1//! Utilities for the `str` primitive type.
2//!
3//! *[See also the `str` primitive type](str).*
4
5#![stable(feature = "rust1", since = "1.0.0")]
6// Many of the usings in this module are only used in the test configuration.
7// It's cleaner to just turn off the unused_imports warning than to fix them.
8#![allow(unused_imports)]
9
10use core::borrow::{Borrow, BorrowMut};
11use core::iter::FusedIterator;
12use core::mem::MaybeUninit;
13#[stable(feature = "encode_utf16", since = "1.8.0")]
14pub use core::str::EncodeUtf16;
15#[stable(feature = "split_ascii_whitespace", since = "1.34.0")]
16pub use core::str::SplitAsciiWhitespace;
17#[stable(feature = "split_inclusive", since = "1.51.0")]
18pub use core::str::SplitInclusive;
19#[stable(feature = "rust1", since = "1.0.0")]
20pub use core::str::SplitWhitespace;
21#[stable(feature = "rust1", since = "1.0.0")]
22pub use core::str::pattern;
23use core::str::pattern::{DoubleEndedSearcher, Pattern, ReverseSearcher, Searcher, Utf8Pattern};
24#[stable(feature = "rust1", since = "1.0.0")]
25pub use core::str::{Bytes, CharIndices, Chars, from_utf8, from_utf8_mut};
26#[stable(feature = "str_escape", since = "1.34.0")]
27pub use core::str::{EscapeDebug, EscapeDefault, EscapeUnicode};
28#[stable(feature = "rust1", since = "1.0.0")]
29pub use core::str::{FromStr, Utf8Error};
30#[allow(deprecated)]
31#[stable(feature = "rust1", since = "1.0.0")]
32pub use core::str::{Lines, LinesAny};
33#[stable(feature = "rust1", since = "1.0.0")]
34pub use core::str::{MatchIndices, RMatchIndices};
35#[stable(feature = "rust1", since = "1.0.0")]
36pub use core::str::{Matches, RMatches};
37#[stable(feature = "rust1", since = "1.0.0")]
38pub use core::str::{ParseBoolError, from_utf8_unchecked, from_utf8_unchecked_mut};
39#[stable(feature = "rust1", since = "1.0.0")]
40pub use core::str::{RSplit, Split};
41#[stable(feature = "rust1", since = "1.0.0")]
42pub use core::str::{RSplitN, SplitN};
43#[stable(feature = "rust1", since = "1.0.0")]
44pub use core::str::{RSplitTerminator, SplitTerminator};
45#[stable(feature = "utf8_chunks", since = "1.79.0")]
46pub use core::str::{Utf8Chunk, Utf8Chunks};
47#[unstable(feature = "str_from_raw_parts", issue = "119206")]
48pub use core::str::{from_raw_parts, from_raw_parts_mut};
49use core::unicode::conversions;
50use core::{mem, ptr};
51
52use crate::borrow::ToOwned;
53use crate::boxed::Box;
54use crate::slice::{Concat, Join, SliceIndex};
55use crate::string::String;
56use crate::vec::Vec;
57
58/// Note: `str` in `Concat<str>` is not meaningful here.
59/// This type parameter of the trait only exists to enable another impl.
60#[cfg(not(no_global_oom_handling))]
61#[unstable(feature = "slice_concat_ext", issue = "27747")]
62impl<S: Borrow<str>> Concat<str> for [S] {
63 type Output = String;
64
65 fn concat(slice: &Self) -> String {
66 Join::join(slice, sep:"")
67 }
68}
69
70#[cfg(not(no_global_oom_handling))]
71#[unstable(feature = "slice_concat_ext", issue = "27747")]
72impl<S: Borrow<str>> Join<&str> for [S] {
73 type Output = String;
74
75 fn join(slice: &Self, sep: &str) -> String {
76 unsafe { String::from_utf8_unchecked(bytes:join_generic_copy(slice, sep.as_bytes())) }
77 }
78}
79
80#[cfg(not(no_global_oom_handling))]
81macro_rules! specialize_for_lengths {
82 ($separator:expr, $target:expr, $iter:expr; $($num:expr),*) => {{
83 let mut target = $target;
84 let iter = $iter;
85 let sep_bytes = $separator;
86 match $separator.len() {
87 $(
88 // loops with hardcoded sizes run much faster
89 // specialize the cases with small separator lengths
90 $num => {
91 for s in iter {
92 copy_slice_and_advance!(target, sep_bytes);
93 let content_bytes = s.borrow().as_ref();
94 copy_slice_and_advance!(target, content_bytes);
95 }
96 },
97 )*
98 _ => {
99 // arbitrary non-zero size fallback
100 for s in iter {
101 copy_slice_and_advance!(target, sep_bytes);
102 let content_bytes = s.borrow().as_ref();
103 copy_slice_and_advance!(target, content_bytes);
104 }
105 }
106 }
107 target
108 }}
109}
110
111#[cfg(not(no_global_oom_handling))]
112macro_rules! copy_slice_and_advance {
113 ($target:expr, $bytes:expr) => {
114 let len = $bytes.len();
115 let (head, tail) = { $target }.split_at_mut(len);
116 head.copy_from_slice($bytes);
117 $target = tail;
118 };
119}
120
121// Optimized join implementation that works for both Vec<T> (T: Copy) and String's inner vec
122// Currently (2018-05-13) there is a bug with type inference and specialization (see issue #36262)
123// For this reason SliceConcat<T> is not specialized for T: Copy and SliceConcat<str> is the
124// only user of this function. It is left in place for the time when that is fixed.
125//
126// the bounds for String-join are S: Borrow<str> and for Vec-join Borrow<[T]>
127// [T] and str both impl AsRef<[T]> for some T
128// => s.borrow().as_ref() and we always have slices
129#[cfg(not(no_global_oom_handling))]
130fn join_generic_copy<B, T, S>(slice: &[S], sep: &[T]) -> Vec<T>
131where
132 T: Copy,
133 B: AsRef<[T]> + ?Sized,
134 S: Borrow<B>,
135{
136 let sep_len = sep.len();
137 let mut iter = slice.iter();
138
139 // the first slice is the only one without a separator preceding it
140 let first = match iter.next() {
141 Some(first) => first,
142 None => return vec![],
143 };
144
145 // compute the exact total length of the joined Vec
146 // if the `len` calculation overflows, we'll panic
147 // we would have run out of memory anyway and the rest of the function requires
148 // the entire Vec pre-allocated for safety
149 let reserved_len = sep_len
150 .checked_mul(iter.len())
151 .and_then(|n| {
152 slice.iter().map(|s| s.borrow().as_ref().len()).try_fold(n, usize::checked_add)
153 })
154 .expect("attempt to join into collection with len > usize::MAX");
155
156 // prepare an uninitialized buffer
157 let mut result = Vec::with_capacity(reserved_len);
158 debug_assert!(result.capacity() >= reserved_len);
159
160 result.extend_from_slice(first.borrow().as_ref());
161
162 unsafe {
163 let pos = result.len();
164 let target = result.spare_capacity_mut().get_unchecked_mut(..reserved_len - pos);
165
166 // Convert the separator and slices to slices of MaybeUninit
167 // to simplify implementation in specialize_for_lengths
168 let sep_uninit = core::slice::from_raw_parts(sep.as_ptr().cast(), sep.len());
169 let iter_uninit = iter.map(|it| {
170 let it = it.borrow().as_ref();
171 core::slice::from_raw_parts(it.as_ptr().cast(), it.len())
172 });
173
174 // copy separator and slices over without bounds checks
175 // generate loops with hardcoded offsets for small separators
176 // massive improvements possible (~ x2)
177 let remain = specialize_for_lengths!(sep_uninit, target, iter_uninit; 0, 1, 2, 3, 4);
178
179 // A weird borrow implementation may return different
180 // slices for the length calculation and the actual copy.
181 // Make sure we don't expose uninitialized bytes to the caller.
182 let result_len = reserved_len - remain.len();
183 result.set_len(result_len);
184 }
185 result
186}
187
188#[stable(feature = "rust1", since = "1.0.0")]
189impl Borrow<str> for String {
190 #[inline]
191 fn borrow(&self) -> &str {
192 &self[..]
193 }
194}
195
196#[stable(feature = "string_borrow_mut", since = "1.36.0")]
197impl BorrowMut<str> for String {
198 #[inline]
199 fn borrow_mut(&mut self) -> &mut str {
200 &mut self[..]
201 }
202}
203
204#[cfg(not(no_global_oom_handling))]
205#[stable(feature = "rust1", since = "1.0.0")]
206impl ToOwned for str {
207 type Owned = String;
208
209 #[inline]
210 fn to_owned(&self) -> String {
211 unsafe { String::from_utf8_unchecked(self.as_bytes().to_owned()) }
212 }
213
214 #[inline]
215 fn clone_into(&self, target: &mut String) {
216 target.clear();
217 target.push_str(self);
218 }
219}
220
221/// Methods for string slices.
222impl str {
223 /// Converts a `Box<str>` into a `Box<[u8]>` without copying or allocating.
224 ///
225 /// # Examples
226 ///
227 /// ```
228 /// let s = "this is a string";
229 /// let boxed_str = s.to_owned().into_boxed_str();
230 /// let boxed_bytes = boxed_str.into_boxed_bytes();
231 /// assert_eq!(*boxed_bytes, *s.as_bytes());
232 /// ```
233 #[rustc_allow_incoherent_impl]
234 #[stable(feature = "str_box_extras", since = "1.20.0")]
235 #[must_use = "`self` will be dropped if the result is not used"]
236 #[inline]
237 pub fn into_boxed_bytes(self: Box<Self>) -> Box<[u8]> {
238 self.into()
239 }
240
241 /// Replaces all matches of a pattern with another string.
242 ///
243 /// `replace` creates a new [`String`], and copies the data from this string slice into it.
244 /// While doing so, it attempts to find matches of a pattern. If it finds any, it
245 /// replaces them with the replacement string slice.
246 ///
247 /// # Examples
248 ///
249 /// ```
250 /// let s = "this is old";
251 ///
252 /// assert_eq!("this is new", s.replace("old", "new"));
253 /// assert_eq!("than an old", s.replace("is", "an"));
254 /// ```
255 ///
256 /// When the pattern doesn't match, it returns this string slice as [`String`]:
257 ///
258 /// ```
259 /// let s = "this is old";
260 /// assert_eq!(s, s.replace("cookie monster", "little lamb"));
261 /// ```
262 #[cfg(not(no_global_oom_handling))]
263 #[rustc_allow_incoherent_impl]
264 #[must_use = "this returns the replaced string as a new allocation, \
265 without modifying the original"]
266 #[stable(feature = "rust1", since = "1.0.0")]
267 #[inline]
268 pub fn replace<P: Pattern>(&self, from: P, to: &str) -> String {
269 // Fast path for replacing a single ASCII character with another.
270 if let Some(from_byte) = match from.as_utf8_pattern() {
271 Some(Utf8Pattern::StringPattern([from_byte])) => Some(*from_byte),
272 Some(Utf8Pattern::CharPattern(c)) => c.as_ascii().map(|ascii_char| ascii_char.to_u8()),
273 _ => None,
274 } {
275 if let [to_byte] = to.as_bytes() {
276 return unsafe { replace_ascii(self.as_bytes(), from_byte, *to_byte) };
277 }
278 }
279 // Set result capacity to self.len() when from.len() <= to.len()
280 let default_capacity = match from.as_utf8_pattern() {
281 Some(Utf8Pattern::StringPattern(s)) if s.len() <= to.len() => self.len(),
282 Some(Utf8Pattern::CharPattern(c)) if c.len_utf8() <= to.len() => self.len(),
283 _ => 0,
284 };
285 let mut result = String::with_capacity(default_capacity);
286 let mut last_end = 0;
287 for (start, part) in self.match_indices(from) {
288 result.push_str(unsafe { self.get_unchecked(last_end..start) });
289 result.push_str(to);
290 last_end = start + part.len();
291 }
292 result.push_str(unsafe { self.get_unchecked(last_end..self.len()) });
293 result
294 }
295
296 /// Replaces first N matches of a pattern with another string.
297 ///
298 /// `replacen` creates a new [`String`], and copies the data from this string slice into it.
299 /// While doing so, it attempts to find matches of a pattern. If it finds any, it
300 /// replaces them with the replacement string slice at most `count` times.
301 ///
302 /// # Examples
303 ///
304 /// ```
305 /// let s = "foo foo 123 foo";
306 /// assert_eq!("new new 123 foo", s.replacen("foo", "new", 2));
307 /// assert_eq!("faa fao 123 foo", s.replacen('o', "a", 3));
308 /// assert_eq!("foo foo new23 foo", s.replacen(char::is_numeric, "new", 1));
309 /// ```
310 ///
311 /// When the pattern doesn't match, it returns this string slice as [`String`]:
312 ///
313 /// ```
314 /// let s = "this is old";
315 /// assert_eq!(s, s.replacen("cookie monster", "little lamb", 10));
316 /// ```
317 #[cfg(not(no_global_oom_handling))]
318 #[rustc_allow_incoherent_impl]
319 #[doc(alias = "replace_first")]
320 #[must_use = "this returns the replaced string as a new allocation, \
321 without modifying the original"]
322 #[stable(feature = "str_replacen", since = "1.16.0")]
323 pub fn replacen<P: Pattern>(&self, pat: P, to: &str, count: usize) -> String {
324 // Hope to reduce the times of re-allocation
325 let mut result = String::with_capacity(32);
326 let mut last_end = 0;
327 for (start, part) in self.match_indices(pat).take(count) {
328 result.push_str(unsafe { self.get_unchecked(last_end..start) });
329 result.push_str(to);
330 last_end = start + part.len();
331 }
332 result.push_str(unsafe { self.get_unchecked(last_end..self.len()) });
333 result
334 }
335
336 /// Returns the lowercase equivalent of this string slice, as a new [`String`].
337 ///
338 /// 'Lowercase' is defined according to the terms of the Unicode Derived Core Property
339 /// `Lowercase`.
340 ///
341 /// Since some characters can expand into multiple characters when changing
342 /// the case, this function returns a [`String`] instead of modifying the
343 /// parameter in-place.
344 ///
345 /// # Examples
346 ///
347 /// Basic usage:
348 ///
349 /// ```
350 /// let s = "HELLO";
351 ///
352 /// assert_eq!("hello", s.to_lowercase());
353 /// ```
354 ///
355 /// A tricky example, with sigma:
356 ///
357 /// ```
358 /// let sigma = "Σ";
359 ///
360 /// assert_eq!("σ", sigma.to_lowercase());
361 ///
362 /// // but at the end of a word, it's ς, not σ:
363 /// let odysseus = "ὈΔΥΣΣΕΎΣ";
364 ///
365 /// assert_eq!("ὀδυσσεύς", odysseus.to_lowercase());
366 /// ```
367 ///
368 /// Languages without case are not changed:
369 ///
370 /// ```
371 /// let new_year = "农历新年";
372 ///
373 /// assert_eq!(new_year, new_year.to_lowercase());
374 /// ```
375 #[cfg(not(no_global_oom_handling))]
376 #[rustc_allow_incoherent_impl]
377 #[must_use = "this returns the lowercase string as a new String, \
378 without modifying the original"]
379 #[stable(feature = "unicode_case_mapping", since = "1.2.0")]
380 pub fn to_lowercase(&self) -> String {
381 let (mut s, rest) = convert_while_ascii(self, u8::to_ascii_lowercase);
382
383 let prefix_len = s.len();
384
385 for (i, c) in rest.char_indices() {
386 if c == 'Σ' {
387 // Σ maps to σ, except at the end of a word where it maps to ς.
388 // This is the only conditional (contextual) but language-independent mapping
389 // in `SpecialCasing.txt`,
390 // so hard-code it rather than have a generic "condition" mechanism.
391 // See https://github.com/rust-lang/rust/issues/26035
392 let sigma_lowercase = map_uppercase_sigma(self, prefix_len + i);
393 s.push(sigma_lowercase);
394 } else {
395 match conversions::to_lower(c) {
396 [a, '\0', _] => s.push(a),
397 [a, b, '\0'] => {
398 s.push(a);
399 s.push(b);
400 }
401 [a, b, c] => {
402 s.push(a);
403 s.push(b);
404 s.push(c);
405 }
406 }
407 }
408 }
409 return s;
410
411 fn map_uppercase_sigma(from: &str, i: usize) -> char {
412 // See https://www.unicode.org/versions/Unicode7.0.0/ch03.pdf#G33992
413 // for the definition of `Final_Sigma`.
414 debug_assert!('Σ'.len_utf8() == 2);
415 let is_word_final = case_ignorable_then_cased(from[..i].chars().rev())
416 && !case_ignorable_then_cased(from[i + 2..].chars());
417 if is_word_final { 'ς' } else { 'σ' }
418 }
419
420 fn case_ignorable_then_cased<I: Iterator<Item = char>>(iter: I) -> bool {
421 use core::unicode::{Case_Ignorable, Cased};
422 match iter.skip_while(|&c| Case_Ignorable(c)).next() {
423 Some(c) => Cased(c),
424 None => false,
425 }
426 }
427 }
428
429 /// Returns the uppercase equivalent of this string slice, as a new [`String`].
430 ///
431 /// 'Uppercase' is defined according to the terms of the Unicode Derived Core Property
432 /// `Uppercase`.
433 ///
434 /// Since some characters can expand into multiple characters when changing
435 /// the case, this function returns a [`String`] instead of modifying the
436 /// parameter in-place.
437 ///
438 /// # Examples
439 ///
440 /// Basic usage:
441 ///
442 /// ```
443 /// let s = "hello";
444 ///
445 /// assert_eq!("HELLO", s.to_uppercase());
446 /// ```
447 ///
448 /// Scripts without case are not changed:
449 ///
450 /// ```
451 /// let new_year = "农历新年";
452 ///
453 /// assert_eq!(new_year, new_year.to_uppercase());
454 /// ```
455 ///
456 /// One character can become multiple:
457 /// ```
458 /// let s = "tschüß";
459 ///
460 /// assert_eq!("TSCHÜSS", s.to_uppercase());
461 /// ```
462 #[cfg(not(no_global_oom_handling))]
463 #[rustc_allow_incoherent_impl]
464 #[must_use = "this returns the uppercase string as a new String, \
465 without modifying the original"]
466 #[stable(feature = "unicode_case_mapping", since = "1.2.0")]
467 pub fn to_uppercase(&self) -> String {
468 let (mut s, rest) = convert_while_ascii(self, u8::to_ascii_uppercase);
469
470 for c in rest.chars() {
471 match conversions::to_upper(c) {
472 [a, '\0', _] => s.push(a),
473 [a, b, '\0'] => {
474 s.push(a);
475 s.push(b);
476 }
477 [a, b, c] => {
478 s.push(a);
479 s.push(b);
480 s.push(c);
481 }
482 }
483 }
484 s
485 }
486
487 /// Converts a [`Box<str>`] into a [`String`] without copying or allocating.
488 ///
489 /// # Examples
490 ///
491 /// ```
492 /// let string = String::from("birthday gift");
493 /// let boxed_str = string.clone().into_boxed_str();
494 ///
495 /// assert_eq!(boxed_str.into_string(), string);
496 /// ```
497 #[stable(feature = "box_str", since = "1.4.0")]
498 #[rustc_allow_incoherent_impl]
499 #[must_use = "`self` will be dropped if the result is not used"]
500 #[inline]
501 pub fn into_string(self: Box<Self>) -> String {
502 let slice = Box::<[u8]>::from(self);
503 unsafe { String::from_utf8_unchecked(slice.into_vec()) }
504 }
505
506 /// Creates a new [`String`] by repeating a string `n` times.
507 ///
508 /// # Panics
509 ///
510 /// This function will panic if the capacity would overflow.
511 ///
512 /// # Examples
513 ///
514 /// Basic usage:
515 ///
516 /// ```
517 /// assert_eq!("abc".repeat(4), String::from("abcabcabcabc"));
518 /// ```
519 ///
520 /// A panic upon overflow:
521 ///
522 /// ```should_panic
523 /// // this will panic at runtime
524 /// let huge = "0123456789abcdef".repeat(usize::MAX);
525 /// ```
526 #[cfg(not(no_global_oom_handling))]
527 #[rustc_allow_incoherent_impl]
528 #[must_use]
529 #[stable(feature = "repeat_str", since = "1.16.0")]
530 #[inline]
531 pub fn repeat(&self, n: usize) -> String {
532 unsafe { String::from_utf8_unchecked(self.as_bytes().repeat(n)) }
533 }
534
535 /// Returns a copy of this string where each character is mapped to its
536 /// ASCII upper case equivalent.
537 ///
538 /// ASCII letters 'a' to 'z' are mapped to 'A' to 'Z',
539 /// but non-ASCII letters are unchanged.
540 ///
541 /// To uppercase the value in-place, use [`make_ascii_uppercase`].
542 ///
543 /// To uppercase ASCII characters in addition to non-ASCII characters, use
544 /// [`to_uppercase`].
545 ///
546 /// # Examples
547 ///
548 /// ```
549 /// let s = "Grüße, Jürgen ❤";
550 ///
551 /// assert_eq!("GRüßE, JüRGEN ❤", s.to_ascii_uppercase());
552 /// ```
553 ///
554 /// [`make_ascii_uppercase`]: str::make_ascii_uppercase
555 /// [`to_uppercase`]: #method.to_uppercase
556 #[cfg(not(no_global_oom_handling))]
557 #[rustc_allow_incoherent_impl]
558 #[must_use = "to uppercase the value in-place, use `make_ascii_uppercase()`"]
559 #[stable(feature = "ascii_methods_on_intrinsics", since = "1.23.0")]
560 #[inline]
561 pub fn to_ascii_uppercase(&self) -> String {
562 let mut s = self.to_owned();
563 s.make_ascii_uppercase();
564 s
565 }
566
567 /// Returns a copy of this string where each character is mapped to its
568 /// ASCII lower case equivalent.
569 ///
570 /// ASCII letters 'A' to 'Z' are mapped to 'a' to 'z',
571 /// but non-ASCII letters are unchanged.
572 ///
573 /// To lowercase the value in-place, use [`make_ascii_lowercase`].
574 ///
575 /// To lowercase ASCII characters in addition to non-ASCII characters, use
576 /// [`to_lowercase`].
577 ///
578 /// # Examples
579 ///
580 /// ```
581 /// let s = "Grüße, Jürgen ❤";
582 ///
583 /// assert_eq!("grüße, jürgen ❤", s.to_ascii_lowercase());
584 /// ```
585 ///
586 /// [`make_ascii_lowercase`]: str::make_ascii_lowercase
587 /// [`to_lowercase`]: #method.to_lowercase
588 #[cfg(not(no_global_oom_handling))]
589 #[rustc_allow_incoherent_impl]
590 #[must_use = "to lowercase the value in-place, use `make_ascii_lowercase()`"]
591 #[stable(feature = "ascii_methods_on_intrinsics", since = "1.23.0")]
592 #[inline]
593 pub fn to_ascii_lowercase(&self) -> String {
594 let mut s = self.to_owned();
595 s.make_ascii_lowercase();
596 s
597 }
598}
599
600/// Converts a boxed slice of bytes to a boxed string slice without checking
601/// that the string contains valid UTF-8.
602///
603/// # Safety
604///
605/// * The provided bytes must contain a valid UTF-8 sequence.
606///
607/// # Examples
608///
609/// ```
610/// let smile_utf8 = Box::new([226, 152, 186]);
611/// let smile = unsafe { std::str::from_boxed_utf8_unchecked(smile_utf8) };
612///
613/// assert_eq!("☺", &*smile);
614/// ```
615#[stable(feature = "str_box_extras", since = "1.20.0")]
616#[must_use]
617#[inline]
618pub unsafe fn from_boxed_utf8_unchecked(v: Box<[u8]>) -> Box<str> {
619 unsafe { Box::from_raw(Box::into_raw(v) as *mut str) }
620}
621
622/// Converts leading ascii bytes in `s` by calling the `convert` function.
623///
624/// For better average performance, this happens in chunks of `2*size_of::<usize>()`.
625///
626/// Returns a tuple of the converted prefix and the remainder starting from
627/// the first non-ascii character.
628///
629/// This function is only public so that it can be verified in a codegen test,
630/// see `issue-123712-str-to-lower-autovectorization.rs`.
631#[unstable(feature = "str_internals", issue = "none")]
632#[doc(hidden)]
633#[inline]
634#[cfg(not(no_global_oom_handling))]
635pub fn convert_while_ascii(s: &str, convert: fn(&u8) -> u8) -> (String, &str) {
636 // Process the input in chunks of 16 bytes to enable auto-vectorization.
637 // Previously the chunk size depended on the size of `usize`,
638 // but on 32-bit platforms with sse or neon is also the better choice.
639 // The only downside on other platforms would be a bit more loop-unrolling.
640 const N: usize = 16;
641
642 let mut slice = s.as_bytes();
643 let mut out = Vec::with_capacity(slice.len());
644 let mut out_slice = out.spare_capacity_mut();
645
646 let mut ascii_prefix_len = 0_usize;
647 let mut is_ascii = [false; N];
648
649 while slice.len() >= N {
650 // SAFETY: checked in loop condition
651 let chunk = unsafe { slice.get_unchecked(..N) };
652 // SAFETY: out_slice has at least same length as input slice and gets sliced with the same offsets
653 let out_chunk = unsafe { out_slice.get_unchecked_mut(..N) };
654
655 for j in 0..N {
656 is_ascii[j] = chunk[j] <= 127;
657 }
658
659 // Auto-vectorization for this check is a bit fragile, sum and comparing against the chunk
660 // size gives the best result, specifically a pmovmsk instruction on x86.
661 // See https://github.com/llvm/llvm-project/issues/96395 for why llvm currently does not
662 // currently recognize other similar idioms.
663 if is_ascii.iter().map(|x| *x as u8).sum::<u8>() as usize != N {
664 break;
665 }
666
667 for j in 0..N {
668 out_chunk[j] = MaybeUninit::new(convert(&chunk[j]));
669 }
670
671 ascii_prefix_len += N;
672 slice = unsafe { slice.get_unchecked(N..) };
673 out_slice = unsafe { out_slice.get_unchecked_mut(N..) };
674 }
675
676 // handle the remainder as individual bytes
677 while slice.len() > 0 {
678 let byte = slice[0];
679 if byte > 127 {
680 break;
681 }
682 // SAFETY: out_slice has at least same length as input slice
683 unsafe {
684 *out_slice.get_unchecked_mut(0) = MaybeUninit::new(convert(&byte));
685 }
686 ascii_prefix_len += 1;
687 slice = unsafe { slice.get_unchecked(1..) };
688 out_slice = unsafe { out_slice.get_unchecked_mut(1..) };
689 }
690
691 unsafe {
692 // SAFETY: ascii_prefix_len bytes have been initialized above
693 out.set_len(ascii_prefix_len);
694
695 // SAFETY: We have written only valid ascii to the output vec
696 let ascii_string = String::from_utf8_unchecked(out);
697
698 // SAFETY: we know this is a valid char boundary
699 // since we only skipped over leading ascii bytes
700 let rest = core::str::from_utf8_unchecked(slice);
701
702 (ascii_string, rest)
703 }
704}
705#[inline]
706#[cfg(not(no_global_oom_handling))]
707#[allow(dead_code)]
708/// Faster implementation of string replacement for ASCII to ASCII cases.
709/// Should produce fast vectorized code.
710unsafe fn replace_ascii(utf8_bytes: &[u8], from: u8, to: u8) -> String {
711 let result: Vec<u8> = utf8_bytes.iter().map(|b: &u8| if *b == from { to } else { *b }).collect();
712 // SAFETY: We replaced ascii with ascii on valid utf8 strings.
713 unsafe { String::from_utf8_unchecked(bytes:result) }
714}
715