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