1 //! The string Pattern API.
3 //! For more details, see the traits [`Pattern`], [`Searcher`],
4 //! [`ReverseSearcher`], and [`DoubleEndedSearcher`].
8 reason = "API not fully fleshed out and ready to be stabilized",
14 use crate::slice::memchr;
21 /// A `Pattern<'a>` expresses that the implementing type
22 /// can be used as a string pattern for searching in a `&'a str`.
24 /// For example, both `'a'` and `"aa"` are patterns that
25 /// would match at index `1` in the string `"baaaab"`.
27 /// The trait itself acts as a builder for an associated
28 /// `Searcher` type, which does the actual work of finding
29 /// occurrences of the pattern in a string.
30 pub trait Pattern<'a>: Sized {
31 /// Associated searcher for this pattern
32 type Searcher: Searcher<'a>;
34 /// Constructs the associated searcher from
35 /// `self` and the `haystack` to search in.
36 fn into_searcher(self, haystack: &'a str) -> Self::Searcher;
38 /// Checks whether the pattern matches anywhere in the haystack
40 fn is_contained_in(self, haystack: &'a str) -> bool {
41 self.into_searcher(haystack).next_match().is_some()
44 /// Checks whether the pattern matches at the front of the haystack
46 fn is_prefix_of(self, haystack: &'a str) -> bool {
47 matches!(self.into_searcher(haystack).next(), SearchStep::Match(0, _))
50 /// Checks whether the pattern matches at the back of the haystack
52 fn is_suffix_of(self, haystack: &'a str) -> bool
54 Self::Searcher: ReverseSearcher<'a>,
56 matches!(self.into_searcher(haystack).next_back(), SearchStep::Match(_, j) if haystack.len() == j)
62 /// Result of calling `Searcher::next()` or `ReverseSearcher::next_back()`.
63 #[derive(Copy, Clone, Eq, PartialEq, Debug)]
65 /// Expresses that a match of the pattern has been found at
68 /// Expresses that `haystack[a..b]` has been rejected as a possible match
71 /// Note that there might be more than one `Reject` between two `Match`es,
72 /// there is no requirement for them to be combined into one.
74 /// Expresses that every byte of the haystack has been visited, ending
79 /// A searcher for a string pattern.
81 /// This trait provides methods for searching for non-overlapping
82 /// matches of a pattern starting from the front (left) of a string.
84 /// It will be implemented by associated `Searcher`
85 /// types of the `Pattern` trait.
87 /// The trait is marked unsafe because the indices returned by the
88 /// `next()` methods are required to lie on valid utf8 boundaries in
89 /// the haystack. This enables consumers of this trait to
90 /// slice the haystack without additional runtime checks.
91 pub unsafe trait Searcher<'a> {
92 /// Getter for the underlying string to be searched in
94 /// Will always return the same `&str`
95 fn haystack(&self) -> &'a str;
97 /// Performs the next search step starting from the front.
99 /// - Returns `Match(a, b)` if `haystack[a..b]` matches the pattern.
100 /// - Returns `Reject(a, b)` if `haystack[a..b]` can not match the
101 /// pattern, even partially.
102 /// - Returns `Done` if every byte of the haystack has been visited
104 /// The stream of `Match` and `Reject` values up to a `Done`
105 /// will contain index ranges that are adjacent, non-overlapping,
106 /// covering the whole haystack, and laying on utf8 boundaries.
108 /// A `Match` result needs to contain the whole matched pattern,
109 /// however `Reject` results may be split up into arbitrary
110 /// many adjacent fragments. Both ranges may have zero length.
112 /// As an example, the pattern `"aaa"` and the haystack `"cbaaaaab"`
113 /// might produce the stream
114 /// `[Reject(0, 1), Reject(1, 2), Match(2, 5), Reject(5, 8)]`
115 fn next(&mut self) -> SearchStep;
117 /// Finds the next `Match` result. See `next()`
119 /// Unlike next(), there is no guarantee that the returned ranges
120 /// of this and next_reject will overlap. This will return (start_match, end_match),
121 /// where start_match is the index of where the match begins, and end_match is
122 /// the index after the end of the match.
124 fn next_match(&mut self) -> Option<(usize, usize)> {
127 SearchStep::Match(a, b) => return Some((a, b)),
128 SearchStep::Done => return None,
134 /// Finds the next `Reject` result. See `next()` and `next_match()`
136 /// Unlike next(), there is no guarantee that the returned ranges
137 /// of this and next_match will overlap.
139 fn next_reject(&mut self) -> Option<(usize, usize)> {
142 SearchStep::Reject(a, b) => return Some((a, b)),
143 SearchStep::Done => return None,
150 /// A reverse searcher for a string pattern.
152 /// This trait provides methods for searching for non-overlapping
153 /// matches of a pattern starting from the back (right) of a string.
155 /// It will be implemented by associated `Searcher`
156 /// types of the `Pattern` trait if the pattern supports searching
157 /// for it from the back.
159 /// The index ranges returned by this trait are not required
160 /// to exactly match those of the forward search in reverse.
162 /// For the reason why this trait is marked unsafe, see them
163 /// parent trait `Searcher`.
164 pub unsafe trait ReverseSearcher<'a>: Searcher<'a> {
165 /// Performs the next search step starting from the back.
167 /// - Returns `Match(a, b)` if `haystack[a..b]` matches the pattern.
168 /// - Returns `Reject(a, b)` if `haystack[a..b]` can not match the
169 /// pattern, even partially.
170 /// - Returns `Done` if every byte of the haystack has been visited
172 /// The stream of `Match` and `Reject` values up to a `Done`
173 /// will contain index ranges that are adjacent, non-overlapping,
174 /// covering the whole haystack, and laying on utf8 boundaries.
176 /// A `Match` result needs to contain the whole matched pattern,
177 /// however `Reject` results may be split up into arbitrary
178 /// many adjacent fragments. Both ranges may have zero length.
180 /// As an example, the pattern `"aaa"` and the haystack `"cbaaaaab"`
181 /// might produce the stream
182 /// `[Reject(7, 8), Match(4, 7), Reject(1, 4), Reject(0, 1)]`
183 fn next_back(&mut self) -> SearchStep;
185 /// Finds the next `Match` result. See `next_back()`
187 fn next_match_back(&mut self) -> Option<(usize, usize)> {
189 match self.next_back() {
190 SearchStep::Match(a, b) => return Some((a, b)),
191 SearchStep::Done => return None,
197 /// Finds the next `Reject` result. See `next_back()`
199 fn next_reject_back(&mut self) -> Option<(usize, usize)> {
201 match self.next_back() {
202 SearchStep::Reject(a, b) => return Some((a, b)),
203 SearchStep::Done => return None,
210 /// A marker trait to express that a `ReverseSearcher`
211 /// can be used for a `DoubleEndedIterator` implementation.
213 /// For this, the impl of `Searcher` and `ReverseSearcher` need
214 /// to follow these conditions:
216 /// - All results of `next()` need to be identical
217 /// to the results of `next_back()` in reverse order.
218 /// - `next()` and `next_back()` need to behave as
219 /// the two ends of a range of values, that is they
220 /// can not "walk past each other".
224 /// `char::Searcher` is a `DoubleEndedSearcher` because searching for a
225 /// `char` only requires looking at one at a time, which behaves the same
228 /// `(&str)::Searcher` is not a `DoubleEndedSearcher` because
229 /// the pattern `"aa"` in the haystack `"aaa"` matches as either
230 /// `"[aa]a"` or `"a[aa]"`, depending from which side it is searched.
231 pub trait DoubleEndedSearcher<'a>: ReverseSearcher<'a> {}
233 /////////////////////////////////////////////////////////////////////////////
235 /////////////////////////////////////////////////////////////////////////////
237 /// Associated type for `<char as Pattern<'a>>::Searcher`.
238 #[derive(Clone, Debug)]
239 pub struct CharSearcher<'a> {
241 // safety invariant: `finger`/`finger_back` must be a valid utf8 byte index of `haystack`
242 // This invariant can be broken *within* next_match and next_match_back, however
243 // they must exit with fingers on valid code point boundaries.
244 /// `finger` is the current byte index of the forward search.
245 /// Imagine that it exists before the byte at its index, i.e.
246 /// `haystack[finger]` is the first byte of the slice we must inspect during
247 /// forward searching
249 /// `finger_back` is the current byte index of the reverse search.
250 /// Imagine that it exists after the byte at its index, i.e.
251 /// haystack[finger_back - 1] is the last byte of the slice we must inspect during
252 /// forward searching (and thus the first byte to be inspected when calling next_back())
254 /// The character being searched for
257 // safety invariant: `utf8_size` must be less than 5
258 /// The number of bytes `needle` takes up when encoded in utf8
260 /// A utf8 encoded copy of the `needle`
261 utf8_encoded: [u8; 4],
264 unsafe impl<'a> Searcher<'a> for CharSearcher<'a> {
266 fn haystack(&self) -> &'a str {
270 fn next(&mut self) -> SearchStep {
271 let old_finger = self.finger;
272 // SAFETY: 1-4 guarantee safety of `get_unchecked`
273 // 1. `self.finger` and `self.finger_back` are kept on unicode boundaries
274 // (this is invariant)
275 // 2. `self.finger >= 0` since it starts at 0 and only increases
276 // 3. `self.finger < self.finger_back` because otherwise the char `iter`
277 // would return `SearchStep::Done`
278 // 4. `self.finger` comes before the end of the haystack because `self.finger_back`
279 // starts at the end and only decreases
280 let slice = unsafe { self.haystack.get_unchecked(old_finger..self.finger_back) };
281 let mut iter = slice.chars();
282 let old_len = iter.iter.len();
283 if let Some(ch) = iter.next() {
284 // add byte offset of current character
285 // without re-encoding as utf-8
286 self.finger += old_len - iter.iter.len();
287 if ch == self.needle {
288 SearchStep::Match(old_finger, self.finger)
290 SearchStep::Reject(old_finger, self.finger)
297 fn next_match(&mut self) -> Option<(usize, usize)> {
299 // get the haystack after the last character found
300 let bytes = self.haystack.as_bytes().get(self.finger..self.finger_back)?;
301 // the last byte of the utf8 encoded needle
302 // SAFETY: we have an invariant that `utf8_size < 5`
303 let last_byte = unsafe { *self.utf8_encoded.get_unchecked(self.utf8_size - 1) };
304 if let Some(index) = memchr::memchr(last_byte, bytes) {
305 // The new finger is the index of the byte we found,
306 // plus one, since we memchr'd for the last byte of the character.
308 // Note that this doesn't always give us a finger on a UTF8 boundary.
309 // If we *didn't* find our character
310 // we may have indexed to the non-last byte of a 3-byte or 4-byte character.
311 // We can't just skip to the next valid starting byte because a character like
312 // ꁁ (U+A041 YI SYLLABLE PA), utf-8 `EA 81 81` will have us always find
313 // the second byte when searching for the third.
315 // However, this is totally okay. While we have the invariant that
316 // self.finger is on a UTF8 boundary, this invariant is not relied upon
317 // within this method (it is relied upon in CharSearcher::next()).
319 // We only exit this method when we reach the end of the string, or if we
320 // find something. When we find something the `finger` will be set
321 // to a UTF8 boundary.
322 self.finger += index + 1;
323 if self.finger >= self.utf8_size {
324 let found_char = self.finger - self.utf8_size;
325 if let Some(slice) = self.haystack.as_bytes().get(found_char..self.finger) {
326 if slice == &self.utf8_encoded[0..self.utf8_size] {
327 return Some((found_char, self.finger));
332 // found nothing, exit
333 self.finger = self.finger_back;
339 // let next_reject use the default implementation from the Searcher trait
342 unsafe impl<'a> ReverseSearcher<'a> for CharSearcher<'a> {
344 fn next_back(&mut self) -> SearchStep {
345 let old_finger = self.finger_back;
346 // SAFETY: see the comment for next() above
347 let slice = unsafe { self.haystack.get_unchecked(self.finger..old_finger) };
348 let mut iter = slice.chars();
349 let old_len = iter.iter.len();
350 if let Some(ch) = iter.next_back() {
351 // subtract byte offset of current character
352 // without re-encoding as utf-8
353 self.finger_back -= old_len - iter.iter.len();
354 if ch == self.needle {
355 SearchStep::Match(self.finger_back, old_finger)
357 SearchStep::Reject(self.finger_back, old_finger)
364 fn next_match_back(&mut self) -> Option<(usize, usize)> {
365 let haystack = self.haystack.as_bytes();
367 // get the haystack up to but not including the last character searched
368 let bytes = if let Some(slice) = haystack.get(self.finger..self.finger_back) {
373 // the last byte of the utf8 encoded needle
374 // SAFETY: we have an invariant that `utf8_size < 5`
375 let last_byte = unsafe { *self.utf8_encoded.get_unchecked(self.utf8_size - 1) };
376 if let Some(index) = memchr::memrchr(last_byte, bytes) {
377 // we searched a slice that was offset by self.finger,
378 // add self.finger to recoup the original index
379 let index = self.finger + index;
380 // memrchr will return the index of the byte we wish to
381 // find. In case of an ASCII character, this is indeed
382 // were we wish our new finger to be ("after" the found
383 // char in the paradigm of reverse iteration). For
384 // multibyte chars we need to skip down by the number of more
385 // bytes they have than ASCII
386 let shift = self.utf8_size - 1;
388 let found_char = index - shift;
389 if let Some(slice) = haystack.get(found_char..(found_char + self.utf8_size)) {
390 if slice == &self.utf8_encoded[0..self.utf8_size] {
391 // move finger to before the character found (i.e., at its start index)
392 self.finger_back = found_char;
393 return Some((self.finger_back, self.finger_back + self.utf8_size));
397 // We can't use finger_back = index - size + 1 here. If we found the last char
398 // of a different-sized character (or the middle byte of a different character)
399 // we need to bump the finger_back down to `index`. This similarly makes
400 // `finger_back` have the potential to no longer be on a boundary,
401 // but this is OK since we only exit this function on a boundary
402 // or when the haystack has been searched completely.
404 // Unlike next_match this does not
405 // have the problem of repeated bytes in utf-8 because
406 // we're searching for the last byte, and we can only have
407 // found the last byte when searching in reverse.
408 self.finger_back = index;
410 self.finger_back = self.finger;
411 // found nothing, exit
417 // let next_reject_back use the default implementation from the Searcher trait
420 impl<'a> DoubleEndedSearcher<'a> for CharSearcher<'a> {}
422 /// Searches for chars that are equal to a given char
423 impl<'a> Pattern<'a> for char {
424 type Searcher = CharSearcher<'a>;
427 fn into_searcher(self, haystack: &'a str) -> Self::Searcher {
428 let mut utf8_encoded = [0; 4];
429 let utf8_size = self.encode_utf8(&mut utf8_encoded).len();
433 finger_back: haystack.len(),
441 fn is_contained_in(self, haystack: &'a str) -> bool {
442 if (self as u32) < 128 {
443 haystack.as_bytes().contains(&(self as u8))
445 let mut buffer = [0u8; 4];
446 self.encode_utf8(&mut buffer).is_contained_in(haystack)
451 fn is_prefix_of(self, haystack: &'a str) -> bool {
452 self.encode_utf8(&mut [0u8; 4]).is_prefix_of(haystack)
456 fn is_suffix_of(self, haystack: &'a str) -> bool
458 Self::Searcher: ReverseSearcher<'a>,
460 self.encode_utf8(&mut [0u8; 4]).is_suffix_of(haystack)
464 /////////////////////////////////////////////////////////////////////////////
465 // Impl for a MultiCharEq wrapper
466 /////////////////////////////////////////////////////////////////////////////
470 fn matches(&mut self, c: char) -> bool;
473 impl<F> MultiCharEq for F
475 F: FnMut(char) -> bool,
478 fn matches(&mut self, c: char) -> bool {
483 impl MultiCharEq for &[char] {
485 fn matches(&mut self, c: char) -> bool {
486 self.iter().any(|&m| m == c)
490 struct MultiCharEqPattern<C: MultiCharEq>(C);
492 #[derive(Clone, Debug)]
493 struct MultiCharEqSearcher<'a, C: MultiCharEq> {
496 char_indices: super::CharIndices<'a>,
499 impl<'a, C: MultiCharEq> Pattern<'a> for MultiCharEqPattern<C> {
500 type Searcher = MultiCharEqSearcher<'a, C>;
503 fn into_searcher(self, haystack: &'a str) -> MultiCharEqSearcher<'a, C> {
504 MultiCharEqSearcher { haystack, char_eq: self.0, char_indices: haystack.char_indices() }
508 unsafe impl<'a, C: MultiCharEq> Searcher<'a> for MultiCharEqSearcher<'a, C> {
510 fn haystack(&self) -> &'a str {
515 fn next(&mut self) -> SearchStep {
516 let s = &mut self.char_indices;
517 // Compare lengths of the internal byte slice iterator
518 // to find length of current char
519 let pre_len = s.iter.iter.len();
520 if let Some((i, c)) = s.next() {
521 let len = s.iter.iter.len();
522 let char_len = pre_len - len;
523 if self.char_eq.matches(c) {
524 return SearchStep::Match(i, i + char_len);
526 return SearchStep::Reject(i, i + char_len);
533 unsafe impl<'a, C: MultiCharEq> ReverseSearcher<'a> for MultiCharEqSearcher<'a, C> {
535 fn next_back(&mut self) -> SearchStep {
536 let s = &mut self.char_indices;
537 // Compare lengths of the internal byte slice iterator
538 // to find length of current char
539 let pre_len = s.iter.iter.len();
540 if let Some((i, c)) = s.next_back() {
541 let len = s.iter.iter.len();
542 let char_len = pre_len - len;
543 if self.char_eq.matches(c) {
544 return SearchStep::Match(i, i + char_len);
546 return SearchStep::Reject(i, i + char_len);
553 impl<'a, C: MultiCharEq> DoubleEndedSearcher<'a> for MultiCharEqSearcher<'a, C> {}
555 /////////////////////////////////////////////////////////////////////////////
557 macro_rules! pattern_methods {
558 ($t:ty, $pmap:expr, $smap:expr) => {
562 fn into_searcher(self, haystack: &'a str) -> $t {
563 ($smap)(($pmap)(self).into_searcher(haystack))
567 fn is_contained_in(self, haystack: &'a str) -> bool {
568 ($pmap)(self).is_contained_in(haystack)
572 fn is_prefix_of(self, haystack: &'a str) -> bool {
573 ($pmap)(self).is_prefix_of(haystack)
577 fn is_suffix_of(self, haystack: &'a str) -> bool
578 where $t: ReverseSearcher<'a>
580 ($pmap)(self).is_suffix_of(haystack)
585 macro_rules! searcher_methods {
588 fn haystack(&self) -> &'a str {
592 fn next(&mut self) -> SearchStep {
596 fn next_match(&mut self) -> Option<(usize, usize)> {
600 fn next_reject(&mut self) -> Option<(usize, usize)> {
606 fn next_back(&mut self) -> SearchStep {
610 fn next_match_back(&mut self) -> Option<(usize, usize)> {
611 self.0.next_match_back()
614 fn next_reject_back(&mut self) -> Option<(usize, usize)> {
615 self.0.next_reject_back()
620 /////////////////////////////////////////////////////////////////////////////
622 /////////////////////////////////////////////////////////////////////////////
624 // Todo: Change / Remove due to ambiguity in meaning.
626 /// Associated type for `<&[char] as Pattern<'a>>::Searcher`.
627 #[derive(Clone, Debug)]
628 pub struct CharSliceSearcher<'a, 'b>(<MultiCharEqPattern<&'b [char]> as Pattern<'a>>::Searcher);
630 unsafe impl<'a, 'b> Searcher<'a> for CharSliceSearcher<'a, 'b> {
631 searcher_methods!(forward);
634 unsafe impl<'a, 'b> ReverseSearcher<'a> for CharSliceSearcher<'a, 'b> {
635 searcher_methods!(reverse);
638 impl<'a, 'b> DoubleEndedSearcher<'a> for CharSliceSearcher<'a, 'b> {}
640 /// Searches for chars that are equal to any of the chars in the array
641 impl<'a, 'b> Pattern<'a> for &'b [char] {
642 pattern_methods!(CharSliceSearcher<'a, 'b>, MultiCharEqPattern, CharSliceSearcher);
645 /////////////////////////////////////////////////////////////////////////////
646 // Impl for F: FnMut(char) -> bool
647 /////////////////////////////////////////////////////////////////////////////
649 /// Associated type for `<F as Pattern<'a>>::Searcher`.
651 pub struct CharPredicateSearcher<'a, F>(<MultiCharEqPattern<F> as Pattern<'a>>::Searcher)
653 F: FnMut(char) -> bool;
655 impl<F> fmt::Debug for CharPredicateSearcher<'_, F>
657 F: FnMut(char) -> bool,
659 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
660 f.debug_struct("CharPredicateSearcher")
661 .field("haystack", &self.0.haystack)
662 .field("char_indices", &self.0.char_indices)
666 unsafe impl<'a, F> Searcher<'a> for CharPredicateSearcher<'a, F>
668 F: FnMut(char) -> bool,
670 searcher_methods!(forward);
673 unsafe impl<'a, F> ReverseSearcher<'a> for CharPredicateSearcher<'a, F>
675 F: FnMut(char) -> bool,
677 searcher_methods!(reverse);
680 impl<'a, F> DoubleEndedSearcher<'a> for CharPredicateSearcher<'a, F> where F: FnMut(char) -> bool {}
682 /// Searches for chars that match the given predicate
683 impl<'a, F> Pattern<'a> for F
685 F: FnMut(char) -> bool,
687 pattern_methods!(CharPredicateSearcher<'a, F>, MultiCharEqPattern, CharPredicateSearcher);
690 /////////////////////////////////////////////////////////////////////////////
692 /////////////////////////////////////////////////////////////////////////////
694 /// Delegates to the `&str` impl.
695 impl<'a, 'b, 'c> Pattern<'a> for &'c &'b str {
696 pattern_methods!(StrSearcher<'a, 'b>, |&s| s, |s| s);
699 /////////////////////////////////////////////////////////////////////////////
701 /////////////////////////////////////////////////////////////////////////////
703 /// Non-allocating substring search.
705 /// Will handle the pattern `""` as returning empty matches at each character
707 impl<'a, 'b> Pattern<'a> for &'b str {
708 type Searcher = StrSearcher<'a, 'b>;
711 fn into_searcher(self, haystack: &'a str) -> StrSearcher<'a, 'b> {
712 StrSearcher::new(haystack, self)
715 /// Checks whether the pattern matches at the front of the haystack
717 fn is_prefix_of(self, haystack: &'a str) -> bool {
718 haystack.as_bytes().starts_with(self.as_bytes())
721 /// Checks whether the pattern matches at the back of the haystack
723 fn is_suffix_of(self, haystack: &'a str) -> bool {
724 haystack.as_bytes().ends_with(self.as_bytes())
728 /////////////////////////////////////////////////////////////////////////////
729 // Two Way substring searcher
730 /////////////////////////////////////////////////////////////////////////////
732 #[derive(Clone, Debug)]
733 /// Associated type for `<&str as Pattern<'a>>::Searcher`.
734 pub struct StrSearcher<'a, 'b> {
738 searcher: StrSearcherImpl,
741 #[derive(Clone, Debug)]
742 enum StrSearcherImpl {
744 TwoWay(TwoWaySearcher),
747 #[derive(Clone, Debug)]
755 impl<'a, 'b> StrSearcher<'a, 'b> {
756 fn new(haystack: &'a str, needle: &'b str) -> StrSearcher<'a, 'b> {
757 if needle.is_empty() {
761 searcher: StrSearcherImpl::Empty(EmptyNeedle {
772 searcher: StrSearcherImpl::TwoWay(TwoWaySearcher::new(
781 unsafe impl<'a, 'b> Searcher<'a> for StrSearcher<'a, 'b> {
783 fn haystack(&self) -> &'a str {
788 fn next(&mut self) -> SearchStep {
789 match self.searcher {
790 StrSearcherImpl::Empty(ref mut searcher) => {
791 // empty needle rejects every char and matches every empty string between them
792 let is_match = searcher.is_match_fw;
793 searcher.is_match_fw = !searcher.is_match_fw;
794 let pos = searcher.position;
795 match self.haystack[pos..].chars().next() {
796 _ if is_match => SearchStep::Match(pos, pos),
797 None => SearchStep::Done,
799 searcher.position += ch.len_utf8();
800 SearchStep::Reject(pos, searcher.position)
804 StrSearcherImpl::TwoWay(ref mut searcher) => {
805 // TwoWaySearcher produces valid *Match* indices that split at char boundaries
806 // as long as it does correct matching and that haystack and needle are
808 // *Rejects* from the algorithm can fall on any indices, but we will walk them
809 // manually to the next character boundary, so that they are utf-8 safe.
810 if searcher.position == self.haystack.len() {
811 return SearchStep::Done;
813 let is_long = searcher.memory == usize::MAX;
814 match searcher.next::<RejectAndMatch>(
815 self.haystack.as_bytes(),
816 self.needle.as_bytes(),
819 SearchStep::Reject(a, mut b) => {
820 // skip to next char boundary
821 while !self.haystack.is_char_boundary(b) {
824 searcher.position = cmp::max(b, searcher.position);
825 SearchStep::Reject(a, b)
827 otherwise => otherwise,
834 fn next_match(&mut self) -> Option<(usize, usize)> {
835 match self.searcher {
836 StrSearcherImpl::Empty(..) => loop {
838 SearchStep::Match(a, b) => return Some((a, b)),
839 SearchStep::Done => return None,
840 SearchStep::Reject(..) => {}
843 StrSearcherImpl::TwoWay(ref mut searcher) => {
844 let is_long = searcher.memory == usize::MAX;
845 // write out `true` and `false` cases to encourage the compiler
846 // to specialize the two cases separately.
848 searcher.next::<MatchOnly>(
849 self.haystack.as_bytes(),
850 self.needle.as_bytes(),
854 searcher.next::<MatchOnly>(
855 self.haystack.as_bytes(),
856 self.needle.as_bytes(),
865 unsafe impl<'a, 'b> ReverseSearcher<'a> for StrSearcher<'a, 'b> {
867 fn next_back(&mut self) -> SearchStep {
868 match self.searcher {
869 StrSearcherImpl::Empty(ref mut searcher) => {
870 let is_match = searcher.is_match_bw;
871 searcher.is_match_bw = !searcher.is_match_bw;
872 let end = searcher.end;
873 match self.haystack[..end].chars().next_back() {
874 _ if is_match => SearchStep::Match(end, end),
875 None => SearchStep::Done,
877 searcher.end -= ch.len_utf8();
878 SearchStep::Reject(searcher.end, end)
882 StrSearcherImpl::TwoWay(ref mut searcher) => {
883 if searcher.end == 0 {
884 return SearchStep::Done;
886 let is_long = searcher.memory == usize::MAX;
887 match searcher.next_back::<RejectAndMatch>(
888 self.haystack.as_bytes(),
889 self.needle.as_bytes(),
892 SearchStep::Reject(mut a, b) => {
893 // skip to next char boundary
894 while !self.haystack.is_char_boundary(a) {
897 searcher.end = cmp::min(a, searcher.end);
898 SearchStep::Reject(a, b)
900 otherwise => otherwise,
907 fn next_match_back(&mut self) -> Option<(usize, usize)> {
908 match self.searcher {
909 StrSearcherImpl::Empty(..) => loop {
910 match self.next_back() {
911 SearchStep::Match(a, b) => return Some((a, b)),
912 SearchStep::Done => return None,
913 SearchStep::Reject(..) => {}
916 StrSearcherImpl::TwoWay(ref mut searcher) => {
917 let is_long = searcher.memory == usize::MAX;
918 // write out `true` and `false`, like `next_match`
920 searcher.next_back::<MatchOnly>(
921 self.haystack.as_bytes(),
922 self.needle.as_bytes(),
926 searcher.next_back::<MatchOnly>(
927 self.haystack.as_bytes(),
928 self.needle.as_bytes(),
937 /// The internal state of the two-way substring search algorithm.
938 #[derive(Clone, Debug)]
939 struct TwoWaySearcher {
941 /// critical factorization index
943 /// critical factorization index for reversed needle
944 crit_pos_back: usize,
946 /// `byteset` is an extension (not part of the two way algorithm);
947 /// it's a 64-bit "fingerprint" where each set bit `j` corresponds
948 /// to a (byte & 63) == j present in the needle.
954 /// index into needle before which we have already matched
956 /// index into needle after which we have already matched
961 This is the Two-Way search algorithm, which was introduced in the paper:
962 Crochemore, M., Perrin, D., 1991, Two-way string-matching, Journal of the ACM 38(3):651-675.
964 Here's some background information.
966 A *word* is a string of symbols. The *length* of a word should be a familiar
967 notion, and here we denote it for any word x by |x|.
968 (We also allow for the possibility of the *empty word*, a word of length zero).
970 If x is any non-empty word, then an integer p with 0 < p <= |x| is said to be a
971 *period* for x iff for all i with 0 <= i <= |x| - p - 1, we have x[i] == x[i+p].
972 For example, both 1 and 2 are periods for the string "aa". As another example,
973 the only period of the string "abcd" is 4.
975 We denote by period(x) the *smallest* period of x (provided that x is non-empty).
976 This is always well-defined since every non-empty word x has at least one period,
977 |x|. We sometimes call this *the period* of x.
979 If u, v and x are words such that x = uv, where uv is the concatenation of u and
980 v, then we say that (u, v) is a *factorization* of x.
982 Let (u, v) be a factorization for a word x. Then if w is a non-empty word such
983 that both of the following hold
985 - either w is a suffix of u or u is a suffix of w
986 - either w is a prefix of v or v is a prefix of w
988 then w is said to be a *repetition* for the factorization (u, v).
990 Just to unpack this, there are four possibilities here. Let w = "abc". Then we
993 - w is a suffix of u and w is a prefix of v. ex: ("lolabc", "abcde")
994 - w is a suffix of u and v is a prefix of w. ex: ("lolabc", "ab")
995 - u is a suffix of w and w is a prefix of v. ex: ("bc", "abchi")
996 - u is a suffix of w and v is a prefix of w. ex: ("bc", "a")
998 Note that the word vu is a repetition for any factorization (u,v) of x = uv,
999 so every factorization has at least one repetition.
1001 If x is a string and (u, v) is a factorization for x, then a *local period* for
1002 (u, v) is an integer r such that there is some word w such that |w| = r and w is
1003 a repetition for (u, v).
1005 We denote by local_period(u, v) the smallest local period of (u, v). We sometimes
1006 call this *the local period* of (u, v). Provided that x = uv is non-empty, this
1007 is well-defined (because each non-empty word has at least one factorization, as
1010 It can be proven that the following is an equivalent definition of a local period
1011 for a factorization (u, v): any positive integer r such that x[i] == x[i+r] for
1012 all i such that |u| - r <= i <= |u| - 1 and such that both x[i] and x[i+r] are
1013 defined. (i.e., i > 0 and i + r < |x|).
1015 Using the above reformulation, it is easy to prove that
1017 1 <= local_period(u, v) <= period(uv)
1019 A factorization (u, v) of x such that local_period(u,v) = period(x) is called a
1020 *critical factorization*.
1022 The algorithm hinges on the following theorem, which is stated without proof:
1024 **Critical Factorization Theorem** Any word x has at least one critical
1025 factorization (u, v) such that |u| < period(x).
1027 The purpose of maximal_suffix is to find such a critical factorization.
1029 If the period is short, compute another factorization x = u' v' to use
1030 for reverse search, chosen instead so that |v'| < period(x).
1033 impl TwoWaySearcher {
1034 fn new(needle: &[u8], end: usize) -> TwoWaySearcher {
1035 let (crit_pos_false, period_false) = TwoWaySearcher::maximal_suffix(needle, false);
1036 let (crit_pos_true, period_true) = TwoWaySearcher::maximal_suffix(needle, true);
1038 let (crit_pos, period) = if crit_pos_false > crit_pos_true {
1039 (crit_pos_false, period_false)
1041 (crit_pos_true, period_true)
1044 // A particularly readable explanation of what's going on here can be found
1045 // in Crochemore and Rytter's book "Text Algorithms", ch 13. Specifically
1046 // see the code for "Algorithm CP" on p. 323.
1048 // What's going on is we have some critical factorization (u, v) of the
1049 // needle, and we want to determine whether u is a suffix of
1050 // &v[..period]. If it is, we use "Algorithm CP1". Otherwise we use
1051 // "Algorithm CP2", which is optimized for when the period of the needle
1053 if needle[..crit_pos] == needle[period..period + crit_pos] {
1054 // short period case -- the period is exact
1055 // compute a separate critical factorization for the reversed needle
1056 // x = u' v' where |v'| < period(x).
1058 // This is sped up by the period being known already.
1059 // Note that a case like x = "acba" may be factored exactly forwards
1060 // (crit_pos = 1, period = 3) while being factored with approximate
1061 // period in reverse (crit_pos = 2, period = 2). We use the given
1062 // reverse factorization but keep the exact period.
1063 let crit_pos_back = needle.len()
1065 TwoWaySearcher::reverse_maximal_suffix(needle, period, false),
1066 TwoWaySearcher::reverse_maximal_suffix(needle, period, true),
1073 byteset: Self::byteset_create(&needle[..period]),
1078 memory_back: needle.len(),
1081 // long period case -- we have an approximation to the actual period,
1082 // and don't use memorization.
1084 // Approximate the period by lower bound max(|u|, |v|) + 1.
1085 // The critical factorization is efficient to use for both forward and
1090 crit_pos_back: crit_pos,
1091 period: cmp::max(crit_pos, needle.len() - crit_pos) + 1,
1092 byteset: Self::byteset_create(needle),
1096 memory: usize::MAX, // Dummy value to signify that the period is long
1097 memory_back: usize::MAX,
1103 fn byteset_create(bytes: &[u8]) -> u64 {
1104 bytes.iter().fold(0, |a, &b| (1 << (b & 0x3f)) | a)
1108 fn byteset_contains(&self, byte: u8) -> bool {
1109 (self.byteset >> ((byte & 0x3f) as usize)) & 1 != 0
1112 // One of the main ideas of Two-Way is that we factorize the needle into
1113 // two halves, (u, v), and begin trying to find v in the haystack by scanning
1114 // left to right. If v matches, we try to match u by scanning right to left.
1115 // How far we can jump when we encounter a mismatch is all based on the fact
1116 // that (u, v) is a critical factorization for the needle.
1118 fn next<S>(&mut self, haystack: &[u8], needle: &[u8], long_period: bool) -> S::Output
1122 // `next()` uses `self.position` as its cursor
1123 let old_pos = self.position;
1124 let needle_last = needle.len() - 1;
1126 // Check that we have room to search in
1127 // position + needle_last can not overflow if we assume slices
1128 // are bounded by isize's range.
1129 let tail_byte = match haystack.get(self.position + needle_last) {
1132 self.position = haystack.len();
1133 return S::rejecting(old_pos, self.position);
1137 if S::use_early_reject() && old_pos != self.position {
1138 return S::rejecting(old_pos, self.position);
1141 // Quickly skip by large portions unrelated to our substring
1142 if !self.byteset_contains(tail_byte) {
1143 self.position += needle.len();
1150 // See if the right part of the needle matches
1152 if long_period { self.crit_pos } else { cmp::max(self.crit_pos, self.memory) };
1153 for i in start..needle.len() {
1154 if needle[i] != haystack[self.position + i] {
1155 self.position += i - self.crit_pos + 1;
1163 // See if the left part of the needle matches
1164 let start = if long_period { 0 } else { self.memory };
1165 for i in (start..self.crit_pos).rev() {
1166 if needle[i] != haystack[self.position + i] {
1167 self.position += self.period;
1169 self.memory = needle.len() - self.period;
1175 // We have found a match!
1176 let match_pos = self.position;
1178 // Note: add self.period instead of needle.len() to have overlapping matches
1179 self.position += needle.len();
1181 self.memory = 0; // set to needle.len() - self.period for overlapping matches
1184 return S::matching(match_pos, match_pos + needle.len());
1188 // Follows the ideas in `next()`.
1190 // The definitions are symmetrical, with period(x) = period(reverse(x))
1191 // and local_period(u, v) = local_period(reverse(v), reverse(u)), so if (u, v)
1192 // is a critical factorization, so is (reverse(v), reverse(u)).
1194 // For the reverse case we have computed a critical factorization x = u' v'
1195 // (field `crit_pos_back`). We need |u| < period(x) for the forward case and
1196 // thus |v'| < period(x) for the reverse.
1198 // To search in reverse through the haystack, we search forward through
1199 // a reversed haystack with a reversed needle, matching first u' and then v'.
1201 fn next_back<S>(&mut self, haystack: &[u8], needle: &[u8], long_period: bool) -> S::Output
1205 // `next_back()` uses `self.end` as its cursor -- so that `next()` and `next_back()`
1207 let old_end = self.end;
1209 // Check that we have room to search in
1210 // end - needle.len() will wrap around when there is no more room,
1211 // but due to slice length limits it can never wrap all the way back
1212 // into the length of haystack.
1213 let front_byte = match haystack.get(self.end.wrapping_sub(needle.len())) {
1217 return S::rejecting(0, old_end);
1221 if S::use_early_reject() && old_end != self.end {
1222 return S::rejecting(self.end, old_end);
1225 // Quickly skip by large portions unrelated to our substring
1226 if !self.byteset_contains(front_byte) {
1227 self.end -= needle.len();
1229 self.memory_back = needle.len();
1234 // See if the left part of the needle matches
1235 let crit = if long_period {
1238 cmp::min(self.crit_pos_back, self.memory_back)
1240 for i in (0..crit).rev() {
1241 if needle[i] != haystack[self.end - needle.len() + i] {
1242 self.end -= self.crit_pos_back - i;
1244 self.memory_back = needle.len();
1250 // See if the right part of the needle matches
1251 let needle_end = if long_period { needle.len() } else { self.memory_back };
1252 for i in self.crit_pos_back..needle_end {
1253 if needle[i] != haystack[self.end - needle.len() + i] {
1254 self.end -= self.period;
1256 self.memory_back = self.period;
1262 // We have found a match!
1263 let match_pos = self.end - needle.len();
1264 // Note: sub self.period instead of needle.len() to have overlapping matches
1265 self.end -= needle.len();
1267 self.memory_back = needle.len();
1270 return S::matching(match_pos, match_pos + needle.len());
1274 // Compute the maximal suffix of `arr`.
1276 // The maximal suffix is a possible critical factorization (u, v) of `arr`.
1278 // Returns (`i`, `p`) where `i` is the starting index of v and `p` is the
1281 // `order_greater` determines if lexical order is `<` or `>`. Both
1282 // orders must be computed -- the ordering with the largest `i` gives
1283 // a critical factorization.
1285 // For long period cases, the resulting period is not exact (it is too short).
1287 fn maximal_suffix(arr: &[u8], order_greater: bool) -> (usize, usize) {
1288 let mut left = 0; // Corresponds to i in the paper
1289 let mut right = 1; // Corresponds to j in the paper
1290 let mut offset = 0; // Corresponds to k in the paper, but starting at 0
1291 // to match 0-based indexing.
1292 let mut period = 1; // Corresponds to p in the paper
1294 while let Some(&a) = arr.get(right + offset) {
1295 // `left` will be inbounds when `right` is.
1296 let b = arr[left + offset];
1297 if (a < b && !order_greater) || (a > b && order_greater) {
1298 // Suffix is smaller, period is entire prefix so far.
1299 right += offset + 1;
1301 period = right - left;
1303 // Advance through repetition of the current period.
1304 if offset + 1 == period {
1305 right += offset + 1;
1311 // Suffix is larger, start over from current location.
1321 // Compute the maximal suffix of the reverse of `arr`.
1323 // The maximal suffix is a possible critical factorization (u', v') of `arr`.
1325 // Returns `i` where `i` is the starting index of v', from the back;
1326 // returns immediately when a period of `known_period` is reached.
1328 // `order_greater` determines if lexical order is `<` or `>`. Both
1329 // orders must be computed -- the ordering with the largest `i` gives
1330 // a critical factorization.
1332 // For long period cases, the resulting period is not exact (it is too short).
1333 fn reverse_maximal_suffix(arr: &[u8], known_period: usize, order_greater: bool) -> usize {
1334 let mut left = 0; // Corresponds to i in the paper
1335 let mut right = 1; // Corresponds to j in the paper
1336 let mut offset = 0; // Corresponds to k in the paper, but starting at 0
1337 // to match 0-based indexing.
1338 let mut period = 1; // Corresponds to p in the paper
1341 while right + offset < n {
1342 let a = arr[n - (1 + right + offset)];
1343 let b = arr[n - (1 + left + offset)];
1344 if (a < b && !order_greater) || (a > b && order_greater) {
1345 // Suffix is smaller, period is entire prefix so far.
1346 right += offset + 1;
1348 period = right - left;
1350 // Advance through repetition of the current period.
1351 if offset + 1 == period {
1352 right += offset + 1;
1358 // Suffix is larger, start over from current location.
1364 if period == known_period {
1368 debug_assert!(period <= known_period);
1373 // TwoWayStrategy allows the algorithm to either skip non-matches as quickly
1374 // as possible, or to work in a mode where it emits Rejects relatively quickly.
1375 trait TwoWayStrategy {
1377 fn use_early_reject() -> bool;
1378 fn rejecting(a: usize, b: usize) -> Self::Output;
1379 fn matching(a: usize, b: usize) -> Self::Output;
1382 /// Skip to match intervals as quickly as possible
1385 impl TwoWayStrategy for MatchOnly {
1386 type Output = Option<(usize, usize)>;
1389 fn use_early_reject() -> bool {
1393 fn rejecting(_a: usize, _b: usize) -> Self::Output {
1397 fn matching(a: usize, b: usize) -> Self::Output {
1402 /// Emit Rejects regularly
1403 enum RejectAndMatch {}
1405 impl TwoWayStrategy for RejectAndMatch {
1406 type Output = SearchStep;
1409 fn use_early_reject() -> bool {
1413 fn rejecting(a: usize, b: usize) -> Self::Output {
1414 SearchStep::Reject(a, b)
1417 fn matching(a: usize, b: usize) -> Self::Output {
1418 SearchStep::Match(a, b)