wtf8.rs - source (original) (raw)
std/sys_common/
wtf8.rs
1//! Implementation of [the WTF-8 encoding](https://simonsapin.github.io/wtf-8/).
2//!
3//! This library uses Rust’s type system to maintain
4//! [well-formedness](https://simonsapin.github.io/wtf-8/#well-formed),
5//! like the `String` and `&str` types do for UTF-8.
6//!
7//! Since [WTF-8 must not be used
8//! for interchange](https://simonsapin.github.io/wtf-8/#intended-audience),
9//! this library deliberately does not provide access to the underlying bytes
10//! of WTF-8 strings,
11//! nor can it decode WTF-8 from arbitrary bytes.
12//! WTF-8 strings can be obtained from UTF-8, UTF-16, or code points.
13
14// this module is imported from @SimonSapin's repo and has tons of dead code on
15// unix (it's mostly used on windows), so don't worry about dead code here.
16#![allow(dead_code)]
17
18#[cfg(test)]
19mod tests;
20
21use core::char::{MAX_LEN_UTF8, MAX_LEN_UTF16, encode_utf8_raw, encode_utf16_raw};
22use core::clone::CloneToUninit;
23use core::str::next_code_point;
24
25use crate::borrow::Cow;
26use crate::collections::TryReserveError;
27use crate::hash::{Hash, Hasher};
28use crate::iter::FusedIterator;
29use crate::rc::Rc;
30use crate::sync::Arc;
31use crate::sys_common::AsInner;
32use crate::{fmt, mem, ops, slice, str};
33
34const UTF8_REPLACEMENT_CHARACTER: &str = "\u{FFFD}";
35
36/// A Unicode code point: from U+0000 to U+10FFFF.
37///
38/// Compares with the `char` type,
39/// which represents a Unicode scalar value:
40/// a code point that is not a surrogate (U+D800 to U+DFFF).
41#[derive(Eq, PartialEq, Ord, PartialOrd, Clone, Copy)]
42pub struct CodePoint {
43 value: u32,
44}
45
46/// Format the code point as `U+` followed by four to six hexadecimal digits.
47/// Example: `U+1F4A9`
48impl fmt::Debug for CodePoint {
49 #[inline]
50 fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
51 write!(formatter, "U+{:04X}", self.value)
52 }
53}
54
55impl CodePoint {
56 /// Unsafely creates a new `CodePoint` without checking the value.
57 ///
58 /// Only use when `value` is known to be less than or equal to 0x10FFFF.
59 #[inline]
60 pub unsafe fn from_u32_unchecked(value: u32) -> CodePoint {
61 CodePoint { value }
62 }
63
64 /// Creates a new `CodePoint` if the value is a valid code point.
65 ///
66 /// Returns `None` if `value` is above 0x10FFFF.
67 #[inline]
68 pub fn from_u32(value: u32) -> Option<CodePoint> {
69 match value {
70 0..=0x10FFFF => Some(CodePoint { value }),
71 _ => None,
72 }
73 }
74
75 /// Creates a new `CodePoint` from a `char`.
76 ///
77 /// Since all Unicode scalar values are code points, this always succeeds.
78 #[inline]
79 pub fn from_char(value: char) -> CodePoint {
80 CodePoint { value: value as u32 }
81 }
82
83 /// Returns the numeric value of the code point.
84 #[inline]
85 pub fn to_u32(&self) -> u32 {
86 self.value
87 }
88
89 /// Returns the numeric value of the code point if it is a leading surrogate.
90 #[inline]
91 pub fn to_lead_surrogate(&self) -> Option<u16> {
92 match self.value {
93 lead @ 0xD800..=0xDBFF => Some(lead as u16),
94 _ => None,
95 }
96 }
97
98 /// Returns the numeric value of the code point if it is a trailing surrogate.
99 #[inline]
100 pub fn to_trail_surrogate(&self) -> Option<u16> {
101 match self.value {
102 trail @ 0xDC00..=0xDFFF => Some(trail as u16),
103 _ => None,
104 }
105 }
106
107 /// Optionally returns a Unicode scalar value for the code point.
108 ///
109 /// Returns `None` if the code point is a surrogate (from U+D800 to U+DFFF).
110 #[inline]
111 pub fn to_char(&self) -> Option<char> {
112 match self.value {
113 0xD800..=0xDFFF => None,
114 _ => Some(unsafe { char::from_u32_unchecked(self.value) }),
115 }
116 }
117
118 /// Returns a Unicode scalar value for the code point.
119 ///
120 /// Returns `'\u{FFFD}'` (the replacement character “�”)
121 /// if the code point is a surrogate (from U+D800 to U+DFFF).
122 #[inline]
123 pub fn to_char_lossy(&self) -> char {
124 self.to_char().unwrap_or('\u{FFFD}')
125 }
126}
127
128/// An owned, growable string of well-formed WTF-8 data.
129///
130/// Similar to `String`, but can additionally contain surrogate code points
131/// if they’re not in a surrogate pair.
132#[derive(Eq, PartialEq, Ord, PartialOrd, Clone)]
133pub struct Wtf8Buf {
134 bytes: Vec<u8>,
135
136 /// Do we know that `bytes` holds a valid UTF-8 encoding? We can easily
137 /// know this if we're constructed from a `String` or `&str`.
138 ///
139 /// It is possible for `bytes` to have valid UTF-8 without this being
140 /// set, such as when we're concatenating `&Wtf8`'s and surrogates become
141 /// paired, as we don't bother to rescan the entire string.
142 is_known_utf8: bool,
143}
144
145impl ops::Deref for Wtf8Buf {
146 type Target = Wtf8;
147
148 fn deref(&self) -> &Wtf8 {
149 self.as_slice()
150 }
151}
152
153impl ops::DerefMut for Wtf8Buf {
154 fn deref_mut(&mut self) -> &mut Wtf8 {
155 self.as_mut_slice()
156 }
157}
158
159/// Formats the string in double quotes, with characters escaped according to
160/// [`char::escape_debug`] and unpaired surrogates represented as `\u{xxxx}`,
161/// where each `x` is a hexadecimal digit.
162///
163/// For example, the code units [U+0061, U+D800, U+000A] are formatted as
164/// `"a\u{D800}\n"`.
165impl fmt::Debug for Wtf8Buf {
166 #[inline]
167 fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
168 fmt::Debug::fmt(&**self, formatter)
169 }
170}
171
172/// Formats the string with unpaired surrogates substituted with the replacement
173/// character, U+FFFD.
174impl fmt::Display for Wtf8Buf {
175 fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
176 if let Some(s) = self.as_known_utf8() {
177 fmt::Display::fmt(s, formatter)
178 } else {
179 fmt::Display::fmt(&**self, formatter)
180 }
181 }
182}
183
184impl Wtf8Buf {
185 /// Creates a new, empty WTF-8 string.
186 #[inline]
187 pub fn new() -> Wtf8Buf {
188 Wtf8Buf { bytes: Vec::new(), is_known_utf8: true }
189 }
190
191 /// Creates a new, empty WTF-8 string with pre-allocated capacity for `capacity` bytes.
192 #[inline]
193 pub fn with_capacity(capacity: usize) -> Wtf8Buf {
194 Wtf8Buf { bytes: Vec::with_capacity(capacity), is_known_utf8: true }
195 }
196
197 /// Creates a WTF-8 string from a WTF-8 byte vec.
198 ///
199 /// Since the byte vec is not checked for valid WTF-8, this function is
200 /// marked unsafe.
201 #[inline]
202 pub unsafe fn from_bytes_unchecked(value: Vec<u8>) -> Wtf8Buf {
203 Wtf8Buf { bytes: value, is_known_utf8: false }
204 }
205
206 /// Creates a WTF-8 string from a UTF-8 `String`.
207 ///
208 /// This takes ownership of the `String` and does not copy.
209 ///
210 /// Since WTF-8 is a superset of UTF-8, this always succeeds.
211 #[inline]
212 pub fn from_string(string: String) -> Wtf8Buf {
213 Wtf8Buf { bytes: string.into_bytes(), is_known_utf8: true }
214 }
215
216 /// Creates a WTF-8 string from a UTF-8 `&str` slice.
217 ///
218 /// This copies the content of the slice.
219 ///
220 /// Since WTF-8 is a superset of UTF-8, this always succeeds.
221 #[inline]
222 pub fn from_str(s: &str) -> Wtf8Buf {
223 Wtf8Buf { bytes: s.as_bytes().to_vec(), is_known_utf8: true }
224 }
225
226 pub fn clear(&mut self) {
227 self.bytes.clear();
228 self.is_known_utf8 = true;
229 }
230
231 /// Creates a WTF-8 string from a potentially ill-formed UTF-16 slice of 16-bit code units.
232 ///
233 /// This is lossless: calling `.encode_wide()` on the resulting string
234 /// will always return the original code units.
235 pub fn from_wide(v: &[u16]) -> Wtf8Buf {
236 let mut string = Wtf8Buf::with_capacity(v.len());
237 for item in char::decode_utf16(v.iter().cloned()) {
238 match item {
239 Ok(ch) => string.push_char(ch),
240 Err(surrogate) => {
241 let surrogate = surrogate.unpaired_surrogate();
242 // Surrogates are known to be in the code point range.
243 let code_point = unsafe { CodePoint::from_u32_unchecked(surrogate as u32) };
244 // The string will now contain an unpaired surrogate.
245 string.is_known_utf8 = false;
246 // Skip the WTF-8 concatenation check,
247 // surrogate pairs are already decoded by decode_utf16
248 string.push_code_point_unchecked(code_point);
249 }
250 }
251 }
252 string
253 }
254
255 /// Appends the given `char` to the end of this string.
256 /// This does **not** include the WTF-8 concatenation check or `is_known_utf8` check.
257 /// Copied from String::push.
258 fn push_code_point_unchecked(&mut self, code_point: CodePoint) {
259 let mut bytes = [0; MAX_LEN_UTF8];
260 let bytes = encode_utf8_raw(code_point.value, &mut bytes);
261 self.bytes.extend_from_slice(bytes)
262 }
263
264 #[inline]
265 pub fn as_slice(&self) -> &Wtf8 {
266 unsafe { Wtf8::from_bytes_unchecked(&self.bytes) }
267 }
268
269 #[inline]
270 pub fn as_mut_slice(&mut self) -> &mut Wtf8 {
271 // Safety: `Wtf8` doesn't expose any way to mutate the bytes that would
272 // cause them to change from well-formed UTF-8 to ill-formed UTF-8,
273 // which would break the assumptions of the `is_known_utf8` field.
274 unsafe { Wtf8::from_mut_bytes_unchecked(&mut self.bytes) }
275 }
276
277 /// Converts the string to UTF-8 without validation, if it was created from
278 /// valid UTF-8.
279 #[inline]
280 fn as_known_utf8(&self) -> Option<&str> {
281 if self.is_known_utf8 {
282 // SAFETY: The buffer is known to be valid UTF-8.
283 Some(unsafe { str::from_utf8_unchecked(self.as_bytes()) })
284 } else {
285 None
286 }
287 }
288
289 /// Reserves capacity for at least `additional` more bytes to be inserted
290 /// in the given `Wtf8Buf`.
291 /// The collection may reserve more space to avoid frequent reallocations.
292 ///
293 /// # Panics
294 ///
295 /// Panics if the new capacity exceeds `isize::MAX` bytes.
296 #[inline]
297 pub fn reserve(&mut self, additional: usize) {
298 self.bytes.reserve(additional)
299 }
300
301 /// Tries to reserve capacity for at least `additional` more bytes to be
302 /// inserted in the given `Wtf8Buf`. The `Wtf8Buf` may reserve more space to
303 /// avoid frequent reallocations. After calling `try_reserve`, capacity will
304 /// be greater than or equal to `self.len() + additional`. Does nothing if
305 /// capacity is already sufficient. This method preserves the contents even
306 /// if an error occurs.
307 ///
308 /// # Errors
309 ///
310 /// If the capacity overflows, or the allocator reports a failure, then an error
311 /// is returned.
312 #[inline]
313 pub fn try_reserve(&mut self, additional: usize) -> Result<(), TryReserveError> {
314 self.bytes.try_reserve(additional)
315 }
316
317 #[inline]
318 pub fn reserve_exact(&mut self, additional: usize) {
319 self.bytes.reserve_exact(additional)
320 }
321
322 /// Tries to reserve the minimum capacity for exactly `additional` more
323 /// bytes to be inserted in the given `Wtf8Buf`. After calling
324 /// `try_reserve_exact`, capacity will be greater than or equal to
325 /// `self.len() + additional` if it returns `Ok(())`.
326 /// Does nothing if the capacity is already sufficient.
327 ///
328 /// Note that the allocator may give the `Wtf8Buf` more space than it
329 /// requests. Therefore, capacity can not be relied upon to be precisely
330 /// minimal. Prefer [`try_reserve`] if future insertions are expected.
331 ///
332 /// [`try_reserve`]: Wtf8Buf::try_reserve
333 ///
334 /// # Errors
335 ///
336 /// If the capacity overflows, or the allocator reports a failure, then an error
337 /// is returned.
338 #[inline]
339 pub fn try_reserve_exact(&mut self, additional: usize) -> Result<(), TryReserveError> {
340 self.bytes.try_reserve_exact(additional)
341 }
342
343 #[inline]
344 pub fn shrink_to_fit(&mut self) {
345 self.bytes.shrink_to_fit()
346 }
347
348 #[inline]
349 pub fn shrink_to(&mut self, min_capacity: usize) {
350 self.bytes.shrink_to(min_capacity)
351 }
352
353 #[inline]
354 pub fn leak<'a>(self) -> &'a mut Wtf8 {
355 unsafe { Wtf8::from_mut_bytes_unchecked(self.bytes.leak()) }
356 }
357
358 /// Returns the number of bytes that this string buffer can hold without reallocating.
359 #[inline]
360 pub fn capacity(&self) -> usize {
361 self.bytes.capacity()
362 }
363
364 /// Append a UTF-8 slice at the end of the string.
365 #[inline]
366 pub fn push_str(&mut self, other: &str) {
367 self.bytes.extend_from_slice(other.as_bytes())
368 }
369
370 /// Append a WTF-8 slice at the end of the string.
371 ///
372 /// This replaces newly paired surrogates at the boundary
373 /// with a supplementary code point,
374 /// like concatenating ill-formed UTF-16 strings effectively would.
375 #[inline]
376 pub fn push_wtf8(&mut self, other: &Wtf8) {
377 match ((&*self).final_lead_surrogate(), other.initial_trail_surrogate()) {
378 // Replace newly paired surrogates by a supplementary code point.
379 (Some(lead), Some(trail)) => {
380 let len_without_lead_surrogate = self.len() - 3;
381 self.bytes.truncate(len_without_lead_surrogate);
382 let other_without_trail_surrogate = &other.bytes[3..];
383 // 4 bytes for the supplementary code point
384 self.bytes.reserve(4 + other_without_trail_surrogate.len());
385 self.push_char(decode_surrogate_pair(lead, trail));
386 self.bytes.extend_from_slice(other_without_trail_surrogate);
387 }
388 _ => {
389 // If we'll be pushing a string containing a surrogate, we may
390 // no longer have UTF-8.
391 if self.is_known_utf8 && other.next_surrogate(0).is_some() {
392 self.is_known_utf8 = false;
393 }
394
395 self.bytes.extend_from_slice(&other.bytes);
396 }
397 }
398 }
399
400 /// Append a Unicode scalar value at the end of the string.
401 #[inline]
402 pub fn push_char(&mut self, c: char) {
403 self.push_code_point_unchecked(CodePoint::from_char(c))
404 }
405
406 /// Append a code point at the end of the string.
407 ///
408 /// This replaces newly paired surrogates at the boundary
409 /// with a supplementary code point,
410 /// like concatenating ill-formed UTF-16 strings effectively would.
411 #[inline]
412 pub fn push(&mut self, code_point: CodePoint) {
413 if let Some(trail) = code_point.to_trail_surrogate() {
414 if let Some(lead) = (&*self).final_lead_surrogate() {
415 let len_without_lead_surrogate = self.len() - 3;
416 self.bytes.truncate(len_without_lead_surrogate);
417 self.push_char(decode_surrogate_pair(lead, trail));
418 return;
419 }
420
421 // We're pushing a trailing surrogate.
422 self.is_known_utf8 = false;
423 } else if code_point.to_lead_surrogate().is_some() {
424 // We're pushing a leading surrogate.
425 self.is_known_utf8 = false;
426 }
427
428 // No newly paired surrogates at the boundary.
429 self.push_code_point_unchecked(code_point)
430 }
431
432 /// Shortens a string to the specified length.
433 ///
434 /// # Panics
435 ///
436 /// Panics if `new_len` > current length,
437 /// or if `new_len` is not a code point boundary.
438 #[inline]
439 pub fn truncate(&mut self, new_len: usize) {
440 assert!(is_code_point_boundary(self, new_len));
441 self.bytes.truncate(new_len)
442 }
443
444 /// Consumes the WTF-8 string and tries to convert it to a vec of bytes.
445 #[inline]
446 pub fn into_bytes(self) -> Vec<u8> {
447 self.bytes
448 }
449
450 /// Consumes the WTF-8 string and tries to convert it to UTF-8.
451 ///
452 /// This does not copy the data.
453 ///
454 /// If the contents are not well-formed UTF-8
455 /// (that is, if the string contains surrogates),
456 /// the original WTF-8 string is returned instead.
457 pub fn into_string(self) -> Result<String, Wtf8Buf> {
458 if self.is_known_utf8 || self.next_surrogate(0).is_none() {
459 Ok(unsafe { String::from_utf8_unchecked(self.bytes) })
460 } else {
461 Err(self)
462 }
463 }
464
465 /// Consumes the WTF-8 string and converts it lossily to UTF-8.
466 ///
467 /// This does not copy the data (but may overwrite parts of it in place).
468 ///
469 /// Surrogates are replaced with `"\u{FFFD}"` (the replacement character “�”)
470 pub fn into_string_lossy(mut self) -> String {
471 if !self.is_known_utf8 {
472 let mut pos = 0;
473 while let Some((surrogate_pos, _)) = self.next_surrogate(pos) {
474 pos = surrogate_pos + 3;
475 // Surrogates and the replacement character are all 3 bytes, so
476 // they can substituted in-place.
477 self.bytes[surrogate_pos..pos]
478 .copy_from_slice(UTF8_REPLACEMENT_CHARACTER.as_bytes());
479 }
480 }
481 unsafe { String::from_utf8_unchecked(self.bytes) }
482 }
483
484 /// Converts this `Wtf8Buf` into a boxed `Wtf8`.
485 #[inline]
486 pub fn into_box(self) -> Box<Wtf8> {
487 // SAFETY: relies on `Wtf8` being `repr(transparent)`.
488 unsafe { mem::transmute(self.bytes.into_boxed_slice()) }
489 }
490
491 /// Converts a `Box<Wtf8>` into a `Wtf8Buf`.
492 pub fn from_box(boxed: Box<Wtf8>) -> Wtf8Buf {
493 let bytes: Box<[u8]> = unsafe { mem::transmute(boxed) };
494 Wtf8Buf { bytes: bytes.into_vec(), is_known_utf8: false }
495 }
496
497 /// Provides plumbing to core `Vec::extend_from_slice`.
498 /// More well behaving alternative to allowing outer types
499 /// full mutable access to the core `Vec`.
500 #[inline]
501 pub(crate) fn extend_from_slice(&mut self, other: &[u8]) {
502 self.bytes.extend_from_slice(other);
503 self.is_known_utf8 = false;
504 }
505}
506
507/// Creates a new WTF-8 string from an iterator of code points.
508///
509/// This replaces surrogate code point pairs with supplementary code points,
510/// like concatenating ill-formed UTF-16 strings effectively would.
511impl FromIterator<CodePoint> for Wtf8Buf {
512 fn from_iter<T: IntoIterator<Item = CodePoint>>(iter: T) -> Wtf8Buf {
513 let mut string = Wtf8Buf::new();
514 string.extend(iter);
515 string
516 }
517}
518
519/// Append code points from an iterator to the string.
520///
521/// This replaces surrogate code point pairs with supplementary code points,
522/// like concatenating ill-formed UTF-16 strings effectively would.
523impl Extend<CodePoint> for Wtf8Buf {
524 fn extend<T: IntoIterator<Item = CodePoint>>(&mut self, iter: T) {
525 let iterator = iter.into_iter();
526 let (low, _high) = iterator.size_hint();
527 // Lower bound of one byte per code point (ASCII only)
528 self.bytes.reserve(low);
529 iterator.for_each(move |code_point| self.push(code_point));
530 }
531
532 #[inline]
533 fn extend_one(&mut self, code_point: CodePoint) {
534 self.push(code_point);
535 }
536
537 #[inline]
538 fn extend_reserve(&mut self, additional: usize) {
539 // Lower bound of one byte per code point (ASCII only)
540 self.bytes.reserve(additional);
541 }
542}
543
544/// A borrowed slice of well-formed WTF-8 data.
545///
546/// Similar to `&str`, but can additionally contain surrogate code points
547/// if they’re not in a surrogate pair.
548#[derive(Eq, Ord, PartialEq, PartialOrd)]
549#[repr(transparent)]
550pub struct Wtf8 {
551 bytes: [u8],
552}
553
554impl AsInner<[u8]> for Wtf8 {
555 #[inline]
556 fn as_inner(&self) -> &[u8] {
557 &self.bytes
558 }
559}
560
561/// Formats the string in double quotes, with characters escaped according to
562/// [`char::escape_debug`] and unpaired surrogates represented as `\u{xxxx}`,
563/// where each `x` is a hexadecimal digit.
564impl fmt::Debug for Wtf8 {
565 fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
566 fn write_str_escaped(f: &mut fmt::Formatter<'_>, s: &str) -> fmt::Result {
567 use crate::fmt::Write;
568 for c in s.chars().flat_map(|c| c.escape_debug()) {
569 f.write_char(c)?
570 }
571 Ok(())
572 }
573
574 formatter.write_str("\"")?;
575 let mut pos = 0;
576 while let Some((surrogate_pos, surrogate)) = self.next_surrogate(pos) {
577 write_str_escaped(formatter, unsafe {
578 str::from_utf8_unchecked(&self.bytes[pos..surrogate_pos])
579 })?;
580 write!(formatter, "\\u{{{:x}}}", surrogate)?;
581 pos = surrogate_pos + 3;
582 }
583 write_str_escaped(formatter, unsafe { str::from_utf8_unchecked(&self.bytes[pos..]) })?;
584 formatter.write_str("\"")
585 }
586}
587
588/// Formats the string with unpaired surrogates substituted with the replacement
589/// character, U+FFFD.
590impl fmt::Display for Wtf8 {
591 fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
592 let wtf8_bytes = &self.bytes;
593 let mut pos = 0;
594 loop {
595 match self.next_surrogate(pos) {
596 Some((surrogate_pos, _)) => {
597 formatter.write_str(unsafe {
598 str::from_utf8_unchecked(&wtf8_bytes[pos..surrogate_pos])
599 })?;
600 formatter.write_str(UTF8_REPLACEMENT_CHARACTER)?;
601 pos = surrogate_pos + 3;
602 }
603 None => {
604 let s = unsafe { str::from_utf8_unchecked(&wtf8_bytes[pos..]) };
605 if pos == 0 { return s.fmt(formatter) } else { return formatter.write_str(s) }
606 }
607 }
608 }
609 }
610}
611
612impl Wtf8 {
613 /// Creates a WTF-8 slice from a UTF-8 `&str` slice.
614 ///
615 /// Since WTF-8 is a superset of UTF-8, this always succeeds.
616 #[inline]
617 pub fn from_str(value: &str) -> &Wtf8 {
618 unsafe { Wtf8::from_bytes_unchecked(value.as_bytes()) }
619 }
620
621 /// Creates a WTF-8 slice from a WTF-8 byte slice.
622 ///
623 /// Since the byte slice is not checked for valid WTF-8, this functions is
624 /// marked unsafe.
625 #[inline]
626 pub unsafe fn from_bytes_unchecked(value: &[u8]) -> &Wtf8 {
627 // SAFETY: start with &[u8], end with fancy &[u8]
628 unsafe { &*(value as *const [u8] as *const Wtf8) }
629 }
630
631 /// Creates a mutable WTF-8 slice from a mutable WTF-8 byte slice.
632 ///
633 /// Since the byte slice is not checked for valid WTF-8, this functions is
634 /// marked unsafe.
635 #[inline]
636 unsafe fn from_mut_bytes_unchecked(value: &mut [u8]) -> &mut Wtf8 {
637 // SAFETY: start with &mut [u8], end with fancy &mut [u8]
638 unsafe { &mut *(value as *mut [u8] as *mut Wtf8) }
639 }
640
641 /// Returns the length, in WTF-8 bytes.
642 #[inline]
643 pub fn len(&self) -> usize {
644 self.bytes.len()
645 }
646
647 #[inline]
648 pub fn is_empty(&self) -> bool {
649 self.bytes.is_empty()
650 }
651
652 /// Returns the code point at `position` if it is in the ASCII range,
653 /// or `b'\xFF'` otherwise.
654 ///
655 /// # Panics
656 ///
657 /// Panics if `position` is beyond the end of the string.
658 #[inline]
659 pub fn ascii_byte_at(&self, position: usize) -> u8 {
660 match self.bytes[position] {
661 ascii_byte @ 0x00..=0x7F => ascii_byte,
662 _ => 0xFF,
663 }
664 }
665
666 /// Returns an iterator for the string’s code points.
667 #[inline]
668 pub fn code_points(&self) -> Wtf8CodePoints<'_> {
669 Wtf8CodePoints { bytes: self.bytes.iter() }
670 }
671
672 /// Access raw bytes of WTF-8 data
673 #[inline]
674 pub fn as_bytes(&self) -> &[u8] {
675 &self.bytes
676 }
677
678 /// Tries to convert the string to UTF-8 and return a `&str` slice.
679 ///
680 /// Returns `None` if the string contains surrogates.
681 ///
682 /// This does not copy the data.
683 #[inline]
684 pub fn as_str(&self) -> Result<&str, str::Utf8Error> {
685 str::from_utf8(&self.bytes)
686 }
687
688 /// Creates an owned `Wtf8Buf` from a borrowed `Wtf8`.
689 pub fn to_owned(&self) -> Wtf8Buf {
690 Wtf8Buf { bytes: self.bytes.to_vec(), is_known_utf8: false }
691 }
692
693 /// Lossily converts the string to UTF-8.
694 /// Returns a UTF-8 `&str` slice if the contents are well-formed in UTF-8.
695 ///
696 /// Surrogates are replaced with `"\u{FFFD}"` (the replacement character “�”).
697 ///
698 /// This only copies the data if necessary (if it contains any surrogate).
699 pub fn to_string_lossy(&self) -> Cow<'_, str> {
700 let Some((surrogate_pos, _)) = self.next_surrogate(0) else {
701 return Cow::Borrowed(unsafe { str::from_utf8_unchecked(&self.bytes) });
702 };
703 let wtf8_bytes = &self.bytes;
704 let mut utf8_bytes = Vec::with_capacity(self.len());
705 utf8_bytes.extend_from_slice(&wtf8_bytes[..surrogate_pos]);
706 utf8_bytes.extend_from_slice(UTF8_REPLACEMENT_CHARACTER.as_bytes());
707 let mut pos = surrogate_pos + 3;
708 loop {
709 match self.next_surrogate(pos) {
710 Some((surrogate_pos, _)) => {
711 utf8_bytes.extend_from_slice(&wtf8_bytes[pos..surrogate_pos]);
712 utf8_bytes.extend_from_slice(UTF8_REPLACEMENT_CHARACTER.as_bytes());
713 pos = surrogate_pos + 3;
714 }
715 None => {
716 utf8_bytes.extend_from_slice(&wtf8_bytes[pos..]);
717 return Cow::Owned(unsafe { String::from_utf8_unchecked(utf8_bytes) });
718 }
719 }
720 }
721 }
722
723 /// Converts the WTF-8 string to potentially ill-formed UTF-16
724 /// and return an iterator of 16-bit code units.
725 ///
726 /// This is lossless:
727 /// calling `Wtf8Buf::from_ill_formed_utf16` on the resulting code units
728 /// would always return the original WTF-8 string.
729 #[inline]
730 pub fn encode_wide(&self) -> EncodeWide<'_> {
731 EncodeWide { code_points: self.code_points(), extra: 0 }
732 }
733
734 #[inline]
735 fn next_surrogate(&self, mut pos: usize) -> Option<(usize, u16)> {
736 let mut iter = self.bytes[pos..].iter();
737 loop {
738 let b = *iter.next()?;
739 if b < 0x80 {
740 pos += 1;
741 } else if b < 0xE0 {
742 iter.next();
743 pos += 2;
744 } else if b == 0xED {
745 match (iter.next(), iter.next()) {
746 (Some(&b2), Some(&b3)) if b2 >= 0xA0 => {
747 return Some((pos, decode_surrogate(b2, b3)));
748 }
749 _ => pos += 3,
750 }
751 } else if b < 0xF0 {
752 iter.next();
753 iter.next();
754 pos += 3;
755 } else {
756 iter.next();
757 iter.next();
758 iter.next();
759 pos += 4;
760 }
761 }
762 }
763
764 #[inline]
765 fn final_lead_surrogate(&self) -> Option<u16> {
766 match self.bytes {
767 [.., 0xED, b2 @ 0xA0..=0xAF, b3] => Some(decode_surrogate(b2, b3)),
768 _ => None,
769 }
770 }
771
772 #[inline]
773 fn initial_trail_surrogate(&self) -> Option<u16> {
774 match self.bytes {
775 [0xED, b2 @ 0xB0..=0xBF, b3, ..] => Some(decode_surrogate(b2, b3)),
776 _ => None,
777 }
778 }
779
780 pub fn clone_into(&self, buf: &mut Wtf8Buf) {
781 buf.is_known_utf8 = false;
782 self.bytes.clone_into(&mut buf.bytes);
783 }
784
785 /// Boxes this `Wtf8`.
786 #[inline]
787 pub fn into_box(&self) -> Box<Wtf8> {
788 let boxed: Box<[u8]> = self.bytes.into();
789 unsafe { mem::transmute(boxed) }
790 }
791
792 /// Creates a boxed, empty `Wtf8`.
793 pub fn empty_box() -> Box<Wtf8> {
794 let boxed: Box<[u8]> = Default::default();
795 unsafe { mem::transmute(boxed) }
796 }
797
798 #[inline]
799 pub fn into_arc(&self) -> Arc<Wtf8> {
800 let arc: Arc<[u8]> = Arc::from(&self.bytes);
801 unsafe { Arc::from_raw(Arc::into_raw(arc) as *const Wtf8) }
802 }
803
804 #[inline]
805 pub fn into_rc(&self) -> Rc<Wtf8> {
806 let rc: Rc<[u8]> = Rc::from(&self.bytes);
807 unsafe { Rc::from_raw(Rc::into_raw(rc) as *const Wtf8) }
808 }
809
810 #[inline]
811 pub fn make_ascii_lowercase(&mut self) {
812 self.bytes.make_ascii_lowercase()
813 }
814
815 #[inline]
816 pub fn make_ascii_uppercase(&mut self) {
817 self.bytes.make_ascii_uppercase()
818 }
819
820 #[inline]
821 pub fn to_ascii_lowercase(&self) -> Wtf8Buf {
822 Wtf8Buf { bytes: self.bytes.to_ascii_lowercase(), is_known_utf8: false }
823 }
824
825 #[inline]
826 pub fn to_ascii_uppercase(&self) -> Wtf8Buf {
827 Wtf8Buf { bytes: self.bytes.to_ascii_uppercase(), is_known_utf8: false }
828 }
829
830 #[inline]
831 pub fn is_ascii(&self) -> bool {
832 self.bytes.is_ascii()
833 }
834
835 #[inline]
836 pub fn eq_ignore_ascii_case(&self, other: &Self) -> bool {
837 self.bytes.eq_ignore_ascii_case(&other.bytes)
838 }
839}
840
841/// Returns a slice of the given string for the byte range \[`begin`..`end`).
842///
843/// # Panics
844///
845/// Panics when `begin` and `end` do not point to code point boundaries,
846/// or point beyond the end of the string.
847impl ops::Index<ops::Range<usize>> for Wtf8 {
848 type Output = Wtf8;
849
850 #[inline]
851 fn index(&self, range: ops::Range<usize>) -> &Wtf8 {
852 // is_code_point_boundary checks that the index is in [0, .len()]
853 if range.start <= range.end
854 && is_code_point_boundary(self, range.start)
855 && is_code_point_boundary(self, range.end)
856 {
857 unsafe { slice_unchecked(self, range.start, range.end) }
858 } else {
859 slice_error_fail(self, range.start, range.end)
860 }
861 }
862}
863
864/// Returns a slice of the given string from byte `begin` to its end.
865///
866/// # Panics
867///
868/// Panics when `begin` is not at a code point boundary,
869/// or is beyond the end of the string.
870impl ops::Index<ops::RangeFrom<usize>> for Wtf8 {
871 type Output = Wtf8;
872
873 #[inline]
874 fn index(&self, range: ops::RangeFrom<usize>) -> &Wtf8 {
875 // is_code_point_boundary checks that the index is in [0, .len()]
876 if is_code_point_boundary(self, range.start) {
877 unsafe { slice_unchecked(self, range.start, self.len()) }
878 } else {
879 slice_error_fail(self, range.start, self.len())
880 }
881 }
882}
883
884/// Returns a slice of the given string from its beginning to byte `end`.
885///
886/// # Panics
887///
888/// Panics when `end` is not at a code point boundary,
889/// or is beyond the end of the string.
890impl ops::Index<ops::RangeTo<usize>> for Wtf8 {
891 type Output = Wtf8;
892
893 #[inline]
894 fn index(&self, range: ops::RangeTo<usize>) -> &Wtf8 {
895 // is_code_point_boundary checks that the index is in [0, .len()]
896 if is_code_point_boundary(self, range.end) {
897 unsafe { slice_unchecked(self, 0, range.end) }
898 } else {
899 slice_error_fail(self, 0, range.end)
900 }
901 }
902}
903
904impl ops::Index<ops::RangeFull> for Wtf8 {
905 type Output = Wtf8;
906
907 #[inline]
908 fn index(&self, _range: ops::RangeFull) -> &Wtf8 {
909 self
910 }
911}
912
913#[inline]
914fn decode_surrogate(second_byte: u8, third_byte: u8) -> u16 {
915 // The first byte is assumed to be 0xED
916 0xD800 | (second_byte as u16 & 0x3F) << 6 | third_byte as u16 & 0x3F
917}
918
919#[inline]
920fn decode_surrogate_pair(lead: u16, trail: u16) -> char {
921 let code_point = 0x10000 + ((((lead - 0xD800) as u32) << 10) | (trail - 0xDC00) as u32);
922 unsafe { char::from_u32_unchecked(code_point) }
923}
924
925/// Copied from str::is_char_boundary
926#[inline]
927pub fn is_code_point_boundary(slice: &Wtf8, index: usize) -> bool {
928 if index == 0 {
929 return true;
930 }
931 match slice.bytes.get(index) {
932 None => index == slice.len(),
933 Some(&b) => (b as i8) >= -0x40,
934 }
935}
936
937/// Verify that `index` is at the edge of either a valid UTF-8 codepoint
938/// (i.e. a codepoint that's not a surrogate) or of the whole string.
939///
940/// These are the cases currently permitted by `OsStr::slice_encoded_bytes`.
941/// Splitting between surrogates is valid as far as WTF-8 is concerned, but
942/// we do not permit it in the public API because WTF-8 is considered an
943/// implementation detail.
944#[track_caller]
945#[inline]
946pub fn check_utf8_boundary(slice: &Wtf8, index: usize) {
947 if index == 0 {
948 return;
949 }
950 match slice.bytes.get(index) {
951 Some(0xED) => (), // Might be a surrogate
952 Some(&b) if (b as i8) >= -0x40 => return,
953 Some(_) => panic!("byte index {index} is not a codepoint boundary"),
954 None if index == slice.len() => return,
955 None => panic!("byte index {index} is out of bounds"),
956 }
957 if slice.bytes[index + 1] >= 0xA0 {
958 // There's a surrogate after index. Now check before index.
959 if index >= 3 && slice.bytes[index - 3] == 0xED && slice.bytes[index - 2] >= 0xA0 {
960 panic!("byte index {index} lies between surrogate codepoints");
961 }
962 }
963}
964
965/// Copied from core::str::raw::slice_unchecked
966#[inline]
967pub unsafe fn slice_unchecked(s: &Wtf8, begin: usize, end: usize) -> &Wtf8 {
968 // SAFETY: memory layout of a &[u8] and &Wtf8 are the same
969 unsafe {
970 let len = end - begin;
971 let start = s.as_bytes().as_ptr().add(begin);
972 Wtf8::from_bytes_unchecked(slice::from_raw_parts(start, len))
973 }
974}
975
976/// Copied from core::str::raw::slice_error_fail
977#[inline(never)]
978pub fn slice_error_fail(s: &Wtf8, begin: usize, end: usize) -> ! {
979 assert!(begin <= end);
980 panic!("index {begin} and/or {end} in `{s:?}` do not lie on character boundary");
981}
982
983/// Iterator for the code points of a WTF-8 string.
984///
985/// Created with the method `.code_points()`.
986#[derive(Clone)]
987pub struct Wtf8CodePoints<'a> {
988 bytes: slice::Iter<'a, u8>,
989}
990
991impl Iterator for Wtf8CodePoints<'_> {
992 type Item = CodePoint;
993
994 #[inline]
995 fn next(&mut self) -> Option<CodePoint> {
996 // SAFETY: `self.bytes` has been created from a WTF-8 string
997 unsafe { next_code_point(&mut self.bytes).map(|c| CodePoint { value: c }) }
998 }
999
1000 #[inline]
1001 fn size_hint(&self) -> (usize, Option<usize>) {
1002 let len = self.bytes.len();
1003 (len.saturating_add(3) / 4, Some(len))
1004 }
1005}
1006
1007/// Generates a wide character sequence for potentially ill-formed UTF-16.
1008#[stable(feature = "rust1", since = "1.0.0")]
1009#[derive(Clone)]
1010pub struct EncodeWide<'a> {
1011 code_points: Wtf8CodePoints<'a>,
1012 extra: u16,
1013}
1014
1015// Copied from libunicode/u_str.rs
1016#[stable(feature = "rust1", since = "1.0.0")]
1017impl Iterator for EncodeWide<'_> {
1018 type Item = u16;
1019
1020 #[inline]
1021 fn next(&mut self) -> Option<u16> {
1022 if self.extra != 0 {
1023 let tmp = self.extra;
1024 self.extra = 0;
1025 return Some(tmp);
1026 }
1027
1028 let mut buf = [0; MAX_LEN_UTF16];
1029 self.code_points.next().map(|code_point| {
1030 let n = encode_utf16_raw(code_point.value, &mut buf).len();
1031 if n == 2 {
1032 self.extra = buf[1];
1033 }
1034 buf[0]
1035 })
1036 }
1037
1038 #[inline]
1039 fn size_hint(&self) -> (usize, Option<usize>) {
1040 let (low, high) = self.code_points.size_hint();
1041 let ext = (self.extra != 0) as usize;
1042 // every code point gets either one u16 or two u16,
1043 // so this iterator is between 1 or 2 times as
1044 // long as the underlying iterator.
1045 (low + ext, high.and_then(|n| n.checked_mul(2)).and_then(|n| n.checked_add(ext)))
1046 }
1047}
1048
1049#[stable(feature = "encode_wide_fused_iterator", since = "1.62.0")]
1050impl FusedIterator for EncodeWide<'_> {}
1051
1052impl Hash for CodePoint {
1053 #[inline]
1054 fn hash<H: Hasher>(&self, state: &mut H) {
1055 self.value.hash(state)
1056 }
1057}
1058
1059impl Hash for Wtf8Buf {
1060 #[inline]
1061 fn hash<H: Hasher>(&self, state: &mut H) {
1062 state.write(&self.bytes);
1063 0xfeu8.hash(state)
1064 }
1065}
1066
1067impl Hash for Wtf8 {
1068 #[inline]
1069 fn hash<H: Hasher>(&self, state: &mut H) {
1070 state.write(&self.bytes);
1071 0xfeu8.hash(state)
1072 }
1073}
1074
1075#[unstable(feature = "clone_to_uninit", issue = "126799")]
1076unsafe impl CloneToUninit for Wtf8 {
1077 #[inline]
1078 #[cfg_attr(debug_assertions, track_caller)]
1079 unsafe fn clone_to_uninit(&self, dst: *mut u8) {
1080 // SAFETY: we're just a transparent wrapper around [u8]
1081 unsafe { self.bytes.clone_to_uninit(dst) }
1082 }
1083}