub fn case_insensitive(mut self, yes: bool) -> Self {
        if yes {
            self.options |= ONIG_OPTION_IGNORECASE;
        } else {
            self.options &= !ONIG_OPTION_IGNORECASE;
        }
        self
    }

    /// Enable or disable multiline mode (`.` matches `\n`).
    pub fn dot_matches_newline(mut self, yes: bool) -> Self {
        if yes {
            self.options |= ONIG_OPTION_MULTILINE;
        } else {
            self.options &= !ONIG_OPTION_MULTILINE;
        }
        self
    }

    /// Enable or disable `^`/`$` matching at every line boundary.
    ///
    /// With the default Oniguruma syntax, this behavior is enabled by default.
    /// When disabled, anchors match only at input boundaries. Calling this
    /// method overrides the selected syntax's default anchor behavior.
    pub fn multi_line_anchors(mut self, yes: bool) -> Self {
        if yes {
            self.options |= ONIG_OPTION_NEGATE_SINGLELINE;
            self.options &= !ONIG_OPTION_SINGLELINE;
        } else {
            self.options &= !ONIG_OPTION_NEGATE_SINGLELINE;
            self.options |= ONIG_OPTION_SINGLELINE;
        }
        self
    }

    /// Enable or disable extended mode (whitespace and `#` comments ignored).
    pub fn extended(mut self, yes: bool) -> Self {
        if yes {
            self.options |= ONIG_OPTION_EXTEND;
        } else {
            self.options &= !ONIG_OPTION_EXTEND;
        }
        self
    }

    /// Set a raw option flag. See `ONIG_OPTION_*` constants.
    pub fn option(mut self, flag: OnigOptionType) -> Self {
        self.options |= flag;
        self
    }

    /// Select the syntax definition to use (default: Oniguruma).
    ///
    /// Pass one of the `OnigSyntax*` statics from [`crate::regsyntax`].
    pub fn syntax(mut self, syntax: &'static OnigSyntaxType) -> Self {
        self.syntax = syntax;
        self
    }

    /// Compile the pattern into a [`Regex`].
    pub fn build(self) -> Result<Regex, RegexError> {
        let inner = onig_new(
            &self.pattern,
            self.options,
            &ONIG_ENCODING_UTF8,
            self.syntax,
        )?;
        Ok(Regex { inner })
    }
}

// === Match ===

/// A single match result referencing the original text.
#[derive(Debug, Clone, Copy)]
pub struct Match<'t> {
    text: &'t [u8],
    start: usize,
    end: usize,
}

impl<'t> Match<'t> {
    /// Build a match from one region entry, or `None` if the entry is not a
    /// range of `text`: unset (negative), past the end, or with start > end.
    /// Oniguruma can report start > end for a capture whose group started
    /// again but failed before closing; every `Match` upholds
    /// `start <= end <= text.len()`, so its accessors never panic on slicing.
    fn from_region(text: &'t [u8], beg: i32, end: i32) -> Option<Self> {
        let start = usize::try_from(beg).ok()?;
        let end = usize::try_from(end).ok()?;
        (start <= end && end <= text.len()).then_some(Match { text, start, end })
    }

    /// Byte offset of the start of the match.
    pub fn start(&self) -> usize {
        self.start
    }

    /// Byte offset of the end of the match (exclusive).
    pub fn end(&self) -> usize {
        self.end
    }

    /// Byte range of the match.
    pub fn range(&self) -> Range<usize> {
        self.start..self.end
    }

    /// The matched text as a byte slice.
    pub fn as_bytes(&self) -> &'t [u8] {
        &self.text[self.start..self.end]
    }

    /// The matched text as a `&str`.
    ///
    /// # Panics
    ///
    /// Panics if the matched bytes are not valid UTF-8.
    pub fn as_str(&self) -> &'t str {
        std::str::from_utf8(self.as_bytes()).expect("match is not valid UTF-8")
    }

    /// Returns the length of the match in bytes.
    pub fn len(&self) -> usize {
        self.end - self.start
    }

    /// Returns `true` if the match is empty (zero-length).
    pub fn is_empty(&self) -> bool {
        self.start == self.end
    }
}

// === Captures ===

/// All capture groups from a single match.
///
/// Group 0 is the entire match. Groups 1..N correspond to `(...)` in the pattern.
pub struct Captures<'t> {
    text: &'t [u8],
    region: OnigRegion,
    regex: &'t Regex,
}

impl<'t> Captures<'t> {
    /// Get capture group `i`, or `None` if the group did not participate.
    ///
    /// Group 0 is the entire match.
    ///
    /// Like Oniguruma, the engine can leave a capture with its start after
    /// its end: a group that matched once, then started again and failed
    /// before closing, keeps the new start and the old end (for example
    /// group 1 of `((?=(a|ab))a?){2}` against `"a"` is `1..0`). Such a
    /// capture is not a range of the text and is reported as not
    /// participating (`None`), here and in [`Captures::iter`] and
    /// [`Captures::name`]. The raw values stay available through the
    /// low-level [`OnigRegion`].
    pub fn get(&self, i: usize) -> Option<Match<'t>> {
        if i >= self.region.num_regs as usize {
            return None;
        }
        Match::from_region(self.text, self.region.beg[i], self.region.end[i])
    }

    /// Get the last capture group with the given name that participated, or `None`.
    pub fn name(&self, name: &str) -> Option<Match<'t>> {
        let num =
            onig_name_to_backref_number(&self.regex.inner, name.as_bytes(), Some(&self.region))
                .ok()?;
        self.get(num as usize)
    }

    /// Number of capture groups (including group 0).
    pub fn len(&self) -> usize {
        self.region.num_regs as usize
    }

    /// Returns `true` if there are no capture groups (should never happen for a valid match).
    pub fn is_empty(&self) -> bool {
        self.region.num_regs == 0
    }

    /// Iterate over all capture groups.
    pub fn iter(&self) -> CapturesIter<'_, 't> {
        CapturesIter {
            captures: self,
            index: 0,
        }
    }
}

impl std::fmt::Debug for Captures<'_> {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        let mut list = f.debug_list();
        for i in 0..self.len() {
            list.entry(&self.get(i));
        }
        list.finish()
    }
}

impl Drop for Captures<'_> {
    fn drop(&mut self) {
        cache_region(std::mem::take(&mut self.region));
    }
}

// === CapturesIter ===

/// Iterator over capture groups in a [`Captures`].
pub struct CapturesIter<'c, 't> {
    captures: &'c Captures<'t>,
    index: usize,
}

impl<'c, 't> Iterator for CapturesIter<'c, 't> {
    type Item = Option<Match<'t>>;

    fn next(&mut self) -> Option<Self::Item> {
        if self.index >= self.captures.len() {
            return None;
        }
        let m = self.captures.get(self.index);
        self.index += 1;
        Some(m)
    }

    fn size_hint(&self) -> (usize, Option<usize>) {
        let remaining = self.captures.len() - self.index;
        (remaining, Some(remaining))
    }
}

impl ExactSizeIterator for CapturesIter<'_, '_> {}

// === FindIter ===

/// Iterator over all non-overlapping matches in a text.
pub struct FindIter<'r, 't> {
    regex: &'r Regex,
    text: &'t [u8],
    last_end: usize,
    last_was_empty: bool,
    region: OnigRegion,
}

impl<'r, 't> Iterator for FindIter<'r, 't> {
    type Item = Match<'t>;

    fn next(&mut self) -> Option<Match<'t>> {
        if self.last_end > self.text.len() {
            return None;
        }

        let (result, region) = onig_search(
            &self.regex.inner,
            self.text,
            self.text.len(),
            self.last_end,
            self.text.len(),
            Some(std::mem::take(&mut self.region)),
            ONIG_OPTION_NONE,
        );
        self.region = region?;

        if result < 0 {
            return None;
        }
        if self.region.num_regs < 1 {
            return None;
        }

        let m = Match::from_region(self.text, self.region.beg[0], self.region.end[0])?;
        let (start, end) = (m.start, m.end);

        // Handle empty matches: advance by one byte to avoid infinite loop.
        if start == end {
            if self.last_was_empty {
                if self.last_end >= self.text.len() {
                    return None;
                }
                // Skip one character to avoid infinite loop on empty match
                self.last_end += self
                    .regex
                    .inner
                    .enc
                    .mbc_enc_len(&self.text[self.last_end..]);
                self.last_was_empty = false;
                return self.next();
            }
            self.last_was_empty = true;
        } else {
            self.last_was_empty = false;
        }

        self.last_end = end;

        Some(m)
    }
}

impl Drop for FindIter<'_, '_> {
    fn drop(&mut self) {
        cache_region(std::mem::take(&mut self.region));
    }
}

/// Iterator over matches that applies [`SearchOptions`] to each search.
///
/// Created by [`Regex::find_iter_with`]. A search that reaches a limit is
/// yielded once as `Err` and ends the iterator.
pub struct TryFindIter<'r, 't> {
    regex: &'r Regex,
    text: &'t [u8],
    last_end: usize,
    last_was_empty: bool,
    region: OnigRegion,
    options: SearchOptions,
    finished: bool,
}

impl<'r, 't> Iterator for TryFindIter<'r, 't> {
    type Item = Result<Match<'t>, RegexError>;

    fn next(&mut self) -> Option<Self::Item> {
        if self.finished || self.last_end > self.text.len() {
            self.finished = true;
            return None;
        }

        let search = self.regex.search_with(
            self.text,
            self.last_end,
            self.text.len(),
            Some(std::mem::take(&mut self.region)),
            self.options,
        );
        let (result, region) = match search {
            Ok(result) => result,
            Err(error) => {
                self.finished = true;
                return Some(Err(error));
            }
        };
        let Some(region) = region else {
            self.finished = true;
            return None;
        };
        self.region = region;

        if result < 0 {
            self.finished = true;
            return None;
        }
        if self.region.num_regs < 1 {
            self.finished = true;
            return None;
        }

        let Some(m) = Match::from_region(self.text, self.region.beg[0], self.region.end[0]) else {
            self.finished = true;
            return None;
        };
        let (start, end) = (m.start, m.end);

        if start == end {
            if self.last_was_empty {
                if self.last_end >= self.text.len() {
                    self.finished = true;
                    return None;
                }
                self.last_end += self
                    .regex
                    .inner
                    .enc
                    .mbc_enc_len(&self.text[self.last_end..]);
                self.last_was_empty = false;
                return self.next();
            }
            self.last_was_empty = true;
        } else {
            self.last_was_empty = false;
        }

        self.last_end = end;
        Some(Ok(m))
    }
}

impl std::iter::FusedIterator for TryFindIter<'_, '_> {}

impl Drop for TryFindIter<'_, '_> {
    fn drop(&mut self) {
        cache_region(std::mem::take(&mut self.region));
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    fn assert_send_sync<T: Send + Sync>() {}

    fn clear_cached_region() {
        CACHED_REGION.with(|cached| *cached.borrow_mut() = None);
    }

    fn cached_region_buffer() -> Option<(*const i32, usize)> {
        CACHED_REGION.with(|cached| {
            cached
                .borrow()
                .as_ref()
                .map(|region| (region.beg.as_ptr(), region.beg.capacity()))
        })
    }

    #[test]
    fn compiled_regex_types_are_send_and_sync() {
        assert_send_sync::<Regex>();
        assert_send_sync::<crate::regset::OnigRegSet>();
        assert_send_sync::<crate::scanner::Scanner>();
    }

    #[test]
    fn find_reuses_the_thread_local_region_buffer() {
        clear_cached_region();
        let re = Regex::new(r"(a)(b)(c)").unwrap();

        assert_eq!(re.find("abc").unwrap().as_str(), "abc");
        let first = cached_region_buffer().expect("find should return its region to the cache");
        assert!(first.1 >= 4);

        assert_eq!(re.find("abc").unwrap().as_str(), "abc");
        let second = cached_region_buffer().expect("find should preserve the cached region");
        assert_eq!(second, first);
    }

    #[test]
    fn captures_returns_its_region_buffer_when_dropped() {
        clear_cached_region();
        let re = Regex::new(r"(a)(b)(c)").unwrap();
        let captures = re.captures("abc").unwrap();
        let buffer = captures.region.beg.as_ptr();

        assert!(cached_region_buffer().is_none());
        drop(captures);

        let cached = cached_region_buffer().expect("drop should return the captures region");
        assert_eq!(cached.0, buffer);
        assert!(cached.1 >= 4);
    }

    #[test]
    fn result_drop_during_thread_local_teardown_does_not_panic() {
        thread_local! {
            static LATE_RESULT: RefCell<Option<Captures<'static>>> = const { RefCell::new(None) };
        }

        std::thread::spawn(|| {
            // Initialize the result holder before CACHED_REGION so it is
            // destroyed afterwards and drops Captures with the cache gone.
            LATE_RESULT.with(|result| assert!(result.borrow().is_none()));
            let regex = Box::leak(Box::new(Regex::new(r"(a)").unwrap()));
            let captures = regex.captures("a").unwrap();
            LATE_RESULT.with(|result| *result.borrow_mut() = Some(captures));
        })
        .join()
        .expect("result drop during thread teardown must not panic");
    }

    #[test]
    fn find_iter_reuses_one_region_for_every_step() {
        clear_cached_region();
        let re = Regex::new(r"(\w+)").unwrap();
        let mut matches = re.find_iter("one two");

        assert_eq!(matches.next().unwrap().as_str(), "one");
        let first = matches.region.beg.as_ptr();
        assert_eq!(matches.next().unwrap().as_str(), "two");
        assert_eq!(matches.region.beg.as_ptr(), first);
        drop(matches);

        let cached = cached_region_buffer().expect("iterator drop should return its region");
        assert_eq!(cached.0, first);
    }

    #[test]
    fn regex_new_and_find() {
        let re = Regex::new(r"\d+").unwrap();
        let m = re.find("hello 42 world").unwrap();
        assert_eq!(m.as_str(), "42");
        assert_eq!(m.start(), 6);
        assert_eq!(m.end(), 8);
        assert_eq!(m.range(), 6..8);
        assert_eq!(m.len(), 2);
        assert!(!m.is_empty());
    }

    #[test]
    fn regex_no_match() {
        let re = Regex::new(r"\d+").unwrap();
        assert!(re.find("no digits here").is_none());
    }

    #[test]
    fn regex_is_match() {
        let re = Regex::new(r"hello").unwrap();
        assert!(re.is_match("say hello"));
        assert!(!re.is_match("say goodbye"));
    }

    #[test]
    fn search_options_methods_return_matches_and_captures() {
        let _lock = crate::regexec::LIMIT_TEST_LOCK.lock().unwrap();
        let re = Regex::new(r"(\w+)").unwrap();
        let options = SearchOptions::new().timeout(Duration::from_secs(1));

        assert_eq!(
            re.find_with("hello 42", options).unwrap().unwrap().as_str(),
            "hello"
        );
        assert_eq!(
            re.find_bytes_with(b"hello 42", options)
                .unwrap()
                .unwrap()
                .as_str(),
            "hello"
        );
        assert!(re.is_match_with("hello", options).unwrap());
        assert!(!re.is_match_bytes_with(b"!!!", options).unwrap());
        assert_eq!(
            re.captures_with("hello 42", options)
                .unwrap()
                .unwrap()
                .get(1)
                .unwrap()
                .as_str(),
            "hello"
        );
        assert!(re.captures_bytes_with(b"!!!", options).unwrap().is_none());
    }

    #[test]
    fn search_options_surface_time_limit_errors() {
        let _lock = crate::regexec::LIMIT_TEST_LOCK.lock().unwrap();
        let re = Regex::new(r"(a+)+b").unwrap();
        let text = "a".repeat(40);
        let options = SearchOptions::new()
            .timeout(Duration::from_millis(1))
            .retry_limit_in_match(0)
            .retry_limit_in_search(0);

        assert!(matches!(
            re.find_with(&text, options),
            Err(RegexError::TimeLimitOver)
        ));
        assert!(matches!(
            re.is_match_with(&text, options),
            Err(RegexError::TimeLimitOver)
        ));
        assert!(matches!(
            re.captures_with(&text, options),
            Err(RegexError::TimeLimitOver)
        ));
    }

    #[test]
    fn search_options_timeout_spans_every_start_position() {
        let _lock = crate::regexec::LIMIT_TEST_LOCK.lock().unwrap();
        // Each start position backtracks only ~400 times; the limit must still
        // fire for the search as a whole.
        let re = Regex::new(r"a{1,400}?(?=b)").unwrap();
        let text = "a".repeat(500_000);
        let options = SearchOptions::new().timeout(Duration::from_millis(10));

        let started = std::time::Instant::now();
        assert!(matches!(
            re.find_with(&text, options),
            Err(RegexError::TimeLimitOver)
        ));
        assert!(started.elapsed() < Duration::from_secs(1));
    }

    #[test]
    fn search_options_surface_retry_limit_errors() {
        let _lock = crate::regexec::LIMIT_TEST_LOCK.lock().unwrap();
        let re = Regex::new(r"(a+)+b").unwrap();
        let text = "a".repeat(30);

        assert!(matches!(
            re.find_with(&text, SearchOptions::new().retry_limit_in_match(1_000)),
            Err(RegexError::RetryLimitInMatchOver)
        ));
        assert!(matches!(
            re.find_with(
                &text,
                SearchOptions::new()
                    .retry_limit_in_match(0)
                    .retry_limit_in_search(1_000)
            ),
            Err(RegexError::RetryLimitInSearchOver)
        ));
    }

    #[test]
    fn search_options_leave_unset_limits_process_wide() {
        let _lock = crate::regexec::LIMIT_TEST_LOCK.lock().unwrap();
        let old_retry = crate::regexec::onig_get_retry_limit_in_match();
        crate::regexec::onig_set_retry_limit_in_match(1_000);
        let re = Regex::new(r"(a+)+b").unwrap();
        let text = "a".repeat(30);

        let unset = re.find_with(&text, SearchOptions::new());
        let disabled = re.find_with(
            &text,
            SearchOptions::new()
                .retry_limit_in_match(0)
                .retry_limit_in_search(0)
                .timeout(Duration::from_millis(50)),
        );
        crate::regexec::onig_set_retry_limit_in_match(old_retry);

        assert!(matches!(unset, Err(RegexError::RetryLimitInMatchOver)));
        assert!(matches!(disabled, Err(RegexError::TimeLimitOver)));
    }

    #[test]
    fn search_options_iterator_yields_matches_and_then_finishes() {
        let _lock = crate::regexec::LIMIT_TEST_LOCK.lock().unwrap();
        let re = Regex::new(r"\w+").unwrap();
        let options = SearchOptions::new().timeout(Duration::from_secs(1));
        let mut matches = re.find_iter_with("one two", options);

        assert_eq!(matches.next().unwrap().unwrap().as_str(), "one");
        assert_eq!(matches.next().unwrap().unwrap().as_str(), "two");
        assert!(matches.next().is_none());
        assert!(matches.next().is_none());
        let bytes: Vec<_> = re
            .find_iter_bytes_with(b"a b", options)
            .map(|m| m.unwrap().as_bytes().to_vec())
            .collect();
        assert_eq!(bytes, vec![b"a".to_vec(), b"b".to_vec()]);
    }

    #[test]
    fn search_options_iterator_surfaces_an_error_once() {
        let _lock = crate::regexec::LIMIT_TEST_LOCK.lock().unwrap();
        let re = Regex::new(r"(a+)+b").unwrap();
        let text = "a".repeat(40);
        let options = SearchOptions::new()
            .timeout(Duration::from_millis(1))
            .retry_limit_in_match(0)
            .retry_limit_in_search(0);
        let mut matches = re.find_iter_with(&text, options);

        assert!(matches!(
            matches.next().unwrap(),
            Err(RegexError::TimeLimitOver)
        ));
        assert!(matches.next().is_none());
    }

    #[test]
    fn search_options_default_keeps_every_limit_unset() {
        assert_eq!(SearchOptions::default(), SearchOptions::new());
        assert_send_sync::<SearchOptions>();
        assert_send_sync::<TryFindIter<'static, 'static>>();
    }

    #[test]
    fn timeout_is_rounded_up_to_milliseconds() {
        assert_eq!(timeout_to_millis(Duration::ZERO), 0);
        assert_eq!(timeout_to_millis(Duration::from_nanos(1)), 1);
        assert_eq!(timeout_to_millis(Duration::from_millis(1)), 1);
        assert_eq!(timeout_to_millis(Duration::from_nanos(1_000_001)), 2);
        assert_eq!(timeout_to_millis(Duration::MAX), u64::MAX);
    }

    #[test]
    fn regex_captures() {
        let re = Regex::new(r"(\d{4})-(\d{2})-(\d{2})").unwrap();
        let caps = re.captures("date: 2026-02-14").unwrap();
        assert_eq!(caps.get(0).unwrap().as_str(), "2026-02-14");
        assert_eq!(caps.get(1).unwrap().as_str(), "2026");
        assert_eq!(caps.get(2).unwrap().as_str(), "02");
        assert_eq!(caps.get(3).unwrap().as_str(), "14");
        assert!(caps.get(4).is_none());
        assert_eq!(caps.len(), 4);
    }

    #[test]
    fn regex_captures_len() {
        let re = Regex::new(r"(a)(b)(c)").unwrap();
        assert_eq!(re.captures_len(), 3);
    }

    #[test]
    fn regex_find_iter() {
        let re = Regex::new(r"\d+").unwrap();
        let matches: Vec<&str> = re.find_iter("1 + 22 = 333").map(|m| m.as_str()).collect();
        assert_eq!(matches, vec!["1", "22", "333"]);
    }

    #[test]
    fn regex_builder_case_insensitive() {
        let re = Regex::builder(r"hello")
            .case_insensitive(true)
            .build()
            .unwrap();
        assert!(re.is_match("HELLO"));
        assert!(re.is_match("Hello"));
    }

    #[test]
    fn regex_invalid_pattern() {
        let err = Regex::new(r"(unclosed").unwrap_err();
        assert!(matches!(err, RegexError::Syntax { .. }));
    }

    #[test]
    fn match_as_bytes() {
        let re = Regex::new(r"world").unwrap();
        let m = re.find("hello world").unwrap();
        assert_eq!(m.as_bytes(), b"world");
    }

    #[test]
    fn captures_iter() {
        let re = Regex::new(r"(a)(b)?").unwrap();
        let caps = re.captures("a").unwrap();
        let items: Vec<_> = caps.iter().collect();
        // group 0 = "a", group 1 = "a", group 2 = None (didn't participate)
        assert_eq!(items.len(), 3);
        assert!(items[0].is_some());
        assert!(items[1].is_some());
        assert!(items[2].is_none());
    }

    #[test]
    fn named_captures() {
        let re = Regex::new(r"(?<year>\d{4})-(?<month>\d{2})").unwrap();
        let caps = re.captures("2026-02").unwrap();
        assert_eq!(caps.name("year").unwrap().as_str(), "2026");
        assert_eq!(caps.name("month").unwrap().as_str(), "02");
        assert!(caps.name("day").is_none());
    }

    #[test]
    fn empty_match_find_iter() {
        let re = Regex::new(r"").unwrap();
        let matches: Vec<_> = re.find_iter("ab").collect();
        // Should yield empty matches at positions 0, 1, 2
        assert_eq!(matches.len(), 3);
        assert_eq!(matches[0].start(), 0);
        assert_eq!(matches[1].start(), 1);
        assert_eq!(matches[2].start(), 2);
    }
}