1 //! Unicode string slices.
3 //! *[See also the `str` primitive type](str).*
5 //! The `&str` type is one of the two main string types, the other being `String`.
6 //! Unlike its `String` counterpart, its contents are borrowed.
10 //! A basic string declaration of `&str` type:
13 //! let hello_world = "Hello, World!";
16 //! Here we have declared a string literal, also known as a string slice.
17 //! String literals have a static lifetime, which means the string `hello_world`
18 //! is guaranteed to be valid for the duration of the entire program.
19 //! We can explicitly specify `hello_world`'s lifetime as well:
22 //! let hello_world: &'static str = "Hello, world!";
25 #![stable(feature = "rust1", since = "1.0.0")]
26 // Many of the usings in this module are only used in the test configuration.
27 // It's cleaner to just turn off the unused_imports warning than to fix them.
28 #![allow(unused_imports)]
30 use core::borrow::{Borrow, BorrowMut};
31 use core::iter::FusedIterator;
34 use core::str::pattern::{DoubleEndedSearcher, Pattern, ReverseSearcher, Searcher};
35 use core::unicode::conversions;
37 use crate::borrow::ToOwned;
38 use crate::boxed::Box;
39 use crate::slice::{Concat, Join, SliceIndex};
40 use crate::string::String;
43 #[stable(feature = "rust1", since = "1.0.0")]
44 pub use core::str::pattern;
45 #[stable(feature = "encode_utf16", since = "1.8.0")]
46 pub use core::str::EncodeUtf16;
47 #[stable(feature = "split_ascii_whitespace", since = "1.34.0")]
48 pub use core::str::SplitAsciiWhitespace;
49 #[stable(feature = "split_inclusive", since = "1.51.0")]
50 pub use core::str::SplitInclusive;
51 #[stable(feature = "rust1", since = "1.0.0")]
52 pub use core::str::SplitWhitespace;
53 #[stable(feature = "rust1", since = "1.0.0")]
54 pub use core::str::{from_utf8, from_utf8_mut, Bytes, CharIndices, Chars};
55 #[stable(feature = "rust1", since = "1.0.0")]
56 pub use core::str::{from_utf8_unchecked, from_utf8_unchecked_mut, ParseBoolError};
57 #[stable(feature = "str_escape", since = "1.34.0")]
58 pub use core::str::{EscapeDebug, EscapeDefault, EscapeUnicode};
59 #[stable(feature = "rust1", since = "1.0.0")]
60 pub use core::str::{FromStr, Utf8Error};
62 #[stable(feature = "rust1", since = "1.0.0")]
63 pub use core::str::{Lines, LinesAny};
64 #[stable(feature = "rust1", since = "1.0.0")]
65 pub use core::str::{MatchIndices, RMatchIndices};
66 #[stable(feature = "rust1", since = "1.0.0")]
67 pub use core::str::{Matches, RMatches};
68 #[stable(feature = "rust1", since = "1.0.0")]
69 pub use core::str::{RSplit, Split};
70 #[stable(feature = "rust1", since = "1.0.0")]
71 pub use core::str::{RSplitN, SplitN};
72 #[stable(feature = "rust1", since = "1.0.0")]
73 pub use core::str::{RSplitTerminator, SplitTerminator};
75 /// Note: `str` in `Concat<str>` is not meaningful here.
76 /// This type parameter of the trait only exists to enable another impl.
77 #[cfg(not(no_global_oom_handling))]
78 #[unstable(feature = "slice_concat_ext", issue = "27747")]
79 impl<S: Borrow<str>> Concat<str> for [S] {
82 fn concat(slice: &Self) -> String {
87 #[cfg(not(no_global_oom_handling))]
88 #[unstable(feature = "slice_concat_ext", issue = "27747")]
89 impl<S: Borrow<str>> Join<&str> for [S] {
92 fn join(slice: &Self, sep: &str) -> String {
93 unsafe { String::from_utf8_unchecked(join_generic_copy(slice, sep.as_bytes())) }
97 #[cfg(not(no_global_oom_handling))]
98 macro_rules! specialize_for_lengths {
99 ($separator:expr, $target:expr, $iter:expr; $($num:expr),*) => {{
100 let mut target = $target;
102 let sep_bytes = $separator;
103 match $separator.len() {
105 // loops with hardcoded sizes run much faster
106 // specialize the cases with small separator lengths
109 copy_slice_and_advance!(target, sep_bytes);
110 let content_bytes = s.borrow().as_ref();
111 copy_slice_and_advance!(target, content_bytes);
116 // arbitrary non-zero size fallback
118 copy_slice_and_advance!(target, sep_bytes);
119 let content_bytes = s.borrow().as_ref();
120 copy_slice_and_advance!(target, content_bytes);
128 #[cfg(not(no_global_oom_handling))]
129 macro_rules! copy_slice_and_advance {
130 ($target:expr, $bytes:expr) => {
131 let len = $bytes.len();
132 let (head, tail) = { $target }.split_at_mut(len);
133 head.copy_from_slice($bytes);
138 // Optimized join implementation that works for both Vec<T> (T: Copy) and String's inner vec
139 // Currently (2018-05-13) there is a bug with type inference and specialization (see issue #36262)
140 // For this reason SliceConcat<T> is not specialized for T: Copy and SliceConcat<str> is the
141 // only user of this function. It is left in place for the time when that is fixed.
143 // the bounds for String-join are S: Borrow<str> and for Vec-join Borrow<[T]>
144 // [T] and str both impl AsRef<[T]> for some T
145 // => s.borrow().as_ref() and we always have slices
146 #[cfg(not(no_global_oom_handling))]
147 fn join_generic_copy<B, T, S>(slice: &[S], sep: &[T]) -> Vec<T>
150 B: AsRef<[T]> + ?Sized,
153 let sep_len = sep.len();
154 let mut iter = slice.iter();
156 // the first slice is the only one without a separator preceding it
157 let first = match iter.next() {
158 Some(first) => first,
159 None => return vec![],
162 // compute the exact total length of the joined Vec
163 // if the `len` calculation overflows, we'll panic
164 // we would have run out of memory anyway and the rest of the function requires
165 // the entire Vec pre-allocated for safety
166 let reserved_len = sep_len
167 .checked_mul(iter.len())
169 slice.iter().map(|s| s.borrow().as_ref().len()).try_fold(n, usize::checked_add)
171 .expect("attempt to join into collection with len > usize::MAX");
173 // prepare an uninitialized buffer
174 let mut result = Vec::with_capacity(reserved_len);
175 debug_assert!(result.capacity() >= reserved_len);
177 result.extend_from_slice(first.borrow().as_ref());
180 let pos = result.len();
181 let target = result.spare_capacity_mut().get_unchecked_mut(..reserved_len - pos);
183 // Convert the separator and slices to slices of MaybeUninit
184 // to simplify implementation in specialize_for_lengths
185 let sep_uninit = core::slice::from_raw_parts(sep.as_ptr().cast(), sep.len());
186 let iter_uninit = iter.map(|it| {
187 let it = it.borrow().as_ref();
188 core::slice::from_raw_parts(it.as_ptr().cast(), it.len())
191 // copy separator and slices over without bounds checks
192 // generate loops with hardcoded offsets for small separators
193 // massive improvements possible (~ x2)
194 let remain = specialize_for_lengths!(sep_uninit, target, iter_uninit; 0, 1, 2, 3, 4);
196 // A weird borrow implementation may return different
197 // slices for the length calculation and the actual copy.
198 // Make sure we don't expose uninitialized bytes to the caller.
199 let result_len = reserved_len - remain.len();
200 result.set_len(result_len);
205 #[stable(feature = "rust1", since = "1.0.0")]
206 impl Borrow<str> for String {
208 fn borrow(&self) -> &str {
213 #[stable(feature = "string_borrow_mut", since = "1.36.0")]
214 impl BorrowMut<str> for String {
216 fn borrow_mut(&mut self) -> &mut str {
221 #[cfg(not(no_global_oom_handling))]
222 #[stable(feature = "rust1", since = "1.0.0")]
223 impl ToOwned for str {
226 fn to_owned(&self) -> String {
227 unsafe { String::from_utf8_unchecked(self.as_bytes().to_owned()) }
230 fn clone_into(&self, target: &mut String) {
231 let mut b = mem::take(target).into_bytes();
232 self.as_bytes().clone_into(&mut b);
233 *target = unsafe { String::from_utf8_unchecked(b) }
237 /// Methods for string slices.
240 /// Converts a `Box<str>` into a `Box<[u8]>` without copying or allocating.
247 /// let s = "this is a string";
248 /// let boxed_str = s.to_owned().into_boxed_str();
249 /// let boxed_bytes = boxed_str.into_boxed_bytes();
250 /// assert_eq!(*boxed_bytes, *s.as_bytes());
252 #[rustc_allow_incoherent_impl]
253 #[stable(feature = "str_box_extras", since = "1.20.0")]
254 #[must_use = "`self` will be dropped if the result is not used"]
256 pub fn into_boxed_bytes(self: Box<str>) -> Box<[u8]> {
260 /// Replaces all matches of a pattern with another string.
262 /// `replace` creates a new [`String`], and copies the data from this string slice into it.
263 /// While doing so, it attempts to find matches of a pattern. If it finds any, it
264 /// replaces them with the replacement string slice.
271 /// let s = "this is old";
273 /// assert_eq!("this is new", s.replace("old", "new"));
274 /// assert_eq!("than an old", s.replace("is", "an"));
277 /// When the pattern doesn't match:
280 /// let s = "this is old";
281 /// assert_eq!(s, s.replace("cookie monster", "little lamb"));
283 #[cfg(not(no_global_oom_handling))]
284 #[rustc_allow_incoherent_impl]
285 #[must_use = "this returns the replaced string as a new allocation, \
286 without modifying the original"]
287 #[stable(feature = "rust1", since = "1.0.0")]
289 pub fn replace<'a, P: Pattern<'a>>(&'a self, from: P, to: &str) -> String {
290 let mut result = String::new();
291 let mut last_end = 0;
292 for (start, part) in self.match_indices(from) {
293 result.push_str(unsafe { self.get_unchecked(last_end..start) });
295 last_end = start + part.len();
297 result.push_str(unsafe { self.get_unchecked(last_end..self.len()) });
301 /// Replaces first N matches of a pattern with another string.
303 /// `replacen` creates a new [`String`], and copies the data from this string slice into it.
304 /// While doing so, it attempts to find matches of a pattern. If it finds any, it
305 /// replaces them with the replacement string slice at most `count` times.
312 /// let s = "foo foo 123 foo";
313 /// assert_eq!("new new 123 foo", s.replacen("foo", "new", 2));
314 /// assert_eq!("faa fao 123 foo", s.replacen('o', "a", 3));
315 /// assert_eq!("foo foo new23 foo", s.replacen(char::is_numeric, "new", 1));
318 /// When the pattern doesn't match:
321 /// let s = "this is old";
322 /// assert_eq!(s, s.replacen("cookie monster", "little lamb", 10));
324 #[cfg(not(no_global_oom_handling))]
325 #[rustc_allow_incoherent_impl]
326 #[must_use = "this returns the replaced string as a new allocation, \
327 without modifying the original"]
328 #[stable(feature = "str_replacen", since = "1.16.0")]
329 pub fn replacen<'a, P: Pattern<'a>>(&'a self, pat: P, to: &str, count: usize) -> String {
330 // Hope to reduce the times of re-allocation
331 let mut result = String::with_capacity(32);
332 let mut last_end = 0;
333 for (start, part) in self.match_indices(pat).take(count) {
334 result.push_str(unsafe { self.get_unchecked(last_end..start) });
336 last_end = start + part.len();
338 result.push_str(unsafe { self.get_unchecked(last_end..self.len()) });
342 /// Returns the lowercase equivalent of this string slice, as a new [`String`].
344 /// 'Lowercase' is defined according to the terms of the Unicode Derived Core Property
347 /// Since some characters can expand into multiple characters when changing
348 /// the case, this function returns a [`String`] instead of modifying the
349 /// parameter in-place.
358 /// assert_eq!("hello", s.to_lowercase());
361 /// A tricky example, with sigma:
366 /// assert_eq!("σ", sigma.to_lowercase());
368 /// // but at the end of a word, it's ς, not σ:
369 /// let odysseus = "ὈΔΥΣΣΕΎΣ";
371 /// assert_eq!("ὀδυσσεύς", odysseus.to_lowercase());
374 /// Languages without case are not changed:
377 /// let new_year = "农历新年";
379 /// assert_eq!(new_year, new_year.to_lowercase());
381 #[cfg(not(no_global_oom_handling))]
382 #[rustc_allow_incoherent_impl]
383 #[must_use = "this returns the lowercase string as a new String, \
384 without modifying the original"]
385 #[stable(feature = "unicode_case_mapping", since = "1.2.0")]
386 pub fn to_lowercase(&self) -> String {
387 let out = convert_while_ascii(self.as_bytes(), u8::to_ascii_lowercase);
389 // Safety: we know this is a valid char boundary since
390 // out.len() is only progressed if ascii bytes are found
391 let rest = unsafe { self.get_unchecked(out.len()..) };
393 // Safety: We have written only valid ASCII to our vec
394 let mut s = unsafe { String::from_utf8_unchecked(out) };
396 for (i, c) in rest[..].char_indices() {
398 // Σ maps to σ, except at the end of a word where it maps to ς.
399 // This is the only conditional (contextual) but language-independent mapping
400 // in `SpecialCasing.txt`,
401 // so hard-code it rather than have a generic "condition" mechanism.
402 // See https://github.com/rust-lang/rust/issues/26035
403 map_uppercase_sigma(rest, i, &mut s)
405 match conversions::to_lower(c) {
406 [a, '\0', _] => s.push(a),
421 fn map_uppercase_sigma(from: &str, i: usize, to: &mut String) {
422 // See https://www.unicode.org/versions/Unicode7.0.0/ch03.pdf#G33992
423 // for the definition of `Final_Sigma`.
424 debug_assert!('Σ'.len_utf8() == 2);
425 let is_word_final = case_ignoreable_then_cased(from[..i].chars().rev())
426 && !case_ignoreable_then_cased(from[i + 2..].chars());
427 to.push_str(if is_word_final { "ς" } else { "σ" });
430 fn case_ignoreable_then_cased<I: Iterator<Item = char>>(iter: I) -> bool {
431 use core::unicode::{Case_Ignorable, Cased};
432 match iter.skip_while(|&c| Case_Ignorable(c)).next() {
439 /// Returns the uppercase equivalent of this string slice, as a new [`String`].
441 /// 'Uppercase' is defined according to the terms of the Unicode Derived Core Property
444 /// Since some characters can expand into multiple characters when changing
445 /// the case, this function returns a [`String`] instead of modifying the
446 /// parameter in-place.
455 /// assert_eq!("HELLO", s.to_uppercase());
458 /// Scripts without case are not changed:
461 /// let new_year = "农历新年";
463 /// assert_eq!(new_year, new_year.to_uppercase());
466 /// One character can become multiple:
468 /// let s = "tschüß";
470 /// assert_eq!("TSCHÜSS", s.to_uppercase());
472 #[cfg(not(no_global_oom_handling))]
473 #[rustc_allow_incoherent_impl]
474 #[must_use = "this returns the uppercase string as a new String, \
475 without modifying the original"]
476 #[stable(feature = "unicode_case_mapping", since = "1.2.0")]
477 pub fn to_uppercase(&self) -> String {
478 let out = convert_while_ascii(self.as_bytes(), u8::to_ascii_uppercase);
480 // Safety: we know this is a valid char boundary since
481 // out.len() is only progressed if ascii bytes are found
482 let rest = unsafe { self.get_unchecked(out.len()..) };
484 // Safety: We have written only valid ASCII to our vec
485 let mut s = unsafe { String::from_utf8_unchecked(out) };
487 for c in rest.chars() {
488 match conversions::to_upper(c) {
489 [a, '\0', _] => s.push(a),
504 /// Converts a [`Box<str>`] into a [`String`] without copying or allocating.
511 /// let string = String::from("birthday gift");
512 /// let boxed_str = string.clone().into_boxed_str();
514 /// assert_eq!(boxed_str.into_string(), string);
516 #[stable(feature = "box_str", since = "1.4.0")]
517 #[rustc_allow_incoherent_impl]
518 #[must_use = "`self` will be dropped if the result is not used"]
520 pub fn into_string(self: Box<str>) -> String {
521 let slice = Box::<[u8]>::from(self);
522 unsafe { String::from_utf8_unchecked(slice.into_vec()) }
525 /// Creates a new [`String`] by repeating a string `n` times.
529 /// This function will panic if the capacity would overflow.
536 /// assert_eq!("abc".repeat(4), String::from("abcabcabcabc"));
539 /// A panic upon overflow:
542 /// // this will panic at runtime
543 /// let huge = "0123456789abcdef".repeat(usize::MAX);
545 #[cfg(not(no_global_oom_handling))]
546 #[rustc_allow_incoherent_impl]
548 #[stable(feature = "repeat_str", since = "1.16.0")]
549 pub fn repeat(&self, n: usize) -> String {
550 unsafe { String::from_utf8_unchecked(self.as_bytes().repeat(n)) }
553 /// Returns a copy of this string where each character is mapped to its
554 /// ASCII upper case equivalent.
556 /// ASCII letters 'a' to 'z' are mapped to 'A' to 'Z',
557 /// but non-ASCII letters are unchanged.
559 /// To uppercase the value in-place, use [`make_ascii_uppercase`].
561 /// To uppercase ASCII characters in addition to non-ASCII characters, use
562 /// [`to_uppercase`].
567 /// let s = "Grüße, Jürgen ❤";
569 /// assert_eq!("GRüßE, JüRGEN ❤", s.to_ascii_uppercase());
572 /// [`make_ascii_uppercase`]: str::make_ascii_uppercase
573 /// [`to_uppercase`]: #method.to_uppercase
574 #[cfg(not(no_global_oom_handling))]
575 #[rustc_allow_incoherent_impl]
576 #[must_use = "to uppercase the value in-place, use `make_ascii_uppercase()`"]
577 #[stable(feature = "ascii_methods_on_intrinsics", since = "1.23.0")]
579 pub fn to_ascii_uppercase(&self) -> String {
580 let mut bytes = self.as_bytes().to_vec();
581 bytes.make_ascii_uppercase();
582 // make_ascii_uppercase() preserves the UTF-8 invariant.
583 unsafe { String::from_utf8_unchecked(bytes) }
586 /// Returns a copy of this string where each character is mapped to its
587 /// ASCII lower case equivalent.
589 /// ASCII letters 'A' to 'Z' are mapped to 'a' to 'z',
590 /// but non-ASCII letters are unchanged.
592 /// To lowercase the value in-place, use [`make_ascii_lowercase`].
594 /// To lowercase ASCII characters in addition to non-ASCII characters, use
595 /// [`to_lowercase`].
600 /// let s = "Grüße, Jürgen ❤";
602 /// assert_eq!("grüße, jürgen ❤", s.to_ascii_lowercase());
605 /// [`make_ascii_lowercase`]: str::make_ascii_lowercase
606 /// [`to_lowercase`]: #method.to_lowercase
607 #[cfg(not(no_global_oom_handling))]
608 #[rustc_allow_incoherent_impl]
609 #[must_use = "to lowercase the value in-place, use `make_ascii_lowercase()`"]
610 #[stable(feature = "ascii_methods_on_intrinsics", since = "1.23.0")]
612 pub fn to_ascii_lowercase(&self) -> String {
613 let mut bytes = self.as_bytes().to_vec();
614 bytes.make_ascii_lowercase();
615 // make_ascii_lowercase() preserves the UTF-8 invariant.
616 unsafe { String::from_utf8_unchecked(bytes) }
620 /// Converts a boxed slice of bytes to a boxed string slice without checking
621 /// that the string contains valid UTF-8.
628 /// let smile_utf8 = Box::new([226, 152, 186]);
629 /// let smile = unsafe { std::str::from_boxed_utf8_unchecked(smile_utf8) };
631 /// assert_eq!("☺", &*smile);
633 #[stable(feature = "str_box_extras", since = "1.20.0")]
636 pub unsafe fn from_boxed_utf8_unchecked(v: Box<[u8]>) -> Box<str> {
637 unsafe { Box::from_raw(Box::into_raw(v) as *mut str) }
640 /// Converts the bytes while the bytes are still ascii.
641 /// For better average performance, this is happens in chunks of `2*size_of::<usize>()`.
642 /// Returns a vec with the converted bytes.
645 #[cfg(not(no_global_oom_handling))]
646 fn convert_while_ascii(b: &[u8], convert: fn(&u8) -> u8) -> Vec<u8> {
647 let mut out = Vec::with_capacity(b.len());
649 const USIZE_SIZE: usize = mem::size_of::<usize>();
650 const MAGIC_UNROLL: usize = 2;
651 const N: usize = USIZE_SIZE * MAGIC_UNROLL;
652 const NONASCII_MASK: usize = usize::from_ne_bytes([0x80; USIZE_SIZE]);
656 while i + N <= b.len() {
657 // Safety: we have checks the sizes `b` and `out` to know that our
658 let in_chunk = b.get_unchecked(i..i + N);
659 let out_chunk = out.spare_capacity_mut().get_unchecked_mut(i..i + N);
662 for j in 0..MAGIC_UNROLL {
663 // read the bytes 1 usize at a time (unaligned since we haven't checked the alignment)
664 // safety: in_chunk is valid bytes in the range
665 bits |= in_chunk.as_ptr().cast::<usize>().add(j).read_unaligned();
667 // if our chunks aren't ascii, then return only the prior bytes as init
668 if bits & NONASCII_MASK != 0 {
672 // perform the case conversions on N bytes (gets heavily autovec'd)
674 // safety: in_chunk and out_chunk is valid bytes in the range
675 let out = out_chunk.get_unchecked_mut(j);
676 out.write(convert(in_chunk.get_unchecked(j)));
679 // mark these bytes as initialised